Downloaded from genome.cshlp.org on January 3, 2023 - Published by Cold Spring Harbor Laboratory Press 1 Bacterial Infection Remodels the DNA Methylation Landscape of Human Dendritic Cells 2 3 Alain Pacis1,2, Ludovic Tailleux3, Alexander M Morin4, John Lambourne5, Julia L Maclsaac4, Vania 4 Yotova1, Anne Dumaine1, Anne Danckaert6, Francesca Luca7, Jean-Christophe Grenier1, Kasper D 5 Hansen8, Brigitte Gicquel3, Miao Yu9, Athma Pai10, Chuan He9, Jenny Tung11,Tomi Pastinen5, Michael S 6 Kobor4, Roger Pique-Regi7, Yoav Gilad12*, Luis B Barreiro1,13,* 7 8 1CHU Sainte-Justine Research Center, Department of Genetics, Montreal, H3T1C5, Canada; 2University 9 of Montreal, Department of Biochemistry, Montreal, H3T1J4, Canada; 3Institut Pasteur, Mycobacterial 10 Genetics Unit, Paris, 75015, France; 4Centre for Molecular Medicine and Therapeutics, Child and Family 11 Research Institute, Department of Medical Genetics, University of British Columbia; 5Génome Québec 12 Innovation Centre, Department of Human Genetics, McGill University, Montréal, H3A0G1, Canada; 13 6Institut Pasteur, Imagopole, Paris, 75015, France; 7Wayne State University, Center for Molecular 14 Medicine and Genetics and Department of Obstetrics and Gynecology, Detroit, MI, 48202; 8Johns 15 Hopkins Bloomberg School of Public Health, Department of Biostatistics and McKusick- 16 Nathans Institute for Genetic Medicine, Baltimore, MD, 21205; 9University of Chicago, Department of 17 Chemistry and Institute for Biophysical Dynamics, Chicago, IL, 60637; 10Department of Biology, 18 Massachusetts Institute of Technology, United States; 11Duke University, Departments of Evolutionary 19 Anthropology and Biology and Duke Population Research Institute, Durham, NC, USA 27708; 20 12University of Chicago, Department of Human Genetics, Chicago, IL, 60637; 13University of Montreal, 21 Department of Pediatrics, Montreal, H3T1J4, Canada. 22 23 *Correspondence to: [email protected] or to [email protected] 24 Running title: Epigenetic changes in response to infection 25 Keywords: Methylation, Infection, Enhancers, Gene regulation 1 Downloaded from genome.cshlp.org on January 3, 2023 - Published by Cold Spring Harbor Laboratory Press 26 ABSTRACT 27 DNA methylation is an epigenetic mark thought to be robust to environmental perturbations on a short 28 time scale. Here, we challenge that view by demonstrating that the infection of human dendritic cells 29 (DCs) with a live pathogenic bacteria is associated with rapid and active demethylation at thousands of 30 loci, independent of cell division. We performed an integrated analysis of data on genome-wide DNA 31 methylation, histone mark patterns, chromatin accessibility, and gene expression, before and after 32 infection. We found that infection-induced demethylation rarely occurs at promoter regions and instead 33 localizes to distal enhancer elements, including those that regulate the activation of key immune 34 transcription factors. Active demethylation is associated with extensive epigenetic remodeling, including 35 the gain of histone activation marks and increased chromatin accessibility, and is strongly predictive of 36 changes in the expression levels of nearby genes. Collectively, our observations show that active, rapid 37 changes in DNA methylation in enhancers play a previously unappreciated role in regulating the 38 transcriptional response to infection, even in non-proliferating cells. 39 2 Downloaded from genome.cshlp.org on January 3, 2023 - Published by Cold Spring Harbor Laboratory Press 40 INTRODUCTION 41 The first immune mechanisms recruited to defend against invading pathogens are those associated with 42 innate immune cells, such as dendritic cells (DCs) or macrophages. Once they sense an intruder, these 43 cells induce sophisticated transcriptional programs involving the regulation of thousands of genes, which κ 44 are coordinated with the help of signal-dependent transcription factors, including NF- B/Rel, AP-1, and 45 interferon regulatory factors (IRFs) (Medzhitov 2001; Smale 2010). The regulation of this program is 46 achieved through a series of epigenetic changes, which are thought to modulate the access of transcription 47 factors to specific DNA regulatory elements (Bierne et al. 2012). 