Genome-Wide Association Study Identifies Novel Loci Associated With Diisocyanate-Induced Occupational Asthma

Abstract

Diisocyanates, reactive chemicals used to produce polyurethane products, are the most common causes of occupational asthma. The aim of this study is to identify susceptibility gene variants that could contribute to the pathogenesis of diisocyanate asthma (DA) using a Genome-Wide Association Study (GWAS) approach. Genome-wide single nucleotide polymorphism (SNP) genotyping was performed in 74 diisocyanate-exposed workers with DA and 824 healthy controls using Omni-2.5 and Omni-5 SNP microarrays. We identified 11 SNPs that exceeded genome-wide significance; the strongest association was for the rs12913832 SNP located on chromosome 15, which has been mapped to the HERC2 gene (p = 6.94 × 10⁻¹⁴). Strong associations were also found for SNPs near the ODZ3 and CDH17 genes on chromosomes 4 and 8 (rs908084, p = 8.59 × 10⁻⁹ and rs2514805, p = 1.22 × 10⁻⁸, respectively). We also prioritized 38 SNPs with suggestive genome-wide significance (p < 1 × 10⁻⁶). Among them, 17 SNPs map to the PITPNC1, ACMSD, ZBTB16, ODZ3, and CDH17 gene loci. Functional genomics data indicate that 2 of the suggestive SNPs (rs2446823 and rs2446824) are located within putative binding sites for the CCAAT/Enhancer Binding Protein (CEBP) and Hepatocyte Nuclear Factor 4, Alpha transcription factors (TFs), respectively. This study identified SNPs mapping to the HERC2, CDH17, and ODZ3 genes as potential susceptibility loci for DA. Pathway analysis indicated that these genes are associated with antigen processing and presentation, and other immune pathways. Overlap of 2 suggestive SNPs with likely TF binding sites suggests possible roles in disruption of gene regulation. These results provide new insights into the genetic architecture of DA and serve as a basis for future functional and mechanistic studies.

Diisocyanates are the leading causes of occupational asthma (OA), estimated to cause asthma in 5%–15% of chronically exposed workers (Bakerly et al., 2008; Bernstein, 2011; Redlich and Karol, 2002). The most common isomers used in industry are: the aliphatic agent 1,6 hexamethylene diisocyanate (HDI), 4, 4-diphenylmethane diisocyanate (MDI), and toluene diisocyanate (TDI). Despite improved industrial hygiene efforts, new cases of diisocyanate asthma (DA) continue to occur (Booth et al., 2009; Campo et al., 2010; Kenyon et al., 2012). The National Institute for Occupational Safety and Health (NIOSH) National Occupational Exposure Survey database showed that at least 280 000 workers were potentially exposed to some form of diisocyanates in the United States alone (NIOSH, 1983). TDI was reported to account for between 2.9% and 13% of all cases of occupational asthma in Korea (Park et al., 2002). Considering the extent of workplace exposure to diisocyanates, identification of genetic variants increasing susceptibility to DA could be important in developing new risk assessment and preventive strategies for chronically exposed workers. In addition, incorporation of genetic data into human health risk assessment will be important for refining occupational exposure limits for diisocyanates to improve protection of worker health.

We have previously identified a number of DA associated variants of: HLA genes (HLA-B, HLA-E, HLA-DOA, HLA-DPB1, and HLA-DQA2) (Yucesoy et al., 2014), Th2 immune response-associated genes (IL-4Rα, IL-13, and CD14) (Bernstein et al., 2006, 2011) and antioxidant defense genes (SOD2, GST, and EPHX1) (Yucesoy et al., 2012). Although genome-wide association studies (GWAS) have been performed in various asthma phenotypes, only one GWAS has investigated workers with OA. This study was conducted in 84 diisocyanate-exposed Korean workers with TDI-asthma confirmed by specific inhalation challenge (SIC) and 263 unexposed healthy controls and identified 4 CTNNA3 (alpha-T catenin) single nucleotide polymorphisms (SNPs) associated with DA (Kim et al., 2009).

In this study, we aimed to increase our understanding of genetic susceptibility to DA by conducting a GWAS in a different background population of Caucasian subjects. Although population admixture, specific diisocyanate exposure, and genotyping platforms differed between the 2 GWAS studies, we searched for common overlapping genetic loci between the Korean population and ours. Our findings revealed several genes/SNPs that may offer promising avenues for future genetic and functional studies of DA susceptibility genes.

