APOE ɛ 4 Allele Moderates the Association Between Basal Forebrain Nuclei Volumes and Allocentric Navigation in Older Adults Without Dementia

INTRODUCTION

Alzheimer’s disease (AD) is a neurogenerative disorder, characterized by gradual cognitive decline resulting in a dementia syndrome, where underlying pathophysiological changes begin decades before onset of clinical symptoms [1]. A major genetic factor for sporadic AD, related to increased cerebral amyloid-β deposition [2], is APOE ɛ4 allele (E4) [3]. One copy of E4 increases the likelihood of developing late-onset AD 3-4-fold, while two copies increase the likelihood 10-12-fold [4]. E4 is also associated with more severe brain atrophy, including the hippocampus (HC) [5] and the entorhinal cortex (EC) [6], lower levels of basal forebrain (BF) cholinergic activity [7], and worse cognitive performance [8].

Spatial navigation and episodic memory impairment [9, 10] have been suggested as promising cognitive markers of early AD [11 –16]. Understanding of the processes underlying spatial navigation impairment in early AD and the individual brain structures involved in spatial navigation may help with development of accurate language— and culture— independent screening tools for detection of early cognitive decline in AD. Generally, two spatial navigation strategies have been described: 1) Egocentric (i.e., self-centered) navigation uses the position of the body and its axes (e.g., up-down, left-right, back-front) within the environment as a frame of reference for tracing a route from one place to another. This strategy depends mostly on the posterior parietal cortex and the caudate nucleus [17]. (2) Allocentric (i.e., world-centered) navigation uses the position of other objects (i.e., distal landmarks) and their relative positions to each other to trace routes within the environment. This strategy depends mostly on the HC and adjacent brain structures [17] but may be influenced by structures beyond the medial temporal lobe (e.g., prefrontal cortex (PFC), BF) [18]. Allocentric strategy is severely impaired in individuals with AD dementia and is also impaired in individuals with mild cognitive impairment (MCI) due to AD and preclinical AD [11, 19]. Egocentric strategy appears to be impaired in individuals with AD dementia and MCI due to AD but not in preclinical AD [11, 19].

The human analogue of the Morris water maze (hMWM) has been shown to be a reliable tool for evaluating spatial navigation abilities of individuals at pre-dementia and dementia stages of AD and for assessing different navigation strategies (e.g., allocentric versus egocentric) [19 –21]. In our previous studies, we found that E4 carriers performed worse in the allocentric spatial navigation (aSN) task of the hMWM compared to non-carriers [22 –24] and their performance was associated with right HC volume [25, 26], thickness of the EC [27], and BF volume [28]. Navigational performance in a virtual analog of the hMWM was associated with grey matter volume in the lateral PFC [29]. Several studies also explored spatial navigation abilities in older adults at risk of AD dementia [19 , 30].

The HC is a key structure for aSN processing [31, 32], where in particular, volume of the posterior HC was reported to be associated with aSN performance [30, 33]. Among the other structures involved in spatial navigation, the EC is known to play an important role in spatial information processing through its multiple spatially and directionally selectively firing cells and strong interconnections with the HC [34 –36]. It has been suggested that with increasing age, spatial learning and navigational performance become more dependent on extrahippocampal regions, undergoing compensatory shift from the medial temporal lobe to frontal cortex, especially to the dorsolateral PFC [37, 38].

The BF has also been shown to affect aSN, likely by interfering with processing of visual information from the environment [39]. As a heterogeneous structure consisting of several nuclei, the BF provides cholinergic input for many cortical and subcortical areas [40 –43]. The individual BF nuclei are referred to as Ch1-Ch4 [40]. In the context of spatial navigation capacity several BF regions are of special interest: 1) the medial septal nucleus (Ch1) loosely interconnected with the nucleus of the vertical limb of diagonal band of Broca (Ch2) whose neurons project primarily through the fornix to the HC; 2) the posterior part of the nucleus basalis of Meynert (Ch4p), projecting to temporal pole and adjacent cortices; and 3) the anterior-intermediate part of the nucleus basalis of Meynert (Ch4ai) projecting to medial regions of the hemispheres, amygdala, and insular, parietal and prefrontal cortices [41]. There are multiple pathways through which the BF may affect aSN performance: first, through disruption of its direct projections from the Ch1-2 to the HC; second, through disruption of its projections from the Ch4p to the EC, that could affect the HC indirectly; and third, through disruption of its projections from the Ch4ai to the PFC, affecting executive functions and associative learning required for successful aSN [44, 45].

