Early life environmental exposures interact with genetic susceptibility variants in pediatric eosinophilic esophagitis

Abstract

Background

While EoE is associated with certain gene variants, the rapidly increasing incidence of eosinophilic esophagitis (EoE) suggests that environmental factors contribute to disease development.

Objective

We tested for gene-environment interaction between EoE-predisposing polymorphisms (within TSLP, LOC283710/KLF13, CAPN14, CCL26, TGF-β) and implicated early life factors (antibiotic use in infancy, cesarean delivery, breastfeeding, neonatal intensive care unit (NICU) admission, and absence of pets in the home).

Methods

We conducted a case-control study using hospital-based cases (n=127) and controls representative of the hospital catchment area (n=121). We computed case-only interaction tests and, in secondary analyses, evaluated the combined and independent effects of genotype and environmental factors on risk of EoE.

Results

Case-only analyses identified interactions between rs6736278 (CAPN14) and breastfeeding (p=0.02) and rs17815905 (LOC283710/KLF13) and NICU admission (p=0.02), but not with any of the factors examined. Case-control analyses suggested that disease risk may be modifiable in individuals with certain gene variants. In particular, breastfeeding, in those with the susceptibility gene variant at rs6736278 (CAPN14), reduced the risk of developing EoE (aOR: 0.08; 95% CI: 0.01, 0.59). Admission to the NICU, in those without the susceptibility gene variant at rs17815905 (LOC283710/KLF13), significantly increased risk of developing disease (aOR 4.83; 95% CI: 1.49, 15.66).

Conclusions

The interplay of gene (CAPN14 and LOC283710/KLF13) and early life environment factors (breastfeeding, NICU admission) may contribute to EoE susceptibility.

Introduction

Over the past two decades, eosinophilic esophagitis (EoE) has transformed from a case-reportable disease to a substantial cause of upper GI morbidity.¹ The reasons for this increase are likely multifactorial including increased disease recognition and changing environmental factors, particularly those interacting with immunological pathways involved in EoE.^{2, 3} Numerous genetic susceptibility variants have been shown to contribute to EoE, including variants at 5q22 (TSLP/WDR36), 2p23 (CAPN14), and 11q13 (LRCC32), but the magnitude of association for disease susceptibility is modest (<2-fold), similar to the magnitude seen in other allergic and immunological diseases. Twin studies have established a role for genetics in disease susceptibility, but an even stronger role for environmental factors, particularly those encountered in early life, as shown by the marked increased concordance for EoE between dizygotic twins compared with non-twin siblings and environmental assessments in EoE case control studies.^{4, 5}

Identification of gene-environment interactions for EoE offers an opportunity to improve understanding of the mechanisms for disease development. It also may offer a means for identifying possible modifiable risk factors for disease, specifically risk factors that may only be relevant in the presence of specific genetic susceptibility variants. Therefore, we conducted a preliminary, case-control study of gene-environment interaction between EoE genetic variants^6–9 and early life exposures that have been associated with EoE and other atopic diseases (Supplementary Table 1).^{5, 10–12} Specifically, we examined interaction between 5 gene variants (rs6736278 within CAPN14 at 2p23, rs2302009 within CCL26 at 7q11, rs3806932 within TSLP at 5q22, rs17815905 within the LOC283710 and KLF13 region at 15q13, and rs1800469 within TGF-β at 19q13) and 6 maternally-reported early life exposures (cesarean delivery, preterm delivery, NICU admission, breastfeeding, antibiotics in infancy, and absence of a furry pet in infancy). We selected single nucleotide polymorphisms (SNPs) reported to be associated with EoE, or in high linkage disequilibrium with SNPs associated with EoE. These included SNPs from genes with demonstrated functionality in gene regulation and association with allergic sensitization. Specifically, these included SNPs from genes with demonstrated functionality in gene regulation and association with allergic sensitization (TSLP at 5q22 [rs3806932], the LOC283710 and LOC283710 region at 15q13 [rs4329885], CAPN14 [rs6736278], CCL26 [rs2302009], and TGF-β [rs1800469])).^6–9 CAPN14 encodes an IL-13-induced calcium-activated cysteine protease involved in maintaining integrity of esophageal barrier function and is overexpressed in EoE.^{13, 14} The rs4329885 SNP, found on the 15q13 chromosome band between LOC283710 and KLF13, is in a region that is influential for gene regulation, may contribute to impaired esophageal epithelial cell development that has been observed to occur in EoE,¹⁵ and has been shown to be involved in Th2 cytokine production.¹⁶ TSLP, a gene that is critical to inducing allergen sensitization, is strongly associated with and overexpressed in EoE.^{9, 17} CCL26 encodes for eotaxin-3, the most upregulated gene in EoE, and acts as a chemoattractant for the activation and recruitment of eosinophils into esophageal tissue.⁸ Transforming growth factor (TGF) - β secretion, produced in part by eosinophils, is implicated in the fibrosis of esophageal tissue observed in EoE, as well as in immunomodulation.¹⁸ While the exact mechanism for a possible interaction is unknown, it may be that environmental factors have epigenetic effects that influence DNA expression.^{19, 20}

