Epistasis between SPINK5 and TSLP Genes Contributes to Childhood Asthma

Abstract

Background

Epithelial genes have previously been associated with asthma, but only explain a small fraction of heritability. In part, this may be due to epistasis that is often not considered.

Objective

To determine independent and epistatic associations between FLG, SPINK5 and TSLP gene variants and childhood asthma.

Methods

Using a candidate gene approach, we genotyped 29 variants in FLG, SPINK5 and TSLP in asthmatic, allergic, and non-allergic-non-asthmatic white and black children participating in the well-phenotyped Greater Cincinnati Pediatric Clinic Repository (GCPCR). Associations with asthma were also assessed in six replication populations.

Results

We observed independent associations of variants in SPINK5 (p=0.003) and TSLP (p=0.006) with childhood asthma; a SPINK5 SNP was replicated. In subjects with one or more SPINK5 risk alleles, the absence of the TSLP protective minor alleles was associated with a significant increase in asthma (67% vs. 53%, p=0.0017). In contrast, the presence or absence of TSLP minor alleles did not affect asthma risk in subjects without the SPINK5 risk alleles. The SPINK5 and TSLP epistasis was replicated in a black population (p=0.036) that did not display independent association with variants in these genes.

Conclusions

Our results support epistasis between SPINK5 and TSLP which contributes to childhood asthma. These findings emphasize the importance of utilizing biology to inform analyses to identify genetic susceptibility to complex diseases. The results from our study have clinical relevance and support that the therapeutic effects of anti-TSLP therapy in asthmatics may be dependent on SPINK5 genotype.

Introduction

Asthma is the most common chronic illness in children in the US, affecting 7.1 million (9.5%) in 2011¹. Heritability estimates for asthma range between 0.36 and 0.77^2-5. Since 1989, >100 candidate genes have been described in >1000 publications on asthma or an associated disorder⁶. However, variants identified thus far confer small increments of risk, leading to questions about how the remaining genetic risk can be explained⁷. These studies have largely neglected genetic interactions including epistasis⁸ and the large sample sizes required for GWAS have often come at the expense of accurate, consistent phenotyping⁹.

Loss of function variation in the filaggrin (FLG) gene is the most replicated genetic risk factor for atopic dermatitis¹⁰, and the data is clear that FLG is associated with asthma in the presence of atopic features^11-17. Thymic stromal lymphopoietin (TSLP), an epithelial-cell derived cytokine that promotes differentiation and activation of T-helper 2 (Th2) cells and their receptors, is a well-validated asthma susceptibility gene¹⁸. Murine studies have suggested that TSLP may share the same genetic pathway as serine protease inhibitor Kazal-type 5 (SPINK5)^{19, 20}, which causes Netherton's Syndrome²¹ and has been associated with asthma, but studies are conflicting^22-25.

To assess the effects of skin-barrier related genes and asthma, we evaluated SNPs in FLG, SPINK5 and TSLP in 973 white and 530 black children. To maximize the impact of the genetic associations in this study, cases and controls were rigorously and objectively phenotyped with regard to asthma and allergies (an important distinction for control selection that we recently reported²⁶), our findings are replicated in six populations and we account for population substructure using ancestry informative markers. In addition, interactions of the candidate genes were evaluated given the mechanistic and biologic plausibility of epistasis.

Methods

Study Populations

The discovery population consisted of a subset of 4 to 17 year old Caucasian/white and African-American/black (the terms white and black will be used for simplicity) participants enrolled in either the Greater Cincinnati Pediatric Clinic Repository (GCPCR) or the Genomic Control Cohort (GCC), both described previously²⁷. The GCPCR includes over 6,500 patients and the GCC has 1,020 children and DNA was available on all participants as previously described^{28, 29}. Case-control definitions including those for asthma in the GCPCR have been previously described²⁶. All asthmatics were rigorously phenotyped by a specialty physician (pediatric allergist or pulmonologist) according to ATS criteria³⁰. Allergic controls (participants with allergic rhinitis, atopic dermatitis or environmental allergies) and non-allergic, non-asthmatic controls were available from both the GCPCR and the GCC. The protocols were approved by the CCHMC Institutional Review Board and participants gave written informed consent.

Among asthmatic children, asthma exacerbation was defined by previous hospitalizations for asthma. Results from skin prick testing (SPT) were available on 56% of asthmatic and allergic white children in the GCPCR. Children were defined as SPT positive if they had a positive test to a pollen (trees, weeds, grass), dust (dust mite, cockroach), animal (cat, dog) or mold at any time up to 6 months after their consent date.

