Journal Logo

Feature Articles

Epigenetic Profiling in Severe Sepsis: A Pilot Study of DNA Methylation Profiles in Critical Illness*

Binnie, Alexandra MD1; Walsh, Christopher J. MD2,3; Hu, Pingzhao PhD4; Dwivedi, Dhruva J. PhD5; Fox-Robichaud, Alison MD5,6; Liaw, Patricia C. PhD5,6; Tsang, Jennifer L. Y. MD6,7; Batt, Jane MD2,3; Carrasqueiro, Gabriela MSc8,9; Gupta, Sahil BSc (Hon)2,3; Marshall, John C. MD2,3; Castelo-Branco, Pedro PhD8,9,10; dos Santos, Claudia C. MD2,3; for the Epigenetic Profiling in Severe Sepsis (EPSIS) Study of the Canadian Critical Care Translational Biology Group (CCCTBG)

Author Information
doi: 10.1097/CCM.0000000000004097


Sepsis is a complex multisystem disease characterized by a dysregulated host response to infection (1) that can persist and evolve even after the inciting infection has been treated. The emerging paradigm is one of altered transcription of thousands of genes involved in fundamental cellular processes including immune and metabolic remodeling. Regulatory mechanisms underlying this response have proven difficult to elucidate, complicated by uncertainty as to whether molecular events are unique to sepsis or represent common features of critical illness (2). Resolving and understanding the pathogenesis of the sepsis “phenotype” is one of the major challenges of intensive care medicine.

Epigenetic marks (including DNA methylation, histone modifications, and regulatory RNAs) are important regulators of gene expression in disease. Disease-specific DNA methylation changes have been identified in conditions as diverse as pre-eclampsia (3), colon cancer (4,5), Crohn’s disease (6), type 2 diabetes mellitus (7,8), and asthma (9). DNA methylation changes may result in gene activation or gene repression, depending on their location relative to genes and regulatory elements (10–12). DNA methylation also regulates expression of key inflammatory cytokines and transcription factors including interferon (IFN) (13), interleukins (13,14), and recognition receptors (15,16). Furthermore, preclinical studies show that DNA methylation inhibitors reduce inflammation and organ failure (17–19). Accordingly, we postulated that epigenome-wide profiling of DNA methylation patterns in whole blood samples might identify DNA methylation changes that distinguish septic from nonseptic critically ill patients and correlate to clinically relevant features (traits) in septic patients.

We have conducted a nested pilot case-control study to compare DNA methylation profiles in whole blood genomic DNA from septic versus nonseptic critically ill patients with similar severity of illness and mortality.


Details are listed in the supplemental methods (Supplemental Digital Content 1,

Study Design

The Epigenetic Profiling in Severe Sepsis (EPSIS) is a nested pilot case-control study of patients recruited from the multicenter prospective observational trial DNA as a Prognostic Marker in ICU patients (DYNAMICS; NCT01355042) (20). Patients were recruited, and samples were collected in two Canadian academic ICUs between November 2010 and May 2014. Written informed consent was obtained from participants or their substitute decision makers. The study protocol was approved by the research ethics boards of the participating institutions (details presented in supplemental material, Supplemental Digital Content 2,

Patient Selection

DNA samples were collected on “day 1” of ICU admission. A total of 141 patient DNA samples were initially selected for DNA extraction and processing. Patient groups were matched for severity of illness and mortality.

DNA Methylation Analysis

Schematic of the workflow is shown (Fig. 1A). Bisulfite-converted DNA (500 ng) was hybridized to the Infinium Human Methylation 450K BeadChip (Illumina, San Diego, CA) (21). The “ChAMP” analysis pipeline (22) was used for preprocessing. Patient samples were excluded if 1) more than 20% of sample probes had detection p value of greater than 0.01 or 2) median bisulfite conversion efficiency control probe signal less than 4,000 in the green channel (23). Seven samples did not meet quality inclusion criteria and were removed leaving 134 high-quality samples (66 septic and 68 nonseptic) for analysis. Probes with the following criteria were excluded: 1) detection p value of less than 0.01 in greater than or equal to 10% of samples; 2) probes mapping to sex chromosomes; 3) probes with nonunique genomic mapping (24); and 4) probes containing single nucleotide polymorphism-introduced artifacts (25). A total of 414,826 probes passed quality control. The quality-filtered cytosine-phosphate-guanine (CpG) probes were analyzed using the adjacent site clustering (A-clustering) algorithm (26) to identify clusters of greater than or equal to two CpGs with correlated methylation levels. The association of the methylation clusters and patient status (septic or nonseptic) was tested by generalized estimating equations. The analysis was adjusted for sex, age, DNA processing method, estimated percent neutrophils, and 28-day mortality. The coMET algorithm (27) was used to assign probes to their gene targets using the human reference genome 19 (28).

