Gene-based analysis of regulatory variants identifies four putative novel asthma risk genes related to nucleotide synthesis and signaling

Abstract

Background

Hundreds of genetic variants are thought to contribute to variation in asthma risk by modulating gene expression. Methods that increase the power of genome wide association studies (GWAS) to identify risk-associated variants are needed.

Objective

To develop a method that aggregates the evidence for association with disease risk across expression quantitative trait loci (eQTLs) of a gene and use this approach to identify asthma risk genes.

Methods

We developed a gene-based test and software package called EUGENE that (1) is applicable to GWAS summary statistics; (2) considers both cis- and trans-eQTLs; (3) incorporates eQTLs identified in different tissues; and (4) uses simulations to account for multiple testing. We applied this approach to two published asthma GWAS (combined N=46,044) and used mouse studies to provide initial functional insights into two genes with novel genetic associations.

Results

We tested the association between asthma and 17,190 genes which were found to have cis-and/or trans-eQTLs across 16 published eQTL studies. At an empirical false discovery rate of 5%, 48 genes were associated with asthma risk. Of these, for 37 the association was driven by eQTLs located in established risk loci for allergic disease, including six genes not previously implicated in disease aetiology (eg. LIMS1, TINF2 and SAFB). The remaining 11 significant genes represent potential novel genetic associations with asthma. The association with four of these replicated in an independent GWAS: B4GALT3, USMG5, P2RY13 and P2RY14, which are genes involved in nucleotide synthesis or nucleotide-dependent cell activation. In mouse studies, P2ry13 and P2ry14 – purinergic receptors activated by ADP and UDP-sugars, respectively – were up-regulated after allergen challenge, notably in airway epithelial cells, eosinophils and neutrophils. Intranasal exposure with the receptor agonists induced the release of IL-33 and subsequent eosinophil infiltration into the lungs.

Conclusion

We identified novel associations between asthma and eQTLs for four genes related to nucleotide synthesis/signaling, and demonstrate the power of gene-based analyses of GWAS.

INTRODUCTION

Asthma is a highly polygenic disease, with potentially hundreds or thousands of risk variants with small effects contributing to variation in disease risk¹. A small number of risk-associated variants has been identified through genome-wide association studies (GWAS), but the majority remain to be mapped. Identifying risk-associated variants is important because these could point to genes that were not previously suspected to be involved in disease pathophysiology (eg.^{2, 3}) or that could represent drug targets with greater probability of clinical success^{4, 5}.

Several approaches have been proposed to increase the power of GWAS to identify variants with a modest but reproducible association with disease risk. These include larger sample sizes, the analysis of more refined phenotypes^{6, 7}, multivariate association analysis of related phenotypes⁸, gene-based association analyses^{9, 10}, and association analyses restricted to functional variants, such as those that regulate gene expression levels¹¹. The aim of this study was to develop a method that combined the two latter approaches and apply it to results from a published asthma GWAS to help identify new genes whose expression was associated with genotype and related to disease risk.

Specifically, we hypothesized that if the expression of a gene is causally related to asthma, and gene expression is regulated by multiple independent expression quantitative trait loci (eQTLs), then a gene-based approach that captures the aggregate signals from these eQTLs would be expected to improve power over the alternative approach of testing each variant individually. Recently, Gamazon et al.¹² described a gene-based association method based on the same concept, called PrediXcan. Briefly, this approach includes three steps: first, for a given gene, eQTLs are identified from transcriptome data sets. Second, a model that can be used to predict gene expression levels based on the aggregate effect of those eQTLs is trained on a reference transcriptome data set. And third, this model is used to infer expression levels for a target GWAS data set that includes individuals genotyped for those eQTLs but for whom actual gene expression levels might not be available. The genetically-inferred gene expression levels can then tested for association with the phenotype of interest (eg. asthma).

As highlighted by Gamazon et al.¹², PrediXcan has several advantages over other gene-based tests, such as VEGAS⁹. However, in our view, it has one major limitation: unlike VEGAS, it is not applicable to GWAS summary statistics, which are typically more readily available, and therefore can be applied to a larger sample size than available GWAS data sets with individual-level genetic data. The TWAS approach developed by Gusev et al.¹³ addresses this caveat, but in its current release is applicable only to a relatively small number of genes (4,284 from two blood eQTL studies), cis- but not trans-acting eQTLs (eg. those located >1 Mb from the target gene), and to a single reference transcriptome dataset at a time.

In this study, we developed a gene-based association approach, called EUGENE, that combines the biological focus of PrediXcan and TWAS, and the versatility of VEGAS. Our approach also considers eQTL evidence across different tissues and estimates empirical false discovery rates (FDR), while accounting for the LD between variants. We applied this new approach to a published asthma GWAS¹⁴ to try to identify novel genes whose genetic component of gene expression is associated with asthma risk. Finally, we investigated whether results from mouse models of experimental acute allergic asthma are consistent with a contribution of two selected genes to disease pathophysiology.

