A functional mechanism for a non-coding variant near AGTR2 associated with risk for preterm birth

The cell line used in EMSA and luciferase assay was an immortalized human endometrium stromal fibroblast cell line (T HESC, Mor lab, Yale University, corresponding to ATCC CRL-4003). Culture and further decidualization conditions are described in [31] with minor modifications. Specifically, the decidualization medium is supplemented with 2% FBS, 0.5 mM cAMP(Sigma 23583–48-4), 1uM MPA (Sigma, M1629), 1*Pen/Strep and 0.24 g NaHCO₃ per 200 mL DMEM medium.

We performed linkage disequilibrium (LD) expansion using the PLINK software package (version 1.90b) with EU ancestry (LD r² > 0.8). This resulted in a total of 49 candidate SNPs. We then annotated these SNPs with eQTL data. We collected functional genomic datasets including ENCODE [28], Cistrome [32], PAZAR [33], Re-Map [34], and NIH Roadmap Epigenomics [35]. All datasets were indexed by their genomic coordinates, which were used to intersect with the LD expanded list of SNPs of interest. To help prioritize functional SNPs, SNPs with overlapping TF ChIP-seq signals were scanned for the putative binding signals using TF binding models based on protein binding microarray data using the Cis-BP database. After cross reference of the above results, we identified rs7889204 as the candidate causal SNP and FOXA1, CEBPB, and HOXA10 as transcription factors with predicted differential allelic binding at this T/C SNP.

We examined the tissue-specific expression of AGTR2 using the Genotype-Tissue Expression (GTEx) Portal (https://​gtexportal.​org/​). Significant single-tissue eQTLs for AGTR2 in all tissues were extracted from GTEx Analysis Release V8. The distribution of the eQTLs and the pairwise LD pattern were visually inspected using the GTEx Locus Browser (Fig. 2A). Single-tissue eQTLs for rs7889204 were extracted and examined using eQTL violin plot (Fig. 2B).

We used rs7889204 as the bait on the HiC page under the interaction visualization tab on the 3D Genome Interactive Viewer and Database [36] (3DIV) (http://​kobic.​kr/​3div/​). We followed the tutorial using default settings with the exception that we set the threshold on distance normalized interaction frequency to be 2.00. We started with a selection of all possible data sets and then ranked them by gene name. Chromatin looping interaction signal was found between this SNP and the promoter region of AGTR2 in a fibroblast cell context (Fig. 2C).

Electrophoretic mobility shift assays (EMSAs) were performed as previously described [29, 30] to determine whether the candidate variant at the AGTR2 locus differentially affected the binding of the predicted transcription factors, namely Forkhead box protein A1 (FOXA1), CCAAT/enhancer-binding protein beta (CEBPB) and Homeobox protein A10 (HOXA10). Briefly, we prepared nuclear protein extracts as previously described with minor modifications [29]. Decidualized hESF cells were trypsinized, washed twice with cold PBS, counted, resuspended in PBS as 10⁷ cells/ml, then centrifuged 3300 g for 2 min, 4 °C. PBS was removed, and 10⁷ cells were resuspended in 400 μl CE buffer (10 mM HEPES (pH7.9), 10 mM KCl, 0.1 mM EDTA, 1 mM dithiothreitol (DTT), and 1X HALT protease/phosphatase inhibitor (ThermoFisher)), then incubated on ice for 15 min. Cells were mixed with 25 μl of 10% Nonidet P-40 by pipetting, and then nuclei were pelleted at maximum speed for 3 min, 4 °C. Nuclei were resuspended in 30 μl NE buffer (20 mM HEPES (pH7.9), 0.4 M NaCl, 1 mM EDTA, 1 mM DTT, 1X protease/phosphatase inhibitor) by vortexing, incubated on ice for 10 min, then centrifuged 3300 g for 2 min, 4 °C. Supernatants (nuclear proteins) were removed from pelleted debris and stored in small aliquots at − 80 °C. Protein concentration was determined by BCA protein assay (ThermoFisher). SNP rs7889204 was used to prepare 5′ IMR700 dye–labeled 35-bp duplexed oligonucleotide probes (Integrated DNA Technologies). SNP oligonucleotide sequences were obtained from the National Center for Biotechnology Information dbSNP database (https://​www.​ncbi.​nlm.​nih.​gov/​snp) by using human genome build 37. The sequence of probes are as follows:

