Postmortem brain MRI reveals differential associations of subcortical and limbic volumes with cortical thinning and neurodegenerative pathologies

Abstract

INTRODUCTION

The impact of different neuropathologies on deep brain structures remains to be understood. We examine subcortical and limbic volumetry in neurodegenerative diseases involving phosphorylated tau (p‐tau), α‐synuclein, and transactive response DNA binding protein 43 (TDP‐43).

METHODS

We acquired neuropathological measures and brain segmentations from postmortem analysis of 132 donors with Alzheimer's disease (AD), Lewy body disease (LBD), frontotemporal lobar degeneration with TDP‐43 (FTLD‐TDP), and FTLD‐tau.

RESULTS

LBD had the least subcortical, limbic, and cortical atrophy compared to AD, FTLD‐TDP, and FTLD‐tau. In donors with both AD and LBD pathologies, primary LBD was associated with less atrophy than primary AD. While AD had cortico‐subcortical and cortico‐limbic morphometric associations, LBD had more limited parieto‐occipital cortico‐limbic associations. FTLD‐TDP had cortico‐subcortical while FTLD‐tau had cortico‐subcortical and cortico‐limbic associations. In AD and FTLD‐tau, hippocampal volumes correlated with p‐tau burden, neuron loss, and gliosis. In LBD, thalamic α‐synuclein severity was associated with subcortical/limbic volumes.

DISCUSSION

Postmortem neuroimaging reveals disease‐ and region‐specific structure–pathology relationships.

Highlights

• Subcortical and limbic volume loss significantly correlates with global brain shrinkage in AD.

• LBD exhibits less postmortem subcortical and limbic atrophy compared to other groups. Parieto‐occipital atrophy corresponds to limbic but not subcortical volume loss in LBD.

• FTLD‐TDP has cortico‐subcortical associations while FTLD‐tau has cortico‐subcortical and cortico‐limbic morphometric associations.

• In AD and FTLD‐tau, hippocampal volume correlates with p‐tau load, neuronal loss, and gliosis semiquantitative measures.

• In LBD, thalamic α‐synuclein burden is associated with subcortical and limbic volumes.

• In FTLD‐TDP, TDP‐43 pathology burden shows no significant volume associations.

1. BACKGROUND

Deep within the brain, subcortical and limbic structures harbor essential neural circuits with critical functions and susceptibilities to different neuropathologies such as neurofibrillary phosphorylated tau (p‐tau), α‐synuclein, and transactive response DNA‐binding protein 43 kDa (TDP‐43). Applying postmortem imaging and pathological evaluation can further parse distinct profiles of subcortical, limbic, and cortical atrophy in structure–structure correspondences and structure–pathology relationships across different neurodegenerative diseases. Thus far, studies converge on structural changes in several key limbic and subcortical structures such as hippocampus, amygdala, thalamus, and striatum in the majority of neurodegenerative diseases. ¹ , ² , ³ , ⁴ These include Alzheimer's disease (AD), ⁵ , ⁶ , ⁷ , ⁸ , ⁹ , ¹⁰ , ¹¹ , ¹² , ¹³ , ¹⁴ , ¹⁵ Lewy body disease (LBD), ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ , ¹⁸ frontotemporal lobar degeneration with TDP‐43 pathology (FTLD‐TDP), ¹⁹ , ²⁰ , ²¹ , ²² , ²³ and FTLD‐tau. ¹⁰ , ²² , ²³ , ²⁴ , ²⁵ , ²⁶

Nonetheless, substantial biological heterogeneity and polypathology exist between patients ²⁷ , ²⁸ with the majority of patients harboring more than one neuropathological diagnosis at autopsy. ³ , ²⁹ , ³⁰ Therefore, to characterize relationships between regional subcortical/limbic volume loss, cortical atrophy, and different neuropathologies, we need accurate measures of brain morphology and pathological burden across patients and diagnostic categories. Several prior studies already investigated structure‐pathology relationships by comparing antemortem imaging with postmortem pathology, but the variable time gap between these assessments may confound observed associations. In this study, we instead examined cortical, subcortical, and limbic associations between postmortem brain magnetic resonance imaging (MRI), histopathology, and immunohistochemistry. Postmortem imaging offers remarkable advantages over antemortem MRI for linking regional neuroanatomy and morphometry, enabling both higher spatial resolution and correlations between brain structure and pathological drivers at the same point in time. ⁶ , ⁸ , ¹² , ³¹ , ³² , ³³ , ³⁴ , ³⁵ , ³⁶

Here, we introduce a unique dataset of 132 postmortem specimens with primary neuropathological diagnoses of AD, LBD, FTLD‐TDP, and FTLD‐tau, along with histopathological markers and high‐resolution 7T postmortem MRI. With a deep learning model SuperSynth, ³⁷ , ³⁸ we obtained segmentations for five subcortical structures (caudate, putamen, nucleus accumbens, thalamus, pallidum) and two limbic regions (hippocampus, amygdala). Cortical segmentations were generated via purple‐mri. ³⁵ , ³⁶

With this data‐rich cohort, we sought to juxtapose subcortical and limbic volumes with cortical thickness and pathology markers to clarify and expand on prior observations of differential volume loss patterns, structure–structure correspondences, and structure–pathology relationships across diseases. For instance, we predicted that donors with FTLD‐TDP and FTLD‐tau would have more subcortical and limbic volume loss than AD, given that non‐AD p‐tau and TDP‐43 pathologies start early in subcortical and limbic regions. ¹⁰ , ²² We hypothesized that donors with primary LBD would have less subcortical and limbic atrophy than other groups, especially AD, given that neuronal α‐synuclein inclusions are linked to changes in neurotransmission and neuronal metabolism and less associated with volume loss. ¹⁶ , ³⁹ , ⁴⁰ , ⁴¹ We aimed to uncover cortical thinning patterns associated with total subcortical and limbic volume loss across diseases. Importantly, we posited that the regional burden of primary neuropathology would correlate with subcortical and limbic volume loss, as it does with cortical atrophy. ³ , ¹⁷ , ¹⁹ , ³³ Finally, we propose that regional histologic neuronal loss and gliosis are associated with gross volume loss on MRI. Overall, by integrating postmortem imaging and pathology, our approach enables us to substantiate these predictions across neurodegenerative diagnoses.

