Genetic Control of Splicing at SIRPG Modulates Risk of Type 1 Diabetes

The signal regulatory protein (SIRP) family of receptor proteins is encoded by a cluster of five genes at chromosome 20p13. The archetypal member of this gene family, SIRPA, encodes SIRPα (CD172A), a receptor expressed on myeloid and neuronal cells that transduces inhibitory signals upon binding CD47 or surfactant proteins (1). SIRPγ, encoded by SIRPG, also binds CD47 but with 10-fold lower affinity than SIRPα and has only a four–amino acid cytoplasmic tail lacking any residues or motifs that might facilitate intracellular signaling (2). Uniquely among SIRP family members, SIRPγ is expressed on lymphocytes, specifically T cells, activated natural killer (NK) cells, and a small subset of CD19⁺ B cells. While SIRPγ is lacking in any apparent capability of signaling intracellularly, SIRPγ engagement by CD47 has been implicated in cell-cell adhesion, proliferation in response to superantigen stimulation, apoptosis, and trans-endothelial migration of T cells (2–4).

Genome-wide association studies originally identified the chromosome 20p13 region in humans as harboring a risk locus for type 1 diabetes (5) (Supplementary Fig. 1). Fine mapping further refined this association, identifying a credible causative variant, rs6043409, a G-to-A substitution that results in the replacement of alanine with valine at position 263 in exon 4 of SIRPγ (6). The minor allele of rs6043409 (A) is associated with reduced risk for type 1 diabetes (P = 3.9 × 10⁻¹⁰; odds ratio 0.87) (7). However, the amino acid substitution encoded by rs6043409 is conservative, the mechanism whereby this variant might affect risk for type 1 diabetes is unknown, and the function of SIRPγ is poorly understood. It has been reported that the level of cell-surface expression of SIRPγ on CD8⁺ T cells varies with genotype at another, noncoding variant in the region, rs2281808 (8), whose alleles are correlated with those at rs6043409 (r² = 0.89).

In a prior study of lymphocyte transcriptomes, we observed that genes associated with risk for autoimmunity in general, and type 1 diabetes, specifically, were significantly enriched for alternative splicing events (9). Multiple alternatively spliced SIRPG transcripts have been identified: the canonical transcript isoform 1, which uses all six exons; isoform 2, which uses alternative transcriptional and translational start sites; and isoforms 3 and 4, which differ in their inclusion or exclusion of exons 3 and 4 (Supplementary Fig. 2). Here, we demonstrate that risk for type 1 diabetes encoded in the 20p13 chromosomal region is modulated through shifts in the distribution of SIRPG transcript isoforms as a consequence of disruption of a predicted exonic splicing enhancer by rs6043409. This results in differential cell-surface expression of SIRPγ on both CD4⁺ and CD8⁺ T cells. We further show that variation in SIRPγ expression affects cell-cell adhesion, gene expression, and cell-surface CD47 levels.

Genomic DNA and viably frozen peripheral blood mononuclear cells (PBMCs) were provided by the Type 1 Diabetes Genetics Consortium (T1DGC). PBMC samples were obtained from individuals of European ancestry ages 35–50 years and collected in North America. All samples had been genotyped previously with the Immunochip, a custom Illumina genotyping array (6). Use of these samples and data was approved by the University of Florida Institutional Review Board.

CD4⁺ T cells were isolated with the CD4⁺ T Cell Isolation Kit, human (Miltenyi Biotec), following the protocol provided by the manufacturer. CD4⁺ cells were activated and expanded with Dynabeads (Thermo Fisher Scientific) according to instructions provided by the manufacturer.

RNA was isolated from CD4⁺ T cells with the RNeasy Plus Mini Kit (QIAGEN). RNA quantity and purity were assessed by spectrophotometry. RNA was reverse transcribed with iScript reverse transcriptase and oligo(dT) primers (Bio-Rad Laboratories).

CD4⁺ T cell cDNA for SIRPG transcript isoforms was tested in quadruplicate by real-time quantitative PCR with TaqMan Gene Expression Probes. Samples were normalized to two reference probes, RPL13A and IPO8, and relative expression was analyzed with use of the efficiency correction method (10).

Vectors for the expression of SIRPG isoforms were prepared in pEF-DEST51 (12285011; Invitrogen) and confirmed by sequencing. SIRPG expression vectors were transfected into HEK293T cells with the X-tremeGENE HP DNA Transfection Reagent (Roche). Cells were harvested 48 h posttransfection for protein analyses.