E4 interferes with brain function on multiple levels. Functional MRI (fMRI) studies have shown that E4 disrupts resting state fMRI connectivity, leading to higher resting state activity in default mode as well as higher HC activation during memory tasks in E4 carriers [46, 47]. It has been hypothesized that these changes are compensatory in nature, and are due to increased neuronal activation required to achieve the equivalent level of performance in E4 carriers [46, 47]. Compensatory recruitment of additional neuronal structures during a demanding working memory task in E4 carriers indicates that different brain structures are involved during neuronal processing in E4 carriers and non-carriers [48]. For detailed discussion of E4 mechanisms in AD we refer readers to [49].

In this study, we aimed to: 1) determine associations between atrophy of the individual BF nuclei and aSN performance in E4 positive and E4 negative older adults without dementia and 2) assess whether the associations operate on aSN through the HC, EC, and PFC or independently from these structures. In particular, we were interested in the following pathways: a) Ch1-2 atrophy affecting aSN either directly, or indirectly through its association with HC volume; b) Ch4p atrophy affecting aSN directly, indirectly through the EC, or indirectly sequentially through the EC and HC; and c) Ch4ai atrophy affecting aSN directly, or indirectly through the PFC.

METHODS

MRI analysis

Volumetry

We used the FreeSurfer 5.3 suite to obtain volumes of the right and left HC and thickness of the right and left EC and PFC as well as estimated total intracranial volume (eTIV). The processing details are described elsewhere [56 –59] and available online at https://surfer.nmr.mgh.harvard.edu/. The PFC thickness was computed as an area-weighted mean of the following cortical regions obtained from FreeSurfer cortical parcellation based on Desikan-Killiany cortical atlas [56] – the rostral and caudal portions of middle frontal gyrus and the superior frontal gyrus.

BF segmentation

Brain volumes were skull-stripped and B1 field intensity inhomogeneity correction was performed using the N4 algorithm [60] implemented within the freely available Advanced Normalization Tools package (ANTs) (https://stnava.github.io/ANTs/). To obtain volumes of individual BF nuclei, we followed the modification of the previously described protocol [61 –63]. MRI data were processed using statistical parametric mapping (SPM8, Wellcome Trust Center for Neuroimaging) and the VBM8-toolbox (https://dbm.neuro.uni-jena.de/vbm/) implemented in MatLab R2015b (MathWorks, Natick, MA). We used the BF mask based on a cytoarchitectonic map of the BF cholinergic nuclei aligned in MNI space, derived from combined histology and MRI of a postmortem brain of a 56 old male, who died from myocardial infarction, with no neurological symptoms at the time of death. Location of the BF nuclei were identified using histological staining that were manually transferred into postmortem MRI space and subsequently transformed into MNI standard space [61, 64] (Fig. 1). The mask included subregions corresponding to the Ch1-2, Ch3, Ch4p (posterior), Ch4ai (anterior and intermediate) nuclei and nucleus subputaminalis (NSP). We non-linearly registered all the images into the MNI152 template and used the resulting DARTEL parameters [65] to warp the cytoarchitectonic map into individual brain scans. The volumes of the Ch1-2 (projecting directly to the HC), Ch4p (projecting to the EC cortex), and Ch4ai (projecting to medial regions of the hemispheres, amygdala and insular, parietal and prefrontal cortices) nuclei [40] were extracted. The warps were visually assessed for accuracy, no volumes were removed. BF and HC volumes were normalized to eTIV prior to the analysis, using the following formula: volume_individual (adjusted) = volume_individual (baseline) – B ^*(eTIV_individual – eTIV_mean), where mean eTIV = average eTIV of all participants and B = the slope of the individual volume regression on eTIV [27, 66].

OPEN IN VIEWER

Fig. 1

Basal forebrain segmentation. Localization of the BF nuclei within the MNI152 brain atlas. Blue clusters depict the Ch1-2 nuclei. Red clusters depict the posterior part of the Ch4 nucleus. Green clusters depict the anterior and intermediate parts of the Ch4 nucleus.