MATERIALS AND METHODS

Study participants

The study population included 88 diisocyanate-exposed workers with DA and 832 self-reported healthy Caucasian individuals (Prahalad et al., 2000). Diisocyanate-exposed workers, referred for clinical evaluation of possible DA by occupational pulmonary disease clinics located in Canada (Sacré-Coeur Hospital, Montreal; Laval Hospital, Sainte-Foy; University Health Network, Toronto) and Spain (Fundación Jiménez Díaz, Madrid and Hospitals Vall D’Hebron, Barcelona), were recruited for participation in this study. To confirm or exclude DA, controlled SIC testing was performed according to published protocols with the relevant diisocyanate chemical to which the worker was exposed (Malo et al., 1999; Sastre et al., 2003). A positive SIC response was defined as a fall in forced expiratory volume in 1 second (FEV₁) of at least 20% from prechallenge baseline following a controlled exposure to a diisocyanate chemical generated in a challenge chamber. For subjects recruited and evaluated in Toronto, lung function data collected during diisocyanate exposure at work was used to establish a DA diagnosis. Data pertinent to demographics, smoking status, duration, and nature of chemical exposure were collected in all subjects. Atopic status was evaluated by skin prick testing to commercial aeroallergen extracts. Atopy was defined by a positive response with one or more common aeroallergens defined by a wheal of 3 mm ≥ saline control. Whole blood was collected for genetic testing.

Ethics

All subjects gave written informed consent. The study protocol conformed to the ethical guidelines of the 1975 Declaration of Helsinki and was approved and renewed annually by the Ethical Research Committee of Hôpital du Sacré Cœur de Montréal (Montréal, Québec, Canada), the Research Ethics Committee of the Institut Universitaire de Cardiologie et de Pneumologie de Québec (Sainte Foy, Québec, Canada), University Health Network Research Ethics Board (Toronto, Ontario, Canada), the Ethics Committee of Clinical Research (IRB) of Fundación Jiménez Díaz-UTE (Madrid, Spain), and Val d’Hebron University Ethics Committee (Barcelona, Spain). The comparator group was recruited from the Cincinnati Genomic Control Cohort (Kovacic et al., 2011). This study was approved by the Institutional Review Board at Cincinnati Children’s Hospital Medical Center. Written informed consent for the purpose of DNA collection and genotyping was obtained from all subjects. Parents gave written informed consent for the children’s participation, and children gave assent.

Genotyping

Whole blood samples were collected for genetic analysis and genomic DNA was extracted using the QIAamp blood kit (QIAGEN, Chatsworth, California). Genome-wide SNP genotyping was performed on 88 cases using Omni-2.5 SNP microarray (Illumina, San Diego, California). The chip contains approximately 2.5 million SNP markers with an average call frequency of >99% and is unbiased with respect to coding and noncoding regions of the genome. 832 controls were run on the Omni-5 microarray which contains approximately 4.3 million SNPs. A total of approximately 2.4 million markers were common to both arrays and used for analysis. As starting material, 250 ng of genomic DNA was used for the assay. Internal quality control measures were used for the data obtained from each chip. Genotypes were auto called using GenomeStudio software (Illumina). A total of 7 individuals with genotyping call rate <95% were removed and the genotyping rate in the remaining individuals was 97%. Genome-wide data were used to infer the top 6 principal components of genetic variation and correct for population stratification using Goldenhelix SVS (Bozeman, Montana). Samples were removed from analysis if they segregated >4 SDs outside of the mean of the first 4 principal components. Figure 1 shows the principal component analysis (PCA) plot of the first 2 components based on 100 000 randomly selected SNPs. Minor allele frequencies of SNPs in the control group were also compared with those reported in the phases 1 and 3 of the 1000 Genomes Project. These analyses indicated that both cases and controls were of European origin and there was no population stratification. We also examined LD patterns around the top-ranked SNPs and identified multiple linked SNPs. LocusZoom association plots showing local LD patterns for the top-ranked SNPs are presented in Supplementary Figure 1.

FIG. 1.

PCA plot showing the first two components based on 100 K randomly selected SNPs.

Of the SNPs assayed on the chip, approximately 835 000 SNPs were excluded because of a low call rate, low minor allele frequency (MAF), or deviation from Hardy–Weinberg equilibrium (HWE) in the controls (P < .0001). About 74 cases and 824 controls satisfying all QC measurements were used for analysis. Markers were excluded if they had a MAF < 0.01 in controls and <0.03 in cases. As a result of this basic QC, 1 556 551 SNPs were included in the final analysis with an average sample call rate of 99.98% and a SNP call rate of 99.84%.

Statistical analyses

HWE was carried out using a Pearson goodness-of-fit test. The association between each SNP and DA was assessed using a chi-square test using Goldenhelix SVS (Bozeman, Montana). In order to investigate previously identified regions within the Korean population, we examined SNPs within 25 kb of reported associations.

