Uncovering distinct spatiotemporal trajectories of T‐N mismatch subtypes with likely co‐pathology in Alzheimer's disease using event‐based modeling

Xueying Lyu; Christopher A Brown; Michael Tran Duong; Emma P Fischer; Emily McGrew; David J Irwin; Bradford C Dickerson; Ilya M Nasrallah; Russell Takeshi Shinohara; Alexandra L Young; Paul A Yushkevich; Sandhitsu R Das; David A Wolk; for the Alzheimer's Disease Neuroimaging Initiative

doi:10.1002/alz.71474

. 2026 May 21;22(5):e71474. doi: 10.1002/alz.71474

Uncovering distinct spatiotemporal trajectories of T‐N mismatch subtypes with likely co‐pathology in Alzheimer's disease using event‐based modeling

Xueying Lyu ¹, Christopher A Brown ², Michael Tran Duong ³, Emma P Fischer ⁴, Emily McGrew ³, David J Irwin ², Bradford C Dickerson ⁵, Ilya M Nasrallah ³, Russell Takeshi Shinohara ⁶, Alexandra L Young ⁷, Paul A Yushkevich ³, Sandhitsu R Das ^2,^✉, David A Wolk ^2,^✉; for the Alzheimer's Disease Neuroimaging Initiative

PMCID: PMC13239684 PMID: 42168780

Abstract

INTRODUCTION

In Alzheimer's disease (AD), tau–neurodegeneration (T–N) mismatch has been proposed to reflect non‐AD processes such as transactive response DNA binding protein 43 kDa and vascular disease. We aimed to characterize the spatiotemporal trajectories of T–N mismatch that may reflect non‐AD progression.

METHODS

We performed T–N regression on 710 Alzheimer's Disease Neuroimaging Initiative participants using cortical thickness and 18F‐flortaucipir uptake across 20 cortical regions. SuStaIn, a data‐driven phenotype discovery and staging algorithm, was applied to standardized T–N residuals in canonical (N∼T) and vulnerable (N > T) cases.

RESULTS

SuStaIn identified three vulnerable subtypes with distinct N > T progression patterns. The posterior and anterior subtypes displayed different, but progressively diffuse mismatch patterns, while the limbic subtype exhibited temporal–limbic progression. Subtypes and SuStaIn stages were associated with distinct clinical features. Their longitudinal trajectories aligned with SuStaIn inferred progression.

DISCUSSION

Findings support that T–N mismatch progression captures specific co‐pathological processes.

Keywords: aging, Alzheimer's disease, co‐pathologies, multi‐modality imaging, neurodegeneration, progression, tau

Highlights

Applying tau–neurodegeneration (T–N) mismatch to SuStaIn identified three vulnerable subtypes characterized by greater neurodegeneration than expected given tau (N > T), each with distinct spatiotemporal mismatch patterns.
The anterior and posterior vulnerable subtypes showed convergence toward a more diffuse pattern of T–N mismatch. While both subtypes exhibited imaging features consistent with vascular involvement, the anterior subtype was more strongly associated with greater white matter hyperintensity burden—commonly used as a marker of arteriosclerotic small‐vessel disease—whereas the posterior subtype showed stronger associations with lobar microbleeds, which are suggestive of cerebral amyloid angiopathy.
The limbic vulnerable subtype demonstrated temporal–limbic mismatch progression and linked to clinical and imaging features suggestive of limbic‐predominant age‐related transactive response DNA binding protein 43 kDa encephalopathy–neuropathologic change (LATE‐NC).
SuStaIn‐inferenced mismatch stages aligned with observed longitudinal trajectories, supporting their biological validity.

1. BACKGROUND

Amyloid plaques and tau neurofibrillary tangles (NFTs) are the two primary pathologies observed in Alzheimer's disease (AD). ¹ , ² Post mortem ³ , ⁴ , ⁵ and ante mortem imaging ⁶ , ⁷ studies indicate that NFTs are more closely associated with downstream neurodegeneration compared to amyloid plaques. ⁶ , ⁷ , ⁸ , ⁹ , ¹⁰ Consequently, the pattern and degree of neurodegeneration due to AD is largely attributed to the local presence and severity of tau pathology.

Nevertheless, concomitant pathologies, ¹¹ , ¹² , ¹³ , ¹⁴ such as cerebrovascular disease and/or other degenerative pathologies, as well as factors that confer resilience, ¹⁵ , ¹⁶ , ¹⁷ , ¹⁸ are common in patients with AD. These non‐AD factors also impact neurodegeneration and cognitive decline. The presence of such heterogeneity and the lack of well‐established markers for some non‐AD factors make it challenging to study the influence of these factors on the effectiveness and risk of therapies in the context of clinical trials and practice.

Given the non‐specific nature of brain atrophy regarding underlying pathology, in our previous work ¹⁹ , ²⁰ , ²¹ we explored deviation from the normative relationship between tau (T) and neurodegeneration (N), described as a mismatch, as an approach for uncovering vulnerabilities and resilience in AD patients. We hypothesized that this T–N mismatch was driven by underlying non‐AD factors. Specifically, participants with the least deviation from the normative T–N relationship are indicative of the typical AD pattern, referred to as the “canonical” T–N relationship. However, patients with disproportionally more atrophy given tau (N > T) may be vulnerable due to presence of co‐pathologies, whereas those with less atrophy than expected (N < T) may exhibit features consistent with resilience potentially due to protective factors.

Our previous findings ¹⁹ , ²⁰ suggest T–N mismatch can define groups enriched in co‐pathologies with AD. We identified distinct data‐driven T–N phenotypes by clustering regional measures of T–N mismatch derived from residuals in T–N regression across brain regions. We found two vulnerable (N > T) groups: one characterized by a diffuse vulnerable pattern and the other by a limbic vulnerable pattern. Consistent with our hypothesis, they appeared enriched in co‐pathologies. We found that the diffuse vulnerable group was linked to vascular disease‐related features. Based on a replication study in a cohort (n = 98) with autopsy‐confirmed diagnoses, ²⁰ the limbic vulnerable group was associated with greater transactive response DNA binding protein 43 kDa (TDP‐43) severity, supporting enrichment of limbic‐predominant age‐related TDP‐43 encephalopathy (LATE) ²² in these cases.

However, the prior analysis was limited by its cross‐sectional and discrete nature, whereas we expect there to be a continuum of resilience and vulnerability. Furthermore, given the lack of well‐validated in vivo biomarkers for co‐pathologies, we currently have limited understanding of their in vivo spatial effects on the pattern of neurodegeneration over time in the context of AD. As we enter an era of precision medicine for dementia led by AD‐specific therapeutics, including amyloid‐targeting immunotherapies, ²³ it is important to identify not only the presence of non‐AD factors, but also their pattern and severity of progression, which may influence therapeutic outcomes.

Therefore, in this work, we aim to leverage T–N mismatch to identify distinct spatial–temporal progression patterns of non‐AD pathologies. We hypothesize that data‐driven T–N subtype staging based on T–N mismatch in vulnerable cases will reflect distinct co‐pathology progression patterns rather than AD progression. Given limited availability of longitudinal data with magnetic resonance imaging (MRI) and tau positron emission tomography (PET) imaging, here we used the SuStaIn algorithm ²⁴ for modeling. SuStaIn provides inference on imaging subtypes along with pseudo‐staging based on only cross‐sectional data. SuStaIn was applied to individuals based on their T–N mismatch which span from low to elevated T–N mismatch (N > T), with the expectation that cases with greater T–N mismatch (N > T) would be enriched for co‐pathologies. Because SuStaIn models only unidirectional progression, individuals showing resilient T–N patterns (N < T) were excluded to best capture trajectories associated with vulnerability and co‐pathology. We then linked the inferred SuStaIn stages to markers suggestive of different pathologies, hypothesizing that later stages would correspond to greater pathological burden. Finally, we validated the SuStaIn pseudo‐staging by examining longitudinal progression of T–N mismatch in a subset of individuals with available follow‐up imaging.

RESEARCH IN CONTEXT

Systematic review: We reviewed the literature on PubMed focusing on applications of SuStaIn in Alzheimer's disease (AD), tau–neurodegeneration (T–N) mismatch, and AD‐related co‐pathologies. Prior work has largely applied SuStaIn to model tau or atrophy trajectories in isolation to characterize AD progression. T–N mismatch has been shown to identify non‐AD processes in AD patients, such as transactive response DNA binding protein 43 kDa (TDP‐43) and vascular disease. However, the spatiotemporal progression of T–N mismatch has not previously been investigated.
Interpretation: Our study demonstrates that applying SuStaIn to regional T–N mismatch reveals three distinct vulnerable subtypes with different spatiotemporal trajectories. These subtypes exhibited differential associations with imaging features suggestive of non‐AD co‐pathological processes, including arteriosclerotic vascular disease, cerebral amyloid angiopathy, and limbic‐predominant age‐related TDP‐43 encephalopathy–neuropathologic change. These findings extend prior cross‐sectional work by reconstructing model‐inferred T–N mismatch progression patterns that are consistent with observed longitudinal changes.
Future directions: Future research should validate these T–N mismatch progressions in more diverse cohorts and with post mortem validation of underlying pathologies. Further work is also needed to assess the prognostic value and relevance of T–N mismatch for predicting therapeutic responses, particularly to amyloid‐targeting therapy.

2. METHODS

2.1. Participants

Data used in the preparation of this article were obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). The ADNI was launched in 2003 as a public–private partnership, led by Principal Investigator Michael W. Weiner, MD. The primary goal of ADNI has been to test whether serial MRI, PET, other biological markers, and clinical and neuropsychological assessment can be combined to measure the progression of mild cognitive impairment (MCI) and early AD.

We included 710 individuals from ADNI who had at least one ¹⁸F‐flortaucipir PET scan and one T1‐weighted structural MRI scan < 1 year apart. The average time difference between the tau PET scan and MRI was 1.7 (± 2.0) months. The characteristics of the cohort is described in Table S1 in supporting information. Of the individuals included in the study, 68.0% (n = 483) had only one scan available. The remaining 31.9% (n = 227) had longitudinal data, among whom 64.3% had two scans, 33.0% had three scans, 2.2% had four scans, and 0.4% had five scans. Among participants with longitudinal scans, the average interval between the baseline and last scan was 1.87 (± 0.93) years, with a maximum observed interval of 5.56 years.