RESEARCH IN CONTEXT

1. Systematic review: The authors conducted a comprehensive literature review using PubMed and Google Scholar to identify studies examining the relationships between postmortem subcortical/limbic volumetry with multiple concomitant neuropathologies (p‐tau, α‐synuclein, and TDP‐43) involved in Alzheimer's and related dementias. Prior work primarily focused on associations between p‐tau and Aβ in subcortical structures but has not systematically evaluated the combined effects of commonly co‐occurring pathologies across distinct neurodegenerative disease groups.

2. Interpretation: Our findings suggest that across subcortical and limbic regions the most significant volume loss was found to be in FTLD‐tau and FTLD‐TDP groups, followed by AD, with LBD showing the least amount of volume loss. Cortical thinning correlated with subcortical and limbic volume loss in AD and FTLD‐tau. More limited cortical thinning was associated with subcortical volume loss in FTLD‐TDP and limbic volume loss in LBD.

3. Future directions: Validation of these findings in independent postmortem cohorts is warranted. Future work will incorporate quantitative pathology measures and investigate regionally localized effects of neuropathology within subcortical structures via shape‐based analyses.

2. METHODS

2.1. Patient donor cohort

We analyzed a dataset of 132 donors from the Penn Integrated Neurodegenerative Disease Database. ⁴² Patients were examined at the Penn Frontotemporal Degeneration Center or Alzheimer's Disease Research Center and followed to autopsy at the Penn Center for Neurodegenerative Disease Research (CNDR). ²⁵ , ⁴³ , ⁴⁴ Human brain specimens were obtained with informed consent from next of kin at time of death. Patients were evaluated per standard antemortem diagnostic criteria and postmortem autopsy and imaging. ⁴² Table 1 details primary and secondary neuropathological diagnoses. Groups included AD, LBD, FTLD‐TDP, and FTLD‐tau. Here, we denote FTLD‐tau to include 3R tauopathies such as Pick's disease and 4R tauopathies including progressive supranuclear palsy (PSP) and corticobasal degeneration (CBD). Disease‐associated genetic variant data were acquired. Study procedures were performed with written informed consent per Institutional Review Board guidelines and the Declaration of Helsinki.

TABLE 1.

Clinical characteristics of donor cohort.

Primary pathological diagnosis	Pathology	Age at death	Sex (M/F)	Education (years)	Postmortem interval (hours)	Secondary pathological diagnoses	Clinical diagnosis	Genetic variant
AD (n = 60)	Aβ, p‐tau (3R/4R)	72.9±10.8 (53.0 to 98.0)	33/27	16.2±3.0 (8 to 20)	18.8±14.1 (3 to 84)	LBD (29) LATE (13) CVD (2) CBD (1)	AD (31) FTLD‐bvFTD (7) FTLD‐PPA (logopenic) (5) PCA (5) Other (3) MCI (amnestic) (2) CBS (2) FTLD‐PPA (PNFA) (2) Normal (2) DLB (1)	1 donor: PSEN1 c.617G>C, p.G206A 1 donor: PSEN1 c.737C>A, p.A246E
LBD (n = 26)	α‐synuclein	76.6±8.4 (57.0‐96.0)	19/7	16.7±2.4 (12 to 20)	19.3±11.3 (3‐56)	AD (21) PART (3) LATE (1) AGD (1)	PDD (10) DLB (7) AD (4) PD (no dementia) (2) Normal (2) CBS (1)	None
FTLD‐TDP (n = 21)	TDP‐43	69.9±9.0 (54.0 to 91.0)	15/6	16.0±2.3 (12 to 22)	20.8±20.2 (5 to 105)	ANX (6) AD (5) Other (5) PART (3) ALS (1)	FTLD‐bvFTD (11) FTLD‐PPA (Semantic) (5) CBS (2) FTLD‐PPA (PNFA) (1) ALS (1) AD (1)	4 donor: C9ORF72 repeat 1 donor: GRN c.1009C>T, p.Q337* 1 donor: GRN c.934‐1G>A
FTLD‐Tau (n = 25) 3R: PiD = 11, 4R: PSP = 8, CBD = 6	p‐tau (3R or 4R)	72.6±8.8 (59.0 to 93.0)	14/11	16.6±2.6 (12 to 22)	23.3±12.2 (7 to 50)	AD (11) AGD (3) CVD (2) LBD (1)	FTLD‐bvFTD (12) PSP (6) CBS (2) FTLD‐PPA (Logopenic) (1) FTLD‐PPA (Semantic) (1) AD (1) LATE (1) PDD (1)	None

Note: The table summarizes group‐level characteristics for donors with neuropathological diagnoses.

Abbreviations: AD Alzheimer's disease; AGD, argyrophilic grain disease; ALS amyotrophic lateral sclerosis. ANX, annexinopathy; bvFTD, behavioral variant frontotemporal dementia; CAA, cerebral amyloid angiopathy; CBD, corticobasal degeneration; CBS, corticobasal syndrome; CVD, cerebrovascular disease; C9ORF72, chromosome 9 open reading frame 72 gene; DLB, dementia with Lewy bodies; F, female; FTLD‐TDP, frontotemporal lobar degeneration with transactive response DNA binding protein 43 kiloDaltons (TDP‐43) pathology; GRN, progranulin gene; LATE, limbic‐predominant age‐related transactive response DNA binding protein 43 kiloDaltons (TDP‐43) encephalopathy; LBD, Lewy body disease; M, male; MCI, mild cognitive impairment; PART, primary age‐related tauopathy; PCA, posterior cortical atrophy; PD, Parkinson's disease; PDD, Parkinson's disease dementia; PiD, Pick's disease; PNFA, progressive non‐fluent aphasia; PSP, progressive supranuclear palsy; PPA, primary progressive aphasia; PSEN1, presenilin 1 gene; 3R, 3‐repeat tau; 3R/4R, mixed 3‐repeat/4‐repeat tau; 4R, 4‐repeat tau.