48 49 The most well-studied epigenetic responses to immune stimuli involve the post-translational modification 50 of histone tails at promoter and enhancer regions (Bierne et al. 2012; Monticelli and Natoli 2013). Histone 51 acetylation has been shown to be essential for the activation of many pro-inflammatory genes (Ghisletti et 52 al. 2010; Qiao et al. 2013), whereas increased activity of histone deacetylases is often associated with 53 gene repression in the context of inflammation (Villagra et al. 2009). Moreover, recent studies suggest 54 that the response of innate cells to different immune challenges can result in the appearance of histone 55 marks associated with de novo enhancer elements (or latent enhancers) (Kaikkonen et al. 2013; Ostuni et 56 al. 2013). These de novo enhancers have been postulated to contribute to a faster and stronger 57 transcriptional response to a secondary stimulus (Ostuni et al. 2013). 58 59 In contrast, we still know remarkably little about the role of other epigenetic changes in controlling 60 responses to infection. DNA methylation has been particularly understudied, as a consequence of the 61 belief that methylation marks are highly stable, and unlikely to respond to environmental perturbations on 62 a short time scale (Bierne et al. 2012; Monticelli and Natoli 2013). Recent work, however, suggests that 63 DNA methylation patterns can rapidly change in response to certain environmental cues (Klug et al. 2010; 64 Guo et al. 2011; Dowen et al. 2012; Marr et al. 2014), raising the possibility that rapid changes in DNA 3 Downloaded from genome.cshlp.org on January 3, 2023 - Published by Cold Spring Harbor Laboratory Press 65 methylation might play a role in innate immune responses. To date, no studies have comprehensively 66 investigated the contribution of rapid, active changes in methylation (in contrast to passive changes 67 during cell replication) to the regulatory programs induced by innate immune cells in response to an 68 infectious agent. More broadly, the few studies in mammalian cells that demonstrate cell division- 69 independent changes in DNA methylation have only focused on a small number of CpG sites and, 70 surprisingly, have suggested that such changes are poorly predictive of changes in gene expression levels 71 (Bruniquel and Schwartz 2003; Klug et al. 2010; Guo et al. 2011; Marr et al. 2014). Here, we report the 72 first comprehensive epigenome and transcriptome analysis of monocyte-derived DCs – professional 73 antigen presenting cells that play a central role in bridging innate and adaptive immunity – before and 74 after in vitro infection with live pathogenic bacteria. All the data generated in this study are freely 75 accessible via a custom web-based browser that enables easy querying and visualization of epigenetic 76 profiles at any genomic region of interest (http://luis-barreirolab.org/EpigenomeBrowser). 77 78 RESULTS 79 MTB infection induces active changes in DNA methylation in human DCs 80 We infected monocyte-derived DCs from six healthy donors with a live virulent strain of Mycobacterium 81 tuberculosis (MTB), the causative agent of tuberculosis (TB) in humans. Monocyte-derived DCs are 82 ideally suited to study active changes in methylation because they are post-mitotic and not expected to 83 proliferate in response to infection (Pickl et al. 1996; Ardeshna et al. 2000). To experimentally confirm 84 this assumption, we performed a Carboxyfluorescein Diacetate Succinimidyl Ester (CFSE) proliferation 85 assay. This method relies on the ability of the highly fluorescent dye carboxyfluorescein to incorporate 86 within cells. Following each cell division, the equal distribution of these fluorescent molecules to progeny 87 cells results in a halving of per-cell fluorescence levels. We did not detect any decrease in per-cell 88 fluorescence at 18 hours post-infection, which confirms that DCs do not proliferate after MTB infection 4 Downloaded from genome.cshlp.org on January 3, 2023 - Published by Cold Spring Harbor Laboratory Press 89 (Fig. 1A). In contrast, we observed high rates of proliferation in our positive control, human monocytic 90 THP-1 cells (Fig. 1A). 91 92 At 18 hours after infection, we obtained paired data on single base-pair resolution DNA methylation 93 levels (using whole genome shotgun bisulfite sequencing: i.e., MethylC-seq) and genome-wide gene 94 expression data (using mRNA sequencing: i.e., mRNA-seq) in non-infected and MTB-infected DCs. For 95 MethylC-seq data, we generated 8.6 billion single-end reads (mean of 648 ± 110 SD million reads per 96 sample; Supplemental Table 1) resulting in an average coverage per CpG site of ~9X for each sample. We 97 detected an average of 24 million CpG sites in each sample, corresponding to over 80% of CpG sites in 98 the human genome. Genome-wide methylation data between biological replicates were strongly 99 correlated, attesting to the high quality of the data (Supplemental Fig. 1; mean r across all samples = 100 0.86). 