2.2. Image acquisition and processing

2.2.1. Image acquisition

We processed T1‐weighted MRI and tau PET (¹⁸F‐flortaucipir) scans to obtain cortical thickness and tau standardized uptake value ratio (SUVR) values for 100 gray matter regions of interest (ROIs). The methods for image acquisition and processing have been detailed previously. ¹⁹ , ²⁰ In summary, the T1‐weighted MRI scans acquired by ADNI had voxel resolutions of either 1.0 × 1.0 × 1.0 mm³ or 1.0 × 1.0 × 1.2 mm³, with 73.9% of scans having a resolution of 1.0 × 1.0 × 1.0 mm³. Fluid‐attenuated inversion recovery MRI scans were also acquired during the same session as the T1‐weighted MRI with spatial resolution varying according to the ADNI protocol. The PET images, initially with variable resolutions, were standardized to a full‐width at half maximum (FWHM) resolution of 8 mm³.

2.2.2. MRI processing

The T1‐weighted MRI scans were processed using the Advanced Normalization Tools (ANTs) cortical thickness pipeline, ²⁵ which includes steps for intensity inhomogeneity correction and tissue segmentation. The scans were then parcellated into cerebellar, cortical, and subcortical ROIs using a multi‐atlas segmentation method. ²⁶ Volumetric thickness maps for each subject were generated through the DiReCT cortical thickness estimation method. ²⁷ The thickness for each ROI was determined by averaging the thickness values across voxels within the gray matter ROIs. We segmented white matter hyperintensities (WMHs) using a nn‐UNet ²⁸ model described elsewhere. ²⁹

Medial temporal lobe subregions were segmented from T1‐weighted MRI scans using the Automated Segmentation of Hippocampal Subfields (ASHS) ³⁰ pipeline. The segmented anatomical subregions included the anterior and posterior hippocampus, amygdala, entorhinal cortex, parahippocampal cortex, BA35, and BA36. Intracranial volume (ICV) was also estimated using ASHS. Thickness estimation was then performed with the Cortical Reconstruction for ASHS (CRASHS) ³¹ pipeline.

2.2.3. PET image processing

ADNI provides post‐processed PET images by averaging co‐registered individual frames. These images were aligned with participants’ T1‐weighted MRI scans using ANTs, ³² with parameters including Mattes mutual information image similarity metric (weight = 1, 32 bins), a rigid transformation model (gradient step = 0.2), smoothing levels of 4 × 2 × 0, and a shrink factor of 4 × 2 × 1. The MRI parcellated ROIs were then mapped to PET space. Mean PET tracer uptake in inferior cerebellar gray matter (for ¹⁸F‐flortaucipir) or in whole cerebellum (for ¹⁸F‐florbetapir or ¹⁸F‐florbetaben) served as the reference to generate SUVR maps. TauMaX, ³³ a global measurement of tau severity reflecting both the magnitude and spatial extent of tau pathology across the brain, was calculated for each subject from their tau PET imaging as previously described.

We processed ¹⁸F‐florbetaben or ¹⁸F‐florbetapir amyloid PET scans to assess amyloid status, based on a composite ROI measure. ³⁴ A positive amyloid scan (A+) was defined by SUVR thresholds of ≥ 1.11 for ¹⁸F‐florbetapir and ≥ 1.08 for ¹⁸F‐florbetaben. ³⁵

2.3. T–N mismatch subtype stratification and trajectory modeling

We calculated regional T–N mismatch for each of the 100 regions across the whole brain and then used the regional T–N mismatch from selected ROIs to explore subtype‐specific trajectories with SuStaIn, a phenotype discovery and stage inference algorithm. This process is summarized in Figure 1 and detailed description is outlined below.

Workflow of SuStaIn modeling on T–N mismatch for subtype stratification and staging. N, neurodegeneration; ROI, region of interest; SD, standard deviation; SUVR, standardized uptake value ratio; T, tau.

2.3.1. T–N mismatch modeling

We modeled regional T–N mismatch using the same approach as in our previous studies. ¹⁹ , ²⁰ The relationship between regional tau SUVR and gray matter cortical thickness was analyzed in each ROI using robust linear regression with a bi‐square weighting function, based on 710 available scan pairs. Tau SUVR values were natural log‐transformed for modeling. We used the resulting regional residuals to represent the T–N mismatch.

2.3.2. SuStaIn modeling

We applied SuStaIn ²⁵ for modeling. Due to known limitations in scaling the SuStaIn model to large numbers of ROIs, we included T–N residuals from 10 bilateral ROIs (20 ROIs) that encompass multiple Braak stages. ³⁶ These selected bilateral ROIs included entorhinal cortex, inferior temporal gyrus, parahippocampus, temporal pole, superior temporal gyrus, medial frontal cortex, anterior cingulate gyrus, middle frontal gyrus, precuneus, and angular gyrus (Figure 1).

To model subtypes and the progression of co‐pathologies in AD, we restricted our cohort to individuals whose regional T–N mismatch patterns were consistent with canonical (N∼T) or vulnerable (N > T) relationships, based on T–N regressions from baseline cross‐sectional scan pairs. Specifically, we excluded scan pairs in which any of the 20 selected regions showed a T–N positive residual > 1.5 standard deviations, reflecting disproportionately preserved cortical thickness relative to tau burden. Exclusion was based on individual scan‐pair characteristics rather than subject‐level classification. Nevertheless, excluded individuals often exhibited multiple regions exceeding this threshold, consistent with an overall resilient phenotype. After applying this criterion, 374 individuals remained, reflecting canonical (N∼T) and vulnerable (N > T) patterns. To align the directionality of T–N mismatch with SuStaIn's assumption of monotonically increasing biomarker values, the residuals were sign‐flipped after exclusion, such that higher residual values indicated greater vulnerability (i.e., greater cortical thinning relative to tau burden). The sign‐flipped residuals were standardized across participants and regions and used as input features for SuStaIn modeling.

SuStaIn was then applied to these standardized residuals across the 20 selected ROIs. We chose z scores of 0.25, 0.5, 1.0, and 1.5 as the severity score cut‐off points for all ROIs, with a maximum z score of 2.0, which approximates the 99th percentile of patients for each region. To ensure a robust and converged solution, the model was initialized with 25 different starting points and run for 100,000 iterations using Markov chain Monte Carlo (MCMC) sampling. To account for the non‐deterministic nature of the SuStaIn algorithm, we ran the model on the same input data for 10 repeated experiments. We then selected the model that had the highest adjusted Rand index ³⁷ (ARI) among all 10 runs (Figure S1A in supporting information) relative to the consensus solution, which enhances the robustness and reproducibility of the SuStaIn inference.

To identify the optimal number of subtypes, we performed 10‐fold cross‐validation using the cross‐validation information criterion (CVIC). We allowed the model to explore up to four subtypes. The model with three subtypes was selected as the optimal solution (Figure S1B) as it demonstrated the highest out‐of‐sample likelihood, which corresponds to the lowest CVIC value, across all folds. Additionally, individuals with a SuStaIn stage of zero were assigned to subtype 0, referred to as the “canonical” subtype.

2.4. Imaging features

2.4.1. Imaging feature suggestive of LATE neuropathologic change

A previous post mortem study ³⁸ of medial temporal lobe comorbidities suggested that the entorhinal‐to‐parahippocampal cortex thickness ratio (ERC/PHC) was reduced in AD patients with TDP‐43 co‐pathology. Based on this observation, we used the ERC/PHC thickness ratio as an imaging surrogate for LATE neuropathologic change (LATE‐NC) and compared this measure across T–N subtypes in the present study.

Furthermore, given that LATE‐NC is associated with severe hippocampal atrophy, as evidenced by previous in vivo ²² , ³⁹ and post mortem studies, ⁴⁰ we also included hippocampal volume as a complementary imaging feature suggestive of LATE‐NC.

2.4.2. Microbleeds

Microbleed counts in both lobar and deep brain regions were obtained from the ADNI database. ⁴¹ , ⁴² Among 374 individuals, 355 have microbleeds data available. For each participant, cerebral microbleeds were quantified separately in lobar and deep/infratentorial regions. Lobar regions included the frontal, temporal, parietal, occipital lobes, and corpus callosum, whereas deep regions included the basal ganglia, thalamus, brainstem, and cerebellum. We categorized the number of microbleeds based on Boston criteria ⁴³ as 0 (none), 1 to 3 (mild), and > 3 (moderate‐to‐severe) for later statistical analyses.

2.5. Domain‐specific cognition

Cognitive domain scores were obtained from the ADNI database, specifically from the Alzheimer's Disease Sequencing Project Phenotype Harmonization Consortium – Derivation of Cognitive Composite Scores. ⁴⁴ This dataset provides harmonized cognitive z scores for domain‐specific cognitive performance across multiple cohorts. The scores provide measures of memory, executive function, language, and visuospatial functions. The visuospatial domain was excluded from our analyses due to floor effects.

2.6. Assessment of longitudinal stability

Among the 374 patients at baseline, 114 had longitudinal tau PET and MRI scan pairs. For each of these individuals, we selected their baseline visit and the subsequent visit closest in time for longitudinal analysis. The average follow‐up period was 1.95 ± 0.97 years for MRI and 1.92 ± 0.99 years for tau PET. The average interval between tau PET and MRI scan for each subject was 1.56 ± 1.9 months.

To evaluate subtype stability, we first obtained longitudinal T–N residuals by applying the linear regression model fitted on baseline scans to the longitudinal scan pairs. These residuals were then standardized to a z score using the mean and standard deviation from the baseline data. The resulting standardized residuals from the 20 selected ROIs were then applied to the SuStaIn model, which was trained on the baseline data, for inferring their T–N subtypes and SuStaIn stages.

We assessed the longitudinal stability of the subtypes by constructing a confusion matrix that compared each individual's subtype classification at baseline to their classification at the next follow‐up visit. The longitudinal stability was evaluated by calculating the proportion of individuals who maintained their subtype assignment between baseline and follow‐up. Biologically plausible transitions were also assessed, which were defined as either remaining in the same subtype or changing from the canonical to a vulnerable subtype. We further examined whether individuals progressed to a more advanced stage within their assigned subtype over time.