EMSA binding reactions contained 6 μg of nuclear extract, 100 fmol of the labeled probe, 2μL of 10 × binding buffer, 1 μg of Poly(dI-dC), 1μL of 40% glycerol, 2μL of 25 mmol/L dithiothreitol/2.5% Tween 20, 1μL of 1% NP-40, 1μL of 100 mM MgCl₂, and ultrapure water (LI-COR Biosciences) in a total reaction volume of 20μL. Products were electrophoresed on native 5% Tris–Borate-EDTA precast polyacrylamide gels (Bio-Rad, Life Sciences, Hercules, Calif). Gels were prerun at 70 V/cm for 30 min in 0.5 × nondenaturing Tris–Borate-EDTA buffer, and samples were loaded, run for 60 min, and then imaged with an Odyssey CLx imager (LI-COR Biosciences). The hESF (human endometrium stromal fibroblast cell) cell extracts were supplemented with purified TF protein as above, as endogenous TF protein was not initially detected on EMSA. All candidate TF proteins examined were generated from an overexpression vector after nuclear lysate purification and tagging with a DDK motif (Origene). 50fmol fluorescent oligo DNAs were then added to the appropriate protein/binding mix and incubated for 20 min at room temp. For supershift assays, anti-DDK antibody (Origine #TA50011-100) was incubated with the nuclear lysate/binding buffer for 20 min prior to the addition and incubation with oligo DNA. 1 × orange loading dye (LI-COR kit) was added to samples, which were then resolved on (pre-cast, pre-run at 100 V for 60 min) 6% TBE gels (Novex, ThermoFisher) in 0.5 × TBE buffer for 80 min at 80 V (4 °C). Fluorescent bands were then imaged using a LI-COR chemiluminescent imaging system. EMSA experiments display representative panels of 2–3 replicates.

Luciferase reporter constructs were cloned by using the pNL1.1 nanoluciferase reporter vector (Promega, Madison, WI) as a backbone. For the AGTR2 promoter reporter construct, PCR primers were used to amplify a 1 kb sequence including the first 816 base pairs upstream of the transcription start site, and the region was subsequently cloned into the pNL1.1 backbone. For the putative AGTR2 regulatory GWAS locus upstream of the AGTR2 promoter, two complementary oligonucleotides were designed to create fragments for insertion into this backbone. The GWAS regulatory region was comprised of a 4.3-kb region surrounding rs7889204 (which includes three other SNPs(dbSNP_153), rs4478734 (which is an eQTL for AGTR2 in uterus and adrenal glands but not predicted to alter TF binding), rs9724136(no significant eQTL data in any tissue) and rs2310841(no entry in GTEx portal, not predicted to alter TF binding). The region was inserted into the pNL1.1 nanoluciferase reporter construct in front of the AGTR2 promoter. Site-directed mutagenesis was performed to generate a construct with all three non-reference SNPs and a construct with the non-reference (risk) allele of rs7889204 using the GeneArt Site-Directed Mutagenesis System (Invitrogen, Darmstadt, Germany). Sequences were confirmed by Sanger sequencing.

