Brain MRI signatures across sex and CSF Alzheimer's disease biomarkers

Abstract

The relationship between cerebrospinal fluid (CSF) biomarkers of Alzheimer's disease and neurodegenerative effects is not fully understood. This study investigates neurodegeneration patterns across CSF Alzheimer's disease biomarker groups, the association of brain volumes with CSF amyloid and tau status and sex differences in these relationships in a clinical neurology sample. MRI and CSF Alzheimer's disease biomarkers data were analysed in 306 patients of the Mass General Brigham healthcare system aged 50+ (mean age = 68.4 ± 8.8 years; 43.1% female), who had lumbar punctures within 1 year of clinical MRI scans. We first analysed neurodegeneration patterns across four biomarker groups: 60 controls (A-T-&CU; amyloid negative, tau negative, cognitively unimpaired), 25 A+T- (amyloid positive, tau negative), 121 A+T+ (amyloid positive, tau positive) and 100 other dementia (A-T-&CI; amyloid negative, tau negative, cognitively impaired). Second, we examined volumetric associations with amyloid (amyloid positive, tau negative versus control) and tau in the presence of amyloid (amyloid positive, tau positive versus amyloid positive, tau negative) across 52 brain areas. Third, we examined sex differences in these relationships. Finally, we validated core analyses across three independent datasets-NACC (National Alzheimer's Coordinating Center), ADNI (Alzheimer's Disease Neuroimaging Initiative) and EPAD (European Prevention of Alzheimer's Dementia)-totalling 3137 participants, and performed meta-analyses to obtain more robust estimates. We observed distinct neurodegeneration patterns across biomarker groups, with disrupted connectivity (brain volume covariance networks) in amyloid positive and other dementia groups, while amyloid and tau negative, cognitively unimpaired controls exhibited the most connected network. Amyloid was associated with subcortical, cerebellar and brainstem atrophy, with consistent association observations in the thalamus and amygdala across all four datasets. Tau in the presence of amyloid demonstrated general brain shrinkage through enlargement of extracerebral CSF, alongside unexpected ventricle shrinkages. Sex-based analyses revealed that A+T+ (amyloid positive, tau positive) had lower sex differences in connectivity patterns compared with other groups. Sex differences were also noted in amyloid-related ventricular volume changes. This study reveals how amyloid and tau affect brain connectivity and volume across sex and CSF biomarker groups, emphasizing global brain changes and sex differences. By leveraging automated pipelines and advanced MRI and biomarker analyses, we extracted meaningful and replicable findings from heterogeneous clinical samples from real-world data. The meta-analyses across four datasets enhance the generalizability of our results.

Keywords: CSF core Alzheimer’s disease biomarkers; brain volumetric change; morphometric connectome; predictive modelling; sex differences.

Graphical Abstract

Figure 1

The MGB dataset. Consort diagram of the patient selection process. ATN, amyloid/tau/neurodegeneration; AD, Alzheimer’s disease; w/o, without; qc, quality control.

Figure 2

Neurodegeneration patterns and morphometric connectome across CSF Alzheimer’s disease biomarker groups in the MGB dataset. Patterns of neurodegeneration in different CSF biomarker categories in the MGB dataset. (A) Partial bivariate Pearson correlation of brain volumes in the control group (N = 60), A+T− group (N = 25), A+T+ group (N = 121) and the A−T−&CI group (N = 100). Compared with groups with positive Alzheimer’s disease biomarkers or cognitive impairment, the control group exhibited more distinct and robustly correlated clusters. Colour annotations above and to the left of each figure represent the brain regions’ categories (i.e. subcortical area, cortical area, cerebellum, ventricle, brain stem and extracerebral CSF). All correlation coefficients were adjusted for age and sex and were Fisher transformed. (B) Visualization of the SCN from the top 10% strongest connections from each group of distinct CSF biomarker categories. The control group had more connections between subcortical and parietal regions, contributing to greater small-worldness than the A+T+ group (pFDR = 0.018, Wilcoxon test), indicating a more integrated network structure. Yet, there appear to be less direct connections between occipital regions and temporal regions as well as between occipital regions and subcortical grey matters in the control group compared with the other three groups. Yellow circle: A+T−, A+T+ and A−T−&CI groups had more connections between occipital lobe and temporal lobe, as well as occipital lobe and subcortical grey matter than the control and A−T−&CI groups. Blue circle: control had more direct connection between parietal lobe and subcortical grey matters. Note: Connection patterns were based on visual inspection only; no statistical comparisons were conducted for regional edge differences. (C) UMAP of whole brain and subregions (N = 306) characterized by CSF biomarker categories. The SGCC quantifies separation between categories, with higher positive values indicating greater separation. The global clustering coefficient at the whole brain level (SGCC = 0.09) was higher than that in subregions (SGCC: 0.026–0.075). Each data point represents the brain volume of an individual participant projected into the UMAP 2D space. AD, Alzheimer’s disease; A+T−, amyloid positive, tau negative; A+T+, amyloid positive, tau positive; A−T−&CI, amyloid negative, tau negative and cognitive impaired; UMAP, uniform manifold approximation and projection; MGB, Mass General Brigham; L, left hemisphere; R, right hemisphere; pFDR, false discovery rate-corrected P-value.