2.7. Statistical analysis

2.7.1. General statistical analyses

We conducted two main sets of analyses: (1) comparisons of demographic, clinical, and imaging characteristics across SuStaIn‐derived subtypes, and (2) associations between SuStaIn stage and clinical or imaging measures. Continuous variables were compared between subtypes using linear regression models adjusted for relevant covariates listed in the figures or tables. Ordinal variables (e.g., sex, cognitively unimpaired [CU]/MCI/dementia status, amyloid status, microbleeds) were compared using Kruskal–Wallis ⁴⁵ tests with Bonferroni correction for multiple comparisons. Associations between SuStaIn stage and clinical characteristics were assessed using Spearman correlation, adjusting for relevant covariates consistent with the group comparisons. Bonferroni correction was applied to all between‐group and association analyses. Statistical analyses were performed in R (v4.5) and SPSS (v28).

2.7.2. Longitudinal T–N residual and cognitive changes

We analyzed longitudinal changes in regional T–N residuals using linear mixed‐effects model, ⁴⁶ with T–N residuals as the dependent variable. This analysis included 114 patients who had longitudinal tau PET and MRI scan pairs, contributing a mean of 2.40 ± 0.58 paired T–N residual observations. The average follow‐up period was 1.96 ± 0.98 years for MRI and 1.93 ± 0.99 years for tau PET, with an average interval between paired MRI and tau PET scans of 1.05 ± 1.37 months. Fixed effects included time × subtype interaction, and covariates (global tau, age, sex, and interval between tau and MRI scans), with a random intercept to account for repeated measures. Longitudinal cognitive trajectories were similarly analyzed with cognition as the dependent variable, adjusting for age, sex, global tau, amyloid status, and years of education. Analyses were restricted to observations within ± 5 years of baseline. Group differences in rates of change were assessed via the time × subtype interaction term. All statistical tests were two sided.

3. RESULTS

3.1. SuStaIn uncovered distinct T–N spatial progression patterns across T–N subtypes

We applied SuStaIn to standardized T–N residuals across 20 selected ROIs in individuals exhibiting canonical or vulnerable T–N patterns (n = 374) with the overview shown in Figure 1. It identified three distinct T–N subtypes with the T–N patterns consistent with what we previously reported in an overlapping dataset. ¹⁹ , ²⁰ , ²¹ Figure 2 illustrates the T–N residual patterns across all ROIs for each identified T–N subtype and their stages. Individuals with inferenced SuStaIn stage of zero were classified as subtype 0 (“canonical”). This canonical subtype (S0; n = 209) encompassed the largest group and was characterized by low residuals (T∼N) across regions, reflecting the normative tau–neurodegeneration relationship (Figure 2A). In addition to the canonical individuals, SuStaIn identified three subtypes with distinct T–N mismatch (N > T) progression profiles (Figure 2B). Two of these vulnerable subtypes (Subtype 1 and 3) ultimately exhibited diffuse T–N mismatch patterns across the brain although they displayed different spatiotemporal staging patterns. Subtype S1 (n = 86) initially had T–N mismatch in the anterior frontal and temporal lobes that then progressively extended to the posterior regions and then the rest of the brain. We therefore labeled S1 “anterior vulnerable.” In contrast, subtype S3 (n = 31), labeled “posterior vulnerable,” demonstrated a diffuse T–N mismatch that began in the temporal and occipital lobes and gradually progressed toward the anterior regions, eventually encompassing the entire brain. Additionally, SuStaIn recapitulated a vulnerable subtype with temporal–limbic T–N mismatch (Subtype S2, n = 48), similar to the limbic phenotype we observed in our previous work ²⁰ through T–N clustering. This subtype S2, which we labeled “limbic vulnerable,” demonstrated T–N mismatch stages beginning in the temporal–limbic regions, progressing to more anterior regions and orbital frontal cortex before eventually involving more diffuse brain regions.

A, Average ROI‐wise residual maps representing spatial T–N mismatch relationships for identified T–N subtypes among 374 individuals from Alzheimer's Disease Neuroimaging Initiative. B, ROI‐wise residual maps showing progression of SuStaIn stages of T–N subtypes. N, neurodegeneration; ROI, region of interest; T, tau.

We also evaluated the two‐subtype solution identified by SuStaIn (Figure S2 in supporting information). In addition to the canonical individuals, this solution consistently revealed an anterior vulnerable subtype characterized by diffuse T–N mismatch progressing from anterior to posterior regions. The second vulnerable group reflected temporal–limbic progression predominantly with some posterior involvement, combining the limbic and posterior patterns observed in the three‐subtype solution, consistent with the transitions illustrated in the Sankey plot (Figure S1C).

3.2. T–N subtypes displayed different clinical and pathological characteristics

We compared the T–N subtypes in terms of different characteristics (Table 1). We observed that all mismatch subtypes were older compared to the canonical subtype (p < 0.001). There were also differences in sex (p < 0.011), with the posterior vulnerable subtype showing a significantly higher proportion of males compared to females. All vulnerable subtypes had a significantly greater proportion of cognitively impaired individuals (MCI or dementia) than the canonical subtype.

TABLE 1.

Comparisons of clinical features between T–N subtypes for 374 individuals at baseline visit.

	General characteristics					Cognitive profile (covariates: age, sex, global tau, amyloid status, years of education)
Group (n)	Age (SD)	Sex (F/M/ unknown)	Diagnosis (CU/MCI/ Dementia)	Amyloid status (A+/A−)	Education (SD)	CDR‐SB (SD)	Memory (SD)	Language (SD)	Executive function (SD)
S0 Canonical (n = 209)	72.4 (6.5)	120/88/1	139/58/12	72/137	16.5 (2.5)	0.621 (1.2)	0.809 (0.66)	0.812 (0.52)	0.758 (0.58)
S1 Anterior vulnerable (n = 86)	76.4 (8.1) ^***	42/44/0	35/39/12 ^***	39/47	16.3 (2.4)	1.59 (2.5) ^*	0.372 (0.81)	0.427 (0.67)	0.275 (0.73)
S2 Limbic vulnerable (n = 48)	78.4 (7.2) ^***	20/27/1	13/16/19 ^***	25/23	16.3 (2.5)	3.48 (3.8) ^***	−0.226 (0.85) ^***	0.0257 (0.64) ^***	0.211 (0.77)
S3 Posterior vulnerable (n = 31)	82.3 (6.6) ^***	9/22/0 ^*	9/12/10 ^***	18/13	16.3 (2.7)	2.42 (2.3) ^**	0.166 (0.82)	0.233 (0.58)	−0.0402 (0.80) ^*
Group diff.	p < 0.001	p = 0.011	p < 0.001	p = 0.015	P = 0.84	p < 0.001	p < 0.001	p < 0.001	p < 0.001

Open in a new tab

Notes: Overall group effects (p values indicated in the bottom row) were tested using the Kruskal–Wallis test for categorical variables and linear regression for continuous variables. The mean (SD) is shown for age, years of education, and cognition. Pairwise comparisons of these variables between T–N subtypes were obtained. Only significant pairwise comparisons between T–N subtypes and subtype 0 (canonical) were marked by asterisks in the table alongside the corresponding value. p values were adjusted by Bonferroni multiple comparisons correction.

Abbreviations: CDR‐SB, Clinical Dementia Rating Sum of Boxes; CU, cognitively unimpaired; F, female; M, male; MCI, mild cognitive impairment; N, neurodegeneration; SD, standard deviation; T, tau.

p < 0.05.

^**

p < 0.01.

^***

p < 0.001.

At baseline (Table 1), the limbic vulnerable subtype demonstrated the most severe global cognitive impairment, as assessed by Clinical Dementia Rating Sum of Boxes (CDR‐SB) ⁴⁷ after controlling by global tau level, age, sex, and years of education (p < 0.001). This limbic vulnerable subtype also showed significantly worse performance in memory and language compared to the canonical subtype (p < 0.001). In contrast, the posterior vulnerable subtype exhibited worse executive function relative to the canonical subtype (P < 0.05). Longitudinally (Figure 3), vulnerable subtypes displayed faster cognitive impairment globally (Figure 3D) and across multiple domains (Figure 3A‐C) compared to the canonical subtype. The decline was significant for CDR‐SB, memory (p < 0.001 for all vulnerable subtypes; Figure 3A) and language (p < 0.001 for anterior and limbic vulnerable subtypes, P < 0.05 for posterior vulnerable subtype; Figure 3B). For executive function, both the anterior and limbic vulnerable subtypes exhibited a significantly faster decline (p < 0.001; Figure 3C), whereas the decline in the posterior vulnerable subtype did not reach statistical significance relative to the canonical subtype (p = 0.089). Furthermore, the limbic vulnerable subtype declined more rapidly than the anterior vulnerable subtype in memory (p < 0.001), language (p < 0.01), and CDR‐SB (p < 0.001).

Longitudinal cognitive impairment across multiple domain functions including memory (A), language (B), executive function (C), and CDR‐SB (D). Comparisons were adjusted for global tau, age, sex, amyloid status, and years of education. Statistical significance levels between vulnerable groups, corrected for multiple comparisons using Bonferroni adjustment, are indicated as *p < 0.05, **p < 0.01, ***p < 0.001. CDR‐SB, Clinical Dementia Rating Sum of Boxes; V, vulnerable.

We next examined their imaging features reflecting AD or non‐AD pathologies (Figure 4). We observed the limbic vulnerable subtype showed higher global tau uptake (p < 0.01) as measured by TauMaX, ³³ which is a global measure of tau severity based on both load and extent of tau (Figure 4A). We therefore accounted for global tau uptake in subsequent analyses by including it as a covariate. The limbic vulnerable subtype differed in ERC/PHC, which is an imaging feature suggestive of comorbid TDP‐43 in AD based on post mortem studies, ³⁸ having a significantly lower ratio than all the other subtypes (p < 0.001, Figure 4B). Moreover, it also showed significantly lower hippocampal volume relative to others after adjusting by age, sex, global tau, and ICV (Figure 4C).

Comparisons of T–N subtypes across imaging features indicative of AD or non‐AD pathologies including global tau uptake (A), ERC/PHC thickness ratio (B), hippocampal volume (C), WMH volume (D), lobar microbleeds (E). and deep microbleeds (F). Comparisons were adjusted for the covariates specified in each panel. Statistical significance was corrected for multiple comparisons using the Bonferroni method and is denoted as *p < 0.05, **p < 0.01, ***p < 0.001. AD, Alzheimer's disease; AV, anterior vulnerable; Can, canonical; ERC, entorhinal cortex; ICV, intracranial volume; LV, limbic vulnerable; N, neurodegeneration; PHC, parahippocampal cortex; PV, posterior vulnerable; T, tau; WMH, white matter hyperintensity.