Replication Cohorts

The replication populations were 1) 334 white trios (1002 individual samples) from the Childhood Asthma Management Program (CAMP)³¹; 2) 95 white trios (285 individual samples) from the Childhood Asthma Research and Education (CARE) Network³²; 3) 382 white children (57 asthmatics, 184 non-asthmatic SPT-controls and 141 non-asthmatic SPT+ controls) participating in the Cincinnati Childhood Allergy and Air Pollution Study (CCAAPS)³³; 4) 418 white individuals (207 GCPCR asthmatics enrolled after the discovery cohort and 211 non-asthmatic controls from the Cincinnati Control Cohort (CCC, described previously²⁶)); 5) 347 white children (207 asthmatics and 140 allergic controls) and 6) 340 black children (272 asthmatics and 68 allergic controls) from the GCPCR enrolled after the discovery cohort. CAMP and CARE data were downloaded with permission from the NIH-based database of Genotypes and Phenotypes (dbGaP) (http://www.ncbi.nlm.nih.gov/gap). Phenotypic description and details about the CAMP CARE data can be found at http://www.ncbi.nlm.nih.gov/gap/?term=asthma. The ‘replication GCPCR’ were white and black asthmatics and allergic controls that were enrolled in the repository after the discovery GCPCR cohort. The CCC is a population-based cohort of white adults with no personal or family history of asthma (by self-report) representative of Greater Cincinnati.

Gene and SNP Selection and Genotyping

The methods for the gene and SNP selection for the discovery array have been previously described^{26, 34}. FLG, SPINK5 and TSLP were chosen for these analyses based on their skin-related role and biologic relevance in the pathogenesis of asthma. To reduce the number of SNPs genotyped while generating the same amount of genetic information, tagging SNPs that maximized genomic coverage and captured the common genetic variation in these genes were selected using Haploview and Tagger (http://www.broad.mit.edu/mpg/haploview). Table 1 includes a description of the selected genes, their reported processes and functions as well as top associated disorders. A total of 40 SNPs, 1 truncation and 1 deletion in the FLG, SPINK5 and TSLP genes were considered for the discovery analyses.

Table 1.

Selected genes, functions and associated disorders.

#SNPsa	Gene	Full Gene Name	Chr.	Reported Processes and Functionc	Reported Associated Disordersd
8b	FLG	filaggrin	1q21.3	Keratinocyte differentiation, multicellular organismal development, cytoplasmic membrane bounded vesicle, intermediate filament, structural molecule activity	atopic dermatitis, eczema, ichthyosis vulgaris, asthma, rheumatoid arthritis, psoriasis, skin diseases, ichythyosis, atopy, inflammation
13	SPINK 5	Serine protease inhibitor Kazal-type 5	5q32	Peptidase inhibitor activity, serine-type endopeptidase inhibitor activity	Netherton syndrome, atopic dermatitis, asthma, skin diseases, congenital ichthyosis, atopy, congenital erythroderma ichthyosiform, ichthyosis, psoriasis, growth retardation
8	TSLP	Thymic stromal lymphopoietin	5q22.1	Extracellular region and space, cytokine activity, protein binding	inflammation, asthma, atopic dermatitis, necrosis, tumors, allergic asthma, skin lesions, mixed cryoglobulinemia, allergy, membranoproliferative glomerulonephritis

^aIndicates the total number of tagging SNPs that entered the analyses.

^bThis includes the 2282del4 deletion and R501X truncation.

^cObtained from Gene Ontology website (www.geneontology.org).

^dReported disease associations were obtained from the top 10 Novoseek disease relationships hits (number of articles in which both the gene's symbol or description and the disease appear) from GeneCards® (www.genecards.org).

Subjects from the GCPCR were selected for genotyping if they had 1) a pulmonary function test or completed asthma symptom questionnaire, 2) a SPT or completed Children's Health Survey for Asthma³⁵ and 3) ≥250ng DNA with an optical density of 1.6-2.2. Allergic and non-asthmatic, non-allergic controls were selected from the GCC that had available DNA. Genomic DNA was isolated from GCPCR buccal swabs with either the Zymo Research Genomic DNA II Kit (Zymo Research Corp., Orange, CA) or the Purgene DNA Purification System (Gentra Systems Minneapolis, MN), and from Oragene saliva samples per the kit's instructions. Genomic DNA was extracted from GCC blood samples using Manual PerfectPure DNA Blood Kit (Invitrogen, Carlsbad, CA). Genotyping of GCPCR and GCC samples was performed using a custom Illumina Golden Gate assay according to the manufacturer's protocol (http://www.illumina.com; San Diego, CA)^{26, 34}. Genotypes were assigned using BeadStudio's genotyping module (BeadStudio v3.2, San Diego, CA). Genotyping of FLG R501X truncation and 2282del4 deletion was performed by PCR and restriction fragment length polymorphism in both CCAAPS and the GCPCR as previously described^{36, 37}. Genotyping of the CCAAPS population for SPINK5 (29 SNPs) and TSLP (10 SNPs) was performed on banked saliva samples using a second custom Illumina Golden Gate assay. The publicly available data downloaded from dbGaP for both the CAMP and CARE replication cohorts were generated using the Affymetrix 6.0 SNP chip. Genotyping data from the Affymetrix 6.0 SNP chip was also available for the CCC. For the replications, all SNPs available in SPINK5 or TSLP for each population were included; only those SNPs that were on the discovery SNP array and/or those that had a p<0.05 are displayed.