METHODS

EUGENE approach

The proposed gene-based approach is described in detail in the Online Repository. Briefly, for a given gene, our approach includes four steps. First, we identify a set of variants that influence gene expression in any cell type or tissue relevant to the disease or trait of interest, based on results from published eQTL studies. Including eQTLs identified in tissues not thought to be relevant for the disease of interest might improve power, but this is something we did not consider in our study. We include in this list eQTLs located in cis (< 1 Mb from the target gene) or trans (> 1 Mb away or in a different chromosome). This list is then reduced to a sub-set of eQTLs with linkage disequilibrium (LD) r²<0.1; we refer to these as “independent eQTLs” for a given gene (see Figure E1 in this article’s Online Repository at www.jacionline.org). Second, we extract association results for these independent eQTLs from a disease or trait GWAS of interest and then calculate a gene-based statistic Q, as the sum of the 1-df chi-squares for the individual eQTLs. This represents the aggregate evidence for association in that GWAS across the independent eQTLs of that gene. Third, we perform simulations using individual-level genetic data to estimate the statistical significance of Q, while accounting for the residual LD between eQTLs. Fourth, false-discovery rate (FDR) thresholds are also estimated empirically to account for multiple testing. Simulations show that the type-I error rate of EUGENE is close to the nominal expectation (Table E1 in the Online Repository). The software and input files required to run EUGENE are freely available at https://genepi.qimr.edu.au/staff/manuelF.

Application of EUGENE to published GWAS of asthma

We applied EUGENE to a published asthma GWAS¹⁴ to illustrate the utility of the proposed approach. This GWAS included 6,685 individuals with both asthma and hayfever and 14,091 asthma- and hayfever-free controls, all of European descent, tested for association with 4.9 million SNPs with a minor allele frequency >1%. In the original analysis of individual SNPs, eleven independent variants were found to be associated with disease risk at a genome-wide significance level of P<3×10⁻⁸. We used EUGENE to identify genes with an association with disease risk in the Ferreira et al. study¹⁴ at an empirical FDR of 0.05 (corresponding to a P-value threshold of 1.9×10⁻⁴). At this FDR level, 5% of genes called significant (ie. with a P<1.9×10⁻⁴) are expected to be false-positive associations due to multiple testing. To confirm putative novel associations, we then applied EUGENE to an independent asthma GWAS, the GABRIEL study¹⁵, for which summary statistics are publicly available. After excluding overlapping samples (the Busselton study), results from the GABRIEL study were based on 9,967 asthmatics and 15,301 controls.

Predicted direction of effect of gene expression on asthma risk

EUGENE can be used to identify a set of genes with expression levels determined by eQTLs, and for which the eQTLs are collectively associated with disease risk. However, unlike PrediXcan¹² or TWAS¹³, EUGENE does not directly provide the predicted direction of effect of gene expression on disease risk. To understand whether a genetically determined increase in gene expression levels was predicted to increase or decrease disease risk, we compared the direction of effect of each eQTL on gene expression reported on the transcriptome GWAS with the effect on asthma risk reported in the Ferreira et al.¹⁴ asthma GWAS. Based on this information, for each eQTL we report whether the allele associated with increased gene expression is associated with an increased or decreased asthma risk.

Functional studies in the mouse

We selected two putative novel asthma risk genes for preliminary functional studies in the mouse: P2RY13 and P2RY14. The criteria used to select these genes for functional follow-up were as follows: (1) significant gene-based association with asthma in the discovery GWAS at an empirical FDR of 5%; (2) the eQTLs that contribute to the significant gene-based association were in low LD (r²<0.1) with established asthma risk variants (those with a P<5×10⁻⁸ in published GWAS of asthma, hayfever, eczema and/or allergies); and (3) the gene-based association replicated (P<0.05) in an independent GWAS. Four genes satisfied all three criteria: P2RY13, P2RY14, USMG5 and B4GALT3. We prioritized the former two for follow-up because functional experiments were feasible with available tools/reagents (both are cell-surface receptors). We performed two sets of experiments, which are described in detail in the Online Repository and were performed in accordance with the Animal Care and Ethics Committees of the University of Queensland (Brisbane, Australia).

First, we used an established mouse model of acute allergic asthma¹⁶ to identify the cell types in the lung that express P2ry13 and P2ry14 in the context of allergen-induced airway inflammation. Two groups of wild-type C57Bl/6 mice were anesthetized and sensitized intranasally (i.n.) with either saline solution (group 1) or 100 μg of HDM extract on day 0. Subsequently, mice were challenged with either saline (group 1) or 5 μg of HDM (group 2) at day 14, 15, 16 and 17 and sacrificed 3 hours later. Total RNA was isolated from the left lung and quantitative real-time PCR performed to measure overall gene expression. To identify individual cell types in the lung expressing P2ry13 and P2ry14, bronchoalveolar lavage fluid (BALF) was collected and cells stained with anti-P2ry13 or anti-P2ry14antibodies. Cells were then stained with cell-type specific fluorescently labeled antibodies and enumerated using a BD LSR Fortessa cytometer. To assess expression in airway epithelial cells, paraffin-embedded lung sections were prepared as previously described¹⁷ and probed with anti-P2ry13 or anti-P2ry14 antibodies. Photomicrographs were taken at 400× and 1000× magnification at room temperature and acquired using Olympus Image Analysis Software.

