Polymorphisms in STING Affect Human Innate Immune Responses to Poxviruses

We conducted a large genome-wide association study (GWAS) of the immune responses to primary smallpox vaccination in a combined cohort of 1,653 subjects. We did not observe any polymorphisms associated with standard vaccine response outcomes (e.g., neutralizing antibody, T cell ELISPOT response, or T cell cytokine production); however, we did identify a cluster of SNPs on chromosome 5 (5q31.2) that were significantly associated (p-value: 1.3 x 10−12 – 1.5x10−36) with IFNα response to in vitro poxvirus stimulation. Examination of these SNPs led to the functional testing of rs1131769, a non-synonymous SNP in TMEM173 causing an Arg-to-His change at position 232 in the STING protein—a major regulator of innate immune responses to viral infections. Our findings demonstrate differences in the ability of the two STING variants to phosphorylate the downstream intermediates TBK1 and IRF3 in response to multiple STING ligands. Further downstream in the STING pathway, we observed significantly reduced expression of type I IFNs (including IFNα) and IFN-response genes in cells carrying the H232 variant. Subsequent molecular modeling of both alleles predicted altered ligand binding characteristics between the two variants, providing a potential mechanism underlying differences in inter-individual responses to poxvirus infection. Our data indicate that possession of the H232 variant may impair STING-mediated innate immunity to poxviruses. These results clarify prior studies evaluating functional effects of genetic variants in TMEM173 and provide novel data regarding genetic control of poxvirus immunity.


INTRODUCTION
Until its eradication in 1980, smallpox (caused by the variola virus) was a deadly, debilitating disease estimated to have killed hundreds of millions of individuals over the last two centuries alone (1). Eradication was made possible by using vaccines based on vaccinia virus (1). These live-virus vaccines elicited robust, long-lasting immunity in nearly all vaccine recipients (2,3). Routine smallpox vaccination was halted in the United States before global eradication due to rare but serious adverse events, including death; however, poxviruses remain a public health issue for several reasons, including zoonotic poxvirus outbreaks (4)(5)(6)(7) and concerns regarding the release of variola virus as a biological weapon and novel poxviruses (8). The increasing use of poxviruses as platform vectors for other vaccines and therapeutics has also enhanced our need for a better understanding of poxvirus immunity. While highly effective, the smallpox vaccine has numerous contraindications as well as rare but serious, potentially life-threatening adverse reactions that limit its widespread use, if needed, in the population. Understanding how poxvirus immunity is controlled may assist in the development of safer yet still effective poxvirusbased vaccines and can provide insights into immunity to other DNA viruses/vaccines.
Although seroconversion rates after smallpox vaccination are high (>97%), antibody titers and cellular immune responses vary widely among recipients (1,(9)(10)(11)(12). We have previously reported on a small subset of individuals who develop the classical vaccine take (i.e., response) but fail to mount vigorous adaptive immune responses (13). Previous research by our lab and others demonstrates that genetic polymorphisms are correlated with immune outcomes to multiple viral vaccines, including the smallpox vaccine (11,12,14,15). To move beyond statistical genetic associations, functional studies are also needed to elucidate the biologic mechanisms underlying these associations and link them to gain a better understanding of how genomic factors contribute to inter-individual variation in immune response.
Recognition of foreign nucleic acids by the cGAS/STING pathway leads to type I IFN responses and is an important component of the innate response to viral and bacterial infection (16). Cytosolic DNA is recognized by cGAS, leading to the generation of the cyclic dinucleotide 2'3'cGAMP, which, in turn binds to STING. STING is also able to directly recognize bacterially produced cyclic dinucleotides. Both direct activation of STING and indirect activation through cGAS trigger the phosphorylation of TBK1 and IRF3, resulting in the transcription of type I interferons, TNF, IL-6, with subsequent activation of interferon regulated genes and inflammation. STING has been found to mediate the IFN response to bacteria and DNA viruses, including herpes simplex viruses, CMV, HPV, and poxviruses such as vaccinia and ectromelia (17)(18)(19)(20)(21).
We have previously reported the findings from the first genome-wide association study (GWAS) examining the association of single nucleotide polymorphisms (SNPs) with immune responses in a cohort of primary smallpox vaccines (11,12,15). Here, we report a cluster of SNPs on chromosome 5 (5q31.2) that were significantly associated with IFNa response following in vitro stimulation of PBMCs with vaccinia virus. We report the results from the functional testing of rs1131769, which is a non-synonymous SNP in TMEM173 that introduces an amino acid change from the arginine at position 232 (R232) to histidine (H232) in the STING protein (18)(19)(20)(21). Our results indicate that the H232 variant of STING is associated with a significant reduction in the IFNa response and that this effect is independent of the effect previously described for SNPs in the STING HAQ haplotype (22). We also report the results of molecular modeling and molecular dynamics (MD) simulations investigating differences in how the H232 and R232 variants interact with the signaling ligand. Overall, our study provides novel and important data regarding genetic control of poxvirus immunity in humans by linking specific genetic polymorphisms in TMEM173 to differential STING pathway activation during innate immune responses (IFNa) to vaccinia virus. These results may also explain inter-individual variations in the innate immune response to other DNA viruses (e.g., HPV, VZV, HSV-1), which also stimulate the STING pathway, as well as the large number of bacterial pathogens that also activate STING.