Statistical Analyses

SNPs failing Hardy Weinberg Equilibrium in the non-allergic control group (p<0.0001, having minor allele frequencies below 10% in the combined controls; or missing call rates greater than 10%) were excluded (n=11). In addition, individuals with more than 20% of their total SNPs missing were excluded. Principal component (PC) analyses were performed using the 30 included ancestry-informative markers (AIMs) in EIGENSTRAT^{38, 39} to account for potential population stratification. Sets of AIMS ranging from 24-128 are useful tools for ascertaining the origin of subjects from particular continents, and to correct for population stratification in admixed population sample sets⁴⁰. After PCs were included in the model the genomic inflation factor was 1.0, suggesting minimal impact of population stratification for these AIMs, which is not unexpected since the general Cincinnati population does not show any North/South or East/West stratification based on larger panels of AIMS used for the larger GCC population (data not shown). Using PLINK⁴¹, associations with asthma were tested adjusting for age, sex and PCs (four PCs used in white and no PCs needed in black analyses) using the additive logistic regression model stratified by race. To address multiple testing, we determined the average pairwise LD (as measured by r²) for all SNP combinations and calculated the Bonferroni correction using Simple Interactive Statistical Analyses Software (http://www.quantitativeskills.com/sisa/).

For the GCPCR whites and blacks, associations with the primary phenotype of asthma were therefore considered significant at 0.0077 and 0.0039, respectively. For the CAMP and CARE cohorts, association analysis was performed using the transmission disequilibrium test (TDT)⁴². The case-control analyses for the GCPCR/CCC replication was accomplished by fitting a logistic regression model adjusted for sex. We used the strictest definition of genetic replication provided by Sullivan et al, which requires that the replication be the same SNP, phenotype and direction of association resulting in a mean false replication rate of 2%⁴³.

Permutation tests provide a computationally intensive approach to generating significance levels empirically. The purpose of the permutation test is to show that it is very unlikely that a permuted dataset will achieve the same significance and the observed effect is not due to chance. It also provides a framework for correction for multiple testing by keeping the patterns of LD between SNPs the same under the observed and permuted samples. We permuted the binary outcome (asthma vs. non-allergic control, or asthma vs. allergic control) 10,000 times and estimated the probability distribution of the statistic under the null hypothesis. The empirical p value was defined as the probability of observing permuted statistics larger than the observed statistics.

After the primary analyses, secondary analyses were performed to ensure that the observed genetic associations with asthma were not due to other co-morbid allergic conditions such as atopic dermatitis (AD) and sensitization. The genetic associations with asthma excluding all cases with known co-morbid AD were performed by fitting a logistic regression model adjusted for age, sex and PCs. Genetic associations with SPT were evaluated for any SPT+ as well as positivity to molds, pollens, animals or dust. Subjects that were SPT+ were compared to non-asthmatic, non-allergic controls as well as SPT+/SPT- restricted to asthmatics. The SPT analyses was carried out in PLINK⁴¹ adjusted for age, sex and PCs using the additive logistic regression model in whites only due to sample size limitations for the blacks.

To evaluate epistasis between SPINK5 and TSLP, we conducted analyses using a gene collapsing technique since the significant SNPs had low minor allele frequencies (∼10%). SNPs included in the analysis were identified by the magnitude of effect (OR) and significance (p-value). Four SNPs in SPINK5 had an OR>1.3 and p<0.01 (rs2303064, rs7445392, rs2303063 or rs9325071) and 3 SNPs in TSLP had an OR<0.70 and p<0.01 (rs10062929, rs11466749, and rs11466750). The children were then identified as having any (SPINK5+) or none (SPINK5-) of the minor alleles in the four risk SNPs from SPINK5, as well as having any (TSLP+) or none (TSLP-) of the minor alleles in the three protective SNPs in TSLP. For FLG, children having one or more copies of the R501X truncation or 2282del4 deletion were defined as FLG+ and those with neither of these two polymoprphisms were defined as FLG-.

