The marked diversity across the transcriptional profiles of the mouse macrophage subpopulations in our scRNA-seq data, coupled with the observation that some expression patterns seem to precede infection, implies a degree of epigenetic programming, as recently reported for AMs (Xu-Vanpala et al., 2020). For example, we were surprised to observe that bystander IM_3 concurrently overexpressed both the complement gene signature (C1qa, C1qb, C1qc, and Apoe; Table S31) that characterizes infected IM_2 and, although at a lower level than the infected cells, the same gene signature associated with control of Mtb infection and hspx′::GFPhigh bacteria (Saa3, Clec4e, Ly6i, Nos2, Ccl5, and Hp; Fig. 6 a and Table S32). In contrast, bystander IM_2s were characterized by a transcriptional signature that, other than the C1qs genes, did not overlap with their infected counterpart (Ccl8, Lgmn, Timp2, Ms4a7, and CD83; Table S33).

Looking at AM subpopulations, we observed similar trends. Focusing on the AM Pro-Infl cluster, we identified a conserved gene signature expressed by this subpopulation irrespective of infection status (Ftl1, Fth1, Fabp5, Lyz2, CD68, and Cd63; Fig. 6, a and b; and Fig. S2 e), suggesting this subpopulation of pro-inflammatory AMs is already present in the lungs of uninfected mice (and humans, as detailed previously). We therefore hypothesized that the different responses to Mtb infection that we observed among the infected cells may be a direct consequence of the preexisting chromatin organization (epigenetic bias) among the different populations.

M. bovis BCG is known to induce Mtb-comparable innate immune activation in myeloid cells (Geisel et al., 2005; Kaufmann et al., 2018; Khan et al., 2020; Rhoades et al., 2005), and there is increased interest in the efficacy of BCG as an inducer of protective immunity against tuberculosis, as recent data showed that i.v. vaccination with BCG prevents or substantially limits Mtb infection in highly susceptible rhesus macaques (Darrah et al., 2020; Nemes et al., 2018). Intravenous infection with BCG induces a reprogramming in macrophage responses known as “trained immunity” that is mediated, at least in part, through epigenetic control of transcriptional activity (Khader et al., 2019). We therefore used live BCG infection to assess the changes in chromatin organization of AMs and IMs and determine if i.v. BCG vaccination generates similar immunological responses to those observed with Mtb infection in mice.

We used ATAC-seq to identify the regions of the genome associated with open chromatin peaks in the two ontologically distinct macrophage subsets, in both BCG-infected and -uninfected mouse lungs (Fig. S5, a–c). An unbiased genome-wide comparison of ATAC-seq peak tag counts data segregated samples by infection status (principal component [PC] 1, 35.4%) and cell type (PC2, 31.9%; Fig. S5 d). Pairwise comparison of ATAC-seq peak tag counts revealed several thousand genomic sites with differentially (fold change [FC] > 2, FDR < 0.05, counts per million [CPM] > 5) opened chromatin between AM and IM in both BCG-infected and -uninfected mouse lungs (Fig. S5 e, top). When looked within ±2 Kb of the transcriptional start site (TSS), a union of 7,815 differentially accessible (DA) peaks was observed (Fig. S5 e, bottom; and Table S34). Most of the DA peaks, i.e., 6,056, were observed in infected IMs over uninfected IMs (Fig. S5 e, bottom). Among these, 921 DA peaks were shared with the comparison of infected AMs versus uninfected AMs (Fig. S5 f), indicating a common response to infection in both macrophage subsets.

Clustering of uninfected and infected IMs and AMs on these 7,815 DA peaks resulted in the identification of seven specific clusters (Fig. 6 c, top), the chromatin accessibility of which was changed drastically upon infection (Fig. 6 c, bottom). We noticed that chromatin-accessible patterns overall coincide with the scRNA-seq datasets. In particular, DA peaks in cluster 2 (infection-independent, open chromatin in AM) were associated with the gene signature that, in the scRNA-seq dataset, characterizes AMs across all infection conditions, while cluster 3 (infection-induced in IM) correlated to the gene signature associated with IM_1. DA peaks in cluster 4 (infection-induced in IM) define genes up-regulated by IMs, while DA peaks in cluster 5 (infection-induced in both AM and IM) were associated with a gene signature up-regulated by both IMs and AMs in infected cells in the scRNA-seq dataset. Finally, DA peaks in cluster 6 (infection-independent in IM) were associated with genes that define the IM_2 and IM_3 subsets (Table S2, Table S3, Table S4, and Table S34). In detail, when looking at IMs, we observed a highly open chromatin pattern for all complement genes (C1qa, C1qb, and C1qc) in both uninfected and BCG-infected IMs relative to their AM counterpart, (cluster 4; Fig. 6 c). IMs also showed an open chromatin pattern among pro-inflammatory genes, such as Ccl5, Ccl4, and Nos2 from infected mice (Fig. S5 g). AMs exhibited specific accessibility in M2 receptors such as Marco and Chil3 (cluster 2; Fig. 6 c and Fig. S5 g) and metabolic genes such as Mgll, Lpl, Trf, and Tkt (cluster 2; Fig. 6 c). The chromatin regions of these genes in IMs were closed (Fig. 6 c and Fig. S5 g). Chromatin accessibility for pro-inflammatory and infection-associated genes such as Il1b, Il6, Nlrp3, Itgam (Cd11b), CD80, CD14, Fos, and Jun were also increased in infected AMs and IMs (cluster 5; Fig. 6 c). Intriguingly, the gene signature that characterizes the AM_Pro-Infl population (Fig. 5 c and Fig. 6, b and d), which was also present in the lung of uninfected mice, had open chromatin in uninfected AMs relative to uninfected IMs and was also opened in infected AMs and IMs (Fig. 6 d), in support of the contention that different macrophage subpopulations are epigenetically constrained in their response to infection. Overall, our analysis suggests that the preexisting chromatin organization of the different subsets of macrophages is, at least in part, involved in driving the resulting gene expression profiles upon infection. Because the ATAC-seq data generated from a different infection model (BCG) show a high degree of correlation with the gene expression patterns observed in Mtb-infected lungs, it is likely that the divergent host cell responses to either BCG or Mtb infection are a combination of epigenetic imprinting that exists before infection, in addition to the subsequent responses to the evolving cytokine environment.