Spatial navigation testing

The aSN performance was assessed using the real-space version of the hMWM [67] that was placed in a real-space navigation apparatus, a fully enclosed cylindrical arena 2.8 meters in diameter and 2.9 meters high surrounded by a dark blue curtain with 8 large digital numerical displays used as distal orientation cues (Fig. 2A) [19]. The procedure was described in detail previously [19, 20]. In short, during the testing participants were requested to enter the arena and locate an invisible goal. The position of participants was tracked by recording the position of an infrared light-emitting diode placed on a top of a tall standing pole held by the participants. The position of the pole was recorded by a TV camera located centrally on the ceiling of the arena. The position of the goal in relation to the starting point and to the orientation cues was first shown at the beginning of the training task and participants were instructed to remember its position. During the training task, participants were requested to locate the hidden goal, placed on a floor of the arena, using either their starting position or two distal orientation cues on the walls of the arena. During the aSN task, participants navigated using only the distal orientation cues; the starting position was placed randomly, and its position was unrelated to the goal, requiring participants to use the aSN strategy (Fig. 2B). The task consisted of 8 trials and the goal was briefly displayed after each trial in order to facilitate the learning process. In each trial, the position of the orientation cues was randomly rotated while the relative positions of the goal to the orientation cues (and/or to the starting position in the training task) remained unchanged across the trials. The performance was recorded, and distance error was computed as a distance between the position of the pole and a correct location of the hidden goal in centimeters. Mean error across 8 trials in the aSN task was used in the statistical analyses as the main measure of aSN performance.

OPEN IN VIEWER

Fig. 2

Human analogue of the Morris water maze task. A) The real-space navigation setting – the “Blue Velvet Arena”. B) Scheme of the allocentric task showing an aerial view of the arena (large white circle) with starting point (red filled circle), orientation cues (red and green lines), and goal (purple circle).

Statistical analysis

We used the R software (R Foundation for Statistical Computing, Vienna, Austria; https://www.r-project.org) to perform the statistical analysis. Student’s t-test was used to assess the between-group differences in age, education, MRI measures, aSN scores and neuropsychology test results.

Linear associations

In order to establish that associations found in the literature hold true within our current dataset, we first performed a correlation and multivariate linear regression analysis. The associations between MRI measures and aSN scores were evaluated using Pearson’s correlation coefficients and multivariate linear regression analysis – type 3. In the multivariate analysis, we controlled for age, sex and education.

Structural equation modeling

To evaluate complex interactions and differential involvement of multiple brain structures on spatial navigation performance, we used structural equation modeling technique (SEM), implemented in the R “lavaan” library [68]. Using SEM, we were able to investigate direct and indirect effects of atrophy in multiple BF nuclei (Ch1-2, Ch4p, and Ch4ai) as well as other brain structures on aSN performance. Moreover, using the complex model we were able to assess multiple pathways through which the associations may be acting (i.e., indirect mediation pathways). SEM, unlike a multiple linear regression or a simple mediation analysis, allows accounting for associations between independent variables and assessing multiple pathways at once. This increases the overall reliability of resulting estimates.

To assess the fit quality of the model we used a combination of recommended indicators [69]. We used the χ ² value to assess overall fit and the discrepancy between the sample covariance matrix and fitted covariance matrix, the null-hypothesis being the perfect fit. Further, we used a comparative fit index (CFI), which compares the fit of the target model to the fit of an independent, or null, model. A CFI of 0.95 indicates the model improves the fit by 95% relative to the null model. We also used root mean square error of approximation (RMSEA) and the standardized root mean square residual (SRMR) metrices.

The model was considered acceptable if it met the following threshold criteria: χ ² > 0.05 (i.e., not rejecting the hypothesis that the models fit perfectly), CFI ≥ 0.90, RMSEA ≤ 0.08, SRMR ≤ 0.08 as recommended by recent guidelines [70]. We first constructed the model by adding all hypothesized associations between variables, based on evidence and theoretical assumptions (Fig. 3A), before analyzing the modification indices to account for potentially omitted interactions. Our initial assumptions were founded on known associations between aSN and analyzed brain structures [17 , 39] and known projections from BF nuclei to different parts of the neocortex [40]. After assessing the initial model modification indices to construct a model best representing the actual associations between variables, we included three additional biologically plausible associations in the final model. First, the effect of Ch1-2 atrophy on EC thickness, as Ch1-2 projects to the EC [71]. Second, the effect of HC and PFC atrophy on Ch4 volumes, given recent evidence of BF afferents from these areas [72], and third, the effect of PFC atrophy on EC cortical thickness [73]. The final model (Fig. 3B) was then used to estimate model parameters using the entire dataset and subsequently separate evaluations of E4 carriers and non-carriers (i.e., the moderation effect of the APOE ɛ4 allele) were performed. The SEM analysis was made in a single step for each group (i.e., the entire dataset, E4 carriers and E4 non-carriers). To assess possible direct effect of education on morphometric measures of the selected brain regions, we also created an alternate SEM model, that included effects of education on Ch1-2, Ch4ai, Ch4p, and HC volumes as well as EC and PFC thickness, but was otherwise identical to our final model.