Results

Demographics of Discovery Subjects

A total of 1503 children from the GCPCR and GCC were eligible for genotyping on the custom SNP chip after inclusion criteria were applied: age (4-17), race (Caucasian or African-American), diagnoses (asthma, AR or AD), DNA quantity (≥250ng), DNA quality (optical density 1.6-2.2) and available pulmonary function test, SPT or questionnaire data (asthmatics only). After excluding 37 genotyped children due to missing SNP call rates >20%, the discovery GCPCR/GCC population consisted of 1,466 asthmatic (402 and 308), allergic (257 and 152) and non-allergic (297 and 50) white and black children respectively (Supplementary Table 1). The non-allergic control children were older than the asthmatics of similar race (p<0.05), and there were more male black asthmatics than black non-allergic controls (p<0.05).

Genetic Associations with Asthma in the Discovery GCPCR Population

The associations of SNPs in FLG, SPINK5 and TSLP were evaluated independently for whites and blacks, adjusted for age, sex and population stratification. After consideration for multiple comparisons, a non-synonymous SNP in SPINK5, rs2303064, was significantly associated with asthma in white subjects compared to non-allergic controls (p=0.003) and a tagging SNP in TSLP (rs11466750) was significantly associated with asthma in whites compared to allergic children (p=0.0056, Table 2). Multiple SNPs in both SPINK5 and TSLP were nominally associated with asthma (p<0.05). No SNPs or deletion variants in FLG achieved significance.

Table 2.

SNP associations between asthmatics and allergic and non-allergic controls in the GCPCR.

			Asthma versus Non-allergic Controls						Asthma versus Allergic Controls
	Population		whitea			blacka			whitea			blacka
	# cases/ # controls		402/297			308/50			402/257			308/152
Gene	SNP	Major/ Minor Allele	MAFb	OR	P-value	MAFb	OR	P-value	MAFb	OR	P-value	MAFb	OR	P-value
FLG	rs2184951	T/G	14%/14%	1.00	0.99	-	-	-	14%/15%	0.93	0.65	-	-	-
	rs11204976	C/T	15%/14%	0.99	0.96	-	-	-	15%/14%	1.04	0.82	-	-	-
	rs12730241	G/A	15%/14%	1.03	0.85	-	-	-	15%/14%	1.06	0.72	-	-	-
	rs2065956	C/T	13%/12%	1.06	0.74	-	-	-	13%/11%	1.09	0.62	-	-	-
	rs2184953	A/Gc	16%/14%	1.12	0.47	34%/31%	1.14	0.61	16%/14%	1.10	0.56	34%/35%	0.95	0.74
	rs11204980	G/C	15%/15%	0.95	0.76	14%/16%	0.84	0.57	15%/14%	1.05	0.78	14%/14%	0.93	0.73
	R501X	na	na	2.01	0.18d	-	-	-	na	0.75	0.57e	-	-	-
	2282del4	na	na	1.20	0.57d	-	-	-	na	1.01	0.99e	-	-	-
SPINK5	rs3822744	C/T	36%/34%	1.09	0.45	9%/15%	0.62	0.12	36%/35%	1.06	0.64	9%/11%	0.87	0.56
	rs1423007	A/G	37%/34%	1.13	0.32	25%/26%	0.94	0.80	37%/35%	1.10	0.42	25%/26%	0.97	0.86
	rs7445392	T/Cc	51%/44%	1.33	0.01	33%/29%	1.21	0.45	51%/47%	1.17	0.18	33%/29%	1.20	0.25
	rs6892205	G/Ac	49%/43%	1.25	0.05	47%/37%	1.59	0.05	49%/46%	1.08	0.49	47%/45%	1.13	0.41
	rs2303063	A/Gc	49%/43%	1.32	0.01	25%/22%	1.14	0.63	49%/46%	1.12	0.33	25%/23%	1.13	0.51
	rs2303064	G/A	14%/9%	1.70	0.003f	48%/48%	1.03	0.90	14%/12%	1.21	0.26	48%/48%	0.98	0.91
	rs2287768	A/G	20%/19%	1.03	0.81	-	-	-	20%/22%	0.90	0.44	-	-	-
	rs9325071	A/G	15%/11%	1.58	0.01	32%/41%	0.65	0.07	15%/12%	1.17	0.35	32%/32%	0.95	0.76
	rs3777134	T/Cc	38%/35%	1.20	0.13	36%/37%	0.92	0.70	38%/36%	1.11	0.40	36%/35%	1.01	0.96
	rs2112767	T/C	39%/35%	1.23	0.09	44%/43%	1.03	0.88	39%/36%	1.12	0.36	44%/44%	1.04	0.79
	rs4263489	C/G	11%/14%	0.80	0.19	-	-	-	11%/12%	0.89	0.51	-	-	-
	rs3815741	A/G	-	-	-	26%/27%	0.89	0.67	-	-	-	26%/27%	0.93	0.69
	rs28408445	C/G	-	-	-	14%/18%	0.76	0.33	-	-	-	14%/12%	1.11	0.63
TSLP	rs3806933	C/T	40%/41%	0.93	0.53	25%/26%	0.96	0.87	40%/42%	0.92	0.46	25%/29%	0.83	0.26
	rs2289276	C/T	28%/28%	0.93	0.57	15%/14%	1.06	0.86	28%/25%	1.13	0.34	15%/17%	0.87	0.45
	rs1898671	C/T	37%/34%	1.10	0.40	-	-	-	37%/32%	1.19	0.15	-	-	-
	rs10062929	C/A	12%/14%	0.93	0.68	10%/12%	0.86	0.67	12%/18%	0.64	0.009	10%/12%	0.85	0.47
	rs2289277	C/G	39%/42%	0.90	0.33	39%/40%	1.01	0.98	39%/43%	0.87	0.24	39%/44%	0.83	0.20
	rs11466749	A/G	15%/16%	0.95	0.73	12%/14%	0.83	0.59	15%/20%	0.67	0.009	12%/14%	0.86	0.50
	rs11466750	G/A	12%/14%	0.92	0.60	26%/27%	0.94	0.81	12%/18%	0.63	0.0056f	26%/25%	1.06	0.73
	rs2289278	C/G	-	-	-	11%/8%	1.60	0.24	-	-	-	11%/12%	1.03	0.89