TF Binding Site Analysis

For identifying SNPs that might impact TF binding events in cell types relevant to DA, we first compiled functional genomics data from several sources, including the ENCODE Project Consortium (2012), Roadmap Epigenomics (Bernstein et al., 2010), and the UCSC Genome Browser (Dreszer et al., 2012). Custom software was used to restrict to cell types relevant to asthma (including cell lines and cells derived from lung tissues) and identify SNPs that intersect these datasets. We then identified those among the remaining SNPs that are located within ChIP-seq binding peaks for TFs. The resulting SNPs were further filtered to those likely to impact TF binding by scanning each ChIP-seq peak for the most likely binding site for the associated TF using position weight matrix (PWM) models taken from Weirauch et al. (2014), and the standard log likelihood scoring system (Stormo, 1990), and restricting to those SNPs that are located within these binding sites. The resulting SNPs are therefore (1) located in likely regulatory regions in relevant cell types; (2) located within a ChIP-seq binding peak for a TF; and (3) located within the most likely binding site for the corresponding TF within that binding peak.

RESULTS

Clinical Characteristics of the Study Subjects

The demographic characteristics of the study groups included in the statistical analyses are described in Table 1. The DA group consisted of 88 subjects of European Caucasian descent and 1 Asian with a mean age of 40.7 years (between 22 and 64 years). The control group included 832 self-reported non-Hispanic Caucasian healthy individuals ranging in age from 2 to 62 years with a mean age of 15.8 years, and consisted of 406 males and 426 females.

Table 1.

Demographic Characteristics of the Study Groups

	DA+	Controls
N	88	832
Sex, M/F	77/11	406/426
Mean age ± SD range	40.7 + 11.9 (22–64)	15.8 ± 12.5 (2–62)
Diisocyanate exposure (HDI/MDI/TDI)	44/19/25	N/A
Duration of exposure mean months ± SD	147.9 ± 137.8	N/A
Skin prick test Positive/negative/unknown	53/31/4	N/A
Smoker (Current/ex/never)	16/34/38	N/A
Mean pack-years ± SD (All subjects)	10.7 ± 14.0	N/A
Mean pack-years ± SD (Current & Ex-smokers)	19.0 ± 13.3	N/A
Race White/non-White	87/1	832/0
Ethnicity (French Canadian/Othera)	60/28	N/A

Abbreviations: DA, diisocyanate asthma; N/A, not applicable.

^aOther includes English Canadian (3), Spanish (18), Italian (3), Polish (1), Moroccan (1), Ecuadorian (1), Indian (1).

Genetic Associations

Top-ranked genes are presented in Table 2. Eleven SNPs exceeded the threshold for genome-wide significance (p < 10⁻⁷) and 3 of them were mapped to the HERC2, CDH17, and ODZ3 genes. As seen in Table 3, the most significant association was observed for the rs12913832 SNP mapping to the HERC2 gene (p = 6.94 × 10⁻¹⁴), located at position 28365618 on chromosome 15q13. The second most significant signal was for the rs12568266 SNP (p = 6.20 × 10⁻¹⁰). This is an intergenic variant and located on chromosome 1 between the hCG_2036596 (left) and LOC644357 (right) genes. The next 2 SNPs (rs74380195 and rs76966929, p = 9.58 × 10⁻¹⁰) were located on chromosome 11 and were not mapped to any gene. The next significant variant (rs7588010, p = 4.08 × 10⁻⁹) was also not mapped to any gene but is located between the TACR1 (left) and FAM176A (right) genes on chromosome 2. Following 2 markers were intron variants and mapped to the ODZ3 (rs908084, p = 8.59 × 10⁻⁹) and CDH17 (rs2514805, p = 1.22 × 10⁻⁸) genes. The rest of the markers reaching a genome-wide significance were rs16867528 (p = 1.65 × 10⁻⁸), rs12143327 (p = 4.57 × 10⁻⁸), rs79252495 (p = 5.85 × 10⁻⁸), and rs76684306 (p = 9.90 × 10⁻⁸) SNPs. They were intergenic variants and located on chromosomes 5, 1, 15, and 5, respectively. Except HERC2 rs12913832 SNP, none of these top SNPs have been previously implicated in any disease.

Table 2.