Study Cohorts
Two previously described study cohorts were combined for our analyses (11,12,15). Briefly, the San Diego cohort consists of 1,076 Dryvax ® recipients (primarily U.S. Navy personnel) recruited in 2003-2006. The U.S. cohort consists of 1,058 ACAM2000 ® or Dryvax ® recipients (primarily U.S. Army personnel) recruited in 2010-2013. For both cohorts, subjects had received their first (and only) smallpox vaccination between 1 and 4 years prior to study enrollment. All participants gave written informed consent for this study. Approval for all study procedures was obtained from the Institutional Review Boards of the Mayo Clinic (Rochester, MN) and the Naval Health Research Center (NHRC; San Diego, CA).

Genotyping and Fine Mapping
DNA from all subjects was extracted using Gentra Puregene kits (QIAGEN, #158445) (11). Genome-wide genotyping for the study cohort was performed as previously described (11,12,15,25). For the SD cohort (recruited in [2003][2004][2005][2006], subjects were genotyped with either the Illumina 550 array or the Illumina 650 array. Genotype quality control (QC) prior to imputation was conducted separately for each platform. QC for the Illumina 550 and 650 arrays involved removing monomorphic SNPs and those on the Y chromosome. We also removed all SNPs with a call rate <95%, and all subjects with a call rate <95%. SNPs were also excluded if they failed Hardy-Weinberg Equilibration (HWE) test p-value > 10 −5 . Genetic sex was verified by PLINK. Subjects in the U.S. cohort were recruited in 2010-2013 and genotyped with the Illumina Omni 2.5 array. For the Omni 2.5 array, mitochondrial SNPs, those on the Y chromosome, and monomorphic SNPs were removed. SNPs were required to have a call rate at least 99%, and subjects had a minimum call rate of 95%. No inconsistencies were found between reported sex and genetic-determined sex. Across these cohorts, a total of 2,062 subjects passed QC for genotyping.
The 1000 Genomes cosmopolitan samples (Build 37: African, AFR; American, AMR; Asian, ASN; European, EUR) served as a reference for SNP imputation. Observed SNPs were eliminated prior to imputation if they could not be converted to the forward strand or if more than one SNP mapped to a given position. The reference genome was filtered to exclude SNPs whose minor allele frequency (MAF) was < 0.005. The data were then phased using SHAPEIT (26) and imputed via IMPUTE2 (27). SNPs were included in analyses if their imputation dosage allele R 2 was at least 0.3 and their MAF was at least 0.01. These GWAS QC restrictions resulted in a dataset with 6,210,296 SNPs for the HumHap550 array; 6,244,529 SNPs for the HumHap650 array; and 6,243,494 SNPs for the Omni 2.5 array.
Fine mapping on the chromosome 5 region was performed using a custom Illumina iSelect panel that included known SNPs in the following gene regions: TMEM173, KCNN2, DNAJC18, and TRIM36 (the coding region, the intronic regions, and 10kb upstream and downstream in order to capture regulatory regions). We then identified all SNPs highly correlated (r2 > 0.9) with each of the target SNPs of interest based on the GWAS results. SNPs were excluded from the fine-mapping effort for the following reasons: low rank on the Illumina design score metric (indicating a low likelihood of successful genotyping); any Illumina error codes; previously genotyped SNPs; and monomorphic SNPs (based on HapMap and 1000 Genomes data). The resulting list of 2,406 SNPs were included on the Illumina iSelect panel. The genotyping was performed in Mayo Clinic's Clinical Genome Facility on 2,208 subjects: 2,011 subjects from the SD and U.S. cohorts; and 197 subjects used for quality control (55 negative controls, 48 trios of father/ mother/child). 1,996 of these subjects passed all QC metrics filters (e.g., call rate at least 99%, duplicates removed, etc.). Of the genotyped SNPs, a total of 580 SNPs were used in the analysis (156 SNPs failed genotyping; 10 SNPs had call rates < 95%; 32 SNPs had HWE p-values < 10E −6 , and 1,500 were monomorphic). For the Caucasian subgroup, an additional 126 SNPs were removed because they were monomorphic in that subgroup.