^aWhite and black asthmatic children were from the GCPCR and non-asthmatic/non-allergic controls and allergic controls were from the GCPCR and GCC. Associations were tested using an additive model. Odds ratios (OR) were determined using logistic regression based on the minor allele adjusted for age, gender and population stratification. Bold indicates association reached linkage disequilibrium-corrected bonferroni cut-off of 0.0077 for whites and 0.0039 for blacks.

^bMAF (minor allele frequency) in cases/controls.

^cMajor/Minor allele are reverse for blacks.

^dSample size was 114/221 for R501X and 401/256 for 2282del4.

^eSample size was 114/72 for R501X and 401/216 for 2282del4.

^fThese associations remained significant after asthmatics with known co-morbid eczema were excluded (p=0.003 and 0.004), indicating an asthma specific association.

To further substantiate our results and ensure that the family wise error rate was appropriately controlled, we performed 10,000 permutations. The point-wise empirical p-value was consistent with the observed p-value (0.0026 for rs2303064 and 0.0046 for rs11466750; results not shown) supporting its validity. The family-wise corrected p-values were 0.039 and 0.069 for rs2303064 and rs11466750, respectively.

In order to ensure that the genetic associations with asthma were not attributed to other co-morbid allergic conditions such as atopic dermatitis (AD), we performed secondary analyses excluding asthmatic cases with AD and evaluated SPT+ as an outcome. These analyses were performed in whites only due to sample size limitations for the blacks. The associations with asthma remained significant after asthmatics with known co-morbid AD were excluded (p=0.003 for rs2303064 and 0.004 for rs11466750; data not shown), indicating an asthma specific association. We did not observe any significant associations with SPT+ (data not shown).

Genetic Associations with Asthma in Replication Cohorts

We evaluated SNP associations in the SPINK5 and TSLP genes with asthma in six additional populations: CAMP, CARE, CCAAPS, replication GCPCR/CCC, and replication GCPCR (whites and blacks). Based on our stringent replication definition (same SNP, phenotype and direction of association), we achieved replication of the SPINK5 rs2303064 variant in the CARE population (p=0.046; Table 3). In addition, we observed additional SNPs in SPINK5 and TSLP to be nominally (p<0.05) associated with asthma in five of the six populations (Table 3).

Table 3.

SNP associations with asthma in replication cohorts.