101 102 As expected for mammalian cells, most CpG sites were highly methylated throughout the genome except 103 near transcription start sites (TSSs), CpG islands, and putative enhancer elements (Supplemental Fig. 104 2A,B). We found a significant negative correlation between gene expression levels and methylation levels 105 around TSSs (r = -0.39; P < 1 × 10-16; Supplemental Fig. 2C,D), highlighting the well-established role of 106 proximal methylation in the stable silencing of gene expression. Principal component analysis of our data 107 along with MethylC-seq data from 21 other purified cell types and tissues revealed that the DC 108 methylome is closely related to that of other blood-derived cells, particularly cells that share a common 109 myeloid progenitor with DCs, such as neutrophils (Supplemental Fig. 2E). 110 111 We next assessed the occurrence and the extent to which the response of DCs to a bacterial infection is 112 accompanied by active changes in DNA methylation, using the BSmooth algorithm (Hansen et al. 2012). 113 We defined MTB-induced differentially methylated regions (MTB-DMRs) as regions of 3 or more 5 Downloaded from genome.cshlp.org on January 3, 2023 - Published by Cold Spring Harbor Laboratory Press 114 consecutive CpG sites exhibiting a significant difference in methylation between the two groups (P < 115 0.01) and an absolute mean methylation difference above 0.1 (Hansen et al. 2014). Using these criteria, 116 we identified 3,271 MTB-DMRs, corresponding to both hypermethylated regions (48%) and 117 hypomethylated regions (52%) (Fig. 1B; Supplemental Table 2). To independently validate these 118 changes, we generated methylation-sensitive pyrosequencing data on control versus MTB-infected DCs 119 from 5 completely new individuals. We targeted 21 CpG sites that were differentially methylated in the 120 MethylC-seq analysis, distributed across 4 hypermethylated (11 CpG sites) and 6 hypomethylated MTB- 121 DMRs (10 CpG sites; Supplemental Table 3). We were able to validate 100% of the hypomethylated CpG 122 sites, with effect sizes similar to or greater than those identified in the original bisulfite sequencing 123 analysis (Fig. 1B,C; Supplemental Fig. 3A). In contrast, we were not able to validate any of the 124 hypermethylated CpG sites (Supplemental Fig. 3B), which indicates that most (if not all) active changes 125 in methylation observed in response to infection are losses rather than gains in methylation, in accordance 126 with previous findings (Klug et al. 2010). 127 128 We found that only 6% of hypomethylated regions overlapped with a promoter (Fig. 1D) and that the vast 129 majority of hypomethylated regions were located distal to TSSs (median distance of ~35 kb from the 130 nearest TSS; Fig. 1E, Supplemental Table 2). Hypomethylated regions occured in genomic regions that 131 show increased levels of evolutionary conservation (Supplemental Fig. 4), a finding that supports their 132 functional importance. Moreover, gene ontology analysis revealed that these regions are significantly 133 enriched (false discovery rate (FDR) < 0.05) near genes known to play a key role in the regulation of 134 immune processes, including the regulation of transcription, signal transduction, and cell apoptosis (Fig. 135 1F; Supplemental Table 4). The set of genes near hypo-DMRs included virtually all of the “master- 136 regulators” of innate immune responses, including CREB5, REL, NFKB1, IRF2, and IRF4. It also 137 included key genes involved in DC-mediated activation of B and T cells (e.g., CD83) and the regulation 138 of cell death (e.g., BCL2). 6 Downloaded from genome.cshlp.org on January 3, 2023 - Published by Cold Spring Harbor Laboratory Press 139 140 Active Changes in Methylation Occur in Regions Enriched for 5-hydroxymethylcytosine 141 The TET family proteins catalyze the conversion of methylated cytosine (5mC) to 5-hydroxy- 142 methylcytosine (5hmC), and are thus key players in the process of active demethylation. To evaluate if 143 5hmC levels dynamically change in response to MTB infection (as expected if 5mC sites must pass 144 through the 5hmC state before demethylation), we generated single base-pair resolution maps of 5hmC 145 across the genome using Tet-assisted bisulfite sequencing (TAB-seq) (Yu et al. 2012) in one of the 5 146 original donor. As previously described for other cell populations (Song et al. 2011; Lister et al. 2013), 147 we found markedly higher levels of 5hmC in gene bodies of highly expressed genes, consistent with a 148 role for 5hmC in maintaining and/or promoting gene expression (Fig. 2A) (Hahn et al. 2013; Hon et al. 149 2014). 