Genetic Ancestry and Population Stratification
Genotypes from the GWAS arrays were used to assign ancestry groups (i.e., Caucasian, African American, or Asian) to participants using the STRUCTURE software (28) and the 1000 Genomes reference data. Genetic ancestry proportions were estimated within cohorts and arrays (San Diego/550, San Diego/650, US/Omni 2.5), as previously described (11,15). A linkage disequilibrium (LD) pruning process (29) was utilized to ensure that the SNPs used for STRUCTURE and for sample eigenvectors were not drawn from small clusters within specific locations (30). Resulting SNPs were entered into to the STRUCTURE program (28), and participant ancestry was classified based on the largest ancestry proportion estimated by STRUCTURE.
Within ancestry groups, eigenvectors were estimated for population-stratification purposes. SNPs with a MAF < 0.01 and those with a HWE p-value < 0.001 were excluded, as were insertion/deletions (INDELS). The remaining SNPs were pruned according to the following variance inflation factors: window size of 50 kilobases; step size of 5; and variance inflation factor threshold of 1.05. SmartPCA was used to create the eigenvectors (31) following the procedures implemented in EIGENSTRAT software. Eigenvectors were included as potential covariates if they had a Tracy-Widom p-value < 0.05.

Selection of Covariates to Adjust for Potential Confounders
For analysis purposes, the immune-response trait of interest (IFNa secretion) was calculated by first computing the difference of the mean stimulated and unstimulated values and then transforming to a normal distribution using normal quantiles. In order to combine data from the two cohorts, potential confounder effects for each ancestry group and cohort were adjusted by linear regression models as described (29). Categorical variables with a very large number of categories were binned using hierarchical clustering. This was achieved by using hierarchical clustering on the estimated regression coefficients for the different categories while binning categories with similar regression coefficients. Categorical variables were included in regression models by using indicator variables for categories, treating the most common category as baseline. Residuals from the linear models were used as the primary adjusted traits for GWAS analyses.

GWAS Analysis Strategy
In order to maximize the power to detect SNPs associated with smallpox vaccine immune response phenotypes, data across genotyping arrays and the two cohorts was pooled after preparing the data as described above. The pooled analyses were then performed using the adjusted traits described above in a regression analysis, along with an indicator of cohort as an additional adjusting covariate. Because the largest ancestry group was Caucasian, we restricted our pooled analysis to the Caucasian subjects (n=1,605). Multiple testing was controlled for by using the standard p-value threshold (p-value< 5x10 −8 ) to determine genome-wide statistical significance (32,33). Statistical analyses were performed with the R statistical software and PLINK (34).

Generation of Stably Transduced BJAB Cell Lines
BJAB cell lines, each expressing one of the rs1131769 variants of interest, were created using custom suCMV promoter-based lentivectors containing a Blasticidin resistance gene (GenTarget Inc.; San Diego, CA). Lentiviral particles were produced in 293T packaging cells using the SureTiter ™ Lentiviral vector system (GenTarget Inc.; San Diego, CA), in DMEM with 10% FBS at approximately 10 7 IFU/per ml. Transduction with lentiviral particles was performed at MOI of 10 in the presence of Polybrene (Millipore Sigma) at 8 µg/mL, and stable cell clones were selected for using Blasticidin (InvivoGen; San Diego, CA) at 10 µg/mL.

Cellular Stimulation With STING Ligands
STING allele-expressing BJAB cells were plated at 300,000 cells/ mL, 0.5 mL/well in a 24-well plate. 50 ug/mL of 2'3' cGAMP or ddH2O were added to stimulated and mock-stimulated wells, respectively. Stimulated and control cells were harvested at the indicated times after 2'3' cGAMP stimulation and centrifuged 5 min at 5,000 rpm in microcentrifuge tubes. Supernatants were collected and frozen for ELISA analysis. Cells were resuspended in 200 uL RNAProtect (Qiagen; Valencia, CA) and frozen at -20°C.

Promoter Reporter Assays
The promoter reporter assays were performed in HEK 293 T cells, stably expressing one of the STING alleles of interest (for rs1131769 -WT/R232 and H232) under blasticidin selection (36). We used pNiFty2-IFNB-SEAP and pNiFty2-56K-SEAP promoter-reporter plasmids (InvivoGen; San Diego, CA), encoding the INFb minimal promoter and the ISG-56K promoter, respectively. Co-expression with constitutively activated IRF3 (or IRF7) leads to promoter induction measured by the inducible expression of the secreted embryonic alkaline phosphatase (SEAP) reporter gene. Promoter assays were performed as previously described (36) but with some modifications. Briefly, 2.5 x 10 4 cells per well (stably expressing STING alleles of interest under blasticidin selection) were cultured overnight in 96-well plates in antibioticfree medium [DMEM (Gibco Invitrogen Corporation; Carlsbad, CA)], containing 10% fetal bovine serum (FBS, HyClone; Logan, UT). On the following day, cells were transfected with Lipofectamine ® LTX (Invitrogen; Carlsbad, CA), according to the manufacturer's protocol, using a constant amount of reporter plasmid (100 ng of either pNiFty2-IFNB-SEAP or pNiFty2-56K-SEAP per well), 0.2 µL PLUS ™ Reagent (Invitrogen; Carlsbad, CA) per well, and 0.25 µL Lipofectamine LTX per well. After overnight incubation, the medium was switched to Opti-MEM (Gibco Invitrogen Corporation; Carlsbad, CA), and cells were stimulated with one of two STING ligands: 2'3' cGAMP (100 µg/ ml), or inactivated vaccinia virus (MOI of 10) at 37°C for different time periods. Promoter induction was measured by the SEAP reporter secretion (quantified at 620 nm following addition of Quanti-Blue ™ media, Invivogen, per the manufacturer's instructions). Experimental conditions were compared using Student's t-test.

