Hippocampal atrophy has long been considered a marker of cognitive decline in aging, but its causes are heterogeneous. Two main culprits: Alzheimer's pathology (amyloid and tau) and LATE-NC (accumulation of TDP-43 protein in limbic regions). Microglia, the brain's immune cells, are activated in both pathologies, but their role in hippocampal atrophy remained unclear. Nearby in the archive are papers on brain age maps, on how a neural network reads medical history, and on beams that heal and maim. This work stands apart: it combines postmortem MRI with quantitative histology of microglia, something its neighbors do not do.
Researchers studied 441 participants from aging cohort studies (ROS, MAP, Clinical Core, MARS). After death, the brain was fixed, and a month later ex vivo MRI was performed on a 3T scanner with high resolution (0.63×0.63×1.5 mm³). The hippocampus was segmented automatically, and volume was normalized to hemisphere volume. Microglia in the CA1/subiculum region were stained with CR3-43 antibody and digitized. Cell density was assessed using machine learning on the HALO AI platform, dividing cells into stages by morphology. Validation against manual stereological counting yielded a correlation of 0.93 for total microglia and 0.83 for activated (stages 2/3). Then linear regressions were built, adding adjustments for demographics, comorbid pathologies, and fixation intervals.
Total microglial density weakly correlated with smaller hippocampal volume (r_s = −0.149, p=0.002), but after accounting for Alzheimer's pathologies, LATE-NC, and vascular lesions, this association disappeared. Amyloid, tau tangles, and LATE-NC remained independent predictors of atrophy. The main result: an interaction between LATE-NC and microglia. For each additional unit of microglial density, LATE-NC was associated with an additional decrease in hippocampal volume of 0.019 units (p<0.05). This effect persisted in people with and without Alzheimer's pathology. No interaction with microglia was found for amyloid or tau.
The work shows that microglia are not a universal amplifier of any neurodegeneration. In Alzheimer's pathology, hippocampal atrophy apparently proceeds through mechanisms independent of microglial density: synaptic dysfunction, tau spread, impaired axonal transport. But in LATE-NC, microglia play a key role in hippocampal damage. This sets the work apart from its neighbors: they studied either only MRI or only histology, whereas here both modalities are combined in the same brain. This approach allows separating the contributions of different pathologies.
The topic could develop toward studying the temporal sequence: what comes first, microglial activation or TDP-43 accumulation? Longitudinal animal models or PET studies of microglia during life will be needed. It would also be interesting to check which specific microglial phenotypes (proinflammatory or senescent) amplify atrophy in LATE-NC. Perhaps therapeutic targets will emerge aimed at modulating microglia specifically in TDP-43 pathology.
The results will affect the interpretation of neuroinflammation biomarkers in clinical studies of Alzheimer's disease and LATE. They will also suggest for which patients anti-inflammatory therapy might be beneficial.
Verify reproducibility in a more diverse population (not only white non-Hispanic women over 90). Study other microglial markers, such as TREM2 or CD68, and their link to atrophy.
In the neurobiology of aging, the question remains open: why do some protein aggregates cause strong inflammation and atrophy, while others do not? This work ties this to the specific pathology LATE-NC and shows that microglia may be not just a witness but an active participant in damage. This is important for understanding the heterogeneity of neurodegeneration and for developing targeted therapy.
The average age of participants at death was 90.1 years, and the average duration of education was 16 years. These are among the oldest and best-educated cohorts in the world.
Not published on arXiv. Here are its page, the PDF and the formalization: you can check it yourself, not only read our account.