Figure 1.
Figure 1.:
Workflow schematic and volcano plot. A, Schematic of experimental design and analysis strategy. B, Volcano plot of cytosine-phosphate-guanine (CpG) sites within differentially methylated regions showing β value difference (x-axis) plotted versus –log10 of adjusted p value (y-axis). Genes associated with the most significantly differentially methylated CpGs and adjusted p value less than 0.05 are indicated. ANGPT2 = angiopoietin 2, C11orf41 = chromosome 11 open reading frame 41, CDC42 = cell division control protein 42, C2 calcium-dependent domain containing protein 2 (C2CD2) L = C2CD2 like, CDC42BPB = CDC42 binding protein kinase β, CDKN1C = cyclin-dependent kinase inhibitor 1C, DLG4 = discs large membrane-associated guanylate kinase scaffold protein 4, ERICH1 = glutamate rich 1, FDR = false discovery rate, FLJ43663 = long intergenic non–protein coding RNA, p53 induced transcript, HCG4P6 = human leukocyte antigen (HLA) complex group 4B (non–protein coding), HLA-DQB1 = major histocompatibility complex, class II, DQ β-I, HTR3A = 5-hydroxytryptamine receptor 3A, INPP5A = inositol polyphosphate-5-phosphatase A, JARID2 = jumonji and AT-rich interaction domain containing 2, LOC148824 = gene identifier 148824, PON1 = paraoxonase 1, RNF39 = ring finger protein 39.

Stratified Comethylation Analysis

Weighted gene coexpression network analysis (WGCNA) was performed to identify clusters of highly correlated CpGs (comethylation modules). Statistical analysis was performed using R v 3.3.2 (R Foundation for Statistical Computing, Vienna, Austria; and Bioconductor v. 3.4 (R Foundation for Statistical Computing) along with the packages described above.


Patient Demographics

Genomic DNA from 141 septic (n = 72) and nonseptic (n = 68) critically ill patients was isolated from banked whole blood specimens collected on day 1 of ICU admission. Patients were matched by severity of illness (multiple organ dysfunction syndrome [MODS] scores) and 28-day mortality, so that equal number of survivors versus nonsurvivors and low versus high severity of illness (MODS scores, ≤ 8 or ≥ 9) were included. After quality control, 134 samples (66 septic and 68 nonseptic) were included in the epigenome-wide association study (EWAS) (Fig. 1A). Clinical characteristics are presented in Supplemental Table 1 (Supplemental Digital Content 3, No significant differences in age, sex, disease severity, comorbidities, or 28-day mortality were identified.

Sepsis-Associated Differentially Methylated Regions

Because methylation changes spanning multiple neighboring CpG sites are more likely to be functionally active than changes at a single CpG site, we clustered CpG sites into regions with correlated DNA methylation (26). We then looked for differential methylation of these regions between septic and nonseptic patients while controlling for baseline and confounding factors including age, gender, severity of illness (MODS score), and 28-day survival status. Using a false discovery rate (FDR) less than 5% and absolute β-difference greater than 2%, a total of 668 differentially methylated regions (DMRs) were identified as significantly differentially methylated between septic and nonseptic patients. Hypermethylated clusters predominated, with 410 hypermethylated (61%) versus 258 hypomethylated DMRs (39%) (Fig. 1B). Supplemental Table 2 (Supplemental Digital Content 4, shows the top DMRs ranked by effect size. A volcano plot of β-methylation difference versus p value for all DMRs reveals individual DMRs with highly significant β-methylation differences (Fig. 1B). Among them are several genes associated with clinically relevant outcomes in sepsis (29).

Functional Enrichment for IFN-Regulated Genes

Initial enrichment analysis was performed using the DAVID functional annotation tools ( Largest fold enrichments were for major histocompatibility complex (MHC) class II proteins and methyltransferases (Supplemental Fig. 1A, Supplemental Digital Content 5,; and legend, Supplemental Digital Content 13, Significant enrichment was also noted for genes encoding for metal binding, adhesion, and cell junction proteins. Importantly, enrichment was identified for genes involved in interferon-gamma [IFNγ]-mediated signaling pathways (adjusted p value [p*] = 3.2E-4), antigen processing and presentation via MHC class II (p* = 3.6E-4), and immunoglobulin production (p* = 6.9E-4). A search in InterHuman disease associations identified overlap between the differentially methylated genes in sepsis and differentially regulated gene expression in various autoimmune diseases, including those regulated by IFN such as ulcerative colitis, diabetes, rheumatoid arthritis, and systemic lupus erythematosus (EnrichR; Supplemental Fig. 1B, Supplemental Digital Content 5,; and legend, Supplemental Digital Content 13,