ELISA Measurement of Secreted Type I and Type III IFNs
IFNa and IFNl production by 2'3' cGAMP-treated STINGtransduced BJAB cells were measured in triplicate using commercial sandwich ELISA assay sets (IFNa: VeriKine-HS ™ Human Interferon Alpha All Subtype ELISA Kit, PBL Assay Science; Piscataway, NJ, and IFNl: Human IL-29/IL-28B [IFNlambda 1/3] DuoSet ELISA set, R&D Systems; Minneapolis, MN) according to the manufacturer's instructions. Standard protein samples were diluted in cell culture media for accurate standard curve construction and calculations. Recombinant IFNs were used as positive controls while cell culture media served as the negative control. Biological duplicate samples from each timepoint were each assayed in technical duplicate. The level of sensitivity for the IFNa and IFN l assays were 12.5 and 62.5 pg/mL, respectively.

Molecular Modeling
The atomic structure of the cyclic dinucleotide binding domain of STING has been experimentally solved (37). As is common for crystallographic structures, mobile loops were not resolved in these structures. To initially place residues within these mobile loops, we used the SwissModel server (38) and template PDB structures 4QXP (39) (open conformation with inhibitor bound) and 4F5Y (40) (closed conformation with cdGMP bound). Mutations present in each template were reverted to WT amino acids according to the UniProt sequence of the canonical transcript (Q86WV6-1). Simulations were run for the apo (un-liganded), cdGMP, and cGAMP ligand states.
We used NAMD (41) and the CHARMM27 with the CMAP (42) force field for Generalized Born implicit solvent molecular dynamics (isMD) simulations using previously optimized conditions (43) that included the following: 1) an interaction cutoff of 15Å; 2) strength tapering (switching) starting at 12Å; 3) a 1fs simulation time step with conformations recorded every 2ps; 4) an initial conformation that was energy minimized for 20,000 steps; and 5) heating to 300K over 300ps via a Langevin thermostat. From each of the 12 conditions (two initial conformations, two alleles, and three ligand states), 100ns of simulation trajectory was generated and the final 70ns analyzed. Three additional and independent 20ns replicates for each condition were generated using the same procedure. All trajectories were aligned to the initial R232 closed conformation using C a atoms. Trajectories were then evaluated using multiple metrics, including C-alpha Root Mean Squared Deviation (RMSD), Root Mean Squared Fluctuation (RMSF), Principal Component (PC) analysis, alignment-free distance difference matrix (44)(45)(46), and distance monitors across the ligand binding site. We quantified variance of atomic C a −C a ( Figure 7A) distances using median absolute difference (MAD). Analysis was performed using custom scripts on the Bio3D R package (46) and VMD (47).

