Figure 1.
Our study design had four stages (sample sizes for cases/controls are in parentheses). In Stage 1, we performed admixture mapping of African Americans (AA) in a case-only analysis with 1032 cases and in a case-control analysis with 1726 controls. In Stage 2, the major admixture mapping signal at 2q22–24 was followed by a candidate gene analysis using case-control association in CCAA (1525/1810) and European Americans (CCEA) (3968/3542), with 737 cases used for stages 1 and 2. In order to focus on our best candidate locus (IFIH1), we used out-of-study controls to increase control sample sizes to 4485 for AA and 9750 for EA. In Stage 3, we performed imputation based analysis on AA (1525/4485) and EA (3968/9750) to confirm our candidate gene analysis. In Stage 4, we performed functional analyses for the three confirmed SNPs. For the coding SNPs rs10930046 and rs1990760, we used an apoptosis assay to assess possible changes in protein function, and a gene expression assay to evaluate the effects of these SNPs on expression of genes related to apoptosis, inflammation and viral response. For the intronic variant rs13023380, we used EMSA to investigate whether the variant affected binding of the local DNA sequence to nuclear proteins.
Figure 2.
Admixture mapping and conditional analysis.
(A) A whole-genome admixture scan on AA SLE cases identified 7 admixture signals that achieved the predefined LOD score >2 (red dashed line). (B) We performed an imputation-based association analysis of IFIH1, which was identified as the most promising candidate gene by a case-control study on 20 candidate genes in the largest peak (2q22–24), followed by 4-SNP haplotype conditional analysis (C). Filled dots indicate the −log10 P values for the association to SLE, and color coding represents inter-marker correlation (r2) between the strongest associated SNP, rs10930046 (“purple diamond”), and the individual SNPs, as shown in the color bar. (C) After conditioning the 4-marker haplotypes for the three markers rs1990760–rs10930046–rs13023380, all individual SNP associations are explained as shown. (D) We analyzed the LD between SNPs on the ImmunoChip, and these LD values were used as a reference panel for imputation in AA. Darker color denotes higher correlation between markers (r2). The LD pattern showed high correlation between markers, making it possible to increase SNP density by imputation. The three independently associated SNPs identified in (B) are denoted by arrows. (E) We performed an imputation based case-control association analysis in EA. Filled dots indicate the −log10 P values for each SNP, and color coding represents the inter-marker correlation (r2) between each individual SNP and the strongest associated SNP, rs13023380 (“purple diamond”), as shown in the color bar. (F) We then performed a two SNP haplotype analysis followed by a three marker haplotype analysis conditioned on the two independent variants rs10930046 and rs13023380. (G) LD analysis of SNPs on the ImmunoChip reference panel showed low inter-marker correlation, which largely precluded imputation based association. Darker color indicates greater r2. Arrows indicate the position of the independent SNPs.
Table 1.
Case-control association for genotyped variants in AA and EA.
Figure 5.
Three risk variants of IFIH1 and the proposed functional model.
(A) IFIH1 protein (1025 amino acids (aa)) structure shows four conserved domains (start-stop aa) in boxes: a caspase recruitment (CARD) domain 115–200 aa); a helicase ATP-binding domain (305–493 aa), a helicase C-terminal domain (743–826 aa), and a RIG-I regulatory domain (901–1022 aa). (B) Domain prediction by Pfam: Gene intron/exon structure is presented below, with exons represented by thick vertical lines. (C) The three associated SNPs are represented by arrowed boxes. SNP rs10930046 is located in the helicase ATP-binding domain encoded by exons 4–7; rs1990760 is located in the RIG-1 regulatory domain encoded by exons 14–16; rs13023380 is on intron 3. We present a model of the functions implicated by the risk alleles in each variant, where each identified variant has an effect on the expression of NFκ-B1, CASP9, MAVS, MX1, IFIT1, NFκ-B2, TNFA and MAPK8, and their impact on inflammation, viral response and transcription.
Figure 3.
Apoptosis and expression assays for exonic SNPs.
