Proefschrift moller

Imaging patterns of tissue destruction Towards a better discrimination between types of dementia

Christiane Mรถller

The studies described in this thesis were carried out at the VUmc Alzheimer Center and embedded in the neurodegeneration research program of the Neuroscience Campus Amsterdam. The VUmc Alzheimer Center is supported by Innovatiefonds Ziektekostenverzekeraars and by unrestricted grants to the Stichting VUmc Fonds from: AEGON Nederland NV, Heer en mevrouw Capitain, Heineken Nederland NV, Gebroeders Boeschen, Genootschap tot Steun VUmc Alzheimercentrum, RABO Bank Amsterdam, Stichting ZABAWAS, Stichting Alzheimer Nederland, 2BikeforAlzheimer, AlzheimerRally, Stichting Dioraphte, Stichting Buytentwist, Purplefield Investments, Stichting Mooiste Contact Fonds (initiatief van KPN), Stichting Noaber Foundation, Twentse Kabel Groep Holding NV, van Leeuwen-Rietberg Stichting, Willem Meindert de Hoop Stichting, and many kind indiviual donors.

ISBN/EAN: 978-94-6108-948-9 Illustratie, cover design en ontwerp uitnodiging: Marijn Groenewoud Lay-out: Christiane Möller Printing: Gildeprint Copyright: © Christiane Möller, Amsterdam, The Netherlands All rights reserved. No part of this publication may be reproduced or transmitted in any form or by any means without permission of the author, or, when applicable, of the publishers of the scientific papers. Printing of this thesis was kindly supported by: Alzheimercentrum VUmc Internationale Stichting Alzheimer Onderzoek Nutricia Advanced Medical Nutrition Vrije Universiteit

VRIJE UNIVERSITEIT

Imaging patterns of tissue destruction Towards a better discrimination between types of dementia

ACADEMISCH PROEFSCHRIFT ter verkrijging van de graad Doctor aan de Vrije Universiteit Amsterdam, op gezag van de rector magnificus prof.dr. F.A. van der Duyn Schouten, in het openbaar te verdedigen ten overstaan van de promotiecommissie van de Faculteit der Geneeskunde op vrijdag 1 mei 2015 om 11.45 uur in de aula van de universiteit, De Boelelaan 1105

door Christiane Mรถller geboren te Mainz, Duitsland

Promotoren:

prof.dr. Ph. Scheltens prof.dr. F. Barkhof Copromotoren: prof.dr. W.M. van der Flier dr.ir. H. Vrenken

Table of contents Chapter 1 – Introduction Chapter 2 – Patterns of gray matter loss in different manifestations of Alzheimer’s disease 2.1 Different patterns of gray matter atrophy in early and lateonset AD

7

22

Neurobiology of Aging 2013 Aug;34(8):2014-22.

2.2

Quantitative regional validation of the visual rating scale for posterior cortical atrophy.

2.3

Relation between subcortical gray matter atrophy and conversion from mild cognitive impairment to Alzheimer’s disease

41

European Radiology 2014 Feb;24(2):397-404

57

Resubmission after major revisions Journal of Neurology, Neurosurgery & Psychiatry

Chapter 3 – Patterns of gray matter loss in different forms of dementia 3.1 More atrophy of deep gray matter structures in Frontotemporal Dementia compared to Alzheimer’s Disease

74

Journal of Alzheimers Disease 2015 Jan 1;44(2):635-47.

3.2

Automatic classification of AD and bvFTD based on cortical atrophy for single-subject diagnosis

3.3

Different patterns of cortical gray matter loss in behavioral variant FTD and AD

3.4

Joint assessment of white matter integrity, cortical and subcortical atrophy to help distinguishing AD from behavioral variant FTD: a two-center study

96

Under review

114

Under review

134

Major revisions NeuroImage:Clinical

Chapter 4 – Discussion

159

Addendum Nederlandse samenvatting Deutsche Zusammenfassung List of publications Hall of fame: List of theses Alzheimercenter VUmc Dankwoord About the author

184 190 197 199 201 205

5

It’s a terrible thing, I think, in life to wait until you’re ready. I have this feeling now that actually no one is ever ready to do anything. There is almost no such thing as ready. There is only now. And you may as well do it now. Generally speaking, now is as good a time as any. Hugh Laurie 6

Chapter 1 – General Introduction

7

Dementia Worldwide, 35.6 million people have dementia. Every year, there are 7.7 million new cases. The total number of people with dementia is projected to almost double every 20 years, to 65.7 million in 2030 and 115.4 million in 2050 [1]. Dementia is a syndrome in which there is deterioration in cognitive function beyond what might be expected for normal ageing. It affects memory, thinking, orientation, comprehension, calculation, learning capacity, language, and judgment. Impairment of cognition interferes with activities of daily living, in a pattern of progressive decline from a previous level of functioning and ultimately leads to a loss of independence. The impairment in cognitive function is commonly accompanied, and occasionally preceded, by deterioration in emotional control, social behavior, or motivation. Dementia is caused by a variety of diseases that primarily or secondarily affect the brain. Alzheimer's disease (AD) is the most common cause of dementia, other major forms include vascular dementia (VaD), dementia with Lewy bodies (DLB), and a group of diseases that contribute to frontotemporal dementia (FTD). The boundaries between different forms of dementia are indistinct and mixed forms often co-exist [2]. Dementia is overwhelming not only for the people who suffer from it, but also for their caregivers and families. There is often a lack of awareness and understanding of dementia, resulting in stigmatization and barriers to diagnosis and care. The impact of dementia on caregivers, family and societies can be physical, psychological, social and economic. There is no treatment currently available to cure dementia or to alter its progressive course. To be able to support and improve the lives of people with dementia and their caregivers and families, it is important to provide information and long-term support. Therefore, one of the principal goals for dementia care is to make a reliable diagnosis as early as possible. Which problems are clinicians facing? It has become clear that there are distinct dementia profiles with different cognitive/behavioral syndromes, reflective of different topographical distribution and types of pathology within the brain. Therefore, a comprehensive analysis of patients’ clinical history, cognition and behavior, together with a full neurological examination have been summarized into consensus clinical diagnostic criteria which ought to lead to a high degree of confidence in clinical diagnosis [3-6]. Nevertheless, underlying pathology can be predicted on clinical grounds with only limited accuracy. This is especially true for the differential diagnosis between dementias such as behavioral variant FTD (bvFTD) and AD which show overlapping clinical features, especially in the beginning of the disease, and an enormous heterogeneity within both diagnoses [7,8]. Indeed, an assessment of the currently used clinical core criteria showed that the sensitivity of discriminating between AD and bvFTD of the new criteria proved to be relatively low [9]. Consequently, bvFTD is often misdiagnosed as AD [10,11] and AD as bvFTD [12,13]. Accurate clinical diagnosis of the dementias in life is critical for proper management, assessment of prognosis, and course of treatment. For instance, whereas AD patients may be treated with acetylcholinesterase inhibitors [14] patients with FTD show differing and in some cases unfavorable responses to these drugs [15]. 8

Notwithstanding the treatment consequences, misdiagnosis may also have severe financial impacts and confound the outcome of therapeutic clinical trials. Alzheimer’s disease Alzheimer’s disease (AD) is the most common form of dementia, accounting for at least 43% of all dementia cases [1]. Although less prevalent before the age of 65 years, it is still the most frequent cause of early-onset dementia, followed by frontotemporal dementia [16]. The neuropathology of AD is characterized by extracellular amyloid deposits (amyloid plaques) and accumulation of the intracellular hyperphosphorilated tau protein (neurofibrillary tangles) [17,18]. Accumulation of these proteins causes cell death which results in a shrinkage of the total brain volume. Late-onset AD (arbitrarily defined as first symptoms after 65 years of age) is mainly characterized by episodic memory impairment at time of initial evaluation. Other types of memory, such as factual information, immediate memory, and procedural memory are relatively preserved in the initial stage, but decline when the disease progresses. Deficits in word-finding skills, visuospatial abilities and executive functioning can also occur in the initial phase, or become more prominent during the progression of the disease [3,19,20]. Early-onset AD (symptom onset before 65 years of age) has a distinct clinical profile, especially in the early disease stage: Impairments in the visuospatial-, executive-, and attention domains are commonly observed, while memory is relatively preserved in this subtype of AD [21,22]. Both subtypes of AD are accompanied with a lack of insight (anosognosia), mood disturbances, delusions, hallucinations, vegetative symptoms and aberrant motor disturbances [23]. Next to differences based on age of disease onset, there are marked phenotypic variations in AD patients [24-26]. Even in late-onset AD, in some patients memory symptoms are minimal or absent at presentation. The dominant presenting symptom may be of problems in language [24,27], visuospatial skills [28], motor abilities [29,30], or frontal and executive capacities [31]. These ‘focal’ presentations of AD are particularly problematic for clinicians because memory impairment is traditionally seen as the hallmark of AD. Indeed, non-amnestic presentations would not fulfill conventional clinical criteria [3,4] for AD, so that reliance on such criteria alone would lead to such cases being missed. The existence of ‘atypical’ variants of AD highlights the more general point: not all patients with dementia exhibit a ‘prototypical’ pattern, and diagnostic boundaries may be blurred. For example, the ‘frontal’ symptoms some AD patients exhibit, might potentially be confused with bvFTD. Behavioral variant frontotemporal dementia The behavioral variant of frontotemporal dementia (bvFTD) is the second most common cause of early onset dementia (<65 years) [32,33]. The syndrome is very heterogeneous, but is mostly characterized by a marked progressive decline in personality and/or behavior. Although the syndrome mostly has a presenile onset, occurrence after the age of 65 accounts for 20-25% of the patients [32]. However, the present rates are likely to be underestimated, since misdiagnoses such as AD or psychiatric disorders are common. The neuropathology of bvFTD can be roughly subdivided in three distinct proteinopathies. Half of the patients suffer from 9

accumulation of the tau protein. Inclusions of this hyperphosphorilated protein are present in neurons and glial cells of bvFTD patients. Another part of the patients suffers from accumulation of the transactive response DNA-binding protein 43 (TDP43 protein), and a relatively small part of patients suffers from accumulation of the fused-in-sarcoma protein (FUS protein) [34,35]. Clinically the syndrome can be classified in three behavioral symptoms: the disinhibited syndrome, the apathetic syndrome, and the stereotypic syndrome. Generally, insight is more impaired compared to AD patients, and symptoms such as loss of empathy, idiosyncratic hoarding and collecting, changing in eating behavior, poor hygiene and hyperorality are common. Furthermore, patients often show deficits in cognitive domains of executive function, attention and working memory [32,36]. Neuropsychological testing is an essential component of early detection of bvFTD. However, this can be very challenging due to the heterogeneity of the syndrome. For instance, the general presentation of patients at the time of initial evaluation can differ considerably, and part of the patients perform within the normal range of the traditional tasks [37]. Moreover, while there are differences in the performance of groups of patients with AD and bvFTD on measures of orientation, memory, language, visuomotor function and general cognitive ability, there is still considerable overlap in the performance of these two groups [38]. Thus, even when the most discriminating measures are used, it can be difficult to differentiate between AD and bvFTD. Indeed, contrary to earlier hypotheses and the currently used clinical criteria [6], episodic memory turned out to not discriminate bvFTD from AD patients [39,40], and visuospatial skills only differentiate the atypical visuospatial AD cases from bvFTD [41]. How is magnetic resonance imaging (MRI) helping us? Structural neuroimaging has dramatically changed our ability to accurately diagnose dementia. The role of neuroimaging extends beyond the detection of potentially treatable causes of dementia (e.g. tumors) to the facilitation of the diagnosis after symptom onset, and shows promise for diagnosis in early or even pre-symptomatic phases. MRI characteristics such as atrophy of the medial temporal lobe, are increasingly used as supportive evidence for a diagnosis of different forms of dementia and some of these characteristics have been incorporated into the clinical guidelines [4,6]. The presence of disproportionate gray matter (GM) atrophy in medial, basal, and lateral temporal lobe, and medial parietal cortex is supportive for a diagnosis of AD, whereas the presence of disproportionate atrophy in medial frontal, orbital–insular and anterior temporal regions help distinguish bvFTD from other conditions. In this way, structural neuroimaging can be helpful in discriminating between normal aging and different forms of dementia and assessments of global cortical atrophy and medial temporal lobe atrophy are of special diagnostic value. These MRI characteristics are detected by the use of visual rating scales [42]. They can be easily used in the daily clinical practice to discover macroscopic brain changes indicative for certain diseases [43-46]. However, many scans differ from the predicted patterns of atrophy, which combined with large between–rater variability results in low sensitivity of these scales. Furthermore, cortical atrophy patterns of AD and bvFTD largely overlap, e.g. frontal atrophy is seen in AD and hippocampal atrophy 10

does not exclude a diagnosis of bvFTD and even appears in normal aging [24,31,47,48]. Moreover, especially in the beginning of the disease, cortical atrophy may not be visible by eyeballing. More sophisticated analytical methods that could detect and quantify more subtle changes of the brain would be helpful. Ideally, methods should focus on patterns of structural changes and/or altered networks, rather than measuring only specific structures. There are different ways to study the brain with automated quantitative image post-processing techniques by measuring different parts of the brain in more detail (e.g. at voxel level). The cerebral cortex (the largest part of the brain) contains approximately 15–33 billion neurons, each connected by synapses to several thousand other neurons [49]. These neurons communicate with one another by means of long fibers called axons, which carry trains of signal pulses to distant parts of the brain or body targeting specific recipient cells. Most of the axons are wrapped in a fatty insulating sheath of myelin, which serves to greatly increase the speed of signal propagation. Myelin is white, making the inner parts of the brain filled exclusively with nerve fibers appear as lightcolored white matter (WM), in contrast to the darker-colored gray matter (GM) at the outside of the cortex, that marks areas with high densities of neuron cell bodies [50]. Relatively compact clusters of neurons at the heart of the brain also show up as gray matter, bordered by white matter or the ventricles. Because of their location and appearance, these clusters of neurons are often called subcortical or deep gray matter (DGM) structures. These deep gray matter structures are thought to be relay stations in the cortical networks throughout the brain. By examining the different parts of the brain with quantitative image analysis methods, it is possible to obtain - next to support for differential diagnosis - important information about the different pathophysiological processes in the brain of AD and bvFTD patients.

Skull White matter Cerebrospinal fluid around the brain and in the ventricles

Deep gray matter structures

Gray matter

Figure 1. Axial MRI scan of the brain, picturing the different parts of the brain.

How do we examine the different parts of the brain with image analysis methods?

11

Gray matter In dementia, neurons, the components of gray matter are dying. The consequence is GM loss and shrinkage of the brain. This GM loss, or atrophy, can be quantified and located by automated image analysis methods. Voxel-based morphometry (VBM) is one such automated technique that has grown in popularity since its introduction [51], largely because of the fact that it is relatively easy to use and has provided biologically plausible results. With this technique it is possible to study patterns of atrophy between different forms of dementia and normal aging. VBM involves a voxel-wise comparison of the local concentration of GM between two groups of subjects. The procedure involves segmenting the GM from the native space images, spatially normalizing the images from all subjects into the same stereotactic space and smoothing the GM segmentations. Voxel-wise parametric statistical tests which compare the smoothed GM images from the two groups are performed. Corrections for multiple comparisons are made using the theory of Gaussian random fields or comparable models. VBM has the advantage of being unbiased because it avoids a priori selection of regions. It is an automated technique that eliminates observer variability and analyses 3D volumes at the voxel level and thus can visualize atrophy patterns throughout the whole brain [51]. Using VBM contributes to the understanding of how the brain changes in different forms of dementia and how brain changes relate to characteristic clinical features.

Figure 2. Workflow of VBM analyses.

Next to cortical atrophy, deep gray matter (DGM) structures in the brain can provide important information about the pathological processes in the different forms of dementia. A number of pathological studies have identified DGM atrophy in AD and 12

bvFTD at autopsy [52,53] and there is emerging evidence that compared to AD, generally regarded as a cortical disease, bvFTD patients have more subcortical brain damage [47,54-56]. Their involvement in frontostriatal circuits could explain clinical symptoms which are difficult to explain by cortical damage. One of the available tools to study DGM structures is FIRST (FMRIB’s integrated registration and segmentation tool), an automated and robust method for the extraction of the subcortical volumes. Volumes of seven structures can be estimated by registration and segmentation of the structures. FIRST has been shown to give accurate and robust results for the segmentation of subcortical structures and that it performs comparable or better to other automatic methods [57-59]. In the discrimination of AD and bvFTD, deep gray matter (DGM) structures have received less attention. However, MRI-based measures of DGM structures could provide important information of the differential distribution of pathology between AD and bvFTD and may explain some of the clinical characteristics typical for these diseases.

Figure 3. Segmentation of deep gray matter structures. Colored structures: Yellow: Hippocampus; Turquoise: Amygdala; Green: Thalamus; Light blue: Caudate Nucleus; Pink: Putamen; Dark blue: Globus Pallidus; Orange: Nucleus Accumbens.

White matter In addition to GM damage, altered anatomical connectivity between white matter pathways may play a major role in dementia [60,61]. The rapid development of newer neuroimaging techniques, especially Diffusion Tensor Imaging (DTI), has enabled researchers to visualize and quantify the integrity of the WM at the microscopic level, which cannot be seen on conventional MRI methods [61,62]. DTI measures water diffusion in white matter tracts and provides details on tissue microstructure. It provides quantitative information regarding white matter tracts and visible and invisible abnormalities. DTI measures differ between dementia and control subjects and seem to correlate with clinical status [63]. When analyzing DTI data, the diffusion measures fractional anisotropy (FA) and mean diffusivity (MD) can be obtained [64]. FA is the measure of the degree of anisotropy, and ranges from 0 (the diffusion is unrestricted in all directions) to 1 (the diffusion goes in one direction and is fully restricted in all other directions). FA varies substantially by anatomical location, but is also affected by many pathological changes of the WM. Healthy WM shows a relatively high FA value, whereas damaged WM leads to less directionality and consequently, a reduction in FA. MD represents the total mean diffusion within a certain voxel and is often used as a measure of WM tissue alteration in addition to FA. 13

In addition, two tensor eigenvalues axial diffusivity and radial diffusivity – representing the water diffusivity respectively parallel and perpendicular to the axis of the fiber tract within a voxel of interest – are obtained as specific biological markers of axonal and myelin degeneration [65-67]. To analyze DTI data, Tract-based spatial statistics (TBSS), as part of FSL, can be used [68]. TBSS is a voxel-wise statistical comparison of the FA data between groups, which is a fully automated, observerindependent multi-subject analysis of whole-brain diffusion data. In TBSS analyses, a (group-wise) mean FA tract skeleton is created, which is thought to represent the centers of all WM tracts common to the group under study. By subsequently projecting each subject’s FA image onto the FA skeleton - resulting in individual skeletonized FA data - and feeding this into voxel-wise statistics, differences between groups can be calculated. Many imaging studies have used FA images in voxelwise statistical analyses, in order to localize brain changes related to development, degeneration and disease. DTI has been used to study normal aging and AD [60,64,69]. But only few DTI studies of bvFTD have been conducted. Furthermore, the direct comparison between AD and bvFTD has been examined even less. Questions like ‘Is the distribution of white matter damage different between bvFTD and Alzheimer’s disease?” or “Is WM damage secondary to the degeneration of cortical neurons or due to primary pathology occurring in the white matter regions?” remain insufficiently answered. To investigate patterns of WM damage between AD and bvFTD and if these patterns contributes to the differentiation between AD and bvFTD, as well as achieving a better understanding about white matter pathology is therefore a desirable research goal.

Figure 4. Results of a TBSS voxelwise statistics displaying areas of white matter skeleton (green) with lower FA (red-yellow) values in the frontal areas of the brain.

Combination of image analysis techniques So far, no single diagnostic biomarker with sufficient sensitivity and specificity to establish an accurate diagnosis, is available. The combination of different and new imaging techniques will help to discover patterns of alterations in connectivity and structure in the brains of patients with AD and bvFTD. Combining structural data of new imaging techniques can provide us with important new information. MRI to study disease progression over time Most information about the underlying neuropathology has been studied in crosssectional designs. Repeated MRI can aid in establishing a diagnosis when brain 14

abnormalities at baseline are insufficient to reach a conclusion and can be used to track brain changes during the course of the disease. Furthermore, rate of decline is less sensitive for inter-subject variability/ noise in the MRI data than one baseline measurement. Therefore longitudinal MRI studies are expected to enhance power to differentiate between dementias as compared to using a single scan. The ultimate goal: Single-subject diagnosis The techniques described above are used for group comparisons of different tissue characteristics. However, an ultimate aim of neuroimaging is to identify diagnoses at a single subject level. Pattern recognition techniques or automated classifiers develop and apply algorithms that recognize patterns in data and can classify new patients into different categories. Pattern Recognition for Neuroimaging Toolbox (PRoNTo) [70] is a software toolbox based on pattern recognition for the analysis of neuroimaging data. In PRoNTo, brain scans are treated as spatial patterns and statistical learning models are used to identify statistical properties of the data that can be used to discriminate between experimental conditions or groups of subjects (classification models) or to predict a continuous measure (regression models). These automated classifiers can be objective, quantitative and easy to implement and potentially satisfy the requirements of a diagnostic tool for single subjects [71].

AD bvFTD

Figure 5. Performance of support vector machine classification of two groups based on gray matter segmentations.

Aims of this thesis The general aim of this thesis was to find early markers of brain changes associated with specific types of dementia for early (differential) diagnosis and to get more insight in the biological causes of brain changes in AD and bvFTD. The following questions will be answered: 1.) How are patterns of gray matter atrophy linked to a specific diagnosis? 15

2.) Do white matter integrity measures improve the diagnostic accuracy? 3.) Are MRI derived measures of GM, DGM structures and WM integrity suitable for a diagnostic tool? Thesis outline In chapter 2 of this thesis we study how patterns of cortical and deep GM loss are related to different manifestations of AD. In chapter 2.1 we use VBM to detect patterns of GM atrophy in early- and late-onset AD patients compared to agematched controls, taking into account the potential modulating effect of APOE status. To get more insight in the diagnostic value of GM we validate the 4-point visual rating scale for posterior cortical atrophy through quantitative GM volumetry and VBM, to determine whether its use in clinical practice is justified (chapter 2.2). In chapter 2.3 we investigated volume differences of the subcortical structures between AD, MCI and controls and their predictive value for progression from MCI to AD. In chapter 3 we investigated the role of GM atrophy in AD compared to bvFTD. In chapter 3.1 we investigate if MRI-based measures of DGM structures could provide important information on the differential distribution of pathology between AD and FTD and if they explain some of the clinical characteristics typical of the diseases. Furthermore, we examine if patterns of GM atrophy can serve as a biomarker in single-subject diagnosis (chapter 3.2). In chapter 3.3 we investigated the decline of gray matter over time in different parts of the brain of patients with AD and bvFTD. In chapter 3.4 we investigate the ability of cortical and subcortical GM atrophy in combination with WM integrity to distinguish bvFTD from AD and from controls using VBM, FIRST, and TBSS.

16

Reference List [1] [2] [3]

[4]

[5]

[6]

[7] [8]

[9]

[10] [11]

[12]

[13]

[14]

[15]

[16]

[17]

World Health Organization and Alzheimer's Disease International (2012) Dementia: a public health priority. Kester MI , Scheltens P (2009) Dementia: the bare essentials. Pract Neurol 9, 241-251. McKhann G, Drachman D, Folstein M, Katzman R, Price D, Stadlan EM (1984) Clinical diagnosis of Alzheimer's disease: report of the NINCDS-ADRDA Work Group under the auspices of Department of Health and Human Services Task Force on Alzheimer's Disease. Neurology 34, 939-944. McKhann GM, Knopman DS, Chertkow H, Hyman BT, Jack CR, Jr., Kawas CH et al. (2011) The diagnosis of dementia due to Alzheimer's disease: recommendations from the National Institute on Aging-Alzheimer's Association workgroups on diagnostic guidelines for Alzheimer's disease. Alzheimers Dement 7, 263-269. Neary D, Snowden JS, Gustafson L, Passant U, Stuss D, Black S et al. (1998) Frontotemporal lobar degeneration: a consensus on clinical diagnostic criteria. Neurology 51, 1546-1554. Rascovsky K, Hodges JR, Knopman D, Mendez MF, Kramer JH, Neuhaus J et al. (2011) Sensitivity of revised diagnostic criteria for the behavioural variant of frontotemporal dementia. Brain 134, 2456-2477. Greicius MD, Geschwind MD, Miller BL (2002) Presenile dementia syndromes: an update on taxonomy and diagnosis. J Neurol Neurosurg Psychiatry 72, 691-700. Miller BL, Diehl J, Freedman M, Kertesz A, Mendez M, Rascovsky K (2003) International approaches to frontotemporal dementia diagnosis: from social cognition to neuropsychology. Ann Neurol 54 Suppl 5, S7-10. Harris JM, Thompson JC, Gall C, Richardson AM, Neary D, du PD et al. (2014) Do NIA-AA criteria distinguish Alzheimer's disease from frontotemporal dementia? Alzheimers Dement Rankin KP, Kramer JH, Miller BL (2005) Patterns of cognitive and emotional empathy in frontotemporal lobar degeneration. Cogn Behav Neurol 18, 28-36. Walker AJ, Meares S, Sachdev PS, Brodaty H (2005) The differentiation of mild frontotemporal dementia from Alzheimer's disease and healthy aging by neuropsychological tests. Int Psychogeriatr 17, 57-68. Forman MS, Farmer J, Johnson JK, Clark CM, Arnold SE, Coslett HB et al. (2006) Frontotemporal dementia: clinicopathological correlations. Ann Neurol 59, 952-962. Hodges JR, Davies RR, Xuereb JH, Casey B, Broe M, Bak TH et al. (2004) Clinicopathological correlates in frontotemporal dementia. Ann Neurol 56, 399-406. McKeith I, Del ST, Spano P, Emre M, Wesnes K, Anand R et al. (2000) Efficacy of rivastigmine in dementia with Lewy bodies: a randomised, double-blind, placebo-controlled international study. Lancet 356, 2031-2036. Mendez MF, Shapira JS, McMurtray A, Licht E (2007) Preliminary findings: behavioral worsening on donepezil in patients with frontotemporal dementia. Am J Geriatr Psychiatry 15, 84-87. Harvey RJ, Skelton-Robinson M, Rossor MN (2003) The prevalence and causes of dementia in people under the age of 65 years. J Neurol Neurosurg Psychiatry 74, 1206-1209. Braak H , Braak E (1991) Neuropathological stageing of Alzheimer-related changes. Acta Neuropathol 82, 239-259. 17

[18] [19] [20] [21]

[22]

[23]

[24]

[25] [26]

[27] [28]

[29]

[30] [31] [32] [33] [34]

[35] [36] [37]

[38]

18

Jellinger KA , Bancher C (1998) Neuropathology of Alzheimer's disease: a critical update. J Neural Transm Suppl 54, 77-95. Nestor PJ, Scheltens P, Hodges JR (2004) Advances in the early detection of Alzheimer's disease. Nat Med 10 Suppl, S34-S41. Hodges JR (2006) Alzheimer's centennial legacy: origins, landmarks and the current status of knowledge concerning cognitive aspects. Brain 129, 2811-2822. Smits LL, Pijnenburg YA, Koedam EL, van der Vlies AE, Reuling IE, Koene T et al. (2012) Early Onset Alzheimer's Disease is Associated with a Distinct Neuropsychological Profile. J Alzheimers Dis 30, 101-108. Koedam EL, Lauffer V, van der Vlies AE, van der Flier WM, Scheltens P, Pijnenburg YA (2010) Early-versus late-onset Alzheimer's disease: more than age alone. J Alzheimers Dis 19, 1401-1408. van Vliet D, de Vugt ME, Aalten P, Bakker C, Pijnenburg YA, Vernooij-Dassen MJ et al. (2012) Prevalence of neuropsychiatric symptoms in young-onset compared to late-onset Alzheimer's disease - part 1: findings of the two-year longitudinal NeedYD-study. Dement Geriatr Cogn Disord 34, 319-327. Galton CJ, Patterson K, Xuereb JH, Hodges JR (2000) Atypical and typical presentations of Alzheimer's disease: a clinical, neuropsychological, neuroimaging and pathological study of 13 cases. Brain 123 Pt 3, 484-498. Alladi S, Xuereb J, Bak T, Nestor P, Knibb J, Patterson K, Hodges JR (2007) Focal cortical presentations of Alzheimer's disease. Brain 130, 2636-2645. Snowden JS, Stopford CL, Julien CL, Thompson JC, Davidson Y, Gibbons L et al. (2007) Cognitive phenotypes in Alzheimer's disease and genetic risk. Cortex 43, 835845. Mesulam MM (2001) Primary progressive aphasia. Ann Neurol 49, 425-432. Mendez MF, Ghajarania M, Perryman KM (2002) Posterior cortical atrophy: clinical characteristics and differences compared to Alzheimer's disease. Dement Geriatr Cogn Disord 14, 33-40. Lleo A, Rey MJ, Castellvi M, Ferrer I, Blesa R (2002) Asymmetric myoclonic parietal syndrome in a patient with Alzheimer's disease mimicking corticobasal degeneration. Neurologia 17, 223-226. Soliveri P, Piacentini S, Carella F, Testa D, Ciano C, Girotti F (2003) Progressive dysarthria: definition and clinical follow-up. Neurol Sci 24, 211-212. Johnson JK, Head E, Kim R, Starr A, Cotman CW (1999) Clinical and pathological evidence for a frontal variant of Alzheimer disease. Arch Neurol 56, 1233-1239. Rabinovici GD , Miller BL (2010) Frontotemporal lobar degeneration: epidemiology, pathophysiology, diagnosis and management. CNS Drugs 24, 375-398. Ratnavalli E, Brayne C, Dawson K, Hodges JR (2002) The prevalence of frontotemporal dementia. Neurology 58, 1615-1621. Sieben A, Van LT, Engelborghs S, Martin JJ, Boon P, Cras P et al. (2012) The genetics and neuropathology of frontotemporal lobar degeneration. Acta Neuropathol 124, 353-372. Salloway S (2012) Clinical and pathological examples of Alzheimer's disease, dementia with Lewy bodies, and frontotemporal dementia. Med Health R I 95, 207-209. Neary D, Snowden J, Mann D (2005) Frontotemporal dementia. Lancet Neurol 4, 771-780. Seelaar H, Rohrer JD, Pijnenburg YA, Fox NC, van Swieten JC (2011) Clinical, genetic and pathological heterogeneity of frontotemporal dementia: a review. J Neurol Neurosurg Psychiatry 82, 476-486. Hutchinson AD , Mathias JL (2007) Neuropsychological deficits in frontotemporal dementia and Alzheimer's disease: a meta-analytic review. J Neurol Neurosurg Psychiatry 78, 917-928.

[39]

[40] [41]

[42]

[43]

[44]

[45]

[46]

[47]

[48]

[49] [50] [51] [52] [53] [54]

[55]

[56]

Bertoux M, de Souza LC, Corlier F, Lamari F, Bottlaender M, Dubois B, Sarazin M (2014) Two distinct amnesic profiles in behavioral variant frontotemporal dementia. Biol Psychiatry 75, 582-588. Hornberger M , Piguet O (2012) Episodic memory in frontotemporal dementia: a critical review. Brain 135, 678-692. Yew B, Alladi S, Shailaja M, Hodges JR, Hornberger M (2013) Lost and forgotten? Orientation versus memory in Alzheimer's disease and frontotemporal dementia. J Alzheimers Dis 33, 473-481. Scheltens P, Fox N, Barkhof F, DeCarli CD (2002) Structural magnetic resonance imaging in the practical assessment of dementia: beyond exclusion. Lancet Neurol 1, 1321. Koedam EL, Lehmann M, van der Flier WM, Scheltens P, Pijnenburg YA, Fox N et al. (2011) Visual assessment of posterior atrophy development of a MRI rating scale. Eur Radiol 21, 2618-2625. Pasquier F, Leys D, Weerts JG, Mounier-Vehier F, Barkhof F, Scheltens P (1996) Interand intraobserver reproducibility of cerebral atrophy assessment on MRI scans with hemispheric infarcts. Eur Neurol 36, 268-272. Scheltens P, Launer LJ, Barkhof F, Weinstein HC, van Gool WA (1995) Visual assessment of medial temporal lobe atrophy on magnetic resonance imaging: interobserver reliability. J Neurol 242, 557-560. Scheltens P, Leys D, Barkhof F, Huglo D, Weinstein HC, Vermersch P et al. (1992) Atrophy of medial temporal lobes on MRI in "probable" Alzheimer's disease and normal ageing: diagnostic value and neuropsychological correlates. J Neurol Neurosurg Psychiatry 55, 967-972. Rabinovici GD, Seeley WW, Kim EJ, Gorno-Tempini ML, Rascovsky K, Pagliaro TA, Allison SC et al. (2007) Distinct MRI atrophy patterns in autopsy-proven Alzheimer's disease and frontotemporal lobar degeneration. Am J Alzheimers Dis Other Demen 22, 474-488. van de Pol LA, Hensel A, van der Flier WM, Visser PJ, Pijnenburg YA, Barkhof F et al. (2006) Hippocampal atrophy on MRI in frontotemporal lobar degeneration and Alzheimer's disease. J Neurol Neurosurg Psychiatry 77, 439-442. Pelvig DP, Pakkenberg H, Stark AK, Pakkenberg B (2008) Neocortical glial cell numbers in human brains. Neurobiol Aging 29, 1754-1762. Kandel ER, Schwartz JH, Jessel T.M. (2000) Principles of neural science., New York: McGraw-Hill, Ashburner J , Friston KJ (2000) Voxel-Based Morphometry - The Methods. Neuroimage 11, 805-821. Braak H , Braak E (1991) Alzheimer's disease affects limbic nuclei of the thalamus. Acta Neuropathol 81, 261-268. Kril JJ, Macdonald V, Patel S, Png F, Halliday GM (2005) Distribution of brain atrophy in behavioral variant frontotemporal dementia. J Neurol Sci 232, 83-90. Halabi C, Halabi A, Dean DL, Wang PN, Boxer AL, Trojanowski JQ et al. (2013) Patterns of striatal degeneration in frontotemporal dementia. Alzheimer Dis Assoc Disord 27, 74-83. Looi JC, Lindberg O, Zandbelt BB, Ostberg P, Andersen C, Botes L et al. (2008) Caudate nucleus volumes in frontotemporal lobar degeneration: differential atrophy in subtypes. AJNR Am J Neuroradiol 29, 1537-1543. Looi JC, Svensson L, Lindberg O, Zandbelt BB, Ostberg P, Orndahl E et al. (2009) Putaminal volume in frontotemporal lobar degeneration and Alzheimer disease: differential volumes in dementia subtypes and controls. AJNR Am J Neuroradiol 30, 1552-1560. 19

[57] [58]

[59]

[60]

[61] [62]

[63]

[64]

[65] [66] [67]

[68]

[69]

[70]

[71]

20

Patenaude B, Smith SM, Kennedy DN, Jenkinson M (2011) A Bayesian model of shape and appearance for subcortical brain segmentation. Neuroimage 56, 907-922. Babalola KO, Patenaude B, Aljabar P, Schnabel J, Kennedy D, Crum W et al. (2009) An evaluation of four automatic methods of segmenting the subcortical structures in the brain. Neuroimage 47, 1435-1447. Morey RA, Selgrade ES, Wagner HR, Huettel SA, Wang L, McCarthy G (2010) Scanrescan reliability of subcortical brain volumes derived from automated segmentation. Hum Brain Mapp 31, 1751-1762. Stadlbauer A, Salomonowitz E, Strunk G, Hammen T, Ganslandt O (2008) Age-related degradation in the central nervous system: assessment with diffusion-tensor imaging and quantitative fiber tracking. Radiology 247, 179-188. Tartaglia MC, Rosen HJ, Miller BL (2011) Neuroimaging in dementia. Neurotherapeutics 8, 82-92. Buchel C, Raedler T, Sommer M, Sach M, Weiller C, Koch MA (2004) White matter asymmetry in the human brain: a diffusion tensor MRI study. Cereb Cortex 14, 945-951. Le Bihan D, Mangin JF, Poupon C, Clark CA, Pappata S, Molko N, Chabriat H (2001) Diffusion tensor imaging: concepts and applications. J Magn Reson Imaging 13, 534-546. Chua TC, Wen W, Slavin MJ, Sachdev PS (2008) Diffusion tensor imaging in mild cognitive impairment and Alzheimer's disease: a review. Curr Opin Neurol 21, 83-92. Beaulieu C (2002) The basis of anisotropic water diffusion in the nervous system - a technical review. NMR Biomed 15, 435-455. Alexander AL, Lee JE, Lazar M, Field AS (2007) Diffusion tensor imaging of the brain. Neurotherapeutics 4, 316-329. Hasan KM , Narayana PA (2006) Retrospective measurement of the diffusion tensor eigenvalues from diffusion anisotropy and mean diffusivity in DTI. Magn Reson Med 56, 130-137. Smith SM, Jenkinson M, Johansen-Berg H, Rueckert D, Nichols TE, Mackay CE et al. (2006) Tract-based spatial statistics: voxelwise analysis of multi-subject diffusion data. Neuroimage 31, 1487-1505. Liu Y, Spulber G, Lehtimaki KK, Kononen M, Hallikainen I, Grohn H et al. (2009) Diffusion tensor imaging and Tract-Based Spatial Statistics in Alzheimer's disease and mild cognitive impairment. Neurobiol Aging Schrouff J, Rosa MJ, Rondina JM, Marquand AF, Chu C, Ashburner J et al. (2013) PRoNTo: pattern recognition for neuroimaging toolbox. Neuroinformatics 11, 319-337. Kloppel S, Stonnington CM, Chu C, Draganski B, Scahill RI, Rohrer JD et al. (2008) Automatic classification of MR scans in Alzheimer's disease. Brain 131, 681689.

Chapter 2 - Patterns of gray matter loss in different manifestations of Alzheimer’s disease

21

Chapter 2.1

Different patterns of gray matter atrophy in early and late-onset AD 1

2, 3

1

Christiane MĂśller MSc , Hugo Vrenken PhD , Lize Jiskoot MSc , 2 2 Adriaan Versteeg , Frederik Barkhof MD PhD , Philip Scheltens MD 1 1,4 PhD , Wiesje M van der Flier PhD 2 1 Alzheimer center & Department of Neurology, Department of 3 Radiology, Departement of Physics & Medical Technology, 4 Departement of Epidemiology & Biostatistics, Neuroscience Campus Amsterdam, VU University medical center, Amsterdam, the Netherlands Neurobiology of Aging 2013 Aug;34(8):2014-22.

Abstract We assessed patterns of grey matter atrophy according to age-at-onset in a large sample of 215 AD patients and 129 controls with voxel-based morphometry using 3Tesla 3DT1 magnetic resonance imaging. Local grey matter amounts were compared between late- and early-onset AD patients and older and younger controls, taking into account the effect of APOE. Additionally, combined effects of age and diagnosis on volumes of hippocampus and precuneus were assessed. Compared to age-matched controls, late-onset AD patients exhibited atrophy of hippocampus, right temporal lobe and cerebellum, whereas early-onset AD patients showed GM atrophy in hippocampus, temporal lobes, precuneus, cingulate gyrus, and inferior frontal cortex. Direct comparisons between late- and early-onset AD revealed more pronounced atrophy of precuneus in early-onset AD patients and more severe atrophy in medial temporal lobe in late-onset AD patients. Age and diagnosis independently affected hippocampus; moreover, the interaction between age and diagnosis showed that precuneus atrophy was most prominent in early-onset AD. Our results suggest that patterns of atrophy may vary in the spectrum of AD.

Key words: Alzheimer’s disease, early-onset, late-onset, MRI, voxel-based morphometry, grey matter atrophy, EOAD, LOAD 22

Introduction The most salient characteristic of Alzheimer’s disease (AD) on MRI is atrophy of the medial temporal lobe, including the hippocampus [1]. Heterogeneity in patterns of atrophy has been suggested however, and age of disease onset may be one of the factors related to the distribution of atrophy [2-4]. The few structural imaging studies published to date on this topic have shown that grey matter (GM) atrophy in earlyonset AD seems to have a predilection for brain regions other than the medial temporal lobe, more located in the posterior and frontoparietal regions of the brain [5-11]. A former study by our group showed however that the hippocampus is similarly affected in younger and older AD patients when compared to age-matched controls [12]. The available literature on this topic lacks clarity due to small sample sizes resulting in lack of power and consequently insufficiently being able to adjust for multiple testing, the absence of direct comparisons between early- and late-onset AD, and the use of different image analysis techniques [5, 7, 9-11]. Some studies worked with a regions-of-interest approach, therefore missing patterns of atrophy in the rest of the brain [5]. Few studies used whole-brain approaches such as voxel-based morphometry (VBM) [13] which permits comparisons of GM volume at every voxel throughout the whole brain with no specific a priori hypothesis [5-7, 9]. An earlier study by our group addressed some of these problems, but still suffered from a small sample size and absence of an age-matched control group [6]. In addition to age at onset, APOE genotype has been suggested to exert regionally specific effects in the brain of AD patients [14-17]. Especially in early-onset patients, APOE genotype seems to modulate the disease [2]. In the present study we used VBM to detect patterns of GM atrophy in a large sample of early- and late-onset AD patients compared to age-matched controls, taking into account the potential modulating effect of APOE status.

Methods Patients We included 276 patients with probable AD and 140 patients with subjective complaints who visited the outpatient memory clinic of the Alzheimer Center of the VU University medical center (VUmc) between August 2008 and January 2011. All patients underwent a standardized one-day assessment including medical history, informant based history, physical and neurological examination, blood tests, neuropsychological assessment, electro-encephalography and magnetic resonance imaging (MRI) of the brain. Diagnoses of probable AD were made in a multidisciplinary consensus meeting according to the NIA-AA criteria [18]. The diagnosis of AD was confirmed by a presenilin 1 mutation in three patients and by autopsy in one patient. As controls, we used patients who were labeled as having subjective complaints (normal clinical investigations, i.e. criteria for mild cognitive impairment (MCI) not fulfilled and no major psychiatric disorder). For inclusion in the present study patients had to fulfil the following inclusion criteria: 1) availability of a T1-weighted threedimensional MRI scan (3DT1) at 3T GE MRI (details below), 2) age between 50-85 years, 3) availability of APOE genotype, and 4) availability of a Mini-mental state 23

examination (MMSE) score. Exclusion criteria were 1) poor MR image quality and/or large image artefacts, 2) failure of the image segmentation pipeline (details below), and 3) gross brain pathology other than atrophy, including severe white matter hyperintensities (WMH). Images of 215 AD patients and 129 controls were available for analysis. Both groups were categorized into a younger (< 65 years) and an older (≥65 years) group. This resulted in a study sample of 95 early-onset AD, 120 late-onset AD, 97 younger controls and 32 older controls. There were four patients who met the criteria of posterior cortical atrophy (PCA, 3 early-onset AD (3%), 1 late-onset AD (0.8%)) [19], three patients were diagnosed with logopenic variant primary progressive aphasia (lv-PPA, 1 early-onset AD (1%), 2 late-onset AD (1.6%)) [20]. The study was approved by the local medical ethics committee. All patients gave written informed consent for their clinical data to be used for research purposes. APOE and cerebrospinal fluid DNA was isolated from 10 ml blood samples in ethylenediaminetetraacetic acid (EDTA). APOE genotype was determined at the Neurological Laboratory of the Department of Clinical Chemistry of the VUmc with the LightCycler APOE mutation detection method (Roche Diagnostics GmbH, Mannheim, Germany). APOE data were available for all participants and were analyzed according to the presence or absence of an APOE ε4 allele. APOE genotype was dichotomized in ε4 carriers versus noncarriers. Cerebrospinal fluid (CSF) was obtained by lumbar puncture. Amyloid-β1−42 (Aβ42), total tau, and tau phosphorylated at threonine-181 (Ptau-181) were measured by sandwich ELISA (Innogenetics, Gent, Belgium) [21]. CSF analyses were performed at the VUmc Department of Clinical Chemistry. Cut-off levels in our lab are as follows: Aβ42< 550, total tau > 375, and ptau > 52 [21]. CSF was available for 268 subjects (early onset: n=81; late onset: n=92; young controls: n=72; old controls: n=23). MR image acquisition and review Imaging was carried out on a 3.0 Tesla scanner (SignaHDxt, GE Healthcare, London, United Kingdom) using a 8-channel head coil with foam padding to restrict head motion. The scan protocol includes a whole-brain 3DT1 fast spoiled gradient echo sequence (FSPGR; TR 708 ms, TE 7 ms, flip angle 12º, 180 sagittal slices, field of view 250 mm, slice thickness 1 mm, voxel size 0.98x0.98x1 mm) which was used for VBM. In addition, our standard MRI protocol includes 3D Fluid Attenuated Inversion Recovery (FLAIR), Dual Echo (PDT2), and Susceptibility Weighted Imaging (SWI). All scans were reviewed for brain pathology other than atrophy by an experienced radiologist. WMH were rated with the Fazekas scale [22], a 4-point rating scale, which provides an overall impression of the presence of WMH. Subjects with Fazekas scale score of 3 were excluded. Atrophy of the medial temporal lobe (MTA) was rated using a a 5point visual rating scale (0=absent – 4=severe) based on the height of the hippocampal formation and the width of the choroid fissure and the temporal horn [23]. Posterior atrophy (PA) was rated using a 4-point visual rating scale (0=absent – 3=end stage atrophy) based on the posterior cingulate- and parieto-occipital sulcus and sulci of the parietal lobes and precuneus [23, 24]. Voxel-based morphometry 24

DICOM images of the FSPGR sequence were corrected for gradient nonlinearity distortions and converted to Nifti format. The linear transformation matrix to MNI space was calculated using FSL-FLIRT [25] and used to place the image coordinate origin (0,0,0) on the anterior commissure by using the Nifti s-form. The structural 3DT1 images were then analyzed using a modified voxel-based morphometry pipeline in Statistical Parametric Mapping (SPM8; Functional Imaging Laboratory, University College London, London, UK) implemented in MATLAB 7.12 (MathWorks, Natick, MA). In the first step VBM automatically identified GM, white matter (WM) and cerebrospinal fluid (CSF) of all scans. Based on these segmentations the volumes (l) of GM, WM and CSF voxels were determined separately for each scan and summed up to calculate total intracranial volume (TIV). After this segmentation process the images were rigidly aligned. Next, a DARTEL template of the grey matter of all scans was created by nonlinearly aligning the GM images to a common space [26]. The native GM and WM segmentations were spatially normalized to the “DARTEL” template by applying the individual flow fields of all scans, using modulation to compensate for volume changes due to compression and/or expansion. Images were smoothed using a 4 mm full width at half maximum (FWHM) isotropic Gaussian kernel. Images were visually inspected at every processing step. Voxelwise statistical comparisons between groups were made to localize GM differences by means of a full factorial design which automatically models interactions between the factors. We used diagnosis (AD vs controls) and age (<65 vs ≥65 years) as factors with independent levels with unequal variance, using absolute threshold masking with a threshold of 0.2 and implicit masking. Sex and TIV were entered as covariates. To avoid the arbitrary dichotomization of age, we additionally used an ANCOVA model with diagnosis (AD vs controls) as factor with independent levels with unequal variance, using absolute threshold masking with a threshold of 0.2 and implicit masking. Age, sex and TIV were entered as covariates. An interaction with age (continuous measure) and diagnosis was modeled. To test group differences post hoc, we used separate two-sample t-tests to compare early-onset AD with young controls, and late-onset AD with old controls, adjusted for sex, age, and TIV. Additionally, we directly compared early-onset AD with late-onset AD and young controls with old controls. In these comparisons, sex, MMSE and TIV were adjusted for. In a second model, we additionally adjusted for APOE genotype. Finally, we assessed correlations between GM atrophy and MMSE by using MMSE score as covariate in the “one-sample t-test” set-up in SPM for the early- and lateonset AD group separately. The threshold for statistical significance in all VBM analyses was set to p<0.05 with family wise error correction (FWE) at the voxel level and an extent threshold of 0 voxels. For visual representation we used an uncorrected threshold of p<0.001. Volumetry of hippocampus and precuneus To visualize the associations between diagnosis and age on the one hand and changes in gray matter volume on the other hand, we extracted the volumes of the hippocampus and the precuneus using the Individual Brain Atlases as implemented in the Statistical Parametric Mapping (IBASPM) toolbox [27], a fully automated method 25

(http://www.thomaskoenig.ch/Lester/ibaspm.htm) based on the SPM software package (http://www.fil.ion.ucl.ac.uk/spm) in the following manner: Each normalized individual voxel of the native segmented GM density maps from the VBM pipeline was anatomically labeled, based on an Automatic Anatomic Labeling (AAL) atlas of predefined structures and taking into account the transformation matrix obtained in the normalization process [28]. An individual brain atlas that consisted of 84 different GM areas was created for each participant. 3 The volume (cm ) of each identified structure was calculated with the IBASPM volume statistic function. Volumes for the left and right hippocampus and left and right precuneus were summed and transferred to SPSS. Segmentation results were visually inspected for accuracy, and none was discarded. Statistical analysis Statistical analyses of clinical data were performed by means of SPSS for Windows version 15.0 (SPSS Inc., Chicago, Ill., USA). We used Student’s t-tests, Mann-Whitney U-tests and Pearson Chi-Square tests to compare groups where appropriate. To estimate the combined associations between age and diagnosis (independent variables) and the volumes of hippocampus and precuneus (dependent variables) we used linear regression analyses. If there was no significant interaction between age and diagnosis, it was left out of the model. Sex and TIV were used as covariates. In a second model, we added APOE genotype as an additional covariate. The level of significance was set at p<0.05.

Results Demographic data for all patients (n=215) and controls (n=129) are summarized in table 1. Late-onset AD patients had a lower education and were more often APOE carrier than older controls. Early-onset AD patients were older and were more often APOE carrier than younger controls. Both AD groups differed from the control groups in all CSF biomarkers. Early-onset AD patients did not differ from late-onset AD patients regarding APOE status or CSF biomarkers. Younger controls had a smaller TIV than early-onset AD patients. Late-onset AD patients had higher MTA scores than early-onset AD patients and than old controls. Old controls had higher MTA scores than the young controls and early-onset AD patients had higher MTA scores than young controls. PA scores differed between old and young controls, between lateonset AD patients and old controls and between early-onset AD patients and young controls with the first groups having the higher scores. Late-onset AD patients did not have higher PA scores than early-onset AD patients. Late-onset AD patients performed worse on the delayed recall task of the Dutch version of the Rey auditory verbal learning task (RAVLT) than early-onset AD patients but there was no difference in performance on the total immediate recall. Late-onset AD patients performed better at the Trail making test A and B (TMT A, B) than early-onset AD. The full factorial design showed main effects of diagnosis and age on patterns of GM reduction. A diagnosis of AD was associated with reduction of GM throughout the brain, especially in medial temporal lobe, hippocampus, and cingulate gyrus. Older 26

age was associated with GM reduction in medial temporal lobe and cerebellum. Furthermore, there was a significant interaction between age and diagnosis (p<0.05 FWE), implicating that early onset AD patients had more pronounced GM reduction in the precuneus than in late onset AD patients. An additional ANCOVA modeling age as a continuous factor confirmed main effects of diagnosis and age on patterns of GM reduction. A diagnosis of AD was associated with reduction of GM throughout the brain, especially in medial temporal lobe, hippocampus, and cingulate gyrus. Furthermore, there was a significant interaction between age and diagnosis (p<0.05 FWE), showing that with an earlier onset of AD, there was more pronounced GM reduction in the precuneus and right temporal gyrus whereas the later the onset of AD, the more pronounced atrophy in medial temporal lobe and cerebellum occurred (figure 1). For post hoc comparisons we stratified the groups based on age and used independent t-tests which showed that compared with older controls, late-onset AD patients showed GM reductions in hippocampus, left parahippocampal gyrus, right temporal lobe (middle temporal gyrus, superior temporal gyrus), left (inferior) parietal lobe and cerebellum (p<0.05 FWE; figure 2). The reverse contrast showed no GM reductions in old controls compared to late-onset AD patients. Compared to younger controls, early-onset AD patients showed widespread reductions in GM in hippocampus, parahippocampal gyri, temporal lobes (middle and inferior gyrus), parietal lobes (primarily precuneus and angular gyrus), posterior and anterior cingulate gyrus, and inferior frontal cortex (p<0.05 FWE; figure 2). The reverse contrast showed no GM reductions. In a direct comparison between early-onset and late-onset AD we found that late-onset AD patients had less GM in hippocampus and parahippocampal gyri, and cerebellum. By contrast, early-onset AD patients had less GM in the right precuneus despite their younger age (p<0.05 FWE). When we compared both control groups, older controls had less GM in hippocampus, parahippocampal gyrus and middle temporal gyrus (p<0.05 FWE). The reverse contrast showed no GM reductions in younger controls compared to older controls. Additional adjustment for APOE did not change the results of any of these comparisons essentially, nor were there any interactions between age and APOE. In an additional VBM analysis, we studied the correlation between regional GM reductions and dementia severity as measured by MMSE in AD patients. Lower MMSE scores were correlated with less GM in left temporal lobe (p<0.05 FWE; figure 4). Separate correlation analyses in the late- and early-onset AD groups revealed comparable patterns in both groups. Subsequently, we performed linear regression analyses to assess the combined effects of age and diagnosis on the extracted volumes of precuneus and the hippocampus. Volumes of the hippocampus were independently affected by diagnosis (β(SE)= -1.944(0.135), p<0.001) and age (β(SE)= -0.068(0.008), p<0.001). Volumes of the precuneus were also predicted by both diagnosis (β(SE)= -14.967(2.764), p<0.001) and age (β(SE)= -0.125(0.036), p=0.001). There was an interaction between diagnosis 27

and age (p<0.001; figure 5), showing that precuneus atrophy is most prominent in early-onset AD. Additional adjustment for APOE did not change these results essentially, nor were there any interactions between age and APOE.

Discussion The main findings of this study are that compared to younger controls, early-onset AD patients showed widespread GM atrophy throughout the brain (medial temporal lobe, precuneus, cingulate gyrus, frontal lobe) whereas late-onset AD patients showed a more specific pattern of GM atrophy, predominantly restricted to the medial temporal lobe and cerebellum. Direct comparisons revealed more pronounced GM atrophy in the precuneus of early-onset AD patients despite their younger age. The generalizability of the few former studies on this topic is hampered by their small sample sizes [7-10]. In the current study we were able to show widespread GM atrophy in posterior and frontoparietal cortex in early-onset AD in a large sample. In this way, we replicated an earlier, preliminary finding from our group in a completely independent sample and found that age-at-onset modulates the distribution of GM involvement [6]. In late-onset AD, we observed a more specific pattern of GM atrophy in the medial temporal lobe and cerebellum. The linear regression analyses showed that hippocampal volume is independently affected by both age and AD, resulting in especially severely affected hippocampi in late-onset AD. Nonetheless, hippocampi of early-onset patients were also smaller than those of controls of their own age, confirming earlier results from our group and others [12, 29]. There are a number of possible explanations that may underlie our finding of more widespread atrophy in early-onset AD. First, knowing that neurofibrillary tangles originate in the medial temporal lobes, later spreading throughout the cortex, a possible explanation of the more widespread atrophy in early-onset AD would be that they were in a more advanced disease stage. Disease severity as measured with MMSE was similar for both age groups however, rendering this an unlikely explanation. Second, early-onset AD patients may have more cognitive reserve, explaining the finding of more widespread atrophy in the presence of similar performance on MMSE [30-32]. If this were true, one would expect that when the burden of neurodegeneration reaches a certain level, clinical performance drops at a faster level, as has also been described for patients with a high education [33-35]. This would fit with the observation that early-onset AD patients show more rapid cognitive decline [36-38]. Third, the different patterns of GM atrophy may reflect a difference in regional vulnerability to the disease. Findings of previous reports provide evidence that the origin and spread of tau pathology originating from the transentorhinal and hippocampal area might not be the only pattern of pathological progression in AD. Spreading of AD pathology might differ between individuals and certain subtypes of AD may have proportionally greater involvement of the cortex than of the hippocampus. Age of onset could be a driving factor in this regional vulnerability as 28

other studies showed that AD patients with a hippocampal sparing subtype were younger at age-at-onset, had a shorter disease duration, and more widespread cortical involvement than the typical and limbic-predominant AD subtypes [4, 39]. This might imply that the Braak stages as described in the early nineties do not hold for all AD patients but need to be adapted for specific subgroups [40, 41]. Differences in clinical profile provide further support for this notion [42-45]. The cognitive profile in early-onset AD often includes prominent non-memory problems such as apraxia, aphasia, and visuospatial dysfunction, seemingly befitting our finding of widespread patterns of GM atrophy in the frontoparietal and posterior cortices. The clinical profile of late-onset AD is typically characterized by memory impairment, in line with their more specific pattern of GM atrophy in the medial temporal lobe. Atypical, focal, clinical syndromes like the logopenic variant of primary progressive aphasia (LPA) or posterior cortical atrophy (PCA), which could be misdiagnosed as early-onset AD, could be another explanation as these groups share largely overlapping patterns of atrophy with early-onset AD patients [46-49]. However, these cases are rare in our sample, not restricted to the early onset AD group and all are underpinned by AD pathology [50-53] rendering it unlikely to be the only cause of determined differences between early- and late-onset AD. Differences in regional vulnerability are probably explained by genetic and/or biological factors that predispose for both an earlier age-at-onset and macroscopic changes in the brain. APOE is a genetic risk factor for AD which has been associated in nondemented populations with more severe brain atrophy of regions typically hit by the disease, such as the medial temporal lobe [14, 15, 17]. It has been postulated that APOE genotype has a modulating effect on the relationship between age at onset and regional GM vulnerability [2]. In the current study, we also investigated this hypothesis. Contrary to our expectation, we found no such effect of APOE. This result is in line with other studies on this topic which also did not find an effect of APOE in combination with age-at-onset [7, 9, 54]. There may be a number of explanations for our negative finding. First, there may have been insufficient statistical power to show an effect of APOE. However, the large sample size of this study renders this explanation unlikely. Second, APOE may act at another stage of the cascade of events leading to clinical AD. In former studies of our group, we have shown effects of APOE genotype on brain activity and cognitive functioning, suggesting that APOE influences brain networks instead of cortical regions [36, 54, 55]. If this is true, than there must be other (genetic) factors that modulate patterns of atrophy. Autosomal dominant mutations could be a driving factor but are rare and their impact on patterns of atrophy has been poorly understood so far [56-58]. The effect of newly identified mutations with small effects on pattern of atrophy and age-at-onset remains to be elucidated [59, 60]. Our finding of more severe cerebellar atrophy in late onset AD patients was rather unexpected, but in line with other studies on dementia and MCI [51-63]. There are several potential explanations for this finding. First, it could be an age effect. It has been shown that in cognitively normal individuals there is GM atrophy in primary motor, sensory, and heteromodal association areas, as well as the cerebellum with 29

chronological age [29, 64, 65]. Atrophy with aging in these areas, particularly in the cerebellum, may reflect altered plasticity in efferent and afferent pathways that are involved in visual, sensory [66] and motor/mobility [67] functions in the elderly. Second, it could be speculated that there is a joint effect of age and AD as we only found cerebellum atrophy with age within the AD group (not between old and young controls). Third, Glodzik et al. found that subjects with hypertension showed significantly less GM in the cerebellum than subjects without hypertension [68]. As hypertension is a risk factor for AD and more common with older age, it is possible that hypertension in the older participants is one reason of the grey matter loss in the cerebellum as it has been found that cerebellar Purkinje cells are particularly susceptible to ischemia [69]. Elaborating on this, it has been shown that AD-type brain pathology, along with hypertension, is associated with white matter changes seen on MRI [70]. As signal differences in the cerebellum are often observed with SPM due to the registration difficulties in this region it could be possible that the results in the cerebellum reflect changes in the white matter rather than in the grey matter. The results of the linear regression analyses clearly show the different relationships of hippocampus and precuneus with age and diagnosis. In contrast to medial temporal lobe atrophy, which increases with ageing and additionally increases with disease, precuneus atrophy is most outspoken in early-onset AD and therefore we find a flat slope with age in AD patients, whereas in controls a steep slope describes the decline of the precuneus with increasing age. The scores of the parietal rating scale show the same pattern. The reason why precuneus atrophy does not seem to increase with age in AD patients is not clear. Possibly, it reflects a very AD-specific phenomenon: independent of age-at-onset, precuneus volume will reach a bottom level. This would be comparable to the pattern which is also seen in AD-specific markers such as amyloid-beta concentration in CSF or amyloid-PET [71-74]. Another potential explanation of the flat slope in AD patients could be that precuneus atrophy is relatively more pronounced in early-onset AD as the difference with their own controls is larger than in late-onset AD patients. One of the strengths of this study constituted our large sample size, due to which we had enough power to detect even subtle differences in GM atrophy in a direct comparison between early-onset AD and late-onset AD patients using FWE-correction to adjust for multiple testing. Another strength was the carefully applied VBM pipeline including visual checks of the results of each step. A possible limitation of this study is that we did not have post-mortem data available, so the possibility of misdiagnosis cannot be excluded. Nevertheless, we have an extensive standardized work-up and all patients fulfilled clinical criteria of probable AD. Furthermore, CSF biomarkers were available for the majority of patients and average biomarker levels were congruent with a diagnosis of AD in both groups, rendering the possibility of misdiagnosis less likely. Another limitation could be the fact that we used persons with subjective memory complaints as control group, as these subjects are known to have an increased risk of progression to dementia. The main question in this study however involved a comparison between early- and late-onset AD, which is not influenced by the control group. The use of IBASPM to select precuneus and hippocampus is not 30

totally independent from our VBM results as it uses the segmentations of the VBM pipeline. Nevertheless, IBASPM relies on the AAL atlas and selects bigger regions than only the significant clusters from the VBM analyses. Therefore, it adds extra support to our VBM results. For the VBM analyses we used masking with an absolute threshold of 0.2 which is considered as a relative high threshold. Because of this strict threshold, it is possible that voxels were excluded from the analysis where the brains are most vulnerable to atrophy [75] and results could be narrowed. However, the location of significant different voxels met our expectations and that of other studies. Furthermore, we repeated the full factorial analysis with an absolute threshold of 0.1. The results did not essentially change compared to those with a threshold of 0.2 and therefore support our results even with a stricter threshold. The same is applicable to the size of our smoothing kernel. Whereas a lot of studies use larger smoothing kernels (SPM8 reports an estimated smoothness of FHWM 8-9 mm) in our study a kernel of 4 mm was chosen because the increased accuracy of the DARTEL registration algorithm means that smaller kernels should be sufficient to correct for misalignment. As the kernel size increases, so does the extent of the findings [76], we believe that the small kernel size underlines the validity of our results. The findings of the current study have important implications. The diagnosis of earlyonset AD is often only made after years of delay. Early-onset patients often have a different clinical presentation, and at first sight, their scan may appear relatively normal. All too often, posterior and frontoparietal atrophy are overlooked in the daily clinical practice of memory clinics, explaining why the disease is often not appropriately diagnosed. To overcome this problem, it is important to compare patients with a reference of their own age category. Moreover, in younger patients, the posterior part of the brain – especially the precuneus – may provide the most valuable information.

Acknowledgements The gradient non-linearity correction was kindly provided by GE medical systems, Milwaukee. Research of the VUmc Alzheimer center is part of the neurodegeneration research program of the Neuroscience Campus Amsterdam. The VUmc Alzheimer center is supported by Alzheimer Nederland and Stichting VUmc fonds. The clinical database structure was developed with funding from Stichting Dioraphte.