150 151 Next, we evaluated if 5hmC marks were enriched within hypomethylated MTB-DMRs. We found that 152 regions that became hypomethylated post-infection were already associated with significantly higher 153 levels of 5hmC prior to infection (3.6-fold enrichment; Wilcoxon test; P < 1 × 10-16). Upon infection, 154 5hmC levels increased even further (Wilcoxon test; P = 1.57 × 10-11; Fig. 2B,C), suggesting that 5hmC 155 plays an important role in the cascade of events leading to active demethylation. The increase in 5hmC 156 appears to be specific to hypomethylated regions since no enrichment was observed genome-wide, a 157 result supported by quantitative immunocytochemistry data (Fig. 2D,E). The striking enrichment of 5hmC 158 within MTB-DMRs prior to infection strongly suggests that, in addition to its role as a transitory 159 demethylation intermediate, 5hmC might also contribute to coordinating the gene expression program 160 induced in response to a microbial stimulus. 161 162 MTB-DMRs overlap with enhancer elements that gain activation marks upon infection 7 Downloaded from genome.cshlp.org on January 3, 2023 - Published by Cold Spring Harbor Laboratory Press 163 Given that MTB-DMRs are primarily found distal to TSSs, we predicted that MTB-DMRs would overlap 164 with enhancer regions. To test this hypothesis and evaluate how the chromatin states associated with 165 MTB-DMRs dynamically change in response to infection, we collected ChIP-seq data for six histone 166 marks (H3K4me1, H3K4me3, H3K27ac, H3K27me3, H3K36me3 and H3K9me3) in non-infected and 167 infected DCs (Supplemental Table 1) from two additional donors. Using these data, we generated 168 genome-wide, gene regulatory annotation maps for non-infected and MTB-infected DCs using the 169 ChromHMM chromatin segmentation program (Fig. 3A; Supplemental Fig. 5) (Ernst and Kellis 2012). 170 We found that 41% of hypomethylated regions overlapped with a ChromHMM-annotated enhancer 171 region (defined by the presence of H3K4me1) already present in non-infected DCs, a 7.4-fold enrichment χ 172 compared to genome-wide expectations ( 2-test; P < 1 × 10-16; Fig. 3B,C; Supplemental Table 2). Slightly 173 higher enrichments (8.1-fold; P < 1 × 10-16) were observed when defining chromatin states in MTB- 174 infected DCs. Given the high-resolution of our histone maps, we could further distinguish between active 175 and inactive/poised enhancer elements based on the presence or absence of the H3K27ac mark, 176 respectively, in addition to H3K4me1 (Heintzman et al. 2007; Creyghton et al. 2010; Rada-Iglesias et al. 177 2011). Overall, we found that MTB infection leads to a significant increase of active enhancer elements 178 (and decrease of inactive/poised enhancers) colocalizing with MTB-DMRs (Fig. 3B,C). 179 180 We next extended our analysis by examining chromatin transition states at hypomethylated regions in 181 response to MTB-infection. We found that 42% of hypomethylated regions occurred in regions that 182 exhibited infection-dependent changes in chromatin state, a significantly higher proportion than expected 183 compared to the rest of the genome (P < 0.001; Fig. 3E). The chromatin state transitions observed resampling 184 within hypomethylated regions were primarily explained by the acquisition of histone activating marks 185 (e.g., H3K27ac) in MTB-infected cells. For example, among hypomethylated regions that overlapped 186 with predefined enhancers (i.e., enhancers observable in non-infected cells), 85% of those that exhibit a 187 change in chromatin state gained an activation mark (H3K27ac or H3K27ac+H3K4me3; Fig. 3F,G; 8 Downloaded from genome.cshlp.org on January 3, 2023 - Published by Cold Spring Harbor Laboratory Press χ 188 Supplemental Fig. 6A). This proportion was markedly larger than that observed genome-wide (37%) ( 2- 189 test; P = 1.1 × 10-59; Fig. 3F). Notably, we also found a large number of hypomethylated regions (n = 190 218; 12.7% of all hypomethylated regions) that overlapped with heterochromatin/repressed regions before 191 infection but gained de novo enhancer marks upon MTB infection (H3K4me1 (+ H3K27ac + H3K4me3)). 