The anterior vulnerable subtype was associated with higher WMH volume compared to the canonical subtype (p < 0.001, Figure 4D). We also evaluated the subtypes in terms of the number of lobar and deep microbleeds using the Boston criteria, ⁴³ which classifies microbleeds severities into three categories based on counts that were 0, 1 to 3, and > 3 (Figure 4E,F). The posterior vulnerable subtype displayed significantly greater lobar microbleed severity compared to both the canonical subtype (p < 0.001) and the anterior vulnerable subtype (p < 0.05), potentially reflecting greater enrichment of cerebral amyloid angiopathy (CAA; ⁴⁸ , ⁴⁹ Figure 4E). Additionally, both the posterior and anterior vulnerable subtypes had a greater burden of deep microbleeds compared to the canonical subtype, though the differences were modest (Figure 4F).

Next, we evaluated the relationship between SuStaIn stages and different clinical characteristics (Figure 5). We found the stages for all three vulnerable subtypes were significantly associated with worse CDR‐SB score (p < 0.001, Figure 5E), even after adjusting for global tau, age, sex, and years of education. SuStaIn stage was associated with higher global tau uptake only for limbic vulnerable groups (p < 0.01, Figure 5A). Stages were also associated with lower ERC/PHC ratios across all three subtypes (p < 0.001), with the strongest effect observed in the limbic vulnerable group (Figure 5B). The posterior vulnerable subtype showed a stage‐dependent decrease in hippocampal volume (p < 0.001, Figure 5C). While the limbic vulnerable group did not display a stage relationship with hippocampal volume, it exhibited consistently lowest hippocampal volume (and ERC/PHC ratio) across stages, suggesting a possible floor effect for further hippocampal atrophy. Moreover, higher stages of both the anterior and limbic vulnerable subtypes were significantly associated with increased WMH burden (p < 0.001, Figure 5D). We further analyzed WMH levels more granularly across different lobes. Both the stages of anterior vulnerable and limbic vulnerable subtypes were significantly associated with WMH across all lobes (Table S2 in supporting information). Interestingly, while the SuStaIn stage of posterior vulnerable subtype did not show a significant association with global WMH, it was significantly associated with higher WMH in the occipital lobe (rho = 0.32, p < 0.01, Table S2).

Associations between SuStaIn stage and global tau (A), ERC/PHC thickness ratio (B), hippocampal volume (C), WMH burden (D), and CDR‐SB score (E). Associations were assessed using Spearman correlation; correlation coefficients (ρ), p values, and covariates are reported in the accompanying table. Statistical significance is denoted as *p < 0.05, **p < 0.01, ***p < 0.001. Can, canonical; CDR‐SB, Clinical Dementia Rating Sum of Boxes; ERC, entorhinal cortex; PHC, parahippocampal cortex; WMH, white matter hyperintensity.

3.3. Longitudinal stability of classifications and T–N progression over time

Because 30% of the participants had longitudinal imaging, we applied the trained SuStaIn model to T–N residuals from follow‐up visits to assess the longitudinal stability of the subtype classification. Among the 144 participants with longitudinal scans, 71.9% stayed in the same subtype over time, and 89.5% demonstrated biologically plausible categorization, defined as either remaining in the same subtype or transitioning from the canonical subtype to one of the vulnerable subtypes between the baseline and follow‐up visits (Figure 6A). Of the participants who displayed longitudinal subtype stability excluding those who remained as canonical, 29 out of 35 individuals (82.9%) experienced an increase in stage from baseline to follow‐up, 2 out of 35 (5.7%) had stable stage and only 4 out of 35 (11.4%) had reduced stage at follow‐up, further indicating the model's reliability for identifying temporal progression. The subtype probability among individuals who transitioned from the canonical to a vulnerable subtype was lower compared to those who remained in the same group across two visits (Figure S3 in supporting information), although this difference did not reach statistical significance (p = 0.22).

A, Confusion matrix showing the longitudinal stability of subtypes between baseline and follow‐up visits with values representing the count of individuals in each subtype at both visits. The red checkmarks represent the biological plausible categorizations. B, T statistics showing longitudinal changes in T–N residuals for the vulnerable subtypes compared to the canonical subtype. MRI, magnetic resonance imaging; N, neurodegeneration; PET, positron emission tomography; T, tau; V, vulnerable.

To further support the SuStaIn‐inferred stages based on T–N mismatch from 20 selected ROIs, we assessed the longitudinal T–N mismatch progression in all 100 ROIs across subtypes. Figure 6B displays the t statistic maps of T–N residual changes in vulnerable subtypes referenced to the canonical subtype. We observed increasing T–N residuals in all three vulnerable subtypes longitudinally. The observed patterns of T–N progression were similar to the data‐driven SuStaIn stagings. Both the anterior and posterior vulnerable subtypes exhibited diffuse T–N progression over time with the posterior vulnerable showing greater extent, while the limbic vulnerable subtype demonstrated increased residuals predominantly in the temporal–limbic regions.

4. DISCUSSION

The presence of co‐pathologies in AD influences the rate of disease progression and, potentially, treatment effects of disease‐modifying medicines. In our previous work, ¹⁹ , ²⁰ , ²¹ we leveraged the mismatch between tau and atrophy to reveal relative vulnerabilities and resilience in AD patients. By clustering regional T–N residuals, we identified distinct data‐driven T–N groups. In addition to canonical and resilient groups, we found two vulnerable groups defined by their spatial pattern of residuals: a diffuse vulnerable group and the limbic vulnerable group. Converging data suggested that these groups were likely enriched in individuals with vascular disease and TDP‐43 pathology, respectively, based on both in vivo and ex vivo studies.

While previous applications of SuStaIn have primarily modeled tau accumulation ⁵⁰ or brain atrophy trajectories ⁵¹ independently to characterize AD progression, here we applied SuStaIn to T–N residuals. Our goal was to investigate the longitudinal dynamics of T–N mismatch using a model‐based approach to infer temporal trajectories associated with distinct T–N subtypes in vivo. We hypothesized that subtype‐specific T–N trajectories reflect non‐AD co‐pathological processes, whereas canonical individuals, characterized by low T–N residuals, represent the typical, tightly coupled T–N relationship in AD.

SuStaIn identified three vulnerable subtypes with more pronounced neurodegeneration than expected for their tau levels. Each of these subtypes exhibited unique SuStaIn stagings. These stages were not associated with tau severity for both anterior and posterior vulnerable subtypes, suggesting that they likely do not represent tau progression. However, the stages of limbic vulnerable subtype, suspected to be enriched in LATE, did show association with higher tau severity. This finding may reflect the fact that both the prevalence and severity of LATE, based on stage, are associated with AD pathologic severity. ⁵² A prior study has also suggested synergistic effects between tau and TDP‐43 pathology. ⁵³ Moreover, the stages of all vulnerable subtypes were associated with progressive cognitive impairment, even after controlling for global tau severity. This finding supports our hypothesis that the stages likely capture the progression of the non‐AD processes which then influence cognition beyond that due to AD pathology.

In this partially overlapping dataset, we recapitulated from our prior study ¹⁹ , ²⁰ , ²¹ the temporal–limbic vulnerable subtype with disproportionally greater atrophy for a given tau load particularly within the limbic region. We believe this subtype is likely enriched in LATE/TDP‐43 co‐pathology given this spatial pattern and prior association with TDP‐43 pathology in an autopsy sample. ²⁰ The current data extend our prior work by showing that this group specifically displayed poorer performance in memory and language at baseline compared to the other two vulnerable subtypes and also displayed faster decline over time. The limbic vulnerable subtype was also older in age compared to the canonical subtype and exhibited lower ERC/PHC ratio and hippocampal volume. These findings are consistent with phenotypic features that have been reported in association with LATE. ²² , ³⁸ , ³⁹ , ⁴⁰ Neuropathology data were available for 10 participants. Although insufficient for statistical analysis, 6 were classified as the limbic vulnerable subtype, and 5 of these exhibited TDP‐43 pathology in both the amygdala and hippocampus at autopsy. Further, we used SuStaIn‐inferred stages to explore temporal progression of this group based on cross‐sectional data under the model assumptions, which appeared to follow the expected spatial progression of LATE, ⁵⁴ starting in anterior medial temporal lobe, then extending to anterior temporal and eventually orbitofrontal regions. This staging was further supported by its alignment with observed longitudinal T–N residual change in a subset of individuals with longitudinal data available. Thus, the T–N approach is likely to provide a means to potentially stage LATE in AD patients.

Expanding on our previous work, ¹⁹ , ²⁰ , ²¹ where we identified a single diffuse vulnerable subtype through T–N clustering, here we identified two subtypes with widespread vulnerable T–N patterns using SuStaIn. However, their pattern of progression differed. The anterior vulnerable subtype exhibited T–N mismatch beginning in the anterior regions of the brain and gradually progressing toward the posterior cortex while the posterior vulnerable subtype showed the opposite pattern. The anterior vulnerable subtype was associated with a significantly greater global burden of WMHs and a strong association between stage and WMHs across the brain lobes. As such, arteriosclerotic cerebrovascular disease may be a driver of the mismatch observed in this group. In contrast, stage in the posterior vulnerable subtype was significantly associated with increased WMHs disproportionally in the occipital lobe. This spatial pattern has been linked to CAA given greater involvement of posterior circulation vessels. ⁵⁵ Indeed, we observed that the posterior vulnerable subtype had a higher number of lobar microbleeds, a characteristic feature of CAA. ⁴³

Our study has several limitations. First, given the high prevalence of multiple comorbid pathologies in AD patients, it is possible that the SuStaIn identified T–N subtypes and stages may not represent completely pure pathological progressions. Second, we limited our analysis to 20 regions given the computational constraints associated with the modeling process with the computational time scales beyond linearly with the number of input features. ⁵⁶ Despite this limitation, the T–N phenotypes identified in our previous work ²⁰ which encompassed 100 regions were recapitulated in this study. Third, parameters for SuStaIn such as z score threshold and number of starting points were chosen empirically. The stability of the inferenced subtypes and stages may change under different parameterizations. Fourth, the presence of co‐pathologies was not able to be confirmed by autopsy and, thus, were based on assumptions about patterns of pathology and linkage to surrogate measures (WMHs and microbleeds). Finally, this study was conducted using data from a single cohort (ADNI), which may not reflect the full spectrum of clinical populations. ADNI participants tend to be less diverse with respect to race, ethnicity, and disease severity compared to real‐world patient populations. ⁵⁷ Future research should validate these findings in larger, independent datasets to ensure generalizability.

