Sander Verfaillie

The aim of the current thesis was to investigate the earliest neurobiological changes related to Alzheimer’s Disease (AD) using amyloid positron emission tomography (PET), structural and functional magnetic resonance imaging (MRI) in patients with Subjective Cognitive Decline (SCD). More specifically, this thesis had three aims: 1) to provide an overview of the Subjective Cognitive Impairment Cohort (SCIENCe), 2) to investigate the predictive value of structural MRI for cognitive decline and clinical progression in individuals with subjective cognitive decline (SCD), 3) to investigate the most optimal kinetic model for the quantification of the amyloid tracer [18F]florbetapir, and to investigate associations between amyloid and resting-state fMRI and patterns of cognitive complaints, cognition and spontaneous speech.

Key findings include:
1) The SCIENCe study confirms that SCD is a heterogeneous group, including individuals with preclinical AD and subthreshold psychiatric features, as well as a relatively large group with complaints of undetermined origin.
2) Thinner temporal and parietal cortex in SCD is associated with increased risk of future clinical progression to dementia.
3) In SCD, thinner cortex, especially in the temporal cortex, is associated with a steeper decline of memory.
4) Less efficient grey matter network organization in SCD is associated with a steeper decline of language and global cognition.
5) SCD is associated with a pattern of brain connectivity that mirrors early AD-related connectivity failure.
6) [18F]florbetapir binding can be robustly quantified by a reversible two tissue compartmental model.
7) Increased amyloid load-β in SCD is associated with worries but not with severity or specific pattern of cognitive complaints.
8) SCD individuals with abnormal amyloid accumulation use fewer specific words in spontaneous speech.

Initial baseline findings of the SCIENCe cohort confirmed that SCD is a heterogeneous group, with preclinical AD (25%) and subthreshold psychiatric features (38%) as well as a relatively large group with complaints of undetermined origin (43%) (chapter 2). In a retrospective study (chapter 3) investigating cortical thickness in relation to clinical progression we found that 16% of individuals with SCD showed clinical progression to mild cognitive impairment (MCI (n=32)), Alzheimer’s Disease (AD (n=9)) or non-AD dementia (n=8) after a follow-up of 3 years. More specifically, lower hippocampal volumes, thinner cortex of the AD-signature regions and various temporal and parietal cortices were associated with an approximate fivefold increased risk of clinical progression to MCI and AD. Subsequently (chapter 4), we aimed to investigate associations between regional cortical thickness and rate of decline over time in four cognitive domains. We found that thinner temporal cortex was related to memory decline over time, which seemed particularly driven by SCD patients who developed MCI and AD on follow-up. In addition (chapter 5), we found that lower grey matter network values (normalized clustering and path length [gamma and lambda]), indicative of a more randomly organized network, were associated with steeper decline in global cognition and language. Furthermore, in a sample of cognitively normal individuals with a family history of AD (chapter 6) we found that SCD was related to altered default mode network (DMN) connectivity in brain regions vulnerable for AD such as the posterior DMN and medial temporal memory system. In addition, higher connectivity between the MTMS and rest of the brain was associated with better baseline immediate memory, attention and global cognition, whereas higher MTMS and pDMN-MTMS connectivity were associated with lower immediate memory over time. In the third part of this thesis we investigated amyloid accumulation. Firstly (chapter 7), we aimed to identify the optimal model for quantifying [18F]florbetapir uptake and to assess test-retest (TRT) reliability of corresponding outcome measures in patients with AD and controls. We found that [18F]florbetapir binding can be robustly quantified by a reversible two tissue compartmental model, independent of brain regions of interest or diagnostic label. In a follow-up study on individuals with SCD (chapter 8) we aimed to investigate whether amyloid pathology is associated with linguistic parameters derived from spontaneous speech. We found that abnormal amyloid accumulation was associated with fewer specific words, but not with lexical diversity or syntactic complexity or conventional neuropsychological language tests. In the last chapter (chapter 9) we aimed to investigate whether amyloid-β load is associated with a specific pattern of cognitive complaints. We found that increased amyloid-β load in all cortical regions was associated with presence of worries, but not with any particular pattern of cognitive complaints or with the severity of complaints, as measured using CCI (informant and self-reported, total scores or individual items) or SCF.

In the following paragraphs these findings will be integrated and placed in the context of the existing literature, followed by a discussion of methodological issues relevant for this thesis. Final paragraphs consist of recommendations for future research and concluding remarks.

Subjective cognitive decline
While SCD portends an increased risk of dementia, a minority (10-15%) of individuals with SCD showed clinical progression to mild cognitive impairment (MCI) or AD dementia in approximately 2-4 years. A vital question remains which individuals with SCD are at increased risk of clinical progression. One difficulty lies in the observation that self-perceived cognitive decline can be caused by a myriad of factors such as preclinical AD, normal aging, depressive symptoms and neuroticism. It is currently difficult to clinically identify preclinical AD in cognitively healthy individuals experiencing memory complaints. In order to investigate the earliest signs related to AD we initiated the Subjective Cognitive Impairment Cohort (SCIENCe) in 2014.