31

Table and Figures Table 1. Demographic characteristics Alzheimer’s disease Controls Late-onset Early-onset Old Young N 120 95 32 97 ## ## # Age, years 72 ± 5 60 ± 4 71 ± 4 58 ± 4 Sex, female 55 (46%) 52 (55%) 16 (50%) 46 (47%) # Level of education* 5±1 5±1 6±2 5±1 # # 28 ± 2 MMSE 21 ± 5 20 ± 6 28 ± 2 § # # RAVLT total immediate 21 ± 7.6 23 ± 8.3 40 ± 7.6 43 ± 8.8 recall § ## # # RAVLT delayed recall 1.4 ± 2 2.4 ± 2.5 9 ± 2.8 9 ± 2.7 § ## # # TMT A 81 ± 51.8 113 ± 79.2 43 ± 14.8 42 ± 39.4 § ## ##,# # TMT B 261 ± 161.1 330 ± 219.3 114 ± 55.1 98 ± 117 # # APOE genotype, ε4 80 (66%) 61 (64%) 7 (22%) 37 (38%) carriers # # Aβ42 494.3 ± 204.5 483.2 ± 140 884.3 ± 272.2 878.3 ± 196.7 ##,# # Tau 719.7 ± 508.1 639.7 ± 446.8 386.1 ± 207.4 224.7 ± 91.1 ##,# # Ptau 99.3 ± 52.1 86.6 ± 34.9 65.6 ± 23.9 45.3 ± 14.9 # TIV (l) 2.12 ± 0.02 2.18 ± 0.02 2.09 ± 0.03 2.11 ± 0.02 ## ##, # # MTA 1.6 ± 0.8 1.2 ± 0.8 0.6 ± 0.6 0.2 ± 0.3 ##,# # PA 1.3 ± 0.8 1.3 ± 0.7 0.6 ± 0.7 0.3 ± 0.6 PCA 1 (0.8%) 3 (3%) lv-PPA 2 (1.6%) 1 (1%) *According to the Verhage system Values presented as mean ± standard deviation or n (%) # Difference between late-onset AD and old controls; or between early-onset AD and young controls with p< 0.05. ## Difference between early- and late-onset AD; or between old and young controls with p< 0.05. Groups were compared using Student’s t-tests, Mann-Whitney U-tests and Pearson ChiSquare tests where appropriate. § RAVLT: Dutch version of the Rey Auditory Verbal Learning Test (number of words), TMT: Trail making Test (seconds), TIV: Total Intracranial Volume

32

Figure 1. The ANCOVA design showed a significant interaction between age and diagnosis, implicating that early-onset AD patients had more pronounced GM reduction in the precuneus than in late-onset AD patients whereas late-onset AD patients had less GM in medial temporal lobe, hippocampus and cerebellum. Brighter colors indicate higher t-values. Figures are displayed with a threshold of p<0.001, uncorrected.

Figure 2. Compared with age-matched controls, late-onset AD patients showed grey matter reductions in hippocampus, left parahippocampal gyrus, left (inferior) parietal lobe, right temporal lobe (middle temporal gyrus, superior temporal gyrus) and cerebellum. Compared to younger controls, early-onset AD showed widespread reductions in grey matter in hippocampus, parahippocampal gyri, temporal lobes (middle and inferior gyrus), parietal lobes (primarily precuneus and angular gyrus), posterior and anterior cingulate gyrus, and inferior frontal cortex. Brighter colors indicate higher t-values. Figures are displayed with a threshold of p<0.001, uncorrected.

33

Figure 3. A: Lower MMSE scores in the late-onset AD group were correlated with less grey matter in left temporal lobe. B: Lower MMSE scores in the early-onset AD group were correlated with less grey matter in left and right temporal lobe, but only the first survived the FWE correction. Figures are displayed with a threshold of p<0.001, uncorrected.

Figure 4. Left: Scatter plot for left and right hippocampus volumes versus age for AD patients (red squares) and controls (blue circles). Lines indicate the regression lines for each group (red for AD, blue for controls). Diagnosis and increasing age independently predicted hippocampal volume. Right: Scatter plot for left and right precuneus volumes versus age for AD patients (red squares) and controls (blue circles). Lines indicate the regression lines for each group (red for AD, blue for controls). The interaction between diagnosis and age showed that precuneus atrophy is most impaired in early-onset AD.

34

Supplemental material. A: Percent difference maps of group means of late-onset AD patients and old controls. Brighter regions illustrate areas of atrophy in late-onset AD patients. The percent difference map show good agreement with the p-maps of figure 2. B: Percent difference maps of group means of early-onset AD patients and young controls. Brighter regions illustrate areas of atrophy in early-onset AD patients. The percent difference map show good agreement with the p-maps of figure 2. C: Percent difference maps of group means of early-onset AD patients and late-onset AD patients. Brighter regions illustrate areas of atrophy in late-onset AD patients, darker regions illustrate areas of atrophy in early-onset AD patients. The percent difference map show good agreement with the p-maps of figure 1. A. Late-onset AD – Old controls

B. Early-onset AD – Young controls

C. Early-onset AD – Late-onset AD

35

Reference list [1] Scheltens, P., Fox, N., Barkhof, F., & DeCarli, C. D. (2002). Structural magnetic resonance imaging in the practical assessment of dementia: beyond exclusion. Lancet Neurol., 1, 13-21. [2] van der Flier, W. M., Pijnenburg, Y. A., Fox, N. C., & Scheltens, P. (2011). Early-onset versus late-onset Alzheimer's disease: the case of the missing APOE epsilon 4 allele. Lancet Neurol., 10, 280-288. [3] Galton, C. J., Patterson, K., Xuereb, J. H., & Hodges, J. R. (2000). Atypical and typical presentations of Alzheimer's disease: a clinical, neuropsychological, neuroimaging and pathological study of 13 cases. Brain, 123 Pt 3, 484-498. [4] Whitwell, J. L., Dickson, D. W., Murray, M. E., Weigand, S. D., Tosakulwong, N., Senjem, M. L. et al. (2012a). Neuroimaging correlates of pathologically defined subtypes of Alzheimer's disease: a case-control study. Lancet Neurol., 11, 868-877. [5] Shiino, A., Watanabe, T., Kitagawa, T., Kotani, E., Takahashi, J., Morikawa, S. et al. (2008). Different atrophic patterns in early- and late-onset Alzheimer's disease and evaluation of clinical utility of a method of regional z-score analysis using voxel-based morphometry. Dement.Geriatr.Cogn Disord., 26, 175-186. [6] Karas, G., Scheltens, P., Rombouts, S., van Schijndel, R., Klein, M., Jones, B. et al. (2007). Precuneus atrophy in early-onset Alzheimer's disease: a morphometric structural MRI study. Neuroradiology, 49, 967-976. [7] Shiino, A., Watanabe, T., Maeda, K., Kotani, E., Akiguchi, I., & Matsuda, M. (2006). Four subgroups of Alzheimer's disease based on patterns of atrophy using VBM and a unique pattern for early onset disease. Neuroimage., 33, 17-26. [8] Frisoni, G. B., Pievani, M., Testa, C., Sabattoli, F., Bresciani, L., Bonetti, M. et al. (2007). The topography of grey matter involvement in early and late onset Alzheimer's disease. Brain, 130, 720-730. [9] Frisoni, G. B., Testa, C., Sabattoli, F., Beltramello, A., Soininen, H., & Laakso, M. P. (2005). Structural correlates of early and late onset Alzheimer's disease: voxel based morphometric study. J.Neurol.Neurosurg.Psychiatry, 76, 112-114. [10] Ishii, K., Kawachi, T., Sasaki, H., Kono, A. K., Fukuda, T., Kojima, Y. et al. (2005). Voxelbased morphometric comparison between early- and late-onset mild Alzheimer's disease and assessment of diagnostic performance of z score images. AJNR Am.J.Neuroradiol., 26, 333-340. [11] Canu, E., Frisoni, G. B., Agosta, F., Pievani, M., Bonetti, M., & Filippi, M. (2012). Early and late onset Alzheimer's disease patients have distinct patterns of white matter damage. Neurobiol.Aging, 33, 1023-1033. [12] van de Pol, L. A., Hensel, A., Barkhof, F., Gertz, H. J., Scheltens, P., & van der Flier, W. M. (2006). Hippocampal atrophy in Alzheimer disease: age matters. Neurology, 66, 236-238. [13] Ashburner, J. & Friston, K. J. (2000). Voxel-Based Morphometry - The Methods. Neuroimage, 11, 805-821. [14] Filippini, N., Rao, A., Wetten, S., Gibson, R. A., Borrie, M., Guzman, D. et al. (2009). Anatomically-distinct genetic associations of APOE epsilon4 allele load with regional cortical atrophy in Alzheimer's disease. Neuroimage., 44, 724-728. [15] Pievani, M., Rasser, P. E., Galluzzi, S., Benussi, L., Ghidoni, R., Sabattoli, F. et al. (2009). Mapping the effect of APOE epsilon4 on gray matter loss in Alzheimer's disease in vivo. Neuroimage., 45, 1090-1098. [16] Hashimoto, M., Yasuda, M., Tanimukai, S., Matsui, M., Hirono, N., Kazui, H. et al. (2001). Apolipoprotein E epsilon 4 and the pattern of regional brain atrophy in Alzheimer's disease. Neurology, 57, 1461-1466. 36

[17] Gutierrez-Galve, L., Lehmann, M., Hobbs, N. Z., Clarkson, M. J., Ridgway, G. R., Crutch, S. et al. (2009). Patterns of cortical thickness according to APOE genotype in Alzheimer's disease. Dement.Geriatr.Cogn Disord., 28, 476-485. [18] McKhann, G. M., Knopman, D. S., Chertkow, H., Hyman, B. T., Jack, C. R., Jr., Kawas, C. H. et al. (2011). The diagnosis of dementia due to Alzheimer's disease: recommendations from the National Institute on Aging-Alzheimer's Association workgroups on diagnostic guidelines for Alzheimer's disease. Alzheimers.Dement., 7, 263-269. [19] Crutch, S. J., Schott, J. M., Rabinovici, G. D., Boeve, B. F., Cappa, S. F., Dickerson, B. C. et al. (2012b). Shining a light on posterior cortical atrophy. Alzheimers.Dement.. [20] Gorno-Tempini, M. L., Brambati, S. M., Ginex, V., Ogar, J., Dronkers, N. F., Marcone, A. et al. (2008). The logopenic/phonological variant of primary progressive aphasia. Neurology, 71, 1227-1234. [21] Mulder, C., Verwey, N. A., van der Flier, W. M., Bouwman, F. H., Kok, A., van Elk, E. J. et al. (2010). Amyloid-beta(1-42), total tau, and phosphorylated tau as cerebrospinal fluid biomarkers for the diagnosis of Alzheimer disease. Clin.Chem., 56, 248253. [22] Fazekas, F., Chawluk, J. B., Alavi, A., Hurtig, H. I., & Zimmerman, R. A. (1987). MR signal abnormalities at 1.5 T in Alzheimer's dementia and normal aging. AJR Am.J.Roentgenol., 149, 351-356. [23] Scheltens, P., Leys, D., Barkhof, F., Huglo, D., Weinstein, H. C., Vermersch, P. et al. (1992). Atrophy of medial temporal lobes on MRI in "probable" Alzheimer's disease and normal ageing: diagnostic value and neuropsychological correlates. J.Neurol.Neurosurg.Psychiatry, 55, 967-972. [24] Koedam, E. L., Lehmann, M., van der Flier, W. M., Scheltens, P., Pijnenburg, Y. A., Fox, N. et al. (2011). Visual assessment of posterior atrophy development of a MRI rating scale. Eur.Radiol., 21, 2618-2625. [25] Jenkinson, M. & Smith, S. (2001). A global optimisation method for robust affine registration of brain images. Med.Image Anal., 5, 143-156. [26] Ashburner, J. (2007). A fast diffeomorphic image registration algorithm. Neuroimage., 38, 95-113. [27] Alemán-Gómez Y, Melie-García L, & Valdés-Hernandez P. (2012). IBASPM: toolbox for automatic parcellation of brain structures. Presented ath the 12th Annual Meeting of the Organization for Human Brain Mapping, 11-15 June 2006, Florence, Italy. Available on CD-Rom in. Neuroimage 27[1]. [28] Tzourio-Mazoyer, N., Landeau, B., Papathanassiou, D., Crivello, F., Etard, O., Delcroix, N. et al. (2002). Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject brain. Neuroimage., 15, 273-289. [29] Raji, C. A., Lopez, O. L., Kuller, L. H., Carmichael, O. T., & Becker, J. T. (2009). Age, Alzheimer disease, and brain structure. Neurology, 73, 1899-1905. [30] Marshall, G. A., Fairbanks, L. A., Tekin, S., Vinters, H. V., & Cummings, J. L. (2007). Earlyonset Alzheimer's disease is associated with greater pathologic burden. J.Geriatr.Psychiatry Neurol., 20, 29-33. [31] Roe, C. M., Mintun, M. A., D'Angelo, G., Xiong, C., Grant, E. A., & Morris, J. C. (2008). Alzheimer disease and cognitive reserve: variation of education effect with carbon 11-labeled Pittsburgh Compound B uptake. Arch.Neurol., 65, 14671471. [32] Kim, E. J., Cho, S. S., Jeong, Y., Park, K. C., Kang, S. J., Kang, E. et al. (2005). Glucose metabolism in early onset versus late onset Alzheimer's disease: an SPM analysis of 120 patients. Brain, 128, 1790-1801. 37

[33] Roselli, F., Tartaglione, B., Federico, F., Lepore, V., Defazio, G., & Livrea, P. (2009). Rate of MMSE score change in Alzheimer's disease: influence of education and vascular risk factors. Clin.Neurol.Neurosurg., 111, 327-330. [34] Stern, Y., Albert, S., Tang, M. X., & Tsai, W. Y. (1999). Rate of memory decline in AD is related to education and occupation: cognitive reserve? Neurology, 53, 19421947. [35] Bruandet, A., Richard, F., Bombois, S., Maurage, C. A., Masse, I., Amouyel, P. et al. (2008). Cognitive decline and survival in Alzheimer's disease according to education level. Dement.Geriatr.Cogn Disord., 25, 74-80. [36] van der Vlies, A. E., Koedam, E. L., Pijnenburg, Y. A., Twisk, J. W., Scheltens, P., & van der Flier, W. M. (2009). Most rapid cognitive decline in APOE epsilon4 negative Alzheimer's disease with early onset. Psychol.Med., 39, 1907-1911. [37] Greicius, M. D., Geschwind, M. D., & Miller, B. L. (2002). Presenile dementia syndromes: an update on taxonomy and diagnosis. J.Neurol.Neurosurg.Psychiatry, 72, 691700. [37] Jacobs, D., Sano, M., Marder, K., Bell, K., Bylsma, F., Lafleche, G. et al. (1994). Age at onset of Alzheimer's disease: relation to pattern of cognitive dysfunction and rate of decline. Neurology, 44, 1215-1220. [39] Murray, M. E., Graff-Radford, N. R., Ross, O. A., Petersen, R. C., Duara, R., & Dickson, D. W. (2011). Neuropathologically defined subtypes of Alzheimer's disease with distinct clinical characteristics: a retrospective study. Lancet Neurol., 10, 785796. [40] Rossor, M. N., Iversen, L. L., Reynolds, G. P., Mountjoy, C. Q., & Roth, M. (1984). Neurochemical characteristics of early and late onset types of Alzheimer's disease. Br.Med.J.(Clin.Res.Ed), 288, 961-964. [41] Braak, H. & Braak, E. (1991). Neuropathological stageing of Alzheimer-related changes. Acta Neuropathol., 82, 239-259. [42] Koedam, E. L., Lauffer, V., van der Vlies, A. E., van der Flier, W. M., Scheltens, P., & Pijnenburg, Y. A. (2010). Early-versus late-onset Alzheimer's disease: more than age alone. J.Alzheimers.Dis., 19, 1401-1408. [43] Smits, L. L., Pijnenburg, Y. A., Koedam, E. L., van der Vlies, A. E., Reuling, I. E., Koene, T. et al. (2012). Early Onset Alzheimer's Disease is Associated with a Distinct Neuropsychological Profile. J.Alzheimers.Dis., 30, 101-108. [44] Balasa, M., Gelpi, E., Antonell, A., Rey, M. J., Sanchez-Valle, R., Molinuevo, J. L. et al. (2011). Clinical features and APOE genotype of pathologically proven earlyonset Alzheimer disease. Neurology, 76, 1720-1725. [45] Koss, E., Edland, S., Fillenbaum, G., Mohs, R., Clark, C., Galasko, D. et al. (1996). Clinical and neuropsychological differences between patients with earlier and later onset of Alzheimer's disease: A CERAD analysis, Part XII. Neurology, 46, 136141. [46] Migliaccio, R., Agosta, F., Rascovsky, K., Karydas, A., Bonasera, S., Rabinovici, G. D. et al. (2009). Clinical syndromes associated with posterior atrophy: early age at onset AD spectrum. Neurology, 73, 1571-1578. [47] Whitwell, J. L., Jack, C. R., Jr., Kantarci, K., Weigand, S. D., Boeve, B. F., Knopman, D. S. et al. (2007). Imaging correlates of posterior cortical atrophy. Neurobiol.Aging, 28, 1051-1061. [48] Lehmann, M., Barnes, J., Ridgway, G. R., Ryan, N. S., Warrington, E. K., Crutch, S. J. et al. (2012). Global gray matter changes in posterior cortical atrophy: A serial imaging study. Alzheimers.Dement..

38

[49] Ridgway, G. R., Lehmann, M., Barnes, J., Rohrer, J. D., Warren, J. D., Crutch, S. J. et al. (2012). Early-onset Alzheimer disease clinical variants: multivariate analyses of cortical thickness. Neurology, 79, 80-84. [50] Zakzanis, K. K. & Boulos, M. I. (2001). Posterior cortical atrophy. Neurologist., 7, 341-349. [51] Mesulam, M., Wicklund, A., Johnson, N., Rogalski, E., Leger, G. C., Rademaker, A. et al. (2008). Alzheimer and frontotemporal pathology in subsets of primary progressive aphasia. Ann.Neurol., 63, 709-719. [52] Gorno-Tempini, M. L., Hillis, A. E., Weintraub, S., Kertesz, A., Mendez, M., Cappa, S. F. et al. (2011). Classification of primary progressive aphasia and its variants. Neurology, 76, 1006-1014. [53] Crutch, S. J., Lehmann, M., Schott, J. M., Rabinovici, G. D., Rossor, M. N., & Fox, N. C. (2012a). Posterior cortical atrophy. Lancet Neurol., 11, 170-178. [54] Ossenkoppele, R., Zwan, M. D., Tolboom, N., van Assema, D. M., Adriaanse, S. F., Kloet, R. W. et al. (2012b). Amyloid burden and metabolic function in early-onset Alzheimer's disease: parietal lobe involvement. Brain, 135, 2115-2125. [55] de Waal, H., Stam, C. J., Blankenstein, M. A., Pijnenburg, Y. A., Scheltens, P., & van der Flier, W. M. (2011). EEG abnormalities in early and late onset Alzheimer's disease: understanding heterogeneity. J.Neurol.Neurosurg.Psychiatry, 82, 6771. [56] Reiman, E. M., Quiroz, Y. T., Fleisher, A. S., Chen, K., Velez-Pardo, C., Jimenez-Del-Rio, M. et al. (2012). Brain imaging and fluid biomarker analysis in young adults at genetic risk for autosomal dominant Alzheimer's disease in the presenilin 1 E280A kindred: a case-control study. Lancet Neurol., 11, 1048-1056. [57] Whitwell, J. L., Weigand, S. D., Boeve, B. F., Senjem, M. L., Gunter, J. L., DeJesusHernandez, M. et al. (2012b). Neuroimaging signatures of frontotemporal dementia genetics: C9ORF72, tau, progranulin and sporadics. Brain, 135, 794806. [58] Ginestroni, A., Battaglini, M., Della, N. R., Moretti, M., Tessa, C., Giannelli, M. et al. (2009). Early structural changes in individuals at risk of familial Alzheimer's disease: a volumetry and magnetization transfer MR imaging study. J.Neurol., 256, 925932. [59] Guerreiro, R., Wojtas, A., Bras, J., Carrasquillo, M., Rogaeva, E., Majounie, E. et al. (2013). TREM2 variants in Alzheimer's disease. N.Engl.J.Med., 368, 117-127. [60] Potkin, S. G., Guffanti, G., Lakatos, A., Turner, J. A., Kruggel, F., Fallon, J. H. et al. (2009). Hippocampal atrophy as a quantitative trait in a genome-wide association study identifying novel susceptibility genes for Alzheimer's disease. PLoS.One., 4, e6501. [61] Spulber, G., Niskanen, E., Macdonald, S., Kivipelto, M., Padilla, D. F., Julkunen, V. et al. (2012). Evolution of global and local grey matter atrophy on serial MRI scans during the progression from MCI to AD. Curr.Alzheimer Res., 9, 516-524. [62] Dos, S., V, Thomann, P. A., Wustenberg, T., Seidl, U., Essig, M., & Schroder, J. (2011). Morphological cerebral correlates of CERAD test performance in mild cognitive impairment and Alzheimer's disease. J.Alzheimers.Dis., 23, 411-420. [63] Grau-Olivares, M., Bartres-Faz, D., Arboix, A., Soliva, J. C., Rovira, M., Targa, C. et al. (2007). Mild cognitive impairment after lacunar infarction: voxel-based morphometry and neuropsychological assessment. Cerebrovasc.Dis., 23, 353361. [64] Smith, C. D., Chebrolu, H., Wekstein, D. R., Schmitt, F. A., & Markesbery, W. R. (2007). Age and gender effects on human brain anatomy: a voxel-based morphometric study in healthy elderly. Neurobiol.Aging, 28, 1075-1087. 39

[65] Good, C. D., Johnsrude, I. S., Ashburner, J., Henson, R. N., Friston, K. J., & Frackowiak, R. S. (2001). A voxel-based morphometric study of ageing in 465 normal adult human brains. Neuroimage., 14, 21-36. [66] Mahncke, H. W., Bronstone, A., & Merzenich, M. M. (2006). Brain plasticity and functional losses in the aged: scientific bases for a novel intervention. Prog.Brain Res., 157, 81-109. [67] Rosano, C., Brach, J., Studenski, S., Longstreth, W. T., Jr., & Newman, A. B. (2007). Gait variability is associated with subclinical brain vascular abnormalities in highfunctioning older adults. Neuroepidemiology, 29, 193-200. [68] Glodzik, L., Mosconi, L., Tsui, W., de, S. S., Zinkowski, R., Pirraglia, E. et al. (2012). Alzheimer's disease markers, hypertension, and gray matter damage in normal elderly. Neurobiol.Aging, 33, 1215-1227. [69] Cervos-Navarro, J. & Diemer, N. H. (1991). Selective vulnerability in brain hypoxia. Crit Rev.Neurobiol., 6, 149-182. [70] Moghekar, A., Kraut, M., Elkins, W., Troncoso, J., Zonderman, A. B., Resnick, S. M. et al. (2012). Cerebral white matter disease is associated with Alzheimer pathology in a prospective cohort. Alzheimers.Dement., 8, S71-S77. [71] Bouwman, F. H., van der Flier, W. M., Schoonenboom, N. S., van Elk, E. J., Kok, A., Rijmen, F. et al. (2007). Longitudinal changes of CSF biomarkers in memory clinic patients. Neurology, 69, 1006-1011. [72] Kester, M. I., Scheffer, P. G., Koel-Simmelink, M. J., Twaalfhoven, H., Verwey, N. A., Veerhuis, R. et al. (2012). Serial CSF sampling in Alzheimer's disease: specific versus non-specific markers. Neurobiol.Aging, 33, 1591-1598. [73] Jack, C. R., Jr., Knopman, D. S., Jagust, W. J., Shaw, L. M., Aisen, P. S., Weiner, M. W. et al. (2010). Hypothetical model of dynamic biomarkers of the Alzheimer's pathological cascade. Lancet Neurol., 9, 119-128. [74] Ossenkoppele, R., Tolboom, N., Foster-Dingley, J. C., Adriaanse, S. F., Boellaard, R., Yaqub, M. et al. (2012a). Longitudinal imaging of Alzheimer pathology using [11C]PIB, [18F]FDDNP and [18F]FDG PET. Eur.J.Nucl.Med.Mol.Imaging, 39, 990-1000. [75] Ridgway, G. R., Omar, R., Ourselin, S., Hill, D. L., Warren, J. D., & Fox, N. C. (2009). Issues with threshold masking in voxel-based morphometry of atrophied brains. Neuroimage., 44, 99-111. [76] Henley, S. M., Ridgway, G. R., Scahill, R. I., Kloppel, S., Tabrizi, S. J., Fox, N. C. et al. (2010). Pitfalls in the use of voxel-based morphometry as a biomarker: examples from huntington disease. AJNR Am.J.Neuroradiol., 31, 711-719.

40

Chapter 2.2

Quantitative regional validation of the visual rating scale for posterior cortical atrophy. 1

1,3

2

Christiane MĂśller , Wiesje M van der Flier , Adriaan Versteeg , 1 2 1 Marije R Benedictus , Mike P Wattjes , Esther L G M Koedam , Philip 1 2 2,4 Scheltens , Frederik Barkhof , Hugo Vrenken 1 Alzheimer center & Department of Neurology , Department of 2 Radiology & Nuclear Medicine , Department of Epidemiology & 3 4 Biostatistics , Department of Physics & Medical Technology , Neuroscience Campus Amsterdam, VU University medical center, P.O. Box 7057, 1007MB Amsterdam, The Netherlands. European Radiology 2014 Feb;24(2):397-404

Abstract Objectives: Validate the four-point visual rating scale for posterior cortical atrophy (PCA) on magnetic resonance images (MRI) through quantitative grey matter (GM) volumetry and voxel-based morphometry (VBM), to justify its use in clinical practice. Methods: 229 patients with probable Alzheimer’s disease and 128 subjective memory complainers underwent 3T MRI. PCA was rated according to the visual rating scale. GM volumes of 6 posterior structures and the total posterior region were extracted using IBASPM and compared among PCA groups. To determine which anatomical regions contributed most to the visual scores we used binary logistic regression. VBM compared local GM density among groups. Results: Patients were categorized according to their PCA scores: PCA-0 (n=122), PCA-1 (n=143), PCA-2 (n=79), and PCA-3 (n=13). All structures except the posterior cingulate differed significantly among groups. Inferior parietal gyrus volume discriminated the most between rating scale levels. VBM showed that PCA-1 had lower GM volume than PCA-0 in the parietal region and other brain regions, whereas between PCA-1 and PCA-2/3 GM atrophy was mostly restricted to posterior regions. Conclusions: The visual PCA rating scale is quantitatively validated and reliably reflects GM atrophy in parietal regions, making it a valuable tool for the daily radiological assessment of dementia. Keywords: visual rating scale, magnetic resonance imaging, posterior cortical atrophy, validation, voxel-based morphometry 41

Key points: • Visual rating scale reflects grey matter atrophy in posterior brain regions. • Different PCA scores corresponded well to different quantitative degrees of atrophy. • Inferior parietal gyrus volume influenced assessment based on the visual rating scale. • This simple visual rating scale makes it useful for radiological dementia assessment. Abbreviations and acronyms: AD – Alzheimer’s disease, PCA – posterior cortical atrophy, VBM – voxel-based morphometry, MTA – medial temporal lobe atrophy, GCA – global cortical atrophy, TIV – total intracranial volume, FWE – family wise error, ROI – region of interest, WMH – white matter hyperintensities, CSF – cerebrospinal fluid, WM – white matter, GM – grey matter

42

Introduction The most salient characteristic of Alzheimer’s disease (AD) on magnetic resonance imaging (MRI) is atrophy of the medial temporal lobe, including the hippocampus [1]. However, atrophy may also occur in the posterior cortex and early posterior cortical involvement is emerging as an important aspect of AD [1-6]. Posterior cortical atrophy (PCA) is often evident from visual inspection on structural MRI. We recently proposed a visual rating scale for assessing the degree of PCA [7]. This scale was found to be robust, reproducible and easily applicable in a clinical setting [6,7]. To establish the validity of this visual PCA rating scale and thereby determine whether its use in clinical practice is justified, it should be compared against quantitative brain volumetry, as has been done previously for other visual rating scales [8-10]. This issue has partially been addressed by comparing PCA scores with the manually determined volumes of the posterior cingulate gyrus, a single structure within the posterior region [6]. The visual rating scale for PCA however covers a larger anatomical area and full validation against volumetry is lacking. Therefore, in the current study, we provide a validation covering the entire anatomical region of interest. We assessed the mutual discriminatory value of the possible scores of the visual PCA rating scale at three different levels of anatomical detail: the entire parietal brain region, the individual anatomical subregions, and in much finer detail using voxel-based morphometry (VBM). Finally, we assessed which anatomical regions contribute most to the discrimination between PCA scores.

Materials and Methods Patients The study was approved by the local institutional medical ethics committee. All patients gave written informed consent. We included 398 patients with probable AD or subjective complaints from our memory clinic-based Amsterdam Dementia Cohort. All patients visited the Alzheimer Centre or the Department of Internal Medicine (COGA) of the VU University medical centre (VUmc) between August 2008 and January 2011 and underwent a standardized one-day assessment including MRI. Diagnoses of probable AD were made in a multidisciplinary consensus meeting according to the NIA-AA criteria [11]. Patients with normal clinical investigations (i.e. not fulfilling criteria for mild cognitive impairment (MCI) [12] or any major psychiatric disorder) were labeled as having subjective memory complaints. For inclusion in the present study a three-dimensional T1-weighted sequence (3D T1) and a 3D fluid-attenuated inversion-recovery (FLAIR) sequence acquired during the same session were required. Exclusion criteria were: 1) Poor MR image quality and/or large image artefacts (n=4), 2) An asymmetric PCA visual rating score (n=34), and 3) Gross abnormalities other than atrophy, including severe white matter hyperintensities (WMH) (n=3) as defined below.

43

MRI acquisition and review Magnetic resonance imaging was performed on a 3.0-Tesla whole-body MRI system (SignaHDxt, GE Healthcare, Milwaukee, WI, United States) using an eight-channel head coil with foam padding to restrict head motion. Imaging included a whole-brain 3D T1 fast spoiled gradient echo sequence (FSPGR; TR 708 ms, TE 7 ms, flip angle 12º, 180 sagittal slices, field of view 250 mm, slice thickness 1 mm, voxel size 0.98 x0.98 x 1 mm) used for VBM and for the visual rating. A 3D FLAIR sequence (TR 8000 ms, TE 125 ms, 132 sagittal slices, field of view 250 mm, slice thickness 1.2 mm, TI = 2349 ms) was acquired for visual rating of WMH. MR Images were reviewed for brain abnormalities other than atrophy by experienced neuroradiologists (FB, 20 years’ experience; MPW, 10 years’ experience). WMH were rated on 3-mm axial reformat of the 3D FLAIR sequences (by FB and MPW) with the Fazekas scale [13], a four-point rating of the overall presence of WMH. Subjects with a maximum Fazekas scale score were excluded. Visual rating of PCA Details about rating of posterior cortical atrophy can be found in the original paper about the visual rating scale [7]. In short, PCA was rated with a 4-point scale with PCA0 = no atrophy, PCA-1 = minimal atrophy, PCA-2 = moderate atrophy, and PCA-3 = severe atrophy based on multiplanar reconstructions (sagittal, coronal, axial) of the 3D T1 and 3D FLAIR images (Fig. 1). Rating of PCA takes approximately two minutes per patient for a skilled rater. All imaging was rated by the same rater (MB, 6 months’ experience). Each tenth acquisition was rated in consensus with an experienced neuroradiologist (MPW, 10 years’ experience). When there was doubt about appropriate scoring, imaging was discussed in a consensus meeting with MPW. Both raters were blinded to the subjects’ age, sex and diagnosis. Intra-rater reliability was determined for 60 acquisitions from the dataset and was very good with weighted kappa (quadratic weights) of 0.95 for left and right scores. Inter-rater reliability between MB and MPW was also very good with weighted kappa (quadratic weights) of 0.85 for left and right scores. Patients were categorized into groups based on the PCA scores of their imaging. MRI were also rated in terms of medial temporal lobe atrophy (MTA) (according to the five-point visual rating scale) [8] and global cortical atrophy (GCA) (according to the four-point visual rating scale) [8,14]. Tissue type segmentation DICOM images of the FSPGR sequence were corrected in three dimensions for gradient non-linearity distortions and converted to the Nifti format. The linear transformation matrix to the MNI space was calculated using FSL-FLIRT [15] and used to place the image coordinate origin (0,0,0) on the anterior commissure by using the Nifti s-form. The structural 3D T1 images were analyzed using Statistical Parametric Mapping (SPM8; Functional Imaging Laboratory, University College London, London, UK) implemented in MATLAB 7.12 (MathWorks, Natick, MA, USA). First, SPM8 was used to automatically quantify probabilities of grey matter (GM), white matter (WM) and cerebrospinal fluid (CSF) for each voxel of each native image. Total intracranial volume (TIV) was calculated by summing up the native space volumes of GM, WM and CSF segmentations. 44

Volumetry of regions of interest We calculated the total GM volumes of six structures covered by the visual rating scale for PCA: bilateral posterior cingulate, postcentral gyrus, superior and inferior parietal gyrus, angular gyrus and precuneus, as follows. Using the fully automated “Individual Brain Atlases as implemented in the Statistical Parametric Mapping” (IBASPM) toolbox [16], (http://www.thomaskoenig.ch/Lester/ibaspm.htm), based on SPM software (http://www.fil.ion.ucl.ac.uk/spm), for each participant, each voxel of the native GM density maps was assigned to one of the 84 predefined cerebral GM structures in the Automatic Anatomic Labeling (AAL) atlas, by applying the inverse of the image transformation obtained in the VBM normalization process (details below) [17]. Segmentation results were visually inspected for accuracy, and none had to be 3 discarded. The GM volume (cm ) of each of the six structures covered by the PCA rating scale was estimated using the IBASPM volume statistic function. GM volumes for the left and right parts of each structure were summed and transferred to SPSS. The total GM volume of the whole parietal region of interest (ROI) was calculated as the sum of the volumes of the six structures. Statistical analysis of volumes Statistical analyses of clinical data were performed using SPSS 20.0 for Windows (IBM SPSS Statistics, Somers, NY, USA), using Student’s t-tests, Mann–Whitney U-tests and Pearson’s Chi-Squared tests to compare groups where appropriate. To test if the total GM volume of the whole parietal ROI (dependent variable) differed among the four PCA scores (independent variable), we used analysis of variance (ANOVA) with TIV as a covariate. As the raters of the degree of PCA were blinded to the subjects’ age and sex, and therefore these did not influence the rating, these two variables were not included as covariates. Bonferroni post-hoc tests were conducted. Additionally, we investigated if the GM volume of the six structures separately (dependent variables) differed among the four PCA scores (fixed factor) using a multivariate model 2 (MANOVA) with TIV as a covariate. Partial eta squared (ηp ) values were calculated to estimate the effect sizes. To assess which anatomical regions contributed most to a higher score on the visual PCA rating scale, we conducted two binary logistic regression analyses between PCA groups (dependent variable): Firstly, between PCA0 and PCA-1, and secondly, between PCA-1 and combined PCA-2/PCA-3. For these logistic regression analyses, volumes of the six structures were transformed to zscores and included as independent variables. PCA scores were coded such that odds ratios reflect the increased risk associated with higher PCA-score per standard deviation smaller volume of a specific structure. We used the forward conditional method. The level of significance was set at P<0.05. Voxel-based morphometry Voxel-based morphometry was performed using a modified pipeline in SPM8 (Functional Imaging Laboratory, University College London, London, UK). After tissue segmentation, images were rigidly aligned. Next, a “DARTEL” GM template of all acquisitions was created by non-linearly aligning the GM images of all participants to a common space [18]. DARTEL is an average group specific template to increase the 45

accuracy of inter-subject alignment. Native GM and WM segmentations were spatially normalized to the DARTEL template by applying the individual flow fields of all acquisitions, using modulation to compensate for volume changes resulting from compression and/or expansion. Images were smoothed using a 4-mm full width at half maximum (FWHM) isotropic Gaussian kernel. Images were visually inspected at every processing step. To localize GM differences, voxelwise statistical comparisons between groups were made, using a full factorial design with PCA as a factor with independent levels, unequal variance, with TIV as a covariate, an absolute threshold of 0.2, and implicit masking. The following post-hoc pairwise comparisons were made: PCA-0 was compared with PCA-1, and with PCA-2; PCA-1 was compared with PCA-2 and PCA-3 together (PCA-2/3). The statistical significance threshold was set to P<0.05 with family-wise error correction (FWE) at the voxel level.

Results Images of 357 patients (229 AD patients, 128 patients with subjective memory complaints) were available for analysis. Patients were categorized according to PCA rating scores, resulting in a study sample of 122 patients with PCA-0, 143 patients with PCA-1, 79 patients with PCA-2, and 13 patients with PCA-3. Demographic data according to PCA groups are summarized in Table 1. AD patients were predominantly classified as PCA-1 (47%), whereas most of the patients with subjective memory complaints (66%) were in the group PCA-0. MTA and GCA scores were lowest in PCA-0 and highest in PCA-3 (P<0.001). Patients with any degree of PCA had lower total brain volume (TBV) than patients categorized in PCA-0 (P<0.001). Analysis of variance (ANOVA) showed that total GM volume of the whole parietal ROI differed among the four PCA-groups (P<0.001) with the largest volumes in PCA-0 and the smallest volumes in PCA-3 (Table 2; Fig. 2). MANOVA for the six posterior cortical structures revealed a significant main effect of the PCA rating scale: All structures except the posterior cingulate differed among the four PCA-groups (posterior 2 2 cingulate gyrus: P=0.054, ηp =0.021; postcentral gyrus: P<0.001, ηp =0.30; superior 2 2 parietal gyrus: P<0.001, ηp =0.29; inferior parietal gyrus: P<0.001, ηp =0.35; angular 2 2 gyrus: P<0.001, ηp =0.25; precuneus: P<0.001, ηp =0.30). Bonferroni-corrected posthoc pairwise comparisons revealed that the volumes of these five structures (all except posterior cingulate gyrus) differed among all PCA group pairs, except between PCA-2 and PCA-3 (Fig. 3, Table 2). The small group size of the PCA-3 group limits the statistical power of comparisons with this group. To investigate which structures contributed most to the discrimination between scoring levels of the PCA rating scale, we used binary logistic regression. Comparing PCA-0 and PCA-1, logistic regression (stepwise forward) revealed that the inferior parietal gyrus (OR=3.7, 95%CI=1.5–9), the precuneus (OR=2.2, 95%CI=1.2-4.1) and the angular gyrus (OR=0.4, 95%CI=0.2–0.9) contributed independently to a higher score. Comparing PCA-1 and PCA-2/3, the inferior parietal gyrus (OR=5.1, 95%CI=2.3–11.6) 46

and the angular gyrus (OR=0.5, 95%CI=0.2–0.99) contributed the most to the higher PCA score. Odds ratios higher than 1 reflect a negative relationship between volume of the area and the PCA score, i.e. the smaller the volume of an area, the higher the odds of a higher PCA score. VBM results Voxel-based morphometry analyses showed that the visual PCA rating scale discriminates well between atrophy (PCA-1, PCA-2) and no atrophy (PCA-0) in the parietal brain regions considered by the rating scale and that with increasing parietal atrophy more atrophy of the whole brain was detected. Compared with PCA-0, the PCA-1 group had less GM in medial cingulate gyrus, middle occipital gyrus, insular cortex, hippocampus, inferior frontal gyrus, precentral gyrus, angular gyrus and middle temporal gyrus (P<0.05 FWE; Fig. 4). Compared with PCA-0, PCA-2 had less GM in the precuneus, postcentral gyrus and cerebellum (P<0.05 FWE). Compared with PCA-1, moderate to severe PCA (PCA-2/3) had less GM in the right precuneus, supplementary motor area, fusiform gyrus, and medial temporal lobe (P<0.05 FWE; Fig. 4).

Discussion We showed with volumetric analysis and VBM that the visual PCA rating scale reliably reflects GM atrophy in parietal cortical regions. There was a clear separation between brains rated as having PCA and those rated as having no atrophy. Moreover, the different severity scores in the rating scale corresponded to different quantitative degrees of atrophy. Finally, the volume of the inferior parietal gyrus in particular affected the visual PCA scoring. Previous studies showed that the presence of PCA may be a helpful additional imaging marker for AD, especially in patients with an early onset [2,19], and in distinguishing AD from fronto-temporal dementia (FTD) [6,20]. The recently developed visual rating scale for PCA detects the wide range of PCA, and is a quick, reproducible, and easily applicable tool for the clinical setting [7]. However, it has not been quantitatively validated so far, which may have hampered its clinical applicability. The current study quantitatively validated the PCA scale at three levels of anatomical detail. First, GM volumes of the entire parietal ROI covered by the scale were smaller for higher scores, and differed significantly among all pairs of PCA groups, except between PCA-2 and PCA-3, which may be explained by insufficient power due to the small size of the PCA-3 group (13 patients). Second, the same behavior was observed for five of the six individual structures. Only the volumes of the posterior cingulate gyrus did not differ among any of the PCA groups, although there was a trend, mainly driven by the subtle difference between PCA-0 and PCA-2. This limited correspondence between PCA rating scale scores and posterior cingulate gyrus volumes is in line with an earlier report where higher PCA scores only moderately corresponded with smaller volumes of the posterior cingulate gyrus [6]. The authors 47

hypothesized that PCA might reflect more than posterior cingulate atrophy alone, which is confirmed by our current findings as the other five anatomical structures did show the expected relation with PCA scores. Another explanation may be that the anatomical proximity of the posterior cingulate gyrus, especially to the retrosplenial cortex and precuneus, makes it rather difficult to assess visually, although limiting its contribution to the PCA score determined by the rater. Finally, the posterior cingulate gyrus is a very small structure, making it more prone to registration errors than larger regions, possibly increasing variability among subjects. Third, VBM analyses showed that the visual PCA rating scale discriminates well between atrophy and no atrophy in the parietal brain regions considered by the rating scale. The visual PCA rating scale also discriminates well between minimal atrophy and moderate/severe atrophy, reflecting that higher visual rating scale scores correspond to a specific pattern of atrophy in the posterior cortex. The VBM findings complement the volumetric measurements with more anatomical detail. Nevertheless, the volumetric measurements revealed differences among the minimal, moderate and severe atrophy groups separately, whereas VBM only detected significant differences in the parietal brain regions when comparing PCA-1 with PCA-2 and PCA-3 together. This may be due to differences among atrophy patterns in individual patients: atrophic regions that do not overlap precisely may go unnoticed by VBM but can be detected when quantifying GM volumes in larger ROIs. Our three-level analysis suggests that higher PCA scores are related to smaller posterior GM volumes, with probably some (limited) variation between patients as regards the exact anatomical distribution of those atrophic changes. Voxel-based morphometry analyses also showed that in this subject group, increasing PCA scores corresponded to more GM atrophy in other brain regions. This finding is driven by the composition of our study group which is a representative sample of patients attending a memory clinic, and emphasizes that isolated atrophy of the posterior cortex is an exception. The discrimination between no and mild atrophy (PCA-0 vs PCA-1), as well as the discrimination between mild and moderate/severe atrophy (PCA-1 vs PCA-2/3), was driven mainly by the inferior parietal gyrus volume, which thus seems to play a major role in the visual scoring of PCA. Inferior parietal gyrus volume was related to cognitive status and predictive of future AD development in a previous study [4]. This is in line with our findings, as the higher the score the more AD patients are categorized into the PCA groups making the inferior parietal gyrus an import subregion of the PCA spectrum. In addition to the inferior parietal gyrus, the precuneus was found to contribute to the discrimination between rating scores 0 and 1, which is not surprising, as it is one of the largest structures in the region. Finally, the angular gyrus contributed to the distinction between 0 and 1, and between 1 and 2. It should be noted that in both cases, the effect was reversed, because of collinearity. A strength of this study was the comprehensive approach with three levels of anatomical detail, including a carefully applied VBM pipeline with visual checks at 48

each step, allowing clear quantitative and visual validation of the PCA rating scale. A possible limitation of this study is that only one rater rated most of the MRI examinations. To ensure that our results can be generalized, our study protocol th included an additional rating by another rater of 10% of all acquisitions (each 10 consecutive acquisition), as well as of all acquisitions where there was doubt about the scoring. The inter-rater reliability between the two raters was very high with weighted kappa (quadratic weights) of 0.85 for left and right scores, indicating the generalizability of our results. Another possible limitation is that strictly speaking, the visual rating scale by definition also assesses the widening of the sulci in the posterior region. However, widening of the sulci is difficult to quantify reliably as the inner boundary of the skull is notoriously difficult to detect on 3D T1 images owing to the limited contrast between bone and CSF. Nevertheless, GM atrophy is supposed to underlie the widening of the sulci, which was reliably measured in this study. In order to limit the number of patient groups in our comparisons, and because most of the patients had equal PCA rating scores for their left and right hemispheres, we only included subjects with symmetric PCA rating scores. Nevertheless, there does not appear to be a plausible biological reason why the current validation of the PCA rating scale would not generalize to asymmetric scores. In spite of the large total number of patients (357), there were only 13 patients in the PCA-3 group, hampering the detection of any putative GM volume differences with the PCA-2 group because of low statistical power. The findings of the current study have important implications. As PCA has been shown to be an additional marker for AD and helps distinguish AD from FTD, it is important to use the visual rating scale in daily practice. We demonstrated that the visual rating scale for PCA reliably reflects GM atrophy in posterior regions. Because alternative approaches involving quantitative image post-processing techniques are time-consuming, require sophisticated post-imaging analysis and may be variable across pulse sequences, the simplicity of this visual rating scale has a great advantage for clinical practice, making it a useful tool in the daily radiological assessment of dementia.

Acknowledgements The gradient non-linearity correction was kindly provided by GE Medical Systems, Milwaukee, WI, USA. Christiane Möller is appointed on a grant from the national project ‘Brain and Cognition’ (“Functionele Markers voor Cognitieve Stoornissen” (# 056-13-001)). Wiesje van der Flier is recipient of the Alzheimer Nederland grant (Influence of age on the endophenotype of AD on MRI, project number 2010-002). Research of the VUmc Alzheimer Centre is part of the neurodegeneration research program of the Neuroscience Campus Amsterdam. The Alzheimer Centre VUmc is supported by Alzheimer Nederland and Stichting VUmc fonds. The clinical database structure was developed with funding from Stichting Dioraphte. All patients were also included in a recent paper on grey matter differences between early- and late-onset Alzheimer’s disease with an unrelated use of the same imaging data. 49

Tables and Figures Table 1. Demographic characteristics Score on visual rating scale for posterior cortical atrophy PCA-0 PCA-1 PCA-2 PCA-3 N 122 143 79 13 Diagnosis, AD/SMC (n)^ 37/85 108/35 72/7 12/1 Sex, female (n) 69 (57%) 68 (48%) 32 (41%) 7 (54%) # # # 67.4 ± 8.6 71.6 ± 10.1 Age, years 62.6 ± 8.1 67.7 ± 8.2 # ## MMSE 25.9 ± 4.3 23 ± 5 21.1 ± 5.5 20.5 ± 6.4 MTA score^ 0.5 ± 0.8 1.1 ± 0.8 1.4 ± 0.9 1.9 ± 1.1 GCA score^ 0.2 ± 0.4 1 ± 0.5 1.6 ± 0.5 2.2 ± 0.6* # # # TBV (l) 1.4 ± 0.1 1.2 ± 0.1 1.2 ± 0.1 1.1 ± 0.1 Values presented as mean ± standard deviation or n (%). Groups were compared using Student’s t tests, Mann Whitney U tests and Pearson’s Chi-Squared tests where appropriate ^Differences between all groups with P≤0.001 # Difference between PCA-0 vs PCA-1, PCA-0 vs PCA-2, and PCA-0 vs PCA-3 with P≤0.01 ## Difference between PCA-1 and PCA-2 with P≤0.05 *No difference between PCA-2 and PCA-3 Key: AD: patients with Alzheimer’s disease, SMC: patients with subjective memory complaints, MMSE: Mini-mental state examination, MTA: score on the visual rating scale for medial temporal lobe atrophy, GCA: score on the visual rating scale for global cortical atrophy, TBV: total brain volume

Table 2. Grey matter (GM) volume of parietal structures Score on visual rating scale for posterior cortical atrophy 3 Volume (cm ) PCA-0 PCA-1 PCA-2 PCA-3 Total posterior ROI^ 89.2 ± 12.9 77.4 ± 11.1 72.1 ± 9.2 66.3 ± 10.6 Posterior cingulate gyrus 2.7 ± 0.5 2.5 ± 0.4 2.5 ± 0.4 2.5 ± 0.5 Postcentral gyrus^ 24.4 ± 3.5 21.5 ± 3.1 20.2 ± 2.6 18.2 ± 3.7 Superior parietal gyrus^ 10.8 ± 2.1 8.9 ± 1.8 8.2 ± 1.5 7.3 ± 1.7 Inferior parietal gyrus^ 15.3 ± 2.4 13.1 ± 2.2 11.8 ± 1.8 10.6 ± 2 Angular gyrus^ 12.0 ± 1.7 10.6 ± 1.6 10.0 ± 1.5 9.5 ± 1.4 Precuneus^ 24.1 ± 3.7 20.7 ± 3.1 19.4 ± 2.7 18.3 ± 2.7 Values presented as mean ± standard deviation. (M)ANOVA with TIV as covariate were conducted between the four PCA-groups ^Differences between all PCA-groups with P≤0.01, except between PCA-2 and PCA-3

50

Figure 1. Scoring of the visual rating scale for posterior cortical atrophy (PCA). In sagittal orientation, widening of the posterior cingulate sulcus (PCS) and parieto-occipital sulcus (POS) and atrophy of the precuneus (PRE) were evaluated. In axial orientation, the widening of the posterior cingulate sulcus and sulcal dilation in the parietal lobes (PAR) were evaluated. In coronal orientation, the widening of the posterior cingulate sulcus and sulcal dilation in the parietal lobes were evaluated. Figure shows from left to right sagittal 3D T1, axial 3D FLAIR and coronal 3D T1 reconstructions with A: 0=no atrophy (38-year-old man with subjective memory complaints), B: 1=minimal atrophy (73-year-old woman with Alzheimer’s disease [AD]), C: 2=moderate atrophy (68-year-old woman with AD), D: 3=severe atrophy (62-year-old woman with AD)

51

Figure 2. Boxplot of GM volume of the whole parietal ROI of the four different PCA groups. Volumes differed significantly among all PCA groups (P<0.001)

52

Figure 3. Boxplots of GM volumes of the six different structures for each PCA-group. Volumes of all structures differed among all PCA groups except between PCA-2 and PCA-3. The volumes of the posterior cingulate did not differ among any of the PCA groups. Colored structures: Red: precuneus, Green: Superior parietal gyrus, Dark blue: postcentral gyrus, Yellow: Inferior parietal gyrus, Light blue: Angular gyrus, Copper: Posterior cingulate gyrus.

53

Figure 4. Voxel-wise comparisons between PCA-0 vs PCA-1 showed that the visual PCA rating scale discriminates well between atrophy and no atrophy (above) as well as between minimal and moderate/severe atrophy (PCA-1 vs PCA-2 and PCA-3, below). In particular, higher scores on the visual scale reflect specific atrophy of the posterior cortex. Brighter colors indicate higher t-values. Figures are displayed with a threshold of P<0.001, uncorrected.

54

Reference List [1]

[2]

[3] [4] [5]

[6]

[7]

[8]

[9]

[10]

[11]

[12]

[13]

[14]

[15] [16]

Sluimer JD, Vrenken H, Blankenstein MA, Fox NC, Scheltens P, Barkhof F et al. (2008) Whole-brain atrophy rate in Alzheimer disease: identifying fast progressors. Neurology 70, 1836-1841. Karas G, Scheltens P, Rombouts S, van Schijndel R, Klein M, Jones B et al. (2007) Precuneus atrophy in early-onset Alzheimer's disease: a morphometric structural MRI study. Neuroradiology 49, 967-976. Benson DF, Davis RJ, Snyder BD (1988) Posterior cortical atrophy. Arch Neurol 45, 789793. Jacobs HI, Van Boxtel MP, Uylings HB, Gronenschild EH, Verhey FR, Jolles J (2011) Atrophy of the parietal lobe in preclinical dementia. Brain Cogn 75, 154-163. Lehmann M, Rohrer JD, Clarkson MJ, Ridgway GR, Scahill RI, Modat M et al. (2010) Reduced cortical thickness in the posterior cingulate gyrus is characteristic of both typical and atypical Alzheimer's disease. J Alzheimers Dis 20, 587-598. Lehmann M, Koedam EL, Barnes J, Bartlett JW, Ryan NS, Pijnenburg YA et al. (2012) Posterior cerebral atrophy in the absence of medial temporal lobe atrophy in pathologically-confirmed Alzheimer's disease. Neurobiol Aging 33, 627-627. Koedam EL, Lehmann M, van der Flier WM, Scheltens P, Pijnenburg YA, Fox N et al. (2011) Visual assessment of posterior atrophy development of a MRI rating scale. Eur Radiol 21, 2618-2625. Scheltens P, Launer LJ, Barkhof F, Weinstein HC, van Gool WA (1995) Visual assessment of medial temporal lobe atrophy on magnetic resonance imaging: interobserver reliability. J Neurol 242, 557-560. Bresciani L, Rossi R, Testa C, Geroldi C, Galluzzi S, Laakso MP et al. (2005) Visual assessment of medial temporal atrophy on MR films in Alzheimer's disease: comparison with volumetry. Aging Clin Exp Res 17, 8-13. Davies RR, Scahill VL, Graham A, Williams GB, Graham KS, Hodges JR (2009) Development of an MRI rating scale for multiple brain regions: comparison with volumetrics and with voxel-based morphometry. Neuroradiology 51, 491-503. McKhann G, Drachman D, Folstein M, Katzman R, Price D, Stadlan EM (1984) Clinical diagnosis of Alzheimer's disease: report of the NINCDS-ADRDA Work Group under the auspices of Department of Health and Human Services Task Force on Alzheimer's Disease. Neurology 34, 939-944. Petersen RC, Stevens JC, Ganguli M, Tangalos EG, Cummings JL, DeKosky ST (2001) Practice parameter: early detection of dementia: mild cognitive impairment (an evidence-based review). Report of the Quality Standards Subcommittee of the American Academy of Neurology. Neurology 56, 1133-1142. Fazekas F, Chawluk JB, Alavi A, Hurtig HI, Zimmerman RA (1987) MR signal abnormalities at 1.5 T in Alzheimer's dementia and normal aging. AJR Am J Roentgenol 149, 351-356. Pasquier F, Leys D, Weerts JG, Mounier-Vehier F, Barkhof F, Scheltens P (1996) Interand intraobserver reproducibility of cerebral atrophy assessment on MRI scans with hemispheric infarcts. Eur Neurol 36, 268-272. Jenkinson M , Smith S (2001) A global optimisation method for robust affine registration of brain images. Med Image Anal 5, 143-156. Alemán-Gómez Y, Melie-García L, Valdés-Hernandez P (2006) IBASPM: toolbox for automatic parcellation of brain structures. Presented ath the 12th Annual Meeting of the Organization for Human Brain Mapping, 11-15 June 2006, Florence, Italy. Available on CD-Rom in. Neuroimage 27. 55

[17]

[18] [19]

[20]

56

Tzourio-Mazoyer N, Landeau B, Papathanassiou D, Crivello F, Etard O, Delcroix N, Mazoyer B, Joliot M (2002) Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI singlesubject brain. Neuroimage 15, 273-289. Ashburner J (2007) A fast diffeomorphic image registration algorithm. Neuroimage 38, 95-113. Frisoni GB, Pievani M, Testa C, Sabattoli F, Bresciani L, Bonetti M, Beltramello A, Hayashi KM, Toga AW, Thompson PM (2007) The topography of grey matter involvement in early and late onset Alzheimer's disease. Brain 130, 720-730. Barnes J, Godbolt AK, Frost C, Boyes RG, Jones BF, Scahill RI, Rossor MN, Fox NC (2007) Atrophy rates of the cingulate gyrus and hippocampus in AD and FTLD. Neurobiol Aging 28, 20-28.

Chapter 2.3

Relation between subcortical gray matter atrophy and conversion from mild cognitive impairment to Alzheimer’s disease 1,6 1 1,5 Hyon-Ah Yi , Christiane Möller , Nikki Dieleman , Femke H. 1 3 1 Bouwman , Frederik Barkhof , Philip Scheltens , Wiesje M van der 1,2 3, 4 Flier , Hugo Vrenken 1 2 Alzheimer center & Department of Neurology, Department of 3 Epidemiology & Biostatistics, Department of Radiology & Nuclear 4 Medicine, Department of Physics & Medical Technology, Neuroscience Campus Amsterdam, VU University medical center, P.O. Box 7057, 5 1007MB Amsterdam, the Netherlands, Department of Radiology, University Medical Center Utrecht, P.O. Box 85500, 3508 GA Utrecht, 6 the Netherlands. Department of Neurology, Keimyung University School of medicine, Daegu, South Korea. Journal of Neurology, Neurosurgery & Psychiatry (resubmission after major revisions)

Abstract Objective: To investigate whether subcortical gray matter atrophy predicts progression from mild cognitive impairment (MCI) to Alzheimer’s disease (AD), and to compare subcortical volumes between AD, MCI and controls. To assess the correlation between subcortical gray matter volumes and severity of cognitive impairment. Methods: We included 773 participants with 3D T1-weighted MRI at 3-Tesla, who were 181 control, who had subjective memory complaints with normal cognition, 201 MCI and 391 AD. During follow-up (2.0±0.9 years), 35 MCIs converted to AD (progressive MCI) and 160 MCIs remained stable (stable MCI). We segmented volumes of six subcortical structures of amygdala, thalamus, caudate nucleus, putamen, globus pallidus and nucleus accumbens, and of hippocampus, using FIRST. Results: ANOVAs, adjusted for sex and age, showed that all structures, except globus pallidus, were smaller in AD than in controls. In addition, amygdala, thalamus, putamen, nucleus accumbens, and hippocampus were smaller in MCI than controls. Across groups, all subcortical gray matter volumes, except globus pallidus, showed positive correlation with cognitive function, as measured by MMSE (0.16<r<0.28, all p<0.05). Cox proportional hazards analyses adjusted for age, sex, education, CAMCOG-R and MMSE showed that smaller volumes of hippocampus and nucleus accumbens were associated with increased risk of progression from MCI to AD (hazards ratio [95% CI] 1.60 [1.15-2.21]; 1.60 [1.09-2.35], p<0.05) Conclusions: In addition to hippocampus, also nucleus accumbens volume loss was associated with increased risk of progression from MCI to AD. Furthermore, volume loss of subcortical gray matter structures was associated with severity of cognitive impairment. Keywords: subcortical atrophy, mild cognitive impairment, Alzheimer’s disease, MRI 57

Introduction Mild cognitive impairment (MCI) is a condition of cognitive decline without significant social or occupational impairment [1]. Patients with MCI have an increased risk of dementia, mostly due to Alzheimer’s disease (AD) and for many patients, MCI represents a predementia stage of AD [1-3]. Early identification of MCI patients who are at risk of progression to AD is important, especially in the context of treatment trials. Prior neuroimaging studies have suggested that hippocampal atrophy, FDG-PET pattern and positive amyloid imaging predicted progression from MCI to AD [2]. These studies have focused on cortical changes, as the neuropathology underlying AD (i.e. senile plaques and neurofibrillary tangles) has a predominantly cortical distribution.[4,5] Senile plaques and neurofibrillary tangles, however, have also been observed to extend into the subcortical structures including thalamus, putamen and amygdala [4,6,7]. Pittsburgh compound B (PiB)-PET studies supported these findings by showing increased amyloid retention in the thalamus or basal ganglia in sporadic AD as well as in presenillin mutation carriers as preclinical AD group [8,9]. In line with these observations, recent in vivo imaging studies showed structural subcortical involvement in AD. Several studies demonstrated subcortical atrophy or shape differences between AD and control, of putamen and thalamus [10], amygdala and thalamus [11], and of the striatum including nucleus accumbens [12]. Also there were few studies of subcortical change between MCI and control. Liu et al found that the baseline volumes of hippocampus, amygdala and nucleus accumbens were reduced in MCI compared to controls [13]. In another cross-sectional study, there were no difference between subcortical volumes of MCI and controls, such as thalamus, caudate nucleus and amygdala [14]. Only a few, small studies have evaluated the value of subcortical atrophy for predicting progression from MCI to AD. Tang et al[15] reported that atrophy in hippocampus and amygdala, and lateral ventricular expansion could discriminate MCI progressing to AD from stable MCI. On the contrary, Liu et al [13] found that the volume of caudate nucleus and amygdala rather than hippocampus were independent predictors of progression from MCI to AD. In the current study, we used cross-sectional imaging and longitudinal clinical followup, aiming, first, to compare volumes of six subcortical structures cross-sectionally, namely thalamus, caudate nucleus, putamen, globus pallidus, amygdala and nucleus accumbens as well as hippocampus, between AD, MCI and controls. Second, we aimed to analyze correlation between subcortical volumes and severity of cognitive impairment. Finally, we aimed to investigate the predictive value of subcortical volumes for progression from MCI to AD with longitudinal clinical follow up. Based on the prior works [13,15] we hypothesized that the volumes of amygdala, caudate nucleus or hippocampus could predict progression from MCI to AD.