Overview of Cohorts and IFNa Response
We conducted a GWAS meta-analysis on two cohorts (n=1,076 and 1,058) of smallpox vaccine recipients (11,12,15) with immune outcome data and genome-wide SNP-typing. We imputed additional SNPs, as described in Materials and Methods section. As the full dataset was available for all subjects, the cohorts were combined to increase statistical power, and a final study sample of 1,653 Caucasian subjects was available for analysis (see Supplementary Table 1 for demographic information). The actual cohort used in the analysis for each immune outcome varied depending on how many of those 1,653 individuals had data for that specific outcome. Our original intent was to determine whether or not there were genetic polymorphisms associated with markers of vaccine-induced cellular immunity, as had been noted in preliminary reports on the San Diego Cohort (24,25,48). VACV was inactivated in order to minimize the immunomodulatory effect of poxvirus-encoded proteins and to allow full development of the cytokine response. Our outcomes of interest included both innate anti-viral outcomes (secretion of IL-1b, IL-6, IFNa) and markers associated with adaptive immune responses (neutralizing antibody titer, IFNg ELISPOT response in PBMCs, IFNg ELISPOT in CD8+ T cells, as well as secretion of IL-2, IFNg, IL-12p40, and TNFa).
Surprisingly, our GWAS analysis found only a strong signal on chromosome 5 associated with IFNa secretion that exceeded the genome-wide significance level ( Figures 1A, B). There were two suggestive signals: one on chromosome 11 associated with the CD8+ T cell IFNg ELISPOT response and another on chromosome 9 associated with IFNg secretion. The remaining outcomes were not associated with any genetic variants. Due to the strength of the IFNa signal and the fact that the other two potential signals did not reach a genome-wide threshold for significance, we focused our efforts on exploring the region on chromosome 5 associated with IFNa secretion.
The locus-zoom plot in Figure 1C depicts the SNPs with the strongest statistical association with IFNa response. We used a genome-wide threshold of p-value <5 x 10 −8 to establish statistical significance. Further details on the SNPs meeting this threshold are provided in Table 1. Of the 1,653 individuals with genotyping available, 1,605 also had IFNa secretion data. This cohort had a median IFNa secretion level of 126 pg/mL (IQR: 48.6-229.6) in PBMC cultures after vaccinia virus stimulation. As illustrated in Figure 2, TT homozygotes (H232 STING) had a median IFNa response of 17.7 pg/mL, while individuals homozygous for the R232 STING allele (CC genotype) had an 8-fold higher response (143.6 pg/mL). Heterozygotes had an intermediate phenotype. We did not identify any SNPs in other genes (including those associated with the STING pathway, such as cGAS, TBK1, or IRF3) associated with variations in IFNa secretion.
Several of the SNPs significantly associated with IFNa secretion were located in TMEM173, which encodes for STING -an adaptor molecule mediating type I IFN responses to cyclic dinucleotides and double-stranded DNA. STING has previously been shown to play an important role in the innate immune response to poxviruses (20,49).
A number of additional non-synonymous polymorphisms potentially affecting STING function have previously been identified, including R71H (rs11554776), G230A (rs78233829), R293Q (rs7380824), and R232H (rs1131769) (22,50). In order to determine which SNPs in the haplotype contributed to the response, while accounting for correlation among SNPs, we used haplo.stats software in R to compute the haplotype frequencies for these four SNPs: rs11554776; rs78233829; rs1131769; and rs7380824 (51).  The haplotype frequencies were very close between the U.S. and San Diego cohorts; therefore, we proceeded to use the combined sample to evaluate the association of haplotypes with IFNa response. The results presented in Table 2 illustrate the effects of the haplotypes on IFNa response and the frequencies of the haplotypes. We compare the haplotype CCTC (H232) with the most frequent haplotype CCCC (R232: treated as the baseline in the linear regression model) in order to focus on the effect of the T allele for rs1131769 while controlling for the effects of the other three SNPs on the haplotype, and identified a statistically significant (p <2E -16 ) decrease in IFNa response. The contrast of the haplotype CCCT with the baseline was not statistically significant, but we had limited power for this comparison because of the low frequency of the haplotype CCCT. We also used likelihood ratio statistics to contrast the effects of the haplotypes CGCT and TGCT, which differed only at the first SNP position, but the effects of these haplotypes were not significantly different (p-value = 0.46), but once again power was limited because of the rarity of haplotype CGCT. These analyses suggest that the SNP rs1131769 is likely the main variant in the haplotype impacting the association with IFNa response. However, the strong association of the haplotype TGCT with IFNa response suggests that there might be additional SNPs in the region that are in linkage disequilibrium with our measured SNPs that are also associated with IFNa response. To further explore this, we computed the dose of the minor allele for each of the four SNPs and performed backward regression, thereby ignoring haplotypes. The two SNPs rs7380824 and rs1131769 remained in the model (each with p-value < 2E −16 ), illustrating that each SNP is strongly associated with IFNa response after adjusting for the other SNPs. The other two SNPs, rs78233829

Promoter Activity of rs1131769 Variants
Two plasmids expressing the secreted embryonic alkaline phosphatase (SEAP) reporter gene, under control of either the INFb promoter or the interferon stimulated gene (ISG)-56K promoter, were used to measure ligand-stimulated promoter activity in HEK293T cells expressing either R232 or H232 STING ( Figure 3). Both 293T variant cell lines expressed high levels of STING mRNA ( Figure 3D) with no significant difference between alleles. Upon 2'3' cGAMP stimulation, IFNb promoter induction was significantly higher at 10 h poststimulation in R232 cells compared to cells expressing H232 ( Figure 3A, p=0.02). Similarly, we observed statistically significant higher induction of the ISG-56K promoter activity in R232 upon stimulation with either 2'3' cGAMP at 4 h and 8 h post-stimulation ( Figure 3B, p=0.006 and p=0.004, respectively) or inactivated vaccinia virus at 8 h post-stimulation ( Figure  3C, p=0.002).   Gene Expression of rs1131769 Variants mRNA was extracted from PBMCs of individuals homozygous for the CC genotype (R232) and the TT genotype (H232), and the two TMEM173 variants were PCR amplified and cloned into lentivirus expression vectors. BJAB cells were transduced with these vectors, creating stable cell lines that constitutively overexpress each STING variant. The PCR products and completed expression vectors were both sequenced to verify the insertion of the correct genetic variants. As illustrated in Figure 4A, stable transfectants express >1,000-fold higher (and comparable between the two variants) STING mRNA than normal BJAB cells. Expression of both TMEM173 variants transiently and minimally (less than 2-fold) increased after 2'3' cGAMP stimulation, indicating the STING protein levels of either variant are unlikely to be significantly affected by cGAMP treatment ( Figure 4B). Finally, we found that MB21D1 (encoding cGAS, an essential upstream nucleotidyltransferase in the STING pathway that generates cyclic cGAMP) gene expression was not significantly different between the two variants ( Figure 4C), indicating that signaling pathway function upstream of STING was not affected by the STING gene variants. Note that the HEK293T lines used in this report were also transfected with MB21D1 as this cell line is known to be deficient in cGAS expression (52).