Methods for quality control, alignment, and normalization of RNA-sequencing (RNA-seq) data have previously been previously (9). The data sets analyzed include existing RNA-seq data from CD4⁺ T cells, CD8⁺ T cells, and CD19⁺ B cells from 82 individuals established as having type 1 diabetes as well as newly generated RNA-seq data from Jurkat cell clones derived from CRISPR/Cas9 targeting experiments (five SIRPG exon 4–targeted clones, four wild-type clones). For samples passing quality control, paired-end sequencing reads were aligned to human Ensembl (11) transcript sequences with RSEM (12) (version 1.2.28) with default settings. References for RSEM were prepared with “rsem-prepare-reference” and transcript sequences for the complete Ensembl transcriptome (release 99) derived from the GRCh38/hg38 reference genome and a tab delimited gene-to-transcript index file, as previously described (9). Estimates of transcript expression were expressed as transcripts per million and log transformed.

Protein from whole cell lysates was separated on Bolt 4–12% Bis-Tris Plus gels (Invitrogen). Protein was transferred to a polyvinyladene difluoride membrane (IPVH20200; Immobilon) with use of the Novex XCell II Blot Module (Life Technologies). Antibodies to SIRPγ (1:2,000; Sigma-Alrich), V5 (1:5,000; Invitrogen), γ-tubulin (1:10,000; Sigma-Aldrich), and GAPDH (1:1,000; Santa Cruz Biotechnology) were used in immunoblotting. CD4⁺ T cell SIRPγ signal was normalized to the GAPDH loading control following densitometric analyses with ImageQuant TL v8.1.0.0 software.

With use of transcript ENST00000303415.7 as a reference, we targeted exon 4 in Jurkat-p2 cells using a single-guide RNA with the sequence GGUGAACAUAUCUGACCAAA. Cloned Jurkat lines were isolated through serial dilution in U-bottom 96-well plates. Clones containing truncating mutations were identified by genomic DNA extraction, PCR amplification, and Sanger DNA sequencing.

SIRPG-targeted clones were expanded in supplemented RPMI medium, and SIRPγ expression was analyzed by flow cytometry. To avoid antibody competition, SIRPγ surface and intracellular staining was performed in separate test wells with the same antibody. After washing twice in PBS, cells were resuspended in PBS–1% FBS at a density of 1 million/tube and stained for viability with the LIVE/DEAD Fixable Near-IR Dead Cell Stain Kit (Thermo Fisher Scientific) for 30 min according to the manufacturer’s protocol. After washing twice in PBS–1% FBS, cells were incubated in a U-bottom 96-well plate with anti-SIRPγ antibody (BV421 Mouse Anti-Human SIRPγ Clone OX-119, BD Biosciences, or PE anti-human CD172g [SIRPγ] Clone LSB2.20, BioLegend) or corresponding isotype control for either 20 min for surface staining or 30 min for intracellular staining with use of Brilliant Stain Buffer (BD Biosciences) and eBioscience Intracellular Fixation & Permeabilization Buffer Set (Thermo Fisher Scientific). Cells were acquired in a BD LSRFortessa cytometer (BD Biosciences) and analyzed with FlowJo software (FlowJo 10.7.1) according to the gating strategy described in Supplementary Fig. 6.

For the analysis of CD47 surface expression, SIRPG-targeted clones were labeled with anti-CD47 antibody (PE Mouse Anti-Human CD47 Clone B6H1 [RUO]; BD Biosciences) or a corresponding isotype control following the manufacturer’s instructions as described above. Cells were acquired in a BD LSRFortessa cytometer and analyzed for CD47 bright cell expression (expression >10⁴) after selection for no debris, morphology, and singlets.

Doublet formation was assayed as described by Piccio et al. (3). Briefly, Jurkat cell clones were labeled with distinguishable cell-permeable dyes, Vybrant DiO and DiD (V-22886 and V-22887, respectively; Molecular Probes). Various combinations of cells (2 × 10⁵ of each) were mixed, centrifuged, and incubated at 37°C for 30 min. Conjugates were gently resuspended in a small volume of medium for flow cytometric analysis on BD LSRFortessa cytometer. Cells that had fused or formed stable clusters were identified through double labeling.

PBMCs were thawed and resuspended in FACS buffer (PBS without calcium and magnesium, 5 mg/mL BSA, 0.1% NaN₃), blocked with TruStain FcX (BioLegend), and stained with directly conjugated antibodies according to standard procedures for cell-surface staining. Antibodies to SIRPγ (BioLegend), CD4 (BioLegend), and CD8 (560179; BD Biosciences) were used. A 7-AAD viability stain (BioLegend) was also applied. Analysis was performed with the BD Accuri C6 (BD Biosciences) and FlowJo 10.0.8r1 software.