Methods Participants This was a retrospective study. From the Amsterdam Dementia Cohort [16], consisting of all patients who visited the memory clinic for evaluation of their 58

cognitive complaints, we considered for inclusion all patients who visited between January 2008 and December 2011. Diagnostic work-up included clinical assessment of medical history, neurological examination, laboratory tests, neuropsychological tests and brain MRI. Diagnoses were made by consensus in a multidisciplinary meeting. The diagnosis of probable AD was based on the criteria of the National Institute of Neurological and Communicative Diseases and Stroke and Alzheimer Disease and Related Disorders Association (NINCDS-ADRDA) [17]. Patients with MCI fulfilled the criteria defined by Petersen [1]. When all clinical investigations were normal (i.e. criteria for MCI or any psychiatric disorder not met), patients were considered to have subjective complaints. Patients visited the clinic annually or according to clinical needs, and their diagnosis was re-evaluated with clinical course and neuropsychological tests including Mini Mental State Examination (MMSE) [18] and the Cambridge Cognitive Examination-Revised (CAMCOG-R) [19]. The CAMCOG-R allows the assessment of eight cognitive domains, including orientation, comprehension, expression, memory, attention and calculation, praxis, abstract reasoning, and perception, with a maximum score of 104. We followed up all participants for 2.0±0.9 years and reviewed their clinical records. Patients were scheduled for a new visit annually or according to clinical needs, which explains the variability in follow-up times in our samples. Not all tests are always performed at every visit. We classified MCI patients as progressive MCI (p-MCI) if they were diagnosed as having a form of dementia at a follow-up clinical examination, and as stable MCI (sMCI) otherwise. Patients with subjective complaints who showed normal cognition through follow-up duration served as controls in this study. We excluded patients if any of the following criteria were met: age <55 or >90, other medical or neurological conditions to affect cognition such as vascular dementia or dementia with Lewy bodies, history of psychiatric episodes or substances abuse, history of being diagnosed as dementia, abnormal brain scans (details given below under “Imaging acquisition and analysis”), or failure of FIRST image analysis algorithm [20]. This study was conducted in accordance with regional research regulations and conformed to the Declaration of Helsinki. This study was approved by medical ethics committee of the VU University medical center, Amsterdam, and written informed consent for their clinical data was obtained from all patients for research purposes. Imaging acquisition and analysis Brain MRI was performed at the initial visit, using a 3.0 Tesla scanner (Signa, HDxt, GE Healthcare, Milwaukee, WI, USA) with an 8-channel head coil. For measurement of the subcortical volumes, a three-dimensional, T1-weighted-fast spoiled gradient echo sequence was obtained with acquisition parameters as follows: repetition time/echo time/inversion time 708/7/450 ms; flip angle, 12˚; matrix size, 256 × 256; field of view, 250mm; 180 sagittal slices; slice thickness, 1mm; voxel size, 0.98 ×0.98×1 mm. Routine T2-weighted, and fast fluid-attenuated inversion recovery (FLAIR) and susceptibility weighted images were also obtained to exclude structural lesions that may affect cognitive function such as mass or vascular lesions of large territorial or strategic infarcts, and severe white matter hyperintensities of Fazekas grade 3 [21]. 59

All the images were processed and analyzed automatically with the tools of the FSL software package (FMRIB Software Library, http://www.fmrib.ox.ac.uk/fsl/fslwiki/). Using the algorithm FIRST (FMRIBs integrated registration and segmentation tool) [20], we segmented bilateral amygdala, thalamus, hippocampus, globus pallidus, nucleus accumbens, caudate nucleus and putamen. Left and right volumes of the same structure were summed. Examples of subcortical segmentation using FIRST are presented in Figure 1. Total brain volume, automatically calculated using SIENAX (Structural Image Evaluation using Normalization of Atrophy Cross-sectional) [22,23], was normalized for head size via volumetric scaling factor (VSF), which was calculated by registering the brain image to MNI152 (Montreal Neurological Institute, Montreal, Canada) space. Similarly, all structural volumes obtained from FIRST were normalized for head size by multiplying by the VSF. Statistical analysis We performed statistical analysis using IBM SPSS statistics for Windows, version 20 (IBM Corp). We compared groups using χ2 test, t-test and one-way analysis of variance (ANOVA) followed by Bonferroni’s post hoc test where appropriate. To evaluate the differences of volumes in each subcortical structure among diagnostic groups, we used ANOVA with post hoc Bonferroni t-test corrected for age and sex. A Pearson’s bivariate correlation analysis and partial correlation corrected for age and sex were performed to examine the correlation of cognitive function and subcortical structural volume. To account for the fact that CAMCOG and MMSE measure similar concepts, we set the threshold for statistical significance of these correlations at p=0.025. We performed Cox proportional hazards model to evaluate the predictive value of subcortical atrophy for progression from MCI to AD. Hazard ratios (HR) with 95% confidence interval (CI) are presented. We normalized volumes of all subcortical structures to z-scores and multiplied this by -1. The resulting HR’s can be interpreted as increased risk of progression to AD for every standard deviation smaller volume. In model 1, unadjusted hazard ratios are presented. In model 2, age, sex and education are corrected for and in model 3, CAMCOG-R, baseline MMSE are additionally included. In addition, to minimize the influence of global atrophy, we controlled for brain volume also. Finally, to assess the combined predictive value of the best combination of markers, we performed a forward stepwise Cox regression analysis including all baseline DGM volumes as possible predictors. Because the diagnosis (MCI or AD) at follow-up is a dichotomous variable and it may in some cases have been influenced by the baseline imaging, we also performed an exploratory analysis investigating the correlations between baseline volumes and follow-up MMSE scores. We assessed bivariate Pearson’s correlations, as well as partial correlations adjusted for age, sex, education level and follow-up duration.

Results Demographic characteristics of the participants Out of the initial 1,012 individuals included in the dataset, 239 patients were excluded for the following reasons: 72 patients had lacunes in the basal ganglia, in 72 patients 60

the FIRST image analysis software failed, 5 patients among the patients who were initially diagnosed as AD were diagnosed with other types of dementia at follow-up visit, 76 controls were below age 55, and 14 controls had deteriorated cognition at follow up. Finally 773 subjects were analyzed; 181 controls (65±8 years; F/M 82/99), 201 MCI (70±9 years; F/M 82/119) and 391 AD (69±9 years; F/M 221/170). For the comparison between s-MCI and p-MCI, 6 MCI patients were additionally excluded, as they progressed to other types of dementia rather than AD. Of the remaining 195 MCI patients, 35 progressed to AD, while 160 remained stable. Demographic and clinical characteristics by group are presented in Table 1. There were significant differences of age, gender, education and cognitive scores between groups (p<0.05). Post hoc comparisons showed that AD and MCI were older and less educated, and had longer duration of follow-up than controls. There were no differences of age, gender, or follow-up duration between AD and MCI. Subcortical volume Volumes of all structures in each clinical group are presented by group in Table 1 and Figure 2. ANOVAs, adjusted for age, sex and education, showed that all subcortical gray matter volumes except globus pallidus differed between groups (p<0.05) and showed a tendency to decrease from controls to MCI and further to AD. Post hoc tests showed that all structures were smaller in AD than in controls, except globus pallidus, and were smaller in MCI than controls except globus pallidus and caudate nucleus. The volumes of hippocampus, thalamus and caudate nucleus were smaller in AD compared to MCI. Correlation between cognitive function and subcortical volumes Pearson’s correlation across groups showed positive correlations of all subcortical structural volumes besides globus pallidus with cognitive function, measured by MMSE (0.13 <r<0.26, all p=0.000) and CAMCOG (0.12<r<0.30, all p<0.025) (Figure 3). Results remained essentially unchanged when we used partial correlations corrected for age, sex and education. In analyses within MCI, thalamus and amygdala volumes were positively correlated with MMSE (respectively r=0.19 and 0.23, p=0.008 and 0.001), and amygdala volume was positively correlated with CAMCOG (r=0.178, p=0.014) although the correlation coefficients were low. On controlling for age, sex and education, there were correlations between amygdala volume and MMSE (r=0.18, p=0.008), and between caudate nucleus volume and CAMCOG (respectively r=0.17, p=0.024) Prediction of progression from MCI to AD The comparison between 160 s-MCI (70±9 years; F/M 64/96) and 35 p-MCI (70±10 years, F/M 18/17) did not show any difference in age, gender, education, baseline cognitive scores, and follow-up duration (Table 2). Adjusted for age, sex and education, atrophy of hippocampus and amygdala was more extensive in p-MCI than in s-MCI (hippocampus p<0.001; amygdala p=0.045), and most of the structures were smaller in p-MCI than in s-MCI. Finally we used Cox proportional hazards analysis to evaluate the predictive value of baseline volume of each subcortical structure for progression to AD in MCI patients. Table 3 shows crude and adjusted progression risk 61

of baseline subcortical volume. In the crude model, smaller volume of hippocampus and nucleus accumbens conferred a higher risk of progression from MCI to AD (HR [95% CI]: hippocampus 1.56 (1.16-2.11), nucleus accumbens 1.59 (1.16-2.18), for both p<0.05). Upon adjustment for age, sex, education, CAMCOG-R and MMSE, these results remained significant (p<0.05). Controlling for normalized brain volume showed similar results (HR (95% CI) 1.55 (1.11-2.16); 1.54 (1.03-2.30) p<0.05, for hippocampus and nucleus accumbens). When using a forward stepwise model to select the best predictors, only hippocampus remained significant (HR (95% CI) 1.59 (1.15-2.21)). Correlation of follow-up MMSE scores and baseline volumes MMSE scores at the follow-up visit were available for 116 of these MCI patients. The bivariate Pearson’s correlations between baseline volumes and follow-up MMSE scores revealed no significant correlations. When adjusting for age, sex and educational level, there was a trend for a weak correlation between MMSE score at follow-up and hippocampus volume at baseline (r=0.182, p=0.054), and no significant correlations.

Discussion The main finding of this study is that in addition to baseline hippocampal volume, also baseline nucleus accumbens volume predicted progression from MCI to AD during 2 years of clinical follow-up. However, when including all baseline volumes as candidate predictors, forward stepwise Cox regression revealed that nucleus accumbens volume had no added independent predictive value over and above that of hippocampal volume. Baseline volumes of other subcortical structures were not predictive of progression to AD, despite the observation of decreasing volumes from controls to MCI to AD, and significant associations with cognitions for all structures except globus pallidus. As early identification of MCI patients at risk for progression to probable AD continues to be an important goal both for clinical treatment trials and ultimately for individual clinical treatment, it is important to identify in vivo imaging measures that can aid this identification. Specifically, baseline subcortical volumes would be easily accessible and suitable measures, because they require just a single MRI and can be determined with acceptable reliability using automated methods [20,24]. We hypothesized that subcortical atrophy contributes to AD pathogenesis as in the previous pathologic and neuroimaging reports [4,10,25] and that it could predict progression of MCI to AD. As expected, we observed that hippocampal volume in MCI predicted subsequent progression to AD. In addition, we observed that nucleus accumbens volume also predicted subsequent progression to AD. Despite the volume difference between AD, MCI and controls, other subcortical gray matter volumes did not predict progression from MCI to AD. There have been few former reports to evaluate subcortical structural volume differences as predictors of progression from MCI to AD. Liu et al demonstrated that baseline amygdala and caudate volume had 69% accuracy in predicting AD in MCI during 1 year follow-up [13]. Although there was a report that emphasized the unique role of hippocampal atrophy as a marker of progression to AD from MCI, they adopted hippocampus and amygdala only using region of interest 62

volumetry and this could not present supportive evidence of subcortical role in progression to AD [26]. Some prior studies showed that subcortical volume decreased as cognitive function became worse in AD [11,27], but they did not investigate the relation with progression of MCI. In the current study, we investigated the role of baseline subcortical volume in progression of MCI during an average of 2-year follow up. We observed significant difference in baseline hippocampus and amygdala volume between s-MCI and p-MCI, but Cox proportional hazard analysis showed that only hippocampus and nucleus accumbens volume predicted clinical progression. Since MMSE scores may be more sensitive to change, and imaging can be used in the diagnostic process, MMSE is a potentially more objective and sensitive outcome measure than diagnostic status. However, our exploratory additional analysis revealed no correlations of follow-up MMSE scores with baseline volumes, except for a trend for hippocampal volume. In comparison between AD and MCI, the volumes of a part of basal ganglia and thalamus were significantly lower in AD than in MCI, but none of these predicted progression from MCI to AD. A previous study also found pathologic striatal involvement in AD [7] and that the nucleus accumbens as a part of ventral striatum as well as hippocampal atrophy is a significant indicator for cognitive decline [28]. The lack of clinical predictive value in MCI patients appears not to be due to any lack of sensitivity of our method, or any absence of atrophy in these structures in our patients. In fact, there were significant subcortical volume differences between controls, MCI and AD. Most of the prior imaging analyses in AD or MCI have paid attention to cortical change rather than subcortical structures and more vulnerable regions in AD such as hippocampus or entorhinal cortex [29-31]. The hypothesis of the subcortical involvement in AD has been based on pathological studies in AD in addition to recent imaging studies. Braak and Braak [4] documented that there were subcortical depositions of neurofibrillary tangles in thalamus and amygdala, in the transentorhinal stages of AD. The subcortical involvement could be a late phenomenon of AD as multiple studies have reported that the symptoms related to frontal-subcortical circuit disruption generally appear in the later stages [32-34]. However, a few studies showed functional and structural involvement of the subcortical regions already in the early stages of AD [10,11,15]. These studies have shown inconsistent results, in which some studies found atrophy of all the subcortical structures, while some observed decreased volumes of specific regions only besides hippocampus, for example, thalamus and caudate nucleus involvement, or putamen and globus pallidus [11, 15]. Most studies focused on the volume difference between AD and healthy control and a few studies included MCI in the comparisons. One of them showed no significant baseline volume difference between AD and MCI [26]. In our large study, however, all the baseline subcortical structures except globus pallidus were significantly atrophied in AD compared to MCI and controls, and also smaller in MCI than controls, which is compatible with some previous studies including a report of pathological topographic distribution in AD [26]. This confirmed subcortical involvement even in an early stage of AD-like disease. Many researchers have tried to find predictive factors of progression of AD with increasing importance of the neuroimaging biomarker for early detection of condition with AD pathology. Most trials have adopted hippocampal volume or cortical 63

thickness as a structural biomarker [35,36]. In additional study to detect progression probability in MCI to AD, investigators suggested that hippocampal atrophy rate [29,30,37] and ventricular expansion rate [30] could be predictive factors for progression of MCI to AD. In presymptomatic familial AD (FAD), the amyloid retention in the thalamus and striatum rather than fronto-temporal cortex using amyloid imaging [8,9] prompted studies of volume reduction in the same regions. A few studies of presymptomatic FAD reported atrophy of thalamus, caudate nucleus or putamen [38]. Although FAD is a condition with genetic background, we speculated that subcortical involvement might happen in the presymptomatic period in sporadic AD and its progression according to the severity of the AD pathology might be worthy of evaluation. Nucleus accumbens is a part of ventral striatum which is prone to cognitive dysfunction through the connections to limbic areas in AD or elderly [28]. In addition to classical pathologic report of striatal involvement in AD [4,7], there were several attempts to clarify the role of striatum in AD pathogenesis. De Jong et al [28] found that volume of the nucleus accumbens was smaller even in preclinical stages of dementia than in a group with normal cognition, and they reported that nucleus accumbens can be a significant indicator for progression to dementia through the steeper slope of cognitive decline. However they included elderly persons instead of patients with MCI, and used conversion to either vascular dementia or AD as an endpoint. Our study instead looked specifically in MCI patients at the value of baseline subcortical volumes in predicting subsequent conversion from MCI to AD. Our findings add to the findings of De Jong et al., a role of cross-sectional nucleus accumbens volume in predicting progression from MCI to AD. In spite of a clear trend towards lower volume in the p-MCI group, there was no baseline volume difference of nucleus accumbens between s-MCI & p-MCI. This absence of a significant difference could be explained by the short follow-up duration in this study and the resulting relatively small group size of p-MCI, or alternatively, by the specific vulnerability of the nucleus accumbens. As a part of limbic circuit, the nucleus accumbens may be vulnerable to neurodegenerative processes already in a pre-clinical stage or early stages of AD, which could lead to similar degree of atrophy in both MCI groups. This hypothesis could be investigated in future studies e.g. by studying patients with subjective memory complaints and comparing those who subsequently progressed to AD to those without progression, to confirm the predictive role and vulnerability of nucleus accumbens. We assessed the correlation between cognitive dysfunction and subcortical structural volume. Basically we expected that subcortical volume reduction might reflect cognitive deterioration as prior studies showed the correlation between cognitive status and structural changes [10,11,36]. Our findings in a large sample, including controls, MCI and AD, also demonstrated a significant positive correlation between MMSE and subcortical volumes with low correlation coefficients, and the role of globus pallidus on the cognition were negligible as in the other reports.[1,10]. From this, we could suggest that positive correlation between structural atrophy and cognitive status does not always simply mean a relation to further progression. Interesting finding was larger volume of globus pallidus in p-MCI comparing with sMCI although it did not have any significance. Some studies in presymptomatic FAD 64

reported similar findings [38] and they explained this increment might reflect reactive neuronal hypertrophy and inflammatory process in the early presymptomatic period. From these results, the lack of volume difference of globus pallidus between groups might not be negligible but structural change of globus pallidus might happen in much later stage in AD-like neurodegeneration. Our study has a number of strengths and limitations. Few studies have compared the subcortical structures between MCI, AD and controls, especially evaluating the predictive role of the subcortical structures besides medial temporal lobe structures in a large cross-sectional cohort. Also to reduce possible variability in other multicenter researches as much as possible, we used imaging data from a single scanner and constant imaging protocol. Among the possible limitations, the first is the fact that we could not study volume changes over time as it was a cross-sectional study. Although our actual goal was to find the value of the point-based subcortical volume reduction, adding more information such as atrophy rate or ventricular enlargement rate measured by longitudinal evaluation of imaging would give more powerful explanations in prediction of progression to AD from MCI on this current crosssectional study. Another potential limitation is the fact that we did not sub-classify MCI patients as being amnestic or non-amnestic, while the future course and relationship with subcortical structures could be affected by clinical subtypes [39]. However, there was a report of longitudinal comparison between MCI subtypes, which showed no difference between the chance of reversion or the risk of conversion to AD, suggesting MCI subtypes were diagnostically unstable [40]. In addition, due to the relatively short follow-up duration, it should be expected that a substantial proportion of the s-MCI patients will later progress to AD. As this was a retrospective study, we accepted variability in follow-up times. Patients were scheduled for a new visit annually or according to clinical needs, leading to variability in follow-up times in our samples. To maximize discriminative power, we considered the last clinical follow-up time point available, leading to additional variability in follow-up duration, and 8 patients with follow-up duration under 1 year. Longer follow-up is needed to further confirm these results. Lastly, we did not include the information of disease duration and APOE genotypes in each patient. As long duration of disease could mean exhausted functional reservoir, information about disease duration would be important regardless of MMSE score. Since APOE genotype was inconsistently available in this patient group, we have decided not to include it in the analyses in order to maintain the maximum number of patients. In conclusion, we confirmed the subcortical volume reduction in the MCI and AD, and found association with cognitive impairment. Nonetheless, besides nucleus accumbens, subcortical volume did not predict progression of MCI to AD, suggesting these structures have limited value as cross-sectional predictors of future disease course. Further study with larger sized sample would reveal more value of the subcortical volume on the given-time point as an imaging biomarker and it would benefit from therapeutic intervention.

65

Acknowledgements The VUmc Alzheimer Center is supported by Alzheimer Nederland and Stichting VUmc Fonds. The study was supported by the Nederlands Organisation for Scientific Research (NWO NIHC; #056-13-001). Research of the VUmc Alzheimer Center is part of the neurodegeneration research program of the Neuroscience Campus Amsterdam. The clinical database structure was developed with funding from Stichting Dioraphte. The gradient non-linearity correction was kindly provided by GE medical systems, Milwaukee.

66

Tables and Figures Table 1. Demographics and normalized brain volume in each group at baseline Demographics Controls (n=181) MCI (n=201) AD (n=391) Age, years 65± 8 70± 9† 69±9† Gender, 82/99, (45.3%) 82/119, 221/170, (56.5%) a women/men (♀ %) (40.8%) b Education, years 5.01 ± 1.81 4.72 ± 1.79 4.66 ±1.58†

p <0.001 <0.001 <0.05

MMSE 28.07 ± 1.78 26.02± 2.60† 19.96 ±5.07†‡ <0.001 CAMCOG 92.34 ± 6.31 84.99 ± 8.06† 66.49 ±15.43†‡ <0.001 CDR 0.07±0.19 0.39±0.24† 1.08±0.51†‡ <0.001 Follow up, years 1.80±0.75 2.11 ±0.96 2.17 ±0.79 0.055 Normalized volume(cm3) Thalamus 19.7±1.7 18.9±1.7† 18.6±1.7†‡ <0.001 Caudate 9.0±1.0 8.8±1.0 8.6±1.0†,‡ <0.001 Putamen 12.3±1.2 11.7±1.3† 11.5±1.2† <0.001 Globus pallidus 4.6±0.6 4.5±0.6 4.6±0.6 NS Hippocampus 9.8±1.2 8.9±1.3† 8.4±1.3†,‡ <0.001 Amygdala 3.9±0.6 3.6±0.5† 3.5±0.6† <0.001 Nucleus accumbens 1.1±0.3 0.9±0.3† 0.9±0.3† <0.001 All data are represented as (m±sd) unless indicated otherwise. a b chi-square test; Kruskal-Wallis test followed by Bonferroni’s post hoc test For the normalized brain volume, ANOVA with Bonferroni’s post hoc test were performed with correcting for age, sex and education † p<0.05 difference between subjects with controls and other groups; ‡ p<0.05 difference between subjects with MCI and AD; NS : non-significant

Table 2. Comparison between s-MCI and p-MCI Demographics s-MCI (n=160) p-MCI (n=35) Age, years 70±8.8 70±9.6 a Gender, women/men (♀ %) 64/96 (40.0%) 18/17 (51.4%) Education, yearsb 4.65±1.82 5.00±1.75 MMSE 26.1±2.6 25.7 ±2.7 CAMCOG 85.0±8.4 84.5 ± 6.6 Follow up, years 2.1±1.0 2.2±1.0 Normalized volume (cm3) Thalamus 18.9±1.8 18.6±1.4 Caudate 8.8±1.0 8.8±0.9 Putamen 11.7±1.4 11.6±1.3 Globus Pallidus 4.5±0.6 4.6±0.7 Hippocampus‡ 9.1±1.3 8.3±1.2 Amygdala† 3.6±0.6 3.4±0.5 Nucleus accumbens 1.0±0.3 0.9±0.3 All data are represented as (m±sd) unless indicated otherwise. a b chi-square test; Kruskal-Wallis test followed by Bonferroni’s post hoc test. For normalized brain volumes analyses, ANOVA with correction for age, sex and education was performed. † p< 0.05; ‡ p <0.001 67

Table 3. Hazard ratios and 95% confidence intervals for progression to Alzheimer’s disease within MCI groups Model 1 Model 2 Model 3 Thalamus 1.26(0.91-1.75) 1.09(0.77-1.55) 1.10(0.77-1.58) Caudate 1.10(0.80-1.51) 1.18(0.87-1.61) 1.26(0.88-1.80) Putamen 1.13(0.85-1.51) 1.02(0.76-1.36) 1.10(0.80-1.51) Globus pallidus 0.88(0.65-1.18) 0.84(0.62-1.14) 0.88(0.64-1.20) Hippocampus 1.56(1.16-2.11) † 1.45(1.07-2.00) † 1.60(1.15-2.21) † Amygdala 1.33(0.98-1.81) 1.14(0.80-1.61) 1.11(0.77-1.60) N. accumbens 1.59(1.16-2.18)† 1.47(1.03-2.09) † 1.60(1.09-2.35) † † : p<0.05 Model 1 : crude model, model 2 : adjusted for age, sex & education, model 3: age, sex, education, CAMCOG-R & MMSE

68

Figure 1. Example of automated segmentation of subcortical gray matter structure using FIRST in MCI patient. Segmentations are shown overlaid on the 3D T1-weighted images in three orthogonal orientations, corresponding to the axial (left column), coronal (middle column) and sagittal (right column) planes. Colored structures; Yellow: Hippocampus, Light orange: Amygdala, Green: Thalamus, Light blue: Caudate nucleus, Pink: Putamen, Dark blue: Globus pallidus, Orange: Nucleus accumbens.

3

Figure 2. Normalized volume (cm ) of six subcortical structures and hippocampus in control, MCI and AD. ANOVAs, adjusted for sex and age, showed that all subcortical gray matter volumes except globus pallidus differed between groups (p<0.05) and showed a tendency to decrease from controls to MCI and further to AD. Post hoc tests showed that all structures were smaller in AD than in controls, except globus pallidus, and were smaller in MCI than controls except globus pallidus and caudate nucleus, the volumes of hippocampus, thalamus and caudate nucleus were smaller in AD compared to MCI.

69

Figure 3. Scatter plots and trend lines show the association between normalized subcortical volumes and MMSE scores at baseline. X-axis represents MMSE score and Y-axis represents 3 volume (cm ).

70

References [1] Petersen RC. Mild cognitive impairment as a diagnostic entity. J Intern Medicine 2004;256:183-94. [2] Petersen RC, Roberts RO, Knopman DS, et al. Mild cognitive impairment: ten years later. Arch Neurol 2009;66:1447-55. [3] Farias ST, Mungas D, Reed BR, et al. Progression of mild cognitive impairment to dementia in clinic- vs community-based cohorts. Arch Neurol 2009;66:1151-7. [4] Braak H, Braak E. Neuropathological stageing of Alzheimer-related changes. Acta Neuropathol 1991;82:239-59. [5] Atiya M, Hyman BT, Albert MS, et al. Structural magnetic resonance imaging in established and prodromal Alzheimer disease: a review. Alzheimer Dis AssocDisord 2003;17:177-95. [6] Braak H, Braak E. Alzheimer’s disease affects limbic nuclei of the thalamus. Acta Neuropathol 1991;81:261-68. [7] Braak H, Braak E. Alzheimer’s disease: striatal amyloid deposits and neurofibrillary changes. J Neuropathol Exp Neurol 1990;49:215-24. [8] Knight WD, Okello AA, Ryan NS, et al. Carbon-11-Pittsburgh compound B positron emission tomography imaging of amyloid deposition in presenilin 1 mutation carriers. Brain2011;134(Pt 1):293-300. [9] Villemagne VL,Ataka S, Mizuno T, et al. Highstriatal amyloid beta-peptide deposition across different autosomalAlzheimer disease mutation types. Arch Neurol 2009;66:1537-44. [10] de Jong LW, van der Hiele K, Veer IM, et al. Strongly reduced volumes of putamen and thalamus in Alzheimer’s disease: an MRI study. Brain 2008;131:3277-85. [11] Roh JH, Qiu A, Seo SW, et al. Volume reduction in subcortical regions according to severity of Alzheimer’s disease. J Neurol 2011;258:1013-20. [12] Pievani M, Bocchetta M, Boccardi M, et al. Striatal morphology in early-onset and late onset Alzheimer's disease: a preliminary study. Neurobiol Aging 2013;34:1728 39. [13] Liu Y, Paajanen T, Zhang Y, et al. Analysis of regional MRI volumes and thickness as predictors of conversion from mild cognitive impairment to Alzheimer's disease. Neurobiol Aging2010;31:1375-85. [14] Cherubini A, Peran P, Spoletini I, et al. Combined Volumetry and DTI in Subcortical Structures of Mild Cognitive Impairment and Alzheimer’s Disease Patients. J Alzheimer’s disease 2010;19:1273-82. [15] Tang X1, Holland D, Dale AM, et al. Shape abnormalities of subcortical and ventricular structures in mild cognitive impairment and Alzheimer's disease: Detecting, quantifying, and predicting. Hum Brain Mapp 2014;35:3701-25. [16] van der Flier WM, Pijnenburg YA, Prins N, et al. Optimizing Patient Care and Research: The Amsterdam Dementia Cohort. J Alzheimers Dis 2014;41:313-27. [17] McKhann G, Drachman D, Folstein M, et al. Clinical diagnosis of Alzheimer’s disease: report of the NINCDS-ADRDA Work Group under the auspices of Department of Health and Human Services Task Force on Alzheimer’s Disease. Neurology 1984;34;939-44. [18] Folstein MF, Folstein SE, McHugh PR. Mini-mental state: a practical method for grading the cognitive state of patients for the clinician. J Psychiatr Res 1975;12:189-98. [19] Roth M, Huppert FA, Mountjoy CQ, et al. The Cambridge Examination for Mental Disorders of the Elderly-Revised. Cambridge: Cambridge University Press; 1999. [20] Patenaude B, Smith SM, Kennedy DN, et al. A Bayesian model of shape and appearance for subcortical brain segmentation. Neuroimage 2011;56:907-22. 71

[21] Fazekas F, Chawluk JB, Alavi A, et al. MR signal abnormalities at 1.5 T in Alzheimer's dementia and normal aging. AJR Am J Roentgenol 1987;149:351-6. [22] Smith SM, Zhang Y, Jenkinson M, et al. Accurate, robust, and automated longitudinal and cross-sectional brain change analysis. Neuroimage 2002;17:479-89. [23] Popescu V, Battaglini M, Hoogstrate WS, et al. Optimizing parameter choice for FSL-Brain Extraction Tool (BET) on 3DT images in multiple sclerosis. Neuroimage 2012;61:1484-94. [24] Babalola KO, Patenaude B, Aljabar P, et al. An evaluation of four automatic methods of segmenting the subcortical structures in the brain. Neuroimage 2009;47:1435 47. [25] Mann DMA. The topographic distribution of brain atrophy in Alzheimer’s disease. Acta Neuropathol 1991;83:81-6. [26] Risacher SL, Saykin AJ, West JD, et al. Baseline MRI predictors of conversion from MCI to probable AD in the ADNI cohort. Curr Alzheimer Res 2009;6:347-361. [27] Stepan-Buksakowska I, Szabo N, Horinek D, et al. Cortical and subcortical atrophy in Alzheimer disease: parallel atrophy of thalamus and hippocampus. Alzheimer Dis Assoc Disord 2014;28:65-72. [28] de Jong LW, Wang Y, White LR, et al. Ventral striatal volume is associated with cognitive decline in older people; a population based MR-study. Neurobiol Aging 2012;33:424.e1-10. [29] Eckerstrom C, Olsson E, Borga M, et al. Small baseline volume of left hippocampus is associated with subsequent conversion of MCI into dementia: The Goteborg MCI study. J Neurol Sci 2008;272:48-59. [30] Leung KK, Bartlett JW, Barnes J, et al. Cerebral atrophy in mild cognitive impairment and Alzheimer disease: rates and acceleration. Neurology 2013;80:648-654. [31] Fleisher AS1, Sun S, Taylor C, et al. Volumetric MRI vs clinical predictors of Alzheimer disease in mild cognitive impairment. Neurology 2008:70:191-9. [32] Bruen PD, McGeown WJ, Shanks MF, et al. Neuroanatomical correlates of neuropsychiatric symptoms in Alzheimer's disease. Brain 2008;131(Pt 9):2455-63. [33] Piccininni M, Di Carlo A, Baldereschi M, et al. Behavioral and psychological symptoms in Alzheimer’s disease: frequency and relationship with duration and severity of the disease. Dement Geriatr Cogn Disord 2005;19(5-6):276-81. [34] Tekin S, CummingsJL. Frontal-subcortical neuronal circuits and clinical neuropsychiatry: an update. J Psychosom Res 2002;53:647-54. [35] Landau SM, Harvey D, Madison CM, et al. Comparing predictors of conversion and decline in mild cognitive impairment. Neurology 2010;75:230-38. [36] Ahn HJ, Seo SW, Chin JH, et al. The cortical neuroanatomy of neuropsychological deficits in mild cognitive impairment and Alzheimer’s disease: A surface-based morphometric analysis. Neuropsychologia 2011;49:3931-45. [37] Henneman WJP, Sluimer JD, Barnes J, et al. Hippocampal atrophy rates in Alzheimer disease. Neurology 2009;72:999-1007 [38] Lee GJ, Lu PH, Medina LD, et al. Regional brain volume differences in symptomatic and presymptomatic carriers of familial Alzheimer’s disease mutations. J Neuol Neurosurg Psychiatry 2013;84:154-62. [39] Klekociuk SZ, Summers MJ. Exploring the validity of mild cognitive impairment (MCI) subtypes: multiple-domain amnestic MCI is the only identifiable subtype at longitudinal follow-up. J Clin Exp Neuropsychol 2014;36:290-301. [40] Han JW, Kim TH, Lee SB, et al. Predictive validity and diagnostic stability of mild cognitive impairment subtypes. Alzheimers Dement 2012;8:553-9.

72

Chapter 3 - Patterns of gray matter loss in different forms of dementia

73

Chapter 3.1

More atrophy of deep gray matter structures in Frontotemporal Dementia compared to Alzheimer’s Disease 1

5

Christiane Möller MSc , Nikki Dieleman MSc , Wiesje M van der Flier 1,2 3 1 PhD , Adriaan Versteeg , Yolande Pijnenburg MD PhD , Philip 1 3 Scheltens MD PhD , Frederik Barkhof MD PhD , Hugo Vrenken 3,4 PhD 1 2 Alzheimer center & Department of Neurology, Department of 3 Epidemiology & Biostatistics, Department of Radiology & Nuclear 4 Medicine, Department of Physics & Medical Technology, Neuroscience Campus Amsterdam, VU University medical center, P.O. Box 7057, 5 1007 MB Amsterdam, the Netherlands, Department of Radiology, University Medical Center Utrecht, P.O. Box 85500, 3508 GA Utrecht, the Netherlands. Journal of Alzheimers Disease 2015 Jan 1;44(2):635-47.

Abstract Background: The involvement of frontostriatal circuits in Frontotemporal Dementia (FTD) suggests that deep gray matter structures (DGM) may be affected in this disease. Objective: We investigated whether volumes of DGM structures differed between patients with bvFTD, Alzheimer’s Disease (AD) and subjective complaints (SC) and explored relationships between DGM structures, cognition and neuropsychiatric functioning. Methods: For this cross-sectional study we included 24 patients with FTD and matched them based on age, sex and education at a ratio of 1:3 to 72 AD patients and 72 patients with SC who served as controls. Volumes of hippocampus, amygdala, thalamus, caudate nucleus, putamen, globus pallidus and nucleus accumbens were estimated by automated segmentation of 3D T1-weighted MRI. MANOVA with Bonferroni adjusted post-hoc tests was used to compare volumes between groups. Relationships between volumes, cognition and neuropsychiatric functioning were examined using multivariate linear regression and Spearman correlations. Results: Nucleus accumbens and caudate nucleus discriminated all groups, with most severe atrophy in FTD. Globus pallidus volumes were smallest in FTD and discriminated FTD from AD and SC. Hippocampus, amygdala, thalamus and putamen were smaller in both dementia groups compared to SC. Associations between amygdala and memory were found to be different in AD and FTD. Globus pallidus and nucleus accumbens were related to attention and executive functioning in FTD. Conclusion: Nucleus accumbens, caudate nucleus, and globus pallidus were more severely affected in FTD than in AD and SC. The associations between cognition and DGM structures varied between the diagnostic groups. The observed difference in volume of these DGM structures supports the idea that next to frontal cortical atrophy, DGM structures, as parts of the frontal circuits, are damaged in FTD rather than in AD. 74

Introduction Alzheimer’s disease (AD) frontotemporal dementia (FTD) are the leading causes of early-onset dementia[1,2]. Cortical atrophy of the two forms of dementia has been extensively studied: Typically, atrophy of the medial temporal lobe (MTL) is seen in AD patients and associated with episodic memory problems [3,4]. However, MTL atrophy is also common in bvFTD [5,6]. Conversely, FTD is typically characterized by atrophy of frontal and anterior temporal lobes, which is associated with changes in behavior and executive functioning [7-9], but atrophy in these regions does not exclude a diagnosis of AD [10,11]. In the discrimination of AD and FTD, deep gray matter (DGM) structures have received less attention. So far, MRI-based studies have demonstrated that AD is associated with atrophy of thalamus, putamen, and caudate nucleus [12-16]. However, results from published studies are difficult to compare as they used different image analysis techniques or focused on different DGM structures [13,15,17,18]. In the context of FTD, DGM structures may be even more important, as they are part of the frontostriatal circuits, known to be affected in FTD [19,20]. Thalamus, neostriatum, nucleus accumbens and globus pallidus connect motor- and cognitive-loops with the prefrontal cortex. Degeneration of these DGM structures may lead to circuit failure and eventually alterations of cognition and behavior [17,19,21]. MRI-based measures of DGM structures could provide important information on the differential distribution of pathology between AD and FTD and may explain some of the clinical characteristics typical of the diseases. Therefore the aim of this study was to investigate atrophy of hippocampus, amygdala and the DGM structures in AD, FTD, and control subjects and to explore the effect of DGM atrophy on cognitive and neuropsychiatric functioning.

Materials and Methods Patients All patients visited the Alzheimer Center of the VU University medical center between 2008 and 2011 where they underwent a standardized one-day assessment for clinical evaluation including medical history, informant-based history, physical and neurological examination, blood tests, neuropsychological assessment, electroencephalography and magnetic resonance imaging (MRI) of the brain. Patients are asked informed consent for the use of their clinical data for research purposes and are as such included in the Amsterdam Dementia Cohort [22]. For the current study, we retrospectively selected 24 FTD patients who met the inclusion criteria as described below. One of the authors matched these patients through visual inspection on a 1:3 basis by age, sex and educational level to 72 AD patients and 72 patients with subjective complaints (SC) who were used as controls. All diagnoses were made in a multidisciplinary consensus meeting according to the core clinical criteria of the National Institute on Aging and the Alzheimer’s Association workgroup for probable AD and according to the clinical diagnostic criteria of FTD based on the results of the one-day assessment as described above [23-25]. Among the AD patients, one patient met the criteria for posterior cortical atrophy (PCA). Six AD patients susceptible for genetic mutations underwent a genetic screening. None of them were positive for 75

genetic mutations. Among the FTD patients there were four patients with ALS and one patient with motor neuron disease. Eight FTD patients susceptible for genetic mutations underwent a genetic screening. None of them were positive for known genetic mutations. On visual inspection of the MRI scans, frontal atrophy was worse than temporal atrophy in all FTD patients. No progressive nonfluent aphasia (PA) or semantic dementia (SD) cases were identified. As controls, we used patients who presented at our memory clinic with subjective complaints. They were labeled as having subjective complaints when they presented with memory complaints, but cognitive functioning was normal and criteria for MCI, dementia or any other neurological or psychiatric disorder known to cause cognitive decline were not met. The diagnosis was also made in a multidisciplinary consensus meeting taking into account results of all examinations as described above. For inclusion in the present study all patients had to fulfill the following criteria: (1) meet the criteria for AD, FTD or subjective complaints based on the core clinical criteria, (2) diagnosis stayed unchanged after 12 months clinical follow-up, (3) availability of a T1-weighted 3dimensional MRI scan (3DT1) at 3 tesla MRI (details see section below) and (4) availability of neuropsychological examination. Exclusion criteria were: (1) age younger than 40 years; (2) failure of segmentation software to analyze DGM volumes due to abnormal tracing of structures (details see section below); and (3) lacunar infarction in the DGM structures. Disease duration was calculated based on the time difference between date of diagnosis and the year patients caregivers noticed the first symptoms. Level of education was rated on a seven-point scale [26]. The local medical ethics committee approved the study. Ethics review criteria conformed to the Helsinki declaration. All patients gave written informed consent for their clinical data to be used for research purposes. Demographic data can be found in table 1. Cerebrospinal fluid (CSF) To gain more diagnostic certainty, CSF was obtained by lumbar puncture. Amyloidβ1−42 (Aβ42), total tau, and tau phosphorylated at threonine-181 (Ptau-181) were measured by sandwich ELISA (Innogenetics, Gent, Belgium) [27]. CSF analyses were performed at the VUmc Department of Clinical Chemistry. Cut-off levels in our lab are as follows: Aβ42< 550, total tau > 375, and ptau > 52 [27]. CSF was available for 140 subjects (SC: n=59; AD: n=62; FTD: n=19). MR image acquisition and review Imaging was carried out on a 3 Tesla scanner (Signa HDxt, GE Healthcare, Milwaukee, WI, USA) using an 8-channel head coil with foam padding to restrict head motion. The scan protocol included a whole-brain isotropic 3DT1 fast spoiled gradient echo sequence (FSPGR; TR 708 ms, TE 7 ms, flip angle 12º, 180 sagittal slices, field of view 250 mm, slice thickness 1 mm, voxel size 0.98x0.98x1 mm) which was used for segmentation. In addition, the MRI protocol included a 3D Fluid Attenuated Inversion Recovery (FLAIR) sequence, dual-echo T2-weighted sequence, and susceptibility weighted imaging (SWI) which were reviewed for brain pathology other than atrophy by an experienced radiologist.

76

Volume measurement of DGM structures DICOM images of the FSPGR sequence were corrected for gradient nonlinearity distortions and converted to NIfTI format. The algorithm FIRST (FMRIB’s integrated registration and segmentation tool, FSL 4.15) [28,29] was applied to estimate left and right volumes of seven structures: hippocampus, amygdala, thalamus, caudate nucleus, putamen, globus pallidus, and nucleus accumbens. Left and right volumes were summed to obtain total volume for each structure. FIRST performed both registration and segmentation of the above mentioned anatomical structures. A twostage linear registration was performed to achieve a more robust and accurate prealignment of the seven structures. During the first-stage registration, the 3DT1 images were registered linearly to a common space based on the Montreal Neurological Institute (MNI) 152 template with 1x1x1 mm resolution. After the first registration, a second stage linear registration using a subcortical mask or weighting image, defined in MNI space, was performed to improve registration for the seven structures. Both stages used 12 degrees of freedom. This 2-stage registration was followed by segmentation based on shape models and voxel intensities. Volumes of the seven structures were extracted in native space, taking into account the transformations matrices during registration. The final step was a boundary correction based on local signal intensities. All registrations and segmentations were visually checked for errors. For examples of automated segmentations see Figure 1. FIRST has been shown to give accurate and robust results for the segmentation of subcortical structures and that it performs comparably to or better than other automatic methods [28,30,31]. To correct the DGM structures for head size, we used the volumetric scaling factor (VSF) derived from SIENAX (Structural Image Evaluation using Normalization of Atrophy Cross-sectional) [32], also part of FSL. In short, first SIENAX extracted skull and brain from the 3DT1 input whole-head image. In our study, brain extraction was performed using optimized parameters [33]. These were then used to register the subject’s brain and skull image to a standard space brain (derived from MNI152 template) to estimate the scaling (volumetric scaling factor (VSF)) between the subject's image and standard space. Normalization for head size differences was done by multiplying the raw volumes of the DGM structures by the VSF. Neuropsychological assessment To assess dementia severity we used the Mini-Mental State Examination (MMSE). Cognitive functioning was assessed using a standardized neuropsychological test battery covering five major domains: memory (immediate recall, recognition and delayed recall of Dutch version of the Rey Auditory Verbal Learning Test and total score of Visual Association Test A), language (Visual Association Test picture naming and category fluency (animals: 1 min)), visuospatial functioning (subtests of Visual Object and Space Perception Battery (VOSP): incomplete letters, dot counting, and number location), attention (Trail Making Test part A (TMT A), Digit Span forward, and Letter Digit Substitution Test (LDST)), and executive functioning (Digit Span backwards, Trail Making Test part B (TMT B), letter fluency, and Stroop Color-Word card subtask [34]). For a detailed description of neuropsychological tests see Smits et al. [35]. For each cognitive task, z-scores were calculated from the raw test scores by the formula z=(x-μ)/σ, where μ is the mean and σ is the standard deviation of the 77

subjective complaints group. The value z = 0 therefore reflects the average test performance of the subjective complaints group in a given domain. Scores of TMT A, TMT B, and Stroop color-word card were inverted by computing -1 x z-score, because higher scores imply a worse performance. Next, composite z-scores were calculated for each cognitive domain by averaging z-scores. Composite z-scores were calculated when at least one neuropsychological task was available in each cognitive domain. There was variability in the number of completed neuropsychological tests. On average, every test was completed by 136 patients, ranging from 158 (DS forward) to 105 (LDST). When tests were not finished, this was because of cognitive impairment or lack of time. Neuropsychiatric assessment To assess psychopathology we used the Neuropsychiatric Inventory (NPI)[36]. The NPI evaluates 12 neuropsychiatric disturbances common in dementia: delusions, hallucinations, agitation/aggression, depression/dysphoria, anxiety, elation/euphoria, apathy/indifference, disinhibition, irritability/lability, aberrant motor behavior, sleep and night-time behavior disturbances, and appetite/eating abnormalities. The severity and frequency of each neuropsychiatric symptom were rated based on scripted questions administered to the patient’s caregiver. Symptom frequency is rated on a scale of 1 (occurs rarely) to 4 (occurs very often) and symptom severity on a scale of 1 (mild) to 3 (severe). Scores for symptom domains are calculated by multiplying the frequency of each symptom by its severity. The total NPI score is derived by summing up all symptom domain scores. The NPI was completed by 93 patients (31 SC, 48 AD, 14 FTD). Statistical analysis SPSS version 20.0 for Windows was used for statistical analysis. Differences between groups for demographics, composite cognitive domain z-scores and NPI domain scores were assessed using ANOVA (VSF), Kruskal-Wallis tests (age, level of education, disease duration, CSF values, MMSE, composite cognitive domain z-scores 2 and NPI domain scores), and χ tests (sex) (table 1) . Multivariate analysis of variance (MANOVA) was used to compare head size adjusted total volumes of the MTL structures (hippocampus and amygdala) and the DGM structures (thalamus, caudate nucleus, putamen, globus pallidus, and nucleus accumbens) between diagnostic groups with Bonferroni adjusted post-hoc tests. Age, sex and disease duration were used as covariates. A second MANOVA was performed separately for left and right volumes of the seven structures. Statistical significance was set at p<0.05. To assess associations between the seven DGM structures (independent variables; entered simultaneously) and cognitive domains (dependent variable) we performed multivariate linear regression analyses. We used five models, one for each cognitive domain. Age, sex, disease duration and diagnosis (using dummy variables) were entered as covariates in the model. To check if associations with DGM structures differed according to diagnosis, interaction terms (dummy-diagnosis x DGM structure) were included in the model . If there was a significant interaction between diagnosis and DGM structure (p<0.10), standardized β are displayed for each 78

diagnostic group separately. When no significant interaction was found, the overall β is reported. Statistical significance was set at p<0.05. As NPI domain scores and NPI total score were not normally distributed we used Spearman correlations to assess associations between the seven DGM structures and NPI domain scores and NPI total score for each diagnostic group separately. Statistical significance was set at p<0.01.

Results Demographics Demographic data, composite cognitive domain z-scores and NPI domain scores are summarized in Table 1. Groups were well matched for age, sex and level of education. Disease duration did not differ between the groups. All CSF biomarkers in AD differed from patients with subjective complaints and FTD, with a CSF profile in line with the clinical diagnosis [27,37,38]. None of the FTD patients had A-beta values under the cutoff, 10 patients with subjective complaints had lower A-beta values and 11 AD patients had higher A-beta values. MMSE scores were different between diagnostic groups, with AD patients having the lowest scores. The VSF did not differ between the groups. AD patients performed worse on memory tasks, visuospatial functioning, and executive functioning than FTD and patients with subjective complaints. For the language and attention domains both patient groups performed worse than the subjective complaints group. FTD patients exhibited most neuropsychiatric symptoms: The total NPI score was higher than in patients with subjective complaints and AD patients. They had higher scores than the two other groups on aberrant motor behavior and appetite/eating abnormalities. Both dementia groups scored higher on apathy/indifference than the subjective complaints group. Deep gray matter structures Volumes of DGM structures are summarized in Table 2 and Figure 2. MANOVA revealed group differences in MTL and DGM structures. Post hoc tests showed that total volumes of hippocampus and amygdala discriminated both dementia groups from the subjective complaints group. Nucleus accumbens and caudate nucleus volume discriminated all groups, with FTD having most severe atrophy (p<0.001). Globus pallidus volumes were smallest in FTD and discriminated FTD from AD (p<0.001) and from patients with subjective complaints (p<0.001). No volume differences between patients with AD and subjective complaints for globus pallidus were found (p=1.00). Volumes of thalamus and putamen were larger in patients with subjective complaints than in both dementia groups. We did not find any left-right differences for the DGM structures. The MTL structures showed some subtle left-right differences: Whereas left hippocampus discriminated only dementia from patients with subjective complaints, the right hippocampus also discriminated AD from FTD (p=0.045), with smaller right hippocampal volume in FTD. Left amygdala volumes differed only between patients with subjective complaints and FTD, whereas right amygdale volumes also discriminated AD from patients with subjective complaints (p=0.021). Left and right volume differences are summarized in the supplementary material table S1. 79

Relationship between volume of DGM structures, cognitive functioning and neuropsychiatric symptoms As the DGM structures did not show any left-right differences and to avoid multiple comparisons, associations between cognition, neuropsychiatric symptoms and DGM structures were only conducted for the total DGM volumes. Multivariate linear regression analysis showed that after adjustment for age, sex, disease duration and diagnosis, there was a trend for an association between hippocampus and memory (standardized β=0.19, p=0.055). In addition, the interaction term for diagnosis and amygdala was significant, implying that for this structure, associations with memory differed according to diagnostic group. For AD, there was a positive association between amygdala and memory (standardized β=0.25, p=0.011), for FTD there was a negative association (standardized β=-0.32, p=0.037), and the association for patients with subjective complaints was not significant (standardized β=-0.20, p=0.113). There were no associations between any of the DGM structures and language. For visuospatial functioning, there appeared to be interactions between diagnosis and hippocampus (p=0.042), but the associations did not reach significance in any of the groups (hippocampus: AD: standardized β=-0.26; FTD: standardized β=0.05; SC: standardized β=0.19; p>0.05), nor were there any associations between any of the other DGM structures and visuospatial functioning. For attention, we observed interactions between diagnosis and globus pallidus (p=0.005). For FTD, there was a positive association for globus pallidus (standardized β=0.75, p=0.010). For patients with AD and subjective complaints no associations between globus pallidus and attention were found (globus pallidus: AD: standardized β=-0.15; SC: standardized β=-0.13; p>0.05). There were no associations between any of the other DGM structures and attention. Finally, for executive functions there were interactions between diagnosis and nucleus accumbens (p=0.042) and globus pallidus (p=0.014). For FTD, there was a positive association for globus pallidus (standardized β=0.93, p=0.002) and a negative association for nucleus accumbens (standardized β=-0.49, p=0.035). Associations for AD and patients with subjective complaints were not significant. Spearman correlations between neuropsychiatric symptoms and DGM structures showed a significant correlation in FTD patients: Disinhibition was negatively correlated with volume of nucleus accumbens (r=-0.69, p=0.007). Although, aberrant motor behavior and appetite/eating abnormalities were reported very frequently in FTD, no correlations with DGM or MTL structures were found.

Discussion We found more prominent volume loss of DGM structures in FTD than in AD and SC. Specifically, caudate nucleus, globus pallidus and nucleus accumbens volumes differentiated between both types of dementia, with FTD being more severely affected. Hippocampus, amygdala, thalamus and putaminal volumes only discriminated dementia from patients with subjective complaints. Associations between cognition and the DGM structures show different effects for the diagnostic groups. In FTD, attention showed a positive association with globus pallidus and executive functioning was positively related to globus pallidus and negatively to 80

nucleus accumbens. Memory was positively related to amygdala in AD and negatively in FTD. Visuospatial functioning showed an overall effect with hippocampal volume. Neuropsychiatric symptoms and DGM structures were only related in FTD patients with volumes of nucleus accumbens correlating negatively with disinhibition. A number of pathological studies have identified DGM atrophy in AD and FTD at autopsy [39,40]. There is emerging evidence that compared to AD, generally regarded as a cortical disease, FTD patients have more subcortical brain damage [5,17,41,42]. This is in line with our findings that volumes of caudate nucleus, globus pallidus and nucleus accumbens are smaller in FTD compared to both AD patients and patients with subjective complaints. One possible explanation for the more severe volume loss of basal nuclei in FTD could be the specific underlying neuropathology. Whereas amyloid deposits are mainly found in the cerebral cortex, tau inclusions are found in subcortical regions as well [43,44]. Next to tau, fused in sarcoma protein inclusions (FUS) is also associated with atrophy of caudate nucleus [45]. Combinations with other clinical phenotypes like amyotrophic lateral sclerosis (ALS) could be another explanation for more atrophy of the DGM structures in the FTD group as ALS is associated with caudate nucleus, hippocampus and nucleus accumbens atrophy [46]. However, when we excluded the 4 patients with ALS and the 1 patient with motor neuron disease from the analysis, as well as the matching patients with subjective complaints and AD, results did not change essentially. Furthermore, we believe that in our sample the number of ALS cases is too small to drive the results of our study. Third, the involvement of basal nuclei as part of the frontostriatal circuits in FTD fits with the signs and symptoms of this disease, that include behavioral abnormalities and extrapyramidal symptoms [20,42]. Structural changes in components of these circuits could lead to Wallerian degeneration of the connecting fibers and eventually to failure of the whole circuit. Indeed, in AD patients, behavioral changes and involvement of the extrapyramidal system tend to develop much later in the disease. Exploration of the contribution of the DGM structures to cognition and neuropsychiatric symptoms revealed that the associations between amygdala and memory were different between AD and FTD patients. For attention and executive functions we also found different effects for the diagnostic groups. As expected there were positive associations between the MTL structures (trend for hippocampus) and memory in the AD group. The negative correlations between amygdala volumes and memory in the FTD group could have the following explanations: The exact neurocognitive mechanisms underlying the episodic memory impairment in bvFTD remain unknown, although it has been suggested that the memory deficits in FTD reflect executive dysfunction [47], most likely due to atrophy in the prefrontal cortices, in particular the orbitofrontal cortex [48,49] rather than to atrophy in MTL structures. Although some studies have pointed to posterior temporal and MTL pathology as a determinant of memory dysfunction in bvFTD [50]. Another explanation could be the type of memory test which place different demands on prefrontal versus medial temporal lobe functioning, such as recall versus recognition. It has been shown that memory dysfunction in bvFTD resulted from defective cognitive (retrieval) control processes rather than true amnesia. If our 81

memory test measures the relatively preserved recognition memory and not the impaired temporal source memory (remembering whether an item was shown in list A or list B) a negative association could occur [51,52]. Different kinds of memory have different underlying neural correlates, therefore a negative association between amygdala and the more frontally controlled recall test results could be the consequence. Another explanation could be an interaction with the underlying pathology. It has been suggested that cases with severe memory disturbance at presentation appear to have pathological changes associated with TDP-43 protein deposition [53]. Therefore it could be possible that patients with TDP-43 pathology present with severe memory disturbances but had a relatively spared amygdala, whereas a patient with tau or FUS pathology performs relatively well on a memory test but had a lot of amygdala atrophy. An alternative explanation could be that this finding is a type 1 error Only in FTD, there were significant relations between attention, executive functioning and the DGM structures. This fits our hypothesis that the DGM structures as relay stations of the frontal circuits play a role in FTD rather than in AD. The observed relation between globus pallidus and attention and executive functioning is in line with symptoms observed in the dorsolateral prefrontal syndrome [20]. The syndrome is characterized primarily by executive function deficits with the globus pallidus as an important relay station. Contrary to our expectations, we found a negative relationship between nucleus accumbens and executive functioning in the FTD group. A possible explanation for the negative association, could be that FTD patients with a small nucleus accumbens have relatively preserved executive functions or there are some interaction effects with the underlying pathological subtypes comparable to the effect we found with the amygdala and memory. An alternative explanation could be that this finding is a type 1 error as the other few studies who looked at the relationship between DGM structures and cognition did not find this association [18,54,55]. As studies examining the relationships between DGM structures and cognition are scarce, our results need to be replicated in larger cohorts. Social, emotional and behavioral changes are frequent symptoms of FTD and these symptoms might be related to volume loss of DGM structures too [56,57]. This theory is confirmed by our results that FTD patients exhibit more neuropsychiatric symptoms than the other groups and the negative correlation between nucleus accumbens and disinhibition which is one of the hallmarks in FTD. These findings are supported by other studies [58-60]. The correlation between nucleus accumbens with response disinhibition in FTD is consistent with theories of response-inhibition networks [61,62] like the fronto-subcortical network, including orbitofrontal cortex, left inferior frontal cortex and bilateral dorsal and ventral striatum. That is, the stopping process is generated by the inferior frontal cortex, leading to activation in the striatum, thereby inhibiting thalamo-cortical output and ultimately reducing motor cortex activity. Animal models suggest that the nucleus accumbens is crucial for response inhibition [63]. Furthermore, the accumbens is part of the mesolimbic dopaminergic system, which plays a key role in motivated and emotional behavior, reinforcement and reward, as well as for goal-directed behaviors [64]. These results together with our findings provide evidence for the hypothesis that pathology in FTD may start in parts 82

of frontostriatal circuits and eventually leads to failure of the whole circuit and may explain that behavioral symptoms proceed cognitive deterioration in FTD. Changes in eating behavior and aberrant motor behavior are common in FTD. Although, FTD patients in our study scored high on these domains, we did not find any relations with DGM or MTL structures. For aberrant motor behavior other studies failed to find an association with DGM or MTL structures too [57,59]. However, overeating and preference for sweet food in bvFTD have been associated with atrophy in the striatum [65,66]. A reason that we did not find any correlations could be the relatively small sample size in our study. Nevertheless despite a large study sample, another study also failed to identify brain regions that were specifically involved in eating changes in FTD [59]. Another reason could be that the causes and effects of eating disturbance in FTD are multi-factorial. It has been shown that the autonomic nervous system plays a role in satiety and central regulation of weight, as well as the hypothalamus [67,68]. Among the strengths of this study is the careful matching of groups enabling comparison of DGM volumes between groups without confounding effects of age and sex. We used FIRST, an automated and robust method for the extraction of the subcortical volumes which has been shown to give accurate results and performs comparable to or better than other automatic methods [28,30,31]. All subcortical volumes segmentations were visually checked, excluding the possibility of large segmentation errors. FIRST has clear advantages compared to voxel-based morphometry (VBM) [5,18] when assessing basal nuclei. VBM is suitable to compare patterns of cortical atrophy, but prone to segmentation errors in subcortical areas. Moreover, FIRST has also advantages over manual segmentations [42,69], since manual segmentation can take a long time and be prone to errors as well. We investigated all DGM structures, as well as the MTL structures – hippocampus and amygdala – and not only concentrated on a few structures. A possible limitation of this study is that we did not have pathological data available, so the possibility of misdiagnosis cannot be excluded. Nevertheless, we used an extensive standardized work-up and all 72 AD patients fulfilled clinical criteria of probable AD, all 24 patients fulfilled the criteria for FTD, i.e. frontal variant. Furthermore, CSF biomarkers were available for the majority of patients and average biomarker levels were congruent with the diagnosis [37,38], rendering the possibility of misdiagnosis less likely. Availability of pathological data would also enable us to study the effects of the different underlying pathology (tau, TDP subtypes, etc.) on volumes of DGM structures. This will be an important next step for future investigations. Another limitation could be the fact that we had a relatively small group of FTD patients, which could hamper the detection of any putative volume differences with the other diagnostic groups or associations with cognition because of low statistical power. Nevertheless, as FTD is a quite rare diagnosis, the sample size corresponds with that of other available studies to date. Moreover, we carefully matched on a 1:3 basis, resulting in optimal use of the cases we had available. Unfortunately, we did not have any cortical data available, which could have been correlated to the DGM structures from which they receive projections from. This would have strengthen our hypothesis about FTD as a network disorder even more. 83

However, our results are a good starting point to elaborate further on anatomical network differences between FTD and AD, In summary, this study yielded a comprehensive description of the differential involvement of DGM structures in AD and FTD patients and how these structures are related to cognitive and neuropsychiatric functioning. Volumes of nucleus accumbens, caudate nucleus, and globus pallidus appear to be more severely impaired in FTD compared to AD patients. These structures play an important role in frontostriatal circuits known to be affected in bvFTD and could explain the behavioral disturbances seen in this patient group, as illustrated by the relation between disinhibiton and nucleus accumbens volumes. DGM atrophy could lead to Wallerian degeneration of connecting fibers in the frontal circuits in FTD. This emphasizes the need to further elucidate the role of these frontal networks by linking the volumes of DGM structures to brain networks (i.e. cortical atrophy and white matter networks as measured by DTI) in FTD patients. Although we did not have pathological data available, the observed difference in volume of these basal nuclei supports the notion of greater involvement of these structures in FTD in contrast to AD and could help to explain and clarify some of the symptomatology in this multifaceted disorder.

Acknowledgements The VUmc Alzheimer Center is supported by Alzheimer Nederland and Stichting VUmc Fonds. The study was supported by the Netherlands Organisation for Scientific Research (NWO, national project ‘Brain and Cognition’ “Functionele Markers voor Cognitieve Stoornissen” (# 056-13-001)). Wiesje van der Flier is recipient of the Alzheimer Nederland grant (Influence of age on the endophenotype of AD on MRI, project number 2010-002).Research of the VUmc Alzheimer Center is part of the neurodegeneration research program of the Neuroscience Campus Amsterdam. The clinical database structure was developed with funding from Stichting Dioraphte. The gradient non-linearity correction was kindly provided by GE medical systems, Milwaukee.