Top-Ranked Genes

Entrez Gene ID	Gene Symbol	Gene Name	Number of SNPs	p Values (range)
8924	HERC2	HECT and RLD domain containing E3 ubiquitin protein ligase 2	1	6.94 × 10−14
55714	ODZ3 (TENM3)	Teneurin transmembrane protein 3	3	8.59 × 10−9–5.64 × 10−7
1015	CDH17	Cadherin 17, liver-intestine cadherin	4	1.22 × 10−8–4.74 × 10−7
130013	ACMSD	Aminocarboxymuconate semialdehyde decarboxylase	1	6.35 × 10−7
26207	PITPNC1	Phosphatidylinositol transfer protein, cytoplasmic 1	6	6.33 × 10−7–7.82 × 10−7
7704	ZBTB16	Zinc finger and BTB domain containing 16	5	1.68 × 10−7–7.03 × 10−7
Suggestive genes
64067	NPAS3	Neuronal PAS domain protein 3	3	2.42 × 10−6–1.69 × 10−5
5578	PRKCA	Protein kinase C, alpha	1	2.55 × 10−6
6539	SLC6A12	Solute carrier family 6 (neurotransmitter transporter), member 12	3	1.47 × 10−6–1.67 × 10−5
6869	TACR1	Tachykinin receptor 1	13	4.08 × 10−9–7.95 × 10−5

Table 3.

Genome-Wide Association Results for Significant (p < 1 × 10⁻⁷) and Suggestive (p < 1 × 10⁻⁶) SNPs

SNP	Chr	Position	Gene	Feature	Risk Allele	MAF Cases	MAF Controls	p value	OR	95% L	95% U
rs12913832	15	28365618	HERC2	Intron	A	0.513	0.233	6.94 × 10−14	3.475	2.468	4.891
rs12568266	1	4193551	NA	NA	A	0.127	0.029	6.20 × 10−10	5.017	2.859	8.803
rs74380195	11	102887229	NA	NA	G	0.173	0.050	9.58 × 10−10	4.018	2.493	6.478
rs76966929	11	102891066	NA	NA	A	0.173	0.050	9.58 × 10−10	4.018	2.493	6.478
rs7588010	2	75473916	NA	NA	A	0.593	0.357	4.08 × 10−9	2.719	1.928	3.836
rs908084	4	183543648	ODZ3	Intron	A	0.153	0.044	8.59 × 10−9	3.970	2.401	6.564
rs2514805	8	95167247	CDH17	Intron	G	0.28	0.118	1.22 × 10−8	2.952	2.004	4.349
rs16867528	5	15607448	NA	NA	A	0.068	0.010	1.65 × 10−8	7.391	3.291	16.598
rs12143327	1	112610895	NA	NA	A	0.213	0.081	4.57 × 10−8	3.142	2.045	4.829
rs79252495	15	26722016	NA	NA	A	0.12	0.032	5.85 × 10−8	4.250	2.415	7.477
rs76684306	5	15616676	NA	NA	G	0.073	0.011	9.90 × 10−8	6.562	2.971	14.492
Suggestive associations
rs118079879	12	41713306	NA	NA	A	0.073	0.013	1.25 × 10−7	5.927	2.815	12.478
rs10769691	11	6375809	NA	NA	A	0.193	0.072	1.57 × 10−7	3.131	2.003	4.895
rs1672691	11	113942151	ZBTB16	intron	A	0.38	0.200	1.68 × 10−7	2.498	1.756	3.553
rs7115199	11	102871623	NA	NA	A	0.1	0.024	1.85 × 10−7	4.534	2.441	8.421
rs74609360	15	31540395	NA	NA	A	0.113	0.030	2.08 × 10−7	4.142	2.323	7.386
rs72974161	2	136802456	NA	NA	A	0.24	0.102	2.11 × 10−7	2.832	1.883	4.258
rs2340023	8	95157476	CDH17	Intron	A	0.267	0.120	2.47 × 10−7	2.712	1.833	4.014
rs76314368	5	116492833	NA	NA	A	0.093	0.022	2.63 × 10−7	4.672	2.459	8.879
rs3018331	11	113940469	ZBTB16	Intron	A	0.347	0.178	2.83 × 10−7	2.505	1.747	3.592
rs11722354	4	183565618	ODZ3	Intron	A	0.122	0.035	2.86 × 10−7	3.925	2.243	6.870
rs2513797	8	95143138	CDH17	CS	G	0.267	0.121	3.39 × 10−7	2.681	1.812	3.967
rs13017967	2	75461177	NA	NA	A	0.54	0.338	3.46 × 10−7	2.368	1.687	3.325
rs6716987	2	136963494	NA	NA	A	0.48	0.287	4.28 × 10−7	2.353	1.676	3.305
rs6442708	3	1994644	NA	NA	A	0.155	0.053	4.41 × 10−7	3.351	2.043	5.496
rs2446824	8	95127574	NA	NA	A	0.267	0.123	4.64 × 10−7	2.651	1.792	3.922
rs2513791	8	95128529	NA	NA	A	0.267	0.123	4.64 × 10−7	2.651	1.792	3.922
rs2446823	8	95127612	NA	NA	C	0.267	0.123	4.64 × 10−7	2.651	1.792	3.922
rs2262592	8	95155550	CDH17	Intron	G	0.267	0.122	4.74 × 10−7	2.678	1.802	3.980
rs11731869	4	183559826	ODZ3	Intron	A	0.12	0.035	5.64 × 10−7	3.796	2.172	6.633
rs2846629	11	113941756	ZBTB16	Intron	G	0.38	0.207	5.72 × 10−7	2.401	1.689	3.413
rs4852364	2	75457783	NA	NA	A	0.3	0.512	5.92 × 10−7	0.404	0.280	0.582
rs12328696	2	75460555	NA	NA	A	0.3	0.512	5.93 × 10−7	0.404	0.280	0.582
rs1784684	11	113939539	ZBTB16	Intron	G	0.347	0.182	6.20 × 10−7	2.44	1.702	3.498
rs7224314	17	65386214	PITPNC1	Intron	C	0.413	0.223	6.33 × 10−7	2.372	1.675	3.359
rs4954192	2	135632981	ACMSD	Intron	A	0.627	0.410	6.35 × 10−7	2.362	1.671	3.340
rs2309284	4	181103729	NA	NA	A	0.093	0.023	6.63 × 10−7	4.427	2.340	8.375
rs62084082	17	65388046	PITPNC1	Intron	G	0.413	0.224	6.98 × 10−7	2.364	1.669	3.348
rs2735199	11	113942191	ZBTB16	Intron	G	0.38	0.208	7.03 × 10−7	2.385	1.677	3.390
rs60244812	9	7267614	NA	NA	A	0.107	0.029	7.05 × 10−7	4.035	2.230	7.302
rs1491921	5	21259138	NA	NA	C	0.068	0.013	7.41 × 10−7	5.690	2.626	12.327
rs62085810	17	65372855	PITPNC1	NG-5	A	0.40	0.219	7.63 × 10−7	2.363	1.667	3.352
rs62084078	17	65386886	PITPNC1	Intron	G	0.413	0.225	7.82 × 10−7	2.355	1.663	3.335
rs62084077	17	65386837	PITPNC1	Intron	A	0.413	0.225	7.82 × 10−7	2.355	1.663	3.335
rs62084080	17	65387014	PITPNC1	Intron	A	0.413	0.225	7.82 × 10−7	2.355	1.663	3.335
rs4954572	2	136926719	NA	NA	A	0.473	0.286	8.19 × 10−7	2.307	1.643	3.241
rs75646498	15	29876649	NA	NA	A	0.115	0.030	8.29 × 10−7	3.996	2.211	7.224
rs6858365	4	12172669	NA	NA	A	0.073	0.015	8.72 × 10−7	5.213	2.511	10.820
rs116146467	6	169558013	NA	NA	A	0.067	0.013	9.32 × 10−7	5.614	2.592	12.160

