The CGEM is a four-tiered approach (Figure 1) comprising the following steps: (1) Generation of a database consisting of genes/proteins with known roles in infectious disease signaling termed infectious disease genes (IDGs); (2) Development of a database of genes/proteins that are modified at an epigenomic, genomic, or proteomic level in response to prenatal iAs or Cd exposure in human pregnancy cohorts termed exposure responsive genes (ERGs); (3) An analytical comparison of the IDG and ERG databases resulting in the identification of a database of infection and exposure-related (IER) genes; and (4) Systems-level analysis to identify pathways prioritized for their likely involvement in infection and exposure-related responses.

The CGEM was used in the present study to elucidate key biological pathways likely involved in environmental exposure-associated changes in infectious disease-related signaling, using prenatal exposure to arsenic and cadmium as the exposures of interest.

In order to evaluate whether there are a common set of genes/proteins related to infectious disease that also overlap with genes that are associated with prenatal exposures to iAs or Cd, a database of n = 1085 unique infectious disease genes (IDGs) was developed (Table S1). Specifically the IDG database was developed by integrating genes/proteins representing 23 pathways related to viral, bacterial, and parasitic infection. To note, amongst these 1085 genes are a set of 15 genes with single nucleotide polymorphisms (SNPs) known to influence response to infectious disease [31] (Table S1).

A separate exposure responsive gene (ERG) database was developed consisting of n = 950 unique genes obtained through a cross-study analysis (Table S2). Specifically, these genes have been identified to have altered levels of 5-methyl cytosine methylation, altered mRNA abundance and/or altered protein expression levels associated with prenatal exposure to iAs and/or Cd in twelve human pregnancy cohorts. To note, environmental exposures can cause shifts in immune cell populations [32], and some of the human pregnancy cohort studies did not report to have controlled for the impact of white blood cell shifts in response to iAs or Cd. Still, many of these prior studies did control for the impact of white blood cell shifts [33,34,35,36]. For instance, Koestler et al. used a statistical methodology, referred to as a cell mixture deconvolution methodology, to infer changes in cord blood leukocyte distributions between exposure quartiles using DNA methylation signatures [34]. In addition, Rager et al., Rojas et al., and Sanders et al. compared genes identified as iAs/Cd-associated to previously published lists of immune cell subset-specific genes [33,35,36]. Rager et al. found a small percentage (2%) of iAs-associated genes with expression levels specific to immune cell subsets [36], while Rojas et al. did not find any iAs-associated genes with differential methylation specific to immune cell subsets [35]. In the study by Sanders et al., only one gene with differential methylation associated with Cd was specific to an immune cell subtype [33]. Thus the observed changes in DNA methylation, mRNA abundance, and/or protein expression are not reflective of changes in immune cell populations.

The metalloid/metal iAs and Cd were selected for evaluation, as there is substantial literature relating prenatal exposure in human cohort studies with changes to the epigenome, genome, and/or proteome. It is important to consider for these studies that while iAs tends to more readily cross the placental barrier, cord blood concentrations of both iAs and Cd increase as maternal exposure increases [21,37,38], thus fetal exposure during critical times of development is occurring. The studies used to generate the ERG database assessed human populations with iAs exposure levels comparable to those that have been associated with increased risk of childhood infection (Table 1 and Table 2). To our knowledge, this represents all studies to date where gene-specific information could be compiled from pregnancy cohorts and where data overlapped with the IDGs.