192 The number of de novo enhancers we observed among hypomethylated regions was significantly higher 193 than expected by chance (P < 0.001; Fig. 3D,E,G; Supplemental Fig. 6A). The identification of resampling 194 enhancers only present in infected DCs resembles recent findings showing that, in response to different 195 immune stimuli, mouse macrophages can gain de novo putative enhancer regions that were absent in 196 naive cells (Kaikkonen et al. 2013; Ostuni et al. 2013). Interestingly, we observed that 5hmC was 197 significantly enriched among de novo hypo-DMRs prior to infection (Wilcoxon test; P = 5.27 × 10-149), 198 suggesting that 5hmC might be an early “pre-marking” mechanism of enhancer activation, even before 199 the deposition of H3K4me1 marks (Supplemental Fig. 6A,B). 200 201 Finally, we found that MTB-induced activation or de novo gain of enhancer elements at hypomethylated 202 regions was associated with the induction of putative enhancer RNAs (eRNAs) (Wang et al. 2011) in 203 these intergenic regions (as measured by whole-transcriptome RNA-seq) as well as with increased levels 204 of histone marks associated with transcriptional activity (Supplemental Fig. 7). Moreover, changes in 205 eRNA levels in response to MTB infection show a striking positive correlation with changes in gene 206 expression levels of nearby genes (r = 0.49, P = 7.6 × 10-13; Supplemental Fig. 7), in support of a 207 mechanistic link between demethylation, eRNA production and the regulation of proximal protein-coding 208 genes (Lam et al. 2014). 209 210 MTB-DMRs are bound by signal-dependent transcription factors 211 We next asked if MTB-infection was associated with changes in the levels of chromatin accessibility in 212 MTB-DMRs. We mapped regions of open chromatin in non-infected and infected DCs based on genome- 9 Downloaded from genome.cshlp.org on January 3, 2023 - Published by Cold Spring Harbor Laboratory Press 213 wide sequencing of regions showing high transposase (Tn5) sensitivity (using ATAC-seq in one 214 additional donor) (Buenrostro et al. 2013). Overall, we observed that MTB-DMRs colocalize with regions 215 of open chromatin, which further reinforces the regulatory potential of these regions (Fig. 4A). 216 Interestingly, we found that the response to MTB-infection was accompanied by a striking increase in 217 Tn5 sensitivity levels in hypomethylated regions, which indicates that the chromatin in these regions 218 became more accessible after infection (Fig. 4A). This observation is commensurate with our data 219 showing the acquisition of active histone marks in these regions, and further supports the idea that 220 hypomethylated regions frequently reflect the presence of regulatory elements that become more active in 221 response to infection. 222 223 An attractive feature of ATAC-seq data is the ability to identify motif instances occupied by transcription 224 factors (TF) within regions of open chromatin (Neph et al. 2012; Buenrostro et al. 2013). We did so by 225 using a modified version of the Centipede algorithm (Pique-Regi et al. 2011) specifically devised to test 226 for aggregate differential binding of TFs between two experimental conditions. This method, which we 227 call CentiDual, compares the intensity of the Tn5 sensitivity-based footprint across all matches to a given 228 motif in the genome, between non-infected and infected samples (see Methods for details on the statistical 229 model). We found compelling evidence for measurable, genome-wide transcription factor activity (i.e., 230 binding to the genome; Bonferroni-corrected P < 0.05) in either non-infected or infected DCs for 264 TF 231 binding motifs, representing over 200 unique transcription factors (some TFs can bind different motifs; 232 Supplemental Table 5). Of these TF binding motifs, we found 55 that were differentially bound between 233 non-infected and infected DCs (Bonferroni-corrected P < 0.05; 27 show increased binding and 28 show 234 decreased binding; Fig. 4B). Among TF binding motifs showing increased genome-wide binding after κ 235 infection, we found several that are associated with NF- B/Rel (e.g., NFKB1, REL) and IRFs (e.g., IRF1, 236 IRF2) family members (Fig. 4B; Supplemental Table 5), both of which play a primary role in the 237 regulation of inflammatory signals in response to infection (Smale 2010). Interestingly, several CTCF 10