We performed a second set of experiments to test the hypothesis that P2ry13 or P2ry14 receptor activation could influence the release of alarmins, such as IL-33, and contribute to airway inflammation. Naïve mice were inoculated i.n. with saline, 10 nM 2-methyl-ADP (P2ry13 agonist), 10 nM UDP-glucose (P2ry14 agonist) or 10 nM ATP (agonist for all P2ry receptors, except P2ry6 and P2ry14), all in 50 uL. For comparison, three additional groups of mice were inoculated with 100 ug of HDM, 100 ug of cockroach extract (Blattella germanica) or 25 ug of Alternaria alternata extract. Two hours post-challenge, BALF was collected as described above and IL-33 levels measured by ELISA. Seventy-two hours post-challenge, BALF was again collected to obtain immune cell counts and stained for flow cytometry as described above.

RESULTS

Application of EUGENE to results from a published asthma GWAS

We applied our proposed gene-based test of association to a published asthma GWAS¹⁴ including 6,685 cases and 14,091 controls to identify genes with eQTLs collectively associated with disease risk. We tested the association with 17,190 genes (Figure 1) which were found to have cis-eQTLs (N=13,557), trans-eQTLs (N=315) or both (N=3,318), across 16 published eQTL studies, representing 12 different cell types or tissues relevant to asthma (Table E2 in the Online Repository).

Figure 1.

Outline of analytical procedure

Of the 17,190 genes tested, 48 genes were associated with asthma at an empirical FDR of 0.05 (Table 1 and Figure 2). Of these, 31 (65%) were located within 1 Mb (or on the MHC region) of established risk variants for allergic disease (highlighted with a ‘+’ in Figure 2 and listed in Table E3 in the Online Repository). For example, for TSLP¹⁸ (gene-based P=7×10⁻⁶), we identified six independent cis-eQTLs in five tissues, of which four were individually associated with asthma risk at P<0.05 (Table E4 in the Online Repository). Multiple genes within the same risk locus had significant associations with asthma: 12 in the MHC¹⁵; 11 on 17q12²; three on 2q12¹⁵; and two on 16p13¹⁴. Some of these associations resulted from eQTLs being shared between neighboring genes, as observed for ORMDL3, GSDMB and ZPBP2 on 17q12 (Table E5 in the Online Repository), and for CLEC16A and SOCS1 on 16p13 (r² between rs35441874 and rs7184491 [cf. Table E3] is 0.64). eQTL sharing could arise, for example, if an underlying causal variant disrupts the activity of a regulatory element that controls the expression of multiple genes. But that was not always the case; in the MHC region, the individual LTA eQTL that was most strongly associated with asthma (P=2×10⁻⁵) was in low LD (r²≤0.02) with the eQTLs for the other 11 significant MHC genes (Table E6 in the Online Repository). Similar results were observed for NEU1. Therefore, at least in the MHC region, the multiple significant associations observed were not entirely explained by eQTLs shared between genes.

Table 1.

Forty eight genes with a significant (FDR < 5%) association with asthma risk in the Ferreira et al.¹⁴ GWAS.