Analysis in INTERFEROME ( identified enrichment for 335 genes involved in IFNγ signaling pathways (Fig. 2A) (p* = 0.0E-15), type I IFN-mediated signaling (p* = 7.41E-15), and negative regulation of IFNγ production (p* = 0.15E-7) as well as regulation of histonehistone 3, lysine 27 methylation (p* = 2.4E-5) and positive regulation of histone 3, lysine 9 methylation (p* = 3.53E-4) (Fig. 2D; and Supplemental Table 3, Supplemental Digital Content 6, Each enrichment term returned was disproportionately associated with hyper- versus hypomethylated DMRs (Fig. 2, B and C; Supplemental Table 4, Supplemental Digital Content 7,; and Supplemental Table 5, Supplemental Digital Content 8, DMR-containing genes that were hypomethylated were more likely involved in stopping, preventing, or reducing the frequency, rate, or extent of IFNγ (p* = 1.04E-7) compared with hypermethylated genes (p* = 1.4E-4). Conversely, regulation of anti-inflammatory, T-helper 1 type immune responses, and tolerance induction of non-self-antigen was enriched among hypermethylated DMRs (Fig. 2, B and C).

Figure 2.
Figure 2.:
Enrichment for interferon (IFN) responsive genes. Venn diagram showing results of enrichment search in INTERFEROME ( identifying significant overlap between 335 genes containing differentially methylated regions and genes known to be involved in IFN signaling pathways (type I, II, and III): A, Shows all differentially methylated region (DMR)-containing genes irrespective of direction of methylation; B, only hypomethylated; and C, only hypermethylated (Hyper Me) genes. Bar graph shows adjusted p values (x-axis) for selected top enrichment results (y-axis) for: hypomethylated (D) and hypermethylated (E) containing genes. By convention, red is hypermethylated enrichment and blue is hypomethylated enrichment. C3a = complement component 3a, IL = interleukin, Rc = receptor, TLR = toll-like receptor, TCF7L2 = transcription factor 7 like 2, VEGF = vascular endothelial growth factor.

Chromosome 6 Enrichment and Predicted Protein Interaction Network

Enrichment analysis further identified enrichment of genes encoded in chromosome 6, cytoband 6p21.3 (11.8-fold enrichment; p* = 1.6E-5) (Fig. 3A). DMR-associated genes in this region included both MHC class proteins and non-MHC proteins. DMR clusters in 6q14-q15 (p* = 6.4E-4), 19q13.32 (p* = 8.3E-4), and 19q13.42 (p* = 4.9E-1) further contributed to enrichment for genes involved in antigen processing and presentation, immunoglobulin production, T cell receptor, and cytokine receptor signaling (Fig. 2D). DMRs on chromosome 6 were disproportionately hypomethylated (141/199) (Fig. 3B). STRING ( was used to determine if the putative gene products could interact in a network (Fig. 3C), and if the purported network had significantly more interactions than would be expected by chance alone (p* < 1.0E-16). Biologic processes predicted to be affected included antigen processing and presentation (FDR, 1.46E-5), IFNγ signaling pathways (FDR, 5.04E-5), and MHC protein complex (FDR, 7.1E-8).

