(a) The tSNE plots of concatenated FCS files from 36 huMG samples analysed by antibody Panel A with colours indicating distribution of each donor’s cells (same as in Fig. 5a). (b) Assessment of subject-to-subject variability by probability binning and bin-wise testing for donor-specific differences between 8 donors on n = 35 biologically independent samples – SVZ (n = 8); THA (n = 8); CER (n = 5); GTS (n = 8) and GFM (n =6) – (using negative binomial generalized linear model and quasi-likelihood test of edgeR). The colour spectrum indicates FDR-adjusted p-values thresholded to 0.05 (blue). Top significant bins were automatically gated (G1) to identify donor #6-specific huMG phenotype. (c) Median bin-expression levels as shown in boxplot representation for bins inside (red boxes, n = 24 bins) and outside (green boxes, n = 488 bins) of G1 and all markers indicate a specifically higher expression of CD64 and EMR1 in huMG samples of donor #6. Box center line and limits represent median, 16th and 84th percentiles; whiskers define the data minimum and maximum. (d) Heatmap representing the pairwise earth mover’s distances (EMD) between cellular density distributions among 31 huMG samples after t-SNE embedding without the 5 samples from donor #6 or after t-SNE-embedding of 36 huMG samples excluding the outlier markers CD64 and EMR1 (e). (f) The tSNE plots of concatenated FCS files (n = 31 biologically independent samples). The colouring indicates five brain regions (left image) and eight donors (middle image, without donor #6). The right image shows the smoothed representation of statistical tSNE map after thresholding to 0.05 FDR-adjusted p-values (blue). Three highly differentially abundant subsets with distinct phenotypes (indicated by arrows) were detected when donor #6 was excluded before embedding. The phenotypes of these subsets are equivalent to subset 1, 2 & 3 seen in Figs. 6 & 7, as shown with expression levels of the key markers (lower boxplot graph representing median bin-expression levels across 31 concatenated samples with subset 1 (n = 73 bins), 2 (n = 23 bins), 3 (n = 8 bins), and all cells – subsets (n = 408 bins). (g) The t-SNE plots of concatenated FCS files (n = 36 biologically independent samples). The colouring indicates five brain regions (left image) and nine donors (middle image). The tSNE embedding was performed without CD64 and EMR1. The right image shows the smoothed representation of statistical tSNE map after thresholding to 0.05 FDR-adjusted p-values (blue). Boxplots represent median bin-expression levels across 36 concatenated samples with subset 1 (n = 56 bins), 2 (n = 53 bins) and all cells – subsets (n = 403 bins). Two highly significant subsets (white lines) were detected when CD64 and EMR1 were excluded from embedding. Subset 1 is phenotypically identical to subset 1 shown in Figs. 6 & 7 with regard to co-expression of CD11c, CD45, CD68, CX3CR1, and HLA-DR, whereas the CD206+ subset 2 is equivalent to subsets 2 & 3 in Figs. 6 & 7, as shown in the boxplot graph below. For the boxplot graph shown in both Supplementary Fig. 8f, g, box center line and limits represent median, 16th and 84th percentiles; whiskers define the data minimum and maximum. (h) Comparable results (as shown in Fig. 6c and Supplementary Fig. 8f, g) are obtained when the analysis was performed using the cydar/edgeR framework for detection of differentially abundant (DA) subsets as hyperspheres in original multiparameter space (on same set of all markers or excluding CD64 and EMR1 in n = 35 biologically independent samples or in n= 30 independent samples excluding samples from donor #6). The colour spectrum indicates FDR-adjusted P-values (edgeR quasi-likelihood test) thresholded to 0.05, overlaid onto the tSNE map (non-significant hyperspheres in blue). (i) Mean signal intensity levels of Ki-67, Cyclin A and Cyclin B staining across four different defined subsets (subset 1, n = 36; subset 2, n = 34; subset 3, n = 36 and subset 4, n = 35). The black line represents the mean value. *P < 0.05, **P < 0.01, ***P < 0.001, ****P <0.0001, one-way ANOVA with Bonferroni correction.