Sex differences in Alzheimer’s disease revealed by free-water DTI and voxel-based morphometry

Abstract

Background

Imaging biomarkers are increasingly used in Alzheimer’s disease (AD), and the identification of sex differences using neuroimaging may provide insight into disease heterogeneity, progression, and therapeutic targets.

Objective

The purpose of this study was to investigate differences in grey matter (GM) volume and white matter (WM) microstructural disorganization between males and females with AD using voxel-based morphometry (VBM) and free-water-corrected diffusion tensor imaging (FW-DTI).

Methods

Data were downloaded from the OASIS-3 database, including 158 healthy control (HC; 86 females) and 46 mild AD subjects (24 females). VBM and FW-DTI metrics (fractional anisotropy (FA), axial and radial diffusivities (AxD and RD, respectively), and FW index) were compared using effect size for the main effects of group, sex, and their interaction.

Results

Significant group and sex differences were observed, with no significant interaction. Post-hoc comparisons showed that AD is associated with reduced GM volume, reduced FW-FA, and higher FW-RD/FW-index, consistent with neurodegeneration. Females in both groups exhibited higher GM volume than males, while FW-DTI metrics showed sex differences only in the AD group. Lower FW, lower FW-FA and higher FW-RD were observed in females relative to males in the AD group.

Conclusion

The combination of VBM and DTI may reveal complementary sex-specific changes in GM and WM associated with AD and aging. Sex differences in GM volume were observed for both groups, while FW-DTI metrics only showed significant sex differences in the AD group, suggesting that WM tract disorganization may play a differential role in AD pathophysiology between females and males.

Introduction

Alzheimer’s disease (AD) is the most prevalent cause of dementia in elderly people and is characterized by the accumulation of amyloid-β (Aβ) within the brain, along with hyperphosphorylated tau. AD is significantly more common in women than in men; in the United States, the estimated AD incidence is split approximately 2:1 between women and men, respectively [1]. Numerous population-based studies have examined sex differences in AD, with some reporting a higher incidence of AD among females [2, 3] while others have reported no such differences [4, 5]. Early explanations of observed sex differences hypothesized that differential life expectancy between men and women was a contributing factor [3]; other explanations have indicated possible protective factors in males, including a higher synaptic density, greater number of neurons, and larger brain size, that might partially explain why males seem to have a greater defense against pathology translating into clinical symptoms of AD [6, 7]. Alternatively, some studies have indicated that females may have higher cognitive resilience, as well as stronger memory abilities, leading to later diagnosis and faster apparent decline [8–11]. While some studies have suggested that people with larger intracranial volume are less likely to develop AD [12], current evidence suggests that intracranial volume does not contribute to so-called cognitive reserve [13].

Sex differences in aging and AD have been documented in epidemiological [14], clinicopathological [15], and, more recently, imaging studies [15, 16]. Imaging-based biomarkers using positron emission tomography (PET) and magnetic resonance imaging (MRI) may provide a window into the aging brain and how the brain changes in males and females with both healthy aging and AD. Similar to findings from post-mortem histopathology [17], studies using Aβ PET tracers, as well as cerebrospinal fluid (CSF) biomarkers, have largely shown a lack of clear sex differences in both cognitively normal and cognitively impaired individuals [18–20]. On the other hand, tau biomarkers of AD pathology, which are more closely related to cognitive decline, have shown sex differences [2, 17, 19]; moreover, observed sex differences in tau biomarkers may be modified directly or indirectly by apolipoprotein E (APOE) genotype and/or Aβ levels [21–23]. Global and regional brain atrophy, as a biomarker of neurodegeneration, can be assessed using high-resolution structural MRI, with some studies finding sex differences in healthy subjects [20, 24] and in subjects with AD [19, 24, 25]. However, these findings related to brain atrophy have been largely inconsistent, which may be attributable to methods for correcting (or normalizing) for head-size and not to true sexual dimorphism [26].

More advanced MRI methods show particular promise for assessing both brain structure and microstructure. For example, subtle structural changes in grey matter (GM) volume, indicative of neurodegeneration, can be assessed using voxel-based morphometry (VBM) with high-resolution anatomical MR images [27]; importantly, this voxel-wise approach minimizes confounding differences in intracranial volume (ICV) [28]. A complementary method, diffusion tensor imaging (DTI), is a powerful tool for assessing white matter (WM) microstructural changes in vivo. Using DTI, the location, orientation, and anisotropy of the brain’s WM tracts can be estimated [29]. Both VBM and DTI have been used in AD and early-AD prodromes as biomarkers of GM volume [30–33] and WM disorganization [34], respectively. In addition to group differences with aging and AD, VBM has revealed unique sex-specific patterns in age-related GM changes in a healthy aging cohort [35]. Previous studies have also investigated sexual dimorphism using microstructural DTI biomarkers in young healthy subjects [36] and in subjects with mild cognitive impairment (MCI) [37]. In the MCI cohort, O’Dwyer et al. showed a significant main effect of sex for DTI metrics, where male HC and MCI subjects showed significantly more microstructural damage than their female counterparts [37]. However, to our knowledge, VBM and DTI have not been previously employed together to study sex differences between HC and patients with AD.

Although DTI is widely used, it is well recognized that there are potential limitations of this technique. One such limitation of DTI involves inaccuracies associated with partial volume effects (PVEs), wherein DTI metrics reflect a weighted average of multiple diffusion components within a voxel rather than characteristics specific to a single tissue type [38]. This limitation may be particularly important in AD where neuronal loss leads to secondary ex vacuo increases in CSF and free water (FW) content across the brain. To minimize this effect, FW-correction algorithms can be implemented [39] and have been shown to improve the accuracy of the resulting DTI metrics in aging populations [40].

In this study, we employed VBM and DTI to investigate differences in GM volumes and WM disorganization between males and females in mild AD and healthy aging subjects. All MRI data were downloaded from the OASIS-3 database (http://oasis-brains.org/), which is a freely available neuroimaging dataset for normal aging and AD. To overcome DTI limitations related to PVEs, a FW correction algorithm was used to quantify and remove the contribution of extracellular FW to DTI data (FW-DTI) [39]. We subsequently quantified differences in GM volumes and WM disorganization between males and females in AD using VBM and FW-DTI. As imaging biomarkers are increasingly used in AD, the identification of sex-specific radiological phenotypes may provide insight into disease heterogeneity, progression, and novel therapeutic targets, thus offering a more patient-centered approach for the clinical management of AD.

Material and Methods

Participants

MRI data were downloaded from the OASIS-3 database (http://oasis-brains.org/). The total number of subjects included in this study were as follows: 158 healthy controls (HC, 86 females and 72 males with a mean age (standard deviation; S.D.) of 71 (8) and 70 (7) years, respectively) and 46 mild AD, as defined clinically and characterized by the Clinical Dementia Rating (CDR) [41], (24 females and 22 males with a mean age (S.D.) of 73 (7) and 74 (6) years, respectively). Inclusion criteria for all participants were availability of 3-Tesla structural T1-weighted images and DTI data with 64 directions. For each subject, only one time-point was used for analysis. All subjects completed the Mini-Mental State Exam (MMSE) [42] and the CDR test [43]. The MMSE is a 30-point assessment commonly used to measure cognitive impairment, while the CDR is a 5-point scale used to characterize multiple domains of cognitive and functional abilities. The mean value of MMSE was 29.3 (S.D. 1.0) for HC and 26.1 (S.D. 4.4) for AD. The mean CDR values for HC and AD were 0 and 0.69 (S.D. 0.46), respectively. Complete subject characteristics are summarized in Table 1.

Table 1.