Figure 3.
Figure 3.:
Enrichment for differentially methylated region (DMR)-containing genes at chromosome 6: (A) localization and correlation of six DMRs within the chromosome 6 human leukocyte antigen (HLA) region (6p21) mapping to the HLA genes major histocompatibility complex, class I, A; major histocompatibility complex, class I, C; major histocompatibility complex, class II, DR β-I; major histocompatibility complex, class II, DQ β-I; and major histocompatibility complex, class II, DQ β-I II. Displayed are the log10(p value) of individual cytosine-phosphate-guanine (CpG) sites within the DMRs (dashed red line indicating false discovery rate 10%), location of gene exons (yellow bars), location of SNPs (red bars), location of CpG islands (black bars), and correlations between differentially methylated CpG sites (bottom panel: red indicates strong positive correlation, white indicates no significant correlation, blue indicates strong negative correlation). B, Effect size versus β difference for all DMR-containing genes on chromosome 6 between septic (n = 66) and nonseptic (n = 68) highlighting disproportionate hypomethylation of genes on chromosome 6. By convention, red is hypermethylated enrichment and blue is hypomethylated enrichment. C, Network analysis in STRING of top differentially methylated genes on chromosome 6 predicts gene product form interaction network. Network nodes represent proteins, that is, each node represents all the proteins produced by a single, protein-coding gene locus. Colored nodes: proteins and first shell of interactors. White nodes: second shell interactors. Node content: empty nodes contain proteins with unknown 3D structures. Filled nodes: proteins with known or predicted 3D structures. Edges represent protein-protein associations. Associations are meant to be specific and meaningful, that is, proteins jointly contribute to a shared function; this does not necessarily mean that they are physically binding each other. Known interactions (light blue from curated databases; pink experimentally determined). Predicted interactions (green [pathway related]; red [gene fusion]; blue [gene co-occurrence]). Others (light green [text mining]; black [coexpressed], gray [protein homology]). B3GALT4 = β-1,3-galactosyltransferase 4, C6orf25 = chromosome 6 open reading frame 25, CLIC1 = chloride intracellular channel 1, DDX43 = DEAD-box helicase 43, DPPA5 = developmental pluripotency-associated protein 5, GABBR1 = gamma-aminobutyric acid type B receptor subunit 1, KHDC1 = KH domain containing protein 1, LY6G5C = lymphocyte antigen 6 family member G5C, PPT2 = palmitoyl-protein thioesterase 2, PRRT1 = proline-rich transmembrane protein 1, RNF39 = ring finger protein, SLC17A5 =solute carrier family 17 (anion/sugar transporter), member 5, TAPBP = transporter associated with antigen-processing binding protein.

Localization of DMRs Relative to DMR-Associated Genes and CpG Islands

A majority of sepsis-associated DMRs were located within known or putative genes and/or their associated regulatory sequences. Of the 668 DMRs, 488 (73.0%) mapped to 443 unique gene IDs, of which 381 coded for known proteins. The location of DMRs was mapped relative to the associated genes: 23.9% of DMRs were located within 1,500 base pairs (bp) of a transcriptional start site, 3.1% in the 5′ untranslated region (UTR), 49% in the gene body, and 1.8% in the 3′UTR (Fig. 4A). Hypomethylated DMRs were more likely than hypermethylated DMRs to be located in upstream regulatory regions (32.9% vs 25.0%; p = 0.029) (Fig. 2A) (30). The location of DMRs relative to CpG islands was also analyzed. Just over a third of DMRs (33.5%) were located entirely within CpG islands while two thirds (67%) were located within 4 kilo-bp (kb) of a CpG island. Hypomethylated DMRs were more likely to be located within CpG islands (41% vs 28%) or within 4 kb of a CpG island (77% vs 58%). Interestingly, 35 of 42 DMRs located near or in genes encoding for methyltransferase activity were located in the 5′UTR or transcriptional start sites (FDR, 9.83e-15) (Fig. 4B). Predicted protein product interactions were derived using STRING (Fig. 4C). β-differences in methylation for histone methyltransferases are shown in Figure 4D.

Figure 4.
Figure 4.:
Localization of differentially methylated regions (DMRs) within DMR-associated genes and histone methyltransferases. A, Bar graph showing Hypomethylated DMRs are overrepresented in 5′ promoter regions relative to hypermethylated DMRs (p = 0.029). B, Localization of DMRs relative to DMR-associated genes and CpG islands versus effect size showing hypermethylation of methyltransferase genes at regulatory 5′ untranslated region (UTR) or close to transcription start sites (TSSs). C, Network analysis in STRING showing predicted protein interaction network of gene products related to methyl transferase activity. Legend for networks and edges is described in detail in Figure 3. D, Box plot showing average and range of negative log10 β differences at key DMR containing methyltransferase-related genes in septic (S) versus nonseptic (NS) critically ill patients (adjusted p values, 0.05). Negative log10 β values for each DMR per individual included in the study are shown as circle. Degree of change in methylation is illustrated by a gradated color progression going from red to blue, where by convention, red is hypermethylation and blue for hypomethylation. gene body = gene body downstream of first exon, NSD1= nuclear receptor binding SET domain protein 1, PRDM1 = PR/SET domain 1, PRDM2 = PR/SET domain 2, PRDM8 = PR/SET domain 8, PRDM9 = PR/SET domain 9, PRDM16 = PR/SET domain 16, SETD1B = SET domain Containing protein 1B, SMYD4 = SET and MYND domain containing protein 4, TSS1500 = within 1,500 base pairs of the transcription start site, TSS200 = within 200 base pairs of the transcription start site.

DMRs Show Transcription Regulation in External Transcriptomic Datasets

We capitalized on external datasets to determine which DMR-associated genes are differentially expressed in existing sepsis cohorts. Whole blood gene expression data from seven datasets comprising septic and nonseptic patients were downloaded from the gene expression omnibus (GEO) (31) (Supplemental Table 6, Supplemental Digital Content 9, In these datasets, critically ill septic patients were compared with healthy controls (three datasets), critically ill nonseptic patients (three datasets), or both (one dataset).