Abbreviations: ACMSD, aminocarboxymuconate semialdehyde decarboxylase; CS, coding synonymous; CDH17, cadherin 17; HERC2, HECT and RLD domain containing E3 ubiquitin protein ligase 2; MAF, minor allele frequency; NG-5, near gene-5; ODZ3 (TENM3), teneurin transmembrane protein 3; PITPNC1, phosphatidylinositol transfer protein, cytoplasmic 1; ZBTB16, zinc finger and BTB domain containing 16; 95% U, upper bound; 95% L, lower bound.

OR, odds ratio in risk genotype relative to common genotype according to each analysis model.

We also prioritized 38 suggestive genome-wide significant SNPs (p < 1 × 10⁻⁶) based on their potential involvement in DA (Table 3). Among them, 17 SNPs map to the PITPNC1, ACMSD, ZBTB16, ODZ3, and CDH17 genes. None of the prioritized candidates have been previously implicated in asthma. The full set of results is available in Supplementary Table 1.

In addition, 4 suggestive genes NPAS3, PRKCA, SLC6A12, and TACR1 were identified based on previous reports related to airway diseases and proximities to our significant SNPs (Table 4). Supplementary Table 2 presents complete genetic association results for these genes.

Table 4.

Suggestive Associations Based on Previous Reports and Proximity to Significant SNPs

SNP ID	Chr	Position	Gene	Chi-squared p value
rs28564912	14	33798766	NPAS3	2.42 × 10−6
rs56962758	14	33769003	NPAS3	1.67 × 10−5
rs9635191	14	33746767	NPAS3	1.69 × 10−5
rs4636942	17	64793346	PRKCA	2.55 × 10−6
rs188610	12	313839	SLC6A12	1.47 × 10−6
rs7138605	12	320510	SLC6A12	1.67 × 10−5
rs16928441	12	321353	SLC6A12	1.67 × 10−5
Gene left to the significant SNPs
rs7588010	2	75473916	TACR1	4.08 × 10−9
rs13017967	2	75461177	TACR1	3.46 × 10−7
rs4852364	2	75457783	TACR1	5.92 × 10−7
rs12328696	2	75460555	TACR1	5.93 × 10−7
rs4853124	2	75449986	TACR1	2.14 × 10−6
rs7583283	2	75478007	TACR1	4.70 × 10−6
rs4853128	2	75491046	TACR1	2.14 × 10−5
rs1861430	2	75470713	TACR1	4.31 × 10−5
rs7576919	2	75471057	TACR1	4.31 × 10−5
rs10490307	2	75482223	TACR1	4.35 × 10−5
rs4508624	2	75479258	TACR1	4.69 × 10−5
rs7605785	2	75453787	TACR1	6.43 × 10−5
rs11684394	2	75444073	TACR1	7.95 × 10−5