Methods

Cases were recruited from the Cincinnati Center for Eosinophilic Disorders (CCED) at Cincinnati Children’s Hospital Medical Center (CCHMC) and met criteria for EoE diagnosis as defined by current consensus guidelines.²¹ Controls, used to assess violation of the assumption of independence between exposure and the genotype in the non-cases, were recruited from the CCHMC Genomic Control Cohort, a population-based control cohort representative of the Greater Cincinnati population.²² Cases and controls were restricted to those less than age 18 at the time of enrollment and all were self-reported Caucasian race.

Early life, environmental factors assessment

Early life environmental factors for cases and controls were assessed through an on-line self-administered questionnaire provided to mothers. The questionnaire, described previously in detail,¹⁰ was developed to study early life exposures, through an interactive process that included cognitive interviewing and testing. Survey domains included prenatal, intrapartum and pediatric exposures, and the exposures of interest for this study were limited to those previously associated with EoE or demonstrating main effect associations in the current study population,⁵ including caesarean delivery, preterm birth, admission to the neonatal intensive care unit (NICU), formula feeding and antibiotic use (Supplementary Table 1).^{5, 10–12, 23}

SNP genotyping and quality control

DNA from cases and controls was obtained from peripheral blood samples. Cases and controls were genotyped using the genome-wide OMNI5 BeadChip array (Illumina, Inc.).²² Examination of call rates identified <0.1% missing genotypes. We identified no evidence of departure from Hardy Weinberg equilibrium in cases, controls or cases and controls combined (Supplementary Table 2). Prior to analyses, we performed principal components analysis using EIGENSOFT²⁴ on 639 ancestry informative markers from the OMNI5 Beadchip (Illumina, Inc.).²²

Statistical analyses

In our primary analysis we conducted a case-only gene-environment interaction analysis. We used a generalized linear model (logit link, binomial distribution) to assess the association between each environmental factor and genotype in cases. We modeled the risk allele using a recessive mode of inheritance if the risk allele was the major allele (e.g. GG or GA vs GG) and used a dominant mode of inheritance if the risk allele was the minor allele (e.g. AA vs AG or GG). If there was marginal evidence of an interactive effect (p<0.10) in the case-only analysis, we tested the assumption of independence between the exposure and genotype in the source population by modelling the association between exposure and genotype in the controls. Lack of independence was identified by evidence of a similar direction and magnitude of association between exposure and genotype in cases and controls.

In secondary analyses, to facilitate interpretation of any interactive effect, we used cases and controls to estimate the odds of disease in those with the susceptibility genotype only, those with the environmental factor only, and those with both the susceptibility genotype and the environmental factor, relative to those with neither the susceptibility genotype nor the environmental factor. Additionally, we modeled disease risk within those with the susceptibility genotype, as a means for assessing the contribution of each early life factor in those with the susceptibility genotype. For all models, we estimated both crude and adjusted (adjusted for maternal education and two principal components for ancestry) associations. The presence of interaction was determined by indication of a case-only interaction (p<0.05) and indication of stratum-specific differences in the magnitude of association in the case-control assessment. Finally, we calculated the relative excess risk due to interaction (RERI) to assess for interaction on the additive scale.^{25, 26} We generated both crude and adjusted estimates for both primary and secondary analyses. Adjusted models included adjustment for maternal education, as a proxy for socioeconomic status, and population stratification. The study was approved by the Cincinnati Children’s Hospital Medical Center IRB.

Results

Of the 237 patients successfully contacted, 169 (71%) consented to participation. Of these, 136 (80%) completed the questionnaire. Similarly, for controls, of the 208 potential subjects successfully contacted, 147 (71%) consented to participation and, of these, 125 (85%) completed the questionnaire. Our final analytic sample, after exclusion of observations with missing data, included 127 cases and 121 controls. The demographic features of the cases and controls were similar, with the exception of atopic disease comorbidities (e.g. 83.3% cases versus 12.4% controls reporting food allergies) and sex (80.3% males in cases versus 52.1% in controls) (Table 1).