OPEN IN VIEWER

Fig. 3A

SEM model diagrams. Initial SEM model diagram. Arrows indicate direction of relationships; Circular boxes, covariates; Square boxes, all other variables; Dotted lines, covariate effects; Solid lines, regressions; Labels a1-c3 correspond to labels of relationships used in the path analysis. aSN, allocentric spatial navigation (distance error in cm); Ch1-2, volume of the Ch1-2 nuclei in mm³; Ch4p, volume of the posterior part of the Ch4 nucleus in mm³; Ch4ai, volume of the anterior and intermediate parts of the Ch4 nucleus in mm³; HC, average hippocampal volume in mm³; EC, average entorhinal cortical thickness in mm; PFC = average prefrontal cortical thickness in mm.

OPEN IN VIEWER

Fig. 3B

Final SEM model diagram. Arrows indicate direction of relationships; Circular boxes, covariates; Square boxes, all other variables; Dotted lines, covariate effects; Solid lines, regressions; Dashed lines, covariances; Labels a1-c3 correspond to labels of relationships used in the path analysis. aSN, allocentric spatial navigation (distance error in cm); Ch1-2, volume of the Ch1-2 nuclei in mm³; Ch4p, volume of the posterior part of the Ch4 nucleus in mm³; Ch4ai, volume of the anterior and intermediate parts of the Ch4 nucleus in mm³; HC, average hippocampal volume in mm³; EC, average entorhinal cortical thickness in mm; PFC, average prefrontal cortical thickness in mm.

RESULTS

Associations between brain structures and allocentric spatial navigation

Volumes of the BF nuclei Ch1-2, Ch4p and Ch4ai as well as HC volume, EC and PFC thickness were negatively correlated with aSN distance error (r = –0.31– –0.59; p < 0.001) (Fig. 4). Using MANOVA controlled for age, sex, and education, all associations except those between Ch4ai volume and aSN remained significant (β= –0.21 – –0.43; p < 0.01) (Table 2).

OPEN IN VIEWER

Fig. 4

Correlations between allocentric spatial navigation and selected brain structures. aSN, allocentric spatial navigation (distance error in cm); Ch1-2, volume of the Ch1-2 nuclei in mm³; Ch4p, volume of the posterior part of the Ch4 nucleus in mm³; Ch4ai, volume of the anterior and intermediate parts of the Ch4 nucleus in mm³; HC, average hippocampal volume in mm³; EC, average entorhinal cortical thickness in mm; PFC, average prefrontal cortical thickness in mm.

OPEN IN VIEWER

Table 2

Associations between allocentric spatial navigation and selected brain structures

Variables	Beta-value	F-value	Df	p
aSN∼Ch1-2	–0.273	9.589	117	0.003**
aSN∼Ch4ai	–0.133	2.438	117	0.121
aSN∼Ch4p	–0.215	6.407	117	0.013*
aSN∼EC	–0.351	15.128	117	< 0.001***
aSN∼HC	–0.361	25.155	117	< 0.001***
aSN∼PFC	–0.430	28.788	117	< 0.001***

^* p < 0.05; ^** p < 0.01; ^*** p < 0.001; We used MANOVA type III, controlling for age, sex, and education in all analyses; aSN, allocentric spatial navigation (distance error in cm); Ch1-2, volume of Ch1-2 nuclei (medial septum and vertical band of diagonal band of Broca) in mm³; Ch4p, volume of posterior part of Ch4 nucleus (nucleus basalis of Meynert) in mm³; Ch4ai, volume of anterior and intermediate part of Ch4 nucleus (nucleus basalis of Meynert) in mm³; HC, average hippocampal volume in mm³; EC, average entorhinal cortical thickness in mm; PFC, average prefrontal cortical thickness in mm.