Statistical analyses were conducted in Prism software (version 7.03). Hypotheses were evaluated with the Kruskal-Wallis test or Mann-Whitney U test. A significance level of 0.05 was defined, and Bonferroni correction was applied where appropriate.

The RNA-seq data set analyzed in the current study is available from dbGaP (https://www.ncbi.nlm.nih.gov/gap) under accession no. phs001426.v1.p1. Genotypes at rs6043409 for all of these individuals with RNA-seq data are also available in dbGaP under accession no. phs000911.v1.p1.

The rs6043409 variant encodes a conservative Ala-to-Val amino acid substitution in exon 4 of SIRPG. Neither computational predictions by SIFT (13) or PolyPhen-2 (14) nor molecular modeling using Phyre2 (15) provided evidence that this substitution would significantly impact SIRPγ protein structure or function. However, an analysis with the program ESEfinder (16) predicted that the minor allele at rs6043409 would disrupt an exonic splicing enhancer by eliminating a predicted binding site for SRSF1, a splicing factor that is highly expressed in lymphocytes. Multiple transcript isoforms of SIRPG, generated by the inclusion or exclusion of exons 3 and 4, have been reported (www.ncbi.nlm.nih.gov/IEB/Research/Acembly) (11) (Supplementary Fig. 2). Each of these exons encodes a complete immunoglobulin homology unit. Their inclusion or exclusion does not alter the reading frame of the protein but could have structural or functional consequences for SIRPγ.

CD4⁺ T cells from each of 24 unaffected subjects representing different rs6043409 genotypes were expanded in vitro, and both RNA and protein were isolated. Quantitative PCR assays were carried out with fluorescently labeled probes specific for SIRPG transcript isoforms 1 and 2, isoform 3, or isoform 4 as well as a probe detecting all isoforms. SIRPG isoforms 1 and 2 exhibited a statistically significant reduction in expression in subjects homozygous for the minor A allele as compared with those homozygous for the major G allele (P = 0.003) (Fig. 1A). Isoform 3 displayed the opposite trend, although differences between genotypes for the expression of this isoform were not statistically significant.

To extend these isoform-specific findings to larger numbers of subjects and explore their disease relevance, we examined SIRPG expression in RNA-seq data from CD4⁺ and CD8⁺ T cells obtained from 82 unrelated subjects with type 1 diabetes (9). We counted reads that mapped uniquely to specific SIRPG transcript isoforms, an approach termed Event Analysis (17). By counting events, it is possible to resolve each of the four SIRPG transcript isoforms and to compare their relative abundance. Expression of the alternative first exon unique to SIRPG transcript isoform 2 showed little or no evidence of expression, whereas isoforms 1, 3, and 4 were predicted as all exons that comprise these transcripts were detected. We then compared the estimated expression of each isoform between rs6043409 genotypes (Fig. 1B). Statistically significant differences in abundance by genotype were observed for SIRPG transcript isoforms 1 and 3 in both CD4⁺ and CD8⁺ T cells. The predominant transcript isoform encoding SIRPγ, in subjects with type 1 diabetes, was dependent on their genotype at rs6043409. Isoforms 2 and 4 did not vary significantly in abundance by genotype but accounted for only 0.74–5.47% of total SIRPG transcripts (Fig. 1B and Supplementary Fig. 3).

To determine whether all four detected SIRPG transcript isoforms produced corresponding protein isoforms, we immunoblotted for SIRPγ using increasing amounts of protein lysate from Jurkat T cells and a B lymphoblastoid cell line, Sweig (Fig. 2A). Proteins corresponding approximately to the predicted molecular weights of isoforms 1 (42 kDa), 2 (38 kDa), and 4 (30 kDa) were detectable at all protein concentrations in Jurkat cells. No protein product corresponding to the predicted molecular weight of isoform 3 (18 kDa) was detected

To evaluate the ability of each transcript isoform to produce a stable protein, the individual isoforms were cloned into expression vectors with a carboxy-terminal V5 epitope tag and transiently transfected into HEK293T cells, and total cellular protein was isolated. Probing immunoblots prepared from these transfected cells with an antibody directed against the V5 epitope tag revealed bands consistent with the production of protein by each of the four transcript isoforms (Fig. 2B). This suggests that the failure to detect an endogenously expressed protein product corresponding to isoform 3 in Jurkat cells was not due to an intrinsic instability in this SIRPγ isoform. Additional bands with molecular weights slightly greater than predicted for isoforms 1 and 4 were also detected. Treatment of lysates with the endoglycosidase PNGaseF eliminated these higher–molecular weight bands, indicating that they most likely represent glycosylated forms of SIRPγ (Supplementary Fig. 4).