84

͘͟ ͘Ǥ͘ ΰ ͘Ǥ͠

͛͞ ͘Ǥ͘ ΰ ͘Ǥ͠

Ǧ

͞͡ ͘Ǥ͘ ΰ ͘Ǥ͠

͘͟ ͘Ǥ͘ ΰ ͘Ǥ͠

͞͠ ͘Ǥ͘ ΰ ͘Ǥ͟

͚͟ ͚͠ ΰ ͚Ǥ͚

Ǧ

͘Ǥ͘͜͞

͘Ǥ͙͜͡

ȋ ͠͞Ȍ

ǡ

ǡ

͚​͚ Ǧ͙Ǥ͛ ΰ ͙Ǥ͛

͚͘ Ǧ͘Ǥ͡ ΰ ͚Ǥ͚

͙͡ Ǧ͙ ΰ ͘Ǥ͠

͚͛ Ǧ͙Ǥ͟ ΰ ͚Ǥ͚

ǡ

͚͜ ͚͜ ΰ ͜Ǥ͝

ǡ

γ͙͡͝Ǥ͝ ζ͘Ǥ͘​͙͘

γ͜​͘͜Ǥ͝ ͘Ǥ͛͘

γ͚͟Ǥ͙͡ ζ͘Ǥ͘​͙͘ γ͜͟͝Ǥ͘ ζ͘Ǥ͘​͙͘ γ͚​͚͟Ǥ͘ ζ͘Ǥ͘​͙͘ γ͙͝͡Ǥ͘ ͘Ǥ͛͘͠

γ͛͝Ǥ͛͡ ζ͘Ǥ͘​͙͘ γ͟͜͞Ǥ͘ ζ͘Ǥ͘​͙͘ γ͛͟͠Ǥ͝ ζ͘Ǥ͘​͙͘ γ͛͘͞Ǥ͘ ͘Ǥ͚͘͟

γ͛͡Ǥ͟͡ ζ͘Ǥ͘​͙͘ γ͙͛͠Ǥ͝ ζ͘Ǥ͘​͙͘ γ͛͘͜Ǥ͝ ͘Ǥ͙͚͘

γ͝͡Ǥ͘͝ ζ͘Ǥ͘​͙͘ γ͙͘͝Ǥ͝ ζ͘Ǥ͘​͙͘ γ͙͛͟Ǥ͝ ζ͘Ǥ͘​͙͘ γ͚͘͞Ǥ͘ ͘Ǥ͠͞͠

γ͛͡Ǥ͠͡ ζ͘Ǥ͘​͙͘ γ͚͘​͘Ǥ͘ ζ͘Ǥ͘​͙͘ γ͙͛​͙Ǥ͘ ζ͘Ǥ͘​͙͘ γ͚͜͜Ǥ͝ ͘Ǥ͘​͙͘

γ͠͞Ǥ͡͠ ζ͘Ǥ͘​͙͘ γ͛͘͡Ǥ͘ ζ͘Ǥ͘​͙͘ γ͚͛͜Ǥ͘ ζ͘Ǥ͘​͙͘ γ͚͝͞Ǥ͝ ͘Ǥ͙͘

͙͡ ͚͜ ȋ͙͡Ǧ͙​͙͚Ȍ γ͛͟Ǥ͟͝ ζ͘Ǥ͘​͙͘ γ͟͞͝Ǥ͘ ζ͘Ǥ͘​͙͘ γ͘͟͝Ǥ͘ ͘Ǥ͛͝​͛

͚͛ Ǧ͙Ǥ͛ ΰ ͙Ǥ͡

͟͞ Ǧ͚Ǥ͙ ΰ ͙Ǥ͟

γ͘Ǥ͙͡

͚͜ ͙͛͞͡ ΰ ͛͠Ǥ͚ γ͚͘Ǥ͙͟ ζ͘Ǥ͘​͙͘ ζ͘Ǥ͘​͙͘ ζ͘Ǥ͘​͙͘ ͘Ǥ͡​͡​͡ ͟͠͞ ȋ͙͞​͙Ǧ ͙͡ ͙͚͜͞Ȍ γ͞​͞Ǥ͟͠ ζ͘Ǥ͘​͙͘ γ͛͜͝Ǥ͝ ζ͘Ǥ͘​͙͘ γ͜͜͡Ǥ͝ ͘Ǥ͜​͚͜ γ͜͝Ǥ͘ ζ͘Ǥ͘​͙͘ ͛​͛͝ ȋ͙͘͞Ǧ ͙͡ ͘͞͠Ȍ γ͚͜Ǥ͟͞ ζ͘Ǥ͘​͙͘ γ͚͟͞Ǥ͘ ζ͘Ǥ͘​͙͘ γ͚͜͜Ǥ͘ ͘Ǥ͙​͙͚ γ͚͟͝Ǥ͘ ζ͘Ǥ͘​͙͘

͘͟ Ǧ͙Ǥ͠ ΰ ͙Ǥ͟

͞͝ Ǧ͚Ǥ͛ ΰ ͛Ǥ͚

͞͝ Ǧ͙Ǥ͛ ΰ ͙Ǥ͞

͙͟ Ǧ͛Ǥ͜ ΰ ͚Ǥ͜

͚͟ ͚͙ ΰ ͜Ǥ͡

͚͟ ͙͜͝͡ ΰ ͙͠Ǥ͛ ͚͟ ͙͛͟͠ΰ ͚͟Ǥ͡ ͚͠͝ ȋ͚͛͡Ǧ ͜​͜​͜Ǥ͝ ȋ͚͞͠Ǧ ͝͡ ͙͛͛͜Ȍ ͚͞ ͙͛͞͝Ȍ ͚͟͜ ȋ͞͡Ǧ ͟͝͞Ǥ͝ ȋ͙͜͠Ǧ ͝͡ ͚͘͠​͠Ȍ ͚͞ ͙͟​͟͡Ȍ ͚͠Ǥ͝ ȋ͛​͛Ǧ ͝͡ ͘͝ ȋ͙͠Ǧ͙͘͡Ȍ ͚͞ ͚͚͞Ȍ

Ⱦ͚͜

ȋ Ȍ

͛

͚͟ ͙Ǥ͚͡ ΰ ͘Ǥ͙

͘Ǥ͚͡​͚

͙Ǥ͘​͘​͘

ɖ͚γ͘Ǥ͘​͘ ͙Ǥ͘​͘​͘ γ͘Ǥ͘​͘

γ͙Ǥ͟͜

͚͟ ͙Ǥ͙͛ ΰ ͘Ǥ͙

͚͛ ͜͠Ǥ͝ ΰ ͘͜Ǥ͟

͚͜ ͙Ǥ͚͠ ΰ ͘Ǥ͚

͚͜ ͛͞ ΰ ͠Ǥ͘

͚͜ ͙͘ ȋ͙͜Ǥ͟άȌ

͚͟ ͛͞ ΰ ͠Ǥ͚

͚͟ ͛͘ ȋ͙͜Ǥ͟άȌ

γ͘Ǥ͙͞

ȋ Ȍ

͚͜ ͝Ǥ͜ ΰ ͙Ǥ͛

͚͟ ͛͘ ȋ͙͜Ǥ͟άȌ

͚͟ ͛͞ ΰ ͠Ǥ͙

ǡ

͘͟ ͝Ǥ͜ ΰ ͙Ǥ͝ ͙͟ ͝Ǥ͝ ΰ ͙Ǥ͙ ȋ Ȍ ͞​͞ ͘͜Ǥ͜ ΰ ͛͝Ǥ͡ ͞͡ ͘͜Ǥ͘ ΰ ͚͜Ǥ͠

͙Ǥ

͠5

͜͠ ͘Ǥ͚ ΰ ͘Ǥ͟ ͜͠ ͘Ǥ͡ ΰ ͚Ǥ͝ ͜͠ ͘Ǥ͞ ΰ ͙Ǥ͟

͙͛ ͟Ǥ͚ ΰ ͟Ǥ͠

͙͛ ͘Ǥ͘ ΰ ͘Ǥ͚

͙͛ ͘Ǥ͘ ΰ ͘Ǥ͘

͙͛ ͘Ǥ͜ ΰ ͙Ǥ͝

͙͛ ͘Ǥ͟ ΰ ͙Ǥ͡

͙͛ ͘Ǥ͡ ΰ ͚Ǥ͝

͙͛ ͘Ǥ͙ ΰ ͘Ǥ͟

͙͛ ͙Ǥ͛ ΰ ͙Ǥ͠

͙͛ ͘Ǥ͝ ΰ ͙Ǥ͜

Ȁ Ȁ

Ȁ Ȁ

ǡ

͙͜ ͙Ǥ͠ ΰ ͚Ǥ͟

͙͜ ͞Ǥ͞ ΰ ͜Ǥ͚

͙͜ ͙Ǥ͙ ΰ ͚Ǥ͞

͙͜ ͙Ǥ͟ ΰ ͛Ǥ͟

͙͜ ͘Ǥ͜ ΰ ͙Ǥ͞

͙͜ ͙Ǥ͜ ΰ ͛Ǥ͜

͙͜ ͘Ǥ͘ ΰ ͘Ǥ͘

͙͜ ͘Ǥ͛ ΰ ͙Ǥ͙

͘Ǥ͙͘͟

͘Ǥ͘͡͞

͘Ǥ͛͞͝

͘Ǥ͚͛͜

͘Ǥ͙͜​͜

͘Ǥ͛͞͠

γ͛͟Ǥ͝

γ͜Ǥ͙͟

͘Ǥ͙͚͜

ζ͘Ǥ͘​͙͘ γ͚͙͛Ǥ͘ ͘Ǥ͘͟͝

ζ͘Ǥ͘​͙͘ γ͚͛͘Ǥ͝ ͘Ǥ͚͘͞

γ͚͜Ǥ͚͜ ζ͘Ǥ͘​͙͘ γ͛͜͟Ǥ͘ ζ͘Ǥ͘​͙͘ γ͚͝Ǥ͝

γ͜Ǥ͜͟

γ͘Ǥ͚͘

γ͘Ǥ͙͡

γ͚Ǥ͚͞

γ͛Ǥ͟͠

γ͘Ǥ͘͡

͙͜ ͚͙Ǥ͞ ΰ ͙͜Ǥ͞ γ͙͞Ǥ͘͜ ζ͘Ǥ͘​͙͘ γ͜͞͝Ǥ͘ ͘ǡ͘​͘͝

Ȁ ͙͛ ͙Ǥ͠ ΰ ͚Ǥ͛ ͜͠ ͚Ǥ͚ ΰ ͛Ǥ͜ ͙͜ ͛Ǥ͟ ΰ ͛Ǥ͡ γ͚Ǥ͙͟ ͘Ǥ͛​͛͠ ǡ γ͡Ǥ͡͠ ͘Ǥ͘​͘͟ γ͞͠͞Ǥ͝ ͘Ǥ͚͡͝ γ͙͛͠Ǥ͝ ͘Ǥ͘​͘͜ γ͚͛​͛Ǥ͝ ͘Ǥ͙͘͝ ͙͛ ͘Ǥ͛ ΰ ͙Ǥ͙ ͜͠ ͘Ǥ͜ ΰ ͙Ǥ͘ ͙͜ ͙Ǥ͡ ΰ ͛Ǥ͛ Ǧ ͙͛ ͘Ǥ͠ ΰ ͙Ǥ͞ ͜͠ ͘Ǥ͟ ΰ ͙Ǥ͠ ͙͜ ͘Ǥ͝ ΰ ͙Ǥ͘ γ͘Ǥ͜​͜ ͘Ǥ͙͘͠ Ȁ ǡ γ͜͠͝ ͘Ǥ͙͘͝ γ͟͞͝Ǥ͘ ͘Ǥ͙͛͝ γ͙͚͛Ǥ͘ ͘Ǥ͘​͘͞ γ͚​͚͡Ǥ͝ ͘Ǥ͚͘͝ ͙͛ ͘Ǥ͝ ΰ ͙Ǥ͚ ͜͠ ͘Ǥ͡ ΰ ͚Ǥ͛ ͙͜ ͚Ǥ͙ ΰ ͚Ǥ͜ ΰ ȋάȌǤ Ǧ Ǥ ͚ Ǧ ǡ Ǧ ɖ ǡ Ǥ Ǧ Ǧ Ǥ ǣ ǣ ǡ ǣ ǯ ǡ ǣ ǡ ǣ ǡ ǣ ǡ ǣ Ǧ Ǥ ȋ ζ͘Ǥ͘͝Ȍǡ ȋ ζ͘Ǥ͘͝Ȍ

͜͠ ͙Ǥ͙ ΰ ͚Ǥ͡

͜͠ ͜Ǥ͜ ΰ ͛Ǥ͡

͜͠ ͘Ǥ͞ ΰ ͙Ǥ͞

͜͠ ͘Ǥ͡ ΰ ͚Ǥ͙

͜͠ ͘Ǥ͙ ΰ ͘Ǥ͞

͜͠ ͙͛Ǥ͘ ΰ ͙͚Ǥ͙

͠6

͜Ǥ͟ ΰ ͘Ǥ͜

͜Ǥ͚ ΰ ͘Ǥ͛

ζ͘Ǥ͘​͙͘

͙​͙Ǥ͝ ΰ ͙Ǥ͚ ζ͘Ǥ͘​͙͘ ͘Ǥ͘͜

͘Ǥ͟͜

͘Ǥ͜͡

͘Ǥ͜͡

͘Ǥ͚͠

͙Ǥ͙͡

η

͙Ǥ͘​͘​͘

͘Ǥ͘​͙͘

͘Ǥ͙͘͝

͘Ǥ͘​͙͘

͘Ǥ͙͘

ζ͘Ǥ͘​͙͘

͘Ǥ͛͝

͘Ǥ͙͡

͙Ǥ͝͠

͙Ǥ͛͡

͘Ǥ͜͠

͙Ǥ͙͠

η

ζ͘Ǥ͘​͙͘

͘Ǥ͘​͘͜

ζ͘Ǥ͘​͙͘

ζ͘Ǥ͘​͙͘

͘Ǥ͘​͚͘

ζ͘Ǥ͘​͙͘

͘Ǥ͜͡

͘Ǥ͙͟

͙Ǥ͘͡

͘Ǥ͜͝

͘Ǥ͙͡

͘Ǥ͚͞

η

ζ͘Ǥ͘​͙͘

͙Ǥ͘​͘​͘

ζ͘Ǥ͘​͙͘

͘Ǥ͝͡͞

͘Ǥ͜͞

͘Ǥ͘͡͠

͙Ǥ͚ ΰ ͘Ǥ͛ ͙Ǥ͘ ΰ ͘Ǥ͚ ͘Ǥ͟ ΰ ͘Ǥ͛ ζ͘Ǥ͘​͙͘ ͘Ǥ͙͟ ζ͘Ǥ͘​͙͘ ͘Ǥ͜ ζ͘Ǥ͘​͙͘ ͘Ǥ͚͛ ζ͘Ǥ͘​͙͘ ͛ ΰ ǡ Ǥ ǡ Ǥ Ǥ ǣ ǣ ǡ ǣ ǯ ǡ ǣ Ǥ

͜Ǥ͟ ΰ ͘Ǥ͞

͙​͙Ǥ͟ ΰ ͙Ǥ͛

ζ͘Ǥ͘​͙͘

͙͚Ǥ͜ ΰ ͙Ǥ͙

͟Ǥ͝ ΰ ͙Ǥ͚

͠Ǥ͝ ΰ ͙Ǥ͙

͘Ǥ͘​͙͘

͡Ǥ͘ ΰ ͘Ǥ͡

͛Ǥ͝ ΰ ͘Ǥ͞

͛Ǥ͟ ΰ ͘Ǥ͞

͛Ǥ͡ ΰ ͘Ǥ͝

ζ͘Ǥ͘​͙͘

͙͡Ǥ͠ ΰ ͙Ǥ͞ ͙͠Ǥ͡ ΰ ͙Ǥ͞ ͙͠Ǥ͝ ΰ ͙Ǥ͝ ζ͘Ǥ͘​͙͘

͠Ǥ͚ ΰ ͙Ǥ͛

͠Ǥ͟ ΰ ͙Ǥ͛

͡Ǥ͡ ΰ ͙Ǥ͙

͛ ͚Ǥ ȋ Ȍ ǡ ǡ Ǥ

͠7

Figure 1. Examples of automated segmentation of deep gray matter structures in the three diagnostic groups using the algorithm FIRST. Segmentations are shown overlaid on the 3D T1weighted images in three orthogonal orientations, corresponding roughly to the axial (left column), coronal (middle column) and sagittal (right column) planes. Top row shows results for a subject with subjective complaints. Second row shows results for an AD patient. Third row shows results for a FTD patients. Last row shows a sagittal view of the nucleus accumbens and surrounding structures. Colored structures: Yellow: Hippocampus; Turquoise: Amygdala; Green: Thalamus; Light blue: Caudate Nucleus; Pink: Putamen; Dark blue: Globus Pallidus; Orange: Nucleus Accumbens. SC

FTD

88

3

Figure 2. Boxplots of volumes (cm ) of hippocampus, amygdala, and deep gray matter structures for each diagnostic group adjusted for head size.

SC FTD

**

*

p≤0.001, p<0.05

89

͞Ǥ͚ ΰ ͘Ǥ͞

͚Ǥ͛ ΰ ͘Ǥ͚

͚Ǥ͜ ΰ ͘Ǥ͚

͘Ǥ͞ ΰ ͘Ǥ͚

Ǥ

͙Ǥ͠ ΰ ͘Ǥ͜

͙Ǥ͟ ΰ ͘Ǥ͛

͜Ǥ͚ ΰ ͘Ǥ͠

͛Ǥ͟ ΰ ͘Ǥ͞

͘Ǥ͝ ΰ ͘Ǥ͙

͚Ǥ͛ ΰ ͘Ǥ͛

͚Ǥ͛ ΰ ͘Ǥ͛

͝Ǥ͡ ΰ ͘Ǥ͟

͝Ǥ͠ ΰ ͘Ǥ͟

͘Ǥ͜ ΰ ͘Ǥ͚

͚Ǥ͙ ΰ ͘Ǥ͚

͚Ǥ͙ ΰ ͘Ǥ͚

͝Ǥ͠ ΰ ͘Ǥ͠

͝Ǥ͟ ΰ ͘Ǥ͞

͜Ǥ͜ ΰ ͘Ǥ͞ ͛Ǥ͡ ΰ ͘Ǥ͞

͜Ǥ͚ ΰ ͘Ǥ͝

͡Ǥ͜ ΰ ͘Ǥ͠ ͡Ǥ͙ ΰ ͘Ǥ͠

͡Ǥ͝ ΰ ͘Ǥ͠ ͡Ǥ͜ ΰ ͘Ǥ͠

͙Ǥ͠ ΰ ͘Ǥ͜

͙Ǥ͠ ΰ ͘Ǥ͛

͜Ǥ͝ ΰ ͘Ǥ͟

͜Ǥ͚ ΰ ͘Ǥ͠ ͜Ǥ͘ ΰ ͘Ǥ͟

ζ͘Ǥ͘​͙͘

ζ͘Ǥ͘​͙͘

ζ͘Ǥ͘​͙͘

͘Ǥ͘​͘͞

ζ͘Ǥ͘​͙͘

ζ͘Ǥ͘​͙͘

ζ͘Ǥ͘​͙͘

ζ͘Ǥ͘​͙͘

ζ͘Ǥ͘​͙͘

͘Ǥ͘​͘͜

͘Ǥ͘​͛͘

ζ͘Ǥ͘​͙͘

ζ͘Ǥ͘​͙͘

η

͘Ǥ͘͡

͘Ǥ͘͜

͘

͘Ǥ͚͟

͘Ǥ͜͞

͘Ǥ͚͛

͘Ǥ͚͞

͘Ǥ͜͟

͘Ǥ͜͟

͘Ǥ͙͞

͘Ǥ͙͚

͘Ǥ͜͝

͘Ǥ͞͝

͘Ǥ͘​͚͘

͙

͙

͘Ǥ͛͘͠

ζ͘Ǥ͘​͙͘

͘Ǥ͘͜͞

͘Ǥ͘​͘͟

͘Ǥ͘​͙͘

͘Ǥ͘​͙͘

͘Ǥ͚͙͘

͘Ǥ͘͝͞

ζ͘Ǥ͘​͙͘

ζ͘Ǥ͘​͙͘

η

͘Ǥ͚͙

͘Ǥ͚͟

͘Ǥ͚͟

͘Ǥ͛͜

͘Ǥ͜͠

͘Ǥ͟​͟

͘Ǥ͙͠

͘Ǥ͟͝

͘Ǥ͜͞

͘Ǥ͚͜

͘Ǥ͚͛

͘Ǥ͚͡

͘Ǥ͠͡

ζ͘Ǥ͘​͙͘

ζ͘Ǥ͘​͙͘

ζ͘Ǥ͘​͙͘

͘Ǥ͙͘͟

͘Ǥ͘​͘͞

ζ͘Ǥ͘​͙͘

ζ͘Ǥ͘​͙͘

ζ͘Ǥ͘​͙͘

͘Ǥ͘​͚͘

͘Ǥ͙͛͘

͘Ǥ͘​͘͜

ζ͘Ǥ͘​͙͘

ζ͘Ǥ͘​͙͘

η

͘Ǥ͙͛

͘Ǥ͚͛

͘Ǥ͚͞

͘Ǥ͙͝

͘Ǥ͚͘

͘Ǥ͜͝

͘Ǥ͝​͝

͘Ǥ͚͠

͘Ǥ͙͟

͘Ǥ͘͠

͘Ǥ͙​͙

͘Ǥ͛͠

͘Ǥ͚͛

͘Ǥ͘​͙͘

͘Ǥ͘​͙͘

ζ͘Ǥ͘​͙͘

͘Ǥ͙͡͠

͙

ζ͘Ǥ͘​͙͘

ζ͘Ǥ͘​͙͘

͘Ǥ͛​͛͜

͙

͙

͘Ǥ͛͝​͝

͘Ǥ͘͜͝

͘Ǥ͛͡​͡

Ǥ ͘Ǥ͝ ΰ ͘Ǥ͙ ͘Ǥ͜ ΰ ͘Ǥ͙ ͘Ǥ͛ ΰ ͘Ǥ͙ ζ͘Ǥ͘​͙͘ ͘Ǥ͘͠ ζ͘Ǥ͘​͙͘ ͘Ǥ͙͡ ζ͘Ǥ͘​͙͘ ͘Ǥ͙​͙ ͘Ǥ͘​͙͘ ͛ ΰ ǡ Ǥ ǡ Ǥ Ǥ ǣ ǣ ǡ ǣ ǯ ǡ ǣ Ǥ

͜Ǥ͞ ΰ ͘Ǥ͝

͞Ǥ͚ ΰ ͘Ǥ͞

͜Ǥ͜ ΰ ͘Ǥ͜

Ǥ

͡Ǥ͠ ΰ ͘Ǥ͠

Ǥ

͚Ǥ͘ ΰ ͘Ǥ͛

͙͘Ǥ͘ ΰ ͘Ǥ͠

͙Ǥ͡ ΰ ͘Ǥ͛

͝Ǥ͘ ΰ ͘Ǥ͞

͜Ǥ͠ ΰ ͘Ǥ͞

͛ ȋ Ȍ ǡ ǡ Ǥ

͡0

Reference List [1]

[2]

[3] [4]

[5]

[6]

[7]

[8]

[9]

[10]

[11] [12]

[13]

[14]

Garre-Olmo J, Genis BD, del Mar FM, Marquez DF, de Eugenio HR, Casadevall T, Turbau RJ, Turon EA, Lopez-Pousa S (2010) Incidence and subtypes of early-onset dementia in a geographically defined general population. Neurology 75, 12491255. Harvey RJ, Skelton-Robinson M, Rossor MN (2003) The prevalence and causes of dementia in people under the age of 65 years. J Neurol Neurosurg Psychiatry 74, 1206-1209. Frisoni GB, Fox NC, Jack CR, Jr., Scheltens P, Thompson PM (2010) The clinical use of structural MRI in Alzheimer disease. Nat Rev Neurol 6, 67-77. Bakkour A, Morris JC, Wolk DA, Dickerson BC (2013) The effects of aging and Alzheimer's disease on cerebral cortical anatomy: specificity and differential relationships with cognition. Neuroimage 76, 332-344. Rabinovici GD, Seeley WW, Kim EJ, Gorno-Tempini ML, Rascovsky K, Pagliaro TA, Allison SC, Halabi C, Kramer JH, Johnson JK, Weiner MW, Forman MS, Trojanowski JQ, Dearmond SJ, Miller BL, Rosen HJ (2007) Distinct MRI atrophy patterns in autopsy-proven Alzheimer's disease and frontotemporal lobar degeneration. Am J Alzheimers Dis Other Demen 22, 474-488. van de Pol LA, Hensel A, van der Flier WM, Visser PJ, Pijnenburg YA, Barkhof F, Gertz HJ, Scheltens P (2006) Hippocampal atrophy on MRI in frontotemporal lobar degeneration and Alzheimer's disease. J Neurol Neurosurg Psychiatry 77, 439442. Whitwell JL, Josephs KA, Rossor MN, Stevens JM, Revesz T, Holton JL, Al-Sarraj S, Godbolt AK, Fox NC, Warren JD (2005) Magnetic resonance imaging signatures of tissue pathology in frontotemporal dementia. Arch Neurol 62, 1402-1408. Rosen HJ, Gorno-Tempini ML, Goldman WP, Perry RJ, Schuff N, Weiner M, Feiwell R, Kramer JH, Miller BL (2002) Patterns of brain atrophy in frontotemporal dementia and semantic dementia. Neurology 58, 198-208. Hornberger M, Yew B, Gilardoni S, Mioshi E, Gleichgerrcht E, Manes F, Hodges JR (2012) Ventromedial-frontopolar prefrontal cortex atrophy correlates with insight loss in frontotemporal dementia and Alzheimer's disease. Hum Brain Mapp Galton CJ, Patterson K, Xuereb JH, Hodges JR (2000) Atypical and typical presentations of Alzheimer's disease: a clinical, neuropsychological, neuroimaging and pathological study of 13 cases. Brain 123 Pt 3, 484-498. Johnson JK, Head E, Kim R, Starr A, Cotman CW (1999) Clinical and pathological evidence for a frontal variant of Alzheimer disease. Arch Neurol 56, 1233-1239. Cherubini A, Peran P, Spoletini I, Di PM, Di IF, Hagberg GE, Sancesario G, Gianni W, Bossu P, Caltagirone C, Sabatini U, Spalletta G (2010) Combined volumetry and DTI in subcortical structures of mild cognitive impairment and Alzheimer's disease patients. J Alzheimers Dis 19, 1273-1282. de Jong LW, van der Hiele K, Veer IM, Houwing JJ, Westendorp RG, Bollen EL, de Bruin PW, Middelkoop HA, van Buchem MA, van der Grond J (2008) Strongly reduced volumes of putamen and thalamus in Alzheimer's disease: an MRI study. Brain 131, 3277-3285. de Jong LW, Ferrarini L, van der Grond J, Milles JR, Reiber JH, Westendorp RG, Bollen EL, Middelkoop HA, van Buchem MA (2011) Shape abnormalities of the striatum in Alzheimer's disease. J Alzheimers Dis 23, 49-59.

91

[15]

[16]

[17]

[18]

[19]

[20] [21]

[22]

[23]

[24]

[25]

[26] [27]

[28]

92

Roh JH, Qiu A, Seo SW, Soon HW, Kim JH, Kim GH, Kim MJ, Lee JM, Na DL (2011) Volume reduction in subcortical regions according to severity of Alzheimer's disease. J Neurol 258, 1013-1020. Chow TW, Izenberg A, Binns MA, Freedman M, Stuss DT, Scott CJ, Ramirez J, Black SE (2008) Magnetic resonance imaging in frontotemporal dementia shows subcortical atrophy. Dement Geriatr Cogn Disord 26, 79-88. Looi JC, Svensson L, Lindberg O, Zandbelt BB, Ostberg P, Orndahl E, Wahlund LO (2009) Putaminal volume in frontotemporal lobar degeneration and Alzheimer disease: differential volumes in dementia subtypes and controls. AJNR Am J Neuroradiol 30, 1552-1560. Garibotto V, Borroni B, Agosti C, Premi E, Alberici A, Eickhoff SB, Brambati SM, Bellelli G, Gasparotti R, Perani D, Padovani A (2011) Subcortical and deep cortical atrophy in Frontotemporal Lobar Degeneration. Neurobiol Aging 32, 875-884. Looi JC, Walterfang M, Styner M, Svensson L, Lindberg O, Ostberg P, Botes L, Orndahl E, Chua P, Kumar R, Velakoulis D, Wahlund LO (2010) Shape analysis of the neostriatum in frontotemporal lobar degeneration, Alzheimer's disease, and controls. Neuroimage 51, 970-986. Cummings JL (1993) Frontal-subcortical circuits and human behavior. Arch Neurol 50, 873-880. Looi JC, Walterfang M, Styner M, Niethammer M, Svensson LA, Lindberg O, Ostberg P, Botes L, Orndahl E, Chua P, Velakoulis D, Wahlund LO (2011) Shape analysis of the neostriatum in subtypes of frontotemporal lobar degeneration: neuroanatomically significant regional morphologic change. Psychiatry Res 191, 98-111. van der Flier WM, Pijnenburg YA, Prins N, Lemstra AW, Bouwman FH, Teunissen CE, van Berckel BN, Stam CJ, Barkhof F, Visser PJ, van EE, Scheltens P (2014) Optimizing patient care and research: the Amsterdam Dementia Cohort. J Alzheimers Dis 41, 313-327. McKhann G, Drachman D, Folstein M, Katzman R, Price D, Stadlan EM (1984) Clinical diagnosis of Alzheimer's disease: report of the NINCDS-ADRDA Work Group under the auspices of Department of Health and Human Services Task Force on Alzheimer's Disease. Neurology 34, 939-944. McKhann GM, Knopman DS, Chertkow H, Hyman BT, Jack CR, Jr., Kawas CH, Klunk WE, Koroshetz WJ, Manly JJ, Mayeux R, Mohs RC, Morris JC, Rossor MN, Scheltens P, Carrillo MC, Thies B, Weintraub S, Phelps CH (2011) The diagnosis of dementia due to Alzheimer's disease: recommendations from the National Institute on Aging-Alzheimer's Association workgroups on diagnostic guidelines for Alzheimer's disease. Alzheimers Dement 7, 263-269. Neary D, Snowden JS, Gustafson L, Passant U, Stuss D, Black S, Freedman M, Kertesz A, Robert PH, Albert M, Boone K, Miller BL, Cummings J, Benson DF (1998) Frontotemporal lobar degeneration: a consensus on clinical diagnostic criteria. Neurology 51, 1546-1554. Verhage F (1964) Intelligentie en leeftijd: onderzoek bij Nederlanders van twaalf tot zevenenzeventig jaar., Van Gorcum, Assen. Mulder C, Verwey NA, van der Flier WM, Bouwman FH, Kok A, van Elk EJ, Scheltens P, Blankenstein MA (2010) Amyloid-beta(1-42), total tau, and phosphorylated tau as cerebrospinal fluid biomarkers for the diagnosis of Alzheimer disease. Clin Chem 56, 248-253. Patenaude B, Smith SM, Kennedy DN, Jenkinson M (2011) A Bayesian model of shape and appearance for subcortical brain segmentation. Neuroimage 56, 907-922.

[29]

[30]

[31]

[32]

[33]

[34] [35]

[36] [37]

[38]

[39] [40] [41]

[42]

[43]

[44]

Smith SM, Jenkinson M, Woolrich MW, Beckmann CF, Behrens TE, Johansen-Berg H, Bannister PR, De LM, Drobnjak I, Flitney DE, Niazy RK, Saunders J, Vickers J, Zhang Y, De SN, Brady JM, Matthews PM (2004) Advances in functional and structural MR image analysis and implementation as FSL. Neuroimage 23 Suppl 1, S208-S219. Babalola KO, Patenaude B, Aljabar P, Schnabel J, Kennedy D, Crum W, Smith S, Cootes T, Jenkinson M, Rueckert D (2009) An evaluation of four automatic methods of segmenting the subcortical structures in the brain. Neuroimage 47, 1435-1447. Morey RA, Selgrade ES, Wagner HR, Huettel SA, Wang L, McCarthy G (2010) Scanrescan reliability of subcortical brain volumes derived from automated segmentation. Hum Brain Mapp 31, 1751-1762. Smith SM, Zhang Y, Jenkinson M, Chen J, Matthews PM, Federico A, De SN (2002) Accurate, robust, and automated longitudinal and cross-sectional brain change analysis. Neuroimage 17, 479-489. Popescu V, Battaglini M, Hoogstrate WS, Verfaillie SC, Sluimer IC, van Schijndel RA, van Dijk BW, Cover KS, Knol DL, Jenkinson M, Barkhof F, De SN, Vrenken H (2012) Optimizing parameter choice for FSL-Brain Extraction Tool (BET) on 3D T1 images in multiple sclerosis. Neuroimage 61, 1484-1494. Stroop JR (1935) Studies of interference in serial verbal reactions. J Exp Psychol 18, 643662. Smits LL, Pijnenburg YA, Koedam EL, van der Vlies AE, Reuling IE, Koene T, Teunissen CE, Scheltens P, van der Flier WM (2012) Early Onset Alzheimer's Disease is Associated with a Distinct Neuropsychological Profile. J Alzheimers Dis 30, 101108. Cummings JL (1997) The Neuropsychiatric Inventory: assessing psychopathology in dementia patients. Neurology 48, S10-S16. van Harten AC, Kester MI, Visser PJ, Blankenstein MA, Pijnenburg YA, van der Flier WM, Scheltens P (2011) Tau and p-tau as CSF biomarkers in dementia: a metaanalysis. Clin Chem Lab Med 49, 353-366. Schoonenboom NS, Reesink FE, Verwey NA, Kester MI, Teunissen CE, van de Ven PM, Pijnenburg YA, Blankenstein MA, Rozemuller AJ, Scheltens P, van der Flier WM (2012) Cerebrospinal fluid markers for differential dementia diagnosis in a large memory clinic cohort. Neurology 78, 47-54. Braak H , Braak E (1991) Alzheimer's disease affects limbic nuclei of the thalamus. Acta Neuropathol 81, 261-268. Kril JJ, Macdonald V, Patel S, Png F, Halliday GM (2005) Distribution of brain atrophy in behavioral variant frontotemporal dementia. J Neurol Sci 232, 83-90. Halabi C, Halabi A, Dean DL, Wang PN, Boxer AL, Trojanowski JQ, Dearmond SJ, Miller BL, Kramer JH, Seeley WW (2013) Patterns of striatal degeneration in frontotemporal dementia. Alzheimer Dis Assoc Disord 27, 74-83. Looi JC, Lindberg O, Zandbelt BB, Ostberg P, Andersen C, Botes L, Svensson L, Wahlund LO (2008) Caudate nucleus volumes in frontotemporal lobar degeneration: differential atrophy in subtypes. AJNR Am J Neuroradiol 29, 1537-1543. Lantos PL, Cairns NJ, Khan MN, King A, Revesz T, Janssen JC, Morris H, Rossor MN (2002) Neuropathologic variation in frontotemporal dementia due to the intronic tau 10(+16) mutation. Neurology 58, 1169-1175. Kim EJ, Rabinovici GD, Seeley WW, Halabi C, Shu H, Weiner MW, Dearmond SJ, Trojanowski JQ, Gorno-Tempini ML, Miller BL, Rosen HJ (2007) Patterns of MRI atrophy in tau positive and ubiquitin positive frontotemporal lobar degeneration. J Neurol Neurosurg Psychiatry 78, 1375-1378. 93

[45]

[46]

[47]

[48]

[49] [50]

[51]

[52]

[53]

[54]

[55]

[56]

[57]

[58]

[59]

[60]

[61]

94

Rohrer JD, Ridgway GR, Modat M, Ourselin S, Mead S, Fox NC, Rossor MN, Warren JD (2010) Distinct profiles of brain atrophy in frontotemporal lobar degeneration caused by progranulin and tau mutations. Neuroimage 53, 1070-1076. Bede P, Elamin M, Byrne S, McLaughlin RL, Kenna K, Vajda A, Pender N, Bradley DG, Hardiman O (2013) Basal ganglia involvement in amyotrophic lateral sclerosis. Neurology 81, 2107-2115. Piolino P, Desgranges B, Belliard S, Matuszewski V, Lalevee C, de lS, V, Eustache F (2003) Autobiographical memory and autonoetic consciousness: triple dissociation in neurodegenerative diseases. Brain 126, 2203-2219. Kramer JH, Rosen HJ, Du AT, Schuff N, Hollnagel C, Weiner MW, Miller BL, Delis DC (2005) Dissociations in hippocampal and frontal contributions to episodic memory performance. Neuropsychology 19, 799-805. Pennington C, Hodges JR, Hornberger M (2011) Neural correlates of episodic memory in behavioral variant frontotemporal dementia. J Alzheimers Dis 24, 261-268. Soderlund H, Black SE, Miller BL, Freedman M, Levine B (2008) Episodic memory and regional atrophy in frontotemporal lobar degeneration. Neuropsychologia 46, 127-136. Simons JS, Verfaellie M, Galton CJ, Miller BL, Hodges JR, Graham KS (2002) Recollection-based memory in frontotemporal dementia: implications for theories of long-term memory. Brain 125, 2523-2536. Collette F, Van der Linden M, Salmon E (2010) Dissociation between controlled and automatic processes in the behavioral variant of fronto-temporal dementia. J Alzheimers Dis 22, 897-907. Graham A, Davies R, Xuereb J, Halliday G, Kril J, Creasey H, Graham K, Hodges J (2005) Pathologically proven frontotemporal dementia presenting with severe amnesia. Brain 128, 597-605. Perry DC, Whitwell JL, Boeve BF, Pankratz VS, Knopman DS, Petersen RC, Jack CR, Jr., Josephs KA (2012) Voxel-based morphometry in patients with obsessivecompulsive behaviors in behavioral variant frontotemporal dementia. Eur J Neurol 19, 911-917. Massimo L, Powers C, Moore P, Vesely L, Avants B, Gee J, Libon DJ, Grossman M (2009) Neuroanatomy of apathy and disinhibition in frontotemporal lobar degeneration. Dement Geriatr Cogn Disord 27, 96-104. Snowden JS, Bathgate D, Varma A, Blackshaw A, Gibbons ZC, Neary D (2001) Distinct behavioural profiles in frontotemporal dementia and semantic dementia. J Neurol Neurosurg Psychiatry 70, 323-332. Liu W, Miller BL, Kramer JH, Rankin K, Wyss-Coray C, Gearhart R, Phengrasamy L, Weiner M, Rosen HJ (2004) Behavioral disorders in the frontal and temporal variants of frontotemporal dementia. Neurology 62, 742-748. Zamboni G, Huey ED, Krueger F, Nichelli PF, Grafman J (2008) Apathy and disinhibition in frontotemporal dementia: Insights into their neural correlates. Neurology 71, 736-742. Rosen HJ, Allison SC, Schauer GF, Gorno-Tempini ML, Weiner MW, Miller BL (2005) Neuroanatomical correlates of behavioural disorders in dementia. Brain 128, 2612-2625. O'Callaghan C, Naismith SL, Hodges JR, Lewis SJ, Hornberger M (2013) Fronto-striatal atrophy correlates of inhibitory dysfunction in Parkinson's disease versus behavioural variant frontotemporal dementia. Cortex 49, 1833-1843. Aron AR (2011) From reactive to proactive and selective control: developing a richer model for stopping inappropriate responses. Biol Psychiatry 69, e55-e68.

[62] [63]

[64] [65]

[66]

[67]

[68] [69]

Dalley JW, Everitt BJ, Robbins TW (2011) Impulsivity, compulsivity, and top-down cognitive control. Neuron 69, 680-694. Ambroggi F, Ghazizadeh A, Nicola SM, Fields HL (2011) Roles of nucleus accumbens core and shell in incentive-cue responding and behavioral inhibition. J Neurosci 31, 6820-6830. Morgane PJ, Galler JR, Mokler DJ (2005) A review of systems and networks of the limbic forebrain/limbic midbrain. Prog Neurobiol 75, 143-160. Whitwell JL, Sampson EL, Loy CT, Warren JE, Rossor MN, Fox NC, Warren JD (2007) VBM signatures of abnormal eating behaviours in frontotemporal lobar degeneration. Neuroimage 35, 207-213. Woolley JD, Gorno-Tempini ML, Seeley WW, Rankin K, Lee SS, Matthews BR, Miller BL (2007) Binge eating is associated with right orbitofrontal-insular-striatal atrophy in frontotemporal dementia. Neurology 69, 1424-1433. Piguet O, Petersen A, Yin Ka LB, Gabery S, Murphy K, Hodges JR, Halliday GM (2011) Eating and hypothalamus changes in behavioral-variant frontotemporal dementia. Ann Neurol 69, 312-319. Dockray G (2004) Gut endocrine secretions and their relevance to satiety. Curr Opin Pharmacol 4, 557-560. Whitwell JL, Sampson EL, Watt HC, Harvey RJ, Rossor MN, Fox NC (2005) A volumetric magnetic resonance imaging study of the amygdala in frontotemporal lobar degeneration and Alzheimer's disease. Dement Geriatr Cogn Disord 20, 238244.

95

Chapter 3.2

Automatic classification of AD and bvFTD based on cortical atrophy for single-subject diagnosis 1

1

Christiane Möller MSc , Yolande A L Pijnenburg MD PhD , Wiesje M 1,2 3 1 van der Flier PhD , Adriaan Versteeg , Betty Tijms PhD , Jan C de 4 6-8 Munck PhD , Anne Hafkemeijer MSc , Serge A R B Rombouts 6-8 8 5,9 PhD , Jeroen van der Grond PhD , John van Swieten MD PhD , 1,6,9 1 Elise Dopper , Philip Scheltens MD PhD , Frederik Barkhof MD 3 3,4 3 PhD , Hugo Vrenken PhD , Alle Meije Wink PhD 1 2 Alzheimer Center & Department of Neurology, Department of 3 Epidemiology & Biostatistics, Department of Radiology & Nuclear 4 5 Medicine, Department of Physics & Medical Technology, Department of Clinical Genetics, Neuroscience Campus Amsterdam, VU University 6 medical center, Amsterdam, the Netherlands, Institute of Psychology, 7 Leiden University, Leiden Institute for Brain and Cognition, Leiden 8 University, Leiden, the Netherlands, Department of Radiology, Leiden 9 University Medical Center, Leiden, the Netherlands, Department of Neurology, Erasmus Medical Center, Rotterdam, the Netherlands. Under review

Manuscript type: original research Advances in Knowledge: • Automated classifiers are able to discriminate between scans of patients with different forms of dementia and controls, based on gray matter (GM) patterns with high accuracy (75.3%-85.4%). • Automated classifiers based on GM patterns can be used for single-subject diagnosis in an independent dataset with good to excellent diagnostic accuracy (AUC 0.81-0.95). • Automated classifiers based on a generally available structural T1-weighted scans, make automated single-subject diagnosis more accessible and easy to use in daily clinical routine. Implications for Patient Care: • Machine learning-based categorization methods could improve the diagnostic process in the daily practice, especially in centers without experienced neuroradiologists. • Machine learning-based categorization methods outperform classification based on a standard neuropsychological test battery. • The application of automatic classification may be used for screening purposes in the future for high-risk groups. Summary statement: Machine learning techniques are able to distinguish disease specific GM atrophy between AD and bvFTD in a standard T1-weighted structural MRI scan for single-subject diagnosis. 96

Abstract Purpose: Assessment of the diagnostic accuracy of a support vector machine (SVM) classifier for individual patients based on common T1-weighted gray matter (GM) images without extensive preprocessing, using two independent data sets. Materials and Methods: The local Institutional Review Board approved the study. 84 patients with Alzheimer’s disease (AD), 51 patients with behavioral variant frontotemporal dementia (bvFTD), and 94 control subjects were divided into independent training (n=115) and test sets (n=114) with identical distributions of diagnosis and scanner type. The training set was entered in a SVM for disease specific predictions of diagnostic status between two groups based on GM patterns. Weightvalues of each voxel for classification were extracted. We conducted discriminant function analyses to determine if the extracted weight-values could be used for singlesubject classification in the independent test-set. Receiver operating characteristic (ROC) curves and area under the curve (AUC) were calculated for MRI classification and neuropsychological z-scores. Threshold for statistical significance was p<0.05. Results: In the training set, the training accuracy of the SVM to discriminate AD from controls was 85.3%, bvFTD from controls 72.3%, and AD from bvFTD 78.7% (p≤0.029). For single-subject diagnosis in the test set, the extracted weight-values discriminated 87.6% of AD and controls, 84.7% of bvFTD and controls, and 82.1% of AD and bvFTD correctly. ROC curves revealed good to excellent AUC (0.81-0.95; p≤0.001). Machine learning-based categorization methods based on GM outperforms classification based on neuropsychological tests. Conclusions: SVMs can be used in single-subject discrimination and help the clinician in making a diagnosis. SVMs can distinguish disease-specific GM patterns in AD and bvFTD from those of normal aging using common T1-weighted structural MRI scans.

97

Introduction Alzheimer’s disease (AD) and behavioral variant frontotemporal dementia (bvFTD) are the most common causes of young onset dementia [1]. Clinical diagnostic criteria are available [2, 3], but the frequent overlap of the clinical symptoms associated with AD and bvFTD poses serious problems in the differential diagnosis. Various studies showed that cognitive tests are not accurate enough to discriminate AD from bvFTD, as bvFTD also show memory deficits and AD patients present with executive dysfunctions [4-9]. Magnetic resonance imaging (MRI), has been shown to be able to detect diseasespecific macroscopic brain changes in an early disease stage. Several studies investigated gray matter (GM) changes to discriminate AD from bvFTD. These studies however typically report group-level differences for various brain structures and not for the single patient [10-12]. Furthermore, it has been shown that atrophy patterns of AD and bvFTD largely overlap e.g. frontal atrophy is seen in AD and hippocampal atrophy does not exclude a diagnosis of bvFTD; it appears in normal aging as well [9, 10, 13-15]. Moreover, especially in mild stages of the disease, cortical atrophy may not be visible by eyeballing. Besides structural MRI, other imaging modalities, as positron emission tomography, functional or diffusion MRI, have been promising in the discrimination between AD and bvFTD. However, some of these techniques are invasive, time-consuming, require the availability of specialized scanners and are difficult to implement in a clinical setting without technical support [16, 17]. Therefore, a generally available, easy-touse analytical imaging method that detects and quantifies more subtle changes of the brain at the single-subject level is needed to support the discrimination between AD and bvFTD. In most memory clinics, patients are usually scanned once during dementia screening, with a standardized protocol generally including a T1-weighted 3-dimensional (3DT1) MRI sequence. This sequence is representative of the disease-specific structural changes, capable of providing similar information irrespective of different scanners and is easy to obtain. An automatic individual patient classification based on a structural 3DT1 scan at one time point could support the clinician’s diagnosis. A support vector machine (SVM) is a machine learning technique that can categorize individual brain images by differentiation of images from two groups. These automated classifiers can be objective, quantitative and easy to implement and potentially satisfy the requirements of a diagnostic tool [18]. The available literature on SVM shares common limitations: Discriminating AD from FTD using the whole spectrum of FTD and not only the behavioral variant [16, 19-21] will be only representative for the language variants as their asymmetric atrophy will determine the classification accuracy [22]. Only using cross-validation of MR scans already known to the classifier in the discrimination of patients may result in biased estimates, especially when the sample size is small [20, 23, 24]. Therefore, we explore the diagnostic accuracy of a SVM for individual patients whose MR images belong to a test sample independent from the sample to train the classifier. The classification is based on a generally available T1-weighted GM image without extensive preprocessing. To increase generalizability we used MRI scans from different scanners and two independent data sets in a cross-sectional design. 98

Materials and Methods Patients In this study, we included 84 patients with AD, 51 patients with bvFTD, and 53 patients with subjective memory complaints who visited either the Alzheimer Center of the VU University medical center or the Alzheimer Center of the Erasmus University Medical Center Rotterdam. All patients underwent a standardized one-day assessment including medical history, informant-based history, physical and neurological examination, blood tests, neuropsychological assessment, and MRI of the brain. Diagnoses were made in a multidisciplinary consensus meeting according to the core clinical criteria of the National Institute on Aging and the Alzheimer’s Association workgroup for probable AD [3, 25] and according to the International FTD Consortium criteria for bvFTD [2] based on the results of the one-day assessment as described above. Patients were diagnosed as having subjective memory complaints when they presented with memory complaints, but cognitive functioning was normal and criteria for MCI [26], dementia or any other neurological or psychiatric disorder known to cause cognitive decline were not met. To minimize center effects, all diagnoses were re-evaluated in a panel including clinicians from both centers. In addition, we included 41 cognitively normal controls, who were recruited by advertisements in local newspapers. They underwent an assessment including medical history, physical examination, neuropsychological assessment, and MRI of the brain comparable to the work-up of patients. Cognitively normal controls and patients with subjective memory complaints served both as controls to obtain a representative control group for the general population. Patients and controls were randomly split into equally sized independent training (n=115) and test sets (n=114) with identical distribution of diagnosis, age, sex and scanner type. Level of education was rated on a seven-point scale [27]. Disease duration was calculated based on the time difference between date of diagnosis and the year patients caregivers noticed the first symptoms. The local medical ethics committee of both centers approved the study. All patients and controls gave written informed consent for their clinical data to be used for research purposes. Neuropsychological assessment To assess dementia severity we used the Mini-Mental State Examination (MMSE) [28] and to assess frontal lobe dysfunction we used the frontal assessment battery (FAB) [29]. Cognitive functioning was assessed using a standardized neuropsychological test battery covering the or domains memory (immediate recall and delayed recall of Dutch version of the Rey Auditory Verbal Learning Test and total score of Visual Association Test A), language (Visual Association Test picture naming and category fluency (animals: 1 min)), attention (Trail Making Test part A (TMT A), Digit Span forward, and Letter Digit Substitution Test (LDST)), and executive functioning (Digit Span backwards, Trail Making Test part B (TMT B), letter fluency, and Stroop ColorWord card subtask [30]). For a detailed description of neuropsychological tests see Smits et al. [31]. For each cognitive task, z-scores were calculated from the raw test scores by the formula z=(x-μ)/σ, where μ is the mean and σ is the standard deviation of the performance of the control group of the whole dataset. The value z = 0 therefore reflects the average baseline test performance of the controls in a given 99

domain. Scores of TMT A, TMT B, and Stroop color-word card were inverted by computing (-1)*z-score, because higher scores imply a worse performance. Next, composite z-scores were calculated for each cognitive domain by averaging z-scores with the MEAN function in SPSS. Composite z-scores were calculated when at least one neuropsychological task was available in each cognitive domain. MR image acquisition and review Imaging at the VUmc was carried out on two 3T scanners (Signa HDxt, GE Healthcare, Milwaukee, WI, USA and TF PET/MR, Philips Medical Systems, Cleveland, Ohio, USA), using an 8-channel head coil with foam padding. Patients and controls from the Erasmus University Medical Center Rotterdam were all scanned at the Leiden University Medical Center (LUMC) on a 3T scanner (Achieva, Philips Medical Systems, Best, the Netherlands) using an 8-channel head coil. The scan protocol included a whole-brain near-isotropic 3DT1-weighted sequence. At the VUmc this was a fast spoiled gradient echo sequence (FSPGR; repetition time TR 7.8 ms, echo time TE 3 ms, inversion time TI 450 ms, flip angle 12º, 180 sagittal slices, voxel size 0.98x0.98x1 mm, total scan time 4.57 minutes) or a turbo field echo sequence (T1TFE; TR 7 ms, TE 3 ms, flip angle 12°, 180 sagittal slices, voxel size 1x1x1 mm, total scan time 6.14 minutes). At the LUMC this was a turbo field echo sequence (T1TFE; TR 9.8 ms, TE 4.6 ms, flip angle 8°, 140 transversal slices, voxel size 0.88x0.88x1.2 mm, total scan time 4.57 minutes). In addition, the MRI protocols included a 3D Fluid Attenuated Inversion Recovery (FLAIR) sequence, dual-echo T2-weighted sequence, and susceptibility weighted imaging (SWI) which were reviewed for brain pathology other than atrophy by an experienced radiologist. MR processing DICOM images of the 3DT1- weighted sequence from the Signa HDxt were corrected for gradient nonlinearity distortions. All scans were converted to Nifti format. The linear transformation matrix to MNI space was calculated using FSL-FLIRT [32] and used to place the image coordinate origin (0,0,0) on the anterior commissure by using the Nifti s-form. The structural 3DT1 images were then analyzed using the voxelbased morphometry toolbox (VBM8; version 435; University of Jena, Department of Psychiatry) in Statistical Parametric Mapping (SPM8; Functional Imaging Laboratory, University College London, London, UK) implemented in MATLAB 7.12 (MathWorks, Natick, MA). Data of the training- and test set were preprocessed separately with VBM8 to avoid any bias [33]. The first module of the VBM8 Toolbox (“Estimate and Write”) segments the 3DT1 volumes into GM, white matter (WM) and cerebrospinal fluid (CSF), applies a registration to MNI space (affine) and subsequently a non-linear deformation. The non-linear deformation parameters are calculated via the high dimensional DARTEL algorithm and the MNI 152 template. Remaining non-brain tissue was removed by the ‘light clean-up’ option. Tissue classes were normalized in alignment with the template with the ‘non-linear only’ option which allows comparing the absolute amount of tissue corrected for individual brain size. The correction is applied directly to the data, which makes a head-size correction to the statistical model redundant. In the second module, images were smoothed using an eight mm 100

full width at half maximum (FWHM) isotropic Gaussian kernel. All images were visually inspected after every processing step. Support vector machine – pattern recognition For the pattern recognition analysis we used the “Pattern Recognition for Neuroimaging Toolbox” (PRoNTo) [33], implemented in MATLAB 7.12 following the standard descriptions of the manual (http://www.mlnl.cs.ucl.ac.uk/pronto). As a first step – the training step – classification of the patients in the training set was done by using PRoNTo. Only the training set was used in PRoNTo to learn the patterns of the MRI scans of the 3 groups. The normalized, modulated, smoothed GM images of all subjects in the training-set were used as inputs. We used a custom mask, which was made by the mean of all smoothed GM segmentations and binarized at a threshold of 0.2. We used a binary SVM to classify (1) AD from controls, (2) bvFTD from controls, and (3) AD from bvFTD, all with leave-one out cross-validation. To estimate how much each voxel contributes to the classification task, we calculated the voxel-wise ‘discrimination maps’ [33]. The weights are the model parameters learned by the SVM projected back to the input space. The weight-value can be positive or negative, where a positive value represents a higher weighted average for class one, a negative value represents a higher weighted average for class two. These maps are shown for each classifier in Figure 1. For illustrative purposes the weight maps are thresholded at 30% of the maximum positive and negative weight values, in line with MouraoMiranda [34]. The performance of the classifier in the training set was expressed in balanced accuracy (class-specific accuracy), sensitivity and specificity. Permutation testing was used to derive a p-value to determine whether the balanced accuracy exceeded chance levels (50%). Class labels were permuted 1000 times. Support vector machine – classification of new subjects As a second step – the test step –, we used the discrimination maps computed in PRoNTo on a new independent set of patients. To test whether the learned weight values from the training set could classify unseen subjects, we extracted the weights of every voxel of the weight map over all folds. These weights were then multiplied with the normalized, modulated, smoothed GM images of each subject from the independent test set. This integrated product of the weight map and the smoothed GM image per subject was averaged and transferred to SPSS for further analyses. This was done separately for the classification between AD-controls, bvFTD-controls, and AD-bvFTD. Statistical analyses SPSS version 20.0 for Windows was used for statistical analysis. Differences between groups were assessed using univariate analysis of variance (ANOVA), Kruskal-Wallis tests and χ2 tests, where appropriate. Composite cognitive domain z-scores were compared using multivariate analysis of variance (MANOVA) with Bonferroni posthoc tests and age, sex, educational level and disease duration as covariates. To determine the discriminative power of the neuropsychological examination in the test set, we conducted a stepwise discriminant function analysis between AD and bvFTD wit leave-one out cross-validation. As predictors we entered the four z-domains memory, 101

language, attention and executive functioning. Additionally, we created a receiver operation characteristic (ROC) and calculated the area under the curve (AUC). To determine the performance of the SVM for single-subject classification in the independent test set, we conducted three discriminant function analyses between two groups with leave-one-out cross validation. As predictor we entered the averaged integrated product of the weight map and the smoothed GM image per subject from the corresponding classification of the SVM (e.g. for the discriminant analysis between AD and controls, the average integrated product from weight map “AD vs. controls� was used for prediction). Additionally, we created a receiver operation characteristic (ROC) and calculated the area under the curve (AUC). Statistical significance was set at p<0.05.

Results Demographics Demographic data for training and test set are summarized in Table 1. As the two sets were split based on age, sex, scanner type and diagnosis, the two sets did not differ in these variables. Furthermore, there were no differences in disease duration and MMSE. Groups based on diagnosis within training and test set did not differ in level of education, sex, scanner type and disease duration (Table 2). AD and bvFTD patients had lower scores on the FAB compared to controls. AD patients were older than controls and had the lowest MMSE scores. The MMSE scores of both patient groups showed that patients were in a mild stadium of the disease. In both datasets, AD patients performed worse on memory compared to bvFTD and controls, and both patient groups had lower scores on the language domain compared to controls. In the training set, AD and bvFTD patients performed worse on attention and executive functioning compared to controls, whereas in the test set only AD patients performed worse than controls on attention and had the worst scores in executive functioning compared to bvFTD and controls. In the test set the four composite cognitive z-domains discriminated 81.0% of AD patients and controls correctly, with correct classification of 23 AD patients (62.2%) and 45 controls (95.7%). Memory had the highest loading. The ROC curve revealed an excellent AUC for the memory domain (0.95; p<0.001). For the discrimination between bvFTD and controls, the four composite cognitive z-domains discriminated 80.6% of all cases correctly, with correct classification of 15 bvFTD patients (60.0%) and 43 controls (91.5%). Language had the highest loading. The ROC curve revealed an excellent AUC for the language domain (0.92; p<0.001). For the discrimination between AD and bvFTD patients, the four composite cognitive z-domains discriminated 65.7% of all cases correctly, with correct classification of 33 AD patients (78.6%) and 11 bvFTD patients (44.0%). Memory had the highest loading. The ROC curve revealed a fair AUC for the memory domain (0.74; p=0.002).

102

Support vector classification Training accuracy Performance of the binary SVM is summarized in Figure 2. The classifier discriminated AD patients from controls with 85.3% balanced accuracy (p=0.001). Sensitivity for classification of AD patients was 83.3% (p=0.001) and specificity 87.2% (p=0.001). A correct distinction between bvFTD patients and controls, was made in 72.3% (balanced accuracy, p=0.001). Sensitivity for classification of bvFTD patients was 61.5% (p=0.001) and specificity 83% (p=0.029). The classifier discriminated AD from bvFTD with 78.7% balanced accuracy (p=0.001). Sensitivity for classification of AD patients was 88.1% (p=0.001) and specificity was 69.2% (p=0.001). Generalizability of the classifiers for single-subject diagnosis Results are summarized in Figure 3. In the independent test set the extracted weights discriminated 87.6% of AD patients and controls correctly, with correct classification of 36 AD patients (85.7%) and 42 controls (89.4%). The ROC curve revealed an excellent AUC for the extracted weights (0.95; p<0.001). For the discrimination between bvFTD patients and controls, the extracted weights predicted 84.7% of all cases correctly, with correct classification of 15 bvFTD patients (60%) and 46 controls (97.9%). The ROC curve revealed a good AUC for the extracted weights (0.87; p<0.001). For the discrimination between AD and bvFTD patients, the extracted SVM weights predicted 82.1% of all cases correctly, with correct classification of 39 AD (92.9%) and 16 bvFTD patients (64%). The ROC curve revealed a good AUC for the extracted weights (0.81; p<0.001).

Discussion In this study we showed that it is possible to discriminate between patients with different forms of mild dementia and between dementia and controls, based on GM patterns with an automated classifier with high accuracy (75.3%-85.4%). Crucially, we have also demonstrated that automated classifiers can be used in single-subject diagnosis, as its diagnostic accuracy in an independent dataset was good to excellent (AUC 0.81-0.95). The accuracy levels in our study are comparable with other studies using SVM in differentiation of AD and FTLD [17-20, 35]. The slightly higher accuracy values found in two other studies can be explained by the use of all subtypes of FTLD. It is possible that the specific atrophy patterns of the language variant of the FTLD spectrum increased the diagnostic power [18, 20]. Besides that, AD and FTD patients in those previous studies had lower scores on the MMSE than in our study, which are indicative of a later disease stage and presumably more disease related GM atrophy, facilitating discrimination. In our study, only patients with bvFTD were included, all patient groups had MMSE scores indicative for mild dementia, as well as comparable age and disease duration, rendering these confounding effects less likely. Furthermore, as the SVM identified brain regions, which contribute highly to the classification, are in 103

agreement with results of literature on GM atrophy in AD and bvFTD, we are confident that our results are valid. The most important difference of this study in comparison to other studies using SVM is, that we used two independent sets of patients to test the generalizability and performance of the classifier. One set, the training set, was used in PRoNTo to learn the atrophy patterns and assign weights to the different patient groups. The learned weights from the training set were then used to classify new patients who were not used in the training phase. As each prediction is done independently for each test subject, this corresponds to a single-subject classification. To our knowledge, other studies using automated classifiers do not use an independent test set and therefore using the cross-validation framework within the classifier to classify patients. This has the disadvantage that classified patients have also been used to train the classifier and are ‘known’ to the machine. This could bias the results. To our knowledge, there is only one other study that used a separate training and test set to evaluate the predictive power of a SVM [35]. The accuracy levels of our study are comparable and demonstrate the robustness of the performance of the automated classifier in independent datasets. Our study extends these results by testing larger sample sizes and fully independent training and test sets. We also used whole brain information instead of a region-of-interest (ROI) approach. The use of a ROI approach may lead to higher accuracy rates but [35, 36] restrain the implementation of the SVM approach in the daily practice, as extraction of ROIs is time-consuming. Another problem of the ROI based approach is the limited generalizability of the trained automatic classifier using single-center data when applied to new data sets acquired on different scanners. When we compare our results with other classification studies using SVM with other modalities, we found slightly higher accuracy rates for SPECT (84-88%) [37], FDG-PET (80-82.9%) [16], and a study on FDG-PET with higher accuracy rates [38]. Even though PET imaging may reveal higher accuracy rates in some studies, it is still not available in most hospital and requires specialized technical support. It may be possible to increase SVM accuracy by adding biomarkers, however for this study we aimed to use only one modality, which is widely available mainstream hospitals. Furthermore, adding modalities does not necessarily lead to higher accuracy rates. A recent study added CBF measurements GM densities, and reported little improvements in diagnostic value for the combined data [17]. While MRI-based methods alone might not show best performance for all applications, our results indicate that the relative ease of data acquisition and applying automated diagnosis methods and the non-invasive nature of MRI make it a useful diagnostic measure for the discrimination between AD and bvFTD, especially in centers, where PET imaging and/or CSF measures are not available. Although the clinical consensus criteria for bvFTD and AD have been improved, underlying pathology can only be predicted on clinical grounds with limited accuracy. Especially in mild stages of the disease, bvFTD and AD show overlapping clinical features and an enormous heterogeneity within both diagnoses [39, 40].Various studies show that neuropsychological tests are not accurate enough to discriminate 104

AD from bvFTD [41], as bvFTD also show memory deficits and AD patients present with executive dysfunctions [5-9, 31, 42, 43]. MMSE and FAB have also been shown to not reliable differentiate AD from bvFTD [44].We confirm these findings by showing that the diagnostic accuracy of the automated classifier based on MRI outperformed the diagnostic accuracy of neuropsychological tests. Although several studies examining cross-sectional and longitudinal effects in volumes of brain regions have shown significant group differences between AD, bvFTD, and controls, the ability to detect structural patterns that enable accurate single-subject predictions will ultimately assist the diagnostic process in the daily practice. Our study focused on making automated single-subject diagnosis more accessible, taking into account multiple factors of the daily clinical practice, such as scans from different MR scanners and a control group consisting of healthy elderly controls and people with subjective memory complaints. Therefore, the method we described has potential in achieving more accurate dementia diagnosis in clinical practice. It is possible that accuracy rates of the classifier may differ in patient samples from a different memory clinic because of heterogeneity within diagnoses. Patients of our test set were selected based on previous diagnoses, which may have improved classification accuracy. We do not expect lower accuracy rates in other memory clinics, as patients with an unclear clinical presentation are often referred to our clinic for a second opinion. At the other end of the spectrum, we predict that in a singleclinic setting with one scanner, and by using healthy controls not SMC subjects, accuracy rates will be even higher than in our study. Another possible limitation is that, we only used binary classifiers, which means that a test case not belonging to one of the two groups will be incorrectly assigned to one of these. Multi-class classifiers software for MRI are currently not widely available. Future releases of available machine learning models will facilitate multi-class classifications, and thereby improve the diagnostic usefulness. However, the current finding indicates that a binary classifier can assist the diagnostic process by predicting different dementia subtypes based on localized changes in GM density. It could be argued that valuable classification power is lost due to the whole brain approach and well-placed ROIs would improve categorization as some brain areas may be more informative about class membership than others. However, a disadvantage of a ROI based approach is that it might not be as generalizable as a classifier that takes into account whole brain information. Nevertheless, the aim of our study was to achieve optimal classification based on whole brain information to build an optimal classifier in a way that it could be easily used in daily practice. Our results are encouraging and describe a role for computerized diagnostic methods in clinical practice. The analytical technique presented here is able to distinguish disease-specific GM atrophy between AD and bvFTD in a standard T1weighted structural MRI scan for single-subjects. The differentiation based on GM outperforms the classification based on neuropsychological tests. A goal of machine learning based automated MR image analysis is higher sensitivity and specificity of ante-mortem diagnosis than is currently possible. A study by Klรถppel et al. [19] 105

showed that computer-based diagnosis is equal to or better than that achieved by radiologists. Together with our results, it is conceivable that in the future machine learning-based categorization methods could improve diagnosis, especially in centers without experienced neuroradiologists and without other supporting diagnostic measures. An important next step will be the application of automatic classification for screening purposes. If machine learning discrimination is sensitive enough to classify subtle differences, early screening of high-risk groups could be easily implemented.

Acknowledgements The VUmc Alzheimer Center is supported by Alzheimer Nederland and Stichting VUmc Fonds. The study was supported by the Netherlands Organization for Scientific Research (NWO). Research of the VUmc Alzheimer Center is part of the neurodegeneration research program of the Neuroscience Campus Amsterdam. The clinical database structure was developed with funding from Stichting Dioraphte. This project is funded by the Netherlands Initiative Brain and Cognition (NIHC), a part of the Netherlands Organization for Scientific Research (NWO) under grant numbers 056-13-014 and 056-13-010. Dr. Wiesje van der Flier is recipient of the Alzheimer Nederland grant (Influence of age on the endophenotype of AD on MRI, project number 2010-002).