Complete subject characteristics for each analysis. For DTI, three subjects were removed due to excessive motion during acquisition. ANOVA was used to test age differences across all sub-groups. Kruskal-Wallis rank sum test was used to evaluate differences in MMSE and CDR. The Post-hoc comparison was performed using Wilcoxon rank sum test with FDR correction for multiple comparisons.

VBM analysis
	N	AGE (y); range (y)	Education	MMSE	CDR
HC-F	86	71 (8); 40–90	15.9 (2.4)	29.38 (0.89)	0
HC-M	72	70 (7); 52–92	16.8 (2.1)	29.21 (1.11)	0
AD-F	24	72 (7); 62–83	14.8 (3.1)	25.17 (5.67)	0.77 (0.59)
AD-M	22	74 (6); 63–81	16.8 (2.0)	26.35 (2.55)	0.59 (0.19)
ANOVA	-	F=1.41; p=0.24	F=5.15; p=0019	-	-
Kruskal-Wallis rank sum	-	-		χ2 =54.05; p<0.0001	χ2 =201.57; p<0.0001
Wilcoxon signed-rank test (p-values)
HC-F vs HC-M	-	-		0.55	-
HC-F vs AD-F	-	-		<0.0001	<0.0001
HC-F vs AD-M	-	-		<0.0001	<0.0001
HC-M vs AD-F	-	-		<0.0001	<0.0001
HC-M vs AD-M	-	-		<0.0001	<0.0001
AD-F vs AD-M	-	-		0.97	0.30
DTI analysis
	N	AGE (y); range (y)	Education	MMSE	CDR
HC-F	86	71 (8); 40–90	15.9 (2.4)	29.38 (0.89)	0
HC-M	71	70 (7); 52–92	16.7 (2.1)	29.21 (1.12)	0
AD-F	23	73 (6); 62–83	14.8 (3.0)	25.34 (5.73)	0.78 (0.60)
AD-M	21	74 (6); 63–81	16.8 (2.0)	26.81 (2.14)	0.60 (0.20)
ANOVA	-	F=1.69; p=0.17	F=5.09; p=0.020	-	-
Kruskal-Wallis rank sum	-	-		χ2 =48.12; p<0.0001	χ2 =197.75; p<0.0001
Wilcoxon signed-rank test (p-values)
HC-F vs HC-M	-	-		0.60	-
HC-F vs AD-F	-	-		<0.0001	<0.0001
HC-F vs AD-M	-	-		<0.0001	<0.0001
HC-M vs AD-F	-	-		<0.0001	<0.0001
HC-M vs AD-M	-	-		<0.0001	<0.0001
AD-F vs AD-M	-	-		0.86	0.33

MRI acquisitions

As part of the OASIS-3 protocol, MRI data were acquired at 3T (Magnetom Trio; Siemens). DTI acquisition was performed using 64 diffusion-encoding directions with the following parameters: b = 1000 s/mm²; TR/TE: 11000/87.0 ms; flip-angle = 90°; matrix: 96 × 96; field of view (FOV): 24.0 × 24.0 cm²; slice thickness: 2.5 mm; 64 axial slices and one non-diffusion-weighted image (B0 image). The structural T1-weighted images (Magnetization Prepared - RApid Gradient Echo; MPRAGE) were acquired with the following parameters: TR/TE: 2400/3.08 ms; flip angle: 8°; matrix: 256 × 256; FOV: 25.6 × 25.6 cm²; slice thickness: 1 mm; 256 axial slices.

VBM pre-processing

FSL-VBM [44] was used to perform VBM analysis on the MPRAGE images [45]. Brain extraction was performed using ROBEX [46] before proceeding with the FSL-VBM pipeline, through which the study-specific GM template was then created via non-linear registration. Finally, all GM images were non-linearly registered to the template and concatenated into a 4D image that was smoothed using an isotropic Gaussian kernel (sigma, 3 mm). The ICV values obtained using FreeSurfer (https://surfer.nmr.mgh.harvard.edu/) were provided in the OASIS-3 database.

DTI pre-processing

DTI pre-processing was performed by Mrtrix3 [47], FSL [45], and ANTs (http://stnava.github.io/ANTs/). The preprocessing included denoising (dwidenoise [48]), eddy current and motion corrections (eddy [49]), and bias field correction (dwibiascorrect [50]). The eddy quality control tools were used to evaluate the quality of the DTI dataset [51]. Slices with signal loss caused by subject movement coinciding with the diffusion encoding were detected and replaced by predictions from a Gaussian process [52] (for this dataset, the mean of the total outliers was 0.36%). Subjects with total outliers greater than 5% and/or an average absolute volume-to-volume head motion greater than 3 mm were removed. Skull stripping involved a multi-step procedure, as follows. Initially, brain extraction was performed on the B0 images using dwi2mask [53]; subsequently, this brain extraction was linearly coregistered to the MNI 1mm B0 image using the FMRIB linear image registration tool (FLIRT) with an affine model [54]. Finally, the coregistration matrix was inverted and applied to the MNI 1mm mask to obtain the brain mask for each native B0 image and thus the final B0 brain extractions. Subsequently, standard fractional anisotropy (FA), FW-corrected FA (FW-FA), FW-corrected axial and radial diffusivity (FW-AxD and FW-RD, respectively), and FW index were calculated in native space using an in-house MATLAB script [39, 55]. All brain-extracted B0 images were used to create a group-wise template using ANTs software. The group template was subsequently non-linearly normalized to the MNI 1-mm space by the ANTs symmetric image normalization (SyN) algorithm [56]. To run voxel-based analysis, all DTI metrics were non-linearly coregistered to the group-wise template and then to the MNI standard space using ANTs. Both FW-FA and FW metrics in MNI space were then smoothed using FSL with an isotropic Gaussian kernel (sigma, 3 mm). A binary WM mask was created from the averaged standard FA maps from all subjects by using a threshold of 0.2. To minimize PVEs, the final WM mask consisted only of voxels that were nonzero for all subjects.

Statistical analysis

The subjects were divided into four sub-groups: female HC (HC-F), male HC (HC-M), female AD (AD-F), and male AD (AD-M). Age, MMSE, and CDR scores are presented as mean and S.D. for each group and sub-group. Differences in age and education were evaluated using the analysis of variance (ANOVA) test. Cognitive test score differences were assessed using the Kruskal-Wallis rank-sum test. Pairwise comparisons across all subjects for the cognitive tests were evaluated using Wilcoxon rank-sum test with the false discovery rate (FDR) correction for multiple comparisons.

Because the number of subjects and male-to-female ratio within each group (AD and HC) differed, imaging results are reported as effect-size which was analyzed at the voxel-based level. For both VBM and DTI analyses, a two-way ANCOVA model (with age, ICV, and education as covariates) was performed using an in-house R script (http://www.R-project.org/) to calculate the partial-eta squared (${\eta }_p^2$) for the main effects of sex and group and for their interaction at the voxel-based level. Post-hoc comparisons between sex and group were assessed using the Hedges’ g effect-size.

For ANCOVA, ${\eta }_p^2>0.06$was considered to be a medium effect size and ${\eta }_p^2>0.14$ was considered to be a large effect size; for post-hoc comparisons, g>0.50 was considered to be a medium effect size and g>0.80 was considered to be a large effect size [57]. The cluster extent size was determined for each comparison using 3dClustSim (AFNI), following estimation of the spatial auto-correlation function (ACF) using 3dFWHMx (AFNI). The minimum cluster sizes across post-hoc comparisons for each parameter ranged from 115 voxels for RD to 418 voxels for VBM (AxD, FA, and FW: 117, 142, and 171 voxels, respectively). The resulting clusters were labeled according to the JHU DTI-based white-matter atlases for DTI [58, 59] and to the Harvard-Oxford cortical and subcortical structural atlases [60], with the probabilistic atlas of the human cerebellum [61] used for VBM.

