Microglia recapitulate a hematopoietic master regulator network in the aging human frontal cortex

Abstract

Microglia form the immune system of the brain. Previous studies in cell cultures and animal models suggest altered activation states and cellular senescence in the aged brain. Instead, we analyzed 3 transcriptome data sets from the postmortem frontal cortex of 381 control individuals to show that microglia gene markers assemble into a transcriptional module in a gene coexpression network. These markers predominantly represented M1 and M1/M2b activation phenotypes. Expression of genes in this module generally declines over the adult life span. This decrease was more pronounced in microglia surface receptors for microglia and/or neuron crosstalk than in markers for activation state phenotypes. In addition to these receptors for exogenous signals, microglia are controlled by brain-expressed regulatory factors. We identified a subnetwork of transcription factors, including RUNX1, IRF8, PU.1, and TAL1, which are master regulators (MRs) for the age-dependent microglia module. The causal contributions of these MRs on the microglia module were verified using publicly available ChIP-Seq data. Interactions of these key MRs were preserved in a protein-protein interaction network. Importantly, these MRs appear to be essential for regulating microglia homeostasis in the adult human frontal cortex in addition to their crucial roles in hematopoiesis and myeloid cell-fate decisions during embryogenesis.

Keywords: Aging; Hematopoiesis; Immune system; Master regulator; Microglia; Transcriptional network.

Fig. 1

Venn diagram showing the overlap of genes in the 3 microarray gene expression data sets and in the age-dependent module. (A) Overlap of genes from the 3 microarray gene expression data sets braincloud, Medical Research Council (MRC/UK) and Harvard Brain Tissue Resource Center (HBTRC). (B) Demographics and coverage of the 3 data sets. (C) Venn diagram showing the age-dependent modules and their gene overlaps.

Fig. 2

The age-dependent modules' eigengene's [gene relative expression: log₂(sample/reference)] negative correlation with age. (A) Bar plot for Pearson's correlation (y-axis) of the expression of all n = 162 genes (x-axis) in the age-dependent module with age, derived from braincloud data ranked by increasing correlations. Each bar reflects the Pearson's correlation of a given gene's expression level with age. The red dashed line indicates the Pearson's correlation for the age-dependent module's eigengene with age (Pearson's correlation = −0.35). (B) A negative slope of a linear model fit with the age-dependent module's eigengene's expression as the dependent variable, and age [years] as the independent variable. (C) and (D) display the same metrics for the network derived from MRC/UK data (n = 141 genes in the age-dependent module); (E) and (F) correspondingly display the same metrics for the network derived from HBTRC data (n = 326 genes in the age-dependent module). Abbreviations: HBTRC, Harvard Brain Tissue Resource Center; MRC, Medical Research Council.

Fig. 3

Microglia surface receptors' and microglia M1 and M1/M2b phenotype markers' expression correlation with age for Medical Research Council/United Kingdom (MRC/UK) data. The regression model fit and confidence interval for normalized gene expression versus age is for (A) the microglia module's eigengene (shown in gray), (B) surface receptors for neuron-microglia crosstalk (blue), (C) toll-like receptors (red), and (D) M1 and M1/M2b activational phenotype markers (green). The x-axis shows age and the y-axis shows normalized gene expression of the microglia module's eigengene. Legends in (B–D) indicate all respective marker genes that are expressed in the MRC/UK data set and therefore were included in the regression. Pearson's correlation coefficient of the mean expression of the expressed marker genes (B–D) or module eigengene versus age (A) is shown in the bottom-left corner of each plot.

Fig. 4

Overlap of MR's (RUNX1, PU.1, and IRF8) predicted target genes in the microglia module (A) compared with genes from the whole data set (Medical Research Council/United Kingdom [MRC/UK]) (B). (A) Predicted regulons (in other words target genes) of RUNX1, PU.1, and IRF8 in the MRC/UK gene expression data. (A) displays only genes that are part of the microglia module. (B) As in (A) but including all predicted target genes of the 3 MRs across all genes in the network. It is of note that n = 14 genes that are regulated by all 3 MRs are present in both (A) and (B). Supplementary Fig. 5 shows the same gene sets as in this figure including all gene names.

Fig. 5

Protein-protein interactions are recapitulated in a MR (MR) network. A protein-protein network, derived from direct and indirect interactions among proteins encoded by MRs' genes. Colors represent Disease Association Protein-Protein Link Evaluator p-values (see legend), calculated from 1000 within-degree node-label permutations. The network was constructed on the basis of physical protein-protein interactions that have been reported in the literature (Rossin et al., 2011). For simplicity, only names for proteins with the most significant interactions are displayed; for the complete set see Supplementary Table 7.

Fig. 6

ChIP-Seq peaks for SPI1 (PU.1), RUNX1, IRF8, and TAL1 near to genes in the combined microglia module. (A) Fold enrichment for SPI1 (PU.1), RUNX1, IRF8, and TAL1 ChIP-Seq peaks in all annotated basal regulatory domains of all genes in the combined microglia module (n = 407 assigned to the microglia module in all 3 data sets combined). Asterisks denote a significant (p-value < 0.05) enrichment. The legend indicates the cell lines that were used for each MR. (B) Illustration of MR binding sites overlapping the basal regulatory domain of AIF1 (taken from the UCSC genome browser).