106

Tables and Figures Table 1. Demographics Training set Test set N 115 114 Scanners GE 84 (73%) 83 (73%) PETMR 9 (8%) 9 (8%) Philips 22 (19%) 22 (19%) Diagnosis AD 42 (37%) 42 (37%) bvFTD 26 (22%) 25 (22%) Controls 47 (41%) (20 HC, 27 SMC) 47 (41%) (21 HC, 26 SMC) Age 62.7 ± 7.5 62.7 ± 6.7 Sex, f 36 (31%) 36 (31%) MMSE 25.2 ± 4.4 25.2 ± 4.3 Disease duration, months 42.6 ± 32.4 43.2 ± 31.0 Values presented as mean ± standard deviation or n (%). Differences between groups for 2 demographics were assessed using ANOVA. Kruskal-Wallis tests and χ tests. where appropriate. Key: SMC: Subjective memory complaints; MMSE: Mini-Mental State Examination

Table 2. Demographics within training- and test-set based on diagnosis Training set Test set AD bvFTD Controls AD bvFTD Controls N 42 26 47 42 25 47 Scanner GE 31 (74%) 18 (69%) 35 (75%) 31(74%) 17 (68%) 35 (75%) PETMR 3 (7%) 3 (12%) 3 (6%) 3 (7%) 3 (12%) 3 (6%) Philips 8 (19%) 5 (19%) 9 (19%) 8 (19%) 5 (20%) 9 (19%) a a Age 64.9 ± 7.1 62.1 ± 7.8 61 ± 7.3 65.3 ± 7.1 61.6 ± 6.4 60.9 ± 5.7 Sex, f 13 (31%) 9 (35%) 14 (30%) 16 (38%) 4 (16%) 16 (34%) Disease 36.6 ± 22.1 44.6 ± 40.3 44.4 ± 30.1 41.0 ± 30.8 duration, months 4.7 ± 1.5 4.9 ± 1.3 5.1 ± 1.2 5.0 ± 1.3 5.0 ± 1.3 5.5 ± 1.2 Level of education a a,b a a,b 24.8 ± 3.4 28.5 ± 1.6 22 ± 3.7 24.6 ± 3.8 28.5 ± 2 MMSE 21.9 ± 4.5 a a a a FAB 12.4 ± 4.0 12.7 ± 4.3 17.1 ± 1.2 13.7 ± 3.1 13.4 ± 3.0 16.8 ± 1.8 a b a b Memory -4.5 ± 3.0 -1.4 ± 1.6 -0.1 ± 0.7 -3.5 ± 3.1 -1.5 ± 1.6 -0.2 ± 0.9 a a a a -1.0 ± 1.1 -0.1 ± 0.8 -1.5 ± 1.7 -1.8 ± 1.3 -0.2 ± 0.9 Language -1.3 ± 1.3 a a a Attention -1.6 ± 1.5 -1.2 ± 1.6 -0.1 ± 0.8 -1.7 ± 1.6 -1.1 ± 2.0 -0.1 ± 0.8 a a a a,b EF -1.9 ± 1.7 -1.7 ± 1.7 0.0 ± 0.6 -2.1 ± 2.0 -1.2 ± 0.9 -0.3 ± 0.8 Values presented as mean ± standard deviation or n (%). Differences between groups for 2 demographics were assessed using ANOVA. Kruskal-Wallis tests and χ tests where appropriate. Cognitive composite z-domains were calculated of the available z-scores of each test by the MEAN function in SPSS and compared by MANOVA with Bonferroni posthoc tests and age, sex, educational level and disease duration as covariates. Key: MMSE: Mini-Mental State Examination; FAB: Frontal Assessment Battery; EF: Executive a b functioning. different from controls (p<0.05), different from AD (p<0.05) 107

Figure 1.Weight maps for each classifier at a threshold of 30% of the maximum positive and negative weight values, superimposed onto a standard brain template from FSL (MNI152_T1_1mm) showing areas of the brain most vital for discriminating the two groups. Red-Yellow: negative values indicative for class 1. Blue-Light blue: positive values indicative for class 2. (A) AD vs. controls, (B) bvFTD vs. controls, (C) AD vs. bvFTD.

108

Figure 2. Performance of support vector machine classification of training set data. (A) AD vs. controls, (B) bvFTD vs. controls, (C) AD vs. bvFTD.

109

Figure 3. Discriminating performance in test-set of averaged integrated product of weight map from the training-set and smoothed GM images from the test-set. Scatterplots showing discrimination between two groups. The ROC curve illustrates the performance of the binary classifier. (A) 87.6% of AD patients and controls were classified correctly, with correct classification of 36 AD patients (85.7%) and 42 controls (89.4%). The ROC curve revealed an excellent AUC for the extracted weights (0.95; p<0.001). (B) 84.7% of FTD patients and controls were classified correctly, with correct classification of 15 bvFTD patients (60%) and 46 controls (97.9%). The ROC curve revealed a good AUC for the extracted weights (0.87; p<0.001). (C) 82.1% of AD and bvFTD patients were classified correctly, with correct classification of 39 AD (92.9%) and 16 bvFTD patients (64%). The ROC curve revealed a good AUC for the extracted weights (0.81; p<0.001).

110

Reference List [1] [2]

[3]

[4]

[5] [6] [7]

[8]

[9]

[10]

[11]

[12] [13] [14]

[15]

[16] [17]

[18]

Ratnavalli E, Brayne C, Dawson K, Hodges JR. The prevalence of frontotemporal dementia. Neurology 2002; 58:1615-1621. Rascovsky K, Hodges JR, Knopman D, et al. Sensitivity of revised diagnostic criteria for the behavioural variant of frontotemporal dementia. Brain 2011; 134:24562477. McKhann GM, Knopman DS, Chertkow H, et al. The diagnosis of dementia due to Alzheimer's disease: recommendations from the National Institute on AgingAlzheimer's Association workgroups on diagnostic guidelines for Alzheimer's disease. Alzheimers Dement 2011; 7:263-269. Koedam EL, Lauffer V, van der Vlies AE, van der Flier WM, Scheltens P, Pijnenburg YA. Early-versus late-onset Alzheimer's disease: more than age alone. J Alzheimers Dis 2010; 19:1401-1408. Bertoux M, de Souza LC, Corlier F, et al. Two distinct amnesic profiles in behavioral variant frontotemporal dementia. Biol Psychiatry 2014; 75:582-588. Hornberger M, Piguet O. Episodic memory in frontotemporal dementia: a critical review. Brain 2012; 135:678-692. Hutchinson AD, Mathias JL. Neuropsychological deficits in frontotemporal dementia and Alzheimer's disease: a meta-analytic review. J Neurol Neurosurg Psychiatry 2007; 78:917-928. Yew B, Alladi S, Shailaja M, Hodges JR, Hornberger M. Lost and forgotten? Orientation versus memory in Alzheimer's disease and frontotemporal dementia. J Alzheimers Dis 2013; 33:473-481. Galton CJ, Patterson K, Xuereb JH, Hodges JR. Atypical and typical presentations of Alzheimer's disease: a clinical, neuropsychological, neuroimaging and pathological study of 13 cases. Brain 2000; 123 Pt 3:484-498. Rabinovici GD, Seeley WW, Kim EJ, et al. Distinct MRI atrophy patterns in autopsyproven Alzheimer's disease and frontotemporal lobar degeneration. Am J Alzheimers Dis Other Demen 2007; 22:474-488. Whitwell JL, Jack CR, Jr. Comparisons between Alzheimer disease, frontotemporal lobar degeneration, and normal aging with brain mapping. Top Magn Reson Imaging 2005; 16:409-425. Du AT, Schuff N, Kramer JH, et al. Different regional patterns of cortical thinning in Alzheimer's disease and frontotemporal dementia. Brain 2007; 130:1159-1166. Johnson JK, Head E, Kim R, Starr A, Cotman CW. Clinical and pathological evidence for a frontal variant of Alzheimer disease. Arch Neurol 1999; 56:1233-1239. van de Pol LA, Hensel A, van der Flier WM, et al. Hippocampal atrophy on MRI in frontotemporal lobar degeneration and Alzheimer's disease. J Neurol Neurosurg Psychiatry 2006; 77:439-442. Mรถller C, Dieleman N, van der Flier WM, et al. More Atrophy of Deep Gray Matter Structures in Frontotemporal Dementia Compared to Alzheimer's Disease. J Alzheimers Dis 2014. Dukart J, Mueller K, Horstmann A, et al. Combined evaluation of FDG-PET and MRI improves detection and differentiation of dementia. PLoS One 2011; 6:e18111. Bron EE, Steketee RM, Houston GC, et al. Diagnostic classification of arterial spin labeling and structural MRI in presenile early stage dementia. Hum Brain Mapp 2014. Klรถppel S, Stonnington CM, Chu C, et al. Automatic classification of MR scans in Alzheimer's disease. Brain 2008; 131:681-689. 111

[19]

[20]

[21]

[22]

[23]

[24] [25]

[26]

[27] [28]

[29] [30] [31]

[32] [33] [34]

[35]

[36]

112

Klรถppel S, Stonnington CM, Barnes J, et al. Accuracy of dementia diagnosis: a direct comparison between radiologists and a computerized method. Brain 2008; 131:2969-2974. Davatzikos C, Resnick SM, Wu X, Parmpi P, Clark CM. Individual patient diagnosis of AD and FTD via high-dimensional pattern classification of MRI. Neuroimage 2008; 41:1220-1227. Vemuri P, Simon G, Kantarci K, et al. Antemortem differential diagnosis of dementia pathology using structural MRI: Differential-STAND. Neuroimage 2011; 55:522-531. Whitwell JL, Josephs KA, Rossor MN, et al. Magnetic resonance imaging signatures of tissue pathology in frontotemporal dementia. Arch Neurol 2005; 62:14021408. Plant C, Teipel SJ, Oswald A, et al. Automated detection of brain atrophy patterns based on MRI for the prediction of Alzheimer's disease. Neuroimage 2010; 50:162174. Zhang D, Wang Y, Zhou L, Yuan H, Shen D. Multimodal classification of Alzheimer's disease and mild cognitive impairment. Neuroimage 2011; 55:856-867. McKhann G, Drachman D, Folstein M, Katzman R, Price D, Stadlan EM. Clinical diagnosis of Alzheimer's disease: report of the NINCDS-ADRDA Work Group under the auspices of Department of Health and Human Services Task Force on Alzheimer's Disease. Neurology 1984; 34:939-944. Petersen RC, Stevens JC, Ganguli M, Tangalos EG, Cummings JL, DeKosky ST. Practice parameter: early detection of dementia: mild cognitive impairment (an evidence-based review). Report of the Quality Standards Subcommittee of the American Academy of Neurology. Neurology 2001; 56:1133-1142. Verhage F. Intelligentie en leeftijd: onderzoek bij Nederlanders van twaalf tot zevenenzeventig jaar. Assen: Van Gorcum, 1964. Folstein MF, Folstein SE, McHugh PR. "Mini-mental state". A practical method for grading the cognitive state of patients for the clinician. J Psychiatr Res 1975; 12:189-198. Dubois B, Slachevsky A, Litvan I, Pillon B. The FAB: a Frontal Assessment Battery at bedside. Neurology 2000; 55:1621-1626. Stroop JR. Studies of interference in serial verbal reactions. J Exp Psychol 1935; 18:643662. Smits LL, Pijnenburg YA, Koedam EL, et al. Early Onset Alzheimer's Disease is Associated with a Distinct Neuropsychological Profile. J Alzheimers Dis 2012; 30:101-108. Jenkinson M, Smith S. A global optimisation method for robust affine registration of brain images. Med Image Anal 2001; 5:143-156. Schrouff J, Rosa MJ, Rondina JM, et al. PRoNTo: pattern recognition for neuroimaging toolbox. Neuroinformatics 2013; 11:319-337. Mourao-Miranda J, Reinders AA, Rocha-Rego V, et al. Individualized prediction of illness course at the first psychotic episode: a support vector machine MRI study. Psychol Med 2012; 42:1037-1047. Raamana PR, Rosen H, Miller B, Weiner MW, Wang L, Beg MF. Three-Class Differential Diagnosis among Alzheimer Disease, Frontotemporal Dementia, and Controls. Front Neurol 2014; 5:71. Duchesne S, Caroli A, Geroldi C, Barillot C, Frisoni GB, Collins DL. MRI-based automated computer classification of probable AD versus normal controls. IEEE Trans Med Imaging 2008; 27:509-520.

[37]

[38] [39] [40]

[41]

[42]

[43]

[44]

Horn JF, Habert MO, Kas A, et al. Differential automatic diagnosis between Alzheimer's disease and frontotemporal dementia based on perfusion SPECT images. Artif Intell Med 2009; 47:147-158. Xia Y, Lu S, Wen L, Eberl S, Fulham M, Feng DD. Automated identification of dementia using FDG-PET imaging. Biomed Res Int 2014; 2014:421743. Greicius MD, Geschwind MD, Miller BL. Presenile dementia syndromes: an update on taxonomy and diagnosis. J Neurol Neurosurg Psychiatry 2002; 72:691-700. Miller BL, Diehl J, Freedman M, Kertesz A, Mendez M, Rascovsky K. International approaches to frontotemporal dementia diagnosis: from social cognition to neuropsychology. Ann Neurol 2003; 54 Suppl 5:S7-10. Gregory CA, Orrell M, Sahakian B, Hodges JR. Can frontotemporal dementia and Alzheimer's disease be differentiated using a brief battery of tests? Int J Geriatr Psychiatry 1997; 12:375-383. Jenner C, Reali G, Puopolo M, Silveri MC. Can cognitive and behavioural disorders differentiate frontal variant-frontotemporal dementia from Alzheimer's disease at early stages? Behav Neurol 2006; 17:89-95. Dickerson BC, Wolk DA. Dysexecutive versus amnesic phenotypes of very mild Alzheimer's disease are associated with distinct clinical, genetic and cortical thinning characteristics. J Neurol Neurosurg Psychiatry 2011; 82:45-51. Castiglioni S, Pelati O, Zuffi M, et al. The frontal assessment battery does not differentiate frontotemporal dementia from Alzheimer's disease. Dement Geriatr Cogn Disord 2006; 22:125-131.

113

Chapter 3.3

Different patterns of cortical gray matter loss in behavioral variant FTD and AD 1*

2-4

Christiane Mรถller MSc , Anne Hafkemeijer MSc , Yolande A L 1 2-4 Pijnenburg MD PhD , Serge A R B Rombouts PhD , Jeroen van der 3 1,3,5,6 5,6 Grond PhD , Elise Dopper , John van Swieten MD PhD , Adriaan 7 7 7 Versteeg , Martijn Steenwijk MSc , Frederik Barkhof MD PhD , Philip 1 7,8 Scheltens MD PhD , Hugo Vrenken PhD , Wiesje M van der Flier 1,9 PhD 1 Alzheimer Center & Department of Neurology, Neuroscience Campus Amsterdam, VU University medical center, Amsterdam, the 2 Netherlands, Institute of Psychology, Leiden University, Leiden, the 3 Netherlands, Department of Radiology, Leiden University Medical 4 Center, Leiden, the Netherlands, Leiden Institute for Brain and 5 Cognition, Leiden University, Leiden, the Netherlands, Department of Clinical Genetics, Neuroscience Campus Amsterdam, VU University 6 medical center, Amsterdam, the Netherlands, Department of Neurology, Erasmus Medical Center, Rotterdam, the Netherlands, 7 8 Department of Radiology & Nuclear Medicine, Department of Physics 9 & Medical Technology, Department of Epidemiology & Biostatistics, Neuroscience Campus Amsterdam, VU University medical center, Amsterdam, the Netherlands. Under review

114

Abstract Objectives: The ability to track regional atrophy changes over time may help refine clinical diagnosis of behavioral variant frontotemporal dementia (bvFTD) and Alzheimer’s disease (AD). We therefore examined the loss of cortical thickness over time in AD, bvFTD and controls in a longitudinal study. Methods: 19 AD, 10 bvFTD patients and 30 control subjects underwent cognitive assessment and MRI twice with a mean interval of 2¹0.4 years. We obtained thicknesses from left and right frontal, parietal, temporal, occipital, insula, cingulate lobe, as well as whole-brain thickness with the longitudinal FreeSurfer pipeline. Group differences in progression of cortical thickness were assessed using 1) MANOVA with diagnosis as independent variable and symmetrized percentage change in cortical thickness per lobe and hemisphere as dependent variable; 2) whole-brain vertex-wise general linear model. Age, sex, center, and disease duration were used as covariates. Results: The groups did not differ in age, sex, center, level of education, or total intracranial volume. BvFTD had a longer disease duration compared to AD. Both, AD and bvFTD showed more cortical thinning per year and showed a steeper decline on MMSE compared to controls with AD showing decline in memory and language. Progression of cortical thickness showed comparable results in whole-brain and lobar measurements: Compared to controls, AD patients progressed over the whole brain with a clear posterior gradient, whereas bvFTD patients only showed progression in frontal cortex en in anterior parts of the temporal lobes. Compared to each other, AD patients showed cortical thinning in the insula, temporal and parietal regions, bvFTD patients only progressed faster in a small frontal region. Conclusions: Decrease of thickness was highest in AD and generalized throughout the whole brain, whereas bvFTD had a more selective loss, predominantly in the frontal and anterior temporal parts of the brain.

115

Introduction Alzheimer’s disease (AD) and behavioral variant frontotemporal dementia (bvFTD) are the two most prevalent early-onset dementias [1]. Patterns of gray matter (GM) atrophy often overlap between AD and bvFTD. Especially early in the disease when the typical patterns of AD and bvFTD are not yet visible, discrimination between the different forms can be challenging. Cross-sectionally, the anatomical differences between AD and bvFTD have been studied extensively but less is known about clinical and anatomical evolution over time. Serial MRI data have been used to quantify rates of cerebral change in AD and bvFTD separately. Compared to controls, AD showed volume loss in temporal, parietal and occipital cortices [2,3] and bvFTD showed volume changes in the anterior cingulate, frontal cortex and insula [4,5]. Rates of atrophy have the potential to be valuable biomarkers of disease progression, and the results of longitudinal studies have been used as outcome measure in clinical trials with AD patients [6,7]. Patterns of atrophy over time, however have only sparsely been compared between bvFTD and AD. The results of the limited number of studies are not uniform. Some studies found that FTD patients progressed faster than AD patients, especially in the frontal parts of the brain [3,8]. Other studies could not support these findings [9-11]. Heterogeneity in results could be explained by different methodological approaches, as some studies either used whole brain atrophy only [10,12], a few predefined regions of interest [3,8,9,11], or included the full clinical spectrum of FTLD [9,10,12]. In this study, we therefore examined the progression over time in cortical thickness in a group of AD and bvFTD patients compared with controls in a whole brain vertexwise fashion and with regions of interest. Furthermore, we explored if cognitive measurements progress in the same fashion.

Methods Patients In this two center study, we included 19 patients with probable AD and 10 patients with bvFTD and 30 controls with repeated MRI, enrolled at the Alzheimer Center of the VU University medical center (VUmc) or the Alzheimer Center of the Erasmus University Medical Center Rotterdam. All patients underwent a standardized one-day assessment including medical history, informant-based history, physical and neurological examination, blood tests, neuropsychological assessment, and MRI of the brain. Diagnoses were made in a multidisciplinary consensus meeting according to the core clinical criteria of the National Institute on Aging and the Alzheimer’s Association workgroup for probable AD [13,14] and according to the clinical diagnostic criteria of FTD for bvFTD [15]. To minimize center effects, all diagnoses were reevaluated in a panel including clinicians from both centers. In addition, we included cognitively normal controls, who were recruited by advertisement in local newspapers. Before inclusion in the present study, controls were screened for memory complaints, family history of dementia, drugs- or alcohol abuse, major psychiatric disorder, and neurological or cerebrovascular diseases. They underwent an assessment including medical history, physical examination, neuropsychological assessment, and MRI of the brain comparable to the work-up of patients. 116

Inclusion criteria for both cohorts were: (1) availability of two T1-weighted 3dimensional MRI (3DT1) scans without artifacts performed on the same scanner using an identical imaging protocol, and (2) age between 50 and 80 years. Subjects were excluded from the study when the imaging analyzing software failed to process MR scans (n=6) or segmentation errors occurred which could not be edited manually (n=9) (details see sections below). Level of education was rated on a seven-point scale [16]. Disease duration was defined as the time between first symptom onset and diagnosis, as reported by the caregiver of the patient. The local medical ethics committee of both centers approved the study. All patients gave written informed consent. Neuropsychological assessment Patients underwent neuropsychological assessment twice. To assess dementia severity we used the Mini-Mental State Examination (MMSE). Cognitive functioning was assessed using a standardized neuropsychological test battery covering five major domains: memory (immediate recall, recognition and delayed recall of Dutch version of the Rey Auditory Verbal Learning Test and total score of Visual Association Test A), language (Visual Association Test picture naming and category fluency (animals: 1 min)), visuospatial functioning (number location (subtests of Visual Object and Space Perception Battery (VOSP))), attention (Trail Making Test part A (TMT A), Digit Span forward, and Letter Digit Substitution Test (LDST)), and executive functioning (Digit Span backwards, Trail Making Test part B (TMT B), letter fluency, and Stroop Color-Word card subtask [17]). For a detailed description of neuropsychological tests see Smits et al. [18]. For each cognitive task, z-scores were calculated from the raw test scores by the formula z=(x-μ)/σ, where μ is the mean and σ is the standard deviation of the baseline performance of the control group. The value z = 0 therefore reflects the average baseline test performance of the healthy controls in a given domain. For cognitive performances at follow-up, z-scores of controls and patients were calculated relative to the baseline z-scores of controls. Scores of TMT A, TMT B, and Stroop color-word card were inverted by computing (1)*z-score, because higher scores imply a worse performance. Next, composite zscores were calculated for each cognitive domain by averaging z-scores. Composite zscores were calculated when at least one neuropsychological task was available in each cognitive domain. MR image acquisition and review All patients underwent two MR scans. Imaging at the VUmc was carried out on a 3T scanner (Signa HDxt, GE Healthcare, Milwaukee, WI, USA), using an 8-channel head coil with foam padding to restrict head motion. Patients and controls from the Erasmus University Medical Center Rotterdam were all scanned at the Leiden University Medical Center (LUMC). Imaging at LUMC was performed on a 3T scanner (Achieva, Philips Medical Systems, Best, the Netherlands) using an 8-channel SENSE head coil. The scan protocol included a whole-brain near-isotropic 3DT1-weighted sequence for cortical segmentation. At VUmc this was a fast spoiled gradient echo sequence (FSPGR; repetition time TR 7.8 ms, echo time TE 3 ms, inversion time TI 450 ms, flip angle 12º, 180 sagittal slices, voxel size 0.98x0.98x1 mm, total scan time 4.57 minutes). At LUMC this was a turbo field echo sequence (T1TFE; TR 9.8 ms, TE 4.6 ms, 117

flip angle 8°, 140 transversal slices, voxel size 0.88x0.88x1.2 mm, total scan time 4.57 minutes). In addition, the MRI protocol included a 3D Fluid Attenuated Inversion Recovery (FLAIR) sequence, dual-echo T2-weighted sequence, and susceptibility weighted imaging (SWI) which were reviewed for brain pathology other than atrophy by an experienced radiologist. MR preprocessing and analysis To extract thickness estimates, images where automatically processed with the longitudinal stream [19] in FreeSurfer 5.3 [20,21]. In general, FreeSurfer uses the 3DT1-weighted images to locate the pial and white matter surface of the cortex. The distance between these surfaces gives the vertex-wise cortical thickness of cortical areas (i.e. the perpendicular thickness at each location). Specifically for the longitudinal analyses, an unbiased within-subject template space and image from both time points [22] was created using a robust, inverse consistent registration [23]. Several processing steps, such as skull stripping, Talairach transforms, atlas registration as well as spherical surface maps and parcellations for baseline and follow-up scan were then initialized with common information from the within-subject template, significantly increasing reliability and statistical power [19]. Areas of cortical thickness were subsequently averaged in the four cortical lobes (frontal, parietal, temporal, occipital) per hemisphere and time point with mri_annotation2label [24]. As the insula and the cingulate cortex are important areas in AD and bvFTD, thickness of cingulate cortex and insula were also extracted using mri_annotation2label [25-29]. Symmetrized percent change (SPC) of thickness of the four lobes and cingulate cortex and insula were obtained from the longitudinal two stage model of FreeSurfer [30]. SPC is the rate with respect to the average thickness (SPC = rate / avg), where rate is the difference in thickness per time unit, so (rate = ( vol2 - vol1 ) / (time2 - time1)), and avg is the average thickness at the midpoint of linear fit (avg = 0.5 * (vol1 + vol2)). SPC already takes the time between baseline and follow-up scan into account. This is a more robust measure than percentage change, because thickness at time point 1 is more noisy than the average. Also SPC is symmetric, what is not true for percentage change. All cortical segmentations were manually checked and re-run after manual editing if errors occurred or excluded if errors could not be edited. Next to SPC per lobe, changes in whole brain cortical thickness were analyzed using FreeSurfer’s vertex-wise general linear modeling testing for a main effect of group on SPC in cortical thickness, taking follow-up time into account. This involved smoothing of the cortical thickness maps using a Gaussian kernel with a full width half maximum of 10-mm. Then the smoothed surface was entered as the dependent in the vertexwise GLM after which cluster-wise correction for multiple comparisons was performed using Monte Carlo Z simulation while thresholding the vertex-wise statistical maps at P<0.001, with a cluster-level threshold of P<0.05, using 5000 iterations. Mean age of age at baseline and follow-up, sex and center were used as covariates. Estimated Total Intracranial Volume (eTIV) as generated by FreeSurfer was used as an estimate for intracranial volume (ICV) in this study. The eTIV measure from FreeSurfer is in good agreement with ICV reference segmentation acquired from proton density 118

weighted images [31] and has previously been used in several studies for normalization. Statistical analysis Statistical analyses were performed using SPSS 20.0 (SPSS Inc). ‘Follow-up time’ for MRI and neuropsychological assessment was defined as the interval in years between first (baseline) and second (follow-up) examination. Demographics between the 2 diagnostic groups were compared with ANOVAs, Kruskal-Wallis and Pearson’s χ tests where appropriate. Neuropsychological test performances between AD, bvFTD and controls at baseline were compared with MANOVA with diagnosis as independent variable, composite zdomains as dependent variable and sex, baseline age, level of education and disease duration as covariates. Subsequently, ANOVA for repeated measures with time and composite z-domain as within-subjects factors, diagnosis as between-subjects factor, and mean age of age at baseline and follow-up, sex, level of education, disease duration, and follow-up time as covariates was conducted. As posthoc test, we conducted MANOVA with ‘annual rate of cognitive decline’ (rate=(follow-up z-domain score - baseline z-domain score) / follow-up time) as dependent variable, diagnosis as independent variable and age, sex, level of education, and disease duration as covariates, to assess the annual rate of cognitive decline between groups for each zdomain. MANOVA with diagnosis as independent variable and lobe per hemisphere as dependent variable were conducted to determine group differences in baseline thickness. If main effects were significant, Bonferroni corrected posthoc comparisons were conducted. Age at baseline, sex, disease duration, and center were used as covariates. Subsequently, we conducted MANOVA with diagnosis as independent variable and SPC in cortical thickness per lobe and hemisphere as dependent variable to determine group differences in progression of cortical thickness for each lobe. If main effects were significant, Bonferroni corrected posthoc comparisons were conducted. Mean age of age at baseline and follow-up, sex, disease duration, and center were used as covariates. Statistical significance for all analyses was set at p<0.05.

Results Demographics Table 1 summarizes the baseline demographics of the diagnostic groups. In general, patients had lower scores on the MMSE, and a shorter length of follow-up compared to controls. There were no differences between the groups in age, sex, center, level of education, or TIV. BvFTD had a longer disease duration compared to AD patients. AD and bvFTD patients showed a steeper decline on MMSE compared to healthy controls. Neuropsychological performance Table 1 summarizes the neuropsychological performances at baseline. MANOVA revealed significant main effects of diagnosis for memory (F=61,161, p<0.001), 119

language (F=5,946, p=0.007), attention (F=2.801, p=0.006), and executive functioning (F=8.265, p=0.001). There was no significant main effect for visuospatial functioning between the diagnostic groups (F=1.716, p=0.194). Posthoc tests showed that AD and bvFTD performed worse on memory and language domain than controls, with AD having the worst memory performances and bvFTD having the lowest scores on the language domain. AD patients had the lowest scores on attention and executive functions, but differed only from controls. Progression of neuropsychological performances Figure 1 and table 1 summarize the progression over time of neuropsychological performances in five cognitive domains. Repeated measures showed significant effects of time*diagnosis (p<0.001), cognitive domain (p=0.004), cognitive domain*diagnosis (p<0.001), and time*cognitive domain*diagnosis (p=0.015) showing that annual cognitive deterioration was different for the three diagnostic groups. Posthoc comparisons of the annual rates of cognitive decline showed only for memory and language significant effects. The performance of AD patients declined the most, but was only significantly different from controls. Cortical thickness at baseline Figure 2 shows patterns of atrophy for whole-brain vertex-wise analyses at baseline. Compared to controls, AD patients showed a clear posterior pattern of atrophy with involvement of the parietal and temporal lobe, whereas bvFTD patients showed a frontal gradient of atrophy with atrophy in the frontal an d temporal lobes. Compared to AD patients, bvFTD showed atrophy in the orbitofrontal area, and AD patients showed atrophy in the precuneus compared to bvFTD. MANOVA on the differences in the four lobes, cingulate cortex and insula showed similar patterns of atrophy (Table 2). AD and bvFTD patients had a thinner cortex of left and right temporal lobe compared to controls. AD patients had thinner left and right parietal cortices than controls, whereas bvFTD patients did not differ from controls or AD patients. BvFTD patients had thinner left, right frontal, right cingulate and left insula cortices than controls, whereas AD patients did not differ from controls or bvFTD patients. Right insular cortex was most severely thinned in bvFTD, differentiating them from AD and controls, whereas AD patients did not differ from controls. In contradiction to volume measurements, thicknesses of the occipital cortices and left cingulate did not differ between the groups. Progression of cortical thickness Figure 3 shows the progression of atrophy for whole-brain vertex-wise analyses. Compared to controls, AD patients progressed over the whole brain with a clear posterior gradient, but also in some frontal parts, whereas bvFTD patients only showed progression in frontal cortex en in anterior parts of the temporal lobes. Compared to each other, AD patients showed cortical thinning in the insula, temporal and parietal regions, bvFTD patients only progressed faster in a small frontal region. Figure 4 and table 2 summarize the SPC of thicknesses of the four lobes, cingulate cortex and insula per hemisphere. MANOVA on the differences in SPC between the groups revealed that compared to controls, AD patients progressed in almost all 120

cortices, whereas bvFTD patients progressed only in bilateral temporal and cingulate cortex, and left frontal cortex. Especially in right frontal (-1.2%) and biparietal (left: 1.5%; right: -2.0%) lobes, AD patients showed a steeper rate of cortical thinning than controls, whereas bvFTD patients did not differ from controls. In left frontal, bitemporal and bicingulate, both, AD (left frontal: -0.8%; left temporal: -2.1%; right temporal: -2.0%; left cingulate: -1.1%; right cingulate: -1.6%) and bvFTD patients (left frontal: -0.7%; left temporal: -2.3%; right temporal: -1.6%; left cingulate: -1.2%; right cingulate: -1.1%) progressed faster than controls (left frontal: -0.4%; left temporal: 0.4%; right temporal: -0.5%; left cingulate: -0.1%; right cingulate: +0.4%). For the right insula, AD patients (-2.1%) lost more cortical thickness than bvFTD (+0.5%) and controls (-0.2%). There were no differences in progression of cortical thinning for bioccipital and left insula.

Discussion The main findings of our study are that annual decrease of thickness was highest in AD patients. They lost GM in a generalized way throughout the whole brain with a posterior gradient, whereas bvFTD patients had a more selective loss of GM, predominantly in the anterior en temporal parts of the brain. Progression of cognitive performances showed also that AD patients deteriorated the most The generalizability of the few former studies on this topic is hampered by the use of heterogeneous patient groups, i.e. using the whole spectrum of FTLD [4,5,9] or patients that were already in a late disease state at baseline [11]. In the current study, we were able to show progression of GM matter thickness in probable bvFTD and probable AD patients, who were included in this study when they first presented at our memory clinic and were scanned with a mean interval of 2 years. Progression of atrophy showed that AD patients had the steepest annual rate of atrophy in almost all areas of the brain, especially in the right insula, where they lost even more GM compared to bvFTD. BvFTD had the steepest annual atrophy rates in left frontal, left temporal and left cingulate, but only compared to controls. Results of other studies on progression of brain atrophy comparing AD with bvFTD are mixed, which can partly be explained by the wide use of different longitudinal MR analysis methods and different investigation methods (whole brain vs ROI, volume vs thickness). Some studies found that rate of atrophy over time was significantly faster in FTD compared to AD [8,11]. Another study found a trend for atrophy rate to be greater in FTLD-U than those observed in AD [10]. The lateral orbitofrontal gyrus was found to separate bvFTD from AD and controls in another study [3]. One study found no differences between AD and FTD [12] and another study found that cingulate atrophy rates were more suggestive for FTD, whereas hippocampal atrophy rates were more suggestive for AD [9]. Our results, that compared to controls, AD patients progressed over the whole brain, whereas bvFTD lost GM restricted to limbic and paralimbic regions add to the mixed findings in the literature. Even though the percentage changes we found in AD match the findings of another study [8], the percentage change for bvFTD patients in our study was lower. The finding that the right insula declined faster in AD compared to bvFTD and that right frontal lobe only 121

declined faster in AD compared to controls seem not to fit other research results either, as these are typical regions involved in bvFTD. A reason for the more outspoken tissue loss in the right insula and right frontal lobe for AD patients could be a floor effect for the bvFTD group as they display already most atrophy in this area at baseline. An explanation for the finding that AD patients showed the steepest annual rate of atrophy in almost all areas of the brain, could be the underlying pathology, which may alter the speed of progression in our patients. It had been shown that taupositive FTD cases had a longer survival rate than FTLD-ubiquitin immunoreactivechanges (FTLD-U) [32]. It could be possible that our bvFTD group consists of more tau-positive cases, which may slow the progression of atrophy. In the same line of argumentation, as FTD is a very heterogeneous disorder, the slower rate of progression could be caused by the large amount of variation that has been observed in this group even though most of our subjects fulfilled criteria for probable bvFTD [4]. Another reason why our AD patients showed higher rates of progression, could be the fact that the majority of our patients are younger AD cases. Other studies reported higher rates of atrophy in younger familial AD cases compared to older subjects [3335]. Another explanation could be the disease stage of our AD patients. It is possible that while the rate of brain loss accelerates in the early phases of the disease, it may slow as the disease progresses. Finally, it is possible that acceleration in the rate of GM loss was also occurring in the bvFTD group but we have failed to detect it due to limitations in statistical power. Decline of cognitive performances followed the anatomical progression and showed that again AD patients had the steepest decline, especially in in memory and language. Given the fact that AD patients also showed the most progression in atrophy these findings support the notion that functional and anatomical changes are connected to each other, increasing the potential for MRI measures to predict clinical decline. Other studies, however, found that bvFTD showed the fastest decline over time [32,36]. A reason why our results do not agree with these findings could be that the majority of our AD patients are younger AD cases, who decline faster than older patients [37]. At baseline, we observed a posterior gradient of atrophy only for AD patients, with parietal and occipital lobes being more atrophied than controls whereas left and right cingulate gyrus were more suggestive for bvFTD. These results support the posterior signature for AD and a damaged fronto-cingulate-insula pattern for bvFTD with the insula making the best discrimination between the dementias. The atrophy patterns we found for AD and bvFTD at baseline are in accordance with the literature [38-40]. Our measurements showed more atrophy in the frontal cortex for AD patients compared with controls, which has also been found in other studies, especially in younger AD patients [29,41]. Against our expectations bvFTD patients exhibited already on baseline more atrophy in right and left insula and right temporal lobe compared to AD patients. In turn, AD patients did not show any areas of more atrophy compared to bvFTD. This is surprising as the baseline scan was conducted at the first moment patients presented at our memory clinic and were thought to be in the beginning of the disease process. Indeed, other literature reports that in the beginning of the disease atrophy patterns of AD and bvFTD often overlap or hardly show any atrophy [38,42]. One reason for these findings could be that clinical consensus criteria 122

for FTD have changed recently. A requirement for a diagnosis of probable bvFTD is that GM atrophy supporting a diagnosis of bvFTD has to be visible [15]. As most of our bvFTD patients have the diagnosis of probable bvFTD (n=8), the detection of atrophy in insula and temporal lobe is not surprising. Another reason could be the fact that our bvFTD patients are in a later disease stage than our AD patients (6.6 years vs. 3.3 years). BvFTD patients present with behavioral disturbances, which is not immediately recognized by caregivers as a sign for dementia. As a consequence, patients often spend years in the medical circuit before finding their way to a memory clinic. However, the performances on the neuropsychological examination showed that bvFTD did not score worse than AD patients. A question, that remains, is the fact that AD patients perform worse on the cognitive tests even though bvFTD patients had more atrophy. A possible answer could be that the areas affected in bvFTD (cingulate, insula and frontal) are less involved in the cognitive processes covered by our neuropsychological tests but are more involved in the frontal circuits, whose failure lead to behavioral symptoms often seen in bvFTD [43,44]. The affected temporal lobe may explain the lower scores on memory and language that are also seen in AD [45,46]. Among the strengths of this study is the careful use of the longitudinal pipeline of FreeSurfer, which has been shown to be reliable across sites regardless of MRI system differences [47]. Moreover, we used a whole-brain vertex-wise and a ROI approach within FreeSurfer, , which both show comparable results.. A possible limitation of this study is that we did not have pathological data available, so the possibility of misdiagnosis cannot be excluded. Nevertheless, we used an extensive standardized work-up and all 19 AD patients fulfilled clinical criteria of probable AD, 8 bvFTD patients fulfilled the criteria for probable bvFTD and 2 patients fulfilled criteria for possible bvFTD. Availability of pathological data would also enable us to study the effects of the different underlying pathology (tau, TDP subtypes, etc.) on progression of atrophy. This will be an important next step for future investigations. Another limitation could be the fact that we had a relatively small group of bvFTD patients, which could hamper the detection of any putative differences with the other diagnostic groups because of low statistical power. Nevertheless, as bvFTD is a quite rare diagnosis, the sample size corresponds with that of other available longitudinal studies to date. Another limitation of this study is, that we only had one follow-up scan available. With more than one follow-up scan, we could have investigated if the longitudinal progression of atrophy follow a linear or a non-linear trend, i.e. if a certain disease group progressed faster in the beginning or in a later disease stage. Finally, our AD patients consists mainly of younger patients, which may limit the generalizability of our findings. To summarize, our study showed that, compared to controls, AD patients have the steepest decline of GM atrophy and progressed throughout the whole brain, whereas bvFTD patients showed a more selective decline limited to cingulate, frontal and temporal regions. Cognitive decline of AD patients was in accordance with the widespread GM loss. 123

Tables and Figures Table 1. Demographics and neuropsychological performances in z-scores. HC AD bvFTD N 34 19 10 Age at baseline (years) 60.6 ± 6.0 64.6 ± 7.5 64.3 ± 6.8 Sex (% f) 15 (44.1%) 6 (31.6%) 2 (20.0%) Center (% VUmc) 20 (58.8%) 17 (89.5%) 7 (70.0%) b Disease duration (years) 3.3 ± 2.2 6.6 ± 4.4 Level of education 5.5 ± 1.1 5.1 ± 1.2 4.6 ± 1.3 a a MRI follow-up length (years) 2.1 ± 0.2 1.8 ± 0.5 1.5 ± 0.5 a a NPSY follow-up length (years) 2.2 ± 0.2 1.8 ± 0.5 1.7 ± 0.6 3 TIV (cm ) 1514.3 ± 1544.5 ± 1588.6 ± 209.6 178.2 116.0 a a MMSE at baseline 28.7 ± 1.5 23.6 ± 2.9 24.8 ± 3.6 a a Annual rate of decline MMSE -0.00 ± 0.79 -2.19 ± 2.32 -1.62 ± 2.38 a a BL memory 34 0.0 ± 0.7 19 -4.0 ± 1.6 10 -2.1 ± 2.1 a a Annual rate of decline in 34 0.1 ± 0.2 19 -1.9 ± 2.4 9 -1.0 ± 1.7 memory a a BL language 34 -0.1 ± 0.9 19 -1.1 ± 1.7 10 -1.9 ± 2.3 a Annual rate of decline in 34 0.1 ± 0.4 19 -1.5 ± 2.3 9 -0.2 ± 1.2 language BL VSF 20 0.0 ± 1.0 16 -0.9 ± 1.6 7 -0.8 ± 1.3 Annual rate of decline in VSF 20 -0.1 ± 0.5 13 -0.7 ± 1.4 7 0.1 ± 0.7 a BL attention 34 -0.1 ± 0.8 19 -1.7 ± 1.9 10 -1.0 ± 1.0 Annual rate of decline in 34 -0.1 ± 0.2 19 0.0 ± 1.6 9 -0.4 ± 0.6 attention a BL EF 34 0.0 ± 0.8 19 -2.3 ± 2.2 10 -1.1 ± 1.3 Annual rate of decline in EF 34 0.1 ± 0.1 19 0.4 ± 1.5 9 0.1 ± 0.7 Demographics are presented as mean ± standard deviation or n (%). Level of education is determined according to the Verhage-system. Differences between groups for demographics 2 were assessed using ANOVA, Kruskal-Wallis tests and χ tests, where appropriate. Annual rate of decline in MMSE between the groups was compared using ANOVA with age, sex, level of education and disease duration as covariates. Annual rate of NBV loss between the groups was compared using ANOVA with age, sex, center and disease duration as covariates. Neuropsychological performances are presented as mean z-scores ± standard deviation Cognitive composite z-domains were calculated of the available z-scores of each test by the MEAN function in SPSS. Key: NPSY: neuropsychological, MMSE: Mini-Mental State Examination, NBV: normalized total brain volume, VSF: visuospatial functioning, EF: executive functioning a different from controls (p<0.05) b different from AD (p<0.05)

124

Table 2. Thickness of left and right frontal, parietal, temporal occipital, cingulate and insula lobe at baseline, the annual thickness change (mm) and symmetrized percentage change. Lobe HC AD bvFTD a Left frontal BL thickness 2.5 ± 0.1 2.4 ± 0.1 2.4 ± 0.1 a a SPC (%) 0.4 ± 0.9 -0.8 ± 1.1 -0.7 ± 2.6 a Right frontal BL thickness 2.5 ± 0.1 2.4 ± 0.1 2.3 ± 0.3 a SPC (%) 0.2 ± 0.9 -1.2 ± 1.6 0.1 ± 3.2 a Left parietal BL thickness 2.3 ± 0.1 2.2 ± 0.1 2.2 ± 0.1 a SPC (%) -0.2 ± 1.0 -1.5 ± 1.2 -0.3 ± 1.5 a Right parietal BL thickness 2.3 ± 0.1 2.2 ± 0.1 2.2 ± 0.2 a SPC (%) 0.1 ± 1.1 -2.0 ± 1.9 -0.3 ± 2.3 a a Left temporal BL thickness 2.8 ± 0.1 2.6 ± 0.1 2.6 ± 0.2 a a SPC (%) -0.4 ± 1.2 -2.1 ± 1.3 -2.3 ± 2.1 a a Right temporal BL thickness 2.8 ± 0.1 2.7 ± 0.2 2.5 ± 0.3 a a SPC (%) -0.5 ± 1.1 -2.0 ± 1.2 -1.6 ± 2.9 Left occipital BL thickness 2.0 ± 0.1 2.0 ± 0.1 2.0 ± 0.1 SPC (%) -0.3 ± 1.1 -0.9 ± 1.8 -0.5 ± 0.9 Right occipital BL thickness 2.0 ± 0.1 2.0 ± 0.1 2.0 ± 0.2 SPC (%) -0.2 ± 1.0 -0.9 ± 1.9 -0.6 ± 0.2 Left cingulate BL thickness 2.6 ± 0.2 2.5 ± 0.2 2.4 ± 0.1 a a SPC (%) 0.1 ± 1.1 -1.1 ± 2.1 -1.2 ± 2.7 a Right cingulate BL thickness 2.5 ± 0.1 2.4 ± 0.2 2.3 ± 0.2 a a SPC (%) 0.4 ± 1.1 -1.6 ± 2.0 -1.1 ± 3.5 a Left insula BL thickness 3.0 ± 0.1 2.9 ± 0.2 2.8 ± 0.2 SPC (%) -0.4 ± 0.9 -1.1 ± 2.1 -1.2 ± 2.5 a,b Right insula BL thickness 2.9 ± 0.1 2.9 ± 0.2 2.5 ± 0.4 b SPC (%) -0.2 ± 0.8 -2.1 ± 1.0 0.5 ± 5.3 Values are presented as mean mm ± standard deviation. MANOVA’s with Bonferroni posthoc tests and age, sex, center and disease duration as covariates. SPC: symmetrized percentage change per year. a b different from HC p<0.05; different from AD p<0.05

125

Figure 2. Patterns of atrophy at baseline. (A) Blue areas show regions of less gray matter in AD patients compared to controls. (B) Blue areas show regions of less gray matter in bvFTD patients compared to controls. (C) Blue areas show regions of less gray matter in AD patients compared to bvFTD patients. Red areas show regions of less gray matter in bvFTD patients compared to AD patients. (A)

(B)

(C)

127

Figure 3. Progression of atrophy. (A) Blue areas show regions of more cortical thinning in AD patients compared to controls. (B) Blue areas show regions of more cortical thinning in bvFTD patients compared to controls. (C) Blue areas show regions of more cortical thinning in AD patients compared to bvFTD patients. Red areas show regions of more cortical thinning in bvFTD patients compared to AD patients. (A)

(B)

(C)

128

Figure 4. Loss of thickness in left and right frontal, parietal, temporal, occipital, cingulate and insula for healthy controls, AD and bvFTD patients. MRI follow up time was 2.1 years for controls, 1.8 years for AD and 1.5 years for bvFTD patients.

ANOVA for repeated measures with time, lobe and hemisphere as within-subjects factors, diagnosis as between-subjects factor, and age, sex, TIV, disease duration, center and follow-up time as covariates was conducted. Cortical thickness measurements of all left and right lobes were dependent variables. As posthoc test, we conducted MANOVA with ‘annual rate of atrophy’ (rate=(thickness follow-up MRI – thickness baseline MRI)/ follow-up time) as dependent variable, diagnosis as independent variable and age, sex, TIV, disease duration, and center as covariates to assess the difference in progression of atrophy between the groups for each lobe, left and right separately. Repeated measures showed that regional distribution of cortical thinning over time was different in the two hemispheres for the three diagnostic groups. AD patients lost most of the cortical thickness in all cortices, except in left temporal and left cingulate, where bvFTD (left temporal: 2.3%; left cingulate: 1.2%) patients lost most of cortical thickness. 129

Especially in right frontal (1.2%) and biparietal (left: 1.5%; right: 2.0%) lobes, AD patients showed a steeper rate of cortical thinning than controls, whereas bvFTD patients did not differ from controls. In left frontal, bitemporal and bicingulate, both, AD (left frontal: 0.8%; left temporal: 2.1%; right temporal: 2.0%; left cingulate: 1.1%; right cingulate: 1.6%) and bvFTD patients (left frontal: 0.7%; left temporal: 2.3%; right temporal: 1.6%; left cingulate: 1.2%; right cingulate: 1.1%) progressed harder than controls (left frontal: 0.4%; left temporal: 0.4%; right temporal: 0.5%; left cingulate: 0.1%; right cingulate: +0.4%). For the right insula, AD patients (2.1%) lost more cortical thickness than bvFTD (+0.5%)and controls (0.2%). There were no differences in progression of cortical thinning for bioccipital and left insula.

130

Reference List [1] [2]

[3]

[4]

[5]

[6]

[7]

[8]

[9]

[10]

[11]

[12]

[13]

[14]

[15]

Ratnavalli E, Brayne C, Dawson K, Hodges JR (2002) The prevalence of frontotemporal dementia. Neurology 58, 1615-1621. Sluimer JD, van der Flier WM, Karas GB, van SR, Barnes J, Boyes RG, Cover KS, et al. (2009) Accelerating regional atrophy rates in the progression from normal aging to Alzheimer's disease. Eur Radiol 19, 2826-2833. Frings L, Yew B, Flanagan E, Lam BY, Hull M, Huppertz HJ, Hodges JR, Hornberger M (2014) Longitudinal grey and white matter changes in frontotemporal dementia and Alzheimer's disease. PLoS One 9, e90814Whitwell JL, Anderson VM, Scahill RI, Rossor MN, Fox NC (2004) Longitudinal patterns of regional change on volumetric MRI in frontotemporal lobar degeneration. Dement Geriatr Cogn Disord 17, 307-310. Brambati SM, Renda NC, Rankin KP, Rosen HJ, Seeley WW, Ashburner J, Weiner MW, Miller BL, Gorno-Tempini ML (2007) A tensor based morphometry study of longitudinal gray matter contraction in FTD. Neuroimage 35, 998-1003. Jack CR, Jr., Slomkowski M, Gracon S, Hoover TM, Felmlee JP, Stewart K et al. (2003) MRI as a biomarker of disease progression in a therapeutic trial of milameline for AD. Neurology 60, 253-260. Fox NC, Black RS, Gilman S, Rossor MN, Griffith SG, Jenkins L, Koller M (2005) Effects of Abeta immunization (AN1792) on MRI measures of cerebral volume in Alzheimer disease. Neurology 64, 1563-1572. Krueger CE, Dean DL, Rosen HJ, Halabi C, Weiner M, Miller BL, Kramer JH (2010) Longitudinal rates of lobar atrophy in frontotemporal dementia, semantic dementia, and Alzheimer's disease. Alzheimer Dis Assoc Disord 24, 43-48. Barnes J, Godbolt AK, Frost C, Boyes RG, Jones BF, Scahill RI, Rossor MN, Fox NC (2007) Atrophy rates of the cingulate gyrus and hippocampus in AD and FTLD. Neurobiol Aging 28, 20-28. Whitwell JL, Jack CR, Jr., Parisi JE, Knopman DS, Boeve BF, Petersen RC, Ferman TJ, Dickson DW, Josephs KA (2007) Rates of cerebral atrophy differ in different degenerative pathologies. Brain 130, 1148-1158. Chan D, Fox NC, Jenkins R, Scahill RI, Crum WR, Rossor MN (2001) Rates of global and regional cerebral atrophy in AD and frontotemporal dementia. Neurology 57, 1756-1763. Whitwell JL, Jack CR, Jr., Pankratz VS, Parisi JE, Knopman DS, Boeve BF, Petersen RC, Dickson DW, Josephs KA (2008) Rates of brain atrophy over time in autopsyproven frontotemporal dementia and Alzheimer disease. Neuroimage 39, 1034-1040. McKhann G, Drachman D, Folstein M, Katzman R, Price D, Stadlan EM (1984) Clinical diagnosis of Alzheimer's disease: report of the NINCDS-ADRDA Work Group under the auspices of Department of Health and Human Services Task Force on Alzheimer's Disease. Neurology 34, 939-944. McKhann GM, Knopman DS, Chertkow H, Hyman BT, Jack CR, Jr., Kawas CH et al. (2011) The diagnosis of dementia due to Alzheimer's disease: recommendations from the National Institute on Aging-Alzheimer's Association workgroups on diagnostic guidelines for Alzheimer's disease. Alzheimers Dement 7, 263-269. Rascovsky K, Hodges JR, Knopman D, Mendez MF, Kramer JH, Neuhaus J, et al. (2011) Sensitivity of revised diagnostic criteria for the behavioural variant of frontotemporal dementia. Brain 134, 2456-2477. 131

[16] [17] [18]

[19] [20] [21] [22] [23] [24]

[25]

[26] [27]

[28]

[29]

[30] [31]

[32]

[33]

132

Verhage F (1964) Intelligentie en leeftijd: onderzoek bij Nederlanders van twaalf tot zevenenzeventig jaar., Van Gorcum, Assen. Stroop JR (1935) Studies of interference in serial verbal reactions. J Exp Psychol 18, 643662. Smits LL, Pijnenburg YA, Koedam EL, van der Vlies AE, Reuling IE, Koene T, Teunissen CE, Scheltens P, van der Flier WM (2012) Early Onset Alzheimer's Disease is Associated with a Distinct Neuropsychological Profile. J Alzheimers Dis 30, 101108. Reuter M, Schmansky NJ, Rosas HD, Fischl B (2012) Within-subject template estimation for unbiased longitudinal image analysis. Neuroimage 61, 1402-1418. Dale AM, Fischl B, Sereno MI (1999) Cortical surface-based analysis. I. Segmentation and surface reconstruction. Neuroimage 9, 179-194. Fischl B, Sereno MI, Dale AM (1999) Cortical surface-based analysis. II: Inflation, flattening, and a surface-based coordinate system. Neuroimage 9, 195-207. Reuter M , Fischl B (2011) Avoiding asymmetry-induced bias in longitudinal image processing. Neuroimage 57, 19-21. Reuter M, Rosas HD, Fischl B (2010) Highly accurate inverse consistent registration: a robust approach. Neuroimage 53, 1181-1196. Desikan RS, Segonne F, Fischl B, Quinn BT, Dickerson BC, Blacker D, Buckner RL, Dale AM, Maguire RP, Hyman BT, Albert MS, Killiany RJ (2006) An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest. Neuroimage 31, 968-980. Seeley WW, Crawford R, Rascovsky K, Kramer JH, Weiner M, Miller BL, Gorno-Tempini ML (2008) Frontal paralimbic network atrophy in very mild behavioral variant frontotemporal dementia. Arch Neurol 65, 249-255. Seeley WW (2010) Anterior insula degeneration in frontotemporal dementia. Brain Struct Funct 214, 465-475. Pan PL, Song W, Yang J, Huang R, Chen K, Gong QY, Zhong JG, Shi HC, Shang HF (2012) Gray matter atrophy in behavioral variant frontotemporal dementia: a metaanalysis of voxel-based morphometry studies. Dement Geriatr Cogn Disord 33, 141-148. Lehmann M, Rohrer JD, Clarkson MJ, Ridgway GR, Scahill RI, Modat M, Warren JD, Ourselin S, Barnes J, Rossor MN, Fox NC (2010) Reduced cortical thickness in the posterior cingulate gyrus is characteristic of both typical and atypical Alzheimer's disease. J Alzheimers Dis 20, 587-598. Mรถller C, Vrenken H, Jiskoot L, Versteeg A, Barkhof F, Scheltens P, van der Flier WM (2013) Different patterns of gray matter atrophy in early- and late-onset Alzheimer's disease. Neurobiol Aging 34, 2014-2022. Reuter M, Schmansky NJ, Rosas HD, Fischl B (2012) Within-subject template estimation for unbiased longitudinal image analysis. Neuroimage 61, 1402-1418. Nordenskjold R, Malmberg F, Larsson EM, Simmons A, Brooks SJ, Lind L, Ahlstrom H, Johansson L, Kullberg J (2013) Intracranial volume estimated with commonly used methods could introduce bias in studies including brain volume measurements. Neuroimage 83, 355-360. Rascovsky K, Salmon DP, Lipton AM, Leverenz JB, DeCarli C, Jagust WJ, Clark CM, Mendez MF, Tang-Wai DF, Graff-Radford NR, Galasko D (2005) Rate of progression differs in frontotemporal dementia and Alzheimer disease. Neurology 65, 397-403. Sluimer JD, Vrenken H, Blankenstein MA, Fox NC, Scheltens P, Barkhof F, van der Flier WM (2008) Whole-brain atrophy rate in Alzheimer disease: identifying fast progressors. Neurology 70, 1836-1841.

[34]

[35]

[36]

[37]

[38] [39]

[40] [41]

[42]

[43]

[44] [45]

[46]

[47]

Wang D, Chalk JB, Rose SE, de ZG, Cowin G, Galloway GJ, Barnes D, Spooner D, Doddrell DM, Semple J (2002) MR image-based measurement of rates of change in volumes of brain structures. Part II: application to a study of Alzheimer's disease and normal aging. Magn Reson Imaging 20, 41-48. Fox NC , Freeborough PA (1997) Brain atrophy progression measured from registered serial MRI: validation and application to Alzheimer's disease. J Magn Reson Imaging 7, 1069-1075. Smits LL, van Harten AC, Pijnenburg YA, Koedam EL, Bouwman FH, Sistermans N, Reuling IE, Prins ND, Lemstra AW, Scheltens P, van der Flier WM (2014) Trajectories of cognitive decline in different types of dementia. Psychol Med, 19. van der Vlies AE, Koedam EL, Pijnenburg YA, Twisk JW, Scheltens P, van der Flier WM (2009) Most rapid cognitive decline in APOE epsilon4 negative Alzheimer's disease with early onset. Psychol Med 39, 1907-1911. Rohrer JD (2012) Structural brain imaging in frontotemporal dementia. Biochim Biophys Acta 1822, 325-332. Rabinovici GD, Seeley WW, Kim EJ, Gorno-Tempini ML, Rascovsky K, Pagliaro TA, Allison SC, Halabi C, Kramer JH, Johnson JK, Weiner MW, Forman MS, Trojanowski JQ, Dearmond SJ, Miller BL, Rosen HJ (2007) Distinct MRI atrophy patterns in autopsy-proven Alzheimer's disease and frontotemporal lobar degeneration. Am J Alzheimers Dis Other Demen 22, 474-488. Raji CA, Lopez OL, Kuller LH, Carmichael OT, Becker JT (2009) Age, Alzheimer disease, and brain structure. Neurology 73, 1899-1905. Frisoni GB, Pievani M, Testa C, Sabattoli F, Bresciani L, Bonetti M, Beltramello A, Hayashi KM, Toga AW, Thompson PM (2007) The topography of grey matter involvement in early and late onset Alzheimer's disease. Brain 130, 720-730. Whitwell JL , Jack CR, Jr. (2005) Comparisons between Alzheimer disease, frontotemporal lobar degeneration, and normal aging with brain mapping. Top Magn Reson Imaging 16, 409-425. Mรถller C, Dieleman N, van der Flier WM, Versteeg A, Pijnenburg Y, Scheltens P, Barkhof F, Vrenken H (2014) More Atrophy of Deep Gray Matter Structures in Frontotemporal Dementia Compared to Alzheimer's Disease. J Alzheimers Dis Cummings JL (1993) Frontal-subcortical circuits and human behavior. Arch Neurol 50, 873-880. Hurley RS, Bonakdarpour B, Wang X, Mesulam MM (2014) Asymmetric Connectivity between the Anterior Temporal Lobe and the Language Network. J Cogn Neurosci, 1-10. Smits LL, Tijms BM, Benedictus MR, Koedam EL, Koene T, Reuling IE, Barkhof F, Scheltens P, Pijnenburg YA, Wattjes MP, van der Flier WM (2014) Regional atrophy is associated with impairment in distinct cognitive domains in Alzheimer's disease. Alzheimers Dement 10, S299-S305. Jovicich J, Marizzoni M, Sala-Llonch R, Bosch B, Bartres-Faz D, Arnold J et al. (2013) Brain morphometry reproducibility in multi-center 3T MRI studies: a comparison of cross-sectional and longitudinal segmentations. Neuroimage 83, 472-484.

133

Chapter 3.4

Joint assessment of white matter integrity, cortical and subcortical atrophy to distinguish AD from behavioral variant FTD: a twocenter study 1*

2-4

Christiane Möller MSc , Anne Hafkemeijer MSc , Yolande A L 1 2-4 Pijnenburg MD PhD , Serge A R B Rombouts PhD , Jeroen van der 3 1,3,5,6 5,6 Grond PhD , Elise Dopper , John van Swieten MD PhD , Adriaan 7 8 7 Versteeg , Petra J W Pouwels PhD , Frederik Barkhof MD PhD , Philip 1 7,8 Scheltens MD PhD , Hugo Vrenken PhD , Wiesje M van der Flier 1,9 PhD 1 Alzheimer Center & Department of Neurology, Neuroscience Campus Amsterdam, VU University medical center, Amsterdam, the Netherlands, 2 Institute of Psychology, Leiden University, Leiden, the Netherlands, 3 Department of Radiology, Leiden University Medical Center, Leiden, the 4 Netherlands, Leiden Institute for Brain and Cognition, Leiden University, 5 Leiden, the Netherlands, Department of Clinical Genetics, Neuroscience Campus Amsterdam, VU University medical center, Amsterdam, the 6 Netherlands, Department of Neurology, Erasmus Medical Center, 7 Rotterdam, the Netherlands, Department of Radiology & Nuclear Medicine, Neuroscience Campus Amsterdam, VU University medical 8 center, Amsterdam, the Netherlands, Department of Physics & Medical Technology, Neuroscience Campus Amsterdam, VU University medical 9 center, Amsterdam, the Netherlands, Department of Epidemiology & Biostatistics, Neuroscience Campus Amsterdam, VU University medical center, Amsterdam, the Netherlands. Major revisions NeuroImage:Clinical

Highlights • There are gray and clear white matter differences between AD and bvFTD • Gray matter atrophy contributed most to distinguish controls from AD and bvFTD patients. • White matter integrity measurements contributed most to distinguish bvFTD from controls and AD. • White matter integrity measures support the hypotheses of a network disorder in bvFTD. • White matter integrity measures allow more precise differentiation between AD and bvFTD.

134

Abstract We investigated the ability of cortical and subcortical gray matter (GM) atrophy in combination with white matter (WM) integrity to distinguish behavioral variant frontotemporal dementia (bvFTD) from Alzheimer’s disease (AD) and from controls using voxel-based morphometry, subcortical structure segmentation, and tract-based spatial statistics. To determine which combination of MR markers differentiated the three groups with the highest accuracy, we conducted discriminant function analyses. Adjusted for age, sex and center, both types of dementia had more GM atrophy, lower fractional anisotropy (FA) and higher mean (MD), axial (L1) and radial diffusivity (L23) values than controls. BvFTD patients had more GM atrophy in orbitofrontal and inferior frontal areas than AD patients. In addition, caudate nucleus and nucleus accumbens were smaller in bvFTD than in AD. FA values were lower, MD, L1 and L23 values were higher, especially in frontal areas of the brain for bvFTD compared to AD patients. The combination of cortical GM, hippocampal volume and WM integrity measurements, classified 97-100% of controls, 81-100% of AD and 67-75% of bvFTD patients correctly. Our results suggest that WM integrity measures add complementary information to measures of GM atrophy, thereby improving the classification between AD and bvFTD. Key words: Alzheimer’s disease, frontotemporal dementia, gray matter atrophy, white matter integrity, discriminant analyses, diagnosis

135

Introduction Alzheimer’s disease (AD) and behavioral variant Frontotemporal dementia (bvFTD) are the most common causes of young onset dementia [1]. Clinical diagnostic criteria have been proposed [2,3], but the frequent overlap of the clinical symptoms associated with AD and bvFTD pose serious problems in the differential diagnosis. Although the definite diagnosis of both types of dementia is only possible at autopsy, magnetic resonance imaging (MRI), providing measurements of gray matter (GM) atrophy and white matter (WM) integrity, have been shown to detect brain changes in an early disease stage. Studies on GM atrophy have shown precuneus, lateral parietal and occipital cortices to be more atrophic in AD than in bvFTD, whereas atrophy of anterior cingulate, anterior insula, subcallosal gyrus, and caudate nucleus was more severe in bvFTD compared to AD [4-6]. GM loss in dorsolateral prefrontal cortex, medial temporal lobes, hippocampus and amygdala is found in both AD and bvFTD and does not help to discriminate between the two disorders [4,7-9]. In addition to local GM damage, a decrease of FA in WM, suggesting WM tract damage has been shown, especially in bvFTD. Previous studies showed that compared to AD, WM integrity was lost in bvFTD especially in the frontal and bilateral temporal regions [10,11]. Most former studies focused on either GM or WM damage, while only a few investigated the extent to which the loss of WM integrity and GM atrophy are related and how they jointly contribute to the clinical classification of patients [12-14]. Generalizability of these findings is limited as in one study patients from the whole FTLD spectrum were compare to AD patients [12] and in other studies the different imaging modalities were only linked to each other but not used for diagnostic discrimination [13,14]. In this multi-center study we compared patterns of cortical and subcortical GM atrophy and of WM integrity between patients with bvFTD, AD and controls with the ultimate goal to facilitate clinical diagnosis. In addition, we investigated the joint discriminative ability of GM atrophy and WM integrity measurement to distinguish both patient groups from controls and from each other.