Comparing the list of IDGs against the ERGs identified a common set of 82 genes representing those that play a role both in infectious disease-related signaling as well as those that are targets for modulation/modification under conditions of prenatal iAs or Cd exposure, hitherto referred to as infection and exposure-related (IER) genes (Table S3). Additionally, 16 of the 82 (20%) IER genes are known to play a role in all three major infectious disease pathways, viral, bacterial, and parasitic pathways, including interferon γ receptor 1 (IFNGR1), interferon γ (IFNG), tumor necrosis factor (TNF), caspase 9, apoptosis-related cysteine peptidase (CASP9), FBJ murine osteosarcoma viral oncogene homolog (FOS), major histocompatibility complex, class II, DP α 1 (HLA-DPA1), major histocompatibility complex, class II, DQ α 1 (HLA-DQA1), intercellular adhesion molecule 1 (ICAM1), interleukin 1 β (IL-1β), interleukin 8 (IL8), jun proto-oncogene (JUN), nuclear factor of κ light polypeptide gene enhancer in B-cells 1 (NF-κB1), nuclear factor of κ light polypeptide gene enhancer in B-cells inhibitor, α (NF-κBIA), phosphoinositide-3-kinase, regulatory subunit 1 (α) (PIK3R1), signal transducer and activator of transcription 1 (STAT1), and toll-like receptor 9 (TLR9) (Table S3).

In order to determine whether the IER genes enrich for specific biological pathways, a systems-level analysis was carried out using the 82 IER genes. Perhaps not surprisingly, given the IDG database into which the information was compiled, the analysis identified enrichment for canonical pathways highly relevant to infectious disease and inflammatory response signaling. Interestingly, many of the IER genes were identified to belong to the glucocorticoid receptor (GR) signal transduction pathway (p = 1.58 × 10⁻¹⁸) where 19 of the 82 IER genes are known to be involved. Another GR-associated pathway was also enriched, namely the tumor necrosis factor receptor 1 (TNFR1) (p = 5.01 × 10⁻¹³) pathway, where nine of the 82 IER genes were present. Also relevant to the GR signal transduction pathway, the toll-like receptor (TLR) pathway was enriched (p = 6.31 × 10⁻¹³) wherein eight of the IER genes were present (Figure 2, Table S4).

The IER genes were analyzed for their known associations with potential cellular regulators including, but not limited to, transcription factors, signaling molecules, and drugs. The 82 IER genes were enriched for their association with numerous upstream regulators including the following six most significant regulators: Lipopolysaccharide (LPS) (p = 4.98 × 10⁻³⁸), colony stimulating factor 2 (CSF2) (p = 2.23 × 10⁻³⁰), TNF (p = 9.72 × 10⁻³⁰), polyinosinic-polycytidylic acid (poly(I:C)) (p = 1.18 × 10⁻²⁸), IL1β (p = 4.74 × 10⁻²⁷), and NF-κB (p = 1.56 × 10⁻²⁴). Of note, LPS has been shown to be associated with 49 of the 82 (60%) IER genes (Table S5). Further supporting both the pathway and predicted upstream regulators analysis, a separate analysis of the DNA sequences within the promoter regions of the IER genes also identified an enrichment for genes with binding sites for the GR, specifically containing “negative” glucocoticoid response elements (NGRE) (3.91 × 10⁻⁶), as well as sites bound by the NF-κB transcription factor family (3.72 × 10⁻¹⁴) (Table S6).

As validation of the relationship between the ERGs and iAs and Cd, the CTD was queried. This manually-curated database is specific for environmental contaminants and their relationships to genes including perturbations at the mRNA and protein level collected from published toxicological and epidemiological studies. At the time of our analyses, the CTD analyzed over 95,000 studies to derive over 15 million toxicogenomic relationships between approximately 11,000 chemicals, 27,000 genes, and 5900 diseases [46]. Thus the CTD includes data from both in vitro and in vivo animal model-based systems while the ERG list was compiled solely from data derived from human pregnancy cohorts.