SIRPG transcript isoform 3 is expressed at significantly higher levels in subjects carrying at least one copy of the minor (A) allele at rs6043409. Thus, the failure to detect a protein corresponding to this transcript isoform could be due to the fact that Jurkat cells are homozygous for the major (G) allele at rs6043409 (18). Whole cell lysates were prepared from CD4⁺ T cells purified from subjects with different genotypes at rs6043409 and immunoblotted for SIRPγ (Fig. 2C). No evidence of a protein corresponding to SIRPG transcript isoform 3 was detected despite the use of an antibody targeting a conserved sequence present in all SIRPG isoforms.

We used CRISPR/Cas9 technology to generate cloned lines of Jurkat cells with truncating mutations in exon 4 of SIRPG (Supplementary Table 1). Exon 4 is included in isoforms 1 and 2 but absent from isoforms 3 and 4 (Supplementary Fig. 2). Cell-surface and intracellular levels of SIRPγ expression in clones with either monoallelic or biallelic truncating mutations in SIRPG were quantified by flow cytometry. In clones with biallelic truncating mutations in SIRPG exon 4, where only proteins corresponding to isoforms 3 and 4 could be expressed, SIRPγ cell-surface expression was undetectable and intracellular expression of SIRPγ was negligible. However, in clones with monoallelic mutations in exon 4, some modest surface and intracellular expression of SIRPγ was detected (Fig. 3). These results were consistent when experiments were repeated with a different anti-SIRPγ antibody (Supplementary Fig. 5). Together with the immunoblotting results, these findings indicate that the majority of cell-surface SIRPγ is derived from transcript isoform 1.

We tested whether ablating SIRPγ cell-surface expression would affect cell-cell adhesion by measuring conjugate formation between CRISPR/Cas9-targeted Jurkat clones that express CD47 but not SIRPγ, and wild-type clones that express both CD47 and SIRPγ. Piccio et al. (3) previously applied this assay to a CD47-deficient Jurkat line to demonstrate the dependence of conjugate formation on CD47 expression. SIRPG-targeted clones consistently formed greater numbers of conjugates when paired with wild-type clones as compared with other targeted clones (Fig. 4A). Targeted clones also displayed significantly increased cell-surface expression of CD47 relative to wild-type clones (Fig. 4B).

RNA-seq analysis of five SIRPG-targeted clones identified a number of genes whose expression was increased or decreased relative to that in four wild-type Jurkat clones (Supplementary Fig. 7). Focusing on downregulated genes that might reflect reduced CD47 signaling in the absence of SIRPγ, we found that there was significant enrichment for TNF-related genes (corrected P = 0.039). Several genes reported to be upregulated by CD47 stimulation, such as BAX, PMAIP1, and DDIT3, were downregulated in SIRPG-targeted clones.

Since variation at rs6043409 affects the proportion of SIRPG transcripts that are spliced to different isoforms, genotypes should correlate with the cell-surface expression of SIRPγ. To test this, we measured the cell-surface expression of SIRPγ in CD4⁺ and CD8⁺ T cells from healthy subjects representing each genotype at rs6043409. For both CD4⁺ and CD8⁺ T cells, the mean fluorescence intensity of SIRPγ differed significantly between subjects with different genotypes at rs6043409 (Fig. 5). The minor allele at rs6043409 (A) was associated with reduced expression of SIRPγ in a dose-dependent manner.

We examined the expansion of CD4⁺ T cells derived from donors with different rs6043409 genotypes treated with bead-bound anti-CD3/CD28 and IL-2. Subjects with the GG genotype, who express higher levels of SIRPγ, showed a trend toward greater proliferation over the 11-day treatment period, but, overall, differences by genotype were not statistically significant (Supplementary Fig. 8).

Alternative splicing is primarily considered a mechanism for creating protein diversity through the differential incorporation of structural or functional motifs in the products of a single gene (19,20). Less well appreciated is its potential for regulating gene expression posttranscriptionally by diverting transcripts derived from a single promoter to alternative isoforms, which may have varying ability to encode functional proteins. When coupled with genetic variants that affect the efficiency or specificity of the splicing process, an opportunity for posttranscriptional quantitative regulation of gene expression is created. We demonstrate here that such a mechanism acts on SIRPG and, specifically, that a credible causative genetic variant for type 1 diabetes, rs6043409, modulates cell-surface expression of SIRPγ by altering splicing efficiency with potential impacts on cell-cell adhesion and gene expression in T cells.