	Population		CAMPa		CAREa		CCAAPSb		replication GCPCR/CCCd		replication GCPCRe		replication GCPCRe
	# cases/ # controls		334 trios		95 trios		57/184c		206/211		206/132		263/67
	Race=		white		white		white		white		white		black
Gene	SNP	Major/ Minor Allele	OR	P-value	OR	P-value	OR	P-value	OR	P-value	OR	P-value	OR	P-value
SPINK5	rs2303064f	G/A	0.82	0.26	2.00	0.046	1.34	0.35	1.22	0.36	-	-	0.93	0.71
	rs7725861	A/T	0.68	0.03	1.53	0.19	-	-	-	-	-	-	-	-
	rs9791022	T/C	0.45	0.004	1.55	0.26	-	-	-	-	-	-	-	-
	rs3777138	A/G	1.00	1.00	7.00	0.03	-	-	-	-	-	-	-	-
	rs3777142	A/T	-	-	-	-	2.14	0.04	-	-	-	-	-	-
	rs17704764	A/G	-	-	-	-	1.75	0.04	-	-	-	-	-	-
	rs6864920	A/G	-	-	-	-	0.62	0.04	-	-	-	-	-	-
	rs2287772	C/T	-	-	-	-	0.63	0.049	-	-	-	-	-	-
	rs9325073	G/C	-	-	-	-	-	-	1.53	0.003	-	-	-	-
	rs7445392f	T/C	-	-	-	-	-	-	-	-	-	-	0.98	0.92
	rs2303063f	G/A	-	-	-	-	-	-	-	-	-	-	1.07	0.76
	rs6892205f	A/G	-	-	-	-	-	-	-	-	-	-	1.23	0.27
TSLP	rs11466750f	G/A	-	-	-	-	1.18	0.65	-	-	0.98	0.93	0.85	0.47
	rs11466749f	T/C	0.92	0.59	0.48	0.02	1.47	0.24	-	-	0.98	0.94	0.82	0.55
	rs11466744	G/T	0.00	0.03	-	-	-	-	-	-	-	-	-	-
	rs2289276f	C/T	-	-	-	-	1.08	0.77	-	-	1.36	0.10	0.64	0.08
	rs3806933f	C/T	-	-	-	-	1.13	0.57	-	-	1.24	0.21	0.64	0.04

^aThe Childhood Asthma Management Program (CAMP) and Childhood Research and Education (CARE) are clinical trials included in the SNP Health Association Resource (SHARe) Asthma Resource project (SHARP) and were used as replication cohorts in this analysis. Whole genome genotyping data from the Affymetrix 6.0 chip was available on 334 CAMP and 95 CARE family trios of asthmatic children. Transmission disequilibrium tests were used for SNP associations.

^bThe Cincinnati Childhood Allergy and Air Pollution (CCAAPS) birth cohort included 57 white asthmatics and 184 white non-asthmatic, SPT negative control children.

^cFor TSLP analysis, controls were 141 non-asthmatic SPT positive children. Asthma was diagnosed at age 7 using pulmonary function and methacholine challenge testing.

^dNew GCPCR cases were asthmatic individuals that were recruited into the GCPCR after the discovery cohort and genotyped on a separate custom Illumina chip. The CCC consisted of white adults with no history of asthma. A logistic model was fitted adjusted for sex.

^eNew GCPCR controls were recruited after the discovery cohort and were non-asthmatic children with allergic rhinitis (both with and without atopic dermatitis). Shading indicates replication defined by same SNP, phenotype and direction of effect. Bolding indicates gene-level replication at p<0.05.

^fIndicates SNP was included on discovery array.

Epistasis between SPINK5 and TSLP

The literature suggests that SPINK5 and TSLP occupy a shared pathway where LEKTI deficiency ultimately leads to TSLP production, increasing the Th2 response and causing allergic lung inflammation^{21, 44-48}. Therefore, we sought to determine if there were epistatic effects of variation in the SPINK5 and TSLP genes on asthma prevalence. Formal gene:gene analyses did not show an interaction between SPINK5 and TSLP (p=0.29). Further analyses were performed stratified by either SPINK5 or TSLP genotype by conducting gene collapsing analyses. We evaluated the effects of having one or more of the 3 protective TSLP alleles (rs10062929, rs11466749, and rs11466750) among subjects with and without one or more of the risk SPINK5 alleles (rs2303064, rs7445392, rs2303063 or rs9325071). There was modest linkage disequilibrium among the SPINK5 SNPs as well as the TSLP SNPs (Figure 1). In subjects with at least one of the SPINK5 risk alleles (SPINK5+), the absence of the TSLP protective minor alleles (TSLP-) was associated with a significant increase in asthma risk (67% vs. 53%, p=0.0017, Figure 2A). In contrast, the presence (TSLP+) or absence of TSLP minor alleles did not affect asthma risk in subjects without any of the SPINK5 risk alleles (SPINK5-, p=0.71).

Figure 1.

Linkage disequilibrium (r²) among SNPs included in the gene collapsing analyses for 407 asthmatics compared to 257 allergic controls in the white discovery GCPCR populations.

Figure 2.

Epistasis of SPINK5 and TSLP. (A) Discovery white GCPCR/GCC children. (B) Replication of epistasis of SPINK5 and TSLP in replication GCPCR black children.