In conclusion, our study explored T–N mismatch progression based on SuStaIn modelling in AD, which are likely associated with underlying co‐pathologies and linked to severity of cognitive impairment controlling for load of tau. This approach provides insights into the potential in vivo progression of these co‐pathologies which may operate independently of AD. Such staging is important because even with emerging biomarkers for non‐AD pathologies, it will be important to not only identify their presence, but also to stage their severity to determine their likely contribution to cognitive symptoms. Moreover, determining the presence and severity of co‐pathologies in vivo will be critical for evaluating the degree to which they influence outcomes with anti‐amyloid treatments and therapeutics in development.

CONFLICT OF INTEREST STATEMENT

D.A.W. has served as a paid consultant to Beckman Coulter and Eisai. He serves on a DSMB for GSK. He was a site investigator for a clinical trial sponsored by Biogen. D.J.I. receives research funding paid to the institution from the National Institutes of Health, Michael J. Fox Foundation, and Lewy Body Dementia Association (LBDA); research funding paid to the institution for clinical trials by Alector, Cervo Med, Denali, Passage Bio, and Prevail. D.J.I. has unpaid positions on the scientific advisory board for the LBDA, the medical advisory board for the Association for Frontotemporal Degeneration, and the board of directors for the International Society for Frontotemporal Dementias. B.C.D. has served as a consultant for Acadia, Arkuda, Biogen, Cervomed, Eisai, Lantheus, Lilly, Merck, Novo Nordisk, and Quanterix. He receives royalties from Cambridge University Press, Elsevier, Oxford University Press, and UpToDate. Consulting relationships with Arkuda (ended December 2024), Biogen (ended November 2024), Lantheus (ended December 2024), Lilly (ended July 2024), Merck (ended October 2024), and Novo Nordisk (ended July 2024) have concluded. I.M.N. was a consultant for Subtle Medical and was on an advisory board for Eisai. R.T.S has received consulting income from Octave Bioscience. X.L., C.A.B., M.T.D., E.P.F., E.M., A.L.Y., P.A.Y., and S.R.D. have no disclosure and declare no conflicts of interest. Author disclosures are available in the Supporting Information.

CONSENT STATEMENT

We confirm that all human subjects provided informed consent.

Supporting information

Supporting Information: alz71474‐sup‐0001‐ICMJE.pdf

ALZ-22-e71474-s001.pdf^{(796.7KB, pdf)}

Supporting Information: alz71474‐sup‐0002‐SuppMat.docx

ALZ-22-e71474-s002.docx^{(1.4MB, docx)}

ACKNOWLEDGMENTS

The study was supported by the funding R01 AG056014, RF1 AG069474, R01AG070592, P30AG072979, and R01AG072796. Data collection and sharing for this project was funded by the Alzheimer's Disease Neuroimaging Initiative (ADNI; National Institutes of Health Grant U01 AG024904) and DOD ADNI (Department of Defense award number W81XWH‐12‐2‐0012). ADNI is funded by the National Institute on Aging, the National Institute of Biomedical Imaging and Bioengineering, and through generous contributions from the following: AbbVie; Alzheimer's Association; Alzheimer's Drug Discovery Foundation; Araclon Biotech; BioClinica, Inc.; Biogen; Bristol‐Myers Squibb Company; CereSpir, Inc.; Cogstate; Eisai Inc.; Elan Pharmaceuticals, Inc.; Eli Lilly and Company; EuroImmun; F.Hoffmann‐La Roche Ltd and its affiliated company Genentech, Inc.; Fujirebio; GE Healthcare; IXICO Ltd.; Janssen Alzheimer Immunotherapy Research & Development, LLC; Johnson & Johnson Pharmaceutical Research & Development LLC; Lumosity; Lundbeck; Merck & Co., Inc.; Meso Scale Diagnostics, LLC; NeuroRx Research; Neurotrack Technologies; Novartis Pharmaceuticals Corporation; Pfizer Inc.; Piramal Imaging; Servier; Takeda Pharmaceutical Company; and Transition Therapeutics. The Canadian Institutes of Health Research is providing funds to support ADNI clinical sites in Canada. Private sector contributions are facilitated by the Foundation for the National Institutes of Health (www.fnih.org). The grantee organization is the Northern California Institute for Research and Education, and the study is coordinated by the Alzheimer's Therapeutic Research Institute at the University of Southern California. ADNI data are disseminated by the Laboratory for Neuro Imaging at the University of Southern California.

Contributor Information

Sandhitsu R. Das, Email: sudas@pennmedicine.upenn.edu.

David A. Wolk, Email: David.Wolk@pennmedicine.upenn.edu.

DATA AVAILABILITY STATEMENT

Data used in preparation of this article were obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). As such, the investigators within the ADNI contributed to the design and implementation of ADNI and/or provided data but did not participate in analysis or writing of this report. A complete listing of ADNI investigators can be found at: http://adni.loni.usc.edu/wp‐content/uploads/how_to_apply/ADNI_Acknowledgement_List.pdf.