The AGTR2 adjacent genetic association locus falls within a non-coding region of the human genome on chromosome X, which suggests that one or more of these variants might functionally act as enhancers by altering gene regulatory mechanisms. Although AGTR2 was the closest gene to this identified region and biologically plausible as the consequential gene affecting preterm birth as previously reviewed, the index SNP from the discovery phase was located over 200 kb from the AGTR2 transcription start site (TSS). As such, we first determined whether this GWAS locus interacts with the gene AGTR2 given the substantial distance between regions. It is promising to know that the locus falls in the same TAD (topologically associated chromatin domains) domain with the AGTR2 gene in human ESC (embryonic stem cells) and IMR90 cell lines [37]. TAD domains are known to be conserved among cell types and even different species. They are the unit of genome organization that allow for coordination of gene regulation programs and integration of transcriptional signals [38]. Next, a total of 49 SNPs were identified as close proxies in linkage disequilibrium (LD) with the index SNP from the discovery phase using criteria r² > 0.8 using the PLINK software package (version 1.90b) [39] with European (EU) ancestry (Fig. 1 and Additional file 1, spreadsheet SNPs in LD). Out of these 49 SNPs, 35 are expression quantitative trait loci (eQTLs) for the AGTR2 gene in the uterus using data from the GTEx database [40] (Fig. 1 and Additional file 1, spreadsheet GTEx V7 eQTLs in uterus). Collectively, these data indicate that AGTR2 is a likely gene target explaining the non-coding GWAS association.

We then used publicly available functional genomics datasets to prioritize the likely causal SNP(s) from the 35 SNPs identified in the previous analysis. The logic underlying this approach is the non-coding region harboring the causal SNP probably functions as an enhancer to regulate AGTR2 gene expression, and thus will be more likely to be marked with genetic and/or epigenetic marks that are typically associated with enhancers. We collected functional genomic datasets from a range of sources, including ENCODE [28], Cistrome [32], PAZAR [33], Re-Map [34], and NIH Roadmap Epigenomics [35]. All datasets were indexed by their genomic coordinates, which were used to intersect with the LD expanded list of SNPs of interest. We prioritized SNPs based on their overlap with both likely regulatory regions (e.g., regions with histone marks such as H3K27ac) and TF ChIP-seq datasets (Fig. 1 and Additional file 1, spreadsheets ChIP-seq no-TF overlaps and ChIP-seq TF overlaps). Collectively, rs7889204 emerged as the most likely functional candidate SNP (Fig. 1 and Additional file 1, spreadsheet ChIP-seq TF-based diff binding). Subsequent functional analysis thus focused on this potential causal SNP. A flowchart is provided (Fig. 1) that documents the complete process — results and intermediate output can be found in Additional file 1.

Differential TF binding analysis for rs7889204 performed using the CisBP database [41] yielded three strong candidates, all of which are predicted to bind more strongly to the T allele: CEBPB, HOXA, and FOX family members. We selected FOXA1 for experimental validation because the SNP is inside FOXA1 ChIP-seq peaks in multiple experiments, and CEBPB and HOXA10 were selected based on known biological roles.

We next examined the effect of rs7889204 on AGTR2 gene expression using the GTEx dataset (Fig. 2). Figure 2A depicts the AGTR2 locus, showing the position of AGTR2 gene (marked in red) and the candidate SNP (middle panel, indicated by a red arrow). A zoomed-in view is also provided showing the effect of this SNP on the expression of the AGTR2 gene in different tissues including the adrenal gland, artery-aorta, lung, and uterus. This eQTL analysis suggests that the region harboring rs7889204 has the largest effect in uterine tissue, with higher AGTR2 expression observed for the reference (non-risk) “T” allele and lower expression for the non-reference (risk) “C” allele (Fig. 2B).

We next examined if the region surrounding this SNP has the potential to form enhancer-promoter interaction with AGTR2, which is essential to regulate target gene expression. We used rs7889204 as the bait on the 3D Genome Interactive Viewer and Database [36] (3DIV) (http://​kobic.​kr/​3div/​). We used default settings on the Hi-C page except we set the threshold on distance normalized interaction frequency to be 2.00. This analysis revealed that the regions harboring rs7889204 do indeed physically interact with the AGTR2 promoter in fibroblasts (Fig. 2C). Although experimental data do not currently exist for examining chromatin interactions between rs7889204 and AGTR2 in the context of preterm/gestational duration, these results suggest that such enhancer-promoter interactions are plausible in disease-relevant cell types.