Gene	Position	N eQTLs	N eQTLs tested	N eQTLs with P<0.05	Best individual eQTL		EUGENE P-value	Potential novel associationa
					SNP	P-value
HLA-DQB1	6:32627244	78	26	15	rs1063355	1.8 × 10−13	<10−6	No
GSDMB	17:38060848	15	11	5	rs2952140	1.2 × 10−8	< 10−6	No
LIMS1	2:109150857	15	14	4	rs1063355	1.8 × 10−13	< 10−6	No*
TLR1	4:38792298	9	6	3	rs12233670	1.4 × 10−11	<10−6	No
ORMDL3	17:38077294	19	14	5	rs2952140	1.2 × 10−8	<10−6	No
IKZF3	17:37921198	9	7	3	rs7207600	4.5 × 10−7	< 10−6	No
IL18RAP	2:103035149	18	16	6	rs13018263	5.0 × 10−6	<10−6	No
CLEC16A	16:11038345	4	4	2	rs35441874	2.9 × 10−8	<10−6	No
ZPBP2	17:38024417	2	2	1	rs9916765	1.9 × 10−9	<10−6	No
GRB7	17:37894180	2	1	1	rs14050	1.4 × 10−7	<10−6	No
TINF2	14:24708849	4	4	2	rs3135006	1.7 × 10−6	< 10−6	No*
TAP2	6:32789610	48	36	11	rs2858312	1.9 × 10−5	2.0 × 10−6	No
TAP1	6:32812986	13	12	4	rs6928482	2.0 × 10−8	2.0 × 10−6	No
HSPA1B	6:31795512	13	13	6	rs13215091	4.7 × 10−4	2.0 × 10−6	No
TSLP	5:110405760	6	6	4	rs17132582	3.2 × 10−4	2.0 × 10−6	No
DYNC1H1	14:102430865	6	5	2	rs4906262	1.1 × 10−5	2.0 × 10−6	Yes
HLA-DRB1	6:32546546	97	46	18	rs3806156	1.4 × 10−4	4.0 × 10−6	No
IL18R1	2:102927989	11	11	5	rs6751967	3.2 × 10−6	4.0 × 10−6	No
SOCS1	16:11348262	7	6	5	rs7184491	3.5 × 10−6	4.0 × 10−6	No
CISD3	17:36886488	4	4	2	rs2941503	1.6 × 10−7	5.0 × 10−6	No
PGAP3	17:37827375	2	2	1	rs903502	1.5 × 10−6	6.0 × 10−6	No
IL1RL2	2:102803433	2	2	1	rs9646944	6.7 × 10−7	9.0 × 10−6	No
CLK3	15:74890841	2	2	1	rs9268853	1.8 × 10−6	9.0 × 10−6	No*
SMAD3	15:67356101	7	7	2	rs17293632	2.0 × 10−7	1.1 × 10−5	No
SAFB	19:5623046	2	2	1	rs9268853	1.8 × 10−6	1.1 × 10−5	No*
P2RY13	3:151044100	8	8	5	rs9877416	1.2 × 10−4	1.2 × 10−5	Yes
AOAH	7:36552456	26	21	3	rs9268853	1.8 × 10−6	1.3 × 10−5	No*
SLC44A4	6:31830969	2	2	1	rs9275141	1.1 × 10−6	1.3 × 10−5	No
STARD3	17:37793318	7	3	1	rs2941503	1.6 × 10−7	1.5 × 10−5	No
LTA	6:31539831	16	13	5	rs2442752	1.5 × 10−5	1.6 × 10−5	No
MED24	17:38175350	5	5	2	rs7502514	4.8 × 10−5	1.7 × 10−5	No
HIBADH	7:27565061	14	12	6	rs6951856	9.6 × 10−5	2.9 × 10−5	Yes
P2RY12	3:151055168	6	6	4	rs17282940	7.0 × 10−5	3.0 × 10−5	Yes
NR1D1	17:38249040	5	4	2	rs12150298	2.8 × 10−6	3.0 × 10−5	No
ZNF707	8:144766622	9	8	3	rs17609240	1.5 × 10−6	4.2 × 10−5	No*
TOP2A	17:38544768	1	1	1	rs2102928	4.1 × 10−5	4.9 × 10−5	No
HLA-DRB6	6:32520490	60	13	6	rs522254	6.3 × 10−4	5.4 × 10−5	No
REEP3	10:65281123	6	4	2	rs7898489	9.1 × 10−6	6.1 × 10−5	Yes
PTCSC3	14:36605314	2	2	2	rs7148603	1.8 × 10−4	6.5 × 10−5	Yes
P2RY14	3:150929905	13	12	5	rs10513393	1.1 × 10−4	7.2 × 10−5	Yes
HLA-DQA1	6:32595956	79	29	9	rs504594	1.7 × 10−5	8.1 × 10−5	No
ACO2	22:41865129	3	3	2	rs960596	1.3 × 10−4	9.4 × 10−5	Yes
HCP5	6:31368479	23	22	4	rs2071595	6.7 × 10−6	9.6 × 10−5	No
NEU1	6:31825436	8	8	5	rs9267901	9.1 × 10−4	9.8 × 10−5	No
MICB	6:31462658	41	30	9	rs9268764	3.3 × 10−5	1.2 × 10−4	No
B4GALT3	1:161141100	10	9	3	rs1668873	1.5 × 10−3	1.2 × 10−4	Yes
USMG5	10:105148798	16	14	4	rs1163073	4.9 × 10−4	1.5 × 10−4	Yes
F12	5:176829141	3	3	2	rs4976765	1.7 × 10−3	1.5 × 10−4	Yes

^aPotential novel genetic associations with asthma (highlighted in grey) are those for which the gene-based association was not driven by eQTLs located in known allergy risk loci. Known allergy loci are defined as those that contain a variant reported to be associated with asthma or other allergic diseases with P<5×10⁻⁸ in published GWAS.

^*Genes not located in an asthma risk locus but for which the gene-based association was driven by trans-eQTLs in LD with allergy risk variants (see Table E7 for more details).

Figure 2. Summary of association results obtained for 17,190 genes by applying the proposed gene-based test of association to a published asthma GWAS¹⁴.

The red horizontal line shows the P-value threshold corresponding to an empirical FDR of 5% (P=1.9×10⁻⁴). Forty eight genes exceeded this threshold, including (1) 31 genes located in established risk loci for allergic disease (denoted by ‘+’); (2) six genes located in new risk loci but with a gene-based association that was driven by trans-eQTLs located in the MHC or near ORMDL3 (denoted by ‘Δ’;gene name shown in black font) and (3) 11 genes with a gene-based association that was not driven by eQTLs located in established allergy risk loci (denoted by ‘○’), including four (green font) for which the association replicated in an independent GWAS¹⁵. The y-axis represents the −log₁₀ of the simulation-derived gene-based P-value, which accounts for the residual LD between eQTLs of a given gene. The P-value was based on up to 1 million simulations, and so it could not exceed a P=10⁻⁶ (dashed grey line).

On the other hand, six (12%; LIMS1, AOAH, ZNF707, CLK3, SAFB and TINF2; ‘Δ’ in Figure 2) of the 48 genes significant at an FDR of 0.05 were not located in established risk loci for asthma but the significant gene-based associations were (in most cases, entirely) driven by trans-eQTLs located in the MHC region or near ORMDL3 (Table E7 in the Online Repository). These include for example variant rs9268853, which is a trans-eQTL for CLK3 (P=7×10⁻¹⁷) in PBMCs¹⁹, SAFB (P=3×10⁻⁶) in whole-blood²⁰ and AOAH in three tissues (best P=10⁻⁶¹,^19–21) (Table E8 in the Online Repository). This variant has also been found to be a cis-eQTL (P<5×10⁻⁸) for HLA-DQ and HLA-DR genes across multiple tissues (not shown). These results suggest that MHC and 17q12 variants might contribute to asthma risk not only by directly modulating the expression of nearby genes, but also by indirectly influencing the expression of genes in different chromosomes (eg. through cis-mediation²²).

Of potential greater interest, 11 (23%) of the 48 significant genes were located in potential novel asthma risk loci and the gene-based associations were not driven by established allergy risk variants (‘○’ in Figure 2 and Table 1). As some of these genes might represent false-positive findings, we studied their association with asthma in an independent GWAS.

Replication of the putative novel gene-based associations in an independent asthma GWAS

To confirm the putative novel associations, we applied EUGENE to an independent GWAS of asthma with publicly available summary statistics¹⁵. Based on results for 9,967 asthmatics and 15,301 controls, four of the 11 genes selected for replication had a significant gene-based association (P<0.05; Table E9 in the Online Repository), when simulations show that on average the expected number of genes significant at this threshold by chance alone given multiple testing was 0.53 (SD=0.77).

We then explored whether the discovery and replication associations for those four genes were consistent by comparing the direction of effect on disease risk for individual eQTLs. Overall, the direction of effect for most eQTLs of a given gene was the same between the two independent GWAS (Table E10 in the Online Repository). For example, of the seven eQTLs for USMG5 that were individually associated with asthma risk in either study, for six the allele that increased asthma risk was the same (or was on the same haplotype) in both studies; one eQTL was not tested in the replication GWAS, and so the direction of effect could not be compared. Therefore, the association between asthma risk and these four genes is generally consistent at the individual eQTL level between the two independent GWAS. Henceforth, we refer to these four genes with a reproducible gene-based association with asthma as “putative novel asthma risk genes”.

Contribution of cis- and trans-eQTLs to significant gene-based associations

For three (P2RY13, P2RY14, USMG5) of the four putative novel risk genes, the gene-based association with asthma was entirely driven by cis-eQTLs. Most of these eQTLs were identified by eQTL studies of whole-blood expression levels (Table E11 in the Online Repository). For the fourth gene, B4GALT3, three cis (in neutrophils, blood and fibroblasts) and one trans (in blood) eQTL contributed to the association with asthma (Table E11 in the Online Repository). The latter (rs1668873) was located 44 Mb away on chromosome 1 and was previously reported to associate with mean platelet volume and count^{23, 24}. This variant is also a cis-eQTL for NUAK2²⁰, a nuclear transcriptional modulator that has been shown to induce the expression of B4GALT5²⁵, a galactosyltransferase related to B4GALT3²⁶. Therefore, these results suggest that both direct (cis-eQTLs) and indirect (via transcriptional modulators such as NUAK2) genetic effects on B4GALT3 expression might contribute to asthma risk.

Genetically predicted direction of effect of gene expression on asthma risk

To assess the direction of effect of gene expression on disease risk, we focused on the independent eQTLs for each gene that were individually associated with asthma in the discovery and/or replication GWAS. These variants had the greatest contribution to the significant gene-based tests. When we compared the direction of effect for each eQTL between asthma risk and expression levels, we found that the allele associated with increased gene expression was also associated with increased asthma risk for all independent eQTLs of P2RY13 and P2RY14 (Table E11 in the Online Repository). The same pattern of results was observed for six of the seven eQTLs of USMG5; for example, the rs1163073:C allele that was associated with asthma risk (OR=1.09, P=0.0005), was associated with increased USMG5 expression in five different cell types or tissues (neutrophils, LCLs, skin, PBMCs and blood). These results suggest that in the tissues or cell types considered in our analysis, a genetically determined increase in gene expression for each of these four genes is associated with increased disease risk. For B4GALT3, there was no clear pattern across multiple eQTLs: of the four alleles associated with increased gene expression, two (in neutrophils and whole-blood) were associated with increased and two (in fibroblasts and whole-blood) with decreased disease risk. Such differences between eQTLs could arise, for example, if B4GALT3 has opposing functional effects on different cell types relevant to asthma (eg. activation in one, inhibition in another). Further studies are required to test this possibility.