Abbreviation: SNP, single nucleotide polymorphism.

Cut-off, p < 1 × 10⁻⁴.

We also investigated chromosome X and identified 8 SNPs that exceed nominal significance (p < 1 × 10⁻⁴). Top-ranked SNPs did not map to any gene. These results are presented in Supplementary Table 3.

Functional Analysis of Genetic Variants

In order to gain insight into molecular mechanisms, we examined whether genes cluster into particular biological pathways, and searched for their expression profiles in the lung using publicly available databases. We also predicted functional relationships between the top 3 genes using the GeneMANIA webserver (Warde-Farley et al., 2010). This analysis suggested that these genes are associated with adaptive immunity, antigen processing, ubiquitination and proteasome degradation, and class I MHC mediated antigen processing and presentation. Figure 2 shows interactive functional association network related to genome-wide significant loci. Considering significant and suggestive genes, we found a significant clustering of genes involved in protein binding, metabolic processes, response to stimulus, and biological regulation using WEB-based GEne SeT AnaLysis Toolkit (Wang et al., 2013). We also used the CGAP EST cDNA library (http://cgap.nci.nih.gov) and the BioGPS gene expression portal to explore the expression profiles of our top-ranked genes in normal human lung tissues (Wu et al., 2013) All candidate genes were expressed in the lung to a varying extent [the number of expressed sequence tag (EST) per 200 000 tags in the CGAP EST cDNA library were 4–7 tags for HERC2, <2 tags for ODZ3, CDH17, AMCSD, ZBTB16, and 2–3 tags for PITPNC1].

FIG. 2.

Interactive functional association network among the top-ranked genes (HERC2, CDH17, and ODZ3). The relationship between the genes in the network includes coexpression, physical and genetic interactions, pathways, colocalization, protein domain similarity, and predicted interactions. The network was filtered by removing all the interactions where weights <0.1.

Effects of SNPs on TF Binding Sites

We used a large collection of TF DNA binding models to predict specific TFs whose binding might be affected by the alleles of each SNP (Weirauch et al., 2014). Based on multiple types of data (including histone marks, ChIP-seq for TFs and chromatin binding proteins, and DNase-seq peaks), this procedure identified a locus 3′ of the CDH17 gene encompassing the rs2446823 and rs2446824 SNPs. This region is likely to play an important regulatory role in several lung cell lines (ie, A549, AG04450, IMR90, and NHLF), as well as lung and fetal lung tissues. These SNPs, which are separated by 37 bases, were located within ChIP-seq peaks for the CEBPA/B (CCAAT/Enhancer Binding Protein) (in U937, A549, and IMR90 cells) and HNF4A (Hepatocyte Nuclear Factor 4, Alpha) (in Caco-2 cells) TFs, respectively.

Overlaps Between Korean and Caucasian GWAS

We detected association signals within 25 kb of significant SNPs in the Korean DA GWAS. These SNPs were mapped to CTNNA1, CTNNA3, TRPM8, KCNIP4, DOCK2, TUSC3, SAMD12, ASTN2, CRTAC1, LHPP, PDGFD, and PCNX genes. Figure 3 shows Manhattan plot of the p values obtained from our and Korean GWAS. 146 SNPs were mapped to chromosome 10 where top hit CTNNA3 gene was located. This gene has been identified as a susceptibility locus in the Korean study and replicated by our group (Bernstein et al., 2013). The overlapping regions have been presented in Supplementary Table 4.

FIG. 3.

Manhattan plot of the p values obtained from Korean (open diamonds) and Caucasian (filled circles) GWAS studies. A total of 146 SNPs were mapped to chromosome 10 (marked in a square) where the top hit gene (CTNNA3) in the Korean study was located.