Results

For DTI analysis, three subjects (two AD (1 F and 1 M) and one HC (1 M)) were removed due to excessive motion during the acquisition. Therefore, a total of 201 subjects were employed for this analysis. For both VBM and DTI, no differences in age were found across all sub-groups (VBM: F=1.41; p=0.24; DTI: F=1.69; p=0.17). Significant differences across the sub-groups were found for education (VBM: F=5.15; p=0.0019; DTI: F=5.09; p=0.0020). Moreover, the sub-groups did differ in MMSE (VBM: χ² =54.05; p<0.0001; DTI: χ² =48.12; p<0.0001) and CDR (VBM: χ² =201.57; p<0.0001; DTI: χ² =197.75; p<0.0001). As expected, the pairwise comparisons showed only differences between HC and AD groups for these cognitive tests (see Table 1).

ANCOVA: main effect of group

Figure 1-a shows the voxel-based effect-size from ANCOVA for the main effect of the group in the MNI standard space. For all metrics (GM volumes, FW index, FW-FA, FW-AxD, and FW-RD), we found differences across all sub-groups with a large effect-size (${\eta }_p^2>0.14$). Using VBM analysis, we found lower GM volumes in AD than HC as expected, principally in the hippocampus, amygdala, anterior and posterior parahippocampal gyrus, and the anterior temporal fusiform cortex. As shown in the violin plots of Figure 1-a, higher FW index was found in AD in several WM locations. Additionally, compared with HCs, the AD group showed lower FW-FA values, mainly in the forceps minor, CC, fornix, and superior cerebellar peduncle (CP), and higher FW-AxD/FW-RD mainly in the forceps minor, genu and body of CC, fornix, uncinate fasciculus (UF), sagittal stratum (SS), and the anterior corona radiata (ACR). For each region showing significant group differences, the effect sizes and cluster volume (as a percent of the total volume of the region) are included in Table 2.

Figure 1.

Effect-size from ANCOVA of the main effects of (a) group and (b) sex. The results are shown at large-effect size (η²_p>0.14) for the main effect of group and at medium effect-size (η²_p>0.06) for the main effect of sex. The violin plots show the mean values of each metric within the significant clusters. No significant difference across groups was found for the interaction term.

Table 2.

ANCOVA effect-size for the main effect of group. Differences across sub-groups were found for all metrics at large effect-size level. ‘%’ columns indicate the percent volume of the cluster inside the corresponding brain area. ${\eta }_p^2$columns are the mean effect-size value inside each cluster.

MAIN EFFECT OF GROUP (Large Effect-Size)
WHITE MATTER	FW-index		FW-FA		FW-AxD		FW-RD		GREY MATTER	VBM
JHU WM tractography	%	${\mathbf{\eta }}_{\mathbf{p}}^{\mathbf{2}}$	%	${\mathbf{\eta }}_{\mathbf{p}}^{\mathbf{2}}$	%	${\mathbf{\eta }}_{\mathbf{p}}^{\mathbf{2}}$	%	${\mathbf{\eta }}_{\mathbf{p}}^{\mathbf{2}}$	Harvard Oxford subcortical	%	${\mathbf{\eta }}_{\mathbf{p}}^{\mathbf{2}}$
Anterior Thalamic Radiation	51.11	0.216	5.20	0.160	7.21	0.174	13.32	0.275	Left Thalamus	1.43	0.181
Cortical spinal tract	7.52	0.196	-	-	7.09	0.176	1.36	0.282	Left Hippocampus	61.12	0.195
Cingulum cingulate gyrus	58.32	0.205	3.36	0.154	-	-	16.33	0.271	Left Amygdala	63.67	0.202
Cingulum Hippo	67.81	0.241	-	-	-	-	31.70	0.302	Right Thalamus	1.66	0.171
Forceps Major	29.19	0.194	-	-	-	-	-	-	Right Hippocampus	41.33	0.184
Forceps Minor	58.19	0.242	21.22	0.179	0.82	0.154	31.67	0.281	Right Amygdala	58.67	0.181
Inferior fronto-occipital fasc	59.07	0.221	3.77	0.159	1.30	0.157	20.09	0.282	Harvard Oxford cortical	%	${\mathbf{\eta }}_{\mathbf{p}}^{\mathbf{2}}$
Inferior Longitudinal fasc	52.56	0.224	-	-	2.09	0.154	17.51	0.285	Insular Cortex	0.69	0.174
Superior Longitudinal fasc	26.28	0.112	-	-	1.04	0.077	8.76	0.143	Temporal Pole	3.38	0.191
Uncinate fasc	81.56	0.221	11.91	0.160	2.59	0.160	39.05	0.280	Parahippocampal Gyrus anterior division	13.71	0.170
ICBM-DTI 81	%	${\mathbf{\eta }}_{\mathbf{p}}^{\mathbf{2}}$	%	${\mathbf{\eta }}_{\mathbf{p}}^{\mathbf{2}}$	%	${\mathbf{\eta }}_{\mathbf{p}}^{\mathbf{2}}$	%	${\mathbf{\eta }}_{\mathbf{p}}^{\mathbf{2}}$	Parahippocampal Gyrus posterior division	12.23	0.182
Genu of corpus callosum	95.85	0.274	61.43	0.181	6.99	0.156	89.81	0.282	Lingual Gyrus	1.35	0.184
Body of corpus callosum	41.02	0.186	19.32	0.151	-	-	20.84	0.272	Temporal Fusiform Cortex anterior division	13.37	0.174
Splenium of corpus callosum	81.33	0.199	6.07	0.154	-	-	1.67	0.254	Temporal Fusiform Cortex posterior division	1.62	0.165
Fornix (column and body of fornix)	97.27	0.218	93.63	0.199	77.23	0.220	56.15	0.291	Planum Polare	4.36	0.177
Inferior cerebellar peduncle	-	-	11.11	0.153	-	-	-	-
Superior cerebellar peduncle	8.11	0.164	28.07	0.164	-	-	-	-
Cerebral peduncle	20.87	0.203	-	-	-	-	6.87	0.278
Anterior limb of internal capsule R	15.14	0.183	-	-	6.82	0.159	1.91	0.263
Anterior limb of internal capsule L	18.75	0.179	-	-	19.45	0.173	-	-
Posterior limb of internal capsule R	2.61	0.166	-	-	31.59	0.205	-	-
Posterior limb of internal capsule L	-	-	-	-	12.39	0.178	-	-
Retrolenticular part of internal capsule	57.24	0.202	-	-	5.32	0.158	9.01	0.271
Anterior corona radiata R	62.41	0.212	21.60	0.167	-	-	39.54	0.277
Anterior corona radiata L	62.45	0.221	8.35	0.155	1.61	0.170	31.79	0.282
Superior corona radiata	11.50	0.173	-	-	18.50	0.179	-	-
Posterior corona radiata	13.01	0.172	-	-	17.91	0.166	-	-
Posterior thalamic radiation	44.01	0.203	-	-	-	-	7.00	0.279
Sagittal stratum	95.11	0.297	-	-	2.73	0.149	73.94	0.307
External capsule R	72.29	0.197	6.95	0.153	2.91	0.153	12.15	0.267
External capsule L	74.21	0.204	-	-	1.18	0.145	28.46	0.271
Fornix (cres) / Stria terminalis	86.22	0.251	-	-	-	-	51.66	0.271
Tapetum R	76.34	0.161	-	-	-	-	-	-
Tapetum L	49.12	0.165	-	-	8.17	0.170	-	-

ANCOVA: main effect of sex

Figure 1-b shows the voxel-based effect-size from ANCOVA for the main effect of sex in the MNI standard space. In this case, significant clusters are shown at the medium (${\eta }_p^2>0.06$) effect-size level because sex differences across groups could be more subtle than group differences. Compared with males, VBM analysis found higher GM volumes in females principally in the cerebellum. FW index and FW-DTI related metrics showed differences between sub-groups mainly in the anterior thalamic radiation (ATR), right superior longitudinal fasciculus (SLF), cerebral peduncle, and the anterior/posterior limb of internal capsule. For each region showing significant sex differences, the effect sizes and cluster volume are provided in Table 3.