2.2. Image acquisition

2.2.1. Antemortem imaging procedure

Antemortem T1‐weighted brain MRI scans enabled intracranial volume (ICV) calculation to normalize regional volumes. Imaging parameters varied because scans were acquired between 2006 and 2021, across a range of scanners with varying field strengths 1.5T (n = 11), 3T (n = 118), or 7T (n = 3) (Magnetom Sonata, Verio, Trio, TIM Trio, Prisma, Prisma Fit, and Terra; Siemens Healthineers, Erlangen, Germany). Median (range) parameters included repetition time 2300 ms (5 to 3000), echo time 2.98 ms (0.002 to 7.72), echo numbers 1.0 (0.0 to 4.0), flip angle 9.0, ¹⁰ , ¹⁴ , ²² , ²³ , ²⁴ , ³⁴ , ³⁷ , ⁴⁵ , ⁴⁶ , ⁴⁷ , ⁴⁸ , ⁴⁹ , ⁵⁰ , ⁵¹ , ⁵² , ⁵³ , ⁵⁴ , ⁵⁵ , ⁵⁶ and number of averages 1.0 (0 to 4.0). ³ The antemortem imaging interval had a median of 4.1 years and range of 0.28 to 15.38 years before death.

2.2.2. Postmortem imaging procedure

After the cerebellum and brainstem were removed, one hemisphere was immersed in 10% neutral buffered formalin for >4 weeks before imaging, with formalin replaced weekly to ensure complete fixation. Meninges and vasculature were removed 2 to 4 days before scanning. The hemisphere was removed from fixative 1 day before scanning and desiccated for 3 h. After fixation, samples were placed in Fomblin (California Vacuum Technology, Freemont, CA, USA) and enclosed in custom‐built 3D printed holders within a sealed bag. Samples were left to rest to allow air bubbles to escape from tissue. Samples were scanned using a custom‐built small solenoid coil or a custom‐modified quadrature birdcage coil (Varian, Palo Alto, CA, USA). ⁴³ , ⁵⁷ Then samples were placed into a whole‐body 7T Siemens Terra scanner with foam shims under the coils to raise them off the table and near the isocenter. T2w images were acquired using a 3D‐encoded T2 SPACE sequence with 0.3 × 0.3 × 0.3 mm³ isotropic resolution, repetition time 3 s, echo time 383 ms, turbo factor 188, echo train duration 951 ms, and bandwidth 348 Hz/pixel in 2 to 3 hours per measurement. Image reconstruction applied vendor reconstruction software, combining signal averages in k‐space and producing magnitude images for each echo. Four repeat measurements were acquired per sample and averaged to generate the final image.

2.3. Image analysis

Volumes of nucleus accumbens, caudate, putamen, thalamus, globus pallidus (pallidum), hippocampus, and amygdala were computed via automated segmentation of postmortem MRI by SuperSynth (https://surfer.nmr.mgh.harvard.edu/fswiki/SuperSynth). For cortical thickness measurements ⁵⁸ , parcellations based on the Desikan‐Killiany‐Tourville (DKT) ⁵⁵ atlas were obtained for each postmortem MRI with purple‐mri (https://github.com/Pulkit‐Khandelwal/purple‐mri; https://purple‐mri.readthedocs.io). SuperSynth has more subcortical and limbic regions included, while purple‐mri has a complete cortical parcellation. All segmentations underwent quality control.

2.4. Semi‐quantitative neuropathology and histology measures

Brain hemispheres were split for imaging and immunohistochemistry. Tissue blocks of approximately 1.5 × 1.5 × 0.5 cm³ were extracted from the contralateral hemisphere and embedded in paraffin, which were then sectioned into 6‐µm slices for immunohistochemistry using monoclonal antibodies to detect amyloid beta (NAB228, 1:8000, CNDR), p‐tau (PHF‐1, 1:1000, Dr. Peter Davies), TDP‐43 (TAR5P‐1D3, 1:500, Drs. Manuela Neumann and E. Kremmer), and α‐synuclein (Syn303, 1:16,000, CNDR). Hematoxylin and eosin staining was performed to assess neuronal and glial density. Semi‐quantitative scores of p‐tau, TDP‐43, α‐synuclein, gliosis, neuronal loss, and global vascular measures were visually assigned by expert neuropathologists (Drs. Edward B. Lee and John Trojanowski) on a scale of 0 to 3, from 0 none, 0.5 rare, 1 mild, 2 moderate, and 3 severe. These ratings are based on consensus criteria agreed upon by the authors and facilitated by visual exemplars. Neuronal loss and gliosis scores were determined by visual semi‐quantitative assessment of different cell morphologies (neurons versus glia) in each field of view relative to the total tissue imaged. Pathology burdens for subcortical and limbic regions were available for tissue samples from caudate, putamen, thalamus, pallidum, amygdala, and hippocampus (computed from a mean between CA1/subiculum, dentate gyrus and entorhinal cortex). Pathology ratings were not available for nucleus accumbens. Pathology ratings were recorded for additional regions including precentral gyrus, postcentral gyrus, middle frontal gyrus, angular gyrus, superior middle temporal gyrus, cingulate gyrus, occipital cortex, entorhinal cortex, midbrain, substantia nigra, pons, locus coeruleus, brainstem, and cerebellum. The ratings across all these regions allowed neuropathologists to comprehensively estimate the magnitude and distribution of different neuropathologies and adjudicate between primary and secondary diagnoses. Figure 1 shows the distribution of the neuropathology ratings across different subcortical and limbic regions along with global markers.

FIGURE 1.

Histopathology ratings across the autopsy cohort. (A) Example immunohistochemistry images. Distributions via stacked bar plots of raw counts for pathology markers, including (B) proteinopathy and histological markers (p‐tau, TDP‐43, α‐synuclein, neuronal loss, and gliosis) across entorhinal cortex, CA1/subiculum, dentate gyrus, amygdala, striatum (caudate/putamen), thalamus, and pallidum, as well as (C) additional global histopathology markers (Aβ Thal phase, Braak neurofibrillary tangle stage, CERAD neuritic plaque stage, cerebral amyloid angiopathy [CAA], arteriolosclerosis, and atherosclerosis measures).