Supporting the analytical approach, the CTD contained information that 50 of the 82 (61%) IER genes have been shown to be modulated by iAs/iAs metabolites and 34 of the 82 (41%) IER genes have been shown to be modulated by Cd in other studies (Table S7). These percentages of genes are much higher than those produced using a randomly generated gene set of the same size (data not shown). It is important to note as further validation of the relationship between iAs and Cd and the IER genes that many of the studies within CTD evaluated the effects of exposure using cells/animals that were treated with iAs/Cd and compared their responses to untreated, negative controls. For instance, an in vitro study within CTD using human bronchial epithelial cells compared gene expression profiles from unexposed cells versus cells exposed to iAs and iAs metabolites and identified four genes associated with exposure that are contained in our IER gene list: BH3 interacting domain death agonist (BID), NF-κB1, PIK3R1, and tumor protein p53 (TP53) [47]. Additionally, many of the IER genes have been shown to be modulated in response to exposure to a range of other environmental toxicants including toxic metals such as beryllium, chromium, cobalt, copper, lead, mercury, nickel, selenium, and zinc. Interestingly there is also evidence for association of the IER genes with other non-metal toxicants including acrolein, asbestos, atrazine, benzene, benzo(a)pyrene, bisphenol A, estradiol, formaldehyde, nitric oxides, particulate matter, and silicon dioxide (Table S7). These data support that, in addition to iAs and Cd, the identified set of 82 IER genes are dysregulated by other environmental contaminants in various model systems. These findings have implications for the putative role that these contaminants play in environmentally-induced infectious disease susceptibility.

To generate a database of key genes involved in infectious disease-related signaling, infectious disease genes (IDGs) were compiled including genes encoding proteins involved in canonical pathways relevant to infectious disease signaling. A total of 23 Kyoto Encyclopedia of Genes and Genomes (KEGG) [75] viral, bacterial and parasitic infectious diseases pathways were included: Vibrio cholera, Helicobacter pylori, pathogenic Escherichia coli, Salmonella, Shigellosis, pertussis, legionellosis, Staphylococcus aureus, tuberculosis, generalized bacterial invasion of epithelial cells, human T-lymphotropic virus, measles, influenza A, hepatitis B, hepatitis C, herpes simplex, epstein-Barr virus, amoebiasis, malaria, toxoplasmosis, leishmaniasis, African trypanosomiasis, and American trypanosomiasis. These genes were associated with the above mentioned KEGG infectious disease pathways to produce a final list of n = 1085 IDGs.

In order to assess the relationship between iAs or Cd exposure and infectious disease, the database of IDGs was compared with genes previously altered in response to prenatal exposure to iAs or Cd. Specifically, twelve datasets were gathered from previous human cohort studies evaluating mRNA changes, DNA methylation changes, or protein expression changes associated with prenatal exposure to iAs or Cd. Five of the twelve gene lists comprise those with altered gene (mRNA) or those that encoded proteins with expression levels associated with in utero exposure to iAs: (1) Fry et al., identified 447 differentially expressed genes from a genome-wide analysis in cord blood leukocytes of 32 pregnant women from the Ron Pibul and Bangkok districts of Thailand [42]; (2) Ahmed et al., performed a gene-specific analysis of 18 inflammation-associated cytokines in cord blood leukocytes and found three differentially expressed proteins in a cohort of 130 pregnant women from Matlab, Bangladesh [39]; (3) Rager et al., conducted a genome-wide analysis in cord blood leukocytes and found 334 differentially expressed mRNAs in a cohort of 40 mother-baby pairs from the Biomarkers of Exposure to ARsenic (BEAR) pregnancy cohort in Gómez Palacio, Mexico [36]; (4) Fei et al., investigated the expression of nine predicted As-associated genes in placental tissue from 133 pregnant women from New Hampshire, U.S [41]; and (5) Bailey et al., investigated As-associated changes to the prenatal proteome and identified 111 proteins associated with prenatal As exposure in the cord blood serum or plasma of 50 newborns from the BEAR cohort in Gómez Palacio, Mexico [76]. Seven of the twelve gene lists comprise genes identified in studies assessing DNA methylation levels: (1) Kippler et al., analyzed genome-wide DNA methylation of cord blood leukocytes in response to Cd exposure from 127 mother-baby pairs from Matlab, Bangladesh and identified 54 genes with differential methylation most associated with Cd [45]; (2) Kile et al., analyzed global DNA methylation and two gene-specific tumor suppressors of cord blood leukocytes in response to prenatal iAs in 113 mother-baby pairs from Sirajdikhan Upazila, Bangladesh [44]; (3) Intarasunanont et al., conducted a global DNA methylation and TP53 specific analysis of cord blood leukocytes from 71 newborns with prenatal iAs exposure from southern Thailand [43]; (4) Koestler et al., performed a genome-wide methylation study of cord blood leukocytes from 134 infants nested in the New Hampshire Birth Cohort Study (NHBCS) and identified 68,353 CpG loci associated with iAs exposure [34]; (5) Sanders et al., performed a genome-wide gene-specific DNA methylation study of 17 mother-baby pairs from Durham, North Carolina and found 61 genes differentially methylated in cord blood leukocytes from prenatal Cd exposure [33]; (6) Broberg et al., carried out a genome-wide methylation analysis of cord blood leukocytes from 127 infants from a pregnancy cohort in Matlab, Bangladesh and identified three CpG sites in boys, but none in girls, that were significantly associated with As exposure after adjustment for multiple comparisons [40]; and (7) Rojas et al., conducted a genome-wide methylation analysis of 38 mother-baby pairs from the previously mentioned BEAR cohort in Gómez Palacio, Mexico and identified 54 genes that were both differentially methylated and differentially expressed in response to iAs in cord blood leukocytes. In total, all twelve gene sets were compiled to form the exposure responsive genes (ERGs) database and compared against the IDGs, where genes in common between the IDG and ERG lists were termed the infection and exposure-related (IER) genes.