Effect of rs1131769 Variants on Downstream Phosphorylation
Transiently transfected HEK293T and stably transfected BJAB cells were stimulated with 2'3' cGAMP for the indicated time periods, and phosphorylation of downstream intermediates TBK1 and IRF3 was evaluated ( Figures 5A, B). In transiently transfected HEK293T cells (also expressing cGAS), H232 STING expression was accompanied by a delay in phosphorylation of both TBK1 and IRF3 until 1 hour after stimulation. Interestingly, we observed substantial pTBK1 and pIRF3 baseline phosphorylation (at timepoint 0) for the R232 STING variant (but not for the H232 variant associated with diminished phosphorylation/activation), which is likely due to STING overexpression. The H232 STING expression at timepoints 0 and 30 min was slightly reduced, but TBK1 and IRF3 protein Frontiers in Immunology | www.frontiersin.org October 2020 | Volume 11 | Article 567348 expression were similar between the two variants, which demonstrates that the observed differences in phosphorylation/ activation are valid. In the stably transduced BJAB line, the delayed phosphorylation was observed with both STING variants and the pattern/kinetics of phosphorylation was similar; however, the magnitude of TBK1 and IRF3 phosphorylation was significantly reduced (in particular at 2h and 4h poststimulation) in cells expressing H232 compared to cells expressing R232 STING. No major differences in STING protein levels, or in the unphosphorylated forms of either TBK or IRF3, were observed between the cells expressing the two STING variants at the observed timepoints.

Effect of rs1131769 Variants on IFN Response
Reasoning that differences in TBK1 and IRF3 phosphorylation between these variants should have downstream consequences, we decided to examine differential pathway activity mediated by the two STING variants. We stimulated R232 and H232 STING variant-expressing stable BJAB cell lines with 2'3' cGAMP and measured gene expression (qPCR, Figure 6A) and protein secretion (ELISA, Figure 6B) over time. cGAMP stimulation induced both type I (IFNa, IFNb) and type III (IFNl1) interferons, with significantly higher levels of IFNs (mRNA and protein) observed in R232 cells. This effect was consistent regardless of the cGAMP isomer used for stimulation (Supplementary Figure S1). Furthermore, the expression of the classical antiviral ISGs, MX1 and OAS1, after 2'3'cGAMP stimulation confirmed the greater STING pathway activation in R232 STING-expressing cells over the H232 STING-expressing cells ( Figure 6C).

Molecular Modeling of STING Variants
In order to begin elucidating the mechanism underlying the greater STING activity in R232-expressing cells, we used molecular modeling and molecular dynamics simulations to examine structural and functional differences between the R232 and H232 variants. H232 exhibited larger overall deviation from the initial experimental structure, as quantified by RMSD ( Figure 7A). This greater deviation (a reflection of mobility) occurred both in the presence or absence of ligand. As RMSD is a global measure, we also quantified per-residue mobility using RMSF ( Figure 7B), which indicated that the ligand-binding loops were more mobile for H232 compared to R232, particularly when the 2'3' cGAMP ligand was bound. We further quantified the displacement of the ligand-binding loops using simple distance measures between residue 232 in each monomer. Regardless of the presence or absence of cGAMP, the ligand-binding loops of H232 were further separated from each other ( Figure 7C) and from the base of the ligand-binding site ( Figure 8) compared to R232. Our initial simulations assumed that the ligand-binding loops were closed over the base of the ligand-binding site; structures displayed in Figures 7D, E highlight the difference in STING conformation between H232 and R232.
Simulations assuming an open ligand-binding loop conformation observed the same effect of H232 compared to R232 (Supplementary Figure S2). In summary, H232 exhibited greater structural flexibility and mobility of the ligand-binding loops in both the open or closed conformations and in the presence or absence of cGAMP.

