Integration, clustering and subclustering analysis were performed using Seurat 3.9.965 . The gene counts were normalized using SCTransform66 , and regressed on mitochondrial read percentage, ribosomal read percentage, and cell cycle score as determined by Seurat. In order to avoid smoothing over depot differences, for integration human and mouse data were grouped by ‘individual’, i.e., if both subcutaneous and visceral adipose tissue for an individual human or mouse were available, they were pooled together during this step. Individuals were integrated with reciprocal PCA, using individuals that had both subcutaneous and visceral samples as references. As a result, the human and mouse references were comprised exclusively from the sNuc seq cohort. To integrate, references were integrated together, then the remaining samples—sNuc seq individuals with only subcutaneous data as well as all Drop-seq samples—were mapped to the reference. For clustering, 5000 variable genes were used, and ribosomal and mitochondrial genes were removed from the variable gene set before running PCA and calculating clusters using a Louvain algorithm, 40 PCs, and a resolution of 0.5. Clusters were identified as adipocytes, preadipocytes, mesothelial cells, vascular cells, or immune cells using marker genes, subset into individual objects, and re-integrated using the above method. Samples with fewer than 50 cells in the subset were removed before re-integration. This led to samples having artificially fewer cells in some instances—for example some Drop-seq samples had cells that clustered with adipocytes, but these cells were removed in subclustering because the small numbers of cells introduced too much variability into the integration. Subclustering was performed using a range of variable genes (1000-2000), PCs (10-40) and resolutions (0.2-0.6). Markers were calculated using a non-parametric Wilcoxon rank sum test with p values adjusted using Bonferroni correction (Supplementary Tables 1, 2), and clusters were evaluated based on the distinctness of called markers to determine the final subclustering conditions. In the subclustered objects, we removed clusters that appeared to represent doublets based on the score assigned by scrublet61 , or that appeared to be driven by high ambient RNA content as determined by percentage of mitochondrial genes and spliced/unspliced RNA ratio. The remaining clusters were annotated based on marker gene expression. In some cases, smaller subclusters (T and NK cells, B cells, monocytes/neutrophils) were further subset and PCA and clustering analysis but not integration was re-run in order to assign clusters. After subcluster annotation, identities were mapped back onto the original object and cells that were removed from the subclustered objects were similarly removed from the all-cell object.
Emont M.P., Jacobs C., Essene A.L., Pant D., Tenen D., Colleluori G., Di Vincenzo A., Jørgensen A.M., Dashti H., Stefek A., McGonagle E., Strobel S., Laber S., Agrawal S., Westcott G.P., Kar A., Veregge M.L., Gulko A., Srinivasan H., Kramer Z., De Filippis E., Merkel E., Ducie J., Boyd C.G., Gourash W., Courcoulas A., Lin S.J., Lee B.T., Morris D., Tobias A., Khera A.V., Claussnitzer M., Pers T.H., Giordano A., Ashenberg O., Regev A., Tsai L.T, & Rosen E.D. (2022). A single cell atlas of human and mouse white adipose tissue. Nature, 603(7903), 926-933.