Of the twelve studies’ gene sets used to generate the ERG database, nine contained genes within the IDG database and were thus included in the final IER gene database. All nine of these studies evaluated gene expression, DNA methylation, or protein expression within human cord blood leukocytes [33,35,36,39,42,43,44,45,76]. It is notable that some of these published studies included weaknesses in design and implementation. Possible limitations include small sample size and incomplete analysis of possible confounders, such as other environmental exposures. Additionally, variations between studies, including exposure duration and demographic information (e.g., subject age, gender, and race), may influence the reported relationships between prenatal exposure and genomic/epigenomic/proteomic alterations. The iAs and Cd exposure levels found in these studies presented a range of levels, and the studies utilized various biological medias to measure for metals exposure including blood, urine, and nail clippings (Table 1). Still, many of the ERGs (n = 67 genes) overlapped across studies, indicating consistency amongst findings and supporting biological plausibility.

Pathway analysis was performed to understand the systems-level relationships amongst the IER genes. For this analysis, IER genes were assessed for their associations to canonical pathways, enabled through Ingenuity Pathway Analysis (IPA) (Ingenuity Systems^®, Redwood City, CA, USA). Canonical pathways were identified as enriched using the right-tailed Fisher’s Exact test, as performed previously [33], where significance was set at p < 0.0001.

An upstream regulator analysis was also carried out using IPA, where transcription factors, growth factors, cytokines, chemicals, drugs, and other molecules with known influences on the IER genes were analyzed as potential regulators. Using a separate in silico database that focuses on transcription factors, the IER genes were analyzed for overrepresented transcription factor binding site enrichment using Genomatix’s Overrepresented TFBS tool (Genomatix Software Inc., Ann Arbor, MI, USA) to determine if the IER genes are potentially regulated by common transcription factors [77]. Promoter regions of the IER genes were selected with preference for experimentally verified 5' complete transcripts, number of cap analysis gene expression (CAGE) tags, and relevant transcripts. Promoter regions were defined as 500 base pairs upstream and 100 base pairs downstream of the transcription start site. Resulting transcription factor families that were significantly (z-score > 1.96, p-value < 0.05) enriched for binding sites within the IER genes were identified.

Gene-chemical interactions were assessed for the IER genes using data retrieved from the Comparative Toxicogenomics Database (CTD) (North Carolina State University, Raleigh, NC, USA) [46]. The CTD curates known interactions between environmental chemicals, genes, and known relationships to disease. As of 2012, the CTD had analyzed over 95,000 studies to derive over 15 million toxicogenomic relationships between approximately 11,000 chemicals, 27,000 genes, and 5900 diseases [46].