Our first aim was to characterize patients with SCD who visited our memory clinic and we showed that SCD is a heterogeneous label, with preclinical AD (i.e. abnormal amyloid accumulation, 25%) and subthreshold psychiatric features (38%). The largest group of individuals with SCD (43%) had neither evidence of amyloid, nor subthreshold psychiatric symptoms (chapter 2). Individuals with preclinical AD were on average older, and had more frequently a family history of dementia, and were more frequently APOE e4 carrier. APOE e4 genotype and age≥60 are part of SCD plus criteria, which have been proposed to facilitate and harmonize SCD research. We now provide evidence that these factors are associated with increased risk of preclinical AD in individuals with SCD. In chapter 2, we also measured the degree of cognitive complaints, and almost all participants reported cognitive decline compared to five years ago, which seems substantially higher than in the general population, but comparable to another memory clinic study, and could be an explanation for these individuals to actively seek medical evaluation.

In cognitively normal individuals with SCD, biomarkers of AD can already be abnormal, such as increased amyloid deposition on PET or brain atrophy on MRI. However, the sequence of neurodegenerative changes eventually leading to AD may vary amongst individuals, and where to position SCD in these pathological sequences remains to be elucidated. For this reason, it is crucial which (neuroimaging) biomarkers are associated with clinical progression and earliest cognitive changes related to AD.

MRI biomarkers associated with subsequent cognitive decline
In the second part of this thesis we investigated associations between baseline MRI measures and incident clinical progression and longitudinal cognitive functioning. So far, memory clinic studies have demonstrated decreased gray matter volumes and cortical thinning in medial temporal regions in SCD compared to healthy controls. It has been suggested that regional thinning of the cortical mantle, coined as the AD-signature (consisting of frontal, temporal and parietal cortices), can be predictive for dementia in cognitively normal individuals until one decade before clinical progression. Central in our studies was further investigation of whether such brain variations were related to clinical progression and cognitive functioning over time.

In chapter 3, we investigated if thinner cortex of the AD-signature region was predictive for clinical progression to MCI or dementia. After a follow-up of approximately 3 years, 49 (16%) individuals with SCD showed clinical progression to MCI (n=32), AD (n=9) or non-AD dementia (n=8). These progression rates are comparable to earlier studies based on data from the Amsterdam Dementia Cohort. We furthermore found that thinner cortex of the AD-signature and mean cortical thickness were both associated with a fivefold increased risk of clinical progression to MCI and dementia (figure 1A). Of these AD-signature subcomponents, temporal and parietal cortices were mostly associated clinical progression. More specifically, the medial temporal cortex showed the strongest predictive value for clinical progression to MCI and dementia, and hippocampal volume did not add any predictive value over this effect. Interestingly, we observed that thinner cortex of the AD-signature was associated with progression to AD dementia as well as non-AD dementia, but not to MCI. This suggests that the predictive value seemed to be non-specific for progression to AD or non-AD, and that AD-signature is likely sensitive for a more general neurodegenerative process.

In two follow-up papers (chapter 4 & 5), including individuals with SCD with repeated neuropsychological assessment, we sought to investigate whether thinner cortical lobar thickness is associated with rate of decline over time in the cognitive domains of memory, attention, language and executive functioning. Although individuals with SCD typically complain about their memory, early cognitive decline may also affect other cognitive domains. In chapter 4, we showed that language functioning decreased over time across the entire group of individuals with SCD, which illustrates that cognitive complaints could also originate from poorer language performance. In chapter 7, we also show that amyloid was related to lower use of specific words during spontaneous speech, whilst conventional neuropsychological tests were still unimpaired (figure 1D). Two recent cross-sectional studies on cognitively intact and mildly cognitively impaired elderly demonstrated that thinner temporal, precuneus and occipital cortex were associated with impaired episodic memory. We found that widespread cortical thinning in SCD patients was associated with faster subsequent decline in memory over time, but not related to concurrent cognitive function.

Global brain atrophy is common in AD, but also occurs with normal aging and both can evoke cognitive complaints. Others demonstrated that over a period of 10 years, those who remained cognitively intact (i.e. normal aging), showed brain volume decreases in the frontal lobes and superior parietal regions, whereas those who progressed to MCI, showed accelerated volume loss in temporal regions. Our findings also point towards involvement of temporal, frontal and occipital cortical regions (figure 1A) in relation to memory decline (chapter 4) and clinical progression (chapter 3). Thinner temporal cortex could be an early reflection of AD, especially since some of the individuals with SCD finally progressed to MCI or AD dementia during the course of the study.