Table 1.

Characteristics of study population

Characteristic	Controls n=121 % or mean; std	Cases n=127 % or mean
Age at enrollment (mean yrs.; std)	13.8; 2.8	10.6; 3.8
Sex (% male)	52.1	80.3
Atopic illnesses (%)
Food allergies	12.4	83.3
Environmental allergies	42.2	69.3
Antibiotic allergies	14.5	25.8
Eczema	21.5	59.1
Asthma	18.2	44.9
Maternal marital status (%)
Married or civil union	93.4	94.4
Single	6.6	5.6
Maternal education (%)
<HS*	0	2.4
HS or GED**	17.4	11.8
Technical or associates	21.5	18.9
Bachelor’s	40.5	37.0
Graduate or professional	20.7	29.9

^*HS=High School

^**GED=General Education Development test certification

In our primary case-only analyses, we observed interaction between rs6736278 within CAPN14 and breastfeeding (p=0.02) (Figure 1). We also observed interaction between rs17815905 in the LOC283710 and KLF13 region and NICU admission (p=0.02). There was no evidence of statistically significant interaction between the SNPs in CCL26, TSLP or TGF-β with any of the early life factors examined (Supplementary Table 3). Where there was evidence of interaction, or marginal evidence of interaction (TSLP and NICU admission and TSLP and no furred pets in infancy), we found no evidence of violation of independence (Supplementary Table 4). Secondary analyses confirmed the case-only interactions observed (Figure 2, Tables 2,3) and stratum-specific assessments identified breastfeeding as a potentially modifiable factor in developing EoE among children with a certain genotype (Table 2). While the exact mechanism of how breastfeeding and CAPN14 may interact and become protective for disease is unknown, the protective effect of breastfeeding was demonstrated in a study of interaction between breastfeeding and the cluster of differentiation 14 (CD14) genotype (CD14C-159T) in relation to atopic sensitization.²⁷ CAPN14 is integral to maintaining barrier function in the esophagus.^{13, 14} Potentially, with impaired barrier function, breastfeeding may confer a higher protective effect than that which is observed when barrier function is preserved, however this must be investigated in follow-up mechanistic studies. Notably, we observed a 92% reduction in the odds of developing EoE among those who were breastfed within the subgroup of participants harbouring the susceptibility genotype (Table 2). Examination of interaction on the additive scale, RERI, identified a strong departure from additivity for CAPN14 and breastfeeding (RERI: −6.27 [95% CI: −17.27, 4.72])) (Table 2) and the LOC283710 and KLF13 region and NICU admission (RERI: −4.53 [95% CI: −10.81, 1.76]), however neither of these reached statistical significance (Table 3).

Figure 1.

Case-only assessment of gene-environment interaction in EoE

Illustration of the association between exposure and genotype in cases only to assess for gene-environment interaction, adjusted for maternal education.

*Suggestive associations examined in case-control data assessment of interaction

**Data too sparse for estimation of association

Figure 2.

Case-control analysis interaction plots for early life factors and susceptibility genotype

Plots denote interaction between selected exposures and genotype variants in the case-control analyses. Interaction models include adjustment for maternal education. G+=exposed to susceptibility genotype variant; G−=unexposed to susceptibility genotype variant. Figure A illustrates the difference in risk for those breastfed and not breastfed, among those with the susceptibility genotype variant. Figure B illustrates the increase in risk for NICU admission and those without the susceptibility variant.

Table 2.

CAPN14 (rs6736278) and breastfeeding

	Cases n	Controls n	OR (95% CI)	aOR (95% CI)
Interaction
G−E−	26	11	referent	referent
G+E−	2	5	5.91 (0.99, 35.21)	5.38 (0.89, 32.71)
G−E+	75	60	1.89 (0.86, 4.14)	1.70 (0.76, 3.77)
G+E+	18	4	0.53 (0.14, 1.91)	0.43 (0.12, 1.62)
Case only p for interaction			0.01	0.02
RERI (95% CI)				−6.27 (−17.27, 4.72)
Effect of exposure among those with susceptibility genotype
G+E−	2	5	referent	referent
G+E+	18	4	0.09 (0.01, 0.63)	0.08 (0.01, 0.59)

Table 3.