REFERENCES

1. Jack CR, Andrews SJ, Beach TG, et al. Revised criteria for the diagnosis and staging of Alzheimer's disease. Nat Med. 2024;30(8):2121‐2124. doi: 10.1038/s41591-024-02988-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
2. Jack CR, Bennett DA, Blennow K, et al. NIA‐AA research framework: toward a biological definition of Alzheimer's disease. Alzheimers Dement J Alzheimers Assoc. 2018;14(4):535‐562. doi: 10.1016/j.jalz.2018.02.018 [DOI] [PMC free article] [PubMed] [Google Scholar]
3. Whitwell JL, Josephs KA, Murray ME, et al. MRI correlates of neurofibrillary tangle pathology at autopsy: a voxel‐based morphometry study. Neurology. 2008;71(10):743‐749. doi: 10.1212/01.wnl.0000324924.91351.7d [DOI] [PMC free article] [PubMed] [Google Scholar]
4. Wisse LEM, Ravikumar S, Ittyerah R, et al. Downstream effects of polypathology on neurodegeneration of medial temporal lobe subregions. Acta Neuropathol Commun. 2021;9(1):128. doi: 10.1186/s40478-021-01225-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
5. Ravikumar S, Wisse LEM, Lim S, et al. Ex vivo MRI atlas of the human medial temporal lobe: characterizing neurodegeneration due to tau pathology. Acta Neuropathol Commun. 2021;9(1):173. doi: 10.1186/s40478-021-01275-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
6. Harrison TM, La Joie R, Maass A, et al. Longitudinal tau accumulation and atrophy in aging and Alzheimer disease. Ann Neurol. 2019;85(2):229‐240. doi: 10.1002/ana.25406 [DOI] [PMC free article] [PubMed] [Google Scholar]
7. Joie RL, Visani AV, Baker SL, et al. Prospective longitudinal atrophy in Alzheimer's disease correlates with the intensity and topography of baseline tau‐PET. Sci Transl Med. 2020;12(524):eaau5732. doi: 10.1126/scitranslmed.aau5732 [DOI] [PMC free article] [PubMed] [Google Scholar]
8. Joie RL, Visani A, Bourakova V, et al. [P2–373]: AV1451‐pet cortical uptake and regional distribution predict longitudinal atrophy in Alzheimer's disease. Alzheimers Dement. 2017;13(7S_Part_15):P769‐P769. doi: 10.1016/j.jalz.2017.06.1028 [DOI] [Google Scholar]
9. Das SR, Xie L, Wisse LEM, et al. Longitudinal and cross‐sectional structural magnetic resonance imaging correlates of AV‐1451 uptake. Neurobiol Aging. 2018;66:49‐58. doi: 10.1016/j.neurobiolaging.2018.01.024 [DOI] [PMC free article] [PubMed] [Google Scholar]
10. Ossenkoppele R, Smith R, Ohlsson T, et al. Associations between tau, Aβ, and cortical thickness with cognition in Alzheimer disease. Neurology. 2019;92(6):e601‐e612. doi: 10.1212/WNL.0000000000006875 [DOI] [PMC free article] [PubMed] [Google Scholar]
11. Kapasi A, Schneider JA. Vascular contributions to cognitive impairment, clinical Alzheimer's disease, and dementia in older persons. Biochim Biophys Acta. 2016;1862(5):878‐886. doi: 10.1016/j.bbadis.2015.12.023 [DOI] [PMC free article] [PubMed] [Google Scholar]
12. Katsumata Y, Fardo DW, Kukull WA, Nelson PT. Dichotomous scoring of TDP‐43 proteinopathy from specific brain regions in 27 academic research centers: associations with Alzheimer's disease and cerebrovascular disease pathologies. Acta Neuropathol Commun. 2018;6(1):142. doi: 10.1186/s40478-018-0641-y [DOI] [PMC free article] [PubMed] [Google Scholar]
13. Lamar M, Boots EA, Arfanakis K, Barnes LL, Schneider JA. Common brain structural alterations associated with cardiovascular disease risk factors and Alzheimer's dementia: future directions and implications. Neuropsychol Rev. 2020;30(4):546‐557. doi: 10.1007/s11065-020-09460-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
14. Robinson JL, Richardson H, Xie SX, et al. The development and convergence of co‐pathologies in Alzheimer's disease. Brain. 2021;144(3):953‐962. doi: 10.1093/brain/awaa438 [DOI] [PMC free article] [PubMed] [Google Scholar]
15. Bartrés‐Faz D, Arenaza‐Urquijo E, Ewers M, et al. Theoretical frameworks and approaches used within the reserve, resilience and protective factors professional interest area of the Alzheimer's association international society to advance Alzheimer's research and treatment. Alzheimers Dement Amst Neth. 2020;12(1):e12115. doi: 10.1002/dad2.12115 [DOI] [PMC free article] [PubMed] [Google Scholar]
16. Stern Y, Arenaza‐Urquijo EM, Bartrés‐Faz D, et al. Whitepaper: defining and investigating cognitive reserve, brain reserve, and brain maintenance. Alzheimers Dement. 2020;16(9):1305‐1311. doi: 10.1016/j.jalz.2018.07.219 [DOI] [PMC free article] [PubMed] [Google Scholar]
17. Ossenkoppele R, Lyoo CH, Jester‐Broms J, et al. Assessment of demographic, genetic, and imaging variables associated with brain resilience and cognitive resilience to pathological tau in patients with Alzheimer disease. JAMA Neurol. 2020;77(5):632. doi: 10.1001/jamaneurol.2019.5154 [DOI] [PMC free article] [PubMed] [Google Scholar]
18. Lesuis SL, Hoeijmakers L, Korosi A, et al. Vulnerability and resilience to Alzheimer's disease: early life conditions modulate neuropathology and determine cognitive reserve. Alzheimers Res Ther. 2018;10(1):95. doi: 10.1186/s13195-018-0422-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
19. Das SR, Lyu X, Duong MT, et al. Tau‐atrophy variability reveals phenotypic heterogeneity in Alzheimer's disease. Ann Neurol. 2021;90(5):751‐762. doi: 10.1002/ana.26233 [DOI] [PMC free article] [PubMed] [Google Scholar]
20. Lyu X, Duong MT, Xie L, et al. Tau‐neurodegeneration mismatch reveals vulnerability and resilience to comorbidities in Alzheimer's continuum. Alzheimers Dement. 2024;20(3):1586‐1600. doi: 10.1002/alz.13559 [DOI] [PMC free article] [PubMed] [Google Scholar]
21. Duong MT, Das SR, Lyu X, et al. Dissociation of tau pathology and neuronal hypometabolism within the ATN framework of Alzheimer's disease. Nat Commun. 2022;13(1):1495. doi: 10.1038/s41467-022-28941-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
22. Wolk DA, Nelson PT, Apostolova L, et al. Clinical criteria for limbic‐predominant age‐related TDP‐43 encephalopathy. Alzheimers Dement J Alzheimers Assoc. 2025;21(1):e14202. doi: 10.1002/alz.14202 [DOI] [PMC free article] [PubMed] [Google Scholar]
23. van Dyck CH, Swanson CJ, Aisen P, et al. Lecanemab in early Alzheimer's disease. N Engl J Med. 2023;388(1):9‐21. doi: 10.1056/NEJMoa2212948 [DOI] [PubMed] [Google Scholar]
24. Young AL, Marinescu RV, Oxtoby NP, et al. Uncovering the heterogeneity and temporal complexity of neurodegenerative diseases with subtype and stage inference. Nat Commun. 2018;9(1):4273. doi: 10.1038/s41467-018-05892-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
25. Tustison NJ, Cook PA, Klein A, et al. Large‐scale evaluation of ANTs and FreeSurfer cortical thickness measurements. NeuroImage. 2014;99:166‐179. doi: 10.1016/j.neuroimage.2014.05.044 [DOI] [PubMed] [Google Scholar]
26. Wang H, Suh JW, Das S, Pluta J, Altinay M, Yushkevich P, Regression‐Based Label Fusion for Multi‐Atlas Segmentation. Proceedings of the Conference on Computer Vision and Pattern Recognition Workshop, IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshop, Colorado Springs, CO, USA, 20‐25 June 2011. IEEE Computer Society; 2011:1113‐1120. doi: 10.1109/CVPR.2011.5995382 [DOI] [PMC free article] [PubMed]
27. Das SR, Avants BB, Grossman M, Gee JC. Registration based cortical thickness measurement. NeuroImage. 2009;45(3):867‐879. doi: 10.1016/j.neuroimage.2008.12.016 [DOI] [PMC free article] [PubMed] [Google Scholar]
28. Isensee F, Jaeger PF, Kohl SAA, Petersen J, Maier‐Hein KH. Net: a self‐configuring method for deep learning‐based biomedical image segmentation. Nat Methods. 2021;18(2):203‐211. doi: 10.1038/s41592-020-01008-z [DOI] [PubMed] [Google Scholar]
29. Taghvaei M, Cook P, Sadaghiani S, et al. Young versus older subject diffusion magnetic resonance imaging data for virtual white matter lesion tractography. Hum Brain Mapp. 2023;44(10):3943‐3953. doi: 10.1002/hbm.26326 [DOI] [PMC free article] [PubMed] [Google Scholar]
30. Yushkevich PA, Pluta JB, Wang H, et al. Automated volumetry and regional thickness analysis of hippocampal subfields and medial temporal cortical structures in mild cognitive impairment. Hum Brain Mapp. 2014;36(1):258‐287. doi: 10.1002/hbm.22627 [DOI] [PMC free article] [PubMed] [Google Scholar]
31. Yushkevich PA, Ittyerah R, Li Y, et al. Morphometry of medial temporal lobe subregions using high‐resolution T2‐weighted MRI in ADNI3: why, how, and what's next?. Alzheimers Dement. 2024;20(11):8113‐8128. doi: 10.1002/alz.14161 [DOI] [PMC free article] [PubMed] [Google Scholar]
32. Avants BB, Epstein CL, Grossman M, Gee JC. Symmetric diffeomorphic image registration with cross‐correlation: evaluating automated labeling of elderly and neurodegenerative brain. Med Image Anal. 2008;12(1):26‐41. doi: 10.1016/j.media.2007.06.004 [DOI] [PMC free article] [PubMed] [Google Scholar]
33. Brown CA, Das SR, Cousins KAQ, et al. Tau burden is best captured by magnitude and extent: tau‐MaX as a measure of global tau. Alzheimers Dement. 2025;21(7):e70346. doi: 10.1002/alz.70346 [DOI] [PMC free article] [PubMed] [Google Scholar]
34. Landau SM, Breault C, Joshi AD, et al. Amyloid‐β imaging with Pittsburgh compound B and florbetapir: comparing radiotracers and quantification methods. J Nucl Med Off Publ Soc Nucl Med. 2013;54(1):70‐77. doi: 10.2967/jnumed.112.109009 [DOI] [PMC free article] [PubMed] [Google Scholar]
35. Landau SM, Mintun MA, Joshi AD, et al. Amyloid deposition, hypometabolism, and longitudinal cognitive decline. Ann Neurol. 2012;72(4):578‐586. doi: 10.1002/ana.23650 [DOI] [PMC free article] [PubMed] [Google Scholar]
36. Braak H, Braak E. Neuropathological stageing of Alzheimer‐related changes. Acta Neuropathol. 1991;82(4):239‐259. doi: 10.1007/BF00308809 [DOI] [PubMed] [Google Scholar]
37. Adjusted Rand Index—an overview. ScienceDirect Topics. Accessed October 13, 2025. https://www.sciencedirect.com/topics/computer‐science/adjusted‐rand‐index [Google Scholar]
38. de Flores R, Wisse LEM, Das SR, et al. Contribution of mixed pathology to medial temporal lobe atrophy in Alzheimer's disease. Alzheimers Dement J Alzheimers Assoc. 2020;16(6):843‐852. doi: 10.1002/alz.12079 [DOI] [PMC free article] [PubMed] [Google Scholar]
39. Nelson PT, Dickson DW, Trojanowski JQ, et al. Limbic‐predominant age‐related TDP‐43 encephalopathy (LATE): consensus working group report. Brain. 2019;142(6):1503‐1527. doi: 10.1093/brain/awz099 [DOI] [PMC free article] [PubMed] [Google Scholar]
40. Agrawal S, Yu L, Leurgans SE, et al. Hippocampal neuronal loss and cognitive decline in LATE‐NC and ADNC among community‐dwelling older persons. Alzheimers Dement. 2025;21(2):e14500. doi: 10.1002/alz.14500 [DOI] [PMC free article] [PubMed] [Google Scholar]
41. Weiner MW, Veitch DP, Aisen PS, et al. The Alzheimer's disease neuroimaging initiative: a review of papers published since its inception. Alzheimers Dement J Alzheimers Assoc. 2013;9(5):e111‐e194. doi: 10.1016/j.jalz.2013.05.1769 [DOI] [PMC free article] [PubMed] [Google Scholar]
42. Chiang GC, Cruz Hernandez JC, Kantarci K, Jack CR, Weiner MW. Alzheimer's disease neuroimaging initiative. cerebral microbleeds, CSF p‐Tau, and cognitive decline: significance of anatomic distribution. AJNR Am J Neuroradiol. 2015;36(9):1635‐1641. doi: 10.3174/ajnr.A4351 [DOI] [PMC free article] [PubMed] [Google Scholar]
43. Charidimou A, Boulouis G, Frosch MP, et al. The Boston Criteria v2.0 for cerebral amyloid angiopathy: a multicentre MRI‐neuropathology diagnostic accuracy study. Lancet Neurol. 2022;21(8):714‐725. doi: 10.1016/S1474-4422(22)00208-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
44. Kang K, Zhang P, Dumitrescu L, et al. The dynamics of cognitive decline toward Alzheimer's disease progression: results from ADSP‐PHC's harmonized cognitive composites. Alzheimers Dement. 2025;21(6):e70335. doi: 10.1002/alz.70335 [DOI] [PMC free article] [PubMed] [Google Scholar]
45. Kruskal WH, Wallis WA. Use of ranks in one‐criterion variance analysis. J Am Stat Assoc. 1952;47(260):583‐621. doi: 10.2307/2280779 [DOI] [Google Scholar]
46. nlme: Linear and nonlinear mixed effects models – ScienceOpen. Accessed February 17, 2022. https://www.scienceopen.com/document?vid=f71afca1‐5770‐4ec9‐8abc‐9f2da9326579
47. O'Bryant SE, Waring SC, Cullum CM, et al. Staging dementia using clinical dementia rating scale sum of boxes scores. Arch Neurol. 2008;65(8):1091‐1095. doi:10.1001/archneur.65.8.1091 [DOI] [PMC free article] [PubMed] [Google Scholar]
48. Kuo PY, Tsai HH, Lee BC, et al. Differences in lobar microbleed topography in cerebral amyloid angiopathy and hypertensive arteriopathy. Sci Rep. 2024;14(1):3774. doi:10.1038/s41598‐024‐54243‐1 [DOI] [PMC free article] [PubMed] [Google Scholar]
49. Jung YH, Jang H, Park SB, et al. Strictly lobar microbleeds reflect amyloid angiopathy regardless of cerebral and cerebellar compartments. Stroke. 2020;51(12):3600‐3607. doi:10.1161/STROKEAHA.119.028487 [DOI] [PubMed] [Google Scholar]
50. Vogel JW, Young AL, Oxtoby NP, et al. Four distinct trajectories of tau deposition identified in Alzheimer's disease. Nat Med. 2021;27(5):871‐881. doi:10.1038/s41591‐021‐01309‐6 [DOI] [PMC free article] [PubMed] [Google Scholar]
51. Baumeister H, Vogel JW, Insel PS, et al. A generalizable data‐driven model of atrophy heterogeneity and progression in memory clinic settings. Brain. 2024;147(7):2400‐2413. doi:10.1093/brain/awae118 [DOI] [PMC free article] [PubMed] [Google Scholar]
52. Nelson PT, Brayne C, Flanagan ME, et al. Frequency of LATE neuropathologic change across the spectrum of Alzheimer's disease neuropathology: combined data from 13 community‐based or population‐based autopsy cohorts. Acta Neuropathol. 2022;144(1):27‐44. doi:10.1007/s00401‐022‐02444‐1 [DOI] [PMC free article] [PubMed] [Google Scholar]
53. Tomé SO, Gawor K, Thal DR. LATE‐NC in Alzheimer's disease: molecular aspects and synergies. Brain Pathol Zurich Switz. 2024;34(4):e13213. doi:10.1111/bpa.13213 [DOI] [PMC free article] [PubMed] [Google Scholar]
54. Nelson PT, Lee EB, Cykowski MD, et al. LATE‐NC staging in routine neuropathologic diagnosis: an update. Acta Neuropathol. 2023;145(2):159‐173. doi:10.1007/s00401‐022‐02524‐2 [DOI] [PMC free article] [PubMed] [Google Scholar]
55. Thanprasertsuk S, Martinez‐Ramirez S, Pontes‐Neto OM, et al. Posterior white matter disease distribution as a predictor of amyloid angiopathy. Neurology. 2014;83(9):794‐800. doi:10.1212/WNL.0000000000000732 [DOI] [PMC free article] [PubMed] [Google Scholar]
56. Young A, Modelling progression and heterogeneity in Alzheimer's disease. University College London. https://discovery.ucl.ac.uk/id/eprint/1546577/ [Google Scholar]
57. Weiner MW, Veitch DP, Miller MJ, et al. Increasing participant diversity in AD research: plans for digital screening, blood testing, and a community‐engaged approach in the Alzheimer's disease neuroimaging initiative 4. Alzheimers Dement J Alzheimers Assoc. 2023;19(1):307‐317. doi:10.1002/alz.12797 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supporting Information: alz71474‐sup‐0001‐ICMJE.pdf