Being the most abundant cell type in the endometrium, endometrial stromal cells undergo a dramatic morphological and functional differentiation during decidualization and are essential for the establishment of a successful pregnancy [42]. Analysis of existing Hi-C data also suggests that such enhancer-promotor interactions are plausible in a fibroblast cell context (Fig. 2C). However, we were not able to detect endogenous AGTR2 expression in the immortalized endometrium stromal fibroblast cell line (ATCC CRL-4003). It is also absent in the transcriptome study using this cell line before and after decidualization [31]. Gene expression profiling using microarray in the basal plate of the human maternal–fetal interface showed the expression of AGTR2 is higher in the 37–40 week samples than in the 14–18 week samples (P = 0.01, BH-FDR = 0.09) [43], indicating the dynamic expression pattern of this gene. We expanded our search to single-cell transcriptomic studies, but AGTR2 seems absent in the datasets we have checked [44‐47]. E.g., re-analysis of single-cell data from the human early maternal–fetal interface showed AGTR2 is only present in 10 cells out of 70,325 cells (Additional file 2). Our attention was then directed to lung research, in which tissue shows biased AGTR2 expression. Re-analysis of single cell/single nuclei RNA-seq in human uterus tissue from this study [48] shows AGTR2 has low but highly selective and enriched expression in Stromal-2 cell populations as shown in Fig. 3, though the biological context of this data-driven unbiased cell type classification needs further investigation.

The computational analysis suggests that the rs7889204 genomic locus might act as an enhancer to regulate AGTR2 gene expression in the uterine tissue. It further suggests that altered binding of HOXA10, CEBPB, and/or FOXA1 might explain the observed allele-dependent expression of AGTR2 in uterine tissue. Previous studies have shown that HOXA10 might play key roles in directing endometrial stromal cell differentiation and is key for normal implantation [49, 50]; CEBPB is expressed at implantation sites in human and non-human primates, where intense nuclear expression of CEBPB has been observed in decidual stromal cells [51]. No direct connection between FOXA1 and pregnancy has been previously reported. Previous transcriptomic studies in immortalized human endometrium stromal cells before and after induced decidualization show expression of both HOXA10 and CEBPB, but not FOXA1 [31]. This finding is also confirmed by the TaqMan gene expression assay by our group (Additional file 2).

Using an electrophoretic mobility shift assay (EMSA), we detected decreased binding of CEBPB to the risk “C” allele of rs7889204 as predicted using decidualized hESC (human endometrium stromal cell) nuclear lysate with recombinant CEBPB protein (Fig. 4). We also detected decreased binding of HOXA10 to the risk “C” allele of rs7889204 as predicted using decidualized hESC nuclear lysate with recombinant HOXA10 protein (Fig. S1). However, we did not detect substantial differences between the risk “C” and non-risk “T” alleles of rs7889204 using decidualized hESC nuclear lysate with recombinant FOXA1 protein (Fig. S2). Collectively, these results confirm the predictions of stronger binding of the HOXA10 and CEBPB transcription factors to the non-risk allele of rs7889204.

To determine whether the region surrounding rs7889204 possesses transcriptional enhancer activity and if the reference non-risk “T” and non-reference risk “C” alleles demonstrate differences in activity, we performed luciferase reporter experiments in both non-decidualized and decidualized hESC cells. The rs7889204 region and promoter region of AGTR2 were placed in front of the nano-luciferase gene. Nano-luciferase expression was responsive to the AGTR2 promoter both in non-decidualized and decidualized hESCs (Fig. 4); however, we observed greater expression in the AGTR2 promoter-containing pNL1.1 vector after decidualization (Fig. 5). Importantly, the regulatory region containing the risk “C” allele for rs7889204 demonstrated significantly less nano-luciferase expression compared to the reference allele reporter construct. This decrease in nano-luciferase expression for the non-reference allele of rs7889204 was not strongly dependent on cellular conditions, as a significant, albeit slightly, weaker decrease was also observed in non-decidualized endometrial cells (Fig. 5). These results are consistent with the reduced transcription factor binding observed in EMSA (Fig. 4), collectively revealing a mechanism involving weaker CEBPB and HOXA10 binding to the rs7889204 risk allele, leading to reduced AGTR2 expression.