Replication of Epistasis between SPINK5 and TSLP

Using biologic plausibility of a shared pathway from the published literature and the results from above, we pursued replication of the observed epistasis of SPINK5 and TSLP in the discovery GCPCR black, replication GCPCR white and replication GCPCR black populations. Notably, there were no independent genetic associations at our LD-adjusted Bonferroni level with any SNPs in SPINK5 or TSLP in these three populations. Therefore, we included SNPs with p<0.10 and OR >1.2 for SPINK5 (rs6892205 and rs2303063) and p<0.10 and OR <0.7 for TSLP (rs3806933 and rs2289276) in the analyses (Table 3). We replicated the epistasis between SPINK5 and TSLP in the replication GCPCR black population (Figure 2B); the other two populations did not replicate. As in the discovery GCPCR white population, SPINK5+ black subjects that were SPINK5+ TSLP+ had a significantly decreased asthma risk compared to subjects that were SPINK5+ TSLP- (73% versus 85%, p=0.036; Figure 2B), whereas there was no effect of TSLP minor alleles among subjects that were SPINK5- (p=0.80).

Discussion

While the role of skin-related genes in asthma has been previously reported, our data support that asthma risk is, in part, a result of epistasis between skin-related genes (TSLP and SPINK5). Using a well characterized cohort of white asthmatic cases and controls, we identified SNPs in these genes to be significantly associated with asthma in children. The finding with SPINK5 was replicated according to the strictest replication definition⁴³. Additionally, we discovered epistasis between TSLP and SPINK5 in whites. This was replicated using a black cohort where neither gene was independently associated. These findings highlight the importance of evaluating epistasis in genetic association studies. Further, our results emphasize the need to utilize biology to inform analyses in order to identify genetic susceptibility to complex diseases, substantiate the value of smaller well-characterized cohorts in genetic association studies, and demonstrate the importance and value of looking at functionally related genes, rather than one at a time as in GWAS, to identify genes with clinical relevance.

This study further implicates skin-related genes in the pathogenesis of childhood asthma. Interestingly, the associations of the skin genes with asthma were sustained when children with co-morbid AD were excluded from the analysis, suggesting an independent asthma effect. Associations between asthma and SPINK5 have been reported in the Chinese²⁴ and German²³ populations. Previous studies have also found that TSLP SNPs are associated with asthma⁴⁹, including a meta-analysis of GWASs in subjects of multiple ancestries⁵⁰. This is the first report of an epistatic effect of SNPs in the SPINK5 and TSLP genes with asthma. We discovered epistasis between TSLP and SPINK5 whereby children that had one or more SPINK5 risk alleles and were wild-type for all TSLP SNPs had the highest risk of asthma. The addition of TSLP minor alleles conferred protection from asthma in the presence of a SPINK5 risk allele, and this protection was stronger than the association between TSLP and asthma alone. Importantly, our primary analyses did not detect independent significant associations with SPINK5 and TSLP in the black population. Only after stratified analyses were we able to uncover the epistatic effect of SPINK5 and TSLP in the black population. Thus, ignoring epistatic effects may mask important genetic susceptibility. While other studies have utilized biologic plausibility^{51, 52}, results from linkage analyses^{53, 54} and bioinformatics^{54, 55} to inform analyses of genetic interactions in asthma, our study is the first to apply this approach in an independent population and replicate findings where no significant independent effects were observed (blacks).

Observations from animal models provide valuable insights into the possible mechanistic basis for the observed epistasis between SPINK5 and TSLP. Mice that lack Spink5 overexpress TSLP in the skin in the absence of lympho-epithelial kazal-type-related inhibitor (LEKTI)²⁰. Overexpression in the skin leads to high systemic availability of TSLP driving susceptibility to allergic inflammation in the lung, even in the absence of any skin pathology¹⁹. TSLP transgenic mice develop severe asthma in the absence of a skin phenotype and express low levels of the TSLP transgene in the trachea but not in the lung¹⁹. Further, the protein is absent in the bronchoalveolar fluid suggesting systemic skin-derived TSLP is sufficient to predispose mice to allergic lung inflammation¹⁹. Models using lung-specific expression of the TSLP transgene or intranasal administration of TSLP show that increased lung TSLP drives an inflammatory response in the presence of antigen, suggesting that TSLP modifies the response to aero-antigen, promoting inflammation^{20, 56}. Murine models of asthma treated with anti-TSLP antibody reversed airway inflammation, prevented structural alterations and decreased airway hyperresponsiveness⁵⁷, while suppression of signaling with anti-TSLP receptor reduces eosinophilic airway inflammation, goblet cell hyperplasia and TH2 cytokine production⁵⁸.