ALZ-22-e71474-s001.pdf^{(796.7KB, pdf)}

Supporting Information: alz71474‐sup‐0002‐SuppMat.docx

ALZ-22-e71474-s002.docx^{(1.4MB, docx)}

Data Availability Statement

[alz71474-bib-0001] 1. Jack CR, Andrews SJ, Beach TG, et al. Revised criteria for the diagnosis and staging of Alzheimer's disease. Nat Med. 2024;30(8):2121‐2124. doi: 10.1038/s41591-024-02988-7 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0002] 2. Jack CR, Bennett DA, Blennow K, et al. NIA‐AA research framework: toward a biological definition of Alzheimer's disease. Alzheimers Dement J Alzheimers Assoc. 2018;14(4):535‐562. doi: 10.1016/j.jalz.2018.02.018 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0003] 3. Whitwell JL, Josephs KA, Murray ME, et al. MRI correlates of neurofibrillary tangle pathology at autopsy: a voxel‐based morphometry study. Neurology. 2008;71(10):743‐749. doi: 10.1212/01.wnl.0000324924.91351.7d [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0004] 4. Wisse LEM, Ravikumar S, Ittyerah R, et al. Downstream effects of polypathology on neurodegeneration of medial temporal lobe subregions. Acta Neuropathol Commun. 2021;9(1):128. doi: 10.1186/s40478-021-01225-3 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0005] 5. Ravikumar S, Wisse LEM, Lim S, et al. Ex vivo MRI atlas of the human medial temporal lobe: characterizing neurodegeneration due to tau pathology. Acta Neuropathol Commun. 2021;9(1):173. doi: 10.1186/s40478-021-01275-7 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0006] 6. Harrison TM, La Joie R, Maass A, et al. Longitudinal tau accumulation and atrophy in aging and Alzheimer disease. Ann Neurol. 2019;85(2):229‐240. doi: 10.1002/ana.25406 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0007] 7. Joie RL, Visani AV, Baker SL, et al. Prospective longitudinal atrophy in Alzheimer's disease correlates with the intensity and topography of baseline tau‐PET. Sci Transl Med. 2020;12(524):eaau5732. doi: 10.1126/scitranslmed.aau5732 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0008] 8. Joie RL, Visani A, Bourakova V, et al. [P2–373]: AV1451‐pet cortical uptake and regional distribution predict longitudinal atrophy in Alzheimer's disease. Alzheimers Dement. 2017;13(7S_Part_15):P769‐P769. doi: 10.1016/j.jalz.2017.06.1028 [DOI] [Google Scholar]

[alz71474-bib-0009] 9. Das SR, Xie L, Wisse LEM, et al. Longitudinal and cross‐sectional structural magnetic resonance imaging correlates of AV‐1451 uptake. Neurobiol Aging. 2018;66:49‐58. doi: 10.1016/j.neurobiolaging.2018.01.024 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0010] 10. Ossenkoppele R, Smith R, Ohlsson T, et al. Associations between tau, Aβ, and cortical thickness with cognition in Alzheimer disease. Neurology. 2019;92(6):e601‐e612. doi: 10.1212/WNL.0000000000006875 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0011] 11. Kapasi A, Schneider JA. Vascular contributions to cognitive impairment, clinical Alzheimer's disease, and dementia in older persons. Biochim Biophys Acta. 2016;1862(5):878‐886. doi: 10.1016/j.bbadis.2015.12.023 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0012] 12. Katsumata Y, Fardo DW, Kukull WA, Nelson PT. Dichotomous scoring of TDP‐43 proteinopathy from specific brain regions in 27 academic research centers: associations with Alzheimer's disease and cerebrovascular disease pathologies. Acta Neuropathol Commun. 2018;6(1):142. doi: 10.1186/s40478-018-0641-y [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0013] 13. Lamar M, Boots EA, Arfanakis K, Barnes LL, Schneider JA. Common brain structural alterations associated with cardiovascular disease risk factors and Alzheimer's dementia: future directions and implications. Neuropsychol Rev. 2020;30(4):546‐557. doi: 10.1007/s11065-020-09460-6 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0014] 14. Robinson JL, Richardson H, Xie SX, et al. The development and convergence of co‐pathologies in Alzheimer's disease. Brain. 2021;144(3):953‐962. doi: 10.1093/brain/awaa438 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0015] 15. Bartrés‐Faz D, Arenaza‐Urquijo E, Ewers M, et al. Theoretical frameworks and approaches used within the reserve, resilience and protective factors professional interest area of the Alzheimer's association international society to advance Alzheimer's research and treatment. Alzheimers Dement Amst Neth. 2020;12(1):e12115. doi: 10.1002/dad2.12115 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0016] 16. Stern Y, Arenaza‐Urquijo EM, Bartrés‐Faz D, et al. Whitepaper: defining and investigating cognitive reserve, brain reserve, and brain maintenance. Alzheimers Dement. 2020;16(9):1305‐1311. doi: 10.1016/j.jalz.2018.07.219 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0017] 17. Ossenkoppele R, Lyoo CH, Jester‐Broms J, et al. Assessment of demographic, genetic, and imaging variables associated with brain resilience and cognitive resilience to pathological tau in patients with Alzheimer disease. JAMA Neurol. 2020;77(5):632. doi: 10.1001/jamaneurol.2019.5154 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0018] 18. Lesuis SL, Hoeijmakers L, Korosi A, et al. Vulnerability and resilience to Alzheimer's disease: early life conditions modulate neuropathology and determine cognitive reserve. Alzheimers Res Ther. 2018;10(1):95. doi: 10.1186/s13195-018-0422-7 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0019] 19. Das SR, Lyu X, Duong MT, et al. Tau‐atrophy variability reveals phenotypic heterogeneity in Alzheimer's disease. Ann Neurol. 2021;90(5):751‐762. doi: 10.1002/ana.26233 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0020] 20. Lyu X, Duong MT, Xie L, et al. Tau‐neurodegeneration mismatch reveals vulnerability and resilience to comorbidities in Alzheimer's continuum. Alzheimers Dement. 2024;20(3):1586‐1600. doi: 10.1002/alz.13559 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0021] 21. Duong MT, Das SR, Lyu X, et al. Dissociation of tau pathology and neuronal hypometabolism within the ATN framework of Alzheimer's disease. Nat Commun. 2022;13(1):1495. doi: 10.1038/s41467-022-28941-1 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0022] 22. Wolk DA, Nelson PT, Apostolova L, et al. Clinical criteria for limbic‐predominant age‐related TDP‐43 encephalopathy. Alzheimers Dement J Alzheimers Assoc. 2025;21(1):e14202. doi: 10.1002/alz.14202 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0023] 23. van Dyck CH, Swanson CJ, Aisen P, et al. Lecanemab in early Alzheimer's disease. N Engl J Med. 2023;388(1):9‐21. doi: 10.1056/NEJMoa2212948 [DOI] [PubMed] [Google Scholar]

[alz71474-bib-0024] 24. Young AL, Marinescu RV, Oxtoby NP, et al. Uncovering the heterogeneity and temporal complexity of neurodegenerative diseases with subtype and stage inference. Nat Commun. 2018;9(1):4273. doi: 10.1038/s41467-018-05892-0 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0025] 25. Tustison NJ, Cook PA, Klein A, et al. Large‐scale evaluation of ANTs and FreeSurfer cortical thickness measurements. NeuroImage. 2014;99:166‐179. doi: 10.1016/j.neuroimage.2014.05.044 [DOI] [PubMed] [Google Scholar]

[alz71474-bib-0026] 26. Wang H, Suh JW, Das S, Pluta J, Altinay M, Yushkevich P, Regression‐Based Label Fusion for Multi‐Atlas Segmentation. Proceedings of the Conference on Computer Vision and Pattern Recognition Workshop, IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshop, Colorado Springs, CO, USA, 20‐25 June 2011. IEEE Computer Society; 2011:1113‐1120. doi: 10.1109/CVPR.2011.5995382 [DOI] [PMC free article] [PubMed]

[alz71474-bib-0027] 27. Das SR, Avants BB, Grossman M, Gee JC. Registration based cortical thickness measurement. NeuroImage. 2009;45(3):867‐879. doi: 10.1016/j.neuroimage.2008.12.016 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0028] 28. Isensee F, Jaeger PF, Kohl SAA, Petersen J, Maier‐Hein KH. Net: a self‐configuring method for deep learning‐based biomedical image segmentation. Nat Methods. 2021;18(2):203‐211. doi: 10.1038/s41592-020-01008-z [DOI] [PubMed] [Google Scholar]