Figure 3

Brain volumes associated with amyloid. (A) Brain areas associated with amyloid status. Significant brain volumes linked to amyloid status using logistic regression in individuals aged 50+ in the MGB dataset (N: A+T− = 25, Control = 60), adjusted for age, sex and intracranial volume (ICV). Beta (log odds): brain stem = −0.512, P = 0.043; thalamus = −0.650, P = 0.03; amygdala = −0.681, P = 0.017; ventral DC = −0.857, P = 0.02; cerebellum cortex = −0.937, P = 0.011; cerebellum white matter = −1.200, P = 0.005. y/o, year old. (B) Feature importance ranking. Top predictors of amyloid status in the ridge logistic regression model for patients aged 50+. The x-axis (Importance) indicates the magnitude of each feature’s standardized coefficient, with features scaled before model fitting and importance values scaled from 0 to 1 for visualization. (C) Model performance. AUROC of the ridge logistic regression model for predicting amyloid status in patients aged 50+ (N: A+T− = 19, Control = 45; AUROC = 0.795, 95% CI: [0.788, 0.802], sensitivity = 0.67 at specificity = 0.87). AUROC, area under receiving operating characteristic; CI, confidence interval. (D) Brain visualization. Overlay of significant brain areas from logistic regression tests across datasets for individuals aged 50+. Darker colours indicate greater overlap (numbers show dataset counts per region); purple indicates mixed associations (positive in some datasets, negative in others). (E) Significant brain areas in logistic regression tests by dataset. A stacked bar plot displaying the count of brain areas showing significant associations across all four datasets, coloured by the association direction (positive/negative). The red box highlights the two brain areas that were significant in all datasets, indicating consistent associations. (F) Functional network overlap. Spin tests of significant brain regions from meta-analyses and functional networks revealed significant overlaps with control C (Dice = 0.08, P = 0.035), dorsal attention A (Dice = 0.16, P = 0.012) and visual associated networks (Dice = 0.19, P = 0.04), indicating consistent amyloid-associated brain volume patterns. ROI, region of interest. Network abbreviations correspond to the functional networks: DorsAttnA/B, dorsal attention network A/B; VisualA/B, visual network A/B; DefaultA/B, default mode network A/B; ControlA/B/C, frontoparietal control network A/B/C; SalVenAttnA/B, salience/ventral attention network A/B; SomatomotorA/B, somatomotor network A/B; LimbicA/B, limbic network A/B; TempPar, temporoparietal network; VisAssoc, visual association network; Default, default mode network (general); SalSubcor, salience/subcortical network; Motor, motor network; FrontPar, frontal-parietal network; MedFront, medial frontal network; InsuB, insular/brainstem network. Asterisks (*) indicate networks with significant overlap. The P-value was based on the spin test permutations of the Dice coefficients. (A) and (D) show sagittal views of cortical areas (left) and coronal, sagittal and sectional views of subcortical and white matter areas (right). A+T−, amyloid positive, tau negative.

Figure 4

Brain volumes associated with tau in the presence of amyloid. (A) Brain areas associated with tau status. Significant brain volumes linked to tau status using logistic regression in amyloid-positive individuals aged 50+ in the MGB dataset (N: A+T+ = 121, A+T− = 25), adjusted for age, sex and intracranial volume (ICV). Beta (log odds): CSF = 0.668, P = 0.006; lateral ventricle = −0.545, P = 0.009. y/o, year old. (B) Feature importance ranking. Top predictors of tau status in the ridge logistic regression model for patients aged 50+. The x-axis (Importance) indicates the magnitude of each feature’s standardized coefficient, with features z-scored before model fitting and importance values scaled from 0 to 1 for visualization. (C) Model performance. AUROC of the random forest model for predicting amyloid status in patients aged 50+ (N: A+T−=19, A+T+=91; AUROC = 0.694, 95% CI: [0.686, 0.702], sensitivity = 0.73 at specificity = 0.5). ROC, receiving operating characteristic; AUROC, area under receiving operating characteristic; CI, confidence interval. (D) Brain visualization. Overlay of significant brain areas from logistic regression tests across datasets for individuals aged 50+. Darker colours indicate greater overlap (numbers show dataset counts per region); purple indicates mixed associations (positive in some datasets, negative in others). (E) Significant brain areas logistic regression tests by dataset. A stacked bar plot displaying the count of brain areas showing significant associations across all four datasets, coloured by the association direction (positive/negative). (F) Functional network overlap. Spin tests of significant brain regions from meta-analyses and functional networks revealed significant overlaps of tau-associated brain volumes in the presence of amyloid revealed with somatomotor A (increased volumes; Dice = 0.26, P = 0.011) and language (decreased volumes; Dice = 0.17, P = 0.004) and auditory networks (decreased volumes; Dice = 0.17, P = 0.049). ROI, region of interest. Network abbreviations correspond to the functional networks: SomatomotorA/B, somatomotor network A/B; VisualA/B/2, visual network A/B/2; DorsAttn, dorsal attention network; Default, default mode network; ControlA/B/C, frontoparietal control network A/B/C; SalVenAttnA/B, salience/ventral attention network A/B; LimbicA/B, limbic network A/B; TempPar, temporoparietal network; FrontPar, frontal-parietal network; VentMulti, ventral multimodal network; PostMulti, posterior multimodal network; CingOperc, cingulo-opercular network; OrbitAffective, orbitofrontal/affective network; Auditory, auditory network; Language, language network. Asterisks (*) indicate networks with significant overlap. The P-value was based on the spin test permutations of the Dice coefficients. (A) and (D) show sagittal views of cortical areas (left) and coronal, sagittal and sectional views of subcortical and white matter areas (right). A+T−, amyloid positive, tau negative; A+T+, amyloid positive, tau positive.