Materials and Methods Patients In this two center study, we included 39 patients with probable AD and 30 patients with bvFTD, who visited either the Alzheimer Center of the VU University medical center (VUmc) (probable AD: n=23; probable bvFTD: n=16; possible bvFTD: n=4) or the Alzheimer Center of the Erasmus University Medical Center Rotterdam (probable AD: n=16 ; probable bvFTD: n=9; possible bvFTD: n=1). All patients underwent a standardized one-day assessment including medical history, informant-based history, physical and neurological examination, blood tests, neuropsychological assessment, and MRI of the brain. Diagnoses were made in a multidisciplinary consensus meeting according to the core clinical criteria of the National Institute on Aging and the Alzheimer’s Association workgroup for probable AD [3,15] and according to the clinical diagnostic criteria of FTD for bvFTD [2]. To minimize center effects, all diagnoses were re-evaluated in a panel including clinicians from both centers. In 136

addition, we included 41 cognitively normal controls (VUmc: n=23; Rotterdam: n=18), who were recruited by advertisement in local newspapers. Before inclusion in the present study, controls were screened for memory complaints, family history of dementia, drugs- or alcohol abuse, major psychiatric disorder, and neurological or cerebrovascular diseases. They underwent an assessment including medical history, physical examination, neuropsychological assessment, and MRI of the brain comparable to the work-up of patients. Inclusion criteria for both cohorts were: (1) availability of a T1-weighted 3-dimensional MRI (3DT1) scan and a diffusion tensor imaging (DTI) image at 3T, and (2) age between 50 and 80 years. Exclusion criteria were: (1) large image artifacts (n=12); (2) failure of imaging analyzing software to process MR scans (n=6) (details see sections below); and (3) gross brain pathology other than atrophy, including severe white matter hyperintensities and/or lacunar infarction in deep gray matter structures. Level of education was rated on a seven-point scale [16]. The study was conducted in accordance with regional research regulations and conformed to the Declaration of Helsinki. The local medical ethics committee of both centers approved the study. All patients gave written informed consent for their clinical and biological data to be used for research purposes. MR image acquisition and review Imaging at the VUmc was carried out on a 3T scanner (Signa HDxt, GE Healthcare, Milwaukee, WI, USA), using an 8-channel head coil with foam padding to restrict head motion. Patients and controls from the Erasmus University Medical Center Rotterdam were all scanned at the Leiden University Medical Center (LUMC). Imaging at LUMC was performed on a 3T scanner (Achieva, Philips Medical Systems, Best, the Netherlands) using an 8-channel SENSE head coil. The scan protocol included a whole-brain near-isotropic 3DT1-weighted sequence for cortical and subcortical segmentation. At VUmc this was a fast spoiled gradient echo sequence (FSPGR; repetition time TR 7.8 ms, echo time TE 3 ms, inversion time TI 450 ms, flip angle 12º, 180 sagittal slices, voxel size 0.98x0.98x1 mm, total scan time 4.57 minutes). At LUMC this was a turbo field echo sequence (T1TFE; TR 9.8 ms, TE 4.6 ms, flip angle 8°, 140 transversal slices, voxel size 0.88x0.88x1.2 mm, total scan time 4.57 minutes). In addition DTI was performed using EPI based sequences. At the VUmc, 2 DTI consisted of five volumes without directional weighting (i.e. b=0 s/mm ) and 30 2 volumes with noncollinear diffusion gradients (i.e. 30 directions, b=1000 s/mm ) and TR 13000 ms, TE 87.8 ms, 45 contiguous axial slices of 2.4 mm, voxel size=2x2x2.4mm, parallel imaging with factor 2, total scan time 7.8 minutes. At the LUMC DTI consisted 2 of one volume without directional weighting (i.e. b=0 s/mm ) and 60 volumes with 2 noncollinear diffusion gradients (i.e. 60 directions, b=1000 s/mm ) and TR 8250 ms, TE 80 ms, 70 axial slices, voxel size=2x2x2mm, parallel imaging with factor 2, total scan time 9 minutes). In addition, the MRI protocol included a 3D Fluid Attenuated Inversion Recovery (FLAIR) sequence, dual-echo T2-weighted sequence, and susceptibility weighted imaging (SWI) which were reviewed for brain pathology other than atrophy by an experienced radiologist. 137

Gray matter volume DICOM images of the 3DT1- weighted sequence were corrected for gradient nonlinearity distortions and converted to Nifti format. The linear transformation matrix to MNI space was calculated using FSL-FLIRT [17] and used to place the image coordinate origin (0,0,0) on the anterior commissure by using the Nifti s-form. The structural 3DT1 images were then analyzed using the voxel-based morphometry toolbox (VBM8; version 435; University of Jena, Department of Psychiatry) in Statistical Parametric Mapping (SPM8; Functional Imaging Laboratory, University College London, London, UK) implemented in MATLAB 7.12 (MathWorks, Natick, MA). The first module of the VBM8 Toolbox (“Estimate and Write”) segments the 3DT1 volumes into GM, WM and cerebrospinal fluid (CSF), apply a registration to MNI space (affine) and subsequently a non-linear deformation. The non-linear deformation parameters are calculated via the high dimensional DARTEL algorithm and the MNI 152 template. Remaining non-brain tissue was removed by the ‘light clean-up’ option. Tissue classes are normalized in alignment with the template with the ‘non-linear only’ option which allows comparing the absolute amount of tissue corrected for individual brain size. The correction is applied directly to the data, which makes a head-size correction to the statistical model redundant. In the second module, images were smoothed using a 8 mm full width at half maximum (FWHM) isotropic Gaussian kernel. Images were visually inspected after every processing step. Voxelwise statistical comparisons between groups were made to localize GM differences by means of a full factorial design with diagnosis (AD, bvFTD, controls) as factor with independent levels with unequal variance, using absolute threshold masking with a threshold of 0.1 and implicit masking. Age, sex and center were entered as covariates. Post hoc, we compared AD with controls, bvFTD with controls, and AD with bvFTD. The threshold for significance in all VBM analyses was set to p<0.05 with family wise error correction (FWE) at the voxel level and an extent threshold of 0 voxels. Volumes of deep gray matter (DGM) structures The algorithm FIRST (FMRIB’s integrated registration and segmentation tool) [18] was applied to estimate left and right volumes of seven structures: hippocampus, amygdala, thalamus, caudate nucleus, putamen, globus pallidus, and nucleus accumbens. Left and right volumes were summed to obtain total volume for each structure. FIRST is integrated in FMRIB’s software library (FSL 4.15) [19] and performed both registration and segmentation of the above mentioned anatomical structures. A two-stage linear registration was performed to achieve a more robust and accurate pre-alignment of the seven structures. During the first-stage registration, the 3DT1 images were registered linearly to a common space based on the Montreal Neurological Institute (MNI) 152 template with 1x1x1 mm resolution using 12 degrees of freedom. After registration, a second stage registration using a subcortical mask or weighting image, defined in MNI space, was performed to improve registration for the seven structures. Both stages used 12 degrees of freedom. This 2-stage registration was followed by segmentation based on shape models and voxel intensities. Volumes of the seven structures were extracted in native space, taking into account the transformations matrices during registration. The final 138

step was a boundary correction based on local signal intensities. All registrations and segmentations were visually checked for errors. To correct the volumes of the DGM structures for head size we used a volumetric scaling factor (VSF) derived from the normalization transform matrix from SIENAX (Structural Image Evaluation using Normalization of Atrophy Cross-sectional) [20], also part of FSL. In short, SIENAX extracted skull and brain from the 3DT1 input whole-head image. In our study, brain extraction was performed using optimized parameters [21]. These were then used to register the subject’s brain and skull image to standard space brain and skull (derived from MNI152 template) to estimate the scaling factor (VSF) between the subject’s image and standard space. Normalization for head size differences was done by multiplying the raw volumes of the DGM structures by the VSF. Next to the VSF, we also obtained brain tissue volumes of GM and WM [22]. Total volumes of the seven DGM structures and volumes of GM and WM, and VSF were transferred to SPSS for further statistical analyses. White matter integrity All preprocessing steps were performed using FSL [19,23], including motion- and eddy-current correction on images and gradient-vectors, followed by diffusion tensor fitting. Fractional anisotropy (FA), mean diffusivity (MD), axial diffusivity (L1), and radial diffusivity (average of L2 and L3, L23) were derived for each voxel. Each subject’s FA image was used to calculate nonlinear registration parameters to the FMRIB58_FA brain, which were then applied to all four parameter images. The registered FA images were averaged into a mean FA image, which was skeletonized for tract-based spatial statistics (TBSS) [24]. The skeleton was thresholded at 0.2 to include only WM and used for TBSS statistics in all diffusion parameters. Each subject's aligned FA data was then projected onto this skeleton and the resulting data fed into voxelwise cross-subject statistics. The projection parameters for each voxel were then also applied to the MD, L1 and L23 data to create skeletonized data in standard space for each subject. Differences in FA, MD, L1 and L23 between controls, AD and bvFTD patients were analyzed in a voxelwise fashion using FSL’s randomise with age, sex and center as covariates. A family wise error (FWE) corrected ThresholdFree Cluster Enhancement (TFCE) significance level of p<0.05 was used to correct for multiple comparisons. Extraction of regions of interest (ROI) As a next step, we extracted ROIs from the VBM and TBSS group analyses, to be able to combine the most promising MR markers in one statistical model. Gray matter: From the resulting T-map from the comparisons AD<controls, bvFTD<controls, and bvFTD<AD from the VBM analyses all significant voxels were extracted as one ROI per groups comparison (in total 3 ROIs: GM ROI AD<Controls, GM ROI bvFTD<Controls, GM ROI bvFTD<AD). This was done by merging the normalized modulated GM segments of all subjects into a 4D file. The T-maps of all contrasts were thresholded at p<0.05 (FWE corrected) and binarised. We then calculated the mean GM fraction of every voxel in the ROI and the size of the ROI 3 (voxels and mm ). By multiplying the GM fraction with the size of the ROI we 139

maintained the GM volume of the whole ROI which was transferred to SPSS for further analysis. White matter integrity: We calculated the mean FA value under the total skeleton using the fslstats function from FSL. Significant voxels (TFCE, FWE corrected p<0.05) from the contrast (statistical) images from the two-group comparisons AD<controls, bvFTD<controls, AD<bvFTD, and bvFTD<AD were extracted in a ROI (in total 3 ROIs: FA ROI AD<Controls, FA ROI bvFTD<Controls, FA ROI bvFTD<AD) using fslmaths and 3 fslstats. We then calculated the mean FA and region size (voxel count and mm ) in the ROI. The same was done for MD, L1 and L23. For these measurements we used the comparisons AD>controls, bvFTD>controls, bvFTD>AD to create the ROIs (in total 3 per diffusivity measurement: MD ROI AD>controls, MD ROI bvFTD>controls, MD ROI bvFTD>AD; L1 ROI AD>controls, L1 ROI bvFTD>controls, L1 ROI bvFTD>AD; L23 ROI AD>controls, L23 ROI bvFTD>controls, L23 ROI bvFTD>AD). Values of the 12 ROIs were transferred to SPSS for further analysis. Statistical analysis SPSS version 20.0 for Windows was used for statistical analysis. Differences between groups were assessed using univariate analysis of variance (ANOVA), Kruskal-Wallis tests and χ2 tests, where appropriate. Multivariate analysis of variance (MANOVA) was used to compare raw total volumes of the medial temporal lobe structures hippocampus and amygdala, and of the DGM structures thalamus, caudate nucleus, putamen, globus pallidus, and nucleus accumbens (dependent variables) between the different diagnostic groups (between-subjects factor). Post hoc, ANOVAs and Bonferroni adjusted t-tests were performed with age, sex and center as covariates. To determine which combination of MR markers based on VBM, DGM structures and TBSS measurements differentiated the three patient groups with the highest accuracy, we conducted a discriminant function analysis with leave-one-out cross validation. As predictors we entered the following variables: three GM ROIs from the comparison AD<controls, bvFTD<controls, bvFTD<AD; DGM volumes of hippocampus, thalamus, caudate nucleus, putamen and nucleus accumbens (as these structures significantly differed between the groups); three FA ROIs from the comparison AD<controls, bvFTD<controls, bvFTD<AD; as well as sex, age, and center. Because of colinearity we performed another discriminant function analyses with the other diffusion parameters L1 and L23 instead of FA. In this discriminant function we used the following variables as predictors: three GM ROIs from the comparison AD<controls, bvFTD<controls, bvFTD<AD; DGM volumes of hippocampus, thalamus, caudate nucleus, putamen and nucleus accumbens; six L1 and L23 ROIs from the comparisons AD>controls, bvFTD>controls, bvFTD>AD; as well as sex, age, and center. In general, a discriminant analysis creates k-1 linear combinations (discriminant functions) of the entered predictor variables which provides the best discrimination between the groups (k). To identify the most optimal combination of variables for best discrimination, stepwise forward analysis was used with a decision scheme based on the F-value of Wilks’ lambda (entry: 3.84; removal: 2.71). Statistical significance for all analyses was set at p<0.05.

140

Results Demographics Demographic data for all patients (AD: n=32; bvFTD: n=24) and controls (n=37) fulfilling inclusion criteria are summarized in Table 1. AD patients were older than controls (p<0.001); there were no differences in gender distribution or education. Both dementia groups had smaller normalized brain volumes than controls (p<0.001). AD patients had lower MMSE scores than both other groups (p<0.05). CDR and GDS scores were lowest in controls (p<0.001) but did not differ between the two dementia groups. Gray matter volume The full factorial design showed main effects of diagnosis (Figure 1). Post hoc comparisons showed that compared to controls, AD patients showed a reduction of GM in superior and middle temporal gyrus, parahippocampal gyrus, hippocampus, posterior cingulate, cuneus, precuneus, superior parietal lobule and inferior frontal gyrus (p<0.05, FWE corrected). BvFTD patients had less GM compared to controls in superior, middle, and inferior frontal gyrus, orbito-frontal gyrus, insular, temporal gyrus, parahippocampal gyrus and hippocampus. Controls did not show any regions with less GM than AD or bvFTD (p<0.05, FWE corrected). Compared to AD patients, bvFTD patients had less GM matter in left inferior and medial frontal gyrus, in right inferior frontal gyrus, and in orbito-frontal gyrus (p<0.05, FWE corrected). AD patients did not show any regions of significantly reduced GM compared to bvFTD patients. Volumes of deep gray matter structures Normalized volumes of MTL and DGM structures are summarized in Table 2. MANOVA adjusted for age, sex and center revealed group differences in hippocampus, thalamus, caudate nucleus, putamen and nucleus accumbens (Figure 2). Post hoc tests showed that nucleus accumbens and caudate nucleus volume discriminated all groups, with bvFTD having most severe atrophy. Hippocampus and thalamus discriminated dementia patients from controls. bvFTD patients had smaller putaminal volumes than controls. White matter integrity Figure 3 shows the mean skeleton with significant regions in FA, MD, L1 and L23 for different group comparisons. Compared with controls, AD patients showed widespread patterns of lower FA values, incorporating 44% of the WM skeleton voxels, in areas including the fornix, corpus callosum, forceps minor, thalamus, posterior thalamic radiation, superior and inferior longitudinal fasciculus. Furthermore, they had higher MD values in 36% of the WM skeleton voxels including the fornix, corpus callosum, forceps minor and forceps major, higher L1 values in 23% of the WM skeleton voxels including the corpus callosum, the corticospinal tract and inferior longitudinal fasciculus, and higher L23 values in 42% of the WM skeleton voxels including the forceps major, inferior fronto-occipital fasciculus, inferior and superior longitudinal fasciculus and the corpus callosum compared with controls. Compared to controls, bvFTD patients showed widespread patterns of lower FA values in 58% of the investigated WM voxels throughout the whole brain, in areas 141

including the fornix, corpus callosum, forceps minor, thalamus, anterior thalamic radiation, superior and inferior longitudinal fasciculus and inferior fronto-occipital fasciculus. Furthermore, they had higher MD values in 55% of the investigated WM voxels including the inferior fronto-occipital fasciculus, uncinate fasciculus and the forceps minor, higher L1 values in 39% of the WM skeleton voxels including the inferior fronto-occiptial fasciculus, inferior longitudinal fasciculus, corticospinal tract and corpus callosum, and higher L23 values in 62% of the investigated WM voxels in the inferior and superior longitudinal fasciculus, corticospinal tract, corpus callosum, fornix, inferior fronto-occiptial fasciculus and the anterior thalamic radiation compared to controls. In direct comparison between the two dementia groups, bvFTD patients had lower FA values in 17% of the investigated voxels, solely located in the frontal parts of the brain, like the rostrum and the genu of the corpus callosum, forceps minor, anterior part of the internal and external capsule, anterior parts of the fronto-occipital fasciculus and superior longitudinal fasciculus. Furthermore, bvFTD patients had higher MD values in 21% and higher radial diffusivity values in 23% of the investigated WM voxels including forceps minor, uncinate fasciculus, inferior fronto-occipital fasciculus and anterior thalamic radiation, higher axial diffusivity values in 14% of the investigated WM voxels including inferior fronto-occipital fasciculus, uncinate fasciculus and forceps minor compared to AD patients. AD patients had no areas of reduced diffusivity of fractional anisotropy compared to bvFTD. Extraction of regions of interest (ROI) Figure 4 illustrates which areas composite the ROIs per contrast. GM ROI AD<Controls was composed of significant voxels in temporal gyrus, posterior cingulate, cuneus, precuneus, parietal lobule and inferior frontal gyrus. GM ROI bvFTD<Controls was composed of significant voxels in frontal gyrus, orbito-frontal gyrus, insular, and temporal gyrus. GM ROI bvFTD<AD was composed of significant voxels in left inferior and medial frontal gyrus, right inferior frontal gyrus, and orbito-frontal gyrus. FA ROI AD<Controls consisted of significant voxels in fornix, corpus callosum, forceps minor, thalamus, posterior thalamic radiation, superior and inferior longitudinal fasciculus. FA ROI bvFTD<Controls consisted of significant voxels in fornix, corpus callosum, forceps minor, thalamus, anterior thalamic radiation, superior and inferior longitudinal fasciculus and inferior fronto-occipital fasciculus. FA ROI bvFTD<AD consisted of significant voxels in rostrum and the genu of the corpus callosum, forceps minor, anterior part of the internal and external capsule, anterior parts of the frontooccipital fasciculus and superior longitudinal fasciculus. MD ROI AD>controls consisted of significant voxels in fornix, corpus callosum, forceps minor and forceps major. MD ROI bvFTD>controls consisted of significant voxels in inferior fronto-occipital fasciculus, uncinate fasciculus and the forceps minor. MD ROI bvFTD>AD consisted of significant voxels in forceps minor, uncinate fasciculus, inferior fronto-occipital fasciculus and anterior thalamic radiation. L1 ROI AD>controls consisted of significant voxels in corpus callosum, the corticospinal tract and inferior longitudinal fasciculus. L1 ROI bvFTD>controls consisted of significant voxels in inferior fronto-occiptial fasciculus, inferior longitudinal fasciculus, corticospinal tract and corpus callosum. L1 ROI bvFTD>AD 142

consisted of significant voxels in inferior fronto-occipital fasciculus, uncinate fasciculus and forceps minor. L23 ROI AD>controls consisted of significant voxels in forceps major, inferior frontooccipital fasciculus, inferior and superior longitudinal fasciculus and the corpus callosum. L23 ROI bvFTD>controls consisted of significant voxels in inferior and superior longitudinal fasciculus, corticospinal tract, corpus callosum, fornix, inferior fronto-occiptial fasciculus and the anterior thalamic radiation. L23 ROI bvFTD>AD consisted of forceps minor, uncinate fasciculus, inferior fronto-occipital fasciculus and anterior thalamic radiation. Predictive value of GM volume, volumes of DGM structures, and white matter integrity Subsequently, we used discriminant analysis to identify the combination of MRmarkers providing optimal classification. Using stepwise forward method, the first discriminant analysis selected the following predictors: (1) GM ROI AD<Controls; (2) hippocampal volume; (3) volume of putamen (4) FA ROI AD<Controls; (5) FA ROI bvFTD<Controls; (6) center; (7) age; and (8) sex. The two resulting discriminant functions had a Wilk’s lambda of 0.082 (p≤0.001) and 0.388 (p≤0.001). Figure 5A shows the projection plot of the two canonical discriminant functions for discrimination of the three groups. Discriminant function 1 discriminated AD from bvFTD and controls. Discriminant function 2 discriminated bvFTD from AD and controls. The loadings of the individual predictors for each function are shown in Table 3A. GM ROI AD<Controls had the highest loading on discriminant function 1. Discriminant function 2 was primarily composed of the variables FA ROI bvFTD<Controls, hippocampal volume, FA ROI AD<Controls, and GM ROI AD<Controls. Crossvalidation successfully classified 91.4 % of all cases correctly, with correct classification of 100% of controls, 100% of AD patients, and 66.7% of bvFTD patients. The second discriminant analysis selected the following predictors: (1) GM ROI AD<Controls; (2) GM bvFTD<AD; (3) L1 ROI AD>Controls; (4) L1 ROI bvFTD>Controls; and (5) L1 ROI bvFTD>AD. The two resulting discriminant functions had a Wilk’s lambda of 0.134 (p≤0.001) and 0.437 (p≤0.001). Figure 5B shows the projection plot of the two canonical discriminant functions for discrimination of the three groups. Discriminant function 1 discriminated AD from bvFTD and controls. Discriminant function 2 discriminated bvFTD from AD and controls. The loadings of the individual predictors for each function are shown in Table 3B. GM ROI AD<Controls and L1 ROI AD<Controls had the highest loadings on discriminant function 1. Discriminant function 2 was primarily composed of GM ROI bvFTD<AD, L1 ROI bvFTD<AD, L1 ROI bvFTD>Controls, GM ROI AD<Controls, and L1 ROI AD>Controls. Cross-validation successfully classified 86% of all cases correctly, with correct classification of 97.3% of controls, 81.3% of AD patients, and 75% of bvFTD patients.

Discussion The main finding of this study is that there are GM and clear WM differences between AD and bvFTD which both independently contributed to the classification of both types of dementia. Despite a comparable disease stage, bvFTD patients had more atrophy in orbitofrontal, medial frontal and inferior frontal areas, caudate nucleus and 143

nucleus accumbens than AD patients. Furthermore, they had more severe loss of FA, higher MD, L1 and L23 values, especially in the frontal areas. Combination of modalities led to 86-91.4% correct classification of patients. GM contributed most to distinguishing AD patients from controls and bvFTD patients, while WM integrity measurements, especially L1, contributed to distinguish bvFTD from controls and AD. A large number of studies investigated the differences between controls and AD or bvFTD patients with regard to either GM or WM pathology. Their results are in line with the current study showing GM atrophy of medial temporal lobe structures and temporoparietal lobes in AD [25-27] and atrophy of orbitofrontal, anterior cingulate, lateral temporal cortices, and caudate nucleus in bvFTD [28-31]. DTI studies on AD reported a rather consistent pattern of FA reductions in widely distributed WM tracts exceeding MTL regions [32-34]. In patients with bvFTD significant FA reductions in the superior and inferior longitudinal fasciculus, as well as additional FA decreases in the uncinate fasciculus and the genu of the corpus callosum have been reported [35,36]. To determine whether GM atrophy or WM integrity have potential diagnostic use, a direct comparison between AD and bvFTD is more important than the comparison with a control group. With respect to GM atrophy, precuneus, lateral parietal and occipital cortices have been shown to be more atrophic in AD than in bvFTD, whereas atrophy of anterior cingulate, anterior insula, subcallosal gyrus, and caudate nucleus are more atrophic in bvFTD compared to AD [4-6]. In our study, we did not find any areas which are more atrophic in AD compared to bvFTD. This could be explained by the strict FWE-correct VBM approach in our study, as in one study the decrease GM areas found in AD compared to FTD did not survive multiple comparisons correction [4]. Another explanation that we did not find any GM reductions in AD could be that our patients are included in an early disease stage, with relatively higher MMSE scores compared to another study [5]. Nevertheless, patterns of GM atrophy often overlap, as there are numerous regions of GM loss which are found in both AD and bvFTD [4,79]. The few existing DTI studies demonstrated WM alterations in FTD compared to AD, including more widespread FA reductions in the frontal, anterior temporal, anterior corpus callosum, inferior fronto-occipital fasciculus and bilateral anterior cingulum [10,12,14,29,37]. One of these studies also investigated the MD, L1 and L23 differences between FTD and AD and found increased L1 and L23 values in FTD compared to AD [10]. Our study is in line with these previous studies, failing to observe reduced FA and increased MD, L1 and/or L23 in AD relative to bvFTD. The same is seen in the DGM structures, where bvFTD patients have more subcortical brain damage compared to AD patients but not the other way around [6,38,39]. This is suggestive that bvFTD is more a network disease, with involvement of the whole frontal-striatal circuits, including the connecting white matter tracts and DGM structures, while AD is seemingly more a cortical disease. We attempted to combine GM and WM measures to increase the discrimination of patient groups and showed that next to GM atrophy, WM integrity measures helped in distinguishing AD from bvFTD. A few earlier studies have combined WM and GM information with the objective to better discriminate between AD and bvFTD. They 144

found that FTD patients exhibited more WM damage than AD patients in an early stage of the disease suggesting that measuring of WM damage add up in the discrimination between these two dementias [13,14]. Another study only linked the two imaging modalities and support the idea of a network disease in FTD but did not examine diagnostic value of GM and WM [37]. Only two studies actually used a multimodal combination of WM and GM. In one study they achieved a classification with 87% sensitivity and 83% specificity between AD and bvFTD [12]. In another study they developed a new metric which gives a measure of the amount of WM connectivity disruption for a GM region and showed classification rates were 8-13% higher when adding WM measurements to GM measurements [40]. The novelty of the study lies in the combination of three measures to separate AD from bvFTD. We combined VBM based measures of cortical atrophy, FIRST based measures of atrophy of DGM structures and DTI based measures of WM integrity to yield an optimal classifier. Both discriminant analyses revealed that cortical GM matter contributed to the separation of AD from the other two groups and WM integrity measurements contributed to the discrimination of bvFTD from the other groups. Especially axial diffusivity increased the discriminatory power for bvFTD. This could be explained by the notion that, despite some involvement of DGM and WM, AD is assumed to be a cortical dementia with specific GM regions being affected whereas bvFTD predominantly affect areas (frontal-insula-anterior cingulate) which are part of structurally and functionally connected neural networks. These networks are connected by specific WM tracts located within damaged GM areas as the frontal lobes and are preferentially affected, contributing to network failure in bvFTD. The finding of more severe damage of DGM structures add up to the hypotheses of bvFTD being a network disorder as DGM structures can be seen as relay stations in the fronto-striatal brain networks. These findings are further supported by the fact that bvFTD had the same disease stage (comparable MMSE, CDR, duration of symptoms) as AD patients but have more WM and DGM structure damage. A possible limitation of this study is that we did not have post-mortem data available, so the possibility of misdiagnosis cannot be excluded. Nevertheless, we used an extensive standardized work-up and all AD patients fulfilled clinical criteria of probable AD, 19 patients fulfilled the criteria for probable bvFTD and 5 patients for possible bvFTD. All diagnoses were re-evaluated in a panel including clinicians from both centers to minimize sample effects. Because this is a multicenter study, the differences in data acquisition parameters between the two centers might introduce some noise in the DTI analysis. However, we adjusted for center in all models and moreover, a recent study showed that when considering scanner effects in the statistical model, no relevant differences between scanners were found [41]. Among the strengths of this study is the sample size and it’s multi-center nature. Most of the studies comparing AD with bvFTD use smaller sample sizes. We had enough power to detect differences using FWE and FWE-TFCE correction to adjust for multiple comparisons. Another strength is the unique combination of three imaging parameters in this study to achieve optimal discrimination between AD and bvFTD. 145

Conclusion Accurate diagnosis of patients in life is increasingly important, both on clinical and scientific grounds. It is a guide to prognosis and prerequisite for optimal clinical care and management. AD and bvFTD are difficult to discriminate due to overlapping clinical and imaging features. Therefore, there is an urgent need to improve diagnostic accuracy in a quantitative manner. This study has shown that DTI measures not only support the hypotheses of a network disorder in bvFTD but also add complementary information to measures of cortical and subcortical atrophy, thereby allowing a more precise diagnosis between AD and bvFTD.

Acknowledgements / Disclosures The VUmc Alzheimer Center is supported by Alzheimer Nederland and Stichting VUmc Fonds. The study was supported by the Netherlands Organisation for Scientific Research (NWO). Research of the VUmc Alzheimer Center is part of the neurodegeneration research program of the Neuroscience Campus Amsterdam. The clinical database structure was developed with funding from Stichting Dioraphte. This project is funded by the Netherlands Initiative Brain and Cognition (NIHC), a part of the Netherlands Organization for Scientific Research (NWO) under grant numbers 056-13-014 and 056-13-010. The gradient non-linearity correction was kindly provided by GE medical systems, Milwaukee. Prof. Dr. Serge Rombouts is supported by The Netherlands Organization for Scientific Research (NWO), Vici project nr. 016.130.677. Dr. Wiesje van der Flier is recipient of the Alzheimer Nederland grant (Influence of age on the endophenotype of AD on MRI, project number 2010-002).

146

Tables and Figures Table 1. Demographics Controls AD bvFTD N 37 32 24 a Age, years 60.4 ± 6.2 66.7 ± 7.7 63.2 ± 7.5 Sex, females 16 (43%) 12 (38%) 6 (25%) Center, Vumc 22 (60%) 22 (69%) 18 (75%) Level of education 5.6 ± 1.0 5.0 ± 1.4 4.8 ± 1.6 Duration of symptoms (months) 40.2 ± 4.6 50.0 ± 8.9 a a,b MMSE 28.9 ± 1.4 23.2 ± 3.1 25.1 ± 3.1 a a CDR 0±0 0.8 ± 0.3 0.8 ± 0.3 a a GDS 1.1 ± 1.3 3.0 ± 3.1 3.8 ± 2.9 3 a a NBV (cm ) 1493.7 ± 64.1 1395.2 ± 76.2 1394.81 ± 87.6 VSF 1.3 ± 0.1 1.3 ± 0.1 1.3 ± 0.1 Values presented as mean ± standard deviation or n (%). Level of education is determined according to the Verhage-system. Differences between groups for demographics were assessed 2 using ANOVA. Kruskal-Wallis tests and χ tests. where appropriate. Key: MMSE: Mini-Mental State Examination; CDR: Dementia Rating Scale; GDS: Geriatric Depression Scale; NBV: normalized brain volume; VSF: volumetric scaling factor a b different from controls (p<0.05), different from AD (p<0.05).

147

͛ ͚Ǥ ȋ Ȍ ǡ Ǥ η η η ͠Ǥ͛ ΰ ͙Ǥ͜ ζ͘Ǥ͘​͙͘ ͙Ǥ͘͟͟ ζ͘Ǥ͘​͙͘ ͙Ǥ͜͟͠ ζ͘Ǥ͘​͙͘ ͘Ǥ͙͘͜ ͙͘Ǥ͛ ΰ ͙Ǥ͘ ͠Ǥ͜ ΰ ͙Ǥ͝ ͛Ǥ͞ ΰ ͘Ǥ͞ ͛Ǥ͜ ΰ ͙Ǥ͝ ͛Ǥ͛ ΰ ͘Ǥ͞ ͘Ǥ͚͘͠ ͙͡Ǥ͠ ΰ ͙Ǥ͟ ͙͠Ǥ͙ ΰ ͚Ǥ͘ ͙͠Ǥ͙ ΰ ͙Ǥ͟ ζ͘Ǥ͘​͙͘ ͙Ǥ͚​͚͟ ͘Ǥ͘​͘͞ ͙Ǥ͞͠͡ ζ͘Ǥ͘​͙͘ ͘Ǥ͙͜͞ ǡ ͡Ǥ͘ ΰ ͘Ǥ͠ ͠Ǥ͟ ΰ ͘Ǥ͡ ͠Ǥ͘ ΰ ͙Ǥ͙ ζ͘Ǥ͘​͙͘ ͘Ǥ͙͚͠ ͙Ǥ͘​͘​͘ ͘Ǥ͟͡͠ ζ͘Ǥ͘​͙͘ ͘Ǥ͘͠͝ ͘Ǥ͘​͙͘ ͘Ǥ͚͙͛ ͘Ǥ͙͛͞ ͙Ǥ͙​͙͠ ͘Ǥ͘​͙͘ ͘Ǥ͜͠͝ ͙͚Ǥ͛ ΰ ͙Ǥ͙ ͙​͙Ǥ͚ ΰ ͙Ǥ͜ ͙​͙Ǥ͙ ΰ ͙Ǥ͜

͜Ǥ͠ ΰ ͘Ǥ͟ ͜Ǥ͞ ΰ ͘Ǥ͠ ͜Ǥ͛ ΰ ͘Ǥ͠ ͘Ǥ͙͘͟ ǡ ͙Ǥ͙ ΰ ͘Ǥ͚ ͘Ǥ͡ ΰ ͘Ǥ͛ ͘Ǥ͠ ΰ ͘Ǥ͚ ζ͘Ǥ͘​͙͘ ͘Ǥ͙͛͜ ͘Ǥ͙͛͘ ͘Ǥ͛​͙͛ ζ͘Ǥ͘​͙͘ ͘Ǥ͙͟͠ ͛ ΰ Ǥ ǡ Ǥ ǡ ȋ ζ͘Ǥ͘͝Ȍ

͙͜8 ͙Ǥ͘​͘​͘ ͘Ǥ͛͟͡ ͘Ǥ͘​͚͘ ͘Ǥ͛͜͞ ͘Ǥ͘​͘͝

Table 3. Structure Matrix showing the discriminant loadings for each predictor. The structure matrix correlation coefficient represents the relative contribution of each predictor to group separation. (A) Discriminant analysis with GM ROIs, DGM structures and FA ROIs. (B) Discriminant analysis with GM ROIs, DGM structures and L1 and L23 ROIs. (A)

GM ROI AD<Controls Hippocampus Putamen FA ROI AD<Controls FA ROI bvFTD<Controls Center Age Sex

Function 1 2 0.469 0.449 0.222 0.475 0.111 0.282 0.232 0.451 0.131 0.476 -0.021 -0.103 0.187 0.111 0.003 -0.121

(B) Function 1 2 GM ROI AD<Controls 0.642 0.400 GM ROI bvFTD<AD 0.039 0.706 L1 ROI bvFTD>AD 0.026 0.478 L1 ROI bvFTD>Controls 0.183 0.433 L1 ROI AD>Controls 0.303 0.329

149

Figure 1. VBM voxel-wise statistical analysis of GM reductions between groups. Figures are displayed with a threshold of p<0.05, FWE corrected. Brighter colors indicate higher t values.

150

3

Figure 2. Boxplot of raw volumes (cm ) of MTL and DGM structures.

**

*

p≤0.001, p<0.05

151

Figure 3. TBSS voxelwise statistics displaying areas of white matter skeleton (green) with lower FA (red-yellow) and higher MD, L1, L23 (blue-light blue) values, overlaid on the MNI-standard brain. Significance level of p<0.05 with correction for multiple comparisons was used. Skeletonized results are thickened to enhance figure clarity. These thickened results are based on the original p-maps. Percentages remained the same.

152

Figure 4. Composition of ROIs per contrast. (A) GM ROI AD<Controls: All significant voxels (p<0.05, FWE corrected) from the VBM group comparison where AD patients had less GM compared to controls are indicated in yellow. FA ROI AD<Controls: All significant areas (p<0.05, FWE TFCE corrected) from the TBSS group comparisons where AD patients had lower FA values compared to controls are indicated in red. (B) GM ROI bvFTD<Controls: All significant voxels (p<0.05, FWE corrected) from the VBM group comparison where bvFTD patients had less GM compared to controls are indicated in yellow. FA ROI bvFTD<Controls: All significant areas (p<0.05, FWE TFCE corrected) from the TBSS group comparisons where bvFTD patients had lower FA values compared to controls are indicated in red. (C) GM ROI bvFTD<AD: All significant voxels (p<0.05, FWE corrected) from the VBM group comparison where bvFTD patients had less GM compared to AD patients are indicated in yellow. FA ROI bvFTD<AD: All significant areas (p<0.05, FWE TFCE corrected) from the TBSS group comparisons where bvFTD patients had lower FA values compared to AD patients are indicated in red. (D) MD, L1, L23 ROI AD>Controls: All significant areas (p<0.05, FWE TFCE corrected) from the TBSS group comparisons where AD patients had higher MD (pink), higher L1 (blue) and higher L23 (green) values compared to controls. (E) MD, L1, L23 ROI bvFTD>Controls: All significant areas (p<0.05, FWE TFCE corrected) from the TBSS group comparisons where bvFTD patients had higher MD (pink), higher L1 (blue) and higher L23 (green) values compared to controls. (F) MD, L1, L23 ROI bvFTD>AD: All significant areas (p<0.05, FWE TFCE corrected) from the TBSS group comparisons where bvFTD patients had higher MD (pink), higher L1 (blue) and higher L23 (green) values compared to AD patients.

153

Figure 5. Projection plot of canonical discriminant functions for discrimination of healthy controls, AD and bvFTD patients. (A) Discriminant function consisted of GM ROI AD<Controls; hippocampal volume; volume of putamen; FA ROI AD<Controls; FA ROI bvFTD<Controls; center; age; and sex. (B) Discriminant function consisted of GM ROI AD<Controls; GM bvFTD<AD; L1 ROI AD>Controls; L1 ROI bvFTD>Controls; and L1 ROI bvFTD>AD. Blue squares indicate individual data of healthy controls, green dots indicate data of individual AD patients, red triangles indicate individual data of bvFTD patients. The black squares represent the group centroids. (A)

(B)

154

Reference List [1] [2]

[3]

[4]

[5]

[6]

[7]

[8]

[9]

[10]

[11]

[12]

[13]

[14]

Ratnavalli E, Brayne C, Dawson K, Hodges JR (2002) The prevalence of frontotemporal dementia. Neurology 58, 1615-1621. Rascovsky K, Hodges JR, Knopman D, Mendez MF, Kramer JH, Neuhaus J et al. (2011) Sensitivity of revised diagnostic criteria for the behavioural variant of frontotemporal dementia. Brain 134, 2456-2477. McKhann GM, Knopman DS, Chertkow H, Hyman BT, Jack CR, Jr., Kawas CH et al. (2011) The diagnosis of dementia due to Alzheimer's disease: recommendations from the National Institute on Aging-Alzheimer's Association workgroups on diagnostic guidelines for Alzheimer's disease. Alzheimers Dement 7, 263-269. Rabinovici GD, Seeley WW, Kim EJ, Gorno-Tempini ML, Rascovsky K, Pagliaro TA, Allison SC, Halabi C, Kramer JH, Johnson JK, Weiner MW, Forman MS, Trojanowski JQ, Dearmond SJ, Miller BL, Rosen HJ (2007) Distinct MRI atrophy patterns in autopsy-proven Alzheimer's disease and frontotemporal lobar degeneration. Am J Alzheimers Dis Other Demen 22, 474-488. Du AT, Schuff N, Kramer JH, Rosen HJ, Gorno-Tempini ML, Rankin K, Miller BL, Weiner MW (2007) Different regional patterns of cortical thinning in Alzheimer's disease and frontotemporal dementia. Brain 130, 1159-1166. Looi JC, Lindberg O, Zandbelt BB, Ostberg P, Andersen C, Botes L, Svensson L, Wahlund LO (2008) Caudate nucleus volumes in frontotemporal lobar degeneration: differential atrophy in subtypes. AJNR Am J Neuroradiol 29, 1537-1543. Munoz-Ruiz MA, Hartikainen P, Koikkalainen J, Wolz R, Julkunen V, Niskanen E, Herukka SK, Kivipelto M, Vanninen R, Rueckert D, Liu Y, Lotjonen J, Soininen H (2012) Structural MRI in frontotemporal dementia: comparisons between hippocampal volumetry, tensor-based morphometry and voxel-based morphometry. PLoS One 7, e52531van de Pol LA, Hensel A, van der Flier WM, Visser PJ, Pijnenburg YA, Barkhof F, Gertz HJ, Scheltens P (2006) Hippocampal atrophy on MRI in frontotemporal lobar degeneration and Alzheimer's disease. J Neurol Neurosurg Psychiatry 77, 439442. Barnes J, Whitwell JL, Frost C, Josephs KA, Rossor M, Fox NC (2006) Measurements of the amygdala and hippocampus in pathologically confirmed Alzheimer disease and frontotemporal lobar degeneration. Arch Neurol 63, 1434-1439. Zhang Y, Schuff N, Du AT, Rosen HJ, Kramer JH, Gorno-Tempini ML, Miller BL, Weiner MW (2009) White matter damage in frontotemporal dementia and Alzheimer's disease measured by diffusion MRI. Brain 132, 2579-2592. Chen TF, Lin CC, Chen YF, Liu HM, Hua MS, Huang YC, Chiu MJ (2009) Diffusion tensor changes in patients with amnesic mild cognitive impairment and various dementias. Psychiatry Res 173, 15-21. McMillan CT, Brun C, Siddiqui S, Churgin M, Libon D, Yushkevich P, Zhang H, Boller A, Gee J, Grossman M (2012) White matter imaging contributes to the multimodal diagnosis of frontotemporal lobar degeneration. Neurology 78, 1761-1768. Mahoney CJ, Ridgway GR, Malone IB, Downey LE, Beck J, Kinnunen KM, Schmitz N, Golden HL, Rohrer JD, Schott JM, Rossor MN, Ourselin S, Mead S, Fox NC, Warren JD (2014) Profiles of white matter tract pathology in frontotemporal dementia. Hum Brain Mapp Zhang Y, Schuff N, Ching C, Tosun D, Zhan W, Nezamzadeh M, Rosen HJ, Kramer JH, Gorno-Tempini ML, Miller BL, Weiner MW (2011) Joint assessment of 155

[15]

[16] [17] [18] [19] [20]

[21]

[22]

[23]

[24]

[25]

[26]

[27]

[28] [29]

[30]

156

structural, perfusion, and diffusion MRI in Alzheimer's disease and frontotemporal dementia. Int J Alzheimers Dis 2011, 546871McKhann G, Drachman D, Folstein M, Katzman R, Price D, Stadlan EM (1984) Clinical diagnosis of Alzheimer's disease: report of the NINCDS-ADRDA Work Group under the auspices of Department of Health and Human Services Task Force on Alzheimer's Disease. Neurology 34, 939-944. Verhage F (1964) Intelligentie en leeftijd: onderzoek bij Nederlanders van twaalf tot zevenenzeventig jaar., Van Gorcum, Assen. Jenkinson M , Smith S (2001) A global optimisation method for robust affine registration of brain images. Med Image Anal 5, 143-156. Patenaude B, Smith SM, Kennedy DN, Jenkinson M (2011) A Bayesian model of shape and appearance for subcortical brain segmentation. Neuroimage 56, 907-922. Jenkinson M, Beckmann CF, Behrens TE, Woolrich MW, Smith SM (2012) FSL. Neuroimage 62, 782-790. Smith SM, Zhang Y, Jenkinson M, Chen J, Matthews PM, Federico A, De SN (2002) Accurate, robust, and automated longitudinal and cross-sectional brain change analysis. Neuroimage 17, 479-489. Popescu V, Battaglini M, Hoogstrate WS, Verfaillie SC, Sluimer IC, van Schijndel RA, van Dijk BW, Cover KS, Knol DL, Jenkinson M, Barkhof F, De SN, Vrenken H (2012) Optimizing parameter choice for FSL-Brain Extraction Tool (BET) on 3D T1 images in multiple sclerosis. Neuroimage 61, 1484-1494. Zhang Y, Brady M, Smith S (2001) Segmentation of brain MR images through a hidden Markov random field model and the expectation-maximization algorithm. IEEE Trans Med Imaging 20, 45-57. Smith SM, Jenkinson M, Woolrich MW, Beckmann CF, Behrens TE, Johansen-Berg H, Bannister PR, De LM, Drobnjak I, Flitney DE, Niazy RK, Saunders J, Vickers J, Zhang Y, De SN, Brady JM, Matthews PM (2004) Advances in functional and structural MR image analysis and implementation as FSL. Neuroimage 23 Suppl 1, S208-S219. Smith SM, Jenkinson M, Johansen-Berg H, Rueckert D, Nichols TE, Mackay CE, Watkins KE, Ciccarelli O, Cader MZ, Matthews PM, Behrens TE (2006) Tract-based spatial statistics: voxelwise analysis of multi-subject diffusion data. Neuroimage 31, 1487-1505. Whitwell JL, Jack CR, Jr., Przybelski SA, Parisi JE, Senjem ML, Boeve BF, Knopman DS, Petersen RC, Dickson DW, Josephs KA (2011) Temporoparietal atrophy: a marker of AD pathology independent of clinical diagnosis. Neurobiol Aging 32, 1531-1541. Mรถller C, Vrenken H, Jiskoot L, Versteeg A, Barkhof F, Scheltens P, van der Flier WM (2013) Different patterns of gray matter atrophy in early- and late-onset Alzheimer's disease. Neurobiol Aging 34, 2014-2022. Frisoni GB, Testa C, Zorzan A, Sabattoli F, Beltramello A, Soininen H, Laakso MP (2002) Detection of grey matter loss in mild Alzheimer's disease with voxel based morphometry. J Neurol Neurosurg Psychiatry 73, 657-664. Rabinovici GD , Miller BL (2010) Frontotemporal lobar degeneration: epidemiology, pathophysiology, diagnosis and management. CNS Drugs 24, 375-398. Hornberger M, Geng J, Hodges JR (2011) Convergent grey and white matter evidence of orbitofrontal cortex changes related to disinhibition in behavioural variant frontotemporal dementia. Brain 134, 2502-2512. Couto B, Manes F, Montanes P, Matallana D, Reyes P, Velasquez M, Yoris A, Baez S, Ibanez A (2013) Structural neuroimaging of social cognition in progressive non-

[31]

[32]

[33]

[34]

[35]

[36]

[37]

[38]

[39]

[40]

[41]

fluent aphasia and behavioral variant of frontotemporal dementia. Front Hum Neurosci 7, 467Looi JC, Walterfang M, Velakoulis D, Macfarlane MD, Svensson LA, Wahlund LO (2012) Frontotemporal dementia as a frontostriatal disorder: neostriatal morphology as a biomarker and structural basis for an endophenotype. Aust N Z J Psychiatry 46, 422-434. Scola E, Bozzali M, Agosta F, Magnani G, Franceschi M, Sormani MP, Cercignani M, Pagani E, Falautano M, Filippi M, Falini A (2010) A diffusion tensor MRI study of patients with MCI and AD with a 2-year clinical follow-up. J Neurol Neurosurg Psychiatry 81, 798-805. Agosta F, Pievani M, Sala S, Geroldi C, Galluzzi S, Frisoni GB, Filippi M (2011) White matter damage in Alzheimer disease and its relationship to gray matter atrophy. Radiology 258, 853-863. Salat DH, Tuch DS, van der Kouwe AJ, Greve DN, Pappu V, Lee SY, Hevelone ND, Zaleta AK, Growdon JH, Corkin S, Fischl B, Rosas HD (2010) White matter pathology isolates the hippocampal formation in Alzheimer's disease. Neurobiol Aging 31, 244-256. Borroni B, Brambati SM, Agosti C, Gipponi S, Bellelli G, Gasparotti R, Garibotto V, Di LM, Scifo P, Perani D, Padovani A (2007) Evidence of white matter changes on diffusion tensor imaging in frontotemporal dementia. Arch Neurol 64, 246-251. Matsuo K, Mizuno T, Yamada K, Akazawa K, Kasai T, Kondo M, Mori S, Nishimura T, Nakagawa M (2008) Cerebral white matter damage in frontotemporal dementia assessed by diffusion tensor tractography. Neuroradiology 50, 605611. Avants BB, Cook PA, Ungar L, Gee JC, Grossman M (2010) Dementia induces correlated reductions in white matter integrity and cortical thickness: a multivariate neuroimaging study with sparse canonical correlation analysis. Neuroimage 50, 1004-1016. Looi JC, Svensson L, Lindberg O, Zandbelt BB, Ostberg P, Orndahl E, Wahlund LO (2009) Putaminal volume in frontotemporal lobar degeneration and Alzheimer disease: differential volumes in dementia subtypes and controls. AJNR Am J Neuroradiol 30, 1552-1560. Halabi C, Halabi A, Dean DL, Wang PN, Boxer AL, Trojanowski JQ, Dearmond SJ, Miller BL, Kramer JH, Seeley WW (2013) Patterns of striatal degeneration in frontotemporal dementia. Alzheimer Dis Assoc Disord 27, 74-83. Kuceyeski A, Zhang Y, Raj A (2012) Linking white matter integrity loss to associated cortical regions using structural connectivity information in Alzheimer's disease and fronto-temporal dementia: the Loss in Connectivity (LoCo) score. Neuroimage 61, 1311-1323. Teipel SJ, Wegrzyn M, Meindl T, Frisoni G, Bokde AL, Fellgiebel A, Filippi M, Hampel H, Kloppel S, Hauenstein K, Ewers M (2012) Anatomical MRI and DTI in the diagnosis of Alzheimer's disease: a European multicenter study. J Alzheimers Dis 31 Suppl 3, S33-S47.

157

158

Chapter 4 – Discussion

159

The main findings of this thesis were that patterns of gray matter (GM) atrophy differ between different manifestations of AD as well as between AD and bvFTD. This is applicable for some cortical areas as well as for certain subcortical structures. In the discrimination of AD and bvFTD, the measurement of white matter (WM) integrity increases the diagnostic accuracy and gives more information about the underlying pathological processes. GM, WM and deep gray matter (DGM) structures have the potential of serving as a diagnostic tool as they allow reliable discrimination of groups. However, at this point, these results are applicable only for groups of patients and not for individual subjects. Image analysis techniques have to become more reliable and suitable for single-subject diagnosis. In the present chapter, the main findings of this thesis are summarized, followed by a discussion of the diagnostic value of MRI derived measures and of the disease pathology underlying AD and bvFTD. The chapter closes with suggestions for further research.

Summary Chapter 2. Patterns of gray matter loss in different manifestations of AD In chapter 2.1 we assessed patterns of GM atrophy according to age-at-onset in a large sample of AD patients and controls with VBM. By comparing AD patients with controls and early-onset with late-onset AD patients, we found that age and diagnosis independently affected hippocampus; moreover, the interaction between age and diagnosis showed that precuneus atrophy was most prominent in early-onset AD. This suggests that patterns of atrophy may vary in the spectrum of AD and that it is important to compare patients with a reference of their own age category. Moreover, in younger patients, the posterior part of the brain – especially the precuneus – may provide the most valuable information when evaluating their MRI scans. In chapter 2.2 we determined if the use of the 4-point visual rating scale for posterior cortical atrophy (PCA) in clinical practice is justified. For this we used quantitative GM volumetry and VBM. The visual PCA rating scale turned out to reliably reflect GM atrophy in posterior cortical regions. There was a clear separation between brains rated as having PCA and those rated as having no atrophy. Moreover, the different severity scores in the rating scale corresponded to different quantitative degrees of atrophy. Finally, especially the volume of the inferior parietal gyrus affected the visual PCA scoring. These results suggest that the visual PCA rating scale is a valuable tool for the daily radiological assessment of dementia. In chapter 2.3 we investigated whether DGM atrophy predicts progression from mild cognitive impairment (MCI) to AD, and compared subcortical volumes between AD, MCI and controls. Furthermore we analyzed associations with severity of cognitive impairment. We found that in addition to baseline hippocampal volume, also baseline nucleus accumbens volume predicted progression from MCI to AD during two years of clinical follow-up. Baseline volumes of other subcortical structures were not predictive of progression to AD, despite the observation of decreasing volumes from controls to MCI to AD, and significant associations with cognition for all structures except globus pallidus.

160

Chapter 3. Patterns of gray matter loss in AD and bvFTD Because the involvement of frontostriatal circuits in bvFTD suggests that DGM structures may be affected in this disease, we investigated whether volumes of DGM structures differed between patients with bvFTD, AD and controls with subjective complaints (SC) and explored the relationships between DGM structures, cognition and neuropsychiatric signs and symptoms in chapter 3.1. The results suggest that nucleus accumbens, caudate nucleus, and globus pallidus are more severely affected in bvFTD than in AD and SC. The associations between cognition and DGM structures varied between the diagnostic groups. The observed difference in volume of these DGM structures supports the idea that in addition to frontal cortical atrophy, DGM structures, as parts of the frontal circuits, are damaged in bvFTD rather than in AD. In chapter 3.2 we explored the diagnostic accuracy of an automated classifier for individual patients based on a generally available 3D T1-weighted from which the GM content of each voxel was quantified using a widely used, standardized procedures. To increase generalizability we used MRI scans from different scanners and centers and two independent data sets in a cross-sectional design. The results showed that with this automated classifier it was possible to discriminate between scans of patients with different forms of dementia and controls with high accuracy, based solely on their GM patterns. We also demonstrated that automated classifiers can be used in single-subject diagnosis, as the diagnostic accuracy of our classifier in an independent dataset of patients and controls was good to excellent. In chapter 3.3 we examined the loss of cortical thickness and cognitive functioning over time in AD, bvFTD and controls in a longitudinal study. We found that both, AD and bvFTD showed more cortical thinning per year compared to controls, with AD showing decline in memory and language. Compared to controls, AD patients lost cortical thickness over the whole brain with a clear posterior gradient, whereas bvFTD patients only showed cortical thinning in the frontal cortex and in the anterior parts of the temporal lobes. Compared to each other, AD patients showed cortical thinning in the insula, temporal and parietal regions, bvFTD patients only progressed faster in a small frontal region. These results suggest that decrease of thickness is more generalized in AD, whereas bvFTD have a more selective loss of thickness. In chapter 3.4 we investigated the ability of cortical and subcortical GM atrophy in combination with WM integrity to distinguish bvFTD from AD and from controls using VBM, subcortical GM structure segmentation, and WM integrity as assessed using TBSS. Furthermore, we were interested in the question which combination of imaging markers differentiated the three groups with the highest accuracy. We showed that there were clear GM and WM differences between AD and bvFTD which independently contributed to the classification of both types of dementia. Despite a comparable disease stage, bvFTD patients had more GM and DGM atrophy, and had more severe loss of fractional anisotropy, and higher mean, axial and radial diffusivity values, especially in the frontal areas. Combining modalities led to 86-91% correct classification of patients. GM contributed most to distinguishing AD patients from controls and bvFTD patients, while WM integrity measurements, especially axial diffusivity, contributed to distinguishing bvFTD from controls and AD. These results suggest that WM integrity measures add complementary information to measures of GM atrophy, thereby improving the classification between AD and bvFTD. 161

General discussion Using the consensus clinical diagnostic criteria for AD and bvFTD, diagnostic certainty is variable, but suboptimal. The criteria for probable AD have a diagnostic sensitivity and specificity compared to the pathological diagnosis, ranging between 50% and 90%, mainly depending on the clinical expertise, the age and other characteristics of the patients studied [1,2]. Diagnostic uncertainty remains in other clinical criteria as well, as the ultimate ‘gold standard’ for diagnosis does not exist [3]. When imaging is included in the criteria a higher degree of specificity (>90%) can be reached. In general, the use of imaging has shifted from an exclusionary to an inclusionary approach over the past decades. Using MRI to identify specific abnormalities that may aid the clinician to diagnose underlying disease became increasingly more relevant and adds positive predictive value to the diagnosis of dementia. Therefore, current international guidelines recommend the use of brain imaging techniques in diagnosing dementia syndromes [4-6].

Diagnostic value of MRI derived measures of brain tissues Gray matter In most memory clinics, patients are usually scanned once during dementia screening, with a standardized protocol generally including a structural T1-weighted 3dimensional (3DT1) MRI sequence. The main features of this sequence include both high spatial resolution and high contrast between GM and WM. Many dementias are characterized by GM atrophy as it reflects a loss of neurons irrespective of the underlying protein defect. GM atrophy may be focal or spread over the whole brain, and the patterns of GM atrophy may be diagnostic in itself as they are linked to specific diagnoses. A 3DT1 image is useful for the detection of these disease-specific patterns of GM atrophy and therefore supports a clinical diagnosis. For the interpretation of structural images, standardized assessment by the use of visual rating scales can improve the diagnostic accuracy [7]. In chapter 2.2 we support this notion by showing that a recently developed rating scale for posterior cortical atrophy is a reliable tool for the daily radiological assessment of dementia. AD is typically associated with medial temporal lobe atrophy (MTA) [8-11]. The presence of MTA on MR imaging improves the discrimination of AD from healthy controls and predicts progression to dementia in patients with mild cognitive impairment (MCI) [11,12]. Visual MTA rating based on established rating scales, has proven to be useful for a good and reproducible assessment in clinical practice and correlates well with volumetric assessments [13,14]. However, MTA is also present in other dementias and may be seen in normal aging [15-17]. While MTA is the hallmark finding in senile-onset AD, especially in APOE4 positive patients with an amnestic presentation, it can be relatively mild in subjects with presenile-onset, or without APOE4 and non-amnestic presentations [18,19]. Occasionally, AD patients present with a striking posterior atrophy pattern. In such patients, the pattern of atrophy will be dominated by posterior/parietal atrophy. In chapter 2.1 we affirm these findings by means of voxel-based comparisons of GM. We demonstrated that, when compared to their age-matched controls, AD patients with an early disease onset also showed more widespread atrophy throughout the brain (medial temporal lobe, precuneus, 162

cingulate gyrus, frontal lobe), whereas late-onset AD patients showed a more specific pattern of GM atrophy, predominantly restricted to the medial temporal lobe and cerebellum. Direct comparisons revealed more pronounced GM atrophy in the precuneus of early-onset AD patients despite their younger age. Posterior cortical atrophy appears to be characteristic of AD in patients with typical and atypical clinical presentations and may assist in the clinical distinction of AD from bvFTD. Moreover, combined with relative sparing of the medial temporal lobe, posterior cortical atrophy has found to be characteristic for patients with atypical clinical presentations [20-24]. As posterior cortical atrophy should be assessed in a standardized manner, a visual rating scale has been developed [25]. To establish the validity of this visual posterior cortical atrophy rating scale and thereby determine whether its use in the clinical practice is justified, it should be compared against quantitative brain volumetry. In chapter 2.2 we addressed this issue to stimulate the clinical applicability of the visual rating scale. We demonstrated that the visual rating scale for PCA reliably reflects GM atrophy in posterior regions and that its simplicity has great advantage for clinical practice, making it a useful tool in the daily radiological assessment of dementia. In the differentiation between AD and bvFTD, GM atrophy only plays a restricted role, as patterns of atrophy are often not exclusive for one diagnosis. The typical AD atrophy of medial temporal lobe, including hippocampus and amygdala [26-28] is also common in bvFTD [16,29-40]. The other way around, bvFTD is typically characterized by atrophy of frontal and anterior temporal lobes [29,41-43], but atrophy in these regions does not exclude a diagnosis of AD [44-46]. This was confirmed by our results in chapter 3.4 where AD patients did not show any regions of significantly reduced GM compared to bvFTD patients. The other way around bvFTD patients show only small areas of reduced GM in left inferior and medial frontal gyrus, in right inferior frontal gyrus, and in orbito-frontal gyrus. On the other hand, studies have also shown areas as the precuneus, lateral parietal and occipital cortices being exclusive for AD, and atrophy of anterior cingulate, anterior insula, subcallosal gyrus and caudate nucleus being more severe in bvFTD compared to AD [20,29,47,48]. This is confirmed by our findings that the insular cortex was more severely atrophied in bvFTD, in the beginning of the disease, and still discriminated bvFTD from AD at follow-up, even though AD patients showed a steeper rate of atrophy compared to bvFTD patients (chapter 3.3). But even if there are small regions found with quantitative image analysis methods, these regions are hard to detect by eyeballing (insula, caudate nucleus) and still refer to results of groups analyses. To be of diagnostic value, patterns of GM atrophy should have the potential to be able to predict a diagnosis for a single subject, taking into account information from the entire brain. As we showed in chapter 3.2, it was possible to discriminate between AD and bvFTD solely based on a T1-weighted GM image with 78.7% accuracy. We used a support vector machine to apply learned GM patterns of groups of AD and bvFTD patients to new individual patients and achieved 82.1% accuracy in deciding with patient has AD or bvFTD. To summarize, GM atrophy has proven to be useful for differentiation of dementia from controls. For the differentiation between forms of dementia however, patterns 163

of GM atrophy often overlap. Furthermore, in the beginning of the disease, when GM atrophy is hardly visible by eyeballing even the discrimination from healthy controls can be challenging. Another point of concern are the atypical cases, as early-onset AD patients or patients with genetic mutations, which display unusual patterns of GM atrophy, will not be recognized based on solely GM atrophy. Quantitative image analysis methods add a lot to the diagnostic certainty, as it allows a closer look at vanishing GM. However, these methods only allow comparisons on a group level, while for application in diagnostic routine, results need to applicable at the single subject level. Automatic classifiers for single-subject diagnosis are available but further research is needed to be able to use them reliably in the daily practice. White matter Dementia has been associated with both macro- as well as microscopic WM pathology [49-51]. Neuroimaging, especially MRI, has played a major role in the identification of macroscopic WM pathology thereby providing a significant contribution to both diagnosis and therapeutics [52,53]. Conventional MRI in neurodegenerative disorders lost in credibility as studies quantifying WM changes resulted in differing outcomes, WM lesions seen on MRI did not correlate well with cognitive disabilities, and differentiating between types of dementia proved problematic [54,55]. A likely explanation for this could be that pathological WM changes occurring at the microscopic level - particularly in the early disease stages - cannot be detected by these relatively insensitive conventional MRI techniques, rendering WM to appear normal [56]. However, the development of novel MRI techniques, especially diffusion tensor imaging (DTI), now provides us with the ability to quantify even the most subtle microstructural alterations in WM in ways that were impossible before [56-58]. By applying a strong gradient magnetic field in a single direction, the so-called diffusionsensitizing gradient, the signal becomes sensitized for diffusion in that direction. Fiber tracts parallel to this gradient field will show maximal signal loss, whereas the effect is minimal if the gradient field is perpendicular to the fiber tracts. By applying gradients in three or more different directions, one can display the anisotropy of tissue, especially in the WM. By using more than six non-collinear diffusion gradients, it is possible to determine the full diffusion tensor, which is the starting point for techniques like fiber tracking and quantitative analyses of the principal eigenvectors of diffusion. Studies of DTI in dementia have consistently shown altered diffusion (tract) properties in accordance with the pattern of neurodegenerative pathology [5961]. For example, widespread abnormalities in the temporal lobe (but also elsewhere in the brain) were found in AD [60]. Microstructural WM changes can even already be detected in cognitively normal individuals in the pre-MCI stage, and may serve as a potential imaging marker of early AD-related brain changes [62,63]. WM tract damage has also been shown in bvFTD. Previous studies showed that compared to AD, WM integrity was lost in bvFTD especially in the frontal and bilateral temporal regions [64,65]. By using DTI, tract-specific pathology can be demonstrated, which may be specifically linked to the clinical syndrome at hand and therefore helps in distinguishing AD from bvFTD. 164

In chapter 3.4 we found that measuring WM damage can improve the discrimination between AD and bvFTD, which is in line with other studies showing increased classification rates between AD and bvFTD in an early disease stage when combining GM with WM measurements [66,67]. In a study on patients with corticobasal degeneration and progressive supranuclear palsy, the authors even showed the potential of DTI as a diagnostic marker at the single-subject level [68]. DTI, therefore, constitutes a promising tool for differentiation of various types of dementia, even in the earliest disease stages, quantifying damage in normal appearing WM. Furthermore, more knowledge regarding the exact timing and anatomical location of pathological WM changes will contribute to more insight into the molecular substrates of the different dementias, thereby expanding treatment options and increasing chances of a good clinical response and, eventually, aid in designing and timing of future interventions for disease prevention. However, DTI sequences are not yet implemented in standard scanning protocols and are not yet suitable for the daily clinical practice as they still lack reliability and stability of measures across multiple sites employing different scanners with different field strengths and scan parameters [69]. Furthermore, post-processing methods are not technically mature enough to use DTI images in the radiological assessment.