DISCUSSION

Using SEM to build a complex model of structures involved in aSN processing and to explore the effect of BF Ch1-2, Ch4p and Ch4ai nuclei volumes, as well as HC volume, EC and PFC thickness on aSN performance in E4 carriers and non-carriers, we found that BF atrophy affects aSN performance differently in E4 carriers and non-carriers.

Spatial navigation, especially allocentric, is modulated by cholinergic projections [39] as well as drugs interfering with the cholinergic system [74, 75]. There are multiple pathways through which efferent cholinergic projections from the BF may affect aSN. Some of these include: 1) tracts originating in the Ch1-2 nucleus and projecting directly to the HC, 2) tracts from the Ch4p nucleus projecting to the medial temporal lobe including the EC, and 3) tracts originating in the Ch4ai projecting to the PFC [40].

Evaluating the entire group of older adults without dementia, we found that after controlling for covariates (sex, age, and education), volumes of Ch1-2 and Ch4p, but not Ch4ai, were significantly associated with aSN performance. In the SEM model, all analyzed BF nuclei affected aSN performance indirectly. Specifically, Ch1-2 volume was associated with aSN performance through HC volume, Ch4ai volume was associated with aSN performance through PFC thickness, and Ch4p volume was associated with aSN performance through EC thickness and subsequently HC volume (“hippocampal path”), rather than through EC thickness alone (“non-hippocampal path”).

Since E4 is a major genetic risk factor for AD, we then analyzed the E4 carriers and non-carriers separately. We observed differential pattern of BF nuclei involvement in each group. In E4 non-carriers, the association between BF and aSN was conveyed through the volume of the Ch1-2 nucleus, which indirectly affected aSN through HC volume. However, in E4 carriers the association with the Ch1-2 nucleus volume was disrupted, while the effect of Ch4p nucleus on aSN though EC thickness and HC volume showed increased relevance. We hypothesize this is due to decreased cholinergic activity in E4 carriers where cholinergic neurons in Ch1-2 are unable to meet increased spatial navigation demands; thus, Ch4p input acts as a compensatory mechanism. Cholinergic dysfunction, as well as recruitment of additional neuronal resources during challenging cognitive tasks, have been previously reported in E4 carriers [7, 48]. It has been suggested this is due to selective impairment of synaptic plasticity and cholinergic integrity [76].

Previously, we analyzed the association between basal forebrain nuclei and aSN in older adults with AD. In our recent study [77] using the same BF mask, Ch1-2 and Ch4p nuclei volumes, unlike Ch4ai and Ch3, were strongly associated with aSN performance in older adults with MCI due to AD and mild AD dementia. Next, in the same study [77], using moderation analysis we found that HC volume moderated the association between aSN and Ch4p volume, where lower right HC volume was indicative of weaker association between Ch4p volume and aSN performance. These results suggested that the compensatory pathway through Ch4p may become less efficient with progression of hippocampal atrophy during the course of AD. In our next study using a different BF mask [78], we found the associations between BF nuclei volumes and aSN performance in cognitively normal (CN) older adults and individuals with aMCI and mild to moderate AD dementia. It should be noted that aMCI participants in this study represented “late” MCI, with an average Mini-Mental State Examination (MMSE) score of 25.3, compared to an average MMSE of 27.0 in the previous study [77]. Analyzing the whole cohort, we found associations between anterior (including Ch1-2 and Ch4ai) and posterior (including Ch4p) BF volumes and aSN. However, in the subgroup analysis, we did not find any association between BF volumes and aSN in participants with MCI and AD dementia. These results suggested that associations of both Ch1-2 and Ch4p with aSN performance are disrupted in individuals with late MCI and mild to moderate AD dementia who have severe hippocampal atrophy [78]. There was an association between anterior BF volume and performance in mixed (combined egocentric and allocentric) navigation in the AD dementia group that was independent of hippocampal volume. It is therefore likely that the BF may still contribute to spatial navigation, especially to navigation that is based on non-hippocampal strategies, which become more prevalent during the course of AD [77]. The BF likely contributes to non-hippocampal strategies through Ch4ai and projections to PFC [38]. In the current study, we did not include participants with dementia and therefore cannot make direct comparisons with these results.