Figure 5

Sex differences in neurodegeneration patterns across CSF Alzheimer’s disease biomarker groups in the MGB dataset. (A–D) Brain volume correlations by sex: partial bivariate Pearson correlations of brain volumes in males and females within each CSF biomarker group: (A) Control group (N = 60), (B) A+T− group (N = 25), (C) A+T+ (N = 121) group and (D) the A−T−&CI group (N = 100). Different connectivity patterns were observed between sexes across all groups. Brain regions are colour-coded (subcortical, cortical, cerebellum, ventricle, brain stem and extracerebral CSF). Correlations are adjusted for age and Fisher transformed. (E and F) SCNs in A+T+ group: Visualization of the top 10% strongest connections in males and females. Connection patterns were based on visual inspection only; no statistical comparisons were conducted for regional edge differences. (E) Female-specific patterns: More connections between the occipital and parietal lobes (yellow circle). (F) Male-specific patterns: More connections between the occipital lobe and subcortical grey matter (yellow circle). (G) Subcortical grey matter clustering by sex. UMAP visualization shows sex-based clustering of subcortical grey matter volumes in control, A+T− and A−T−&CI groups, but not in the A+T+ group. A+T−, amyloid positive, tau negative; A+T+, amyloid positive tau positive; A−T−&CI, amyloid negative, tau negative and cognitive impaired; UMAP, uniform manifold approximation and projection; MGB, Mass General Brigham; L, left hemisphere; R, right hemisphere.

Figure 6

Sex-differentiated brain regions in A+T− and A+T+ groups. (A) Overlap of sex-differentiated brain volumes in A+T−. Visualization of overlapping regions across four datasets. A+T−, amyloid positive, tau negative. (B) Count of sex-differentiated brain volumes in A+T−. Stacked bar plot showing the count of significant brain volumes in individuals aged 50+, coloured by association direction (Female > Male or Male > Female). (C) Functional network overlap in A+T−. Spin tests of significant brain regions from meta-analyses and functional networks revealed significant overlap with default A network (larger brain volumes in females; Dice = 0.21, P = 0.016) and control B network (larger brain volumes in males; Dice = 0.13, P = 0.037). (D) Overlap of sex-differentiated brain volumes in A+T+. Visualization of overlapping regions across four datasets. (E) Count of sex-differentiated brain volumes in A+T+. Stacked bar plot showing significant regions for individuals aged 50+, with colours indicating association direction. The red box highlights the consistently significant brain area (temporal pole) across all datasets. A+T+, amyloid positive, tau positive. (F) Functional network overlap in A+T+. Spin tests of significant brain regions from meta-analyses and functional networks revealed overlap with anterior medial temporal lobe (larger brain volumes in females; Dice = 0.16, P = 0.04), posterior medial temporal lobe (larger brain volumes in females; Dice = 0.019, P = 0.001) and medial visual networks (larger brain volumes in females; Dice = 0.012, P = 0.044) and default B network (larger brain volumes in males; Dice = 0.017, P = 0.025). Network abbreviations correspond to the functional networks: DefaultA/B/C, default mode network A/B/C; ControlA/B/C, frontoparietal control network A/B/C; VisualA/B/2, visual network A/B/2; DorsAttnA/B, dorsal attention network A/B; SalVenAttnA/B, salience/ventral attention network A/B; SomatomotorA/B, somatomotor network A/B; LimbicA/B, limbic network A/B; TempPar, temporoparietal network; Language, language network; Auditory, auditory network; VentMulti, ventral multimodal network; PostMulti, posterior multimodal network; VisualCs, visual central strip network; VisualCb, visual cerebellar network; CingOperc, cingulo-opercular network; OrbitAffective, orbitofrontal/affective network; MedVis, medial visual network; LatVis, lateral visual network; Context, contextual association network; ParMemory, parietal memory network; FrontPar, frontal-parietal network; Premotor, premotor network; PostMTL, posterior medial temporal lobe network; TLMN, temporal lobe midline network; FootSM, HandSM, FaceSM, somatomotor subregions (foot, hand, face). Asterisks indicate networks with significant overlap. The P-value was based on the spin test permutations of the Dice coefficients. In (A) and (D), darker colours indicate greater overlap (numbers show dataset counts per region); purple indicates mixed associations (positive in some datasets, negative in others).