DISCUSSION

In this study, we conducted a GWAS for DA in a Caucasian population of European background and identified SNPs reaching genome wide significant associations with confirmed occupational asthma caused by isocyanates. The top ranked SNP mapped to the HERC2 (HECT and RLD domain containing E3 ubiquitin protein ligase 2) gene. Although there were 3 SNPs mapping to this gene, only 1 (rs12913832) exceeded genome-wide significance (p = 6.9 × 10⁻¹⁴). As one of the E3 ubiquitin ligases, HERC2 is involved in the addition of ubiquitin to proteins undergoing degradation. It has also been implicated in DNA replication and DNA damage response (Bekker-Jensen et al., 2010; Izawa et al., 2011). Protein ubiquitination is an important regulatory posttranslational process that controls intracellular signaling and antigen processing. Due to its fundamental role in cell signaling, the ubiquitin system plays an important role in cancer immunity (Sun, 2006). The HERC2 rs12913832 SNP has been implicated in immune (eg, ulcerative colitis) and developmental conditions (eg, pigmentation, iris color) (Franke et al., 2008; Visser et al., 2012). In a recent study, N-succinimidyl N-methylcarbamate (NSNM), a surrogate chemical containing a functional isocyanate group, was used to understand the mechanism behind isocyanate toxicity in the model organism Saccharomyces cerevisiae. The biological effect of NSNM on the growth of yeast deletion mutants of various pathways was investigated. The deletion mutants of ubiquitylation factors and mitochondrial chaperones were found to be hypersensitive to NSNM in a growth assay suggesting that isocyanates specifically target these pathways (Azad et al., 2014). Our results support this observation and suggest an interaction between genetic variability within this gene involved in the ubiquitylation pathway and DA. Elucidation of the functional role of this SNP may provide new insights into the incompletely defined pathogenesis of DA.

The next marker (rs908084) reaching genome-wide significance (p = 8.59 × 10⁻⁹) mapped to the ODZ3 (teneurin transmembrane protein 3, also called TENM3) gene. The protein encoded by ODZ3 (4q34.3) belongs to the teneurin subfamily and is involved in homophilic cell adhesion and regulation of neuronal development. The role of this gene in the asthmatic process is unknown. Due to its role in cellular adhesion, it is possible that ODZ3 variants could play a functional role in the airway epithelium response to diisocyanate exposure. An external search of the catalog of published genome-wide association studies did not reveal associations of this SNP with other asthma phenotypes (http://www.genome.gov/gwastudies).

The other genome-wide significant marker (rs2514805, p = 1.22 × 10⁻⁸) mapped to the CDH17 (cadherin 17) gene. CDH17 gene is located on chromosome 8q22.1 and encodes liver–intestine–cadherin. Cadherins are cell–cell adhesion molecules that play an important role in maintaining tissue structure and morphology. CDH17 functions as a peptide transporter and cell adhesion molecule to maintain tissue integrity in epithelia. In an Australian GWAS, another CDH17 SNP (rs11776675) was found to be 1 of 5 independent SNPs that were associated with heightened asthma risk at a pre-defined cut-off of p ≤ 5 × 10⁻⁶ (OR = 1.21, p = 2.7 × 10⁻⁶). However, this result was not replicated in a follow up meta-analysis of 4 independent asthma studies (Ferreira et al., 2011). We found 9 SNPs in this region with only 1, rs2514805, reaching genome-wide significance (p = 1.22 × 10⁻⁸). Although expression of CDH17 is low in the lungs, significant markers mapping to the CDH17 gene in Australian and current GWAS studies suggests potential involvement of this gene in the asthmatic process by possibly impacting epithelial barrier integrity. Existence of additional linked SNPs around the top-ranked variants further supports the association of these genomic regions with DA (see Supplementary Fig. 1).

Based on previously reported associations with other asthma phenotypes and proximities to our significant and suggestive genes in Table 3, we identified 4 additional suggestive genes, NPAS3, PRKCA, SLC6A12, and TACR1. NPAS3 SNPs have been identified in asthmatics of different backgrounds as one of the shared genetic factors (Ding et al., 2013). NPAS3 is known to activate or suppress multiple distinct signaling pathways in lung development and repair, and its deficiency was reported to be associated with emphysema and asthma (Zhou et al., 2009). In our analysis, we identified 7 SNPs mapping to this gene and only 3 of them achieved suggestive p values. PRKCA that plays a role in cellular transformation has been identified as a positional candidate gene for asthma (Murphy et al., 2009). We obtained one significant signal mapping to the PRKCA gene with a suggestive p value. SLC6A12 SNPs have been associated with aspirin-intolerant asthma in a Korean population (Pasaje et al., 2010). We identified 3 suggestive SNPs mapping to this gene which is known to play a critical role in mucus production in asthma. TACR1 (tachykinin receptor 1) gene was located left of 13 SNPs that were not mapped to any gene (Table 3). Tachykinins have been found to activate NF-kB and stimulate proinflammatory cytokine expression in epithelial cells following lung injury (Williams et al., 2007). In animal models for TDI-induced asthma, tachykinins were found to play a role in the development of airway hyperresponsiveness (Mapp et al., 1998; Scheerens et al., 1996). Our results provide novel insights about potential involvement of these genes in the DA process.