The functional domain of SIRPγ that binds CD47 is encoded entirely within exon 2 of SIRPG, which is present in all transcript isoforms studied here. Exons 3 and 4 can be used alternatively or eliminated. We observed that isoform 3, which lacks both exons 3 and 4, is readily detectable as a transcript; indeed, in subjects with type 1 diabetes homozygous for the minor allele at rs6043409, it is the predominant SIRPG transcript produced. Transcript isoform 3 can produce protein after transient transfection of HEK293T cells, but no corresponding protein is detectable in primary T cells by immunoblotting or cell-surface staining. Isoform 4 is transcribed at a relatively low level and does produce a small amount of endogenous protein detectable by immunoblotting. However, this protein is not detected on the cell surface where it could engage CD47.

Like most membrane-bound proteins in the secretory pathway, SIRPγ is glycosylated. Four N-linked glycosylation sites have been reported in SIRPγ (21); all four sites are present in isoforms 1 and 2, none in isoform 3, and one site in isoform 4. N-linked glycosylation plays a significant role in quality control of protein folding in the endoplasmic reticulum as well as in stabilizing protein structure (22). One possible explanation for the failure to detect endogenous protein corresponding to transcript isoform 3 is that lack of glycosylation results in its turnover.

The major allele at rs6043409 confers risk for type 1 diabetes and, in our study, is associated with higher cell-surface levels of SIRPγ on both CD4⁺ and CD8⁺ T cells. SIRPγ lacks discernable internal signaling motifs. Its interaction with PBMCs is mediated solely by CD47 (2,3), suggesting that binding to CD47 is key to its effects. CD47 is not implicated as a risk locus for type 1 diabetes. SIRPα, which is expressed on myeloid and neuronal cells, binds CD47 with an ∼10-fold greater affinity than SIRPγ (2). Thus, it likely is not the engagement of CD47 per se that confers risk for type 1 diabetes but, rather, its engagement by a ligand, SIRPγ, expressed on T cells. Indeed, SIRPγ expression provides the only opportunity for T cells to engage and signal through CD47. Using SIRPG CRISPR/Cas9-targeted Jurkat clones, we show here that SIRPγ increases cell-cell affinity and that in its absence, genes normally upregulated by CD47 engagement are expressed at lower levels.

SIRPγ-CD47 engagement has been reported to increase CD4⁺ T cell proliferation in response to suboptimal concentrations of anti-CD3 (3). We observed a trend toward greater expansion of purified CD4⁺ T cells from donors homozygous for the major allele at rs6043409 upon anti-CD3/CD28 stimulation. However, differences between genotypes were not statistically significant, suggesting that any effect of SIRPG genotype on proliferation is likely secondary to its effects on cell-cell adhesion or only apparent in the absence of CD28 signaling. Increased SIRPγ expression might also contribute to autoimmunity by signaling through CD47 expressed on nonlymphoid cell types. For example, engagement of CD47 on endothelial cells by SIRPγ is required for trans-endothelial migration of T cells (4). The increased expression of SIRPγ associated with the risk allele at rs6043409 might enhance this process, which, in the context of type 1 diabetes, could increase access of T cells to sites of ongoing inflammation such as the pancreas.

Unlike the majority of genetic variants associated with risk for type 1 diabetes, the major allele at rs6043409 is the risk allele. Our study indicates that this allele results in increased cell-surface expression of SIRPγ on T cells. This creates a potential therapeutic opportunity, as strategies that either reduce SIRPγ expression or block its access to ligand could potentially interfere with type 1 diabetes pathogenesis.

Funding. This work was supported by grants from the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) (R01DK116954, R01DK106718) and Division of Intramural Research, National Institute of Allergy and Infectious Diseases (NIAID) (P01AI042288), National Institutes of Health. This research was performed using resources developed by the Type 1 Diabetes Genetics Consortium, a collaborative clinical study sponsored by the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK), National Institute of Allergy and Infectious Diseases (NIAID), National Human Genome Research Institute (NHGRI), National Institute of Child Health and Human Development (NICHD), and Juvenile Diabetes Research Foundation International (JDRF).

Duality of Interest. P.C. is a stockholder in Amgen, Inc. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. M.J.S. carried out experiments, directed the study, and edited the manuscript. L.P. carried out the CRISPR/Cas9 experiments and edited the manuscript. J.R.B.N. carried out the RNA-seq analyses and edited the manuscript. P.C. oversaw the study and wrote and edited the manuscript. P.C. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.