Evidence is accumulating that brain changes leading to cognitive decline and dementia are not restricted to specific regions such as the medial temporal lobe, but rather include widespread changes in structure, function and organization of the brain. Brain morphology can be considered as a large-scale network consisting of multiple small regions of grey matter (i.e. nodes) that work in harmony. In chapter 5, we aimed to investigate whether grey matter networks are related to decline in specific cognitive functioning over time. Our main finding (figure 1B) was that SCD subjects with more randomly organized grey matter networks show steeper decline in language and global cognition over time (chapter 5). We found associations with network parameters that are often reported in AD literature, i.e. normalized clustering and path length. In addition, we found that lower path length values in precuneus and fronto-occipital brain regions were associated with global cognitive decline. Our findings suggest that at very early, preclinical stages, less efficient network organization may prelude subsequent cognitive impairment. There was a relatively large overlap with respect to neuropsychological and structural MRI data between chapter 4 and 5. In both chapters we found associations between cortical thickness and grey matter network organization in relation to memory and language decline.

After correction for multiple comparisons, however, associations between cortical thickness and language, and grey matter network organization and memory did not survive more stringent thresholds. While the precise implications and biological relevance of grey matter networks are not yet fully understood, these results provide evidence that network parameters provide additional information over atrophy measures.

In chapter 6 we took a different approach. In this chapter we sought to investigate whether SCD was associated with functional brain connectivity. In this chapter we investigated SCD in pre-symptomatic individuals with a parental or multiple-sibling family history of AD dementia from the Pre-symptomatic Evaluation of Novel or Experimental Treatments for Alzheimer’s Disease (PREVENT-AD) program (Montreal, Quebec, Canada). The primary goal of this program is to test whether serial determination of multi-modal biomarkers of AD may be measured and used in pre-symptomatic persons at high risk of subsequent AD dementia to trace the progression of the disease. Functional brain changes may precede structural abnormalities and clinical symptoms and could therefore serve as a potential early AD biomarker. It has been suggested that brain hyperconnectivity in the posterior default mode network (pDMN) compensates for early pathophysiological processes, but later gives way to global brain hypoconnectivity, perhaps resulting from sustained excitotoxicity. So far studies have shown altered DMN connectivity in SCD patients compared to AD patients and controls, but not yet investigated this in cognitively normal individuals at risk for AD. We found that subjects with SCD had increased connectivity between the pDMN and medial temporal memory system (MTMS) if compared to subjects without SCD (figure 1C). Moreover, we found that hyperconnectivity, particularly between the pDMN and MTMS, was related to better concurrent cognitive performance (e.g. compensatory mechanism), but could have a detrimental effect over time. These findings were independent of cortical atrophy, and demonstrate that functional brain connectivity could be considered as an early disease marker in the future.

Figure 1. Overview of findings. Thinner cortex in memory clinic patients with subjective cognitive decline (SCD) is associated with increased risk of clinical progression, and steeper decline of memory (A). Less efficient grey matter brain network organization is associated with a steeper decline of language and global cognition (B). SCD in cognitively normal individuals with a family history of AD is associated with altered resting-state fMRI connectivity and cognitive functioning (C). Higher amyloid-β levels are associated with fewer specific words during spontaneous speech, and SCD-related worries (D).

Amyloid imaging, spontaneous speech and cognitive complaints
In the third part of this thesis we investigated associations between amyloid and spontaneous speech and cognitive complaints. Firstly, to more accurately quantify amyloid deposition for subsequent research projects in this thesis, we have investigated various kinetic models in relation to [18F]florbetapir binding (chapter 7). A validated tracer kinetic model is important, not only for identification of early (subtle) amyloid accumulation, but also for studying its relation with early cognitive changes related to AD in individuals with SCD. In chapter 7, we found that in vivo kinetics of [18F]florbetapir binding could best be described by a reversible two tissue compartmental model with fitted blood volume parameter. These results are in line with the kinetic model of a widely used first generation amyloid imaging agent ([11C]PiB), but also converge with recent findings on [18F]florbetapir quantification. We extend previous findings by showing that this model produced robust and consistent test-retest results, with comparable findings for AD patients and controls. In addition, we found that simplified reference tissue models can be used to quantify amyloid-β load, with substantially less bias if compared to standardized uptake values.