The transcriptomic datasets were used to validate gene expression changes in DMR-associated genes during sepsis. Among DMR-associated genes, 81% showed significant (FDR, < 5%) gene expression change in at least one dataset and 31% in three or more datasets (data not shown). DMR-associated genes were significantly more likely to be differentially expressed in these datasets relative to genes not associated with a DMR (p = 0.009) (Supplemental Fig. 2A, Supplemental Digital Content 10,; and legend, Supplemental Digital Content 13, When the analysis was restricted to genes with greater than or equal to 1.5-fold average change in expression (averaged across seven datasets), DMR-containing genes were much more likely to be differentially expressed in septic versus nonseptic critically ill patients (p = 3 × 10–5) (Supplemental Fig. 2B, Supplemental Digital Content 10,; and legend, Supplemental Digital Content 13,

DMR-associated genes that are differentially expressed in sepsis might be targets for epigenetic therapy. Supplemental Figure 2C (Supplemental Digital Content 10,; and legend, Supplemental Digital Content 13, shows the DMR-associated genes plotted as β-methylation difference versus (mean) log-fold gene expression change (determined as mean expression change across seven GEO datasets: potential targets include MHC, class II, DQ β-I [HLA-DQB1]; cluster of differentiation 177; paraoxonase 1 [PON1]; and jumonji and AT-rich interaction domain containing 2 [JARID2]) (32–34).

Functional Gene Analysis

A number of DMR-associated genes are known to play a role in sepsis. Among the 367 DMR-associated genes showing differential expression in sepsis transcriptomic datasets, we identified 56 genes (15.2%) that were previously associated with sepsis in the literature (Supplemental Table 6, Supplemental Digital Content 9, These included complement component 3; myeloperoxidase; angiopoietin 2; lactoperoxidase; MHC, class I, A; MHC, class I, C; MHC, class II, isotype DQ β-I and MHC, class II, isotype DR β-I; lipase A lysosomal acid type; PON1, fragment crystallizable region of of immunoglobulin G receptorIIa (FCG2A); interleukin 23 A; tissue inhibitor of metalloproteinases 2 (TIMP2); and cathepsin G. A further subset of genes had known epigenetic activity including the pro-apoptotic protein arginine methyltransferase 2 (PRMT2), the histone deacetylase histone deacetylase 4 (HDAC4) (35), the transcription factor JARID2 (34), and the histone methyltransferase and chromatin modifier PR/SET domain 2 (PRDM2) (36).

Correlation of Methylation With Clinical Phenotypes

Coexpression network analysis has been used to uncover clinically relevant methylation modules that correlate with phenotype (37,38). We used WGCNA to detect coexpression modules that correlated with relevant phenotypic traits available for the EPSIS cohort including MODS score, vasopressor requirement, 28-day survival, ICU discharge, and hospital discharge. Two separate coexpression networks were built for 1) septic and 2) nonseptic patients (Supplemental Fig. 3, A and B, Supplemental Digital Content 11,; and legend, Supplemental Digital Content 13, Each module was summarized by its first principle component of the CpG methylation values (module eigengenes). Correlation with clinical phenotypes on day 1 of ICU admission was identified for eight septic and five nonseptic modules (p < 0.05).

Module preservation analysis (Supplemental Fig. 3C, Supplemental Digital Content 11,; and legend, Supplemental Digital Content 13, demonstrated nonseptic modules preserved in septic patients, whereas among septic modules, only eight of 10 were preserved in nonseptic patients.


We report the first epigenome-wide DNA methylation analysis of whole blood samples from septic and nonseptic critically ill adults. A single previous study in neonates profiled three septic patients and three healthy infants (39). Our analysis of 134 patients (68 septic and 66 nonseptic) revealed the presence of 668 DMRs, that correlated with sepsis status. Functional enrichment analysis of the 443 DMR-associated genes showed these encoded for MHC proteins and methyltransferases as well as proteins involved in cell adhesion and cell junctions. These functions are highly plausible in the context of sepsis and lend support to the validity of our analysis. Disease association analysis reveals a strong overlap with genes involved in autoimmune diseases, including systemic lupus erythematosus, rheumatoid arthritis, and ulcerative colitis. Importantly, the most significant area of overlap was in IFN-regulated genes (40,41).