Table 3.

ANCOVA effect-size for the main effect of sex. The table shows the clusters at medium effect-size level (${\eta }_p^2>0.06$) for all metrics. ‘%’ columns indicate the percent volume of the cluster inside the corresponding brain area. ${\eta }_p^2$columns are the mean effect-size value inside each cluster.

MAIN EFFECT OF SEX (Medium Effect-Size)
WHITE MATTER	FW-index		FW-FA		FW-AxD		FW-RD		GREY MATTER	VBM
JHU WM tractography	%	${\mathbf{\eta }}_{\mathbf{p}}^{\mathbf{2}}$	%	${\mathbf{\eta }}_{\mathbf{p}}^{\mathbf{2}}$	%	${\mathbf{\eta }}_{\mathbf{p}}^{\mathbf{2}}$	%	${\mathbf{\eta }}_{\mathbf{p}}^{\mathbf{2}}$	Harvard Oxford subcortical	%	${\mathbf{\eta }}_{\mathbf{p}}^{\mathbf{2}}$
Anterior Thalamic Radiation	4.52	0.080	16.95	0.088	7.46	0.078	3.20	0.078	Left Cerebral Cortex	3.25	0.115
Cortical spinal tract L	-	-	9.46	0.085	2.00	0.067	-	-	Left Thalamus	9.12	0.091
Cortical spinal tract R	0.73	0.076	7.90	0.079	-	-	-	-	Brain Stem	1.91	0.112
Forceps Minor	2.31	0.077	-	-	-	-	-	-	Left Hippocampus	14.21	0.093
Inferior fronto-occipital fasc L	3.55	0.078	-	-	-	-	-	-	Cerebral Cortex	2.58	0.100
Inferior Longitudinal fasc L	5.26	0.077	-	-	-	-	-	-	Right Thalamus	6.74	0.089
Superior Longitudinal fasc L	-	-	1.45	0.078	-	-	-	-	Right Hippocampus	2.64	0.083
Superior Longitudinal fasc R	4.00	0.086	2.76	0.074	1.31	0.075	5.60	0.084	Harvard Oxford cortical	%	${\mathbf{\eta }}_{\mathbf{p}}^{\mathbf{2}}$
ICBM-DTI 81	%	${\mathbf{\eta }}_{\mathbf{p}}^{\mathbf{2}}$	%	${\mathbf{\eta }}_{\mathbf{p}}^{\mathbf{2}}$	%	${\mathbf{\eta }}_{\mathbf{p}}^{\mathbf{2}}$	%	${\mathbf{\eta }}_{\mathbf{p}}^{\mathbf{2}}$	Superior Frontal Gyrus	2.36	0.096
Middle cerebellar peduncle	6.98	0.081	-	-	-	-	-	-	Precentral Gyrus	2.37	0.092
Pontine crossing tract	-	-	34.18	0.084	-	-	-	-	Inferior Temporal Gyrus temporooccipital part	12.85	0.117
Fornix	42.01	0.074	-	-	-	-	2.12	0.065	Postcentral Gyrus	17.61	0.119
Medial lemniscus	-	-	6.07	0.075	-	-	-	-	Superior Parietal Lobule	13.11	0.093
Superior cerebellar peduncle R	2.32	0.064	2.02	0.077	-	-	1.61	0.072	Supramarginal Gyrus anterior division	7.04	0.099
Superior cerebellar peduncle L	-	-	2.92	0.066	-	-	-	-	Supramarginal Gyrus posterior division	3.07	0.091
Cerebral peduncle R	1.62	0.075	27.74	0.082	-	-	-	-	Angular Gyrus	4.21	0.103
Cerebral peduncle L	1.84	0.064	29.54	0.077	1.27	0.064	-	-	Lateral Occipital Cortex superior division	2.91	0.095
Anterior limb of internal capsule	-	-	64.52	0.091	44.79	0.080	-	-	Lateral Occipital Cortex inferior division	1.88	0.111
Posterior limb of internal capsule	-	-	37.74	0.082	9.82	0.071	-	-	Frontal Medial Cortex	24.31	0.111
Superior corona radiata R	-	-	6.41	0.082	-	-	-	-	Subcallosal Cortex	1.74	0.097
Superior corona radiata L	-	-	7.49	0.075	3.56	0.070	-	-	Cingulate Gyrus anterior division	3.36	0.088
Posterior corona radiata R	-	-	1.31	0.073	-	-	-	-	Precuneous Cortex	1.17	0.092
Posterior thalamic radiation L	8.47	0.070	-	-	-	-	-	-	Parahippocampal Gyrus posterior division	17.97	0.095
Sagittal stratum L	16.22	0.084	-	-	-	-	-	-	Lingual Gyrus	1.24	0.094
External capsule R	-	-	18.71	0.080	-	-	-	-	Temporal Fusiform Cortex posterior division	1.08	0.084
External capsule L	-	-	22.58	0.079	1.93	0.068	-	-	Cerebellum	%	${\mathbf{\eta }}_{\mathbf{p}}^{\mathbf{2}}$
									Left Crus I	10.40	0.104
									Right Crus I	12.14	0.099
									Left Crus II	71.20	0.138
									Vermis Crus II	21.19	0.094
									Right Crus II	52.41	0.141
									Left VIIb	71.52	0.142
									Vermis VIIb	4.42	0.091
									Right VIIb	47.20	0.130
									Left VIIIa	47.05	0.129
									Vermis VIIIa	48.98	0.092
									Right VIIIa	38.00	0.123
									Left VIIIb	44.63	0.097
									Vermis VIIIb	29.19	0.086
									Right VIIIb	45.72	0.118
									Left IX	75.03	0.127
									Vermis IX	87.76	0.125
									Right IX	60.40	0.106
									Vermis X	61.13	0.141

ANCOVA: group-by-sex interaction

We did not find any significant group-by-sex interaction at either medium or large effect-size levels.

Post-hoc comparison: AD vs. HC

The post-hoc comparisons between AD and HC are reported for large effect size (g>0.80) in Figure 2 for females (Figure 2-a) and males (Figure 2-b). Compared with HC, VBM analysis revealed lower GM volumes in the AD group for both sexes (see Table 4). As expected, the FW index demonstrated clusters in several WM areas with higher FW values in the AD group compared with HCs; no clusters with lower FW values were observed. For both sexes, the AD groups were associated with lower FW-FA and higher FW-RD than the HC groups. For FW-AxD we found clusters with both higher and lower values in AD than HC. FW clusters were detected principally in the ATR, cingulum of the cingulate gyrus, forceps minor, inferior fronto-occipital fasciculus (IFOF), UF, CC, fornix, inferior CP, superior CP, anterior corona radiata (CR), and EC. Though similar regions of group differences were observed between males and females for FW-DTI, larger clusters were observed in most regions for males (for AD vs. HC). More specifically, the percent volume of clusters from both WM atlases covered 5.0% for FW-FA, 38.5% for FW, and 4.9% for FW-RD for males, compared to 2.4% for FW-FA, 28.2% for FW, and 2.8% for FW-RD for females. The complete list of clusters for FW index, FW-FA, FW-AxD, and FW-RD is reported in Table 5.