2.5. Statistical analysis

To evaluate volumetric and pathological relationships across disease groups, we performed a series of group‐level analyses (Figure 2). For all analyses, regional volumetric measures from postmortem MRI were normalized by ICV to reduce inter‐subject variability. All models included covariates of age at death, sex, education, and postmortem interval to fixation. Antemortem comparison models included antemortem imaging interval rather than postmortem interval. All statistical tests were two‐tailed, with significance set at p < 0.05 and corrected for multiple comparisons using the Benjamini–Hochberg false discovery rate (FDR) procedure. All statistical analyses were conducted using statsmodel ⁵⁹ in Python. We subdivide the FTLD‐tau group into 3R tauopathy (Pick's disease) and 4R tauopathy (PSP and CBD) for further comparative analysis. All the statistical analysis code for the current study is available at: https://github.com/Pulkit‐Khandelwal/postmortem‐subcortical‐limbic‐pathologies.

FIGURE 2.

Deep learning 3D subcortical and limbic structure segmentations and whole‐hemisphere surface‐based parcellation. (A) The automated segmentations for hippocampus, amygdala, nucleus accumbens, putamen, caudate, thalamus, and globus pallidus were obtained via SuperSynth. Shown are MRIs with segmentation overlaid for axial, coronal, and sagittal planes along with 3D renderings for the lateral and medial views for four specimens on T2w postmortem MRI. (B) Cortical gray matter was parcellated based on the Desikan‐Killiany‐Tourville (DKT) atlas with purple‐mri followed by cortical thickness estimation in native subject‐space resolution. Shown are gray matter (pink) and white matter (cyan) surfaces on the left along with the cortical parcellation on the right and computed cortical thickness.

2.5.1. Structural group differences and subcortical–cortical associations

To assess the differential volumetric patterns of hippocampus, amygdala, nucleus accumbens, thalamus, caudate, putamen, and pallidum, we computed ANOVA group comparisons and then pairwise within‐group differences with likelihood‐ratio tests via nested linear models across groups. To examine regional relationships between cortical and subcortical matter, we computed partial Spearman correlations between mean thickness of each DKT atlas‐based cortical region and the total weighted composite subcortical and limbic volume.

2.5.2. Structure–pathology associations

To determine the relationships between regional semi‐quantitative neuropathology ratings and subcortical/limbic volumes within each diagnostic group, we fitted an independent ordinary least‐squares (OLS) regression model with volume as the dependent variable and the region's predominant primary disease‐specific pathology measure as the independent variable accounting for covariates as above, with slope coefficients β and residual error ε. For primary pathology models, AD and FTLD‐tau were compared with p‐tau, FTLD‐TDP with TDP‐43, and LBD with α‐synuclein regional burden. Next, we fitted several polypathology OLS models to examine the effects of volume loss on secondary concomitant molecular pathologies. ³

Primary pathology model:

$$\begin{array}{ccc}\hfill \text{Volume}& =& {\beta }_0+{\beta }_1(\text{primary}\text{pathology})+{\beta }_2(\text{age})\hfill \\ & & +{\beta }_3(\text{sex})+{\beta }_4(\text{education})+{\beta }_5(\text{PMI})+\epsilon \hfill \end{array}$$

Polypathology model:

$$\begin{array}{ccc}\hfill \text{Volume}& =& {\beta }_0+{\beta }_1(\mathrm{p}\text{tau})+{\beta }_2(\alpha \text{synuclein})+{\beta }_3(\text{TDP}43)\hfill \\ & & +{\beta }_4(\text{age})+{\beta }_5(\text{sex})+{\beta }_6(\text{education})+{\beta }_7(\text{PMI})+\epsilon \hfill \end{array}$$

2.5.3. Structure–histology feature correlations

To test whether MRI‐derived volumetry reflects histopathological markers of neurodegeneration independent of pathology type, we fitted an OLS model with volume as the dependent variable and regional neuronal loss and gliosis measures as the independent variables accounting for covariates as above. Furthermore, we performed mediation analyses to determine whether neuronal loss and gliosis mediated the interaction between primary disease‐specific regional pathology and its effects on structural volumetry. Finally, we assessed relationships between global vascular disease burden and structural volume for each structure by computing OLS regression adjusting for covariates between volume and global pathology scores in CAA, arteriolosclerosis, and atherosclerosis.

3. RESULTS

3.1. Pipeline for postmortem imaging, segmentation, and histopathology

Our dataset comprised 132 postmortem donor specimens spanning four primary neuropathological diagnoses: AD (n = 60), LBD (n = 26), FTLD‐tau (n = 25), and FTLD‐TDP (n = 21) (Table 1). While most donors had sporadic disease, nine had disease‐associated genetic variants. We found two donors with familial AD associated with PSEN1 exon mutations, four donors with FTLD‐TDP due to C9ORF72 hexanucleotide repeat expansion, and two donors with FTLD‐TDP due to GRN non‐coding mutations. Across the cohort, mean age‐at‐death was 73 ± 10 years with an average of 16 ± 3 years of education. The postmortem interval (PMI) averaged 20 ± 14 hours. MRI segmentations were performed for subcortical and limbic regions via SuperSynth along with the cortical gray matter DKT atlas‐based volumetric segmentations and surface‐based parcellations via purple‐mri (Figure 2).

3.2. AD, FTLD‐TDP, and FTLD‐tau have lower subcortical and limbic volumes than LBD

First, we compared relationships between postmortem subcortical/limbic volumes and diagnostic groups adjusting for covariates and multiple comparisons (Figure 3A, Table S1). Donors with AD had significantly lower volumes than those with LBD in all subcortical and limbic regions studied (p < 0.05), except caudate and globus pallidus. Donors with FTLD‐tau and FTLD‐TDP had lower postmortem volumes than those with LBD across all subcortical and limbic regions queried (p < 0.05). Donors with FTLD‐TDP displayed the lowest volumes across the groups and had significantly worse volume loss than those with AD for caudate, putamen, and nucleus accumbens (p < 0.05). When we split the FTLD‐tau group into 3R tauopathy and 4R tauopathy, we found significant relationships of worse subcortical volume loss compared to AD in these same regions (Figure S1). Postmortem findings paralleled antemortem imaging (Figure S2). In comparisons that covaried for age, sex, education, and antemortem imaging interval, we similarly found that patients with LBD had higher antemortem volumes across subcortical and limbic regions than AD, FTLD‐tau, and FTLD‐TDP.