(A) K652 cells were transfected with IFIH1 full length cDNA containing the protective ‘G’ or risk ‘A’ allele for rs10930046 and rs1990760 in the following combinations: ‘A-G’, ‘G-A’ and ‘G-G’. After transfection, the percentage of apoptotic cells was quantified by FACS for annexin V and AAD positivity among GFP+ cells (transfection positive) at seven different time points (in hours). At all time points, the risk allele ‘A’ for rs10930046 produced a significant increase in the proportion of apoptotic cells compared to the ‘G’ allele (mean increase 14.6%, P≤1.4×10−3, ‘A-G’ vs ‘G-G’). In contrast, the ‘A’ allele for rs1990760 had no significant effect on apoptosis compared to the ‘G’ allele at any time point (P = 1, ‘G-A’ vs ‘G-G’). IFIH1 cDNA encoding the risk or protective allele of rs10930046 or rs1990760 was transiently transfected, GFP+ cells were sorted by FACS and total RNA was isolated. cDNAs were subjected to RT-qPCR for quantification of expression of NFκ-B1, NFκ-B2, CASP8, CASP9, TNFA, MAPK8, MAVS, IFNA, IFIT1 and MXI (B–K). All of these genes showed significantly altered expression with at least one of the risk alleles. Further, we observed that IFIT1 and MX1 (L, M) expression both increased with risk alleles when transfected cells were treated with IFN beta. Also, IFN beta stimulation increased IFIH1 expression (N), and IFIH1 over-expression induced IFNA expression (O), suggesting a positive feedback loop between IFNA and IFIH1 (P). Although it was known that IFNA induces IFIH1 expression, it was not previously shown that IFIH1 induces IFNA expression, which here is designated by “X”. Colors used in the expression figures are blue for ancestral ‘G-G’, red for rs10930046-‘A’+rs1990760-‘G’; and green shows rs10930046-‘G’+rs1990760-‘A’.
Figure 4.
Binding assay for rs13023380 and molecular model of IFIH1.
(A) EMSA was performed using nuclear protein extracts from K562 cells (A) with 141-bp PCR products including either the protective (‘G’) or risk (‘A’) sequence at rs13023380. Both ‘G’ and ‘A’ allele-containing PCR products bound to a protein complex in the nuclear extracts. However, the ‘A’ allele bound with at least 2-fold reduced efficiency compared to the ‘G’ allele-carrying PCR product, as measured by the intensity of the shifted band relative to the free DNA band in the same lane. As a nonspecific (NS) DNA control, a 140-bp DNA sequence not present in the genome was created by PCR amplification of bisulfite-modified genomic DNA. (B, C) EMSA for purified recombinant Nucleolin and Ku70/80 protein with PCR products carrying the ‘G’ or ‘A’ allele of rs13023380. In both the cases, the ‘G’ allele binds both of these proteins with increased efficiency. +signs are used to denote the increasing amount of protein added in the reaction. Numbers below EMSA pictures denote the ratio between the intensities of protein bound DNA to the free DNA. (D) Luciferase activities of intronic DNA sequences carrying ancestral ‘G’ or risk allele ‘A’. The protective allele has approximately 2-fold higher promoter activity (luciferase units) than risk allele ‘A’ carrying sequences. Tkmin-only vector, MCS-vector with multiple cloning sites, 380G-protective allele, 380A-risk allele. (E) Crystal structure of RIG-I in complex with dsRNA (from PDB 3TMI) [27]. Side-chains are shown in red for the positions corresponding to the two coding SNPs in IFIH1. Both mutations are in close proximity to the dsRNA-binding pocket. (F) Close-up of the side-chain of Ala946, modeled from 3TMI. The side-chain makes close contact with the opposing helicase “cap” domain; together these two domains regulate dsRNA entry and processing. Threonine is shown in transparent colors. (G) Superimposition of the RIG-I ATP-binding domain (PDB 4A2W) in blue, and the human IFIH1 ATP-binding domain (PDB 3B6E) in green. The IFIH1 structure contains the histidine side-chain resulting from the rs10930046 risk allele. Large portions of the IFIH1 structure are absent in the 3B6E model, and the two helices are shifted by 1.5 Å. In the ancestral protein, Arg460 likely interacts with the Leu421 main-chain oxygen, as well as the negative helix dipole and the side-chains of Gln433, and Glu425 and 428 (not present in 3B6E).