Figure 2.

Effect-size from post-hoc comparison between AD and HC in females (a) and males (b). Similar trends were found for each group. All significant clusters in figure are shown at higher effect-size level. Similar trends were observed for both males and females, where the AD cohort showed lower GM volumes, higher FW index, lower FW-FA, and higher FW-RD than the HC group.

Table 4.

Hedges’ g effect-size from the post-hoc analysis for AD vs HC for females and males using VBM across GM areas. All results are reported at large effect size level (|g|>0.80). ‘% Vol’ columns indicate the percent volume of the cluster inside the corresponding brain area. g columns are the mean effect-size value inside each cluster.

AD vs HC (GM)	females		males
	AD<HC		AD<HC
Harvard Oxford subcortical	% Vol	g	% Vol	g
L Cerebral Cortex	3.74	−0.966	4.16	−1.084
L Thalamus	3.92	−1.014	1.95	−0.998
L Caudate	-	-	9.50	−0.938
L Putamen	6.19	−0.925	6.36	−0.967
L Hippocampus	71.33	−1.111	71.11	−1.205
L Amygdala	42.79	−0.990	88.45	−1.407
L Accumbens	-	-	81.11	−0.959
R Cerebral Cortex	2.17	−0.949	3.89	−0.988
R Thalamus	3.14	−0.991	5.75	−1.036
R Caudate	-	-	3.15	−0.903
R Putamen	-	-	8.66	−0.945
R Hippocampus	32.78	−1.070	87.38	−1.189
R Amygdala	21.34	−0.933	91.14	−1.335
R Accumbens	-	-	71.81	−0.930
Harvard Oxford cortical	% Vol	g	% Vol	g
Insular Cortex	4.74	−0.938	12.16	−1.015
Temporal Pole	6.11	−0.938	13.17	−1.108
Superior Temporal Gyrus anterior division	10.13	−0.918	−	-
Superior Temporal Gyrus posterior division	2.53	−0.939	12.98	−0.933
Middle Temporal Gyrus anterior division	17.77	−0.951	-	-
Middle Temporal Gyrus posterior division	5.69	−0.936	5.82	−0.943
Middle Temporal Gyrus temporooccipital part	2.94	−0.913	-	-
Inferior Temporal Gyrus anterior division	1.84	−0.882	-	-
Inferior Temporal Gyrus posterior division	-	-	3.65	−0.968
Supramarginal Gyrus posterior division	2.37	−0.919	1.72	−0.928
Lateral Occipital Cortex inferior division	-	-	-	-
Subcallosal Cortex	-	-	21.53	−0.983
Cingulate Gyrus posterior division	7.95	−0.975	-	-
Precuneous Cortex	17.10	−0.993	-	-
Frontal Orbital Cortex	-	-	14.66	−1.057
Parahippocampal Gyrus anterior division	21.67	−0.981	45.11	−1.062
Parahippocampal Gyrus posterior division	26.57	−1.024	19.02	−1.059
Temporal Fusiform Cortex anterior division	31.61	−0.987	24.44	−1.046
Temporal Fusiform Cortex posterior division	21.78	−0.958	5.69	−0.950
Temporal Occipital Fusiform Cortex	2.02	−0.938	-	-
Occipital Fusiform Gyrus	-	-	14.62	−1.052
Frontal Operculum Cortex	-	-	1.98	−0.933
Central Opercular Cortex	4.06	−0.906	1.67	−0.946
Parietal Operculum Cortex	12.19	−0.917	12.24	−0.936
Planum Polare	10.10	−0.955	13.69	−1.067
Heschls Gyrus includes H1 and H2	17.41	−0.922	23.32	−0.987
Planum Temporale	9.26	−0.888	16.11	−1.005
Occipital Pole	-	-	17.82	−1.079
Cerebellum	% Vol	g	% Vol	g
Left V	9.25	−0.992	-	-
Right V	8.98	−0.999	-	-
Left VI	5.03	−0.982	-	-
Vermis VI	9.31	−0.944	-	-
Right VI	6.81	−0.962	-	-
Left Crus II	0.91	−0.898	-	-
Left VIIb	8.47	−0.920	-	-
Vermis VIIb	4.42	−0.867	-	-
Right VIIb	8.65	−0.997	-	-
Left VIIIa	11.67	−0.961	-	-
Vermis VIIIa	6.15	−0.908	-	-
Right VIIIa	10.36	−0.951	-	-
Left VIIIb	2.55	−0.923	-	-
Vermis VIIIb	2.24	−0.902	-	-
Right VIIIb	0.92	−0.895	-	-

Table 5.

Hedges’ g effect-size from the post-hoc analysis for AD vs HC for females and males using DTI across WM areas. All results are reported at large effect size level (|g|>0.80). ‘% Vol’ columns indicate the percent volume of the cluster inside the corresponding brain area. g columns are the mean effect-size value inside each cluster.

AD vs HC (WM)	females										males
	FW-index		FW-FA		FW-AxD		FW-AxD		FW-RD		FW-index		FW-FA		FW-AxD		FW-AxD		FW-RD
	AD>HC		AD<HC		AD>HC		AD<HC		AD>HC		AD>HC		AD<HC		AD>HC		AD<HC		AD>HC
JHU white matter tractography	%	g	%	g	%	g	%	g	%	g	%	g	%	g	%	g	%	g	%	g
Anterior Thalamic Radiation L	25.99	0.91	-	-	7.51	0.94	-	-	9.08	0.92	41.11	1.01	5.31	−0.92	3.79	0.92	2.26	−0.89	11.74	0.95
Anterior Thalamic Radiation R	31.11	0.89	2.63	−0.83	0.81	0.93	1.08	−0.91	7.25	0.93	42.63	1.00	4.52	−0.88	0.93	0.89	1.98	−0.90	14.14	0.91
Cortical spinal tract L	3.90	0.88	-	-	1.65	0.84	-	-	-	-	8.05	1.02	-	-	3.31	0.87	-	-	0.98	0.86
Cortical spinal tract R	5.15	0.89	-	-	6.15	0.92	-	-	3.26	0.94	5.33	0.99	-	-	8.48	0.90	-	-	12.22	0.94
Cingulum cingulate gyrus L	38.11	0.95	-	-	-	-	-	-	-	-	49.58	0.98	0.78	−0.88	-	-	-	-	-	-
Cingulum cingulate gyrus R	34.04	0.90	-	-	-	-	-	-	-	-	41.22	0.96	0.70	−0.83	-	-	-	-	-	-
Cingulum Hippo L	45.64	0.94	-	-	-	-	0.83	−0.85	27.98	0.98	76.44	1.18	-	-	-	-	-	-	5.86	0.89
Cingulum Hippo R	46.30	0.97	-	-	-	-	-	-	17.92	0.94	65.64	1.07	-	-	-	-	-	-	11.92	0.97
Forceps Major	22.28	0.93	-	-	-	-	-	-	-	-	11.05	0.92	4.97	−0.90	-	-	-	-	-	-
Forceps Minor	51.86	0.97	9.03	−0.85	-	-	-	-	-	-	54.10	1.09	18.59	−0.95	-	-	-	-	0.69	0.88
Inferior fronto-occipital fasc L	39.97	0.92	-	-	-	-	-	-	-	-	50.91	1.05	7.33	−0.94	-	-	5.71	−0.92	1.82	0.89
Inferior fronto-occipital fasc R	37.80	0.91	2.39	−0.84	-	-	-	-	1.77	0.90	48.39	1.07	4.00	−0.89	-	-	1.75	−0.88	4.28	0.99
Inferior Longitudinal fasc	29.56	0.93	-	-	-	-	-	-	3.97	0.90	44.28	1.11	-	-	-	-	5.06	−0.89	6.75	0.90
Superior Longitudinal fasc L	5.90	0.88	-	-	-	-	-	-	0.67	0.94	28.99	0.98	-	-	-	-	-	-	2.95	0.92
Superior Longitudinal fasc R	5.34	0.88	-	-	-	-	-	-	-	-	30.12	0.97	-	-	1.33	0.86	-	-	1.71	0.90
Uncinate fasc L	57.71	0.91	-	-	-	-	-	-	2.92	0.86	75.81	1.11	9.60	−0.95	-	-	4.58	−0.91	4.47	0.96
Uncinate fasc R	64.95	0.93	10.41	−0.84	-	-	-	-	4.40	0.89	77.05	1.06	14.59	−0.89	-	-	14.25	−0.93	14.06	0.96