In two follow-up papers (figure 1D) we investigated associations between amyloid and spontaneous speech (chapter 8) and cognitive complaints (chapter 9) respectively in individuals with SCD from the SCIENCe cohort. Individuals with subjective cognitive decline often experience word-finding problems, but these are difficult to verify with standard neuropsychological language tests and it is unknown whether spontaneous speech deficiencies are related to underlying AD pathology. In chapter 8, we aimed to investigate whether preclinical AD in individuals with SCD predisposes for less use of specific words, lexical diversity or syntactic complexity. AD dementia is characterized by gradual deterioration in various cognitive domains, including language. Lexical diversity and the number of content words in spontaneous speech are impaired in AD. Moreover, a recent case study showed that increased use of conversational fillers and decreased number of content words can be observed years before onset of dementia. We found that higher amyloid-β load was associated with decreased use of content and concrete words, and to a lesser extent with lexical diversity or syntactic complexity (chapter 8). Of note, we did not find any differences between conventional neuropsychological language tests in individuals with and without preclinical AD, suggesting that subtle changes in spontaneous speech occur very early in the AD continuum.

Inconsistent findings have been reported in literature regarding which cognitive complaint questions can be used to unmask cognitively normal individuals at risk for AD. Therefore, it remains to be elucidated whether amyloid load is associated with a specific pattern of cognitive complaints. In chapter 9 we used all available questionnaires (total scores and single items or questions) that are used in the SCIENCe cohort to assess severity and content of cognitive complaints, and we found that increased amyloid-β load was associated with worries about the self-perceived cognitive decline, but not with any specific item on cognitive complaints or questionnaire. One explanation is that inquiries about cognitive complaints in a memory clinic setting could lead to receiving false-positive answers, which do not necessarily relate to SCD due to preclinical AD. Notwithstanding, these findings suggest that worries about self-perceived cognitive decline is a strong indicator of preclinical AD. These findings are also in line with results described in chapter 6, showing that a dichotomous SCD question can provide an unrefined but effective risk stratification for individuals at risk for AD.

METHODOLOGICAL CONSIDERATIONS

There are a number of methodological considerations. First, we studied patients who visited a memory clinic because they perceived deficits in their cognitive functioning. Cognitive complaints in elderly could be caused by a myriad of factors, including normal aging, preclinical AD, cerebrovascular disease, psychological factors, mood disorders and many more. Others have shown that memory clinic individuals with SCD exhibit additional subclinical depressive symptomatology if compared to SCD in the general population. In this thesis we aimed to investigate the earliest changes related to AD, and the majority of the observed associations between neuroimaging, cognition, and cognitive complaints seemed independent of age, gender, education and (subclinical) depressive symptoms. It is nearly impossible to adjust for all factors that could potentially influence cognitive complaints. In addition, factors influencing cognitive complaints will greatly vary from one individual to another, and very large datasets with refined psychometric measures will be necessary to adjust for human variation. Of note, in most of our chapters we additionally adjusted for common causes of cognitive complaints such as: normal aging (chapters 3-9), depressive symptoms (chapter 5 and 6), cerebrovascular disease (chapter 5) and neuroticism (chapter 6). Furthermore, because the majority of research presented in this thesis was performed in a memory clinic setting, it is unclear to which extent our results can be extrapolated to the general population. Others observed increased anxiety and amyloid-β deposition in both community-dwelling as well as memory clinic SCD, whereas subclinical depression and (hippocampal) atrophy was mostly observed in memory clinic SCD. Notwithstanding, our results are highly relevant for cognitively normal individuals with SCD in memory clinics. Moreover, referred SCD patients could be a clinically relevant group for future preventive trials as they self-perceive cognitive decline and actively seek help. Third, because we investigated cognitively normal individuals, we cannot rule out that (subtle) practice effects may have occurred in relation to our cognitive data. We expect that these effects were especially present in “lower-risk” cognitively intact individuals, but that practice effect may have been less apparent in individuals with future progression.

In addition, in our longitudinal studies individuals were tested once each year, where available with parallel versions of the test (chapter 6), which reduces the risk for practice effects. Others have demonstrated that preclinical AD could reduce the magnitude of gain from retesting, especially with regard to episodic memory. Fourth, ideally one would use repeated neuroimaging measures to investigate brain changes related to AD, and many of our studies made use of a single neuroimaging measure to predict a certain clinical outcome. Therefore, we cannot fully disentangle the brain changes that might have occurred during the time of study. Nonetheless, single neuroimaging measures are commonly used for diagnostic decision making, recruitment for intervention studies, and generally reflect a realistic clinical setting. Fifth, we have used different measures of amyloid-β load in this thesis (radiotracers [18F]florbetapir and [18F]florbetaben, and Aβ1-42 cerebrospinal fluid (CSF) levels), which could have impacted our findings. Notwithstanding, both imaging agents are able to detect beta-sheet fibrillar deposits of amyloid β-protein in plaques, and both CSF and PET provide comparable diagnostic accuracies in AD patients, but lower accuracies in cases with less amyloid-β deposition. More importantly, for all our clinical PET studies (chapter 8 and 9), we used state-of-the-art quantification techniques to estimate specific binding of [18F]florbetapir. Compared to standardized uptake values