Introduction

Asthma is a chronic disease characterized by airway inflammation and narrowing. In recent decades, asthma prevalence has increased1 and it has become a significant public health problem in the US, affecting quality of life, work absenteeism, health care utilization, and health expenditures2,3.

Gene-environment interactions play an important role in the development and severity of asthma4. Air pollutants activate innate immunity, particularly the TLR4 pathway5,6. This is especially true with ozone and nitrogen dioxide but has been shown with other air pollutants such as vanadate7. TLR4 recognizes lipopolysaccharide (LPS) and protects against gram-negative infection. There is evidence that traffic pollutants can interfere with LPS-induced signaling8. Maladaptive activation of TLR4 can importantly contribute to acute or chronic inflammation9. SNPs that perturb TLR4 responses are associated with asthma10,11,12,13,14.

Precision medicine seeks to personalize predictive and therapeutic interventions by accounting for individual genetic, exposure, and disease profiles. However, little attention has been paid to personalized gene-environment interactions in complex diseases like asthma. We, and others15, hypothesized that population-level data may be exploited to generate predictions for individual gene-environment interactions. To test a proof-of-principle of this hypothesis with asthma, we examined associations between combinations of specific SNPs along the TLR4 inflammatory pathway, air pollution exposure, and asthma-related health outcomes. We reasoned that individuals who carry SNP combinations rendering them hyper-responsive to TLR4 activation may be particularly sensitive to air pollution exposures. Conversely, individuals with SNP combinations that render them hypo-responsive to TLR4 activation may be less sensitive to air pollution exposures.

We examined three genes in the TLR4 complex (TLR4, CD14 and TIRAP) and TNFα which is activated downstream of TLR4. SNPs were selected based on relatively high minor allele frequency (MAF) and previous evidence relating the minor alleles to differential health outcomes. They included two putative gain-of-function SNPs (rs2569190 in CD14 and rs1800629 in TNFα) and two putative loss-of-function SNPs (rs4986791 in TLR4 and rs8177374 in TIRAP). The CD14 SNP minor allele (MAF up to 22%) is associated with higher circulating CD14 levels and asthmatic exacerbations10. The TNFα SNP minor allele, rs1800629 (MAF up to 15%), leads to increased TNFα transcriptional activation16, and is associated with chronic obstructive pulmonary disease17. A synergistic effect has been shown between polymorphisms of TNFα and CD14 and bronchial responsiveness in children with asthma18. The TLR4 SNP minor allele rs4986790, which is in linkage disequilibrium with rs4986791 (MAF up to 6.6% in Caucasians), confers hypo-responsiveness to inhaled LPS19. The TIRAP SNP minor allele rs8177374 is associated with decreased response to LPS20. A synergistic effect has been shown between polymorphisms of TLR4 and TIRAP in risk for severe infections following cardiac surgery21.

We explored possible interactions between air pollution exposure and underlying genotype by examining the residential distance to nearest road as a proxy for traffic-related air pollution exposure. Our outcomes were likelihood of asthma diagnosis and asthma exacerbations. We hypothesized that: (1) asthma exacerbations are associated with specific genotype combinations; and (2) roadway proximity, as a proxy for air pollution exposure, may act as an effect modifier in the relationship between genotype and asthma.

Results

Individual SNP genotype associations with asthma diagnosis

We first examined the relationship of genotype and asthma at the individual SNP/gene level (TLR4, CD14, TIRAP, and TNFα). All SNPs followed Hardy-Weinberg Equilibrium for the control (no asthma) samples.

Tables 1 and 2 show the number and percentage of individuals with each SNP by sex, race, ethnicity, smoking status, and income. Table 3 shows mean age of asthma diagnosis by SNP. There were no significant differences in mean age of asthma diagnosis for TNFα and CD14 or for TLR4 heterozygotes. There were too few individuals homozygous for the TLR4 minor allele to perform this comparison. The mean age of asthma diagnosis was significantly younger for individuals with the wildtype (WT) TIRAP allele (22.4 years) compared to those who are carriers for this allele (26.1 years; p = 0.008) indicating that the onset of asthma may be delayed in individuals with at least one copy of the minor TIRAP allele of rs8177374.

We did not detect significant differences in the proportions of carriers vs. WT for each SNP between individuals with and without asthma, using χ2 tests. Allele frequencies were also examined and showed no significant differences.

Individual SNP genotype associations with asthma exacerbations

There were no significant differences in reports of asthma-related sleeplessness or activity limitations by SNP during the previous 14-days (Tables S1S4). TLR4 carriers vs. WT had lower odds ratio (OR) of asthma-related emergency room (ER) visits in the past 12-months (0.39 [0.15, 0.97]) in an unadjusted model. This association was not significant after adjustment for sex, race, smoking status, body mass index (BMI) and income.

Association of responder status, asthma diagnosis and exacerbations, and distance to road

Of the 2,704 participants classified by responder status, 369 (13.7%) were hyper-responders [carriers of CD14 and TNFa minor SNPs] and [WT for TLR4 and TIRAP SNPs], 132 (4.9%) hypo-responders [carriers of TLR4 and/or TIRAP minor SNPs] and [WT for CD14 and TNFa SNPs], and 2,203 (81.5%) neither [all others]. There was no significant variation in the distribution of the responder categories among those with and without asthma (Table S5), no difference in age of asthma diagnosis (Table S6), or in prevalence of asthma-related exacerbations in 14-days (Table S7) or 12-months (Table S8).