ICBM-DTI 81	%	g	%	g	%	g	%	g	%	g	%	g	%	g	%	g	%	g	%	g
Middle cerebellar peduncle	-	-	-	-	-	-	-	-	-	-	2.09	0.93	-	-	-	-	-	-	-	-
Pontine crossing tract	-	-	-	-	-	-	-	-	-	-	1.20	0.90	-	-	-	-	-	-	-	-
Genu of corpus callosum	95.22	0.98	27.78	−0.85	-	-	-	-	-	-	97.96	1.14	65.11	−0.98	-	-	-	-	-	-
Body of corpus callosum	28.60	0.89	11.76	−0.84	-	-	-	-	-	-	36.94	0.94	9.88	−0.86	-	-	0.93	−0.85	-	-
Splenium of corpus callosum	64.13	0.96	9.51	−0.86	-	-	-	-	-	-	46.57	0.91	-	-	-	-	-	-	-	-
Fornix	65.61	0.92	43.40	−0.86	-	-	41.35	-0.96	33.54	0.96	59.94	0.93	70.55	−0.98	-	-	66.16	−1.04	28.11	1.02
Inferior cerebellar peduncle R	-	-	17.67	−0.87	-	-	-	-	-	-	-	-	12.71	−0.87	-	-	-	-	-	-
Inferior cerebellar peduncle L	-	-	3.31	−0.85	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
Superior cerebellar peduncle R	-	-	23.29	−0.86	-	-	-	-	-	-	-	-	21.15	−0.92	-	-	-	-	-	-
Superior cerebellar peduncle L	11.39	0.90	31.77	−0.91	-	-	-	-	-	-	7.86	0.89	26.61	−0.92	-	-	-	-	-	-
Cerebral peduncle	12.53	0.90	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-	-
Anterior limb of internal capsule	6.34	0.87	-	-	21.11	0.91	-	-	12.90	0.90	16.28	0.90	-	-	-	-	-	-	4.73	0.86
Posterior limb of internal capsule	-	-	-	-	15.36	0.91	-	-	9.90	0.92	1.08	0.87	-	-	18.81	0.91	-	-	21.45	0.94
Retrolenticular part of internal capsule R	39.40	0.86	-	-	1.15	0.88	-	-	12.56	0.83	61.11	0.98	-	-	12.65	0.99	-	-	15.23	0.90
Retrolenticular part of internal capsule L	7.94	0.84	-	-	-	-	-	-	11.06	0.90	36.78	1.00	-	-	3.00	0.87	18.19	−0.87	5.10	0.85
Anterior corona radiata R	49.69	0.92	7.61	−0.83	-	-	-	-	-	-	53.51	1.00	12.28	-0.87	-	-	-	-	-	-
Anterior corona radiata L	43.17	0.92	0.72	−0.81	1.97	0.92	-	-	0.99	0.87	62.20	1.05	1.20	-0.83	0.53	0.84	-	-	-	-
Superior corona radiata R	13.07	0.87	-	-	13.08	0.95	-	-	8.25	0.96	5.04	0.89	-	-	12.49	0.86	-	-	19.60	0.88
Superior corona radiata L	1.72	0.83	-	-	6.86	0.83	-	-	1.37	0.83	4.25	0.88	-	-	12.91	0.88	-	-	-	-
Posterior corona radiata R	8.02	0.85	-	-	3.81	0.86	-	-	-	-	7.78	0.86	1.02	-0.86	26.37	0.87	-	-	21.98	0.89
Posterior corona radiata L	-	-	-	-	16.32	0.83	-	-	-	-	3.66	0.88	-	-	19.17	0.91	-	-	-	-
Posterior thalamic radiation R	19.91	0.91	-	-	-	-	-	-	-	-	31.22	0.96	3.60	-0.88	-	-	-	-	-	-
Posterior thalamic radiation L	43.69	0.94	-	-	-	-	-	-	-	-	18.80	0.94	-	-	-	-	-	-	-	-
Sagittal stratum	91.14	1.00	-	-	-	-	-	-	11.34	0.92	91.23	1.23	-	-	-	-	9.17	−0.88	22.41	0.95
External capsule R	26.38	0.88	6.40	−0.83	-	-	-	-	-	-	63.91	0.99	5.10	-0.87	-	-	8.95	−0.91	20.14	0.99
External capsule L	30.03	0.89	-	-	-	-	-	-	8.82	0.94	83.28	0.98	6.05	-0.87	-	-	6.23	−0.90	-	-
Fornix (cres)/Stria terminalis R	61.14	0.92	-	-	-	-	-	-	32.46	0.98	85.15	1.14	2.49	-0.84	-	-	16.45	−0.91	30.73	0.98
Tapetum R	-	-	-	-	-	-	-	-	-	-	9.40	0.86	-	-	-	-	-	-	-	-
Tapetum L	10.83	0.85	-	-	4.31	0.86	-	-	-	-	1.33	0.85	-	-	15.00	0.93	-	-	-	-

Post-hoc comparison: females vs. males

The post-hoc comparisons between females and males are all reported for large effect size (g>0.80) (with results for medium effect size (g>0.50) shown in Supplementary Figure 1). Figure 3 shows the post-hoc results between females and males for the AD group (Figure 3-a) and for the HC group (Figure 3-b). For both groups, VBM analysis detected sex differences at large effect-size, where these differences were primarily observed in the cerebellum. For GM differences with VBM, a complete list of the locations of the significant clusters is reported in Table 6.

Figure 3.

Effect-size from post-hoc comparison between females and males within the AD (a) and HC (b) groups. In the AD group, males showed lower GM volumes, higher FW index, higher FW-FA, lower FW-RD than females at large effect-size. Differences between AD males and AD females were also observed using FW-AxD. In the HC group, only VBM detected differences between females and males at the large effect-size level.

Table 6.

Hedges’ g effect-size from the post-hoc analysis for females vs males using VBM across GM areas. All results are reported at the large effect size level (|g|>0.80). In the cerebellum, we found significant clusters where females had higher GM volumes than males in both groups. ‘% Vol’ columns indicate the percent volume of the cluster inside the corresponding brain area. g columns are the mean effect-size value inside each cluster.