A very recent study [79] evaluating aSN performance in a computerized task in CN older adults with (SCD) and without cognitive complaints found significant association between Ch4p and aSN performance, but did not report associations with other BF nuclei, supposedly finding them non-significant. This finding seems to be unexpected. However, the study was done on a different population (CN older adults) than our current work (individuals at genetic risk for AD). In addition, the study included 55 participants and may have been underpowered (the estimated power, given moderate strength of association, β= 0.3, was 0.54), which could explain the absence of associations between other BF nuclei and aSN performance.

Based on the current and previous findings, we hypothesize that in older E4 non-carriers, BF affects aSN performance through Ch1-2 and its direct projections to the HC. In individuals with mild cholinergic dysfunction, such as in E4 carriers, Ch4p is recruited to compensate for insufficient function of Ch1-2 and affects aSN through the EC and subsequently the HC. However, with increasing BF and hippocampal atrophy in late MCI and AD dementia stages, the compensation becomes less and less effective, until the association is disrupted. To the best of our knowledge, there are no studies assessing the relationship between BF and aSN in E4 positive older adults. This is also the first study showing, that Ch4p involvement in aSN processing (through the EC and HC), may in fact be compensatory for Ch1-2 dysfunction.

We also explored the associations between aSN performance and structural measures of brain regions involved in aSN processing. The HC is one of the first structures affected early in the course of AD [80] and is closely associated with spatial memory and navigation [17, 81]. Our results are in line with previous findings, showing association between HC volume and aSN performance. These findings are consistent with a growing body of evidence showing a prominent role of the HC in aSN processing [33]. Next, our results showed that EC thickness was associated with aSN performance and that this association was conveyed through HC volume. AD-related neurodegeneration starts in the EC [80]. Its medial part is strongly interconnected with the HC and contains grid cells encoding information about positions in space [82]. Here we confirm our previous findings of the association between EC thickness and aSN performance in the real-space version of the hMWM in individuals with aMCI [27]. Our current results are consistent with previous studies showing that close interplay between the EC and HC is essential for aSN processing [83]. The PFC is one of the regions that may compensate for worse aSN performance caused by HC dysfunction in early AD [37]. The PFC receives projections from multiple brain regions, including cholinergic projections from the Ch4ai [40]. In this study we observed a moderately strong association of the PFC with aSN performance. Our results are consistent with findings of studies reporting PFC activation during navigation and associative learning tasks [84, 85] and the association between lateral PFC grey matter volume and aSN performance in a virtual hMWM [29].

We used the groundwork of established methods of structural MRI and aSN assessment in conjunction with advanced statistical analysis tools (i.e., SEM) and sufficient sample size (n = 120 versus n = 55–66 in previous comparable studies [25 , 78]) to assess the interaction effects of E4 allele presence and atrophy of multiple brain structures on aSN performance. We acknowledge limitations of our study. In order to achieve sufficient sample size and to represent a wide spectrum of preclinical AD, we used study groups including both SCD and aMCI patients together. While these syndromes may differ in underlying pathology, possibly contributing to increased heterogeneity within each group, we tried to minimize the impact of this choice by matching patients in each study group using multiple criteria (syndrome, gender, age). In the subgroup analysis, we used half of our total sample, which lowered estimated statistical power of our analysis. In this study, we did not have longitudinal data available, therefore we were not able to determine the increased risk of conversion to AD dementia in the E4 positive group. These concerns should be a focus of the future studies.

Further, SEM, as well as other statistical methods alone, cannot be used to draw new causal inferences but rather to support the hypotheses based on the study design and theoretical knowledge, aligned with known biological mechanisms and anatomical relationships. Here we examined complex associations between brain structures and aSN performance that were limited to the structures with well documented anatomical connections, and we intentionally omitted the analyses without biological plausibility. The established BF mask used in this study is based on histological maps of cholinergic neurons derived from postmortem histology and MRI scanning. However this mask is not specific to cholinergic neurons and contains other neuronal populations [86]. Next, although methodology used to measure the volume of BF nuclei is well established and was used in several publications [62 , 87–89], the mask used in this study was based on the brain of a single individual. Although a probabilistic mask based on multiple subjects is available [90] we decided not to use this mask since it does not differentiate between anterior-intermediate and posterior portions of the nucleus basalis of Meynert, a distinction relevant for the purpose of this study.