IFNs are a family of pleiotropic cytokines that typically exhibit antiviral, antiproliferative, antitumor, and immunomodulatory properties. Although their complex mechanisms of action remain unclear, the IFN-mediated innate immune response, selected by evolution, is hardwired within genomes and provides a robust first line of defense against invading pathogens (42). Analysis of methylation changes between septic and nonseptic patients identified overlap between the methylation profile present in septic patients and profiles commonly associated with IFN response ( It should be noted, however, that there are far more datasets for genes regulated by type I than types II or III. This imbalance introduces the risk of false negatives and bias for the underrepresented types II and III. In this instance, the DMRs identified in our study overlapped with genes involved in antigen presentation and histone methyltransferases. This raises the possibility that fundamental differences exist in the epigenetic regulation of the innate immune response and adaptive immune system in septic versus nonseptic critically ill patients. Using STRING, we were able to determine not only that the putative gene products could potentially interact and form interaction networks but also that the purported networks had significantly more interactions than would be expected by chance alone. Further studies will be critical in elucidating the nature, trajectory, and impact of this overlap and putative interactions in septic critically ill patients.

Although RNA samples were not available for the EPSIS cohort, we identified seven external transcriptomic cohorts from the GEO repository comprising both septic and nonseptic patients, with which to validate our data. Over 80% of DMR-associated genes showed differential expression in at least one external sepsis cohort. Furthermore, DMR-associated genes were significantly more likely to be differentially expressed in these cohorts than non-DMR–associated genes. We did not take into account “direction” of expression change because differential methylation can both increase and decrease expression, making predictions regarding direction of change unreliable (12,43).

Adoption studies have demonstrated that susceptibility to infection is only slightly correlated in biologic siblings although risk of death is strongly correlated (44); emphasizing that although susceptibility may be a product of the chance of exposure, survival is determined at the level of the genome. Genome-wide associated screens in sepsis (as well as other complex diseases) have shown limited utility in explaining this heritability (45,46). It is tempting to speculate that the “missing heritability” is at least in part explained by epigenetics. Epigenetic changes have already been linked to immune activation and tolerance (47,48) in vivo and may contribute to protracted inflammation, organ failure, persistent immune suppression, development of severe secondary infections, and death.

In spite of the heterogeneity of the EPSIS cohort, comprising patients with a variety of infectious and noninfectious illnesses, we were able to identify a large number of DMRs that correlated with the septic state. By matching patients for severity of illness and mortality, we attempted to isolate methylation changes associated with sepsis rather than critical illness.

The use of whole blood samples is both a strength and limitation of our study. Analysis of mixed cell populations has greater power to detect net changes in common immune cell types and less power to detect small changes within less numerous cell subtypes. As an easily accessible compartment of the immune system, however, whole blood has been the tissue of choice in sepsis transcriptomic studies to date. For biomarker studies—in which the question of causality is not of primary concern—this straightforward design may be sufficient. In the EPSIS cohort, we did not detect significant differences in cell proportions between septic and nonseptic patients using cell counts collected on day 1 of ICU admission or inferred cell counts based on the Houseman deconvolution method (49) (Supplemental Fig. 4, Supplemental Digital Content 12,; and legend, Supplemental Digital Content 13, Furthermore, variation in cell proportions was specifically addressed as a confounder in our analysis (49). It is, therefore, unlikely that cell proportion differences are driving the DNA methylation differences identified between the septic and nonseptic patient populations. Further studies will be required to address the issue of DNA methylation over time.

WGCNA represents a powerful systems biology technique to identify groups of genes whose behavior correlates with each other and with specific clinical traits, such as severity of illness, shock, length of ICU stay, and survival. In this study, comethylation modules correlated with important clinical traits in both septic and nonseptic patients. Nonseptic modules showed a higher degree or preservation in the sepsis patients than vice versa, suggesting that although many methylation pathways are common to septic and nonseptic critical illness, some methylation pathways may be unique to sepsis. Accordingly, our work lends support to the hypothesis that epigenetic changes play a role in sepsis, not only as putative novel biomarkers but potentially as causative agents underscoring phenotypic features of severity and outcomes. DNA methylation represents a new avenue for diagnosis, prognosis, and possibly therapy, of sepsis.


We gratefully acknowledge the administrative assistance from the DNA as a Prognostic Marker in ICU patients (DYNAMICS) research coordinators and the support from the patients and families who made this study possible.