Females vs Males (GM)	AD		HC
	M<F		M>F		M<F
Harvard Oxford subcortical	% Vol	g	% Vol	g	% Vol	g
L Cerebral Cortex	0.58	0.989	-	-	0.78	0.941
R Cerebral Cortex	0.66	0.982	0.17	−0.924	-	-
Harvard Oxford cortical	% Vol	g	% Vol	g	% Vol	g
Superior Frontal Gyrus	2.71	1.012	-	-	-	-
Inferior Temporal Gyrus temporooccipital part	-	-	3.69	-0.924	-	-
Postcentral Gyrus	0.34	0.907	-	-	6.25	0.947
Superior Parietal Lobule	1.68	0.962	-	-	-	-
Supramarginal Gyrus anterior division	-	-	-	-	3.31	0.897
Lateral Occipital Cortex superior division	0.28	1.034	-	-	-	-
Lateral Occipital Cortex inferior division	1.19	0.924	0.55	−0.911	-	-
Precuneous Cortex	0.51	1.040	-	-	-	-
Occipital Fusiform Gyrus	8.05	0.985	-	-	-	-
Occipital Pole	4.81	0.971	-	-	-	-
Cerebellum	% Vol	g	% Vol	g	% Vol	g
Left VI	2.37	0.904	-	-	-	-
Right VI	2.52	0.934	-	-	-	-
Left Crus I	15.11	0.979	-	-	-	-
Right Crus I	13.81	0.955	-	-	-	-
Left Crus II	51.23	1.060	-	-	15.58	0.899
Right Crus II	71.36	1.134	-	-	16.57	0.901
Left VIIb	45.81	1.120	-	-	23.81	0.936
Right VIIb	54.32	1.113	-	-	13.11	0.909
Left VIIIa	18.48	0.975	-	-	12.11	0.953
Right VIIIa	12.51	1.001	-	-	12.25	0.905

Interestingly, for all FW-DTI-related metrics, we found sex differences at large effect-size level (g>0.80) in the AD group, while no significant differences between males and females at the same effect level were found for HC group. In the AD group, we found that males had higher values of FW-index and FW-FA and lower values FW-RD compared with females. For FW index, differences were observed in the right ATR, right CST, right SLF, and right CP (see Table 7). Differences for FW-FA and FW-RD were found mainly in the left ATR, right CST, pontine crossing tract (only for FW-FA), CP, and left posterior LIC.

Table 7.

Hedges’ g effect-size from the post-hoc analysis for females vs males using DTI across WM areas. Interestingly, differences were observed at large effect size level (|g|>0.80) only for the AD group. ‘% Vol’ columns indicate the percent volume of the cluster inside the corresponding brain area. g columns are the mean effect-size value inside each cluster.

Females vs Males (WM)	AD group
	FW-index		FW-FA		FW-AxD		FW-AxD		FW-RD
	F<M		F<M		F<M		F>M		F>M
JHU WM tractography	% vol	g	% vol	g	% vol	g	% vol	g	% vol	g
Anterior Thalamic Radiation L	-	-	5.44	0.931	1.70	0.892	-	-	9.11	−1.01
Anterior Thalamic Radiation R	0.84	0.905	1.32	0.848	-	-	-	-	4.22	−0.88
Cortical spinal tract L	-	-	1.80	0.922	0.38	0.848	-	-	-	-
Cortical spinal tract R	0.34	0.863	4.84	0.920	-	-	-	-	3.53	−0.91
Cingulum cingulate gyrus	-	-	-	-	-	-	-	-	3.17	−0.90
Forceps Minor	-	-	-	-	-	-	-	-	1.74	−0.88
Inferior fronto-occipital fasc R	-	-	-	-	-	-	-	-	0.65	−0.87
Superior Longitudinal fasc L	-	-	1.70	0.854	-	-	-	-	1.51	−0.92
Superior Longitudinal fasc R	0.64	0.890	0.87	0.939	-	-	1.86	−0.88	3.98	−0.91
Uncinate fasc R	-	-	-	-	-	-	-	-	5.56	−0.91
ICBM-DTI 81	% vol	g	% vol	g	% vol	g	% vol	g	% vol	g
Pontine crossing tract	-	-	14.53	0.898	-	-	-	-	-	-
Body of corpus callosum	-	-	-	-	-	-	-	-	18.89	−0.95
Fornix	-	-	-	-	-	-	-	-	0.61	−0.81
Medial lemniscus	-	-	4.63	0.855	-	-	-	-	-	-
Cerebral peduncle R	1.61	0.836	20.11	0.862	-	-	-	-	8.34	−0.84
Cerebral peduncle L	-	-	14.12	1.002	7.02	0.867	-	-	0.92	−0.82
Anterior limb of internal capsule L	-	-	-	-	-	-	-	-	3.05	−0.87
Posterior limb of internal capsule R	-	-	1.60	0.832	-	-	-	-	-	-
Posterior limb of internal capsule L	-	-	16.71	0.944	3.01	0.851	-	-	5.11	−0.82
Superior corona radiata R	-	-	-	-	-	-	-	-	2.60	−0.92

At the medium effect-size level (g>0.50) (see Supplementary Figure 1), we found differences across the sub-groups for VBM and all FW-DTI metrics for the main effect of sex. A similar trend was found in the HC group (that is, males had higher values of FW-index and FW-FA and lower values FW-RD compared with females), although these clusters were much smaller in the HC group than in the AD group. Interestingly, at medium effect-size level, we did not find differences in FW-FA and FW-AxD for the main effect of sex in the CC or fornix, although a large cluster in the fornix was found for the FW index (volume ≈ 42.0%), as well as a smaller FW-RD cluster (volume ≈ 2.10%).

Discussion

In this study, we analyzed the differences in GM volume and WM structural disorganization between males and females in healthy aging and mild AD cohorts. For this purpose, we used two complementary MRI techniques: VBM (for GM volume) and DTI (for WM disorganization via FW-FA, FW-AxD, FW-RD, and FW index). As standard DTI has limitations associated with PVEs [38], we applied an algorithm to correct for FW effects [39]. Using ANCOVA analysis, we found large differences for the main effect of group, as expected, and significant differences for the main effect of sex, though no significant group-by-sex interaction was observed in this dataset.

For both males and females, we found regionally lower GM volumes, higher FW index, lower FW-FA, both higher and lower FW-AxD, and higher FW-RD in the AD group compared to HC. Using VBM, GM volume changes were observed in the hippocampus, thalamus, and amygdala. These findings are consistent with previous VBM studies in AD [30, 62], as well as with known AD neuropathological patterns [63, 64]. Consistent with our results, several studies have shown that AD cohorts have significantly higher FW than HC cohorts [40, 65–67], and FW corrected metrics may be more accurate in aging cohorts [40]. Additionally, the FW index has been associated with CSF AD biomarkers [68] and may predict cognitive decline [67]. Lower FW-FA, and higher FW-AxD and FW-RD, are also consistent with neurodegenerative mechanisms. For FW-DTI, large differences were found in the CC and fornix. Notably, several studies have reported significant atrophy of the CC in AD [69–71], and progression of clinical disease severity has been correlated with atrophy of the CC [72]. The fornix carries afferent fibers to the hippocampus [73] and is critical for episodic memory function [74]. In this study, we found large clusters of differences between AD and HC in the fornix (FW index covered 97%, FW-FA covered 94%, FW-AxD covered 77%, and FW-RD covered 56% of the fornix) that are in line with other studies [75]. Overall, the structures showing both group and sex differences in this study are largely consistent with AD-related pathology and with prior studies.