Results from recent studies suggest that as many as 93% of disease- and trait-associated SNPs are located in regulatory regions (Hindorff et al., 2009; Maurano et al., 2012), indicating that many SNPs might act by affecting the binding of TFs. We therefore used functional genomics data to identify SNPs located in likely regulatory regions in cell types relevant to DA. We found that 2 suggestive intergenic SNPs (rs2446823 and rs2446824) were located within ChIP-seq peaks for the CEBPA/B and HNF4A TFs. Both SNPs are likely located within the specific region bound by each TF, suggesting that each might function by impacting the binding of their respective TF(s). Intriguingly, CEBPB and HNF4A are well known interaction partners in a variety of cellular contexts (Schmidt et al., 2010) and previous studies have linked asthma to impaired translation of CEBPs (Borger et al., 2007; Miglino et al., 2012). It has been suggested that reduced CEBP levels stimulate cell proliferation and the release of proinflammatory cytokines, resulting in airway inflammation.

Based on systems-level functional analysis (Warde-Farley et al., 2010), our top 3 genes (CDH17, HERC2, and ODZ3) are all involved in pathways related to antigen presentation/binding and the immune response (Fig. 2). Although the pathology of DA has been considered similar to common environmental asthma, recent evidence suggests that immune responses elicited by low molecular weight agents such as diisocyanates are distinct from those induced by common environmental allergens (high molecular weight) (Wisnewski et al., 2008). Associations of DA with variants of genes involved in antigen processing and adaptive immunity is biologically plausible considering the different nature of mechanisms involved. We also searched our genes to see whether they are related to a particular molecular/biological function. The findings showed that significant genes (genome-wide and suggestive) were mainly related to protein binding and biological regulation. Although unproven, genetic variations influencing their ability to either bind or transport small (usually charged) molecules could be mechanistically relevant to poorly understood immune mechanisms contributing to DA (Redlich and Karol, 2002).

In addition, we searched GWAS results from a Korean population for any overlapping signals. This study reported a novel locus on chromosome 10q21 encompassing the CTNNA3 gene (encoding catenin alpha 3), and an additional 54 suggestive SNPs. Although there were no exactly overlapping SNPs with our study, we detected significant signals in the proximity of their top ranked SNPs (Supplementary Table 4). The sheer number of signals in this same region is highly unlikely to be a chance observation, and suggests that this genomic region might contain common susceptibility variants. Mapping these regions in detail to identify causal variants, followed by exploration of causal mechanisms may yield new insights into the pathogenesis of DA.

It should be noted that this study has several limitations. Although this was the largest GWAS study of DA to date, our sample size is small due to the relative rarity of DA compared to other types of asthma. However, rigorous phenotypic characterization of this population helps to maximize the discriminatory potential between study groups and with a larger control group, this study was better powered to identify statistically significant signals. Although the controls were genetically homozygous to the cases based on PCA, they were from a younger, healthy population not exposed to diisocyanates. DA occurs as a consequence of exposure to diisocyanates and our control population had negligible potential for future exposure. Considering this and the difficulty in recruiting age-matched workplace controls, we took advantage of an existing large cohort recruited from the general population. Another limitation of this study is the lack of a replication stage for those loci that were most highly associated with DA. As such, the highly associated loci remain suggestive until confirmed in future studies.

Although we detected signals around significant SNPs, we were unable to replicate results of the previous GWAS in Korean workers with DA. Most GWAS findings do not replicate consistently, which may be related to ancestry variations in the study populations, differences in asthma phenotype definitions, type of exposure, unaccounted environmental factors, and the coverage of genotyping platform used. The previous GWAS platform (Affymetrix 500) had a lower coverage as compared to Illumina Omni-2.5 and Omni-5 SNP microarrays which may result in missing several genes that might interact with diisocyanate exposure. In addition, the Korean population was exposed only to TDI whereas our population was exposed to HDI, TDI, and MDI.

In summary, this study identified 3 novel genome-wide significant loci and prioritized an additional 5 genome-wide suggestive candidate genes for DA. In addition to several novel candidate genes, we found that rs2446823 and rs2446824 SNPs could affect binding of CEBP and HNF4A TFs. Although this study is exploratory and novel signals need to be followed up in a large independent sample, these results offer new avenues for future studies of genes contributing to DA susceptibility.

FUNDING

NIOSH/CDC (R01 OH 008795), University of Cincinnati Research Council, Interdisciplinary Faculty Research Support Grant (2012).

SUPPLEMENTARY DATA

Supplementary data are available online at http://toxsci.oxfordjournals.org/.