Reduced monocyte and macrophage TNFSF15/TL1A expression is associated with susceptibility to inflammatory bowel disease

Chronic inflammation in inflammatory bowel disease (IBD) results from a breakdown of intestinal immune homeostasis and compromise of the intestinal barrier. Genome-wide association studies have identified over 200 genetic loci associated with risk for IBD, but the functional mechanisms of most of these genetic variants remain unknown. Polymorphisms at the TNFSF15 locus, which encodes the TNF superfamily cytokine commonly known as TL1A, are associated with susceptibility to IBD in multiple ethnic groups. In a wide variety of murine models of inflammation including models of IBD, TNFSF15 promotes immunopathology by signaling through its receptor DR3. Such evidence has led to the hypothesis that expression of this lymphocyte costimulatory cytokine increases risk for IBD. In contrast, here we show that the IBD-risk haplotype at TNFSF15 is associated with decreased expression of the gene by peripheral blood monocytes in both healthy volunteers and IBD patients. This association persists under various stimulation conditions at both the RNA and protein levels and is maintained after macrophage differentiation. Utilizing a “recall-by-genotype” bioresource for allele-specific expression measurements in a functional fine-mapping assay, we localize the polymorphism controlling TNFSF15 expression to the regulatory region upstream of the gene. Through a T cell costimulation assay, we demonstrate that genetically regulated TNFSF15 has functional relevance. These findings indicate that genetically enhanced expression of TNFSF15 in specific cell types may confer protection against the development of IBD.


Discussion of cell-type specificity of the eQTL at TNFSF15
While we cannot formally exclude the possibility of a TNFSF15 eQTL with a small effect size that we were under-powered to detect in stimulated T cells, we do not have evidence to reject the null hypothesis of no eQTL. In our recall-by-genotype eQTL analysis (Fig 2B), we had 10 versus 9 samples homozygous for each allele in stimulated CD4 + T cells and 9 versus 9 in stimulated CD8 + T cells. We conducted a t-test power calculation to determine the effect size (here, change in mean expression between minor and major allele homozygotes divided by the pooled standard deviation) that we would be able to detect at a significance threshold of p 0.05.
We emphasize that the effect size described here is distinct from the beta estimate from eQTL linear regression. We determined that we had 80% power to detect an effect size of 1.37 and 1.41 in stimulated CD4+ T cells and CD8+ T cells, respectively. For comparison, the observed effect size (as defined above) in ex vivo monocytes, where we did find a significant eQTL, was 2.58. In addition to conventional eQTL analysis, we also examined allele-specific expression (ASE) in stimulated T cells from heterozygous individuals. This assay compares the ratio of transcripts within each heterozygous individual and is therefore not affected by inter-individual variation in gene expression, thus providing greater power. Even with the ASE assay, we found no significant allelic imbalance in either stimulated T cell subset. Furthermore, visual inspection of the ASE plots did not support an eQTL. Therefore, our data point to a strong monocyte eQTL but do not support a substantial stimulated T cell eQTL.

Additional discussion of previous functional studies of the IBD risk locus at TNFSF15
Previous studies describing gene expression association with the IBD-associated locus at TNFSF15 have yielded conflicting results. We identified an association of decreased monocyte TNFSF15 expression with the IBD risk allele. We here discuss how our data relate to additional previous studies not described in the main text.
Our eQTL results are concordant with the data from two previous reports by Michelsen et al [1] and Sun et al [2], but these studies incorrectly attributed IBD risk to the opposite allele.
Michelsen et al [1] focused on the Ashkenazi Jewish population where CD susceptibility 2 haplotypes at TNFSF15 were suggested to be reversed [3], but subsequent genetic association studies in this population have not found such an association [4,5]. Thus, the haplotype associated with increased TNFSF15 expression in Michelsen et al is the haplotype associated with IBD protection in the general European population, which we also find to be associated with elevated TNFSF15. Sun et al [2] examined genotype association with mRNA from whole blood in a healthy Chinese cohort and reported the rs6478109:A allele to be associated with increased TNFSF15 expression, consistent with our results. However, they reported this allele to be part of the IBD risk haplotype; this risk attribution is inconsistent with previous GWAS results where rs6478109:A has been directly reported as protective for CD in Japanese individuals [6], as well as in the Liu et al GWAS meta-analysis and a previous studiy of pediatric onset IBD in European individuals [7,8]. Furthermore, the T risk allele that Sun et al reported at rs6478108, the SNP most strongly associated with CD in their study [2], is out of phase with the A risk allele reported at the eQTL SNP rs6478109 in both East Asian and European individuals (S1B Fig), suggesting an error in their analysis. Together, these results clearly indicate that the protective allele rs64718109:A is associated with increased monocyte TNFSF15 expression as we have demonstrated.
Two additional small studies have reported the opposite direction of effect of the TNFSF15 eQTL. Zucchelli et al observed increased TNFSF15 mRNA in peripheral blood leukocytes (n=25) and rectal biopsies (n=15) from healthy donors with the rs4263839:G risk allele (in phase with the rs6478109-G risk allele) [9]. However, it is difficult to compare these data from mixed cell populations with our data from purified monocytes, and findings from a small sample size must be interpreted with caution. In contrast to a report by Kakuta et al, we found no allelespecific expression of TNFSF15 in stimulated T cells [10]. Kakuta et al activated whole PBMC with PHA and total T cells with PMA and ionomycin [10], whereas we activated separate populations of CD4 + and CD8 + T cells with anti-CD3 and anti-CD28. It is possible that the results differ due to cell type composition or stimulus, but we believe that our additional purification provides strong evidence against a substantial T cell eQTL.
Finally, two large-scale studies found a TNFSF15 eQTL at the same narrowed LD block that we identified in resting monocytes [11] and influenza-stimulated monocyte-derived dendritic cells [12], but the reference allele was not specified in either study.

Assays with elutriated monocytes and cells sorted from leukocyte cones:
Genotype-independent assays were performed using positively selected subsets from leukocyte cones obtained through the National Health Service (NHS) Blood and Transplant service at Addenbrooke's Hospital or elutriated monocytes from the National Institutes of Health Department of Transfusion Medicine. Leukocyte cones were processed as described in Methods "Whole blood processing". Elutriated monocytes were passed through a 100 μm strainer and washed in PBS before plating.
Allele-specific expression allelic ratio calculation: Allelic ratios were calculated from VIC and FAM Ct values by fitting a line to the standard curve according to the following rationale. For allele F amplified by the FAM-labeled probe with efficiency EF from initial concentration Fi to cross the threshold concentration at cycle CtF, and likewise for allele V amplified by the VIC-labeled probe, Rearranging and taking the natural logarithm reveals that Assuming equal efficiencies for VIC and FAM probe amplification increases robustness to pipetting error as each VIC versus FAM comparison may be calculated within each individual reaction. Thus with this assumption We fitted a line to this equation using the Ct values from VIC and FAM probes in the standard curve dilutions. This intercept and slope were then used to then calculate starting allelic ratios in the cDNA and gDNA samples. To adjust for slight measurement errors possible with such dilute DNA samples in the standard curve, all allele-specific expression ratios were normalized to the average gDNA ratio measured in the individual assay. Immunophenotyping: Flow cytometry immunophenotyping of PBMC was performed using 5 different panels, as detailed in S5