Imaging brain tissue for a better understanding of disease pathology The characteristic lesions of AD at the microscopic level are extracellular neuritic plaques, consisting of a core composed of beta amyloid (Aβ), and intracellular neurofibrillary tangles (NFT) consisting of hyperphosphorylated tau-protein (tau). Neuritic plaques are amyloid plaques surrounded by degenerating neuritis filled with tau pathology and are the best histological markers of AD. Amyloid plaques are found in both non-demented and demented patients, while neuritic plaques are only found in demented patients. The staging system developed by Braak and Braak describes the extent, location and the presumed sequence of accumulating neurofibrillary tangle pathology, which in AD is thought to start in the transentorhinal and entorhinal areas, before spreading to the hippocampus, the association cortices, and the rest of the cortex [37]. Imaging GM, could help us to draw conclusions about the distribution of microscopic pathology as plaques and tangles are responsible for GM atrophy. Besides this prototypic distribution of Aβ and tau, which is often found in AD patients with the typical clinical presentation and MRI profile, atypical variants, e.g. early-onset patients with posterior cortical atrophy, suggest that the origin and spread of tau pathology originating from the transentorhinal and hippocampal area might not be the only pattern of pathological progression in AD. As we showed in chapter 2.1, spreading of AD pathology might differ between individuals and certain subtypes of AD may have proportionally greater involvement of the cortex than of the hippocampus. Age of onset could be a driving factor in this regional vulnerability as other studies showed that AD patients with a hippocampal sparing subtype were younger at age-at-onset, and had more widespread cortical involvement than the typical and limbic-predominant AD subtypes [70,71]. This might imply that the Braak stages as described in the early nineties do not hold for all AD patients but need to be adapted for specific subgroups [37,72]. 165

Imaging of GM atrophy can also give information about the chronology of pathological processes, as we showed in chapter 2.3. Baseline volumes of DGM structures in MCI were not predictive of progression to AD, despite the observation of decreasing volumes from controls to MCI to AD. These results could indicate that subcortical involvement is a late phenomenon of AD and that AD could be regarded as a primarily cortical disease. This hypothesis is supported by clinical evidence that symptoms related to frontostriatal circuit disruption in AD generally appear in the later disease stages [73-75]. FTD is genetically and pathologically heterogeneous without a clear relationship between the clinical phenotypes and the underlying pathogenetics. Up to a third of patients with FTD will have an autosomal dominant family history of the disease. Mutations in six genes have been associated with genetic FTD although only two of these, progranulin (GRN) and microtubule-associated protein tau (MAPT) are common causes. FTD clinical syndromes are usually associated with one of the frontotemporal lobar degeneration (FTLD) pathologies. Two major pathological pathological types of FTLD were described, those with tau-positive (FTLD-tau) and those with tau-negative, ubiquitin-positive pathology (FTLD-U). However, it has been shown that FTLD-U actually consists of three separate groups: TDP-43-postitive pathology (FTLD-TDP), fused-in-sarcoma protein positive pathology (FTLD-FUS) and cases which are both TDP-43 and FUS-negative. Each of these major pathological types also has a number of subtypes [76,77]. Because of this wide heterogeneity of patients with FTD, it is extremely challenging to study this group. Neuroimaging may help in explaining the underlying pathology. In contrast to AD, FTD is associated with prominent volume loss of the DGM structures, even in earlier disease stages [29,48,78,79]. This is in line with our findings in chapter 3.1 that volumes of caudate nucleus, globus pallidus and nucleus accumbens are smaller in FTD compared to both AD patients and patients with subjective complaints. These findings help to explain some of the symptomatology of this multifaceted disorder, as the involvement of basal nuclei as part of the frontostriatal circuits in FTD fits with the signs and symptoms of this disease, that include behavioral abnormalities and extrapyramidal symptoms [48,80]. Structural changes in components of these circuits could lead to Wallerian degeneration of the connecting fibers and eventually to failure of the whole circuit. Furthermore, these imaging findings can also give more information about the different underlying neuropathology of FTD. Whereas amyloid deposits are mainly found in the cerebral cortex, tau inclusions are found in subcortical regions as well [81,82]. Next to tau, FUS is also associated with atrophy of caudate nucleus [83], Pick bodies (PiD) have been found in the putamen [84] and MAPT mutations are associated with basal ganglia atrophy [85]. These findings point to an important crux of determining whether neuroanatomical patterns can be useful to predict pathology in groups of patients who present with the same clinical diagnosis, especially in syndromes that have heterogeneous pathology as FTD. Correlations between imaging and pathology have been observed in patients with bvFTD: FTLD-TDP type 1 pathology was associated with a different pattern of atrophy from those seen in CBD and PiD pathologies [86166

88].These findings suggest that imaging in patients with FTD syndromes could be used to predict pathology in these individuals, regardless of their clinical presentation. It is also important to recognize that each of the FTD clinical syndromes can also be associated with AD pathology i.e. a prominent and primary behavioral syndrome indistinguishable from bvFTD may occur [89,90]. Also, the term ‘frontal AD’ is used by some people to indicate a clinical syndrome in which patients have features of episodic memory impairment characteristic of AD and also early behavioral symptoms characteristic of bvFTD [91]. Some patients that present with the clinical picture of bvFTD may turn out to have AD. An imaging signature of AD, showing temporoparietal or parietal atrophy with relative sparing of the medial temporal lobes, can be seen on MRI scans of patients who have a clinical presentation suggestive of an FTD syndrome [23,24]. In these cases, neuroimaging showed that a combination of reduced temporoparietal volumes and large hippocampal volumes enabled discrimination of patients with FTLD-like clinical symptoms and AD pathology from individuals with identical clinical diagnoses but with FTLD pathology [24]. In addition, in individuals with CBS, reduced temporoparietal volume could be used to predict which patients had AD pathology, as opposed to CBD pathology [92]. Imaging the WM could also provide more information about the underlying pathology and how this pathology spreads through the brain. Neurotoxic processes may target the neuron first, leading to neuronal death, followed by degeneration of dendrites and axons. In this case, axonal degeneration in the WM thought to result from decreased axonal transport subsequent to dysfunction or degeneration of cell bodies in the GM. This process of WM degeneration secondary to cortical neurodegeneration is comparable to Wallerian degenerative processes. Wallerian degeneration assumes preferential WM loss or disconnection along the posterior-anterior axis of the brain, secondary to GM pathology in neighboring cortical areas [93,94]. On the other hand, WM could also be damaged more directly [95,96]. This type of WM pathology, known as demyelination, could interfere with transmission velocity by conduction delay and increased refractory period of the axon [97]. Delaying this transmission mechanism may influence synchronization of impulses which are essential for integration of information across the distributed neuronal networks underlying higher cognitive functions [98]. Elaborating on this, GM could be secondary to disconnected WM pathways. This may be the case in bvFTD, where DTI measurements show widespread FA reductions in the frontal, anterior temporal, anterior corpus callosum, inferior frontooccipital fasciculus and bilateral anterior cingulum [64,66,99-101] and increased L1 and L23 values in FTD compared to AD [64]. We showed in chapter 3.4 that this WM integrity loss was more widespread than the GM atrophy found in bvFTD. This could be indicative for a direct targeting of the WM instead of a secondary consequence of GM damage. Together with the results on DGM structures from chapter 3.1, where bvFTD patients have more subcortical brain damage compared to AD patients [48,78,79] these results are suggestive that bvFTD is more a network disease, with involvement of the whole frontal-striatal circuits, including the connecting WM tracts and DGM structures. These network failures may also explain the clinical symptoms of bvFTD, as impaired self-monitoring, theory of mind capabilities, perception of 167

emotions, and changes in personality, which are all coordinated by frontal networks [80]. DTI measurements of WM could shed more light on the anatomical networks and at which point of the pathological cascade these networks start to fail. Finally, a combination of DTI and resting-state functional MRI (rs-fMRI) data would offer a unique opportunity to improve the understanding of the structural basis underlying brain functional changes. Even though DTI has not been used so far to study the underlying pathology, there is some evidence that changes in the individual eigenvalues of the tensor can provide information about the specifics of WM damage. Increased diffusivity found in the WM is indicative for tissue degeneration and findings of reduced degree of anisotropy are considered to reflect cytoarchitecture changes i.e. axonal degeneration and/or demyelination. Decreased axial diffusion has been associated with axonal damage in mouse models [102], perhaps reflecting increased barriers to organized diffusion in the axial plane. Increased radial diffusion has been associated with damage to myelin [103], perhaps reflecting increased diffusion in the plane orthogonal to the axial plane. However it is still unclear what pathological mechanism causes such microstructural WM damage. The question of whether such tissue loss and cytoarchitecture changes are secondary to or independent of GM pathology remains. To summarize, discovering different patterns of GM and WM atrophy with MRI sheds light on the microscopic histological changes in the neurodegenerative diseases as they are inevitably associated with each other. Especially in patients where the clinicopathological correlations are weak, a strong imaging signature consisting of cortical, subcortical and WM tract information could be vital to help target future treatments and provide an accurate prognosis.

Methodological considerations Patient selection All patients who were included in the studies presented in this thesis were included from the Amsterdam Dementia cohort and collected within the framework of the National Program Brain and Cognition (both VU university medical center and Leiden University medical center/Erasmus medical center Rotterdam). They underwent a one-day thorough examination for clinical evaluation. All diagnoses were made in a multidisciplinary consensus meeting according to the core clinical criteria of the National Institute on Aging and the Alzheimer’s Association workgroup for probable AD and according to the clinical diagnostic criteria of FTD [1,2,4,5]. Even though, patients were carefully screened, the possibility of misdiagnosis cannot be ruled out. The absence of pathological and genetic data also contributes to possible misdiagnoses. Due to the focus of the Alzheimer Center VUmc on early-onset dementia, AD patients that participated in the studies were relatively young, which may hamper generalizability of our results to an average AD patient, who on average is ten years older. Some of the studies in this thesis are multicenter studies as patients were included in the VUmc and LUMC. A potential strength of multicenter studies is 168

that the generalizability of the results is strong. However it also has limitations, as patient groups will be less homogenous than in one-center studies. When studying atrophy patterns of AD or bvFTD it is important to be aware of the composition of the groups one intends to examine, as different pathological and genetic subtypes can present with different forms of atrophy. In the case of bvFTD patients can also have a ‘benign’ or ‘nonprogressive’ form [104]. These so called phenocopy syndromes should not be included when studying patterns of atrophy in FTD. In the current project, we took care only to include patients with probable bvFTD. Data acquisition and analysis In many clinical situations, a qualitative visual assessment of images is sufficient for diagnostic purposes. With trained eyes, it is reproducible and differentiates reasonably well a normal from a diseased state [11,25,105-107]. However, many scans differ from the predicted patterns of atrophy, which combined with large between– rater variability results in low sensitivity of these scales. More complex analytical tools such as volume measures of specific structures are more sensitive but are very timeconsuming, and also do not provide enough accuracy for differential diagnosis. Because of their high sensitivity and detailed information on a more microscopic level, quantitative image post-processing techniques, may overcome the shortcomings of visual rating scales and volumetric measurements. Besides a lot of advantages, these image analysis techniques should be used only with a sufficient amount of expertise and results should always be interpreted with caution. In general, one should bear in mind that segmentation of brain images (e.g. VBM, FreeSurfer) critically depends on the quality of the input image, as well as on availability of adequate computing power, and that the image analysis is suited for the type of image available. Furthermore, quantitative image post-processing techniques require sophisticated post-imaging analysis and even though they are mostly automatic, they still require a certain amount of human interaction. Gray matter analysis – VBM, FreeSufer and FIRST Voxel-based morphometry involves a voxel-wise comparison of the local concentration of GM between groups of subjects. It calculates brain volume by segmenting tissue into GM and WM using prior probability maps, whereas FreeSurfer calculates cortical thickness using an estimate of the width of the cortical GM. Although these techniques assess two different metrics - volume and thickness - they are both essentially markers of neurodegeneration and of atrophy of the cortex. The advantages of these techniques are that they assess atrophy throughout the whole brain and do not require prespecified and prelocalized regions or features of interest. Therefore a priori hypotheses are not required for these data-driven methods. In addition, it reduces the likelihood of a Type II error that can occur when ROIs are not placed in the brain region that differs most between groups or that correlates most strongly with some dependent variable (e.g. cognitive performance). However, evaluation on a voxel-by-voxel basis is complicated by Type 1 error inflation due to multiple comparisons. Other disadvantages arise from the fact that the exact imageprocessing is not uniform across studies, and that the spatial extent of the results will 169

depend on how user-specified options are changed. Varying the type and level of statistical correction, e.g. using uncorrected data versus family-wise error corrected data, largely influences the results of a study. In all studies in this thesis the results are corrected for multiple-comparisons minimizing the probability of Type 1 error. Another problem is the need for spatial smoothing, and arbitrariness of choosing the spatial smoothing extent. It has been shown that the extent of the significant findings depends on the size of the smoothing kernel [108]. Especially when studying neurodegeneration, greater smoothing tends to increase sensitivity at the expense of specificity and makes it harder to localize an effect anatomically [109]. Next to user-specific changes, another disadvantage of voxel-based morphometry is that it only assesses mean differences between groups, and does not allow for variability within groups or provide data at the individual level, which makes precise prognostication in a given individual difficult. However, as we have shown, the techniques provide valuable data on imaging signatures across types of dementia and provide the necessary first step to the development of useful biomarkers. The algorithm FIRST (FMRIB’s integrated registration and segmentation tool) [110] is a model-based segmentation/registration tool for 15 different subcortical structures. For the segmentation, it uses shape/appearance models, which are constructed from 336 manually segmented images. FIRST is easy to use and no deviation from the standard instructions are necessary to obtain good results. However, visual checks after every step of the image analysis pipeline, which has been done in all of our studies, is advisable. FIRST has been shown to give accurate and robust results for the segmentation of subcortical structures and that it performs comparable or better to other automatic methods [110-112]. White matter analysis – TBSS TBSS bring together the strengths of voxel-based and tractography-based approaches. It is fully automated, and investigates WM in the whole brain, not requiring prespecification of tracts of interest. It overcomes alignment problems by working in the space of individual subjects’ tractography results and do not necessarily require presmoothing [113]. Next to these advantages, careful interpretation is needed if there are confounding effects, such as within-scan head motion. The most obvious effects of increased head motion are increased image blurring and biased FA. This could lead to misinterpretation of apparent subject group differences, if for example a patient group had greater head motion than a control group. This problem is not in general resolved through the use of the TBSS approach. Another area where careful interpretation is needed is in regions of crossing tracts or tract junctions. Voxelwise statistics are difficult to estimate and interpret at tract junctions or crossings. An apparent reduction in FA at junctions can in fact be due to an increase in one of the tracts feeding into the junction [113,114]. Furthermore, there is the possibility that pathology could reduce FA so strongly that potential areas of interest may be wrongly excluded from analysis. Pathologies like gross stroke or large tumors, which are likely to seriously disrupt tracts (and FA) are unlikely to be suitable for TBSS analysis. 170

However, considering these shortcomings, TBSS is a reliable method for estimating localized change in fractional anisotropy, a useful marker for anatomical brain connectivity across different subjects. As the above discussed problems are not in general be resolved through the use of the TBSS approach, we excluded subjects with severe WM damage and movement artifacts. Furthermore, all images were carefully inspected after every analysis step. Diagnostic Systems for Single-Subject Diagnosis - PRONTO The holy grail of computational techniques may be the one of diagnostic assistants. Multivariate pattern recognition approaches can be extremely useful in distinguishing between groups of subjects (e.g. healthy controls versus patients), and their predictive power can potentially be used as a diagnostic tool for single subjects in a clinical setting [115]. PRoNTo automatically discovers regularities in data through the use of computer algorithms, and with the use of these regularities takes actions such as classifying the data into different categories [116]. Not only diagnosis but also prognosis can be performed [117]. One common mistake, when using linear models, relates to the temptation of interpreting the model weights images as statistical parametric maps (SPMs). Contrary to SPMs, it is the combination of all weights that defines the model and therefore the weights at each voxel are dependent on one another. No voxel-wise statistical tests assuming independence can be performed on them. This leads to interpretability issues, since most neuroscientists look to find not only how information is encoded in the brain but also where in the brain this information resides. Even though PRONTO has the potential to be used as a multiclass classifier, one should bear in mind that it can only classify cases according to already learned patterns. New cases which do not belong to one of the learned groups will be incorrectly assigned to one of these. In chapter 3.2 we showed that PRoNTo provides a comprehensive and user-friendly software framework for multivariate analysis based on machine learning models for neuroimaging data. It constitutes a promising diagnostic assistant for the daily practice and showed good to excellent accuracy in classifying new subjects to a diagnostic group. Advanced MR techniques and second line imaging When structural imaging is equivocal or does not lead to the diagnosis, functional imaging may add diagnostic value. Newer imaging techniques such as ASL and resting state fMRI are likely to be increasingly used as studies already showed their additional value in the workup for dementia. In an earlier study we showed that region-ofinterest-based cerebral blood flow comparisons showed different perfusion patterns in AD and FTD patients. Partial volume corrected cortical cerebral blood flow values were lowest in the frontal lobes in FTD patients, and in the temporal lobes in AD patients [118]. Within the framework of the Brain & Cognition project, resting-state functional MRI (rs-fMRI) was used to study functional connectivity between AD and bvFTD. We showed that whole-brain functional connectivity differed between patients with bvFTD and patients with AD. Compared to AD bvFTD patients had 171

decreased functional connectivity between the lateral visual cortical network and the lateral occipital cortex and cuneal cortex, and between the auditory system network and the angular gyrus. Patients with AD showed decreased functional connectivity between the dorsal visual stream network and lateral occipital cortex and opercular cortex [119]. These findings support the idea that imaging of resting state functional connectivity is sensitive to detect disease-related functional connectivity network changes in neurodegenerative diseases and could be useful for differentiation at the group level. Two other studies have shown that changes in functional connectivity precede the onset of clinical symptoms and atrophy in individuals with MAPT (microtubule-associated protein tau) or GRN (progranulin) mutations, suggesting that functional MRI could provide useful biomarkers for FTLD at the preclinical stage [120,121]. Nevertheless, more research is needed to validate these sequences on group level. For the implication in the daily clinical practice as well as for single-subject diagnosis, acquisition of rs-fMRI has to be optimized and analysis techniques have to be validated . Imaging techniques other than volumetric MRI could also provide promising biomarker options. Second-line neuroimaging investigation includes metabolic information obtained by using single-photon emission computed tomography (SPECT) or positron emission tomography (PET). As in the early stages of FTD there may not exist any discernible atrophy, Fludeoxyglucose(18F)-PET or SPECT might demonstrate decreased metabolism or hypoperfusion preceding tissue loss on structural imaging. Abnormalities on molecular imaging could also be useful in predicting the presence of AD pathology. In a study employing SPECT, a characteristic AD-like pattern of parietal hypoperfusion was identified in patients with CBS, and could distinguish these individuals from those with CBD pathology [122]. However, one should note that temporoparietal abnormalities on functional imaging are not restricted to patients with AD, and have also been observed in patients with FTLD pathology [123]. The predictive value of patterns observed using PET and SPECT imaging may, therefore, be limited. Although detection of amyloid by means of PET provides an excellent biomarker for AD pathology in patients with FTD syndromes, for FTLD pathology such markers are insufficiently investigated so far. However, newer PET imaging methods have recently been investigated (e.g. cholinergic imaging in FTD, PSP and CBS [124] and newer 18F amyloid labeling compounds) and in the future the implementation of PET ligands that would bind to tau, TDP-43 or FUS may offer a huge advance in the ability to make a molecular diagnosis for FTLD [125-128].

Conclusions and future perspectives The standard MRI sequences generally used in dementia screening are very useful in discriminating between dementia and healthy aging. In the differentiation between AD and bvFTD these sequences are sufficient when the radiological diagnosis fits well with the clinical criteria and patients present with the typical atrophy patterns. However, there can be a gray area where clinical and imaging data do not fit, patients are showing with atypical presentations or overlapping imaging signatures. Nevertheless it is important to use these sequences in the radiological assessment in a 172

standardized manner, e.g. by the use of visual rating scales to quantify objectively what can be seen by eyeballing. DTI has been shown to be promising for an early and differential diagnosis between AD and bvFTD but is not yet eligible for the daily radiological practice for singlesubject diagnosis as this sequence needs post-processing and image analysis before conclusions can be drawn. As DTI measures have been shown to be sensitive for early WM damage not visible on conventional MRI sequences, DTI could elucidate the earliest point at which structural changes occur by focusing on patients with normal structural GM scans. The combination of patterns of GM atrophy and WM integrity sheds light on the microscopic histological changes in AD and bvFTD. Especially in patients where the clinicopathological correlations are weak, a strong imaging signature consisting of cortical, subcortical and WM tract information could be vital to help target future treatments and provide an accurate prognosis. The combination of DTI and rs-fMRI data would offer a unique opportunity to improve the understanding of the structural basis underlying brain functional changes. Furthermore, a multimodal approach that combines different imaging sequences and techniques, or a combination of neuropsychological testing and MRI, can improve non-invasive, in vivo distinction between AD and bvFTD. For clinical implications it is therefore important to develop and evaluate image analysis methods which can be used in the daily practice and for single subject diagnosis to eventually support the clinical diagnosis on a daily basis. To do this it is necessary to a) find MRI signatures for all pathological subtypes of AD and FTD based on GM and WM - these imaging signatures have the potential to help in the diagnosis and prognosis of dementia as it can provide information on which abnormal protein is causing the disease; b) to use pattern recognition approaches more to eventually serve as diagnostic assistants; and c) to conduct larger longitudinal studies to find regional distribution of tissue loss that would provide measurements for the assessment of disease modifying treatments.

173

Reference List [1]

[2]

[3] [4]

[5]

[6]

[7]

[8]

[9] [10]

[11]

[12]

[13]

[14]

174

McKhann G, Drachman D, Folstein M, Katzman R, Price D, Stadlan EM (1984) Clinical diagnosis of Alzheimer's disease: report of the NINCDS-ADRDA Work Group under the auspices of Department of Health and Human Services Task Force on Alzheimer's Disease. Neurology 34, 939-944. Neary D, Snowden JS, Gustafson L, Passant U, Stuss D, Black S, Freedman M, Kertesz A, Robert PH, Albert M, Boone K, Miller BL, Cummings J, Benson DF (1998) Frontotemporal lobar degeneration: a consensus on clinical diagnostic criteria. Neurology 51, 1546-1554. Scheltens P , Rockwood K (2011) How golden is the gold standard of neuropathology in dementia? Alzheimers Dement 7, 486-489. Rascovsky K, Hodges JR, Knopman D, Mendez MF, Kramer JH, Neuhaus J et al. (2011) Sensitivity of revised diagnostic criteria for the behavioural variant of frontotemporal dementia. Brain 134, 2456-2477. McKhann GM, Knopman DS, Chertkow H, Hyman BT, Jack CR, Jr., Kawas CH et al. (2011) The diagnosis of dementia due to Alzheimer's disease: recommendations from the National Institute on Aging-Alzheimer's Association workgroups on diagnostic guidelines for Alzheimer's disease. Alzheimers Dement 7, 263-269. Dubois B, Feldman HH, Jacova C, DeKosky ST, Barberger-Gateau P, Cummings J, Delacourte A, Galasko D, Gauthier S, Jicha G, Meguro K, O'Brien J, Pasquier F, Robert P, Rossor M, Salloway S, Stern Y, Visser PJ, Scheltens P (2007) Research criteria for the diagnosis of Alzheimer's disease: revising the NINCDS-ADRDA criteria. Lancet Neurol 6, 734-746. Boelaarts L, Scheltens P, de JJ (2014) Using Magnetic Resonance Imaging in Diagnosing Dementia: A Dutch Outpatient Memory Clinics Survey. Dement Geriatr Cogn Disord 38, 281-285. Erkinjuntti T, Lee DH, Gao F, Steenhuis R, Eliasziw M, Fry R, Merskey H, Hachinski VC (1993) Temporal lobe atrophy on magnetic resonance imaging in the diagnosis of early Alzheimer's disease. Arch Neurol 50, 305-310. Jack CR, Jr., Petersen RC, O'Brien PC, Tangalos EG (1992) MR-based hippocampal volumetry in the diagnosis of Alzheimer's disease. Neurology 42, 183-188. Kesslak JP, Nalcioglu O, Cotman CW (1991) Quantification of magnetic resonance scans for hippocampal and parahippocampal atrophy in Alzheimer's disease. Neurology 41, 51-54. Scheltens P, Leys D, Barkhof F, Huglo D, Weinstein HC, Vermersch P, Kuiper M, Steinling M, Wolters EC, Valk J (1992) Atrophy of medial temporal lobes on MRI in "probable" Alzheimer's disease and normal ageing: diagnostic value and neuropsychological correlates. J Neurol Neurosurg Psychiatry 55, 967972. Korf ES, Wahlund LO, Visser PJ, Scheltens P (2004) Medial temporal lobe atrophy on MRI predicts dementia in patients with mild cognitive impairment. Neurology 63, 94-100. Bresciani L, Rossi R, Testa C, Geroldi C, Galluzzi S, Laakso MP, Beltramello A, Soininen H, Frisoni GB (2005) Visual assessment of medial temporal atrophy on MR films in Alzheimer's disease: comparison with volumetry. Aging Clin Exp Res 17, 8-13. Scheltens P, Launer LJ, Barkhof F, Weinstein HC, van Gool WA (1995) Visual assessment of medial temporal lobe atrophy on magnetic resonance imaging: interobserver reliability. J Neurol 242, 557-560.

[15]

[16]

[17] [18]

[19]

[20]

[21]

[22]

[23]

[24]

[25]

[26] [27]

[28]

[29]

Barber R, McKeith I, Ballard C, O'Brien J (2002) Volumetric MRI study of the caudate nucleus in patients with dementia with Lewy bodies, Alzheimer's disease, and vascular dementia. J Neurol Neurosurg Psychiatry 72, 406-407. van de Pol LA, Hensel A, van der Flier WM, Visser PJ, Pijnenburg YA, Barkhof F, Gertz HJ, Scheltens P (2006) Hippocampal atrophy on MRI in frontotemporal lobar degeneration and Alzheimer's disease. J Neurol Neurosurg Psychiatry 77, 439-442. Raji CA, Lopez OL, Kuller LH, Carmichael OT, Becker JT (2009) Age, Alzheimer disease, and brain structure. Neurology 73, 1899-1905. Cherbuin N, Leach LS, Christensen H, Anstey KJ (2007) Neuroimaging and APOE genotype: a systematic qualitative review. Dement Geriatr Cogn Disord 24, 348-362. Schott JM, Ridha BH, Crutch SJ, Healy DG, Uphill JB, Warrington EK, Rossor MN, Fox NC (2006) Apolipoprotein e genotype modifies the phenotype of Alzheimer disease. Arch Neurol 63, 155-156. Lehmann M, Koedam EL, Barnes J, Bartlett JW, Ryan NS, Pijnenburg YA, Barkhof F, Wattjes MP, Scheltens P, Fox NC (2012) Posterior cerebral atrophy in the absence of medial temporal lobe atrophy in pathologically-confirmed Alzheimer's disease. Neurobiol Aging 33, 627-627. Lehmann M, Crutch SJ, Ridgway GR, Ridha BH, Barnes J, Warrington EK, Rossor MN, Fox NC (2011) Cortical thickness and voxel-based morphometry in posterior cortical atrophy and typical Alzheimer's disease. Neurobiol Aging 32, 14661476. Frisoni GB, Pievani M, Testa C, Sabattoli F, Bresciani L, Bonetti M, Beltramello A, Hayashi KM, Toga AW, Thompson PM (2007) The topography of grey matter involvement in early and late onset Alzheimer's disease. Brain 130, 720-730. Lehmann M, Rohrer JD, Clarkson MJ, Ridgway GR, Scahill RI, Modat M, Warren JD, Ourselin S, Barnes J, Rossor MN, Fox NC (2010) Reduced cortical thickness in the posterior cingulate gyrus is characteristic of both typical and atypical Alzheimer's disease. J Alzheimers Dis 20, 587-598. Whitwell JL, Jack CR, Jr., Przybelski SA, Parisi JE, Senjem ML, Boeve BF, Knopman DS, Petersen RC, Dickson DW, Josephs KA (2011) Temporoparietal atrophy: a marker of AD pathology independent of clinical diagnosis. Neurobiol Aging 32, 1531-1541. Koedam EL, Lehmann M, van der Flier WM, Scheltens P, Pijnenburg YA, Fox N, Barkhof F, Wattjes MP (2011) Visual assessment of posterior atrophy development of a MRI rating scale. Eur Radiol 21, 2618-2625. Frisoni GB, Fox NC, Jack CR, Jr., Scheltens P, Thompson PM (2010) The clinical use of structural MRI in Alzheimer disease. Nat Rev Neurol 6, 67-77. Bakkour A, Morris JC, Wolk DA, Dickerson BC (2013) The effects of aging and Alzheimer's disease on cerebral cortical anatomy: specificity and differential relationships with cognition. Neuroimage 76, 332-344. Thompson PM, Hayashi KM, de ZG, Janke AL, Rose SE, Semple J, Herman D, Hong MS, Dittmer SS, Doddrell DM, Toga AW (2003) Dynamics of gray matter loss in Alzheimer's disease. J Neurosci 23, 994-1005. Rabinovici GD, Seeley WW, Kim EJ, Gorno-Tempini ML, Rascovsky K, Pagliaro TA, Allison SC, Halabi C, Kramer JH, Johnson JK, Weiner MW, Forman MS, Trojanowski JQ, Dearmond SJ, Miller BL, Rosen HJ (2007) Distinct MRI atrophy patterns in autopsy-proven Alzheimer's disease and frontotemporal lobar degeneration. Am J Alzheimers Dis Other Demen 22, 474-488. 175

[30]

[31]

[32]

[33]

[34]

[35]

[36]

[37] [38] [39]

[40]

[41]

[42]

[43]

[44]

[45]

176

Barnes J, Godbolt AK, Frost C, Boyes RG, Jones BF, Scahill RI, Rossor MN, Fox NC (2007) Atrophy rates of the cingulate gyrus and hippocampus in AD and FTLD. Neurobiol Aging 28, 20-28. Du AT, Schuff N, Kramer JH, Rosen HJ, Gorno-Tempini ML, Rankin K, Miller BL, Weiner MW (2007) Different regional patterns of cortical thinning in Alzheimer's disease and frontotemporal dementia. Brain 130, 1159-1166. Whitwell JL, Jack CR, Jr., Baker M, Rademakers R, Adamson J, Boeve BF, Knopman DS, Parisi JF, Petersen RC, Dickson DW, Hutton ML, Josephs KA (2007) Voxel-based morphometry in frontotemporal lobar degeneration with ubiquitin-positive inclusions with and without progranulin mutations. Arch Neurol 64, 371-376. Rosen HJ, Perry RJ, Murphy J, Kramer JH, Mychack P, Schuff N, Weiner M, Levenson RW, Miller BL (2002) Emotion comprehension in the temporal variant of frontotemporal dementia. Brain 125, 2286-2295. Boccardi M, Pennanen C, Laakso MP, Testa C, Geroldi C, Soininen H, Frisoni GB (2002) Amygdaloid atrophy in frontotemporal dementia and Alzheimer's disease. Neurosci Lett 335, 139-143. Whitwell JL, Sampson EL, Watt HC, Harvey RJ, Rossor MN, Fox NC (2005) A volumetric magnetic resonance imaging study of the amygdala in frontotemporal lobar degeneration and Alzheimer's disease. Dement Geriatr Cogn Disord 20, 238-244. Barnes J, Whitwell JL, Frost C, Josephs KA, Rossor M, Fox NC (2006) Measurements of the amygdala and hippocampus in pathologically confirmed Alzheimer disease and frontotemporal lobar degeneration. Arch Neurol 63, 1434-1439. Braak H , Braak E (1991) Neuropathological stageing of Alzheimer-related changes. Acta Neuropathol 82, 239-259. LaBar KS , Cabeza R (2006) Cognitive neuroscience of emotional memory. Nat Rev Neurosci 7, 54-64. Basso M, Yang J, Warren L, MacAvoy MG, Varma P, Bronen RA, van Dyck CH (2006) Volumetry of amygdala and hippocampus and memory performance in Alzheimer's disease. Psychiatry Res 146, 251-261. Poulin SP, Dautoff R, Morris JC, Barrett LF, Dickerson BC (2011) Amygdala atrophy is prominent in early Alzheimer's disease and relates to symptom severity. Psychiatry Res 194, 7-13. Whitwell JL, Josephs KA, Rossor MN, Stevens JM, Revesz T, Holton JL, Al-Sarraj S, Godbolt AK, Fox NC, Warren JD (2005) Magnetic resonance imaging signatures of tissue pathology in frontotemporal dementia. Arch Neurol 62, 1402-1408. Rosen HJ, Gorno-Tempini ML, Goldman WP, Perry RJ, Schuff N, Weiner M, Feiwell R, Kramer JH, Miller BL (2002) Patterns of brain atrophy in frontotemporal dementia and semantic dementia. Neurology 58, 198-208. Hornberger M, Yew B, Gilardoni S, Mioshi E, Gleichgerrcht E, Manes F, Hodges JR (2012) Ventromedial-frontopolar prefrontal cortex atrophy correlates with insight loss in frontotemporal dementia and Alzheimer's disease. Hum Brain Mapp Galton CJ, Patterson K, Xuereb JH, Hodges JR (2000) Atypical and typical presentations of Alzheimer's disease: a clinical, neuropsychological, neuroimaging and pathological study of 13 cases. Brain 123 Pt 3, 484-498. Graham A, Davies R, Xuereb J, Halliday G, Kril J, Creasey H, Graham K, Hodges J (2005) Pathologically proven frontotemporal dementia presenting with severe amnesia. Brain 128, 597-605.

[46]

[47]

[48]

[49]

[50]

[51] [52] [53] [54] [55] [56]

[57]

[58]

[59] [60]

[61] [62]

[63]

Johnson JK, Head E, Kim R, Starr A, Cotman CW (1999) Clinical and pathological evidence for a frontal variant of Alzheimer disease. Arch Neurol 56, 12331239. Du AT, Schuff N, Kramer JH, Rosen HJ, Gorno-Tempini ML, Rankin K, Miller BL, Weiner MW (2007) Different regional patterns of cortical thinning in Alzheimer's disease and frontotemporal dementia. Brain 130, 1159-1166. Looi JC, Lindberg O, Zandbelt BB, Ostberg P, Andersen C, Botes L, Svensson L, Wahlund LO (2008) Caudate nucleus volumes in frontotemporal lobar degeneration: differential atrophy in subtypes. AJNR Am J Neuroradiol 29, 1537-1543. Gootjes L, Teipel SJ, Zebuhr Y, Schwarz R, Leinsinger G, Scheltens P, Moller HJ, Hampel H (2004) Regional distribution of white matter hyperintensities in vascular dementia, Alzheimer's disease and healthy aging. Dement Geriatr Cogn Disord 18, 180-188. Smith CD, Snowdon DA, Wang H, Markesbery WR (2000) White matter volumes and periventricular white matter hyperintensities in aging and dementia. Neurology 54, 838-842. Englund E (1998) Neuropathology of white matter changes in Alzheimer's disease and vascular dementia. Dement Geriatr Cogn Disord 9 Suppl 1, 6-12. Malloy P, Correia S, Stebbins G, Laidlaw DH (2007) Neuroimaging of white matter in aging and dementia. Clin Neuropsychol 21, 73-109. Husted C (1994) Contributions of neuroimaging to diagnosis and monitoring of multiple sclerosis. Curr Opin Neurol 7, 234-241. Bronge L , Wahlund LO (2003) Prognostic significance of white matter changes in a memory clinic population. Psychiatry Res 122, 199-206. Kalaria RN , Ballard C (1999) Overlap between pathology of Alzheimer disease and vascular dementia. Alzheimer Dis Assoc Disord 13 Suppl 3, S115-S123. Chua TC, Wen W, Slavin MJ, Sachdev PS (2008) Diffusion tensor imaging in mild cognitive impairment and Alzheimer's disease: a review. Curr Opin Neurol 21, 83-92. Abe O, Aoki S, Hayashi N, Yamada H, Kunimatsu A, Mori H, Yoshikawa T, Okubo T, Ohtomo K (2002) Normal aging in the central nervous system: quantitative MR diffusion-tensor analysis. Neurobiol Aging 23, 433-441. Rose SE, Chen F, Chalk JB, Zelaya FO, Strugnell WE, Benson M, Semple J, Doddrell DM (2000) Loss of connectivity in Alzheimer's disease: an evaluation of white matter tract integrity with colour coded MR diffusion tensor imaging. J Neurol Neurosurg Psychiatry 69, 528-530. Smith CD (2010) Neuroimaging through the course of Alzheimer's disease. J Alzheimers Dis 19, 273-290. Damoiseaux JS, Smith SM, Witter MP, Sanz-Arigita EJ, Barkhof F, Scheltens P, Stam CJ, Zarei M, Rombouts SA (2009) White matter tract integrity in aging and Alzheimer's disease. Hum Brain Mapp 30, 1051-1059. Bozzali M , Cherubini A (2007) Diffusion tensor MRI to investigate dementias: a brief review. Magn Reson Imaging 25, 969-977. Zhuang L, Sachdev PS, Trollor JN, Kochan NA, Reppermund S, Brodaty H, Wen W (2012) Microstructural white matter changes in cognitively normal individuals at risk of amnestic MCI. Neurology 79, 748-754. Chua TC, Wen W, Chen X, Kochan N, Slavin MJ, Trollor JN, Brodaty H, Sachdev PS (2009) Diffusion tensor imaging of the posterior cingulate is a useful biomarker of mild cognitive impairment. Am J Geriatr Psychiatry 17, 602613. 177

[64]

[65]

[66]

[67]

[68]

[69]

[70]

[71]

[72]

[73] [74]

[75] [76] [77] [78]

[79]

178

Zhang Y, Schuff N, Du AT, Rosen HJ, Kramer JH, Gorno-Tempini ML, Miller BL, Weiner MW (2009) White matter damage in frontotemporal dementia and Alzheimer's disease measured by diffusion MRI. Brain 132, 2579-2592. Chen TF, Lin CC, Chen YF, Liu HM, Hua MS, Huang YC, Chiu MJ (2009) Diffusion tensor changes in patients with amnesic mild cognitive impairment and various dementias. Psychiatry Res 173, 15-21. McMillan CT, Brun C, Siddiqui S, Churgin M, Libon D, Yushkevich P, Zhang H, Boller A, Gee J, Grossman M (2012) White matter imaging contributes to the multimodal diagnosis of frontotemporal lobar degeneration. Neurology 78, 1761-1768. Kuceyeski A, Zhang Y, Raj A (2012) Linking white matter integrity loss to associated cortical regions using structural connectivity information in Alzheimer's disease and fronto-temporal dementia: the Loss in Connectivity (LoCo) score. Neuroimage 61, 1311-1323. Sajjadi SA, Acosta-Cabronero J, Patterson K, Diaz-de-Grenu LZ, Williams GB, Nestor PJ (2013) Diffusion tensor magnetic resonance imaging for single subject diagnosis in neurodegenerative diseases. Brain 136, 2253-2261. Teipel SJ, Reuter S, Stieltjes B, Acosta-Cabronero J, Ernemann U, Fellgiebel A, Filippi M, Frisoni G, Hentschel F, Jessen F, Kloppel S, Meindl T, Pouwels PJ, Hauenstein KH, Hampel H (2011) Multicenter stability of diffusion tensor imaging measures: a European clinical and physical phantom study. Psychiatry Res 194, 363-371. Murray ME, Graff-Radford NR, Ross OA, Petersen RC, Duara R, Dickson DW (2011) Neuropathologically defined subtypes of Alzheimer's disease with distinct clinical characteristics: a retrospective study. Lancet Neurol 10, 785-796. Whitwell JL, Dickson DW, Murray ME, Weigand SD, Tosakulwong N, Senjem ML, Knopman DS, Boeve BF, Parisi JE, Petersen RC, Jack CR, Jr., Josephs KA (2012) Neuroimaging correlates of pathologically defined subtypes of Alzheimer's disease: a case-control study. Lancet Neurol 11, 868-877. Rossor MN, Iversen LL, Reynolds GP, Mountjoy CQ, Roth M (1984) Neurochemical characteristics of early and late onset types of Alzheimer's disease. Br Med J (Clin Res Ed) 288, 961-964. Bruen PD, McGeown WJ, Shanks MF, Venneri A (2008) Neuroanatomical correlates of neuropsychiatric symptoms in Alzheimer's disease. Brain 131, 2455-2463. Piccininni M, Di CA, Baldereschi M, Zaccara G, Inzitari D (2005) Behavioral and psychological symptoms in Alzheimer's disease: frequency and relationship with duration and severity of the disease. Dement Geriatr Cogn Disord 19, 276-281. Tekin S , Cummings JL (2002) Frontal-subcortical neuronal circuits and clinical neuropsychiatry: an update. J Psychosom Res 53, 647-654. Rohrer JD (2012) Structural brain imaging in frontotemporal dementia. Biochim Biophys Acta 1822, 325-332. Whitwell JL , Josephs KA (2011) Neuroimaging in frontotemporal lobar degeneration-predicting molecular pathology. Nat Rev Neurol 8, 131-142. Halabi C, Halabi A, Dean DL, Wang PN, Boxer AL, Trojanowski JQ, Dearmond SJ, Miller BL, Kramer JH, Seeley WW (2013) Patterns of striatal degeneration in frontotemporal dementia. Alzheimer Dis Assoc Disord 27, 74-83. Looi JC, Svensson L, Lindberg O, Zandbelt BB, Ostberg P, Orndahl E, Wahlund LO (2009) Putaminal volume in frontotemporal lobar degeneration and Alzheimer disease: differential volumes in dementia subtypes and controls. AJNR Am J Neuroradiol 30, 1552-1560.

[80] [81]

[82]

[83]

[84]

[85]

[86]

[87]

[88]

[89] [90]

[91]

[92]

[93]

Cummings JL (1993) Frontal-subcortical circuits and human behavior. Arch Neurol 50, 873-880. Lantos PL, Cairns NJ, Khan MN, King A, Revesz T, Janssen JC, Morris H, Rossor MN (2002) Neuropathologic variation in frontotemporal dementia due to the intronic tau 10(+16) mutation. Neurology 58, 1169-1175. Kim EJ, Rabinovici GD, Seeley WW, Halabi C, Shu H, Weiner MW, Dearmond SJ, Trojanowski JQ, Gorno-Tempini ML, Miller BL, Rosen HJ (2007) Patterns of MRI atrophy in tau positive and ubiquitin positive frontotemporal lobar degeneration. J Neurol Neurosurg Psychiatry 78, 1375-1378. Rohrer JD, Ridgway GR, Modat M, Ourselin S, Mead S, Fox NC, Rossor MN, Warren JD (2010) Distinct profiles of brain atrophy in frontotemporal lobar degeneration caused by progranulin and tau mutations. Neuroimage 53, 1070-1076. Rohrer JD, Lashley T, Schott JM, Warren JE, Mead S, Isaacs AM, Beck J, Hardy J, de SR, Warrington E, Troakes C, Al-Sarraj S, King A, Borroni B, Clarkson MJ, Ourselin S, Holton JL, Fox NC, Revesz T, Rossor MN, Warren JD (2011) Clinical and neuroanatomical signatures of tissue pathology in frontotemporal lobar degeneration. Brain 134, 2565-2581. Slowinski JL, Schweitzer KJ, Imamura A, Uitti RJ, Strongosky AJ, Dickson DW, Broderick DF, Wszolek ZK (2009) Brainstem atrophy on routine MR study in pallidopontonigral degeneration. J Neurol 256, 827-829. Whitwell JL, Jack CR, Jr., Parisi JE, Knopman DS, Boeve BF, Petersen RC, Dickson DW, Josephs KA (2011) Imaging signatures of molecular pathology in behavioral variant frontotemporal dementia. J Mol Neurosci 45, 372-378. Rankin KP, Mayo MC, Seeley WW, Lee S, Rabinovici G, Gorno-Tempini ML, Boxer AL, Weiner MW, Trojanowski JQ, Dearmond SJ, Miller BL (2011) Behavioral variant frontotemporal dementia with corticobasal degeneration pathology: phenotypic comparison to bvFTD with Pick's disease. J Mol Neurosci 45, 594608. Lee SE, Rabinovici GD, Mayo MC, Wilson SM, Seeley WW, Dearmond SJ, Huang EJ, Trojanowski JQ, Growdon ME, Jang JY, Sidhu M, See TM, Karydas AM, Gorno-Tempini ML, Boxer AL, Weiner MW, Geschwind MD, Rankin KP, Miller BL (2011) Clinicopathological correlations in corticobasal degeneration. Ann Neurol 70, 327-340. Kertesz A, McMonagle P, Blair M, Davidson W, Munoz DG (2005) The evolution and pathology of frontotemporal dementia. Brain 128, 1996-2005. Knopman DS, Boeve BF, Parisi JE, Dickson DW, Smith GE, Ivnik RJ, Josephs KA, Petersen RC (2005) Antemortem diagnosis of frontotemporal lobar degeneration. Ann Neurol 57, 480-488. Johnson JK, Head E, Kim R, Starr A, Cotman CW (1999) Clinical and pathological evidence for a frontal variant of Alzheimer disease. Arch Neurol 56, 12331239. Josephs KA, Whitwell JL, Boeve BF, Knopman DS, Petersen RC, Hu WT, Parisi JE, Dickson DW, Jack CR, Jr. (2010) Anatomical differences between CBScorticobasal degeneration and CBS-Alzheimer's disease. Mov Disord 25, 1246-1252. O'Dwyer L, Lamberton F, Bokde AL, Ewers M, Faluyi YO, Tanner C, Mazoyer B, O'Neill D, Bartley M, Collins DR, Coughlan T, Prvulovic D, Hampel H (2011) Multiple indices of diffusion identifies white matter damage in mild cognitive impairment and Alzheimer's disease. PLoS One 6, e21745179

[94]

[95] [96] [97] [98] [99]

[100]

[101]

[102]

[103]

[104]

[105]

[106]

[107]

[108]

[109]

[110]

180

Di PM, Di IF, Cherubini A, Blundo C, Casini AR, Sancesario G, Passafiume D, Caltagirone C, Spalletta G (2010) When, where, and how the corpus callosum changes in MCI and AD: a multimodal MRI study. Neurology 74, 1136-1142. Malone MJ , Szoke MC (1985) Neurochemical changes in white matter. Aged human brain and Alzheimer's disease. Arch Neurol 42, 1063-1066. Xu J, Chen S, Ahmed SH, Chen H, Ku G, Goldberg MP, Hsu CY (2001) Amyloid-beta peptides are cytotoxic to oligodendrocytes. J Neurosci 21, RC118Felts PA, Baker TA, Smith KJ (1997) Conduction in segmentally demyelinated mammalian central axons. J Neurosci 17, 7267-7277. Srinivasan R (1999) Spatial structure of the human alpha rhythm: global correlation in adults and local correlation in children. Clin Neurophysiol 110, 1351-1362. Zhang Y, Schuff N, Ching C, Tosun D, Zhan W, Nezamzadeh M, Rosen HJ, Kramer JH, Gorno-Tempini ML, Miller BL, Weiner MW (2011) Joint assessment of structural, perfusion, and diffusion MRI in Alzheimer's disease and frontotemporal dementia. Int J Alzheimers Dis 2011, 546871Hornberger M, Geng J, Hodges JR (2011) Convergent grey and white matter evidence of orbitofrontal cortex changes related to disinhibition in behavioural variant frontotemporal dementia. Brain 134, 2502-2512. Avants BB, Cook PA, Ungar L, Gee JC, Grossman M (2010) Dementia induces correlated reductions in white matter integrity and cortical thickness: a multivariate neuroimaging study with sparse canonical correlation analysis. Neuroimage 50, 1004-1016. Song SK, Sun SW, Ju WK, Lin SJ, Cross AH, Neufeld AH (2003) Diffusion tensor imaging detects and differentiates axon and myelin degeneration in mouse optic nerve after retinal ischemia. Neuroimage 20, 1714-1722. Song SK, Sun SW, Ramsbottom MJ, Chang C, Russell J, Cross AH (2002) Dysmyelination revealed through MRI as increased radial (but unchanged axial) diffusion of water. Neuroimage 17, 1429-1436. Kipps CM, Hodges JR, Hornberger M (2010) Nonprogressive behavioural frontotemporal dementia: recent developments and clinical implications of the 'bvFTD phenocopy syndrome'. Curr Opin Neurol 23, 628-632. Pasquier F, Leys D, Weerts JG, Mounier-Vehier F, Barkhof F, Scheltens P (1996) Interand intraobserver reproducibility of cerebral atrophy assessment on MRI scans with hemispheric infarcts. Eur Neurol 36, 268-272. Wahlund LO, Almkvist O, Blennow K, Engedahl K, Johansson A, Waldemar G, Wolf H (2005) Evidence-based evaluation of magnetic resonance imaging as a diagnostic tool in dementia workup. Top Magn Reson Imaging 16, 427-437. Kloppel S, Stonnington CM, Barnes J, Chen F, Chu C, Good CD, Mader I, Mitchell LA, Patel AC, Roberts CC, Fox NC, Jack CR, Jr., Ashburner J, Frackowiak RS (2008) Accuracy of dementia diagnosis: a direct comparison between radiologists and a computerized method. Brain 131, 2969-2974. Henley SM, Ridgway GR, Scahill RI, Kloppel S, Tabrizi SJ, Fox NC, Kassubek J (2010) Pitfalls in the use of voxel-based morphometry as a biomarker: examples from huntington disease. AJNR Am J Neuroradiol 31, 711-719. Reimold M, Slifstein M, Heinz A, Mueller-Schauenburg W, Bares R (2006) Effect of spatial smoothing on t-maps: arguments for going back from t-maps to masked contrast images. J Cereb Blood Flow Metab 26, 751-759. Patenaude B, Smith SM, Kennedy DN, Jenkinson M (2011) A Bayesian model of shape and appearance for subcortical brain segmentation. Neuroimage 56, 907922.

[111]

[112]

[113]

[114]

[115] [116] [117]

[118]

[119]

[120]

[121]

[122]

[123]

Babalola KO, Patenaude B, Aljabar P, Schnabel J, Kennedy D, Crum W, Smith S, Cootes T, Jenkinson M, Rueckert D (2009) An evaluation of four automatic methods of segmenting the subcortical structures in the brain. Neuroimage 47, 1435-1447. Morey RA, Selgrade ES, Wagner HR, Huettel SA, Wang L, McCarthy G (2010) Scanrescan reliability of subcortical brain volumes derived from automated segmentation. Hum Brain Mapp 31, 1751-1762. Smith SM, Jenkinson M, Johansen-Berg H, Rueckert D, Nichols TE, Mackay CE, Watkins KE, Ciccarelli O, Cader MZ, Matthews PM, Behrens TE (2006) Tractbased spatial statistics: voxelwise analysis of multi-subject diffusion data. Neuroimage 31, 1487-1505. Tuch DS, Salat DH, Wisco JJ, Zaleta AK, Hevelone ND, Rosas HD (2005) Choice reaction time performance correlates with diffusion anisotropy in white matter pathways supporting visuospatial attention. Proc Natl Acad Sci U S A 102, 12212-12217. Kloppel S, Abdulkadir A, Jack CR, Jr., Koutsouleris N, Mourao-Miranda J, Vemuri P (2012) Diagnostic neuroimaging across diseases. Neuroimage 61, 457-463. Bishop C.M. (2006) Pattern Recognition and Machine Learning (Information Science and Statistics), Springer, New York. Mourao-Miranda J, Almeida JR, Hassel S, de OL, Versace A, Marquand AF, Sato JR, Brammer M, Phillips ML (2012) Pattern recognition analyses of brain activation elicited by happy and neutral faces in unipolar and bipolar depression. Bipolar Disord 14, 451-460. Binnewijzend MA, Kuijer JP, van der Flier WM, Benedictus MR, Moller CM, Pijnenburg YA, Lemstra AW, Prins ND, Wattjes MP, van Berckel BN, Scheltens P, Barkhof F (2014) Distinct perfusion patterns in Alzheimer's disease, frontotemporal dementia and dementia with Lewy bodies. Eur Radiol 24, 2326-2333. Hafkemeijer A., Mรถller C, Dopper EG, Jiskoot L, Schouten T.M., van Swieten JC, van der Flier WM, Vrenken H, Barkhof F, Scheltens P, van der Grond J, Rombouts S (2014) Resting state functional connectivity differences between behavioral variant frontotemporal dementia and Alzheimer's disease. Whitwell JL, Josephs KA, Avula R, Tosakulwong N, Weigand SD, Senjem ML, Vemuri P, Jones DT, Gunter JL, Baker M, Wszolek ZK, Knopman DS, Rademakers R, Petersen RC, Boeve BF, Jack CR, Jr. (2011) Altered functional connectivity in asymptomatic MAPT subjects: a comparison to bvFTD. Neurology 77, 866874. Dopper EG, Rombouts SA, Jiskoot LC, den HT, de Graaf JR, de K, I, Hammerschlag AR, Seelaar H, Seeley WW, Veer IM, van Buchem MA, Rizzu P, van Swieten JC (2014) Structural and functional brain connectivity in presymptomatic familial frontotemporal dementia. Neurology 83, e19-e26. Hu WT, Rippon GW, Boeve BF, Knopman DS, Petersen RC, Parisi JE, Josephs KA (2009) Alzheimer's disease and corticobasal degeneration presenting as corticobasal syndrome. Mov Disord 24, 1375-1379. Womack KB, Diaz-Arrastia R, Aizenstein HJ, Arnold SE, Barbas NR, Boeve BF, Clark CM, DeCarli CS, Jagust WJ, Leverenz JB, Peskind ER, Turner RS, Zamrini EY, Heidebrink JL, Burke JR, DeKosky ST, Farlow MR, Gabel MJ, Higdon R, Kawas CH, Koeppe RA, Lipton AM, Foster NL (2011) Temporoparietal hypometabolism in frontotemporal lobar degeneration and associated imaging diagnostic errors. Arch Neurol 68, 329-337. 181

[124]

[125]

[126]

[127]

[128]

182

Hirano S, Shinotoh H, Shimada H, Aotsuka A, Tanaka N, Ota T, Sato K, Ito H, Kuwabara S, Fukushi K, Irie T, Suhara T (2010) Cholinergic imaging in corticobasal syndrome, progressive supranuclear palsy and frontotemporal dementia. Brain 133, 2058-2068. Fodero-Tavoletti MT, Okamura N, Furumoto S, Mulligan RS, Connor AR, McLean CA, Cao D, Rigopoulos A, Cartwright GA, O'Keefe G, Gong S, Adlard PA, Barnham KJ, Rowe CC, Masters CL, Kudo Y, Cappai R, Yanai K, Villemagne VL (2011) 18F-THK523: a novel in vivo tau imaging ligand for Alzheimer's disease. Brain 134, 1089-1100. Okamura N, Furumoto S, Fodero-Tavoletti MT, Mulligan RS, Harada R, Yates P, Pejoska S, Kudo Y, Masters CL, Yanai K, Rowe CC, Villemagne VL (2014) Non-invasive assessment of Alzheimer's disease neurofibrillary pathology using 18F-THK5105 PET. Brain 137, 1762-1771. Xia CF, Arteaga J, Chen G, Gangadharmath U, Gomez LF, Kasi D, Lam C, Liang Q, Liu C, Mocharla VP, Mu F, Sinha A, Su H, Szardenings AK, Walsh JC, Wang E, Yu C, Zhang W, Zhao T, Kolb HC (2013) [(18)F]T807, a novel tau positron emission tomography imaging agent for Alzheimer's disease. Alzheimers Dement 9, 666-676. Maruyama M, Shimada H, Suhara T, Shinotoh H, Ji B, Maeda J, Zhang MR, Trojanowski JQ, Lee VM, Ono M, Masamoto K, Takano H, Sahara N, Iwata N, Okamura N, Furumoto S, Kudo Y, Chang Q, Saido TC, Takashima A, Lewis J, Jang MK, Aoki I, Ito H, Higuchi M (2013) Imaging of tau pathology in a tauopathy mouse model and in Alzheimer patients compared to normal controls. Neuron 79, 1094-1108.

Addendum

183

Nederlandse samenvatting

Visualiseren van patronen van weefselverlies Naar een beter onderscheid tussen verschillende vormen van dementie Dementie is een verzamelnaam voor verschillende ziektes. De bekendste vormen van dementie zijn de ziekte van Alzheimer, vasculaire dementie, frontotemporale dementie en dementie met Lewy lichaampjes. In Nederland zijn er op dit moment ruim 260.000 mensen met een vorm van dementie. Door de vergrijzing van de bevolking zal dit aantal alleen maar gaan stijgen: in 2040 zal ruim een half miljoen mensen lijden aan dementie. Ook op jongere leeftijd kunnen mensen dementie krijgen. Naar schatting zijn er in Nederland 12.000 mensen met dementie die jonger zijn dan 65 jaar. Dit noemen we preseniele dementie. Bij dementie sterven in de loop van de ziekte steeds meer zenuwcellen en verbindingen in de hersenen af. De hersenen kunnen niet meer goed functioneren waardoor een patiënt steeds minder kan en uiteindelijk volledig afhankelijk wordt van de zorg van anderen. Nog altijd zijn er geen geneesmiddelen voor de ziekte van Alzheimer en de andere vormen van dementie maar om alsnog adequate zorg te kunnen leveren, is het van belang in een zo vroeg mogelijk stadium een diagnose te stellen. De ziekte van Alzheimer is de meest voorkomende vorm van dementie en is verantwoordelijk voor 43% van alle gevallen van dementie. Hoewel het voor het 65ste levensjaar minder vaak voorkomt, is de ziekte van Alzheimer nog steeds de meest frequente oorzaak van preseniele dementie, gevolgd door frontotemporale dementie. De eerste prominente symptomen van de seniele vorm (ziekte begint na het 65 levensjaar) van de ziekte van Alzheimer zijn problemen met het geheugen. Daarnaast hebben patiënten ook soms moeite met het vinden van woorden, problemen met het besef van tijd, ruimtelijke oriëntatie en met de hogere uitvoerende functies, zoals probleemoplossend denken, inhibitie, planning en overzicht houden. Deze vaardigheden zijn in het begin van de ziekte nog subtiel aangetast maar verergeren naarmate de ziekte vordert. Patiënten met de preseniele vorm hebben in het begin van de ziekte vooral moeite met ruimtelijk oriëntatie, met de hogere uitvoerende functies en problemen met de aandacht. Het geheugen is in de meeste gevallen relatief gespaard. Helaas hebben niet alle patiënten deze ‘prototypische’ symptomen wat de diagnosestelling in veel gevallen moeilijk maakt. Vooral bij relatief jonge patiënten zonder geheugenproblemen wordt die diagnose ziekte van Alzheimer nog vaak gemist. Naast dat er veel heterogeniteit binnen één vorm van dementie optreed, is het onderscheid met andere vormen van dementie ook vaak een uitdaging. Vooral in het begin van de ziekte, waar symptomen nog niet uitgesproken zijn en vaak overlappen tussen verschillende vormen van dementie. Een van de vormen van dementie die in het begin vaak moeilijk te onderscheiden is van de ziekte van Alzheimer is de gedragsvariant van frontotemporale dementie (bvFTD).

