Second, we analyzed the compositional changes of three microglia cell types in an Alzheimer’s disease (AD) mouse model19 (Fig. 3b and Supplementary Data 2). Here, the number of replicates of sorted cells from cortex and cerebellum was low (n = 2 per group), thus challenging standard statistical testing scenarios. In the cortex, scCODA identified statistically credible changes both in microglia 2 and disease-associated microglia (DAM) using the most abundant tissue-resident microglia 1 as reference cell type, or a credible change in microglia 1 when using one of the other two types of microglia as the reference. By contrast, scCODA detected no statistically credible change in the cerebellum, which is known to be unperturbed in AD. Keren-Shaul et al.19 quantified the increase of DAM in the cortex of the AD mouse model via staining. While DAM localize in close proximity to amyloid-beta plaques and show a distinct inflammatory gene expression pattern, microglia 2 tend to represent an intermediate state between DAM and homeostatic microglia 119 (Supplementary Fig. 6). Therefore, our analysis with scCODA supports the contribution of DAM in AD. For comparison, ANCOM identified all three types of microglia as significantly changing in the cortex, and none in the cerebellum.