These animal data, along with our results, support a shared pathway for SPINK5 and TSLP (and possibly FLG) leading to airway inflammation and asthma development (Figure 3). SPINK5 overexpression in human epithelial cells leads to increased inflammatory cytokines including IL6, IL8 and RANTES, indicating it may have a role in the pathogenesis of asthma that is independent of protease inhibition⁵⁹ (Figure 3A). This suggests that SPINK5 stimulates the involvement of B and T lymphocytes, neutrophils and eosinophils active in the inflammatory process through its effects on pro-inflammatory cytokines and chemokines. In humans, LEKTI deficiency results in a lack of kallikrein-related peptidase 5 (KLK5) and KLK7 inhibition^{21, 60} and unopposed KLK14 activity²¹, activating through protease-activated receptor-2 (PAR-2) which is found at the surface of keratinocytes. PAR-2 then activates nuclear factor kappa B (NF-κB) leading to the production of the TSLP cytokine^{21, 60} causing the differentiation of Th0 cells to Th2 which produce IL-4, IL-13 and IL-5, amplifying the Th2 response and promoting inflammation, allergy and asthma (Figure 3B). LEKTI deficiency also promotes increased profilaggrin cleavage⁴⁴ leading to a compromised barrier⁴⁵ promoting sensitization and inflammation (Figure 3C).

Figure 3.

Hypothesized shared pathways leading to asthma and inflammation. (A) SPINK5 overexpression increases inflammatory cytokines. (B) LEKTI deficiency increases KLK5/7/14, activating through PAR-2, and NF-κB leading to TSLP production, causing allergy and lung inflammation. (C) LEKTI deficiency increases profillagrin cleavage compromising the skin barrier and contributing to sensitization and subsequent asthma. Solid lines represent published pathways, dashed lines represent hypothesized pathways.

While we were successful in identifying both main and epistatic effects with TSLP and SPINK5, no associations with FLG were identified. This is not surprising given the low frequency of the functional FLG variants in our population resulting in inadequate power to detect an association. However, our data does support that common variation in FLG, represented by our 6 tagging SNPs, is not associated with asthma in these populations. A meta-analysis of asthma regardless of eczema revealed a significant association with the combined R501X and 2282del4 genotype (pooled OR 1.48, 95%CI 1.32-1.66)⁶¹, indicating that FLG deficiency may predispose to asthma. However, the asthma association was driven by the asthma plus eczema phenotype, supporting that FLG is only associated with asthma in the presence of atopic features.

The identification of epistasis between SPINK5 and TSLP is has immediate clinical implications; there are clinical trials underway testing AMG157, a human monoclonal antibody that inhibits the action of TSLP. Phase 1 escalation studies to evaluate the safety, tolerability and pharmacokinetics in both healthy subjects and subjects with atopic dermatitis have been completed^{62, 63} and another study evaluating the effects of AMG157 on FEV1 after allergen inhalation challenge in mild atopic asthmatics is currently recruiting⁶⁴. The results from our study suggest that the therapeutic effects of anti-TSLP therapy in asthmatics may be dependent on SPINK5 genotype.

In conclusion, our results support a novel epistasis between SPINK5 and TSLP, which contributes to childhood asthma. These findings emphasize the importance of utilizing biology to inform analyses to identify genetic susceptibility to complex diseases, especially in populations that do not exhibit independent genetic associations. The results from our study are clinically relevant with respect to anti-TSLP therapy and suggest that therapeutic effects in asthmatics may be dependent on SPINK5 genotype.

Key Messages.

• Our study supports epistasis between SPINK5 and TSLP which contributes to childhood asthma.

• Genetic association studies should use biologic plausibility to inform analyses to identify genetic susceptibility to complex disease.

• The therapeutic effects of anti-TSLP therapy in asthmatics may be dependent on SPINK5 genotype.

Abbreviations

FLG

filaggrin

TSLP

thymic stromal lymphopoietin

Th2

T-helper 2

SPINK5

serine protease inhibitor Kazal-type 5

GCPCR

Greater Cincinnati Pediatric Clinic Repository

GCC

Genomic Control Cohort

SPT

skin prick testing

CAMP

Childhood Asthma Management Program

CARE

Childhood Asthma Research and Education

CCAAPS

Cincinnati Childhood Allergy and Air Pollution Study

CCC

Cincinnati Control Cohort

dbGaP

database of Genotypes and Phenotypes

AIMs

ancestry-informative markers

PCs

principal components

TDT

transmission disequilibrium test

OR

odds ratio

LEKTI

lympho-epithelial kazal-type-related inhibitor

KLK5

kallikrein-related peptidase 5

KLK7

kallikrein-related peptidase 7

KLK14

kallikrein-related peptidase 14

PAR-2

protease-activated receptor-2

NF-κB

nuclear factor kappa B