[alz71474-bib-0029] 29. Taghvaei M, Cook P, Sadaghiani S, et al. Young versus older subject diffusion magnetic resonance imaging data for virtual white matter lesion tractography. Hum Brain Mapp. 2023;44(10):3943‐3953. doi: 10.1002/hbm.26326 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0030] 30. Yushkevich PA, Pluta JB, Wang H, et al. Automated volumetry and regional thickness analysis of hippocampal subfields and medial temporal cortical structures in mild cognitive impairment. Hum Brain Mapp. 2014;36(1):258‐287. doi: 10.1002/hbm.22627 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0031] 31. Yushkevich PA, Ittyerah R, Li Y, et al. Morphometry of medial temporal lobe subregions using high‐resolution T2‐weighted MRI in ADNI3: why, how, and what's next?. Alzheimers Dement. 2024;20(11):8113‐8128. doi: 10.1002/alz.14161 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0032] 32. Avants BB, Epstein CL, Grossman M, Gee JC. Symmetric diffeomorphic image registration with cross‐correlation: evaluating automated labeling of elderly and neurodegenerative brain. Med Image Anal. 2008;12(1):26‐41. doi: 10.1016/j.media.2007.06.004 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0033] 33. Brown CA, Das SR, Cousins KAQ, et al. Tau burden is best captured by magnitude and extent: tau‐MaX as a measure of global tau. Alzheimers Dement. 2025;21(7):e70346. doi: 10.1002/alz.70346 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0034] 34. Landau SM, Breault C, Joshi AD, et al. Amyloid‐β imaging with Pittsburgh compound B and florbetapir: comparing radiotracers and quantification methods. J Nucl Med Off Publ Soc Nucl Med. 2013;54(1):70‐77. doi: 10.2967/jnumed.112.109009 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0035] 35. Landau SM, Mintun MA, Joshi AD, et al. Amyloid deposition, hypometabolism, and longitudinal cognitive decline. Ann Neurol. 2012;72(4):578‐586. doi: 10.1002/ana.23650 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0036] 36. Braak H, Braak E. Neuropathological stageing of Alzheimer‐related changes. Acta Neuropathol. 1991;82(4):239‐259. doi: 10.1007/BF00308809 [DOI] [PubMed] [Google Scholar]

[alz71474-bib-0037] 37. Adjusted Rand Index—an overview. ScienceDirect Topics. Accessed October 13, 2025. https://www.sciencedirect.com/topics/computer‐science/adjusted‐rand‐index [Google Scholar]

[alz71474-bib-0038] 38. de Flores R, Wisse LEM, Das SR, et al. Contribution of mixed pathology to medial temporal lobe atrophy in Alzheimer's disease. Alzheimers Dement J Alzheimers Assoc. 2020;16(6):843‐852. doi: 10.1002/alz.12079 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0039] 39. Nelson PT, Dickson DW, Trojanowski JQ, et al. Limbic‐predominant age‐related TDP‐43 encephalopathy (LATE): consensus working group report. Brain. 2019;142(6):1503‐1527. doi: 10.1093/brain/awz099 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0040] 40. Agrawal S, Yu L, Leurgans SE, et al. Hippocampal neuronal loss and cognitive decline in LATE‐NC and ADNC among community‐dwelling older persons. Alzheimers Dement. 2025;21(2):e14500. doi: 10.1002/alz.14500 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0041] 41. Weiner MW, Veitch DP, Aisen PS, et al. The Alzheimer's disease neuroimaging initiative: a review of papers published since its inception. Alzheimers Dement J Alzheimers Assoc. 2013;9(5):e111‐e194. doi: 10.1016/j.jalz.2013.05.1769 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0042] 42. Chiang GC, Cruz Hernandez JC, Kantarci K, Jack CR, Weiner MW. Alzheimer's disease neuroimaging initiative. cerebral microbleeds, CSF p‐Tau, and cognitive decline: significance of anatomic distribution. AJNR Am J Neuroradiol. 2015;36(9):1635‐1641. doi: 10.3174/ajnr.A4351 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0043] 43. Charidimou A, Boulouis G, Frosch MP, et al. The Boston Criteria v2.0 for cerebral amyloid angiopathy: a multicentre MRI‐neuropathology diagnostic accuracy study. Lancet Neurol. 2022;21(8):714‐725. doi: 10.1016/S1474-4422(22)00208-3 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0044] 44. Kang K, Zhang P, Dumitrescu L, et al. The dynamics of cognitive decline toward Alzheimer's disease progression: results from ADSP‐PHC's harmonized cognitive composites. Alzheimers Dement. 2025;21(6):e70335. doi: 10.1002/alz.70335 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0045] 45. Kruskal WH, Wallis WA. Use of ranks in one‐criterion variance analysis. J Am Stat Assoc. 1952;47(260):583‐621. doi: 10.2307/2280779 [DOI] [Google Scholar]

[alz71474-bib-0046] 46. nlme: Linear and nonlinear mixed effects models – ScienceOpen. Accessed February 17, 2022. https://www.scienceopen.com/document?vid=f71afca1‐5770‐4ec9‐8abc‐9f2da9326579

[alz71474-bib-0047] 47. O'Bryant SE, Waring SC, Cullum CM, et al. Staging dementia using clinical dementia rating scale sum of boxes scores. Arch Neurol. 2008;65(8):1091‐1095. doi:10.1001/archneur.65.8.1091 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0048] 48. Kuo PY, Tsai HH, Lee BC, et al. Differences in lobar microbleed topography in cerebral amyloid angiopathy and hypertensive arteriopathy. Sci Rep. 2024;14(1):3774. doi:10.1038/s41598‐024‐54243‐1 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0049] 49. Jung YH, Jang H, Park SB, et al. Strictly lobar microbleeds reflect amyloid angiopathy regardless of cerebral and cerebellar compartments. Stroke. 2020;51(12):3600‐3607. doi:10.1161/STROKEAHA.119.028487 [DOI] [PubMed] [Google Scholar]

[alz71474-bib-0050] 50. Vogel JW, Young AL, Oxtoby NP, et al. Four distinct trajectories of tau deposition identified in Alzheimer's disease. Nat Med. 2021;27(5):871‐881. doi:10.1038/s41591‐021‐01309‐6 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0051] 51. Baumeister H, Vogel JW, Insel PS, et al. A generalizable data‐driven model of atrophy heterogeneity and progression in memory clinic settings. Brain. 2024;147(7):2400‐2413. doi:10.1093/brain/awae118 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0052] 52. Nelson PT, Brayne C, Flanagan ME, et al. Frequency of LATE neuropathologic change across the spectrum of Alzheimer's disease neuropathology: combined data from 13 community‐based or population‐based autopsy cohorts. Acta Neuropathol. 2022;144(1):27‐44. doi:10.1007/s00401‐022‐02444‐1 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0053] 53. Tomé SO, Gawor K, Thal DR. LATE‐NC in Alzheimer's disease: molecular aspects and synergies. Brain Pathol Zurich Switz. 2024;34(4):e13213. doi:10.1111/bpa.13213 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0054] 54. Nelson PT, Lee EB, Cykowski MD, et al. LATE‐NC staging in routine neuropathologic diagnosis: an update. Acta Neuropathol. 2023;145(2):159‐173. doi:10.1007/s00401‐022‐02524‐2 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0055] 55. Thanprasertsuk S, Martinez‐Ramirez S, Pontes‐Neto OM, et al. Posterior white matter disease distribution as a predictor of amyloid angiopathy. Neurology. 2014;83(9):794‐800. doi:10.1212/WNL.0000000000000732 [DOI] [PMC free article] [PubMed] [Google Scholar]

[alz71474-bib-0056] 56. Young A, Modelling progression and heterogeneity in Alzheimer's disease. University College London. https://discovery.ucl.ac.uk/id/eprint/1546577/ [Google Scholar]

[alz71474-bib-0057] 57. Weiner MW, Veitch DP, Miller MJ, et al. Increasing participant diversity in AD research: plans for digital screening, blood testing, and a community‐engaged approach in the Alzheimer's disease neuroimaging initiative 4. Alzheimers Dement J Alzheimers Assoc. 2023;19(1):307‐317. doi:10.1002/alz.12797 [DOI] [PMC free article] [PubMed] [Google Scholar]

PERMALINK

Uncovering distinct spatiotemporal trajectories of T‐N mismatch subtypes with likely co‐pathology in Alzheimer's disease using event‐based modeling

Xueying Lyu

Christopher A Brown

Michael Tran Duong

Emma P Fischer

Emily McGrew

David J Irwin

Bradford C Dickerson

Ilya M Nasrallah

Russell Takeshi Shinohara

Alexandra L Young

Paul A Yushkevich

Sandhitsu R Das

David A Wolk

Abstract

INTRODUCTION

METHODS

RESULTS

DISCUSSION

Highlights

1. BACKGROUND

RESEARCH IN CONTEXT

2. METHODS

2.1. Participants

2.2. Image acquisition and processing

2.2.1. Image acquisition

2.2.2. MRI processing

2.2.3. PET image processing

2.3. T–N mismatch subtype stratification and trajectory modeling

FIGURE 1.

2.3.1. T–N mismatch modeling

2.3.2. SuStaIn modeling

2.4. Imaging features

2.4.1. Imaging feature suggestive of LATE neuropathologic change

2.4.2. Microbleeds

2.5. Domain‐specific cognition

2.6. Assessment of longitudinal stability

2.7. Statistical analysis

2.7.1. General statistical analyses

2.7.2. Longitudinal T–N residual and cognitive changes

3. RESULTS

3.1. SuStaIn uncovered distinct T–N spatial progression patterns across T–N subtypes

FIGURE 2.

3.2. T–N subtypes displayed different clinical and pathological characteristics

TABLE 1.

FIGURE 3.

FIGURE 4.

FIGURE 5.

3.3. Longitudinal stability of classifications and T–N progression over time

FIGURE 6.

4. DISCUSSION

CONFLICT OF INTEREST STATEMENT

CONSENT STATEMENT

Supporting information

ACKNOWLEDGMENTS

Contributor Information

DATA AVAILABILITY STATEMENT

REFERENCES

Associated Data

Supplementary Materials

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases