Association of ST2 polymorphisms with atopy, asthma and leukemia

To the editors

The phenome-wide association study (PheWAS) is a new, validated, reverse genetics approach that associates genetic variants of interest with phenotypes by linking a database of de-identified genotyping to a broad range of electronic medical record (EMR)-derived clinical phenotypes.¹ While the more common genome-wide association study (GWAS) is useful in identifying genetic polymorphisms that may relate to one particular disease phenotype, the PheWAS’s ability to evaluate these polymorphisms against an entire database of clinical phenotypes greatly expands the scope of genomic association research. Using a PheWAS, one is able to examine polymorphisms detected in the context of a disease to not only confirm this established association with greater certainty, but one may uncover links to entirely new phenotypes that had not been previously considered or investigated at the genetic level.¹

IL-33 is one of the most consistently associated gene candidates for asthma identified by GWASs in diverse ethnic groups.² IL-33 is a central mediator of both innate and adaptive immunity regulated allergic inflammation in the lung.^3,4 The primary binding partner of IL-33, ST2, is a member of the IL-1 receptor family that is expressed on a number of immune cells, including ILC2, CD4+ Th2 cells, mast cells, and eosinophils.⁵ GWASs investigating polymorphisms related to asthma have identified a number of single nucleotide polymorphisms (SNPs) in the ST2 receptor gene.^6,7 We sought to investigate these polymorphisms to assess the role of ST2 signaling in other, non-asthma related allergic diseases. In addition, we hoped to identify new diseases in which genetic links to the ST2 receptor had not been reported. We performed a PheWAS on 15 ST2 genetic variants implicated in asthma to assess their associations with other allergic, and non-allergic diseases. The Exome BeadChip that we used to perform the PheWAS genotyped 38 variants in IL1R1. We only included variants with a minor allele threshold (MAF) of greater than 1%; this filtering is necessary to support a PheWAS analysis using a logistic regression. The PheWAS utilized a cohort of 36,400 adult (age > 18 years) subjects who had genotyping on Illumina HumanExome BeadChip version 1.1 and available electronic health record (HER) data from the Vanderbilt BioVU DNA biobank that was ascertained previously for five different criteria: (1) eligibility as case or control in one of 31 pharmacogenetics studies (2) available longitudinal data with primary care visits (3) presence in the cancer registry (4) old age with longitudinal data (5) presence of rare diseases or conditions ascertained via billing codes. Of the 15 SNPs, six had mean allele frequencies (MAFs) greater than 1% in the total cohort population, rendering them sufficiently powered for further analysis.¹ All of these six SNPs are in the ST2 gene (human chromosome 2q11.2).⁶ The PheWAS yields results as a list of phenotype ICD9 codes that best associated with each particular SNP.

Our study revealed that the minor allele of ST2 SNP rs1041973 associated with allergic conjunctivitis (OR=1.27), an allergic disease that had not been linked genetically in the context of IL-33/ST2 signaling. We further demonstrate that the minor allele of rs1041973 exhibits an inverse association with eosinophilia, as indicated by an odds ratio (ORs) < 1 (0.686, Table 1). In addition, the PheWAS confirmed the previously reported inverse relationship of 5 ST2 SNPs with asthma (rs10192036 OR=0.876, rs1041973 OR=0.505, rs4988956 OR=0.876, rs10192157 OR=0.876, rs10206753 OR=0.876), and it confirmed the formerly identified association of rs1420101 (OR=1.20) with asthma.^6,7 We calculated the pair-wise LD for each of the 6 variants tested and found that the minor alleles of rs10192036, rs10206753, rs4988956, and rs10192157 are in perfect linkage (r²=1). For this reason, the associations of these 4 SNPs are reported under only rs4988956. These genomic investigations with study populations entirely separate from our own provide ORs in similar directions for all six SNPs investigated in this study, confirming the previously described association of each.^6,7 This further supports the validity of our method by demonstrating that genetic and phenotypic data from Vanderbilt University Medical Center patients is representative of national and international populations.⁶ Finally, the PheWAS uncovered a novel association of 2 SNPs in the ST2 gene with various forms of leukemia (Table 2). The minor allele of the SNP rs1420101 associated with leukemia and acute myeloid leukemia, while the minor allele of rs1041973 associated with leukemia, acute myeloid leukemia, and acute lymphoid leukemia.

Table I.

Allergic and asthmatic phenotypes significantly associate with polymorphisms in the ST2 gene.

SNP	MAF (%)	PheWAS Phenotype	PheWAS Phenotype Code	Cases	MAF (%) Cases	Controls	MAF (%) Controls	OR	P value	P value (FDR)
rs1041973	22.41	Allergic conjunctivitis	371.21	420	26.79	25276	22.31	1.27	2.0×10−3	0.50
		Wheezing	512.1	470	19.26	14672	22.66	0.813	1.4×10−2	0.654
		Eosinophilia	288.3	122	16.53	21272	22.40	0.686	3.0×10−2	0.67
		Chronic obstructive asthma with exacerbation	495.11	43	12.79	23477	22.54	0.505	3.4×10−2	0.67
rs4988956	37.78	Asthma	495	1886	34.89	23477	37.92	0.876	2.0×10−4	0.28
		Allergies, other	949	116	27.16	23713	37.68	0.614	1.0×10−3	0.36
rs1420101	38.05	Asthma	495	1886	34.89	23477	37.92	1.20	1.074×10−7	1.5×10−4
		Allergies, other	949	116	27.16	23713	37.68	1.45	5.1×10−3	0.77
		Asthma with exacerbation	495.2	316	42.88	23477	37.82	1.25	6.2×10−3	0.77

Definition of abbreviations: OR= Odds Ratio, MAF= Mean Allele Frequency, FDR=False Discovery Rate Correction, PheWAS= Phenome- Wide Association Study

Table II.

Leukemia, acute Myeloid Leukemia, and acute Lymphoid Leukemia significantly associate with polymorphisms in the ST2 gene.

SNP	MAF (%)	PheWAS Phenotype	PheWAS Phenotype Code	Cases	MAF (%) Cases	Controls	MAF (%) Controls	OR	P value	P value (FDR)
rs1420101	38.05	Leukemia	204	1006	40.95	26852	37.92	1.13	5.9×10−3	0.77
		Myeloid Leukemia, acute	204.21	323	41.95	26852	37.92	1.18	3.6×10−2	0.92
		Myeloid Leukemia, acute	204.21	323	26.16	26852	22.23	1.24	1.8×10−2	0.66
rs1041973	22.41	Lymphoid Leukemia, acute	204.11	323	25.85	26852	22.23	1.23	2.4×10−2	0.66
		Leukemia	204	1006	24.20	26852	22.23	1.12	3.7×10−2	0.66

Definition of abbreviations: OR= Odds Ratio, MAF= Mean Allele Frequency, PheWAS= Phenome-wide association study, FDR= False Discovery Rate

The results of this study inform our understanding of the ST2 pathway as it relates to other, non-asthmatic forms of allergic disease. We are also the first to identify rs1041973 as a genetic variant that associates inversely with eosinophilia. In an analysis of peripheral blood eosinophilia for eosinophilia-associated rs1041973 cases and controls, there was a significant increase in the maximum recorded percentage of peripheral blood eosinophilis in cases (13.5%) compared to controls (3.4%, p= 3.15E-165, Supplemental Table 1). Since SNPs examined in this study were originally identified in a GWAS as relevant to asthma, the associations that we report here with asthma serve as another indicator of the reliability of PheWAS as an analytical technique, and of the validity of the asthma-ST2 connection. We repeated our study in an independent cohort of over 500,000 subjects, the UK Biobank.⁸ We verified our previously identified allergic and asthmatic phenotype associations with ST2 polymorphisms in the UK Biobank using self- reported phenotype codes (Supplemental Table 2). Finally, our findings establish the first genetic link between the ST2 gene and leukemia. We were unable to confirm our associations to lymphoma and leukemia due to an insufficient number of these disease cases in the UK Biobank cohort. However, the association between ST2 polymorphisms and multiple forms of leukemia can be supported by a handful of in vivo and in vitro studies that have reported IL-33 and ST2 to be potential mediators of leukemic cell proliferation.⁹ These findings identify novel genetic risk factors for leukemia, and provide further support for the investigation of this signaling pathway in leukocyte tumorigenesis. The significance of these results is further demonstrated by the fact that 5 of the 6 SNP’s discussed in this study have MAFs over 37%. These polymorphisms are present in a substantial portion of the United States population, and their continued investigation will no doubt yield widespread benefit.

While PheWASs such as this one offer inexpensive and efficient insight into the association of disease with specific genetic variants, the model itself has certain intrinsic limitations. Firstly, a selection bias exists, as the genotyping information available for the study comes from patients who have had blood drawn for laboratory testing. Additionally, a majority of the subjects who provided blood were admitted to the emergency department or hospital, further compromising the true randomness of the test samples. Furthermore, the study reports uncorrected P values as well as P values adjusted with the false discovery rate correction. A Bonferroni correction may be used to reduce the likelihood of Type I statistical error. However, this study was performed as an exploratory analysis. Therefore, we excluded the use of multiple testing corrections to reduce our Type II error rate. In situations such as this study where the number of tests is large, multiple testing corrections can be overly conservative, possibly contributing to missed or overlooked results. A limitation of our approach is that it depends on the prior identification of SNPs in other studies or the sufficiently high prevalence of SNPs in our cohort of study. For this reason, the number of SNPs available for PheWAS is limited, and important genetic associations in the ST2 gene likely remain undiscovered. In addition, inherent limitations exist in the study’s dependence on disease phenotype diagnosis with ICD codes. ICD-9 billing codes are often a poor and unspecific diagnostic tool, and variation exists in the application of these codes by physicians and the ability of the algorithms to assign code designations to more narrative, textual diagnoses. This is particularly true in the case of asthma in which the diagnoses are difficult to confirm in the EHR, as not every patient will have lung function or methacholine challenge tests performed.

In conclusion, these data demonstrate the utility of PheWAS by identifying links between polymorphisms in the ST2 gene and multiple forms of leukemia. Further, they implicate specific ST2 SNPs in allergic diseases not previously tied genetically to ST2, and they confirm the important role of ST2 signaling in asthma.