sc4D: spatio-temporal single-cell transcriptomics analysis through embedded optimal transport identifies joint glial response to Alzheimer’s disease pathology

How can researchers better model the complex, non-linear cellular responses to disease-associated pathologies by integrating spatial transcriptomics with longitudinal data? While current methods often analyze spatial and temporal dimensions in isolation, Alzheimer’s disease (AD) progression involves tightly coupled changes in cellular identity, spatial environment, and gene expression over several years. The authors aim to bridge this gap with a new computational framework.

To address these challenges, the researchers developed sc4D, a 4D analysis framework that integrates three core components:

1. Local-Neighborhood Analysis: Investigates cellular composition and gene expression within a 10-micron radius of amyloid-beta (Aβ) plaques and tau tangles in 3D brain tissue.
2. Embedded Optimal Transport: Utilizes a deep learning autoencoder to compress scRNA-seq data into a 100-dimensional latent manifold, applying unbalanced optimal transport to predict how cell states shift between two time points (8 and 13 months) in AD mouse models.
3. In Silico Perturbations: Leverages the autoencoder’s decoding capability to simulate gene expression knockdowns or overexpression, predicting how these molecular changes influence cellular stability and lineage transitions during disease progression.

Applying sc4D to TauPS2APP mouse models revealed that Aβ plaques and tau tangles create distinct local microenvironments, inducing specific reactive glial states. Notably, the framework identified a late-stage microglia-astrocyte syncytium surrounding plaques, supporting the hypothesis that these cells form protective networks to isolate pathological toxicity.

Longitudinal analysis revealed that reactive glial states (such as Micro3 and Astro3) show increased stability as the disease progresses, even when total cell abundance remains relatively unchanged. By simulating drug perturbations, the authors predicted that targeting specific genes—such as CAMTA1 activation or the administration of Donepezil—could shift glial populations away from these disease-associated, inflamed states toward more homeostatic, neuro-protective phenotypes.