DISCUSSION
Our GWAS across two cohorts of smallpox vaccine recipients, totaling just over 1,600 individuals, identified a highly significant (p < 1 x 10 −30 ) association signal from a region on chromosome 5 that was linked to significant inter-individual variations in IFNa response to in vitro stimulation with vaccinia virus. The lack of genetic association with vaccine response markers (e.g., neutralizing antibody titer and IFNg ELISPOT) indicates that the signal observed is likely reflective of an innate response to poxvirus infection rather than a T cell response. This suggests that our findings are broadly applicable to the response to poxvirus infection rather than smallpox vaccination. Fine-mapping analysis identified a number of putative causal SNPs, including several in TMEM173, which encodes for the signaling adaptor protein STING. STING mediates IFN responses to dsDNA and cyclic dinucleotides through a pathway involving cGAS and the phosphorylation of TBK1 and IRF3. Our regression modeling also indicated that multiple SNPs within TMEM173 have independent effects on the phenotypic outcome. Homozygotes for the H232 allele of rs1131769 in TMEM173 exhibit a 90% reduction in IFNa secretion compared to R232 homozygotes. This is a highly significant effect that may have significant downstream consequences for poxvirus immunity. We have previously reported that a small percentage of smallpox vaccine recipients have impaired innate immune responses to vaccinia virus and that these same individuals also have suboptimal cellular immunity (13). Our current results provide additional support to existing data (53,54) suggesting that appropriate innate immune responses are necessary for robust adaptive immunity to vaccinia virus. Further investigation of this effect on smallpox immunity is warranted.
We conducted a series of experiments with the intention of elucidating functional effects of this SNP that might be underlying the identified genotype-phenotype association. We assessed gene expression of both variants of TMEM137 by PCR and STING expression by western blot and did not detect significant differences between variants in either gene or protein expression. We observed higher promoter activity of downstream IFN-inducible genes in the R232 variant compared to the H232 variant, which suggests that there are differences in activation of the STING pathway. Upon stimulation with 2'3' cGAMP, the R232 STING variant elicited faster phosphorylation of both TBK1 and IRF3 as well as resulted in a greater quantity of phosphorylated TBK1 and IRF3 in the cells. These changes led to a significant increase in IFN and IFN-stimulated gene expression in R232-expressing cells, confirming that the statistical association was rooted in differential biological activity. As is true of most transfection systems, the TMEM173 gene was overexpressed in our cell lines, with the HEK293T cells expressing~10,000 times as much TMEM173 as untransfected cells. The BJAB transfectants also expressed high levels of TMEM173, but the overexpression was an order of magnitude lower. More relevant to our results, protein expression was similar to the expression levels of endogenous IRF3, TBK1, and tubulin, suggesting that protein expression was within normal limits despite the upstream overexpression of STING observed at the gene level. We note that the expression (at the gene and protein level) of both variants was consistent; therefore, differences in activity are not a result of differential gene or protein expression between the variants. Multiple genetic variants of TMEM173 have been described, including three non-synonymous SNPs: rs1131769 (H232R), which was the focus of our study; rs11554776 (R71H); and rs7380824 (R293Q). R71H and R293Q, together with a fourth SNP, rs78233829 (G230A), form the HAQ haplotype (22). Zhang et al. have previously reported that expression of the H232 variant results in reduced IFNb transcription (55). Our results confirm and extend these initial findings, demonstrating that IFNa secretion is also affected, as is the expression of multiple interferon-stimulated genes. Regarding the 3 alleles in the HAQ haplotype, there has been some controversy over the biological effects of these alleles (56,57). In one study, cells carrying the G230 variant had fully functional STING activity and the HAQ haplotype effect was attributed to the R71H and R293Q SNPs (50). A similar study evaluating TMEM173 variants found that the R232H, R293Q, and AQ (G230A, R293Q) variants had minimal effects on endogenous STING activity while the reduced STING function of the HAQ (R71H, G230A, R293Q) haplotype was attributed to the R71H variant (22). Our analysis supported previous findings that possession of the HAQ haplotype leads to reduced STING activity, demonstrated that the H232 variant also leads to reduced STING activity, and verified that this functional effect is independent of the HAQ haplotype. Thus, our haplotype and regression model results indicate that multiple SNPs/haplotypes are independently associated with variations in IFNa secretion. Our study also provides a potential biochemical mechanism for the reduced IFNa activity mediated by the R232H variant; however, further work will be required to tease apart the contributions of each individual SNP to the resulting immune response phenotype.
The crystal structure of STING has been resolved, as has the structure of the H232 variant bound to cGAMP (58). Our molecular modeling simulations, using these structure data, revealed that the ligand-binding loops of STING were more mobile for H232 compared to WT; that is, H232 loop conformations were more open and flexible than R232, even in the presence of bound cGAMP. We speculate that this may reflect a failure of the ligand-binding loops to either stay closed when beginning from a closed conformation, or to close when beginning from an open conformation. This may be indicative of weaker binding (and/or faster disassociation) between cGAMP and the H232 variant of STING. Our data suggest that the alterations in loop dynamics and weaker affinity of H232 STING for its ligand are the underlying molecular mechanism for the reduced STING activity that we observed for this variant. This hypothesis will require additional experimental data to confirm or refute.
With regard to poxviruses, it has been demonstrated that Modified Vaccinia Ankara (MVA) infection of dendritic cells (DCs) triggers type I IFN production through a STING and cGAS-dependent pathway involving cyclic dinucleotides (49). This type I IFN response is not seen during wild type vaccinia virus infection, likely due to the presence of viral immunomodulatory proteins such as C6L, E3L, and N1L in the wild type virus but not in MVA (20,21). A recent report confirmed that MVA activated IRF3 in a cGAS-and STINGdependent manner, whereas wild type vaccinia strains failed to do so (19). Our experiments found clear differences in R232 and H232 STING activity in the presence of cyclic dinucleotides and inactivated vaccinia virus. It is possible that possession of the H232 STING variant may alter the effects of viral immunomodulation of this innate immune pathway during infection. This may happen through differential interactions with viral proteins, or indirectly as reduced secretion of type I IFNs may render viral immunomodulation more effective.
We have demonstrated that carriage of the H variant of rs1131769 results in a 90% decrease in innate immune response (secreted IFNa) to vaccinia virus. This data helps resolve prior conflicting reports regarding functional effects of STING polymorphisms. We hypothesize that the effect of this polymorphism is due to different flexibility/mobility in STING H232 loop conformations, which results in reduced ability of H232 STING to phosphorylate downstream signaling intermediates and mediate effective STING pathway activation.
Poxviruses represent a continuing public health concern due to the risk of bioterrorism use, zoonotic outbreaks (e.g., monkeypox, buffalopox, vaccinia-like viruses, and novel poxviruses), the increasing use of poxviruses for oncolytic viral therapy, and their use as vectors for vaccine antigens against HIV, rabies, Ebola, Zika, and other pathogens. STING also plays an essential role in triggering protective innate responses to DNA viruses (e.g., poxviruses, herpes simplex viruses, varicella zoster, EBV, HPV, and others) and multiple bacterial pathogens. Polymorphisms that reduce the effectiveness of the innate response to these threats are likely to enhance disease susceptibility and may have a deleterious effect on vaccine immunogenicity in the~15% of the population with this genotype. Given the broad potential impact of this pathway, this is an area that merits additional investigation.
Understanding how genetic factors control the immune response to poxviruses will have important clinical implications in how, when, and in whom these vectors can be safely and effectively used. Furthermore, this information may inform the use of adjuvants to overcome this defect and enhance vaccine responses or the development of therapeutic drugs that can be used to enhance the innate antiviral response during an infection.