FIGURE 3.

Postmortem limbic and subcortical volumes across neuropathological groups. (A) Boxplots show ICV‐normalized postmortem single‐hemisphere volumes for the seven subcortical and limbic regions across four primary diagnostic categories: Alzheimer's disease (AD), Lewy body disease (LBD), frontotemporal lobar degeneration with TDP‐43 pathology (FTLD‐TDP), and FTLD‐tau. (B) Boxplots compare ICV‐normalized postmortem volumes across regions for primary AD and secondary LBD (n = 29) versus primary LBD and secondary AD (n = 21) in donors with both AD and LBD pathology. Pairwise differences between groups were assessed using likelihood‐ratio tests adjusting for age at death, sex, education, and postmortem interval. Statistically significant pairwise comparisons with multiple test adjustment are denoted by *p < 0.05, **p < 0.01, ***p < 0.001. In each boxplot, the center line indicates median, box edges represent 25th to 75th percentiles inter‐quantile range (IQR), whiskers extend to 1.5 × IQR, and overlaid points correspond to individuals. Detailed statistical results are reported in Table S1.

3.3. Primary AD is linked to worse subcortical and cortical atrophy than primary LBD in donors with both AD and LBD copathologies

Since AD and LBD pathologies often coexist, we aimed to better delineate the relationships between volume loss and primary AD versus primary LBD pathology (Figure 3, 5). Our autopsy cohort included 29 donors with primary AD and secondary LBD and 21 donors with primary LBD and secondary AD. Similar to the omnibus findings, donors with primary AD and secondary LBD had significantly lower volumes than donors with primary LBD and secondary AD in all subcortical and limbic regions studied (p < 0.05), except globus pallidus and caudate. We also assessed postmortem cortical thicknesses from purple‐mri. Across almost all cortical regions except medial occipital cortex, sensorimotor cortex, superior parietal lobule, and pars orbitalis, donors with primary AD and secondary LBD had lower postmortem cortical thickness than donors with primary LBD and secondary AD (Figure S3).

FIGURE 5.

Primary pathology model with ordinary least‐squares model regression analysis between postmortem volumes and primary neuropathological burden. Heatmap shows standardized β coefficients from regression models on the relation between normalized postmortem subcortical and limbic matter volumes and corresponding regional predominant molecular pathology measures within each diagnostic group: AD (p‐tau), LBD (α‐synuclein), FTLD‐TDP (TDP‐43), and FTLD‐tau (p‐tau). Correlations were adjusted for age at death, sex, postmortem interval, and years of education. For the OLS regression analysis, an independent linear model was fit for each structure within each diagnostic group, including the region's predominant pathology measure together with the same covariates and multiple‐comparisons adjustment. Each box displays standardized β coefficients from OLS models predicting each structure's volume from its predominant regional pathology measure while adjusting for the same covariates with asterisks denoting significance after false discovery rate correction applied separately within each diagnostic group (*p < 0.05 [and outlined], **p < 0.01, ***p < 0.001). Pathology burden ratings were not available for the nucleus accumbens. Table S2 tabulates detailed statistical results respectively.

Comparing cortical thickness across the four main disease groups, we found LBD also had less cortical atrophy than FTLD‐tau and FTLD‐TDP in nearly all 35 cortical regions (Figure S4). Intriguingly, we found that superior temporal gyrus and insula had significantly lower postmortem thickness in FTLD‐TDP than FTLD‐tau, suggesting these regions may help differentiate these clinical entities. Overall, these results strengthen the notion that AD more so than LBD pathology drives cortical, subcortical, and limbic atrophy, even in patients with mixed AD and LBD.

3.4. Subcortical and limbic volume loss tracks with cortical thinning differently in each disease

Since volume changes occurred across several subcortical and limbic structures in a fairly similar manner for these diseases, we first assessed how cortical thinning correlated with a composite subcortical and limbic meta‐region (composed of hippocampus, amygdala, nucleus accumbens, thalamus, caudate, putamen, and pallidum) (Figure 4A). Composite subcortical and limbic volume loss indicated positive pathology–volume correlations with thinning of nearly all cortical regions in donors with AD (p < 0.05). However, donors with FTLD‐TDP and FTLD‐tau exhibited positive correlations limited to frontal regions (such as superior frontal gyrus and cingulate gyrus) and the inferior parietal lobule (p < 0.05). Donors with LBD had less atrophy such that associations between cortical thinning and composite subcortical/limbic volume loss were not found. Separately evaluating the 3R and 4R tauopathy subgroups individually also did not show significant correlations between subcortical volume and cortical thinning (Figure S5).

FIGURE 4.

Associations between Desikan‐Killiany‐Tourville atlas‐based regional mean thickness and weighted subcortical and limbic volumes. Disease‐specific regional correlations between cortical thickness and (A) a meta‐region of subcortical and limbic structures computed by a volume‐weighted, ICV‐normalized composite (including caudate, putamen, pallidum, thalamus, nucleus accumbens, hippocampus, and amygdala), (B) subcortical structures only (caudate, putamen, pallidum, thalamus, nucleus accumbens), and (C) limbic structures only (hippocampus and amygdala). For each neuropathological group, partial Spearman correlations were calculated between mean ROI‐level Desikan‐Killiany‐Tourville cortical thickness and normalized weighted volume, adjusting for age at death, sex, education, and postmortem interval. Partial Spearman coefficients (ρ) are reported with regions outlined in black indicating p < 0.05.

To ascertain whether the correspondence between cortical thinning and total subcortical/limbic volume loss was largely driven by variation in the medial temporal lobe, we then compared regional cortical thickness with volume for a second subcortical‐only meta‐region that included only subcortical regions (caudate, putamen, accumbens, thalamus, pallidum) (Figure 4B). This modified subcortical composite volume showed significant correlations with most frontal, parietal, and temporal regions in AD (p < 0.05) and frontal and lateral parietal regions in FTLD‐TDP and FTLD‐tau (p < 0.05). LBD did not show cortical–subcortical associations were not significant.