LOC283710 and KLF13 region (rs17815905) and NICU admission

	Cases n	Controls n	OR (95% CI)	aOR (95% CI)
Interaction
G−E−	38	51	referent	referent
G+E−	63	56	1.47 (0.85, 2.55)	1.45 (0.83, 2.53)
G−E+	16	4	5.23 (1.62, 16.89)	4.83 (1.49, 15.66)
G+E+	9	10	1.18 (0.44, 3.17)	1.27 (0.46, 3.53)
Case only p for interaction			0.02	0.02
RERI (95% CI)				−4.53 (−10.81, 1.76)
Effect of exposure among those with susceptibility genotype
G+E−	63	56	referent	referent
G+E+	9	10	0.80 (0.30, 2.11)	0.88 (0.32, 2.38)

Discussion

In this study we examined interaction between 5 gene variants (rs6736278 within CAPN14 at 2p23, rs2302009 within CCL26 at 7q11, rs3806932 within TSLP at 5q22, rs17815905 within the LOC283710 and KLF13 region at 15q13, and rs1800469 within TGF-β at 19q13) and 6 maternally-reported early life exposures (cesarean delivery, preterm delivery, NICU admission, breastfeeding, antibiotics in infancy, and absence of a furry pet in infancy). We conducted case-only tests of interaction (with subsequent confirmation of results with the case-control analysis) and observed evidence of statistical interaction between rs6736278 (CAPN14) and breastfeeding and rs17815905 (LOC283710 and KLF13 region) and NICU admission. Case-control analyses of these associations provided additional evidence to support the presence of an interaction. The combined effect of susceptibility genotype and environmental exposure was less than expected (e.g. antagonistic) after examining the independent effect of genotype and environmental exposure. For example, the combined effect for breastfeeding and genotype was significantly less (aOR: 0.43; 95% CI: 0.12, 1.62) than the effect of genotype (aoR: 5.38; 95% CI: 0.89, 32.71) or breastfeeding alone (aOR: 1.70; 95% CI: 0.76, 3.77). Indeed, breastfeeding had a strong protective effect (aOR 0.08; 95% CI: 0.01, 0.59) in those with the susceptibility genotype (CAPN14 in rs6736278). While the exact mechanism of how breastfeeding and CAPN14 may interact and become protective for disease is unknown, the protective effect of breastfeeding was demonstrated in a study of interaction between breastfeeding and the cluster of differentiation 14 (CD14) genotype (CD14C-159T) in relation to atopic sensitization.²⁷ CAPN14 is integral to maintaining barrier function in the esophagus.¹³ It is tempting to speculate that with impaired barrier function, breastfeeding may confer a higher protective effect than that which is observed when barrier function is preserved, but this would need to be investigated in mechanistic studies.

The early life factors examined in this study have been associated with dysbiosis in gut colonization and early life.^28–36 Indeed, the esophagus harbors its own microbiome that is dysregulated in EoE patients.³⁷ It is interesting to speculate that the interacting SNPs might mediate their effects by modifying esophageal microbiota content and/or their interaction with EoE-related immunological pathways, even before the onset of EoE.^{28, 38–40} While this study was not designed to directly assess the influence of these exposures on the microbiome, the results provide the first potential evidence of gene-environment interactions in EoE, and support future work examining the microbiome and EoE.

Our results must be interpreted with caution as this preliminary study had a relatively small sample size and we have not sought to replicate these findings in an independent sample. Further, the associations reported required only nominal significance (p< 0.05) and did not account for multiple testing. To try to minimize the risk of false positives, we limited the analyses and environmental effects which had been associated with EoE risk in prior studies. Given many of the exposures are correlated (e.g. preterm delivery and NICU admission), each exposure-genotype interaction assessment did not represent a completely independent test and the exposures and SNPs selected were selected a priori, based on hypotheses of an interactive effect, as opposed to an agnostic study of gene-environment interaction where multiple correction testing may be more relevant.⁴¹ While we observed interaction for several of the environmental exposures examined, these exposures may be a proxy for some other factor(s) that are interacting with the identified genes. Additionally, while early life factors such as cesarean delivery, breastfeeding, NICU admission, and having pets in the home are likely not hampered by recall bias, recall of antibiotic use and, possibly, preterm delivery, could be biased.

In conclusion, we have preliminary evidence that the etiology of EoE may involve the interplay of gene and environmental factors. Further understanding of these interactions provides an opportunity to potentially alter the development of EoE, as environmental factors can be modified in individuals harbouring specific genetic variants.

Clinical Implications.

EoE-predisposing polymorphisms interact with early life factors to modify risk of EoE disease development. Risk for EoE disease may be modifiable in individuals with certain environmental exposures and gene variants.

Abbreviations

EoE

eosinophilic esophagitis

RERI

relative excess risk due to interaction

aOR

adjusted odds ratio

NICU

neonatal intensive care unit

SNP

single nucleotide polymorphism