Functional studies in the mouse

The four putative novel asthma risk genes identified in our genetic association analysis are involved in nucleotide synthesis (B4GALT3, USMG5) and nucleotide-dependent cell activation (P2RY13, P2RY14). Based on this observation, we hypothesise that genetic dysregulation of nucleotide signaling contributes to asthma risk. In depth functional experiments that comprehensively test this hypothesis were beyond the scope of this study. Nonetheless, we carried out two sets of experiments in the mouse to provide preliminary functional support for the involvement in allergic asthma for two of these four nucleotide-related genes: P2RY13 and P2RY14. Both are cell-surface receptors with known agonists, and so were well suited for functional studies.

First, we used an established experimental model of acute allergic asthma¹⁶ to study P2ry13 and P2ry14 expression in the lungs of C57Bl/6 mice sensitized and subsequently challenged with house dust mite (HDM) allergen. In this model, mice develop granulocytic airway inflammation that has a predominant eosinophil contribution¹⁶. When considering overall lung expression, HDM challenge resulted in a significant increase in P2ry13 and P2ry14 expression, relative to control mice challenged with a saline solution (Figure 3A). To understand which lung cell types contributed to this increase in gene expression, we used flow cytometry to measure protein expression in airway epithelial cells (AECs) and major immune cell types collected through bronchoalveolar lavage. There was widespread expression of both receptors in AECs, both at baseline and after HDM challenge (Figure 3B and 3C). Most eosinophils collected in BALF after HDM challenge stained positive for both receptors (Figure 3D); expression in neutrophils was also high (Figure 3E). Lymphocytes and dendritic cells had low expression of both receptors (Figure E2 in the Online Repository).

Figure 3. Expression levels of P2ry13 and P2ry14 in lung of C57Bl/6 mice sensitized and then challenged with a saline solution or a house dust mite extract.

(A) Overall gene expression in lung. Expression levels were normalized to Hprt and are expressed as fold-change over saline challenge group. Results show mean +/− SD in each group. * Wilcoxon rank sum test P < 0.005 when comparing HDM and saline groups. (B, D and E) Expression of P2ry13 and P2ry14 based on flow cytometry analysis in lung epithelial cells, or eosinophils and neutrophils collected in BALF after saline or HDM challenge. (C) Expression of P2ry13 and P2ry14 in lung sections of mice challenged with saline or HDM.

Secondly, given the high level of P2ry13 and P2ry14 expression observed in AECs at baseline (ie. in the absence of allergen challenge), and the previously reported pro-inflammatory effect of the respective agonists (eg. ^{27, 28}), we postulated that receptor activation could promote airway inflammation by inducing the release of alarmins. To test this possibility, we collected BALF from naïve mice 2 and 72 hours after intra-nasal challenge with saline, ADP (selective P2ry13 agonist) or UDP-glucose (selective P2ry14 agonist). At 2 hours post challenge, BALF levels of the alarmin IL-33 were significantly greater in mice exposed to the receptor agonists than in control mice (Figure 4A). Of note, nucleotide-induced IL-33 levels were comparable to allergen-induced IL-33 levels, indicating that both ADP and UDP-glucose are sufficient to induce IL-33 release. Furthermore, 72 hours after challenge, the number of BALF eosinophils and lymphocytes were significantly higher in agonist-treated mice (Figure 4B and 4C), but this was not the case for neutrophils, dendritic cells and monocytes (Figure E3 in the Online Repository). These results demonstrate that selective agonists of P2ry13 and P2ry14 can promote airway inflammation, even in the absence of allergen stimulation.

Figure 4. In vivo exposure to P2ry13 and P2ry14 receptor agonists in naïve C57Bl/6 mice.

Mice were challenged with either vehicle, one of three allergens (HDM, CRE, Alt) or one of three nucleotides (ATP, UPD-glucose, ADP), and euthanized 2 and 72 hours after challenge. (A) IL-33 expression in bronchoalveolar lavage fluid (BALF) collected 2 hours post challenge. Total number of eosinophils (B) and lymphocytes (C) recruited to the BALF at 72 hours post challenge, based on flow cytometry analysis. Veh: vehicle. HDM: house dust mite. CRE: cockroach. Alt: alternaria.

DISCUSSION

Dysregulation of gene expression is thought to be a common mechanism by which genetic variants can influence cellular function and, ultimately, variation in human traits and disease risk. This proposition is supported by the observation that eQTLs are more likely to be trait-associated when compared to random variants¹⁸ and was the motivation for the gene-based approach developed in this study.