Closing the loop, we tested how cortical thinning patterns change with limbic volume loss across diseases with a limbic‐only region (hippocampus and amygdala) (Figure 4C). We recapitulated the finding of diffuse cortical–limbic associations in AD (p < 0.05). Interestingly, compared to the subcortical volumes, donors with LBD had strong associations between frontal and parieto‐occipital thinning and limbic volume loss (p < 0.05). Moreover, we observed robust significant cortico‐limbic morphometric associations in FTLD‐tau (p < 0.05) but not FTLD‐TDP, juxtaposed to the significant cortico‐subcortical associations in FTLD‐TDP.

3.5. Primary neuropathology is associated with subcortical and limbic volume loss

Next, we assessed how semi‐quantitative primary pathology burden is associated with subcortical and limbic volumes across regions in their corresponding disease group (p‐tau in AD and FTLD‐tau, α‐synuclein in LBD, and TDP‐43 in FTLD‐TDP) with regression models with covariates (Figure 5, Figure S6, Table S2). In AD, we found that regional p‐tau burden had significant negative correlations with volumes for both hippocampus (p = 0.002) and putamen (p = 0.002) after multiple‐comparison adjustments. In LBD, thalamic α‐synuclein pathology correlated with MRI volumes of thalamus, putamen, and pallidum (all p < 0.02), after multiple‐test adjustment (Figure S7). For FTLD‐tau, regional p‐tau ratings indicated inverse correlations with hippocampus and amygdala volumes but did not withstand multiple‐comparisons correction. For FTLD‐TDP, TDP‐43 severity had an inverse relationship with putamen volume but not after multiple testing correction. Additionally, 3R and 4R tauopathy subgroups did not present significant correlations between semi‐quantitative p‐tau measures and subcortical/limbic volumes (Figure S8).

3.6. Polypathological models suggest primary pathology mostly drives subcortical and limbic volume loss

Approximately 82% (108/132) of donors had secondary neuropathological diagnoses, so we assessed a polypathology regression with multiple pathologies (tau, α‐synuclein, and TDP‐43) for each primary diagnostic group, ³ also correcting for covariates (Figure 6, Table S3). For the AD polypathology models, hippocampal p‐tau pathology had significant correlations with hippocampal volume (p = 0.007), which did not remain after multiple‐comparison adjustments. Donors with AD also exhibited partial correlations between amygdala volumes and TDP‐43 burden that did not survive multiple testing corrections. Regional p‐tau burden associated with hippocampal and amygdala volume in FTLD‐tau and regional α‐synuclein pathology associated with thalamic volume in LBD, though these correlations did not withstand multiple testing correction. Similarly to results from the primary pathology models, for FTLD‐TDP, putamen TDP‐43 severity and volume had a trend that did not survive multiple testing correction. Distinguishing between 3R and 4R tauopathy in the polypathology model highlighted trends in lower pallidum volume with greater regional p‐tau and α‐synuclein involvement, though these did not remain after multiple testing correction (Figure S9). These findings were similar in polypathology models with covariates of neuronal loss and gliosis (Figure S10).

FIGURE 6.

Polypathology ordinary least‐squares (OLS) models for structure–pathology relationships. Heatmaps show standardized β coefficients from OLS models quantifying the association between regional postmortem MRI volumes (rows) and multiple semi‐quantitative pathology ratings (columns) across the four diagnostic groups (AD, LBD, FTLD‐TDP, FTLD‐tau). Each cell reports the standardized effect size (β) and the corresponding uncorrected p value; boxes highlight uncorrected p < 0.05, and asterisks denote significance after false discovery rate correction using the Benjamini–Hochberg procedure applied separately within each diagnostic group across all structures and pathology predictors (*p < 0.05 (also outlined), **p < 0.01, ***p < 0.001). All models included age at death, sex, years of education, and postmortem interval as covariates. Pathology burden ratings were not available for the nucleus accumbens. Table S3 shows detailed statistical results. Note: Some predictors appear as “Not applicable” because several histology markers exhibited minimal or zero variability within disease groups. Predictors lacking sufficient within‐group variance cannot yield interpretable regression estimates and may otherwise produce β values near zero with spuriously small p‐values. Such predictors were removed prior to model fitting and excluded from FDR correction.

3.7. Histologic neuronal loss and gliosis mediate relationships between subcortical and limbic pathology and volume loss

Because neurodegeneration is classically associated with neuron loss and glial proliferation, ²⁷ we compared gross subcortical and limbic volumes with semi‐quantitative histologic measures of neuronal loss and gliosis accounting for covariates (Figures 7A and S11, Table S4). In AD, microscopic neuronal loss ratings significantly corresponded to macroscopic volume loss in amygdala (p = 0.004), caudate (p = 0.029), and putamen (p = 0.020). Similarly in AD, local gliosis scores correlated with volume loss in amygdala (p = 0.025). In FTLD‐tau, significant associations were found between microscopic neuronal loss and volume loss in hippocampus (p < 0.001) and amygdala (p < 0.001). In parallel, for FTLD‐tau, gliosis was associated with volume loss in hippocampus (p < 0.001) and amygdala (p < 0.001). This finding did not remain after splitting FTLD‐tau into 3R and 4R tauopathy subgroups (Figure S12). Donors with LBD and FTLD‐TDP had no significant relationships between these microscopic and macroscopic measures.

FIGURE 7.

Structure‐histology relationships and mediation analyses for gliosis and neuronal loss. (A) Correlations between postmortem MRI volumes and regional mediation markers were calculated between postmortem subcortical and limbic MRI volumes and semi‐quantitative ratings of gliosis and neuronal loss for each group. All correlations were adjusted for age at death, sex, postmortem interval, and education. Heatmap colors reflect slope (β) with corresponding uncorrected p value (in parentheses, boxes for p < 0.05) with asterisks denoting significance after FDR correction applied within group (*p < 0.05, **p < 0.01, ***p < 0.001). (B) Mediation analysis with the bar plots showing effect sizes for direct (purple) and indirect (green) effects from regional pathology burden to subcortical/limbic volume, mediated by gliosis (top row) or neuronal loss (bottom row). Pathology burden ratings were not available for the nucleus accumbens, so this structure is not shown. Tables S4 and S5 tabulate detailed statistical results.