To the best of our knowledge, this is the first study to assess sex differences in AD using both VBM and FW-DTI. In this study, post-hoc comparisons showed significant sex differences in GM volumes in both AD and HC groups, with females showing primarily higher GM volumes than males. Significantly larger GM volume in females was previously observed using VBM in young healthy adults, particularly in the pars opercularis and inferior parietal lobule [76]. Previous studies have also considered sex differences using standard DTI. Menzler et al. showed that young healthy males had significantly higher FA and lower RD in the thalamus, CC, and cingulum than young females [36]. In contrast, another study found that females had higher FA in the CC, while males had higher FA in cerebellar WM and in the left SLF [77]. In terms of aging, whole-brain FA has been negatively correlated with age, with females showing significantly lower FA in the right deep temporal regions relative to males [78]. To investigate sexual dimorphism in MCI, O’Dwyer et al. analyzed differences in WM tracts using standard DTI between males and females in MCI and healthy older participants [37]. In that study, a significant main effect of sex was reported for FA and RD, with males of both groups showing significantly more microstructural change than female participants, with no interaction reported between sex and group. These results are consistent with the present study, where we found significant sex differences for FW-DTI for the AD group; more specifically, we found lower FW index, lower FW-FA and higher FW-RD in females than in males in the AD group, which is generally consistent with previous studies. Interestingly, the HC group showed no sex differences in FW-DTI (large effect-size) and minimal differences at a lower effect-size.

Sex differences are increasingly recognized as a factor in the care and management of patients with AD. Emerging evidence suggests that males and females exhibit unique patterns in disease manifestation and cognitive decline (with females exhibiting faster decline [79, 80]), as well as behavioral and psychiatric symptoms (with males exhibiting more aggressive behaviors [81] and females experiencing more mood disorders [82]). Sex differences have previously been reported in cognitive profiles, where studies show that cognitively normal females have higher performance on verbal memory and males have higher visuospatial scores [15, 16, 83]. Interestingly, recent studies have suggested that females have higher cognitive resilience that enables preservation of cognitive function despite increasing AD biomarker abnormalities [8], including volumetric [23], metabolic [9, 10], and Aβ [84] changes. At some point, the pathological burden overcomes any advantage in cognitive resilience, which is supported by clinical studies showing that females progress faster after diagnosis [80, 85], with lower cognitive performance compared to males at the same clinical stage [86, 87], and by postmortem studies showing females have higher levels of AD pathology than males at similar levels of clinical impairment [17]. As such, standard cut-off values for clinical assessments may be insensitive to subtle changes in cognition in females [8, 17], and sex-specific normative data may improve diagnostic accuracy [11, 85], particularly for MCI. Relatedly, the majority of neuroimaging studies in AD have grouped subjects on the basis of diagnostic category, with age and/or sex included as covariates; consequently, there are limited reports of sex differences using neuroimaging biomarkers [15]. Additionally, sex differences in regional brain volumes have been largely inconsistent, though longitudinal imaging studies have more consistently shown that females experience higher annual rates of atrophy than males [88, 89].

Recent studies using more advanced biomarkers of metabolism and functional connectivity have revealed sex differences in cognitively intact older adults with subjective memory complaints [90]; more specifically, males showed higher Aβ load, lower metabolism, and lower connectivity in AD-implicated regions. Males with AD have similarly shown lower connectivity compared to females [91], while higher metabolism in females has been associated with cognitive resilience [9]. In the present study, male and female healthy aging and AD cohorts were compared using advanced structural and microstructural biomarkers with complementary sensitivities to GM and WM, respectively. We found that females and males showed similar regional results between female HC and AD and between male HC and AD. Interestingly, the group comparison in females showed more GM volume changes in the cerebellum, while males had minimal cerebellar changes.

In this study, we found that AD is associated with lower GM volume, higher FW index, lower FW-FA, and higher FW-RD, all of which are likely indicative of neurodegeneration. Although females are disproportionately affected by AD, females in both groups exhibited higher GM volume than males. This difference was larger in the AD group relative to the HC group, which may suggest that GM volume is not a strong factor in the higher incidence of AD in females. Alternatively, this finding may support the hypothesis of increased ‘cognitive reserve’ in males [90], though further study is needed to verify this premise. Additionally, females in the AD group showed lower FW index than males in only a few small, isolated clusters, which further suggests that primary neurodegenerative mechanisms may not differ between males and females. In contrast, lower FW-FA was observed in females relative to males in the AD group, including in large clusters covering the pontine crossing tract, cerebral peduncle, and the left PLIC. These findings suggest that FA as a biomarker of WM tract disorganization may be reflected in the disproportionate incidence rates of AD between females and males.

The underlying mechanisms that lead to sex differences in both healthy aging and AD cohorts are multifactorial. Studies have shown that sex differences in the brain begin early in development, and the impact of sex hormones over the lifespan is likely a major contributing factor [92, 93]. Additionally, a range of environmental and lifestyle factors may further contribute to these differences. For example, males are more likely to have adverse health factors, such as smoking, alcohol use, and hypertension, but are also more likely to have attained higher education levels in older cohorts due to generational and gender-specific factors [15, 90]. Genetic factors such as APOE genotype may also play a role in sex differences, both as a primary factor and as a secondary modulator [93, 94]. More recent evidence has shown that tau levels are increased in female APOE-ε4 carriers, but that this difference may be related to testosterone levels, where higher testosterone is associated with lower tau in both sexes [95]. These findings highlight the complex and multifaceted role of sex differences in aging and AD, and more studies are needed to ascertain the impact of sex- and gender-specific factors on neuroimaging biomarkers.

Finally, the concept of sexual dimorphism in the human brain is not without controversy [96, 97]. In healthy subjects, the predominant neuroimaging sex difference is the well-accepted difference in brain size (males > females), which may lead to perceived differences in volumetric and connectivity measures. However, a recent review of decades of MRI data found limited evidence of sex differences after accounting for brain volume [96]. This is consistent with the findings of the present study, where the HC cohort showed almost no sex differences at large effect size (small clusters were observed with VBM, while no clusters survived for any FW-DTI metrics). Interestingly, the presence of sex differences in the AD cohort is juxtaposed with a lack of sex differences in the HC cohort, which could be indicative of overlooked sex-specific pathophysiology in AD. However, further studies are warranted to verify these findings in larger independent cohorts. Until then, caution is warranted in the interpretation of these observed sex differences in AD.

There are a few limitations in this study. The first limitation relates to potential contributing factors that could not be accounted for; more specifically, we did not have information related to genetic and environmental risk factors. As these factors may contribute further to sex differences, future studies should include these factors. The second limitation is that the number of subjects within each group was not matched. As Table 1 shows, the number of HC subjects was higher than the number of AD subjects. Since standard statistical testing (e.g., t-test) and, more specifically, p-values are strongly affected by the sample size of a study, we used the effect-size for both ANCOVA and post-hoc comparisons to overcome this limitation. More specifically, effect size is a quantitative measure of the difference between two or more groups [98]. The final limitation is related to DTI acquisition, where the diffusion data in the OASIS database were acquired with a single-shell acquisition. Two well-known confounding factors for DTI analysis are PVEs and crossing fiber differentiation. The limitation associated with PVEs was minimized using an algorithm to remove the FW components from DTI data. However, the FW correction algorithm is ill-posed for single-shell DTI acquisitions [99], thus requiring regularization, and this method does not address limitations related to crossing fibers, as FW-FA may be underestimated in regions with crossing fibers [100]. To completely overcome these two limitations, different diffusion MRI acquisitions (such as multi-shell) and analysis methods must be employed.

In conclusion, this study supports the existence of substantial volumetric and microstructural changes between healthy aging and AD, with similar spatiotemporal trends observed between males and females in regions critical to the development and symptomatology of AD. We further examined sex differences in AD beyond that seen in healthy brain aging in both white and grey matter, demonstrating subtle volumetric and microstructural changes in females and males. Notably, higher cerebellar GM volume and lower WM microstructural disorganization in females with AD were observed and warrant further investigation. Lastly, these findings suggest that sex-specific radiological AD phenotypes should be considered in the development and assessment of neuroimaging biomarkers in AD cohorts.

Supplementary Material

Supplemental File

Supplementary Figure 1

Effect-size from post-hoc comparison between females and males within the AD (a) and HC (b) groups at medium effect-size level.