DATA AVAILABILITY STATEMENT
The datasets presented in this study can be found in online repositories. The names of the repository/repositories and A B FIGURE 8 | Changes in the ligand binding loop between R232 and H232 are consistent across conditions. (A) Diagram of the STING structure showing one monomer of the closed conformation in white and the second monomer in gray, with the three C a atoms used to compute distances marked by black spheres. The bound cGAMP is shown as a ball-and-stick representation. A rotated view, with the second monomer omitted for clarity, more clearly shows the relationship between these C a positions and the ligand binding site. (B) H232 lead to a consistent expansion of the distances between the ligand binding loops and from the ligand binding loops to the base of the ligand binding site. Probability density plots are shown with the most frequently sampled regions colored black and scaling through red and yellow as data become more sparse. The medians for additional independent replicates are indicated by Xs and a circle indicating the level of variability. accession number(s) can be found below: https://immport.niaid. nih.gov, SDY28.

ETHICS STATEMENT
The studies involving human participants were reviewed and approved by Institutional Review Boards of the Mayo Clinic (Rochester, MN) and the Naval Health Research Center (NHRC; San Diego, CA). The patients/participants provided their written informed consent to participate in this study.

AUTHOR CONTRIBUTIONS
RK contributed to the conception and design of the study; participated in the acquisition of data and the analysis and interpretation of the results; prepared the initial draft; revised draft for intellectual content. IH contributed to the study design; participated in data acquisition and interpretation of study results; assisted in drafting the manuscript; revised draft for intellectual content; approved final version. IO contributed to the study design; participated in data acquisition and interpretation of study results; assisted in drafting the manuscript; revised draft for intellectual content; approved final version. EV contributed to the study design; participated in data acquisition and interpretation of study results; assisted in drafting the manuscript; revised draft for intellectual content; approved final version. BL contributed data analysis and interpretation; assisted in drafting the manuscript; revised draft for intellectual content; approved final version. DS supervised and contributed to the data analysis and interpretation; participated in drafting and revising the manuscript; approved final version. MZ contributed to the interpretation of study results, molecular modeling, data analysis, drafting the manuscript; provided critical review; approved final version. AO contributed to the design of the study; participated in data analysis and interpretation; assisted in drafting the manuscript; revised the manuscript for intellectual content; approved the final version. GP contributed to the conception and design of the study; participated in data analysis and interpretation; secured funding for the project; participated in drafting and revising the manuscript; approved final version. All authors contributed to the article and approved the submitted version.