184

BvFTD is een andere vorm van dementie die relatief vaak voorkomt bij mensen die dementie op jonge leeftijd (<65 jaar) krijgen. De symptomen van bvFTD zijn zeer heterogeen maar meestal staan veranderingen in het gedrag en de persoonlijkheid op de voorgrond. Gedragsveranderingen kunnen bestaan uit passiviteit, initiatiefverlies of het verlies van het empathisch vermogen, maar ook kan er bijvoorbeeld onrust, agressie en ontremming optreden. Vaak zijn ook de hogere uitvoerende functies en het behouden van aandacht aangedaan. Het geheugen is meestal relatief gespaard en in het begin van de ziekte presteren patiĂŤnten met bvFTD binnen de normen op testen die de cognitieve functies meten. Alhoewel bvFTD meestal op jongere leeftijd optreed, zijn 20-25% van de patiĂŤnten met bvFTD ouder dan 65. Tot op heden worden de diagnoses van Alzheimer en bvFTD gemaakt aan de hand van klinische criteria, maar het sluipende begin van de ziekte en de overlap van symptomen tussen de verschillende ziektebeelden staan een vroege en accurate diagnose in de weg. Het onderscheidend vermogen tussen de ziekte van Alzheimer en bvFTD op basis van deze klinische criteria is tot op heden nog teleurstellend laag. Daarom is er een grote behoefte aan meetmethoden (markers) die vroeg in het ziektebeloop bij de verschillende types dementie optreden. Een benadering van deze behoefte is het onderzoeken van veranderingen in de hersenen met behulp van Magnetic resonance imaging (MRI). MRI is de laatste jaren een steeds prominentere rol gaan spelen bij de diagnostiek van dementie. Waar MRI eerst met name gebruikt werd om andere oorzaken van cognitieve stoornissen uit te sluiten, wordt MRI steeds vaker gebruikt om het met dementie samenhangende weefselverlies (atrofie) aan te tonen. Inmiddels kunnen bepaalde patronen van atrofie al toegeschreven worden aan een bepaalde vorm van dementie en helpt MRI op deze manier bij het stellen van een diagnose. Het is bekend dat atrofie van de mediale temporaal kwab of de pariĂŤtale kwab de diagnose van de ziekte van Alzheimer ondersteunt en atrofie van de frontale en temporale delen van de hersenen past bij de diagnose van bvFTD. Helaas laten niet alle MRI scans deze typische patronen van atrofie zien en is er veel overlap binnen een ziekte maar ook tussen de ziekte van Alzheimer en bvFTD. Daarom zijn er meer gedetailleerde meetmethodes voor MRI scans nodig, die meer informatie over weefselverlies kunnen geven, dan zichtbaar is op het oog. Er zijn verschillende methoden om een hersenscan kwantitatief te analyseren. Meestal richt men zich hierbij op de verschillende onderdelen van de hersenen. De hersenen zijn opgebouwd uit grijze en witte stof. De eerste wordt ook wel hersenschors genoemd en is de buitenste laag van de grote hersenen. De grijze stof is samengesteld uit neuronen en hun dendrieten en bevat gyri, of hersenwindingen, die worden gescheiden door fissuren, diepe groeven, en sulci, ondiepe groeven. De witte stof bestaat uit de verbindingen tussen de neuronen (axonen) en is wit vanwege de isolerende myelineschedes eromheen. In de grijze stof vindt de gegevensopslag plaats; de witte stof dient voor de connecties tussen hersengebieden. In het binnenste van de hersenen bevinden zich dicht op elkaar gepakte groepen van neuronen, die op basis van hun locatie en hun uiterlijk, diepe 185

grijze stof structuren worden genoemd. Zij lijken een belangrijke schakelfunctie te hebben in de netwerken van de hersenen. De grijze stof en het verlies van neuronen kan in kaart gebracht worden met behulp van verschillende technieken. Eén daarvan is voxel-based morphometry (VBM). Met deze techniek is het mogelijk om de hoeveelheid grijze stof in de hersenen op voxelniveau te meten. Het begrip voxel is een samentrekking van volumetric en pixel en is het driedimensionale equivalent van de pixel. Met behulp van VBM is het mogelijk om patronen van weefselverlies te lokaliseren, de verschillen in het verlies van grijze stof tussen verschillende diagnoses te bepalen en een gebied van grijze stof te linken aan cognitieve functies, zoals het geheugen. Met andere technieken, zoals FIRST is het mogelijk om de volumes van de diepe grijze stof structuren te bestuderen en een uitspraak te doen over functies, die niet verklaard kunnen worden door het afsterven van de hersenschors. Naast schade aan de corticale en de diepe grijze stof, speelt de beschadiging van de verbindingsbanen in de witte stof ook een belangrijke rol bij dementie. Door nieuwe beeldvormingstechnieken, zoals Diffusion Tensor Imaging (DTI), is het mogelijk om de beschadiging van de witte stof banen in kaart te brengen. Dit gebeurt op microscopisch niveau want deze beschadigingen zijn op het oog niet zichtbaar. Er is al bekend dat DTI metingen in dementie anders zijn dan bij gezonde veroudering en er lijkt een verband te zijn met de symptomen van dementie. Met Tract-based spatial statistics (TBSS) kan DTI data geanalyseerd worden. Hiermee kan een uitspraak gedaan worden over in welke regio’s in the hersenen de witte stof bij een groep van patiënten meer beschadigd is dan bij een andere groep. Voor de vergelijking tussen de ziekte van Alzheimer en bvFTD is DTI nog niet vaak gebruikt. Wellicht dat deze aanpak kan bijdragen om de twee vormen beter van elkaar te kunnen onderscheiden. Tot op heden is gebleken dat het onderzoeken van alleen één onderdeel van de hersenen ontoereikende informatie oplevert voor een betrouwbaar onderscheid tussen de ziekte van Alzheimer en bvFTD. Ook zouden gegevens over veranderingen in de tijd waardevolle informatie over het startpunt en het verloop van het ziekteproces kunnen opleveren. Tot slot is het van groot belang dat de verkregen resultaten ook toepasbaar zijn in de dagelijkse klinische praktijk. Resultaten gebaseerd op groepsanalyses, zeggen nog weinig over de individuele patiënt en te ingewikkelde MRI analyse programma’s zijn in de dagelijkse routine in een ziekenhuis niet inzetbaar. In de laatste jaren zijn er programma’s ontwikkeld, die precies deze knelpunten proberen op te lossen. Zogenaamde Support Vector Machines (SVM) proberen door het leren van patronen van weefselbeschadiging die typisch zijn voor een bepaald ziektebeeld, nieuwe individuele patiënten te classificeren in één van de van het systeem geleerde ziektebeelden. De doelstelling van dit proefschrift was het bestuderen van patronen van corticale en diepe grijze stof atrofie en de beschadiging van witte stof banen bij de ziekte van Alzheimer en bvFTD, in vergelijking met gezonde controlepersonen. Hiermee hopen we beter onderscheid tussen de twee vormen van dementie te kunnen maken en 186

meer inzicht in de onderliggende processen van hersenverandering te verkrijgen. We zullen inzoomen op de vragen: 1. Zijn patronen van grijze stof atrofie gerelateerd aan specifieke vorm van dementie? 2. Kan het meten van witte stof banen bijdragen aan het beter onderscheiden tussen de ziekte van Alzheimer en bvFTD? 3. Kan de informatie die we verkrijgen door het analyseren van een MRI scan ingezet worden om een diagnostische marker te ontwikkelen? Hoofdstuk 2 richt zich op de vraag hoe patronen van corticale en diepe grijze stof atrofie gerelateerd zijn aan verschillende types van de ziekte van Alzheimer. In hoofdstuk 2.1 hebben we patronen van grijze stof atrofie bij patiënten met de preseniele en de seniele variant van de ziekte van Alzheimer in kaart gebracht. Hierbij vergeleken we de twee groepen met elkaar en met een oude en een jonge controle groep met behulp van VBM. De resultaten lieten zien dat leeftijd en diagnose onafhankelijk van elkaar het volume van de hippocampus beïnvloed. De interactie tussen leeftijd en diagnose liet zien dat de precuneus het meest beschadigd was bij de preseniele vorm van de ziekte van Alzheimer. De resultaten bevestigen dat patronen van atrofie zelfs binnen de ziekte van Alzheimer kunnen variëren en dat het belangrijk is om Alzheimer patiënten altijd te vergelijken met een controlegroep van dezelfde leeftijd. Bovendien is het belangrijk om vooral bij jongere patiënten op de gebieden in de achterste delen van de hersenen – in het bijzonder de precuneus – te letten bij het beoordelen van een MRI scan. In hoofdstuk 2.2 hebben we de reeds bestaande visuele beoordelingsschaal voor Posterior Cortical Atrophy (PCA) – weefselverlies in de achterste delen van de hersenen – gevalideerd om haar gebruik in de klinische praktijk te stimuleren en te rechtvaardigen. De visuele PCA-schaal geeft met scores van 0-geen atrofie tot 3eindstadium atrofie aan hoeveel weefselverlies aanwezig is. Hiervoor hebben we grijze stof volumetrie en VBM gebruikt. Het bleek dat de visuele PCA schaal betrouwbaar de atrofie in de achterste delen van de hersenen meet. Er was een duidelijk onderscheid te maken tussen hersenen die een PCA score van 1 of hoger hadden vergeleken met hersenen die geen weefselverlies in de achterste delen van de hersenen hadden (PCA score 0). Ook de scores 1, 2 en 3 correspondeerden met de hoeveelheid grijze stof gemeten met volumetrie en VBM. Het bleek dat vooral de hoeveelheid grijze stof in de lobulus parietalis inferior de visuele scoring beïnvloedde. Onze resultaten maken duidelijk dat radiologen de visuele beoordelingsschaal voor PCA betrouwbaar in de dagelijkse praktijk kunnen toepassen. In hoofdstuk 2.3 hebben we onderzocht of volumes van de diepe grijze stof structuren de conversie van Mild Cognitive Impairment (MCI, milde cognitieve stoornis) naar de ziekte van Alzheimer kunnen voorspellen. Hiervoor hebben we patiënten met MCI, de ziekte van Alzheimer en controles met elkaar vergeleken. We hebben ook de relatie tussen volumes van de diepe grijze stof structuren met de mate van cognitieve beperkingen onderzocht. We vonden dat de volumes van de hippocampus en de nucleus accumbens op de eerst gemaakte MRI scan konden voorspelden welke MCI patiënten een hoger risico hadden om te converteren naar de ziekte van Alzheimer na een tijd van twee jaar. Alhoewel de andere diepe grijze stof structuren tussen 187

controles, MCI en de ziekte van Alzheimer in volume significant verschilden, voorspelden zij niet of iemand met MCI naar de ziekte van Alzheimer ging converteren. Behalve voor de globus pallidus, vonden we voor alle diepe grijze stof structuren een positieve associatie met de cognitieve metingen. Hoofdstuk 3 richt zich op de verschillende patronen van grijze stof atrofie tussen de ziekte van Alzheimer en bvFTD. Het is bekend dat bij bvFTD de frontostriatale circuits zijn aangedaan. Daarom onderzochten we in hoofdstuk 3.1 of de diepe grijze stof structuren, als schakelstations van deze circuits, bij bvFTD meer aangedaan zijn dan bij patiënten met de ziekte van Alzheimer en bij controles. Ook keken we naar de relatie tussen volume van de diepe grijze stof structuren, cognitieve functies en neuropsychiatrische symptomen. We vonden dat de nucleus accumbens, nucleus caudatus en de globus pallidus meer zijn aangedaan bij bvFTD vergeleken met patiënten met de ziekte van Alzheimer en controles. De samenhang tussen volumes van de diepe grijze stof structuren en cognitie was verschillend bij patiënten met bvFTD, de ziekte van Alzheimer en controles. Met dit onderzoek hebben we laten zien dat naast frontale corticale atrofie de diepe grijze stof structuren in bvFTD meer aangedaan zijn dan in de ziekte van Alzheimer. In hoofdstuk 3.2 onderzochten we hoe goed een SVM individuele patiënten op basis van hun grijze stof patroon kan indelen. Hiervoor hebben we gebruik gemaakt van een standaard beschikbare 3D T1 MRI scan, die vervolgens automatisch scans van verschillende patiënten, gebaseerd op de hoeveelheid grijze stof in een voxel, ging indelen in één van twee groepen. Om de resultaten zo veelzeggend mogelijk te maken hebben we gebruik gemaakt van MRI scans van verschillende MRI machines uit verschillende ziekenhuizen. Het was mogelijk om met een SVM een accurate indeling te maken tussen twee groepen. Dit is een eerste stap in richting van een diagnose voor een individuele patiënt op basis van MRI. In hoofdstuk 3.3 hebben we grijze stof verlies in de loop van de tijd in patiënten met de ziekte van Alzheimer, bvFTD en gezonde controles bestudeerd. Het bleek dat patiënten met de ziekte van Alzheimer en bvFTD meer grijze stof verliezen dan controles. Patiënten met de ziekte van Alzheimer verloren ook meer grijze stof dan patiënten met bvFTD, maar dit was alleen statistisch significant in de rechter insula. Vergeleken met controles, zagen wij grijze stof verlies over het gehele brein met een voorkeur voor de achterste gebieden bij de ziekte van Alzheimer en een specifiek grijze stof verlies in de frontale hersengebieden bij bvFTD. De cognitieve achteruitgang sloot aan bij het grijze stof verlies. Alhoewel het alleen significant was in vergelijking met gezonde controles, gingen patiënten met de ziekte van Alzheimer het meest op de cognitieve functies achteruit. BvFTD patiënten gingen vooral op het gebied van aandacht achteruit. In hoofdstuk 3.4 hebben we gekeken naar de diagnostische waarde van corticale en diepe grijze stof en witte stof integriteit. Hiervoor hebben we VBM, FIRST en TBSS gebruikt. Bovendien hebben we gekeken welke meetmethode het beste het onderscheid maakt tussen patiënten met de ziekte van Alzheimer, bvFTD en gezonde controles en of de combinatie van de drie methodes het beste resultaat behaalt bij het onderscheid van deze drie groepen. We vonden zowel verschillen in de corticale en 188

diepe grijze stof en ook in de witte stof, die onafhankelijk van elkaar bijdroegen aan de scheiding tussen de twee vormen van dementie en van gezonde controles. Hoewel patiënten met bvFTD dezelfde ziekteduur als patiënten met de ziekte van Alzheimer hadden, waren bij bvFTD alle drie de metingen het meest aangedaan. Vooral de witte stof was duidelijk meer beschadigd in bvFTD, met name in de frontale gebieden. De combinatie van alle drie meetmethodes kon met 86-91% precisie patiënten met de ziekte van Alzheimer van de andere twee groepen onderscheiden. Metingen van de corticale grijze stof leverde hier de grootste bijdrage. Voor de differentiatie van bvFTD van de andere twee groepen, leverde vooral de afwijkende witte stof de grootste bijdrage. De resultaten suggereren dat het meten van de witte stof met DTI waardevolle aanvullende informatie oplevert in het onderscheid tussen patiënten met de ziekte van Alzheimer en bvFTD. Concluderend kunnen we stellen dat patronen van corticale en diepe grijze stof verschillen tussen controles, patiënten met de ziekte van Alzheimer en bvFTD. Echter, deze verschillen zijn in het begin van de ziekte nog vrij klein, zodat ze alleen met meetmethodes op voxel niveau gedetecteerd kunnen worden. Voor patiënten die al meer grijze stof atrofie vertonen, is de inzet van visuele beoordelingsschalen aan te raden, om atrofie op een betrouwbare en systematische manier in kaart te brengen. Hoewel men zich altijd moet realiseren dat niet elke patiënt een stereotypisch atrofiepatroon, passend bij de klinische symptomen, zal hebben. Het meten van de witte stof beschadiging speelt een belangrijke rol in het onderscheid tussen de ziekte van Alzheimer en bvFTD en geeft ons meer inzicht in de onderliggende pathologische processen die een rol spelen bij deze ziektes. Het grote voordeel is, dat afwijkingen in de witte stof, gemeten met DTI, al vroeg in het ziekteproces zichtbaar zijn en daarom het beginpunt van een ziekte kunnen tonen en een vroeg onderscheid van ziektes kan ondersteunen. Maar ook hier gaat het om afwijkingen die op het oog niet zichtbaar zijn. Bovendien is de DTI scan nog geen onderdeel van het standaard MRI protocol in de meeste ziekenhuizen. Ook vergen de analyses veel expertise en precisie. Samen hebben corticale en diepe grijze en witte stof het potentieel om een diagnostische marker te vormen in het onderscheid tussen patiënten met de ziekte van Alzheimer en bvFTD. Op dit moment zijn deze resultaten echter alleen toepasbaar op groepen van patiënten. De toepasbaarheid op de individuele patiënt bevindt zich nog in de kinderschoenen, maar de eerste hoopgevende stappen zijn gezet met de SVM die goede resultaten opleveren in het onderscheid tussen twee groepen. Voor de toekomst is het belangrijk om MRI evaluatie methodes te ontwikkelen, die in de dagelijkse praktijk en voor de individuele patiënt toegepast kunnen worden. Hiermee kan het diagnostisch proces versneld en nauwkeuriger worden. Het is daarom nodig om a) patronen van MRI afwijkingen te vinden voor alle subtypes van de ziekte van Alzheimer en bvFTD gebaseerd op grijze en witte stof, b) SVMs die patronen van afwijkingen kunnen detecteren, meer te gaan gebruiken en te verbeteren om deze uiteindelijk op de individuele patiënt toe te kunnen passen en c) in grote patiënten cohorten met herhaalde scans de gevonden patronen van hersenafwijkingen duidelijker in kaart te brengen en te verifiëren. 189

Deutsche Zusammenfassung

Visualisierung von strukturellem Gewebeverlust – Auf dem Weg zu besserer Differenzialdiagnostik bei verschiedenen Formen von Demenz Demenz ist ein Oberbegriff für verschiedene Krankheiten. Die bekanntesten Arten von Demenz sind die Alzheimer-Krankheit, vaskuläre Demenz, Frontotemporale Demenz und Demenz mit Lewy-Körpern. In den Niederlanden gibt es derzeit mehr als 260.000 Menschen mit einer Form von Demenz. Durch die Vergreisung der Bevölkerung wird diese Zahl weiter steigen: im Jahr 2040 schätzt man, dass mehr als eine halbe Million Menschen in den Niederlanden die Diagnose Demenz haben. Auch können Menschen in jüngeren Jahren an Demenz erkranken. Es wird geschätzt, dass in den Niederlande 12.000 Menschen, die jünger als 65 Jahre sind, eine Form von Demenz haben. Das Auftreten von Demenz vor dem 65. Lebensjahr nennt man präsenile Demenz. Bei Demenz sterben im Verlauf der Erkrankung mehr und mehr Nervenzellen und Verbindungen im Gehirn ab. Das Gehirn kann nicht mehr ordnungsgemäß funktionieren, wodurch ein Patient immer weniger kann und letztendlich vollkommen abhängig ist von der Hilfe anderer. Noch gibt es keine Medikamente gegen die Alzheimer-Krankheit und anderen Formen der Demenz. Trotzdem ist es wichtig so früh wie möglich die richtige Diagnose zu stellen, um die angemessene Versorgung der Patienten gewährleisten zu können. Die Alzheimer-Krankheit ist die häufigste Form der Demenz und sie ist verantwortlich für 43% aller Fälle von Demenz. Obwohl die Alzheimer-Krankheit vor dem 65. Lebensjahr weniger häufig auftritt, ist sie immer noch die häufigste Ursache für präsenile Demenz, gefolgt von Frontotemporaler Demenz. Die ersten bekannten Symptome der senilen Form (Krankheit beginnt nach dem 65. Lebensjahr) der Alzheimer-Krankheit sind Gedächtnisstörungen. Zusätzlich haben Patienten manchmal auch Schwierigkeiten bei der Suche nach Worten, Probleme mit dem Begriff von Zeit, räumlicher Orientierung und mit komplexeren Tätigkeiten, wie bspw. Probleme zu lösen, primäre Reaktionen zu unterdrücken, Pläne zu machen und den Überblick zu behalten. Diese Fähigkeiten sind am Anfang der Krankheit noch subtil beeinflusst, aber verschlimmern sich im Verlauf der Krankheit. Patienten mit der präsenilen Form haben zu Beginn der Krankheit vor allem Schwierigkeiten mit der räumlichen Orientierung, mit dem Ausführen komplexer Tätigkeiten und Probleme mit der Aufmerksamkeit. Das Gedächtnis ist in den meisten Fällen relativ verschont. Leider weisen nicht alle Patienten diese prototypischen Symptome auf, was die Diagnose in vielen Fällen erschwert. Vor allem bei jüngeren Patienten ohne Gedächtnisprobleme wird oft nicht an die Alzheimer-Krankheit gedacht. Nicht nur die Heterogenität innerhalb einer Form von Demenz erschwert die Frühdiagnose, die Unterscheidung von anderen Formen der Demenz wird damit auch zu einer großen Herausforderung. Vor allem zu Beginn der Krankheit, wenn die Symptome noch nicht ausgeprägt sind und sich oftmals überlappen mit anderen 190

Formen der Demenz. Eine andere Form der Demenz, die am Anfang schwer von der Alzheimer-Krankheit zu unterscheiden ist, ist die Verhaltens-Variante der Frontotemporalen Demenz (BvFTD). BvFTD ist eine andere Form der Demenz, die relativ häufig bei jüngeren Menschen (< 65 Jahre) auftritt. Die Symptome von BvFTD sind sehr heterogen, aber in der Regel fallen Veränderungen im Verhalten und der Persönlichkeit am meisten auf. Verhaltensänderungen können aus Passivität, Initiativverlust oder Schwund von Empathie bestehen, aber auch Unruhe, Aggression und Enthemmung können auftreten. Patienten haben auch oft Probleme mit der Ausführung komplexer Tätigkeiten und der Aufmerksamkeit. Das Gedächtnis ist in der Regel relativ verschont und Patienten mit BvFTD schneiden bei den meisten kognitiven Tests mit normalen Ergebnissen ab. Obwohl BvFTD meist in jüngeren Jahren auftritt, sind 20-25 % der Patienten mit BvFTD älter als 65 Jahre. Bis heute wird die Diagnose der Alzheimer-Krankheit und von BvFTD auf Grund von diagnostischen Kriterien gestellt, aber der schleichende Beginn der Krankheit und die Überlappung der Symptome zwischen den verschiedenen Formen, stehen einer frühzeitigen und genauen Diagnose im Weg. Die diagnostische Präzision bei der Unterscheidung zwischen BvFTD und der Alzheimer-Krankheit, basierend auf diesen diagnostischen Kriterien, ist immer noch enttäuschend niedrig. Daher besteht ein großes Bedürfnis nach Messmethoden, die früh im Verlauf der Erkrankung bei den verschiedenen Arten von Demenz durchgeführt werden können. Eine mögliche Annäherung an diese Notwendigkeit ist die Möglichkeit, Veränderungen im Gehirn mit Magnetresonanztomographie (MRT) festzustellen. MRT spielt in den letzten Jahren eine zunehmend wichtigere Rolle bei der Diagnose von Demenz. Zunächst wurde MRT insbesondere eingesetzt um andere Ursachen für kognitive Beeinträchtigungen auszuschließen. Neuerdings wird zunehmend auf das MRT zurückgegriffen um Gewebsschwund (Atrophie) sichtbar zu machen, der indikativ für eine bestimmte Form von Demenz ist. Inzwischen sind bestimmte Muster von Atrophie ein Hinweis für eine bestimmte Form von Demenz und auf diese Weise unterstützt das MRT den diagnostischen Prozess. Mittlerweile ist bekannt, dass Atrophie des medialen Teils des Temporallappens und des Parietallappens bezeichnend sind für die Alzheimer-Krankheit, wohingegen die Atrophie des Frontalund Temporallappens BvFTD zugeschrieben wird. Leider weisen nicht alle MRT-Bilder diese typischen Muster der Atrophie auf und es gibt viele Überschneidungen innerhalb einer Form von Demenz, ebenso zwischen der Alzheimer-Krankheit und BvFTD. Deshalb werden detailliertere Messverfahren für MRT-Bilder benötigt, die mehr Informationen zum Verlust von Gewebe liefern können, als man mit dem bloßen Auge sehen kann. Es gibt verschiedene Methoden um einen Gehirn-Scan quantitativ zu analysieren. In der Regel untersucht man die verschiedenen Bestandteile des Gehirns. Das Gehirn besteht aus der grauen und weißen Substanz. Die erste ist auch bekannt als die Hirnrinde und ist die äußerste Schicht des Großhirns. Die graue Substanz besteht aus 191

Neuronen und ihren Dendriten und enthält Hirnwindungen (Gyri), die durch tiefe Gräben (Fissuren) und flachere Furchen (Sulci) getrennt sind. Die weiße Substanz besteht aus den Verbindungen zwischen den Neuronen (Axone) und ist aufgrund der isolierenden Schicht aus Myelin weiß gefärbt. In der grauen Substanz befindet sich der Datenspeicher; die weiße Substanz ist für die Verbindungen zwischen den Hirnarealen zuständig. Im Inneren des Gehirns befinden sich dicht aufeinander gepackte Neuronengruppen, die wegen ihrer Lage und ihres Aussehens als subkortikale graue Kerne bezeichnet werden. Sie spielen eine wichtige Funktion in den Regelschleifen des Gehirns. Der Schwund der graue Substanz kann mit unterschiedlichen Methoden gemessen werden, zum Beispiel mit Voxel-based Morphometry (VBM). Mit dieser Technik ist es möglich, die Menge der grauen Substanz im Gehirn auf Voxel-Ebene zu messen. Der Begriff Voxel ist eine Zusammensetzung aus volumetric und Pixel und ist das dreidimensionale Äquivalent des Pixels. Mit VBM ist es möglich Strukturen von Gewebeverlust zu lokalisieren, die Unterschiede des Verlusts der grauen Substanz zwischen unterschiedliche Diagnosen aufzudecken und Gehirnareale an kognitive Funktionen wie das Gedächtnis zu verbinden. Mit anderen Techniken, z. B. FIRST ist es möglich, das Volumen der subkortikalen grauen Kerne zu berechnen und Aussagen über Funktionsstörungen von Patienten zu machen, die durch den Schwund der grauen Substanz alleine nicht erklärt werden können. Neben Schäden an der Hirnrinde und den subkortikalen grauen Kernen spielen Schäden an den Verbindungen in der weißen Substanz auch eine wichtige Rolle bei Demenz. Durch neue bildgebende Verfahren, wie bspw. Diffusion Tensor Imaging (DTI), ist es möglich, Schäden an der weißen Substanz sichtbar zu machen. Dies passiert auf mikroskopischer Ebene, weil diese Läsionen nicht für das Auge zu sehen sind. Es ist bereits bekannt, dass DTI-Messungen bei Demenz anders sind als bei gesundem Altern und es scheint eine Verbindung zwischen beschädigter weißer Substanz und den Symptomen der Demenz zu geben. Mit dem Programm Tractbased spatial statistics (TBSS) können DTI-Daten analysiert werden. Dies ermöglicht es, ein Urteil zu fällen in welchen Regionen die weiße Hirnsubstanz bei einer Gruppe von Patienten mehr beschädigt ist als bei einer anderen Gruppe. Für die Differenzierung zwischen der Alzheimer-Krankheit und BvFTD wurde diese Methode noch nicht oft angewendet. Vielleicht kann dieser Ansatz dazu beitragen die beiden Formen besser voneinander unterscheiden zu können. Bis heute hat sich gezeigt, dass nur ein Bestandteil des Gehirns nicht genügend Informationen für eine zuverlässige Unterscheidung zwischen Alzheimer-Krankheit und BvFTD liefert. Außerdem würden Daten über Veränderungen im Laufe der Zeit wertvolle Informationen über den Startpunkt und den Verlauf des Krankheitsprozesses erbringen. Und letztendlich ist es von großer Bedeutung, dass die gefundenen Ergebnisse auch für die tägliche klinische Praxis geeignet sind. Ergebnisse auf der Grundlage von Gruppenanalyse sagen wenig über einzelne Patienten und komplizierte MRI-Analyse-Programme können nicht in den Alltag in einem Krankenhaus integriert werden. In den letzten Jahren wurden Programme 192

entwickelt, die versuchen genau diese Engpässe zu lösen. So genannte Support Vector Machines (SVM) lernen Mustern von Gewebeschäden, die typisch für ein bestimmtes Krankheitsbild sind. Danach wird ein MRT-scan van einem neuen individuelle Patienten mit dem gelernten Muster verglichen und der neue Patient wird in die Kategorie eingeordnet, zu der der MRT-scan am besten passt. Das Ziel dieser Dissertation ist, Atrophie der grauen Substanz und subkortikalen grauen Kernen und Schäden an der weißen Substanz in der Alzheimer-Krankheit und BvFTD, im Vergleich zu gesunden Kontrollpersonen zu erforschen. Hiermit hoffen wir, die zwei Arten von Demenz besser unterscheiden zu können und ein besseres Verständnis für die zugrunde liegenden Prozesse der Gehirnveränderung bei der Alzheimer-Krankheit und BvFTD zu entwickeln. Die folgenden Fragen werden wir versuchen zu beantworten: 1. Gibt es typische Muster von Atrophie der grauen Substanz, die einer spezifischen Form der Demenz zugeordnet werden können? 2. Kann die Messung der weißen Substanz dazu beitragen, die AlzheimerKrankheit und BvFTD besser zu unterscheiden? 3. Sind die Informationen, die wir durch die Analyse eines MRT-Gehirnscans erhalten, ausreichend um ein diagnostisches Instrument für individuelle Patienten zu entwickeln? Kapitel 2 konzentriert sich auf die Frage, wie Muster von Atrophie der grauen Substanz und der subkortikalen grauen Kerne in Zusammenhang stehen mit verschiedenen Arten der Alzheimer-Krankheit. In Kapitel 2.1 haben wir Muster von Atrophie der grauen Substanz bei Patienten mit präseniler und seniler Variante der Alzheimer-Krankheit identifiziert. Wir verglichen beide Gruppen miteinander und mit einer alten und einer jungen Kontrollgruppe mit VBM. Die Ergebnisse zeigten, dass Alter und Diagnose unabhängig voneinander das Volumen des Hippocampus beeinflussen. Die Interaktion zwischen Alter und Diagnose zeigte, dass der Precuneus am meisten beschädigt ist bei der präsenilen Form der Alzheimer-Krankheit. Die Ergebnisse bestätigen, dass Muster von grauer Substanz Atrophie sogar innerhalb der Alzheimer-Krankheit variieren können und dass es wichtig ist, Alzheimer-Patienten immer mit einer Kontrollgruppe des gleichen Alters zu vergleichen. Darüber hinaus ist es wichtig, vor allem bei jüngeren Patienten auf den hinteren Bereich des Gehirns – vor allem den Precuneus – zu achten, wenn ein MRT begutachtet wird. In Kapitel 2.2 haben wir die visuelle Beurteilungsskala für Posterior Cortical Atrophy (PCA)– Gewebeverlust in den hinteren Gebieten des Gehirns – validiert, um seine Verwendung in der klinischen Praxis zu fördern und zu rechtfertigen. Die PCA-Skala misst mit Punktzahlen von 0-Keine Atrophie bis 3-Endphase Atrophie, wie viel Gewebeverlust in den hinteren Gehirnarealen vorhanden ist. Zu diesem Zweck haben wir VBM und Volumetrie der grauen Substanz verwendet. Es stellte sich heraus, dass die visuelle PCA-Skala zuverlässig die Atrophie im hinteren Teil des Gehirns misst. Es gab eine deutliche Unterscheidung zwischen Gehirnen, die mit einer 1 oder höher beurteilt wurden im Vergleich zu Gehirnen, die keinen Gewebe-Verlust in den hinteren Teilen des Gehirns (PCA 0) aufwiesen. Auch korrespondierten die Stufen 1, 2 und 3 mit 193

der Menge der grauen Substanz, gemessen mit Volumetrie und VBM. Es stellte sich heraus, dass vor allem die Menge der grauen Substanz in dem lobulus parietalis inferior die Höhe der visuellen Bewertung beeinflusst. Unsere Ergebnisse machen deutlich, dass die visuelle Beurteilungsskala für PCA in der täglichen Praxis zuverlässig angewendet werden kann. In Kapitel 2.3 haben wir geprüft, ob das Volumen der subkortikalen grauen Kerne die Konversion von Mild Cognitive Impairment (MCI, milde kognitive Beeinträchtigung) die Alzheimer-Krankheit voraussagen kann. Zu diesem Zweck haben wir die Volumen der subkortikalen grauen Kerne von Patienten mit MCI, der Alzheimer-Krankheit und Kontrollpersonen miteinander verglichen. Wir haben auch den Zusammenhang zwischen dem Volumen der subkortikalen grauen Kerne mit dem Grad des kognitiven Verfalls untersucht. Wir fanden heraus, dass das Volumen des Hippocampus und des Nucleus Accumbens voraussagen konnten welche MCI-Patienten ein höheres Risiko haben nach einem Zeitraum von zwei Jahren die Alzheimer-Krankheit zu entwickeln. Obwohl die anderen subkortikalen grauen Kerne zwischen Kontrollpersonen, MCI und der Alzheimer-Krankheit unterschiedlich groß waren, konnten sie nicht vorhersagen, ob jemand mit MCI später die Alzheimer-Krankheit entwickelt. Mit Ausnahme des Globus Pallidus fanden wir für alle subkortikalen grauen Kerne einen positiven Zusammenhang mit den kognitiven Messungen. In Kapitel 3 konzentrierten wir uns auf die verschiedenen Muster der Atrophie der grauen Substanz im Vergleich zwischen der Alzheimer-Krankheit und BvFTD. Es ist bekannt, dass die Frontostriatalen Kreisläufe bei BvFTD betroffen sind. Daher haben wir in Kapitel 3.1 untersucht ob die subkortikalen grauen Kerne, als Zwischenstationen dieser Kreisläufe, bei BvFTD-Patienten mehr beschädigt sind als bei Patienten mit der Alzheimer-Krankheit und Kontrollpersonen. Wir haben auch den Zusammenhang zwischen dem Volumen der subkortikalen grauen Kerne mit dem Grad des kognitiven Verfalls und neuropsychiatrischen Symptomen untersucht. Wir fanden heraus, dass der Nucleus caudatus, der Nucleus Accumbens und der Globus Pallidus bei BvFDT-Patienten mehr beschädigt ist als bei Patienten mit der AlzheimerKrankheit und Kontrollpersonen. Das Verhältnis zwischen dem Volumen der subkortikalen grauen Kerne und Kognition war bei allen Gruppen unterschiedlich. Mit dieser Studie haben wir gezeigt, dass neben der frontalen Atrophie der grauen Substanz, die subkortikalen grauen Kerne in BvFTD mehr betroffen sind als bei der Alzheimer-Krankheit. In Kapitel 3.2 haben wir geprüft, wie gut ein SVM individuelle Patienten mit der Alzheimer-Krankheit und BvFTD auf der Grundlage ihrer atrophierten grauen Substanz in eine Kategorie einteilen kann. Dafür haben wir einen 3D T1 MRT Scan benutzt, der bei einem normalen MRT-Protokoll immer verfügbar ist. Die SVM teilt automatisch die Patienten, basierend auf der Menge der grauen Substanz in einem Voxel, in zwei Gruppen ein. Um die Ergebnisse allgemein anwendbar zu machen, haben wir MRT-Scans von verschiedenen MRT-Maschinen aus mehreren Krankenhäusern verwendet. Es war möglich mit der SVM eine genaue Klassifizierung zwischen den beiden Gruppen zu machen. Dies ist ein erster Schritt in Richtung Diagnose für den individuellen Patienten, auf der Grundlage eines MRT. 194

In Kapitel 3.3 haben wir den Verlust der grauen Substanz im Laufe der Zeit bei Patienten mit der Alzheimer-Krankheit, BvFTD und gesunden Kontrollpersonen untersucht. Es stellte sich heraus, dass Patienten mit der Alzheimer-Krankheit mehr graue Substanz verlieren als Kontrollpersonen über einen Zeitraum von ungefähr zwei Jahren. Patienten mit der Alzheimer-Krankheit verloren auch mehr graue Substanz als Patienten mit BvFTD, aber dies war nur in der rechten Inselrinde statistisch signifikant. Im Vergleich mit Kontrollpersonen verloren Patienten mit der AlzheimerKrankheit im gesamten Gehirn die graue Substanz, vermehrt im hinteren Bereich des Gehirns. Die graue Substanz von BvFTD-Patienten verschwand vor allem in den frontalen Bereichen des Gehirns. Der kognitive Rückgang passte sich dem VolumenVerlust der grauen Substanz an. Obwohl es nur statistisch signifikant war im Vergleich mit gesunden Kontrollpersonen, verschlechterten sich Patienten mit der AlzheimerKrankheit am meisten in den kognitiven Messungen. BvFTD Patienten büßten vor allem im Bereich der Aufmerksamkeit ein. In Kapitel 3.4 betrachteten wir den diagnostischen Wert der grauen Substanz, der subkortikalen Kerne und der weißen Substanz. Zu diesem Zweck verwendeten wir die Programme VBM, FIRST und TBSS. Außerdem untersuchten wir, welche Messmethode die beste Unterscheidung zwischen Patienten mit der AlzheimerKrankheit, BvFTD und gesunden Kontrollpersonen machte, oder ob die Kombination der drei Methoden das beste Ergebnis bei der Unterscheidung dieser drei Gruppen erzielte. Wir fanden Unterschiede in der grauen Substanz, den subkortikalen grauen Kernen und auch in der weißen Substanz, die unabhängig voneinander zur Unterscheidung zwischen den zwei Formen der Demenz und gesunden Kontrollgruppen beitrugen. Obwohl Patienten mit BvFTD die gleiche Krankheitsdauer wie Patienten mit der Alzheimer-Krankheit hatten, waren BvFTD-Patienten am meisten betroffen in allen drei Messungen. Vor allem die weiße Substanz war deutlich mehr in den frontalen Gehirnbereichen bei BvFTD beschädigt. Die Kombination aus allen drei Messmethoden konnte mit 86-91 % Genauigkeit, die Patienten mit der Alzheimer-Krankheit von den anderen beiden Gruppen unterschieden. Messungen von der grauen Substanz spielten hier die größte Rolle. Bei Patienten mit BvFTD lieferte die Beschädigung der weißen Substanz den größten Beitrag bei der Unterscheidung von den anderen zwei Gruppen. Die Ergebnisse legen nahe, dass die Messung der weißen Substanz mit DTI wertvolle Zusatzinformationen bei der Unterscheidung von Patienten mit Alzheimer-Krankheit und BvFTD liefert. Zusammenfassend können wir sagen, dass die Muster der grauen Substanz und der subkortikalen grauen Kerne verschieden sind zwischen Kontrollpersonen, Patienten mit der Alzheimer-Krankheit und BvFTD. Diese Unterschiede sind jedoch am Anfang der Krankheit noch so klein, dass sie nur mit Messmethoden auf Voxel-Ebene erkannt werden können. Bei Patienten, die bereits eine deutliche Atrophie der grauen Substanz vorweisen, ist die Verwendung von visuellen Beurteilungsskalas zu empfehlen, die auf zuverlässige und systematische Weise die Menge der Atrophie feststellen. Allerdings muss man sich dessen bewusst sein, dass nicht jeder Patient ein stereotypisches Atrophie-Muster aufweist, das auch zu den klinischen Symptomen passt. 195

Die Messung der weißen Substanz spielt eine wichtige Rolle in der Unterscheidung zwischen Alzheimer und BvFTD und gibt uns tiefere Einblicke in die zugrunde liegenden pathologischen Prozesse, die bei diesen Krankheiten eine Rolle spielen. Der große Vorteil ist, dass Anomalien in der weißen Substanz, gemessen mit DTI, früh im Krankheitsverlauf festgestellt werden und damit den Startpunkt einer Krankheit sichtbar machen. Dies vereinfacht eine frühe Unterscheidung von BvFTD und der Alzheimer-Krankheit. Aber auch hier geht es um Beschädigungen, die für das Auge nicht sichtbar sind. Darüber hinaus ist der DTI-scan noch kein Bestandteil der MRTStandardprotokolle in den meisten Krankenhäusern. Auch die Akquisition und Analyse der DTI-Bilder erfordert viel Sachverstand und Präzision. Zusammen haben die graue Substanz, die subkortikalen grauen Kerne und die weiße Substanz das Potenzial, ein diagnostisches Instrument in der Unterscheidung von Patienten mit der Alzheimer-Krankheit und BvFTD zu bilden. Die Ergebnisse sind jedoch derzeit nur auf Gruppen von Patienten anwendbar. Die Anwendbarkeit für den einzelnen Patienten steckt noch in den Kinderschuhen, aber die ersten viel versprechenden Schritte wurden mit den SVMs, die gute Ergebnisse in der Unterscheidung zwischen beiden Gruppen zeigten, gemacht. Für die Zukunft ist es wichtig, MRT-Auswertungsmethoden zu entwickeln, die in der täglichen Praxis und für den einzelnen Patienten angewendet werden können. Dies ermöglicht, dass der diagnostische Prozess beschleunigt und verbessert werden kann. Um dies zu erreichen müssen wir a) die Muster von MRT-Anomalien für alle Subtypen der Alzheimer-Krankheit und BvFTD auf der Grundlage der grauen und der weißen Substanz genau erforschen, b) SVMs mehr nutzen und verbessern, um schließlich in der Lage zu sein, diese Methode für den individuellen Patient anzuwenden, c) in großen Patienten-Kohorten mit wiederholten MRT-scans die gefundenen Muster von Gehirnveränderungen identifizieren und überprüfen.

196

List of publications Möller C, Vrenken H, Jiskoot L, Versteeg A, Barkhof F, Scheltens P, van der Flier, WM. Different patterns of gray matter atrophy in early- and late-onset Alzheimer’s disease. Neurobiology of Aging 2013 Aug;34(8):2014-22. Tijms BM, Möller C, Vrenken H, Wink AM, de Haan W, van der Flier WM, Stam CJ, Scheltens P, Barkhof F. Single-subject grey matter graphs in Alzheimer's disease. PLoS One. 2013;8(3):e58921. Epub 2013 Mar 11. Tijms BM, Yeung HM, Sikkes SA, Möller C, Smits LL, Stam CJ, Scheltens P, van der Flier WM, Barkhof F. Single-subject gray matter graph properties and their relationship with cognitive impairment in early- and late-onset Alzheimer's disease. Brain Connect. 2014 Jun;4(5):337-46. Krudop WA, Kerssens CJ, Dols A, Prins ND, Möller C, Schouws S, Barkhof F, van Berckel BNM, Teunissen CE, van der Flier WM, Scheltens P, Stek ML, Pijnenburg YA. Building a new paradigm for the early recognition of behavioral variant frontotemporal dementia: Late Onset Frontal Lobe Syndrome study. Am J Geriatr Psychiatry. 2014 Jul;22(7):735-40. Möller C, van der Flier WM, Versteeg A, Benedictus MR, Wattjes MP, Koedam EL, Scheltens P, Barkhof F, Vrenken H. Quantitative regional validation of the visual rating scale for posterior cortical atrophy. Eur Radiol. 2014, 24 (2), 397-404. Binnewijzend MA, Kuijer JP, van der Flier WM, Benedictus MR, Möller C, Pijnenburg YA, Lemstra AW, Prins ND, Wattjes MP, van Berckel BN, Scheltens P, Barkhof F. Distinct perfusion patterns in Alzheimer's disease, frontotemporal dementia and dementia with Lewy bodies. European Radiology. 2014 Sep;24(9):2326-33. Möller C, Dieleman N, van der Flier WM, Versteeg A, Pijnenburg YAL, Scheltens P, Barkhof F, Vrenken H. More atrophy of deep gray matter structures in behavioral variant Frontotemporal Dementia compared to Alzheimer’s Disease. J Alzheimer’s Dis. 2015 Jan 1;44(2):635-47. Möller C, Hafkemeijer A, Pijnenburg YAL, Rombouts SARB, van der Grond J, Dopper E, van Swieten J, Versteeg A, Pouwels PJW, Barkhof F, ScheltensP, Vrenken H, van der Flier WM. Joint assessment of white matter integrity, cortical and subcortical atrophy to distinguish AD from behavioral variant FTD: a multi-center study. Major revisions NeuroImage – Clinical. Möller C, Pijnenburg YAL, van der Flier WM, Versteeg A, Tijms B, de Munck JC, Hafkemeijer A, Rombouts SARB, van der Grond J, van Swieten J, Dopper E, Scheltens P, Barkhof F, Vrenken H, Wink AM. Automatic classification of AD and bvFTD based on cortical atrophy for single-subject diagnosis. Under review Radiology 197

Yi HA, Möller C, Dieleman N, Bouwman FH, Barkhof F, Scheltens P, van der Flier WM, Vrenken H. Relation between subcortical gray matter atrophy and conversion from mild cognitive impairment to Alzheimer’s disease. Resubmission after major revisions Journal of Neurology, Neurosurgery, Psychiatry. Hafkemeijer A, Möller C, Dopper E, Jiskoot L, Schouten T, van Swieten J, van der Flier WM, Vrenken H, Barkhof F, Scheltens P, van der Grond J, Rombouts SARB. Resting state functional connectivity in frontotemporal dementia and Alzheimer's disease. Submitted Alzheimer's & Dementia: Diagnosis, Assessment & Disease Monitoring Krudop WA, Kerssens CJ, Dols A, Prins ND, Möller C, Schouws S, van der Flier WM, Scheltens P, Sikkes S, Stek ML, Pijnenburg YA. Identifying bvFTD within the wide spectrum of the late onset frontal lobe syndrome. A clinical approach. Major revisions American Journal of Geriatric Psychiatry Ossenkoppele R, Cohn-Sheehy BI, La Joie R, Vogel JW, Möller C, Lehmann M, van Berckel BNM, Seeley WW, Pijnenburg YA, Gorno-Tempini ML, Kramer JH, Barkhof F, Rosen HJ, van der Flier WM, Jagust WJ, Miller BL, Scheltens P, Rabinovici GD. Atrophy Patterns in Early Clinical Stages Across Distinct Phenotypes of Alzheimer's Disease. Submitted Brain Möller C, Hafkemeijer A, Pijnenburg YAL, Rombouts SARB, van der Grond J, Dopper E, van Swieten J, Versteeg A, Steenwijk M, Barkhof F, Scheltens P, Vrenken H, van der Flier WM. Different patterns of gray matter loss in behavioral variant FTD and AD. Submitted

198

Hall of fame: List of theses Alzheimercenter VUmc 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30.

L. Gootjes: Dichotic Listening, hemispheral connectivity and dementia (1409-2004) K. van Dijk: Peripheral Nerve Stimulation in Alzheimer’s Disease (16-01-2005) R. Goekoop: Functional MRI of cholinergic transmission (16-01-2006) R. Lazeron: Cognitive aspects in Multiple Sclerosis (03-07- 2006) N.S.M. Schoonenboom: CSF markers in Dementia (10-11-2006) E.S.C. Korf: Medial Temporal Lobe atrophy on MRI: risk factors and predictive value (22-11-2006) B. van Harten: Aspects of subcortical vascular ischemic disease (22-12-2006) B. Jones: Cingular cortex networks: role in learning and memory and Alzheimer’s disease related changes (23-03-2007) L. van de Pol: Hippocampal atrophy from aging to dementia: a clinical and radiological perspective (11-05-2007) Y.A.L. Pijnenburg: Frontotemporal dementia: towards an earlier diagnosis (05-07-2007) A. Bastos Leite: Pathological ageing of the Brain (16-11-2007) E.C.W. van Straaten: Vascular dementia (11-01-2008) R.L.C. Vogels: Cognitive impairment in heart failure (11-04-2008) J. Damoiseaux: The brain at rest (20-05-2008) G.B. Karas: computational neuro-anatomy (19-06-2008) F.H. Bouwman: Biomarkers in dementia: longitudinal aspects (20-06-2008) A.A. Gouw: Cerebral small vessel disease on MRI: clinical impact and underlying pathology (20-03-2009) H. van der Roest: Care needs in dementia and interactive digital information provisioning (12-10-2009) C. Mulder: CSF Biomarkers in Alzheimer’s disease (11-11-2009) W. Henneman. Advances in hippocampal atrophy measurement in dementia: beyond diagnostics (27-11-2009) S.S. Staekenborg: From normal aging to dementia: risk factors and clinical findings in relation to vascular changes on brain MRI (23-12-2009) N. Tolboom: Imaging Alzheimer’s disease pathology in vivo: towards an early diagnosis (12-02-2010) E. Altena: Mapping insomnia: brain structure, function and sleep intervention (17-03-2010) N.A. Verwey: Biochemical markers in dementia: from mice to men. A translational approach (15-04-2010) M.I. Kester: Biomarkers for Alzheimer’s pathology; Monitoring, predicting and understanding the disease (14-01-2011) J.D. Sluimer: longitudinal changes in the brain (28-04-2011) S.D Mulder: Amyloid associated proteins in Alzheimer’s Disease (07-10-2011) S.A.M. Sikkes: measuring IADL in dementia (14-10-2011) A. Schuitemaker: Inflammation in Alzheimer’s Disease: in vivo quantification (27-01-2012) 199

31. K. Joling: Depression and anxiety in family caregivers of persons with dementia (02-04-2012) 32. W. de Haan: In a network state of mind (02-11-2012) (Cum Laude) 33. D. van Assema: Blood-brain barrier P-glycoprotein function in ageing and Alzheimer’s disease (07-12-2012) 34. J.D.C. Goos: Cerebral microbleeds: connecting the dots (06-02-2013) 35. R. Ossenkoppele: Alzheimer PEThology (08-05-2013) 36. H.M. Jochemsen: Brain under pressure: influences of blood pressure and angiotensin-converting enzyme on the brain (04-10-2013) 37. A.E. van der Vlies: Cognitive profiles in Alzheimer’s disease: Recognizing its many faces (27-11-2013) 38. I. van Rossum: Diagnosis and prognosis of Alzheimer’s disease in subjects with mild cognitive impairment (28-11-2013) 39. E.I.S. Møst: Circadian rhythm deterioration in early Alzheimer’s disease and the preventative effect of light (03-12-2013) 40. M.A.A. Binnewijzend: Functional and perfusion MRI in dementia (21-03-2014) 41. H. de Waal: Understanding heterogeneity in Alzheimer’s disease: A neurophysiological perspective (25-04-2014) 42. W. Jongbloed: Neurodegeneration: Biochemical signals from the brain (0805-2014) 43. E.L.G.E. Poortvliet-Koedam: Early-onset dementia: Unraveling the clinical phenotypes (28-05-2014) 44. A.M. Hooghiemstra: Early-onset dementia: With exercise in mind (03-122014) 45. L.L. Sandberg-Smits: A cognitive perspective on clinical manifestations of Alzheimer’s disease (20-03-2015) 46. F.H. Duits: Biomarkers for Alzheimer’s disease, current practice and new perspectives (01-04-2015) 47. S.M. Adriaanse: Integrating functional and molecular imaging in Alzheimer’s disease (07-04-2015) 48. C. Möller: Imaging patterns of tissue destruction – Towards a better discrimination of types of dementia (01-05-2015)

200

Dankwoord Het schijnt dat dit het meest gelezen onderdeel van een proefschrift is, daarom voel ik de druk om hier niemand te kort te doen. Ik ben blij om dit onderdeel van mijn proefschrift, want in mijn beleving zeggen we nog te weinig ‘Dank je wel’ tegen elkaar en zonder de hulp en inzet van velen was dit proefschrift nooit tot stand gekomen. Dus hier gaan we. Allereerst wil ik alle patiënten, mantelzorgers en gezonde proefpersonen vanwege hun inzet bedanken. Zonder u was er geen data geweest en mijn dank is groot dat iedereen drie jaar lang bereid was opnieuw in die verschrikkelijk lawaaierige MRI apparaat te gaan liggen en urenlang neuropsychologische testen te ondergaan. Bij deze gelegenheid wil ik ook mijn arts collega’s bedanken, die hun vrije zondag voor me hebben opgeofferd om gezonde proefpersonen te zien. Beste prof. dr. Scheltens, beste Philip, hoe doe je dit toch allemaal? Onderzoek, patiënten, directeur van het Alzheimercentrum, en ondertussen komt niemand om jouw naam heen als je aan Alzheimer denkt. Ik ben heel dankbaar, dat je me toen hebt aangenomen en ik het voorrecht had om in één van de beste onderzoekscentra te mogen leren hoe je onderzoek doet. En wat heb je allemaal mogelijk gemaakt voor mij: Ajax kijken in de VIP lounge, Youp van ‘t Hek in Carré, tv opnames met BNers, op congresbezoek in de hele wereld, ik mocht Princess Beatrix, de koning en de koningin ontmoeten. You name it! Bedankt voor alles! Beste prof. dr. Barkhof, beste Frederik, bedankt voor jouw ongelofelijk grote kennis op het gebied van de Radiologie. Dat ik verantwoordelijk mocht zijn voor de MRI scans tijdens het MDO vond ik niet alleen geweldig leuk maar ook een grote eer. Ik heb veel van je geleerd, over verschillende radiologische afwijkingen en de anatomie van de hersenen. Beste prof. dr. van der Flier, beste Wiesje, bedankt voor jouw begeleiding tijdens mijn promotietraject. Hoewel je erg druk bent, kon ik altijd op je rekenen, reageerde je altijd meteen op vragen en kreeg ik altijd op tijd commentaar over mijn stukken. Zonder jouw structuur en jouw vermogen de hoofdboodschap helder te krijgen, was dit proefschrift er niet gekomen. Beste dr. Vrenken, beste Hugo. Zonder jouw onuitputtelijke kennis over MRI analyse had ik dit nooit gekund. Door jouw blik voor details hebben we vaak de mooie puntjes op de ‘i’ kunnen zetten. Ook voor jou geldt, dat je altijd klaar stond als ik vragen had en er altijd reactie kwam, hoe laat het ook was. Bedankt voor je inzet en begeleiding! De leden van de leescommissie: Sehr geehrter Prof. Teipel, vielen Dank, dass Sie sich die Zeit und Mühe genommen haben um meine Doktorarbeit zu lesen. Es ist mir eine Ehre, dass sie zu meiner Verteidigung nach Amsterdam kommen. Geachte Prof. Durston, geachte Prof. Rombouts, geachte Prof. van Swieten, geachte Dr. Wattjes, geachte Dr. 201

Pijnenburg, bedankt voor het zorgvuldig lezen van mijn proefschrift en uw bereidheid om bij mijn verdediging aanwezig te zijn. Ik kijk erna uit. Collega’s van het Hersenen & Cognitie project uit Leiden: Beste Anne, Jeroen en Serge. Bedankt voor een goede samenwerking voor vier jaar lang. Fijn dat we onze data hebben gedeeld en voor jullie kritische blik op mijn artikelen. Anne, we hebben het gewoon maar lekker gedaan! Dat we maar nog lange coauteurs van elkaars mogen blijven. Succes met de laatste lootjes. En natuurlijk uit Rotterdam: Elise, bedankt voor jouw hulp bij het opstarten van het project. John, wat ongelofelijk leuk dat jij in mijn leescommissie zit! Jouw enthousiasme en jouw grote kennis over FTD werken aanstekelijk en het is indrukwekkend wat jij in FTD land betekend. Ik hoop dat onze wetenschappelijke paden zich in de toekomst nog gaan kruisen. Collega’s van de Radiologie: Beste Martijn en Adriaan, Ik weet niet hoe ik jullie moet bedanken voor alle hulp op het vlak van Linux, SPM, FSL en MRI analyses. Zonder jullie was ik nog steeds met de hand bezig om honderden scans te bewerken en was ik er nooit uitgekomen met de resultaten. Sorry, dat ik soms wat gestrest bij jullie om hulp kwam vragen. Beste Ton, jarenlang heb je me minimaal 1 keer per week aan de telefoon gehad of was ik te gast bij de MRI scanner. Bedankt voor jouw geduld en jouw vriendelijkheid bij het plannen van de scans en voor jouw tijd op de zondagen. Beste Petra, bedankt voor jouw kritische blik en jouw enorme kennis bij de DTI analyses. Alle Meije, het was me een genoegen om met jouw samen een paper te mogen schrijven. Bedankt dat je me hebt geïntroduceerd in de wonderbaarlijke wereld van de support vector machines. Beste Mike, ik was erg blij toen ik hoorde dat jij deel van mijn leescommissie uit ging maken. Naast dat ik je een uitstekende radioloog vind, kon ik jouw betrokkenheid bij mijn projecten en carrière en jouw waardering voor wat ik deed altijd erg waarderen. Dank je wel! En verder wil ik natuurlijk niet vergeten: Menno, Hanneke, Oliver, Milou, Veronica, Ronald, Joost en Regina. Collega’s van het Alzheimercentrum: Bedankt aan al mijn bunkergenoten in de loop van de jaren. Uit bunker 2 Dr. Goos, Dr. Rikki, Danielle, Maja, Lieke (Kopenhagen rules), Eva, Zwannie Mannie, Saskia, Annebet en Rosalinde. Ik vond het altijd ontzettend leuk met jullie. En in de seniorenbunker naast de dodekast: Eddy, Els, Ellen, Welmoed, Daniela en Sofie. Aan jullie moet ik eigenlijk mijn excuses aanbieden, want nadat mijn proefschrift af was, heb ik toch een eind weggekletst en jullie van het werk gehouden. Gelukkig kwam ik altijd pas laat binnen en er kon wat werk verzet worden in de vroege uurtjes. Verder wil ik iedereen van de oude garde bedanken voor feestjes, uitjes, gesprekken, advies, roddels en de goede samenwerking: Hanneke (bedankt voor alles wat jij en Jelle voor mij hebben gedaan! Jullie zijn geweldig), Argonde, Ineke, Willem, Sietske, Betty, Astrid, Marije, Floor, Marjolein (bedankt dat je foto’s wilt maken), Aafke, Eveline, Femke, Niels, Pieter-Jelle, Annelies, Freek, Tanja, Karin, Anita, Nicole, Elizabeth en Karlijn. En uiteraard de twee dames van het secretariaat (pardon projectbureau): Elisa en Ilse. Jullie zijn toppers! Het LOF team: Welmoed, Annemiek, Flora, en Cora. Bedankt voor de goede samenwerking met een heel bijzondere patiëntenpopulatie. Ik 202

heb veel van jullie geleerd en een top tijd met jullie in Vancouver gehad. Welmoed, bedankt voor de toffe tijd in de VS. Beste Yolande, ik weet eigenlijk niet zo goed waar ik moet beginnen: Jouw enthousiasme met patiënten en in de wetenschap en jouw betrokkenheid bij mijn project, hebben mijn promotietijd veel, veel leuker gemaakt. Ik zal nooit vergeten hoe we ‘ stiekem’ bij de halve marathon van Vancouver over de finish zijn gerend. Jammer dat het voor de nabije toekomst met onze samenwerking niet gaat lukken maar ik hoop dat wij elkaar niet uit het oog gaan verliezen. Een ook bedankt aan de nieuwe garde. Het zijn er gewoon te veel om hier op te noemen maar ik vind jullie allemaal erg fijne collega’s en ben blij om te zien, dat het Alzheimercentrum verder gaat met zo’n enthousiaste groep onderzoekers. Geachte Prof. Dr. Scherder, beste Erik, van 2007 tot 2010 was ik student bij jou. Ik heb nog nooit iemand gezien, die zo betrokken is bij zijn studenten als jij het bent. Je bent een inspirerende professor met onuitputtelijke energie en ik zal nooit vergeten hoe betrokken je was toen het niet goed met me ging. Jouw enthousiasme en jouw geloof in mij als wetenschapper maken dat ik er vandaag nu sta. Mijn dank is groot. Geachte Prof. Dr. Dröes, lieve Rose-Marie. In 2008 kwam ik bij jou aankloppen voor een studentenbaan bij de helpdesk ontmoetingscentra. Ik ben ontzettend blij dat je me de kans gaf, want anders zou onze vriendschap niet bestaan en was ik uiteindelijk waarschijnlijk niet bij het Alzheimercentrum terechtgekomen. Ik wil jou bedanken voor jouw emotionele steun, een luisterend oor en jouw geloof in mij. Naast het werk bestaat er gelukkig ook nog een ander leven en zonder mijn lieve vrienden was mijn leven zo veel minder leuk. Allereerst natuurlijk bedankt aan mijn paranimfen Celeste, Lotte en Cornelie. Jullie zijn geweldige vriendinnen en jullie gaan het op 1 mei prachtig doen. De jaren met jullie zijn voorbij gevlogen. Ik hoop dat er nog heel veel gaan volgen. Corrie en Lotte, jullie delen de paranimfen-plek omdat Postimus zo nodig de aardbol moest omzeilen maar ik kon me geen betere vervanger voorstellen. Marijke en Sana, de drie musketiers, blond-rood-zwart, huisgenoten, praatpaal, gaypride, Twente, Duitsland, Marokko. Dank voor een geweldige vriendschap. Sanne, bloed, zweet en tranen en liefde op het eerste spinningfiets. We zijn het beste voorbeeld dat sport verbindt. Ook al sporten we niet meer samen, de vriendschap blijft en daar ben ik ontzettend blij mee. Sofie, collega, hardloopmaatje, vriendin, vakantieganger en spiegelbeeld. Wie had gedacht dat twee zo verschillende personen zo goed met elkaar kunnen opschieten. Bedankt dat je me af en toe erop wijst, hoe rete irritant ik kan zijn en bedankt dat je me zo accepteert hoe ik ben. Lieve Eva, wat ben jij een stoer wijf zeg! Je studeert Neuropsychologie, Kinder & Jeugdpsychologie, volgt een opleiding tot fotografe en inmiddels heb je een dikke baan bij de KPN. Jij bent het beste voorbeeld, dat out of your comfort zone het beste in je naar voren kan halen. Mag ik iets van jouw moed en courage? Mariëlle, nog zo’n kanjer in mijn vriendenkring, die gewoon doet, wat haar hart haar zegt. Ik kijk naar je op. Lieve Ndedi, levenskunstenaar en stapmaatje, fijn om jou als vriendin te hebben. 203

De boys Thijs, Guido, Ramon en Ferry. Het was in een zwoele zomernacht in 2012 tijdens de groepsfase van het EK voetbal. Wat ben ik blij dat bier en burgers Duitsers en Nederlanders toch bij elkaar kan brengen en dat nog tijdens een voetbalwedstrijd. En ik ben nog blijer dat we er na drie jaar nog steeds bier en burgers kunnen delen, samen feesten, helpen met verhuizen en ijzersterke hardloopprestaties kunnen neerzetten. Lieve Marcel, jij hebt het begin van mijn promotietraject van dichtbij meegemaakt. Ik wil je bedanken voor de fijne jaren en jouw steun en toeverlaat. Onze wegen hebben zich gescheiden maar ik ben blij dat we nog steeds contact met elkaar hebben. Geert en Linde, Gabi en Lotti. Ik ben erg blij dat onze vriendschap nog steeds bestaat en dat jullie mij dat gevoel geven dat dit ook altijd zo blijft. Lieve JeeWee, mijn lieve huisgenoot uit Huize Riz in Enschede. Bijna tegelijkertijd gingen we naar de grote stad en we zien elkaar nog steeds. Ik heb altijd het gevoel dat we elkaar precies snappen zonder veel te moeten zeggen. Dat is ontzettend fijn. Lieve Guido, ik kan me geen betere wederhelft voor JeeWee voorstellen. Meine Freunde aus Deutschland: Antje, Tamara, und Jens. In 2003 haben wir zusammen Abi gemacht in der Weltmetropole Höhr-Grenzhausen und ob es Frankfurt, Mainz, Shanghai oder Amsterdam ist, wir verlieren uns hoffentlich nicht aus den Augen. Liebe Eva (liebster Franz), seit ich in Holland wohne, verbreitet das Beachvolleyball Team Möller-Kamenz leider keine Angst und Schrecken mehr auf den Feldern von Deutschland, aber zum Glück schaffen wir das noch manchmal bei euch auf dem Feld hinterm Haus. Freya und Katja, studieren-in-holland.de hat uns in Enschede zusammengeführt und das ist jetzt schon wieder 10 Jahre her. Schön, dass wir immer noch Kontakt haben. Liebes Bruderherz, lieber Olli. Unsere Beziehung ist nie ganz einfach gewesen, aber ich bin froh, dass wir uns mittlerweile so gut verstehen. Margareta, Augustin und mein Patenkind Hilda sind eine große Bereicherung für unsere Familie und ich bin stolz wie du das alles hinbekommst und Familie und Arbeit so gut unter einen Hut bringst. Auch bin ich die kleine, besserwissrige Schwester, ich hab mehr von dir gelernt, als du denkst: Ohne mein Durchsetzungsvermögen, was ich bei meinem großen Bruder an den Tag legen musste, stände ich nun nicht hier. Danke. Liebe Mama, lieber Papa. Es ist jetzt fast 10 Jahre her, dass ich in das Land der Käsköpfe ausgewandert bin. Ich bin so froh, dass ihr mich ohne Wenn und Aber habt ziehen lassen in das Land der Wohnmobile und des wässrigen Biers, und dass ihr mich immer unterstützt habt: War es mich in einer Nacht und Nebelaktion wieder nach Hause zu holen oder wenn ich mal knapp bei Kasse war. Danke für euer Vertrauen und ich bin froh, dass ich mittlerweile weiß, dass ihr stolz auf mich seid. Ich hab euch lieb. En tenslotte, lieve Stach, ik heb je lief!

204

About the author Christiane Möller was born in Mainz, Germany, on March the 5th 1984. She grow up in Vallendar and graduated in 2003 at the Gymnasium im Kannenbäckerland. From 2003 to 2005 she finished an apprenticeship at an advertising agency as an advertising consultant. Because of her interest in Psychology she decided to move to the Netherlands. In 2005 Christiane started studying Psychology at the University of Twente, Enschede. Because of a fascination with the brain and the biological reasons behind behavioral disturbances, she continued studying Neuropsychology at the Free University in Amsterdam, where she acquired her Master of Science in Clinical Neuropsychology in 2010. Because of her student work at the Valeriuskliniek and an enthusiastic professor at the VU University her interest in dementia was stimulated. In 2010 she started her PhD project “Dementia – structural markers for cognitive impairment” embedded in the Dutch National project “Brain and Cognition” (NWO) at the Alzheimer center at the VU University medical center in Amsterdam under supervision of Prof. Dr. Philip Scheltens, Prof. Dr. Wiesje van der Flier, Prof. Dr. Frederik Barkhof and Dr. Hugo Vrenken. At time of writing, Christiane is finishing her work as a PhD student and will explore Asia after her defense.

205

“...you'll see that things will turn out like they do, because that is what usually happens - almost always, in fact� Jonas Jonasson The Hundred-Year-Old Man Who Climbed Out of the Window and Disappeared 206