Uncovering regional atrophy not explained by tau deposition using Additional Pathology Inference (AddiPath)

Lawrence P. Binding, Mihaela Croitor, C. Parker, Isaac Llorente Saguer, Neil P. Oxtoby, Daniel C. Alexander, Al YoungAlzheimer 2025 · DOI

TL;DR

This paper introduces AddiPath, a novel computational method to disentangle the effects of tau pathology from other brain pathologies on cortical atrophy in Alzheimer's disease (AD). By analyzing tau-PET and MRI data, AddiPath identifies atrophy patterns not explained by tau, suggesting the presence of other disease processes like TDP-43, which could improve patient stratification for clinical trials.

What Problem Does This Paper Address?

Alzheimer's disease (AD) is often characterized by mixed pathologies, meaning multiple disease processes contribute to brain damage. While tau deposition is a hallmark of AD, other proteins like TDP-43 can also play a significant role. Distinguishing the specific contributions of these different pathologies in vivo is challenging, especially as biomarkers for non-tau pathologies are less developed. This makes it difficult to accurately assess disease progression and select appropriate patients for clinical trials. Cortical thickness (CT) atrophy, measurable via MRI, is known to reflect both tau deposition and other contributing pathologies. The authors aim to develop a method to separate these influences and identify atrophy patterns specifically linked to non-tau pathologies.

Methods

The core innovation is the Additional Pathology Inference (AddiPath) software. AddiPath works by:

  1. Subtyping Tau Pathology: Using the SuStaln algorithm, the researchers first analyzed tau-PET data from 444 ADNI subjects to identify distinct patterns and progression subtypes of tau deposition (posterior, limbic, and cortical).
  2. Inferring Additional Pathology: AddiPath then takes the tau-PET data and corresponding cortical thickness (CT) data to learn the relative contributions of tau and additional pathologies to regional CT changes. It simultaneously estimates:
  3. A scaling parameter that quantifies how much the tau-PET signal predicts CT changes.
  4. The progression pattern of CT changes not explained by tau.
  5. Regression Analysis: Finally, regression analysis is used to examine the association between the AddiPath-inferred stages of additional pathology and clinical variables.

This approach allows for the deconvolution of atrophy signals, separating the tau-driven component from other potential drivers.

Datasets

The study utilized data from the Alzheimer's Disease Neuroimaging Initiative (ADNI). Specifically, they analyzed paired tau-PET and CT data from 444 ADNI subjects. No new benchmark datasets were introduced.

Key Insights and Ideas

  • Deconvoluting Atrophy: The primary insight is the successful development of a method (AddiPath) to disentangle the effects of tau pathology from other, unidentified pathologies on brain atrophy. This is crucial for a more precise understanding of AD progression.
  • Identifying Non-Tau Pathology Patterns: AddiPath uncovered a distinct progression pattern of atrophy that was not explained by tau deposition. This pattern originated in the entorhinal cortex and spread to the medial temporal lobe and then across the cortex, which the authors suggest resembles the spread of TDP-43 pathology.
  • Clinical Relevance: The severity of AddiPath-inferred additional pathology was significantly associated with worse cognitive function (memory, language, and executive function), highlighting its clinical importance.
  • Independence from Tau Load: Crucially, the AddiPath-inferred additional pathology stages were not correlated with CSF total tau or tau-PET levels, strongly supporting the idea that this identified pathology is independent of the measured tau burden.

Key Results

  • Tau Subtypes: SuStaln identified three distinct tau pathology subtypes: posterior, limbic, and cortical.
  • Additional Pathology Pattern: AddiPath identified an "additional pathology" progression pattern, labeled as "limbic+" due to its starting point.
  • Prevalence of Additional Pathology: 84 out of 444 subjects (approximately 19%) were identified by AddiPath as having significant additional pathology.
  • Cognitive Associations: Increased AddiPath stages were strongly associated with worse composite memory (p<0.001), language (p=0.001), and executive function (p<0.001) scores.
  • Independence Confirmation: AddiPath stages showed no significant correlation with CSF total tau (p=0.860) or tau-PET (p=0.475), confirming its independence from tau load.

Limitations and Future Work

  • Nature of Additional Pathology: While the pattern suggests TDP-43, the exact nature of this "additional pathology" remains inferred and not definitively confirmed by direct biomarkers in this study.
  • Replication and Generalizability: The authors plan to replicate these findings in additional cohorts to ensure the robustness and generalizability of AddiPath.
  • Broader Biomarker Application: Future work will also explore the applicability of AddiPath to other types of biomarkers beyond tau-PET and MRI, potentially expanding its utility in understanding complex neurodegenerative diseases.

Summary updated

Abstract

We introduced AddiPath, a method for disentangling the contributions of additional pathology from primary pathology. AddiPath uncovered changes in CT not explained by tau deposition in the limbic lobe, possibly indicating TDP-43 deposition. This approach could improve patient selection for clinical trials and treatment stratification. Future work will replicate this finding across additional cohorts and test the applicability of AddiPath to other biomarkers.
BibTeX
@inproceedings{binding2025uncovering,
  author = {Lawrence P. Binding and Mihaela Croitor and C. PARKER and Isaac Llorente Saguer and Neil P. Oxtoby and Daniel C. Alexander and AL Young},
  title = {{Uncovering regional atrophy not explained by tau deposition using Additional Pathology Inference (AddiPath)}},
  booktitle = {Alzheimer},
  year = {2025},
  doi = {10.1002/alz70862\_110289},
}
Figure