1. Singer M, Deutschman CS, Seymour CW, et al. The Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3). JAMA 2016; 315:801–810
2. Tompkins RG. Genomics of injury: The glue grant experience. J Trauma Acute Care Surg 2015; 78:671–686
3. Wright ML, Anderson CM, Uthus EO, et al. Validation of DNA methylation patterns: Potential biomarker for heritable risk of preeclampsia. West J Nurs Res 2012; 34:1074–1075
4. Lofton-Day C, Model F, Devos T, et al. DNA methylation biomarkers for blood-based colorectal cancer screening. Clin Chem 2008; 54:414–423
5. Warren JD, Xiong W, Bunker AM, et al. Septin 9 methylated DNA is a sensitive and specific blood test for colorectal cancer. BMC Med 2011; 9:133
6. Nimmo ER, Prendergast JG, Aldhous MC, et al. Genome-wide methylation profiling in Crohn’s disease identifies altered epigenetic regulation of key host defense mechanisms including the Th17 pathway. Inflamm Bowel Dis 2012; 18:889–899
7. Dayeh T, Volkov P, Salö S, et al. Genome-wide DNA methylation analysis of human pancreatic islets from type 2 diabetic and non-diabetic donors identifies candidate genes that influence insulin secretion. PLoS Genet 2014; 10:e1004160
8. Toperoff G, Aran D, Kark JD, et al. Genome-wide survey reveals predisposing diabetes type 2-related DNA methylation variations in human peripheral blood. Hum Mol Genet 2012; 21:371–383
9. Liang L, Willis-Owen SAG, Laprise C, et al. An epigenome-wide association study of total serum immunoglobulin E concentration. Nature 2015; 520:670–674
10. Castelo-Branco P, Choufani S, Mack S, et al. Methylation of the TERT promoter and risk stratification of childhood brain tumours: An integrative genomic and molecular study. Lancet Oncol 2013; 14:534–542
11. Jaenisch R, Bird A. Epigenetic regulation of gene expression: How the genome integrates intrinsic and environmental signals. Nat Genet 2003; 33 (Suppl):245–254
12. Leenen FAD, Muller CP, Turner JD. DNA methylation: Conducting the orchestra from exposure to phenotype? Clin Epigenet 2017; 1–15
13. Gamper CJ, Agoston AT, Nelson WG, et al. Identification of DNA methyltransferase 3a as a T cell receptor-induced regulator of Th1 and Th2 differentiation. J Immunol 2009; 183:2267–2276
14. Poplutz MK, Wessels I, Rink L, et al. Regulation of the Interleukin-6 gene expression during monocytic differentiation of HL-60 cells by chromatin remodeling and methylation. Immunobiology 2014; 219:619–626
15. Shuto T, Furuta T, Oba M, et al. Promoter hypomethylation of Toll-like receptor-2 gene is associated with increased proinflammatory response toward bacterial peptidoglycan in cystic fibrosis bronchial epithelial cells. FASEB J 2006; 20:782–784
16. Unterberg M, Kreuzer MJ, Schäfer ST, et al. NFKB1 Promoter DNA from nt+402 to nt+99 is hypomethylated in different human immune cells. PLoS One 2016; 11:e0156702
17. Singer BD, Mock JR, Aggarwal NR, et al. Regulatory T cell DNA methyltransferase inhibition accelerates resolution of lung inflammation. Am J Respir Cell Mol Biol 2015; 52:641–652
18. Thangavel J, Samanta S, Rajasingh S, et al. Epigenetic modifiers reduce inflammation and modulate macrophage phenotype during endotoxemia-induced acute lung injury. J Cell Sci 2015; 128:3094–3105
19. Shih CC, Liao MH, Hsiao TS, et al. Procainamide inhibits DNA methylation and alleviates multiple organ dysfunction in rats with endotoxic shock. PLoS One 2016; 11:e0163690
20. Jackson Chornenki NL, Coke R, Kwong AC, et al. Comparison of the source and prognostic utility of cfDNA in trauma and sepsis. Intensive Care Med Exp 2019; 7:29
21. Bibikova M, Barnes B, Tsan C, et al. High density DNA methylation array with single CpG site resolution. Genomics 2011; 98:288–295
22. Morris TJ, Butcher LM, Feber A, et al. ChAMP: 450k chip analysis methylation pipeline. Bioinformatics 2014; 30:428–430
23. Teschendorff AE, Menon U, Gentry-Maharaj A, et al. An epigenetic signature in peripheral blood predicts active ovarian cancer. PLoS One 2009; 4:e8274
24. Chen YA, Lemire M, Choufani S, et al. Discovery of cross-reactive probes and polymorphic CpGs in the Illumina Infinium HumanMethylation450 microarray. Epigenetics 2013; 8:203–209
25. Zhou W, Laird PW, Shen H. Comprehensive characterization, annotation and innovative use of Infinium DNA methylation BeadChip probes. Nucleic Acids Research 2016; 2:gkw967