EUGENE has some advantages when compared to other gene-based association approaches, of which we highlight three: VEGAS⁹, PrediXcan¹² and TWAS¹³. When compared to VEGAS, EUGENE avoids the requirement to use an arbitrary distance (same for all genes) from the known gene boundaries to define which SNPs to include in the gene-based analysis. If the distance is too small (eg. +/−5 kb), then the contribution of important more distantly located eQTLs might be missed, while a large distance (eg. +/− 1 Mb) could result in testing a large number of variants, many of which are likely to be unrelated to gene expression/function; in both cases, the power to detect a significant gene-based association is reduced. Also, because the analysis in EUGENE is restricted to variants previously shown to influence gene expression, whether in cis or in trans, a significant gene-based trait association directly implies that genetically-determined differences in gene expression contribute to trait variation. On the other hand, when compared to the recently described PrediXcan approach¹², the main advantage of EUGENE is that it is applicable to GWAS summary statistics, which are typically easier to share than data sets with individual level genetic data. TWAS¹³, which is conceptually very similar to PrediXcan, is applicable to summary statistics but its current release only includes weights for a relatively small number of genes obtained from three eQTL studies, although this is likely to increase in the future. Another major difference when compared to both of these approaches is that with EUGENE, eQTLs identified in transcriptome studies of different cell types (and/or upon cell stimulation) can be included in the same association analysis, all contributing with equal weight to the gene-based statistic. This might be important for traits or diseases for which multiple cell types or tissues are known to play a role in the underlying pathophysiology. In the PrediXcan and TWAS approaches, the weights assigned to different eQTLs in the model used to predict gene expression levels are based on the effect (eg. regression coefficient) of those variants on expression levels measured in a single reference transcriptome data set. The extent to which those weights remain appropriate (ie. yield good prediction) if the reference transcriptome and target data sets are very different (eg. in age composition) is unclear. A disadvantage of EUGENE is that the direction of effect between gene expression and disease risk (or trait variation) is not directly inferred. To do so with the EUGENE approach, the effect of individual eQTLs that contribute to the gene-based test needs to be compared ad-hoc between the transcriptome and trait GWAS. These eQTLs also provide a specific small group of variants to test in validation studies. Lastly, EUGENE (but not the other three approaches) estimates FDR thresholds empirically, taking into account the LD between eQTLs of the same or different genes. This is important to account for multiple testing.

When we applied EUGENE to a published GWAS of asthma, we identified 48 genes with a significant gene-based association at an FDR of 0.05, including 11 associations that were not driven by established genetic risk variants for allergic disease. For four of these genes (B4GALT3, USMG5, P2RY13 and P2RY14), the association was nominally significant in an independent asthma GWAS and so we refer to these as putative novel asthma risk genes.

B4GALT3 encodes the widely-expressed enzyme β-1,4-galactosyltransferase III that catalyzes the transfer of galactose from UDP-galactose to N-acetylglucosamine, to form N-acetyllactosamine and UDP^{26, 29}. How variation in B4GALT3 expression might contribute to asthma risk is unclear, but potential mechanisms include activation of β1 integrin³⁰, which is important in the initiation of T-cell inflammatory responses³¹, or by influencing extracellular release of UDP-galactose³², a P2RY14 agonist.

USMG5 encodes a small subunit of ATP synthase³³, an enzyme responsible for ATP synthesis in the mitochondria. USMG5 knockdown in HeLa cells causes the loss of ATP synthase, resulting in lower ATP synthesis and slower cell growth³⁴. In CD4+ T-cells, mitochondria produce the ATP that is rapidly released into the extracellular space upon cell stimulation³⁵. In turn, this ATP establishes an autocrine feedback through purinergic receptors that is essential for proper T-cell activation³⁶. Given these observations, we speculate that genetically-determined increased USMG5 expression results in increased mitochondrial production of ATP, increased extracellular ATP release and increased T-cell activation. In turn, this would translate into an increased risk of asthma. ATP synthase has also been detected at the surface of different cell types³⁷, where it is thought to play different physiological roles, for example, HDL endocytosis in hepatocytes via P2RY13 activation³⁸ and non-conventional T-cell activation³⁹. Whether USMG5 associates with membrane ATP synthase, and so could potentially influence its ectopic roles, remains to be determined.

P2RY13, also known as GPR86 or GPR94, is a purinergic receptor highly expressed in the immune system, lung and skin, but also in the brain^40–42; it displays a significant homology with the nearby P2RY12 and P2RY14 genes, sharing 48 and 45% amino acid identity⁴⁰. P2RY13 is strongly activated by ADP⁴⁰, a degradation product of ATP. Airway epithelial goblet cells are a major source of extracellular ADP, which is released as a co-cargo molecule from mucin-containing granules⁴³. In turn, ADP has been shown to enhance antigen-induced degranulation in mast cells, through a P2RY13-dependent mechanism²⁷. ADP has also been reported to promote IL-6 release from keratinocytes⁴², inhibit TNF-alpha and IL-12 production by mature DCs⁴⁴ and promote chemotaxis of immature DCs⁴⁵; however, these studies did not specifically test if the observed ADP effects were mediated by P2RY13. Results from our genetic association analyses indicate that a genetically-determined increase in P2RY13 expression increases asthma risk, which is consistent with the pro-inflammatory effect suggested for ADP and P2RY13 by these functional studies.