Mediation analyses showed both direct and indirect relationships between subcortical/limbic neuropathology and volume loss, mediated in part by neuronal loss and gliosis, even after adjusting for covariates and multiple comparisons (Figure 7B, Table S5). For FTLD‐tau, the indirect associations between local p‐tau pathology and volume loss were mediated by hippocampal neuronal loss (p < 0.001), hippocampal gliosis (p < 0.001), and amygdala neuronal loss (p < 0.05).

3.8. Cerebrovascular pathology is not associated with subcortical and limbic volume loss

Given that subcortical regions are common sites of ischemic injury, we investigated how microvascular features may relate to deep brain atrophy. Vascular pathology, as measured by global cerebral amyloid angiopathy, atherosclerosis, and arteriolosclerosis scores, did not significantly correlate with subcortical or limbic volumes in groups after multiple testing correction (Figures S13, S14).

4. DISCUSSION

We examined structure–structure and structure–pathology relationships between subcortical, limbic, and cortical regions and p‐tau, α‐synuclein, and TDP‐43 neuropathologies by incorporating postmortem imaging and histopathology. With our novel pipeline, we found that brain donors with FTLD‐tau and FTLD‐TDP had lower postmortem volumes across most subcortical and limbic structures than donors with either AD or LBD. Similarly, striatum and accumbens had lower volume in FTLD‐TDP than in AD. Strikingly, AD had diffuse cortico‐subcortical and cortico‐limbic morphometric associations, while LBD had limited parieto‐occipital cortico‐limbic associations. FTLD‐tau exhibited frontal cortico‐limbic and cortico‐subcortical structural associations, while FTLD‐TDP only showed cortico‐subcortical associations. These findings in FTLD‐tau remained significant after stratifying by 3R and 4R tauopathies. In AD and FTLD‐tau, hippocampal volume loss correlated with p‐tau, neuronal loss, and gliosis measures, but not for TDP‐43 or α‐synuclein. In LBD, regional α‐synuclein was associated with thalamic volume, independent of other pathologies. In FTLD‐TDP, local TDP‐43 had no significant volume associations. Overall, our study demonstrates the utility of postmortem imaging to reveal relationships between deep brain structure and pathologies across neurodegenerative diseases.

We compared postmortem MRI cortical mantle segmentation using three methods. Qualitative evaluation of SuperSynth and NextBrain‐based ³⁷ , ³⁸ , ⁶⁰ , ⁶¹ segmentations revealed cortical gray matter under‐segmentation and over‐segmentation into neighboring gyri or cerebrospinal fluid spaces (Figure S15). These errors likely reflect the challenges of a domain randomization‐based method (SuperSynth) and atlas‐based segmentation (NextBrain) in postmortem MRI, where altered tissue contrast, intensity heterogeneity, and large anatomical deformations are common. In contrast, purple‐mri was developed specifically for postmortem imaging and generated anatomically plausible tissue boundaries that supported reliable cortical surface reconstruction, anatomical parcellation, cortical thickness estimation, and downstream morphometric analyses. For the subcortical and limbic structures’ segmentations, SuperSynth was the most robust.

A prior study applied postmortem 3T MRI with shape analysis and pathology in AD and limbic‐predominant age‐related TDP‐43 encephalopathy (LATE) cohorts to uncover structure–pathology insights. ¹¹ We build on this work with ultra‐high‐resolution 7T MRI and deep learning segmentations of limbic, subcortical, and cortical regions in a diverse cohort with AD, LBD, FTLD‐TDP, and FTLD‐tau. Our cohort also compared copathologies, histologic measurements, and genetic data in donors with sporadic and genetic disease.

Subcortical and limbic volume loss was significantly associated with diffuse cortical atrophy in AD and a more restricted atrophy pattern limited to frontal and temporal regions in FTLD‐TDP and FTLD‐tau. This is consistent with prior studies suggesting that pathologies arise early in specific subcortical and limbic regions and spread through neural circuits to specific cortical regions in amyotrophic lateral sclerosis (ALS‐TDP) and perhaps FTLD‐TDP, ³ , ⁴⁸ , ⁴⁹ but not necessarily in other cases such as FTLD‐tau. ²⁵ , ⁶² , ⁶³ , ⁶⁴ FTLD‐TDP harbored cortico‐subcortical associations, in keeping with evidence of subcortical‐to‐cortical spread, while both tau‐related disorders FTLD‐tau and AD portrayed both cortico‐subcortical and cortico‐limbic morphometric associations. Some of this lack of correlation in FTLD‐TDP may be attributed to a floor effect in hippocampal volumes in FTLD‐TDP compared to FTLD‐tau, the latter of which had greater dynamic range of variation to derive associations. Nevertheless, these findings suggest differing regional patterns involved in misfolded protein propagation specific to TDP‐43 versus p‐tau. LBD demonstrated parieto‐occipital cortico‐limbic associations but not cortico‐subcortical relationships. Since most of our LBD group had a secondary AD neuropathology (21/26), this cortico‐limbic association is concordant with prior studies ³⁹ , ⁴¹ and could represent a mixed AD with LBD pattern of atrophy.

We contrasted histopathological markers and subcortical/limbic volume loss with regression analyses for primary pathology and polypathology models. Power was limited in our polypathology analyses, leading to observed relationships between volume and primary pathology and trends for secondary pathologies that did not survive multiple‐comparison adjustments. Donors with AD had significant partial correlations of p‐tau burden with volumes for hippocampus, amygdala, caudate, and putamen after accounting for covariates. In LBD, thalamic α‐synuclein pathology correlated with all subcortical and limbic volumes. For FTLD‐tau, regional p‐tau burden did not correlate with subcortical or limbic volumes after multiple testing correction. For FTLD‐TDP, TDP‐43 severity scores did not significantly correlate with subcortical and limbic volumes. From modeling regressions of primary pathology and polypathology, the association between hippocampal structure and p‐tau pathology was supported in AD, with a hint of comorbid TDP‐43 pathology contributing to atrophy. The observed relationship between hippocampal shrinkage and p‐tau pathology in AD was mediated by hippocampal neuronal loss and gliosis, but not atherosclerosis or arteriolosclerosis. Although many donors had cerebrovascular pathology, vascular disease may reflect a pathological factor that is not differentially associated with unique subcortical volume loss patterns. These findings support a strong tandem relationship between p‐tau burden and cellular injury (neuron loss, gliosis, and gross volume), ⁴⁷ , ⁵¹ , ⁵⁴ , ⁶⁵ , ⁶⁶ though it is not as clear‐cut for α‐synuclein.