26. Sofer T, Schifano ED, Hoppin JA, et al. A-clustering: A novel method for the detection of co-regulated methylation regions, and regions associated with exposure. Bioinformatics 2013; 29:2884–2891
27. Martin TC, Yet I, Tsai PC, et al. coMET: visualisation of regional epigenome-wide association scan results and DNA co-methylation patterns. BMC Bioinformatics 2015; 16:131
28. Lander ES, Linton LM, Birren B, et al.; International Human Genome Sequencing Consortium: Initial sequencing and analysis of the human genome. Nature 2001; 409:860–921
29. Scholz A, Plate KH, Reiss Y. Angiopoietin-2: A multifaceted cytokine that functions in both angiogenesis and inflammation. Ann N Y Acad Sci 2015; 1347:45–51
30. Esteller M. Cancer epigenomics: DNA methylomes and histone-modification maps. Nat Rev Genet 2007; 8:286–298
31. Edgar R, Domrachev M, Lash AE. Gene expression omnibus: NCBI gene expression and hybridization array data repository. Nucleic Acids Res 2002; 30:207–210
32. Demaret J, Venet F, Plassais J, et al. Immunology letters. Immunol Lett 2016; 178:122–130
33. Xiao W, Mindrinos MN, Seok J, et al.; Inflammation and Host Response to Injury Large-Scale Collaborative Research Program: A genomic storm in critically injured humans. J Exp Med 2011; 208:2581–2590
34. Merkenschlager M. Jarid2 links microRNA and chromatin in Th17 cells. Immunity 2014; 40:855–856
35. Shakespear MR, Halili MA, Irvine KM, et al. Histone deacetylases as regulators of inflammation and immunity. Trends Immunol 2011; 32:335–343
36. Pandzic T, Rendo V, Lim J, et al. Somatic PRDM2 c.4467delA mutations in colorectal cancers control histone methylation and tumor growth. Oncotarget 2017; 8:98646–98659
37. Voineagu I, Wang X, Johnston P, et al. Transcriptomic analysis of autistic brain reveals convergent molecular pathology. Nature 2011; 474:380–384
38. Bettencourt C, Ryten M, Forabosco P, et al.; United Kingdom Brain Expression Consortium: Insights from cerebellar transcriptomic analysis into the pathogenesis of ataxia. JAMA Neurol 2014; 71:831–839
39. Dhas DB, Ashmi AH, Bhat BV, et al. Comparison of genomic DNA methylation pattern among septic and non-septic newborns - an epigenome wide association study. Genom Data 2015; 3:36–40
40. Takeyama N, Tanaka T, Yabuki T, et al. Effect of interferon gamma on sepsis-related death in patients with immunoparalysis. Crit Care 2004; 8:P207
41. Rackov G, Shokri R, De Mon MÁ, et al. The role of IFN-β during the course of sepsis progression and its therapeutic potential. Front Immunol 2017; 8:493
42. González-Navajas JM, Lee J, David M, et al. Immunomodulatory functions of type I interferons. Nat Rev Immunol 2012; 12:125–135
43. Castelo-Branco P, Leão R, Lipman T, et al. A cancer specific hypermethylation signature of the TERT promoter predicts biochemical relapse in prostate cancer: A retrospective cohort study. Oncotarget 2016; 7:57726–57736
44. Petersen L, Andersen PK, Sørensen TI. Genetic influences on incidence and case-fatality of infectious disease. PLoS One 2010; 5:e10603
45. Rautanen A, Mills TC, Gordon AC, et al.; ESICM/ECCRN GenOSept Investigators: Genome-wide association study of survival from sepsis due to pneumonia: An observational cohort study. Lancet Respir Med 2015; 3:53–60
46. The CARDIoGRAMplusC4D Consortium, DIAGRAM Consortium, CARDIOGENICS Consortium, Large-scale association analysis identifies new risk loci for coronary artery disease. Nat Genet 2012; 45:25–33
47. Saeed S, Quintin J, Kerstens HH, et al. Epigenetic programming of monocyte-to-macrophage differentiation and trained innate immunity. Science 2014; 345:1251086
48. Ansel KM, Djuretic I, Tanasa B, et al. Regulation of Th2 differentiation and Il4 locus accessibility. Annu Rev Immunol 2006; 24:607–656
49. Houseman EA, Accomando WP, Koestler DC, et al. DNA methylation arrays as surrogate measures of cell mixture distribution. BMC Bioinformatics 2012; 13:86

critical illness; deoxyribonucleic acid methylation; epigenetics; epigenome-wide association study; sepsis; weighted gene coexpression network analysis

Supplemental Digital Content

Copyright © 2020 by the Society of Critical Care Medicine and Wolters Kluwer Health, Inc. All Rights Reserved.