P2RY14, also known as GPR105, encodes a G protein-coupled receptor that is potently and selectively activated by UDP-sugars, especially UDP-glucose⁴⁶. UDP-glucose is thought to be an extracellular pro-inflammatory mediator⁴⁷, constitutively released by different cell types including airway epithelial cells⁴⁸. Importantly, infection with respiratory syncytial virus (RSV) or treatment with IL-13 significantly increases UDP-glucose release by airway epithelial cells⁴⁹, and this coincides with increased mucus secretion⁵⁰. Known pro-inflammatory effects of UDP-glucose acting through P2RY14 include inhibition of TLR9-dependent IFN-alpha production⁵¹, increased chemotaxis of neutrophils⁵², induction of mast cell degranulation⁵³, production of IL-8^{28, 54} and STAT3-dependent epidermal inflammation⁵⁵. A small molecule antagonist for P2RY14 was recently developed and shown to effectively block chemotaxis of freshly isolated human neutrophils⁵⁶. Of note, plasma UDP-glucose levels are elevated in mice fed a high-fat diet⁵⁷, which raises the possibility that obesity might contribute to chronic P2RY14 activation and, in that way, increase asthma risk and/or severity. Studies that investigate this possibility are underway.

When we studied the expression of both P2ry13 and P2ry14 in mice, we found that both genes were highly expressed in AECs, both at baseline and after allergen challenge. High expression was also observed in infiltrating eosinophils after challenge and, to a smaller extent, neutrophils and monocytes. High expression in AECs suggested that receptor activation could contribute to airway inflammation even in the absence of allergen stimulation. This was indeed what we observed when naïve mice were challenged intra-nasally with either ADP or UDP-glucose: 72 hours after challenge, the numbers of BALF eosinophils were significantly increased when compared to control mice. Interestingly, eosinophil influx into the airways was preceded by a significant increase in the levels of the alarmin IL-33. The effect of both receptor agonists on IL-33 release was comparable in magnitude to that observed with allergens known to have a potent effect on IL-33 production, namely the fungus Alternaria alternata⁵⁸. These results demonstrate that activation of P2ry13 and P2ry14, in addition to P2y2⁵⁸, can strongly induce IL-33 release in mice. Interestingly, P2ry13 expression was observed in the nuclei of AECs after allergen challenge. Given that IL-33 is constitutively stored in the nuclei of AECs⁵⁹, it is possible that intracellular activation of P2ry13 expressed on the nuclear membrane plays a role in allergen-induced IL-33 release.

In conclusion, our genetic findings establish an association between asthma risk and genes involved in nucleotide synthesis (B4GALT3, USMG5) and nucleotide-dependent cell activation (P2RY13, P2RY14). In mice, in vivo activation of P2ry13 and P2ry14 induced IL-33 release and subsequent eosinophilic airway infiltration. These observations suggest that genetic dysregulation of nucleotide signaling contributes to the risk of asthma (allergic and, potentially, also non-allergic) and other related conditions; studies that test this possibility are now warranted. Our results also show that re-analysis of published GWAS with a gene-based test that exclusively focuses on documented eQTLs has the potential to identify novel associations.

KEY MESSAGES.

• In humans, asthma risk is associated with genetically-determined expression of four genes related to nucleotide synthesis (B4GALT3, USMG5) and nucleotide-dependent cell activation (P2RY13 and P2RY14).

• In mice, intranasal exposure with selective agonists for P2ry13 (ADP) or P2ry14 (UDP-glucose) induced the release of IL-33 and eosinophil infiltration into the lungs, in the absence of allergen stimulation.

CAPSULE SUMMARY.

Using a new method for gene-based analysis of GWAS results, we identified a genetic association between asthma risk and eQTLs for B4GALT3 and USMG5, which are involved in the production of UDP-galactose and ATP respectively, and P2RY13 and P2RY14, two G protein-coupled receptors activated respectively by ADP and UDP-sugars. Functional studies in the mouse show that activation of P2ry13 or P2ry14 induces the release of IL-33 and eosinophil infiltration into the lungs, in the absence of allergen stimulation. Functional studies that characterize in depth the contribution of these four genes to asthma pathophysiology are warranted.

ABBREVIATIONS

AECs

airway epithelial cells

ADP

adenosine 5-diphosphate

Alt

Alternaria allergen

ATP

adenosine 5-triphosphate

BALF

bronchoalveolar lavage fluid

CRE

cockroach allergen

DC

dendritic cells

eQTL

expression quantitative trait locus

FDR

false discovery rate

GWAS

Genome Wide Association Study

HDM

house dust mite. IL: interleukin

LCLs

lymphoblastoid cell lines

LD

linkage disequilibrium

SNP

single nucleotide polymorphism

UDP

uridine-diphosphoglucose (UDP-glucose)