Compared to tau‐related diseases, LBD had less subcortical and limbic volume loss and less robust relationships between cortical and subcortical neurodegeneration. Moreover, thalamic α‐synuclein pathology in LBD correlated with subcortical and limbic volumes in primary pathology models, though this finding did not extend to the polypathology model after multiple‐test adjustment. Indeed, in donors with both AD and LBD pathology, those with primary LBD and secondary AD diagnoses had less atrophy than those with primary AD and secondary LBD diagnoses, across limbic, subcortical, and cortical regions (except for caudate, globus pallidus, calcarine cortex, and a few additional areas). The presence of neuronal α‐synuclein pathology is associated with hypometabolism and neurotransmission changes, but not as much atrophy. ¹⁶ , ¹⁸ , ³⁹ , ⁴¹ , ⁵⁰ This suggests that LBD has less subcortical and limbic volume loss than other diseases, and/or non‐LBD pathology may contribute to volume loss more than α‐synuclein does in LBD, ¹⁷ , ⁵³ , ⁶⁷ though we did not find specific evidence of this in our dataset. Evaluation of the effects of LBD pathology on subcortical structures may require different imaging tools to probe metabolism or synaptic density, as well as different queries of α‐synuclein regional patterns. ⁶⁸

The lack of significant partial correlations between TDP‐43 burden and limbic/subcortical volume in FTLD‐TDP or other groups has several explanations. First, this group may be underpowered. Second, similar studies demonstrated low‐magnitude correlations between TDP‐43 severity and cortical region thickness. ⁸ , ³³ However, some groups found moderate associations between TDP‐43 burden and volume loss. ¹⁹ , ²⁰ , ²¹ , ²³ TDP‐43 load may be less associated with cortical and limbic volume loss in primary pathology in FTLD‐TDP than in LATE or comorbid AD with LATE. ⁵² While the lack of evidence may be due to the absence of biological effects, additional technical sources can contribute, such as sampling error. This is relevant for TDP‐43 pathology, which has low dynamic range and is sparse even in advanced stages. To this point, intraneuronal TDP‐43 inclusions disappear once affected neurons die, unlike ghost tau tangles that remain after the neuron disintegrates.

This study had several limitations and next steps. Although the cohort included 132 donors, some groups were likely underpowered, particularly FTLD‐TDP and polypathology models in which volume–pathology relationships were subdivided by diagnosis. This reduced our ability to detect additional polypathology associations that prior studies tested in omnibus cohorts. ³ Pathology and imaging were obtained from different hemispheres, a common practice in postmortem studies, ⁶ , ¹² , ³³ , ³⁴ , ³⁵ , ⁶⁹ because many neuropathologies are bilateral, though not universally so. ¹⁹ , ⁷⁰ In AD, prior work found largely similar MTL thickness correlations with ipsilateral and contralateral p‐tau scores. ⁸ Any unmeasured hemispheric asymmetry in pathology or volume would reduce power, likely most affecting FTLD‐tau. Another limitation of the study is its reliance on semi‐quantitative pathology scores, which are subjective, variable across readers, and may not scale linearly with true burden. Cerebrovascular measures were whole‐brain measures and may not capture local subcortical effects. We are now obtaining same‐hemisphere immunohistochemistry and developing machine‐learning‐based quantitative ratings ⁴⁶ to validate these findings. Monoaminergic and cholinergic nuclei are dysregulated in neurodegenerative disease, ¹ including basal forebrain, ⁶⁹ , ⁷¹ substantia nigra, ⁷² and locus coeruleus, ⁷³ so segmentation of additional subcortical regions is planned. Future work should examine more granular clinicopathological entities, including LATE, ²⁸ , ³⁰ FTLD‐TDP type C with annexin A11 co‐aggregates, ⁴⁵ , ⁷⁴ and genetic variants. We will investigate cortico‐limbic versus cortico‐subcortical associations in FTLD‐tau and FTLD‐TDP and local shape deformations of subcortical and limbic structures.

Overall, our study provides a comprehensive postmortem framework for linking brain tissue structure to pathological markers of neurodegenerative disease across primary and secondary diagnoses, brain regions, and antemortem versus postmortem imaging. We found evidence of limbic, subcortical, and cortical volume loss in primary AD over primary LBD, even among donors with comorbid AD and LBD. We also demonstrated selective cortico‐limbic associations in LBD and cortico‐subcortical associations in FTLD‐TDP, compared with more widespread associations in AD and, to a lesser extent, FTLD‐tau. Regional p‐tau burden was associated with atrophy, partly mediated by neuron loss and gliosis. Together, these findings advance the characterization of the pathological basis of structural imaging biomarkers and neurodegenerative disease mechanisms.

CONFLICT OF INTEREST STATEMENT

J.A.V. reports honoraria/fees from the International Parkinson and Movement Disorder Society and the University City Science Center outside this work. I.M.N. reports fees from Biogen and Eisai outside this work. S.R.D. reports grants/fees from Nia Therapeutics and Rancho Biosciences outside this work. D.J.I. receives funding to the institution for clinical trials by Alector, CervoMed, Denali, PassageBio, and Prevail outside this work. D.A.W. reports grants/fees from Beckman Coulter, Biogen, Eli Lilly, Functional Neuromodulation, GE Healthcare, GSK, and Qynapse outside of this work. No additional disclosures. Author disclosures are available in the Supporting Information.

CONSENT STATEMENT

Human brain specimens were obtained in accordance with local laws and regulations and include informed consent from next of kin at time of death or, where possible, pre‐consent during life.

Supporting information

Supporting Information

Supporting Information