Summary information for the individuals in our analysis, by asthma diagnosis and exacerbation, is provided in Table 4. For distance to road analyses, 36 participants were not used because of missing BMI data (2,668 participants analyzed). Figure 1 shows the distribution of participants at the county level across census tracts within North Carolina, the state with the majority (94%) of participants. Mean (median) distance to primary and secondary roads was 6.91 km (3.67 km) and 3.98 km (2.90 km), respectively. Classifying proximity in this way, 200 (7.5%) subjects lived <250 m from the nearest road and 2,468 (92.5%) subjects lived ≥250 m from the nearest road. Distributions of covariates (sex, race, smoking status, BMI, income) by responder-type and distance to roadway categorization are provided in Tables S9 and S10. African-Americans were less likely to be hypo-responders (2.4%) compared to Caucasians (5.5%), and a higher percentage of African-Americans (8.9%) resided in the <250 m category compared to Caucasians (6.9%, p = 0.10). Percentages of hyper-, hypo-, and neither-responders residing in the <250 m distance category were not significantly different.

We examined asthma diagnosis, with distance to roadway and responder-type as the exposures of interest, using a logistic regression model. Unadjusted and adjusted ORs for the association of asthma diagnosis with responder-type and distance-to-roadway are presented in Table 5. Adjusted models controlled for smoking status, sex, race, BMI, and income. The likelihood ratio test comparing interaction and adjusted models for asthma diagnosis provided the suggestion of an interaction between responder-type and distance to road (p = 0.12). Thus, we focus on results from interaction models for asthma diagnosis. The reference group was chosen as a sub-group representative of the source population: Caucasian, female, non-smoking, high income (≥$60,000/year), BMI in the normal or underweight range (i.e. not obese or overweight), neither-responders, and residing ≥250 m from the nearest major (primary or secondary) road. Both hyper- and hypo-responders ≥250 m from a major road had similar odds of reporting an asthma diagnosis as neither-responders ≥250 m from a major road (reference group). The OR of an asthma diagnosis among hyper-responders <250 m from a major road was 2.37 (0.97, 6.01) compared to the reference group. Holding other variables constant at their reference value, the OR of asthma diagnosis among individuals with incomes <$20,000 was 1.58 (1.24, 2.01) compared to individuals with incomes ≥$60,000. African-Americans and men were less likely to report an asthma diagnosis, with adjusted ORs of 0.57 (0.47, 0.69) and 0.41 (0.34, 0.49), respectively. The OR of an asthma diagnosis among other races was 1.40 (0.98, 1.99) compared to Caucasians. Asthma exacerbations For individuals that reported an asthma diagnosis, we examined road proximity, responder-type, and each asthma exacerbation (asthma-related activity limitations, sleeplessness, and ER visits) using logistic regression. Additionally, we looked at the outcome of “any” asthma exacerbation, i.e., individuals reporting any one or more of these three exacerbations. We fit unadjusted, adjusted, and interaction logistic regression models for each asthma exacerbation. The likelihood ratio test indicated that there was no evidence of interaction between responder-type and distance to road on any of the asthma exacerbations examined; here we present results from the adjusted model (Table 6). Hyper-responders ≥250 m from a major road had similar odds of reporting any/all of the asthma exacerbations examined as neither-responders (Table 6). Hypo-responders ≥250 m from the nearest road, however, had lower odds of reporting activity limitations (0.46 [0.21, 0.95]), sleeplessness (0.36 [0.12, 0.91]), and “any” exacerbations (0.50 [0.24, 0.99]) compared to neither-responders ≥250 m from the nearest road (all p < 0.05). Holding other variables constant at their reference value, the OR of sleeplessness among smokers was 1.72 (1.23, 2.42) compared to nonsmokers. Odds of activity limitations among smokers was suggestively higher (1.26 [0.95, 1.68]) when compared to nonsmokers. Men were less likely to report activity limitations, sleeplessness, and “any” exacerbation compared to women. African-Americans were more likely to report sleeplessness (1.57 [1.07, 2.29]) and ER visits (1.71 [1.08, 2.68]) than Caucasians. Odds of all of the exacerbations examined were higher among individuals with incomes <$20,000/year and individuals with incomes $20,000 to$39,999/year compared to individuals with income ≥$60,000/year. Also, odds of all of the exacerbations examined were higher among individuals who were overweight or obese, based on BMI, compared to individuals who were normal weight or underweight. Sensitivity analysis We repeated our main analysis but classified proximity to road as: (1) <300 m or ≥300 m, <400 m or ≥400 m and <500 m and ≥500 m from a primary or secondary road; and (2) <250 m or ≥250 m from a primary, secondary, or tertiary road. Tertiary roads included local, neighborhood, and rural roads22. Overall, findings from the sensitivity analyses were generally consistent with those of the main analysis: hypo-responders had lower odds of sleeplessness, activity limitations, and “any” exacerbation compared to neither-responders (Table S11). Full results of the sensitivity analysis are presented in Tables S11S17. Discussion This study set out to test, as a proof-of-principle, the ability to use refined population analysis of gene-environment interactions to create personalized predictions for the activity of a complex disease. We first explored relationships between genotype and asthma. There was some evidence that individual SNPs related to asthma diagnosis and exacerbations. For example, the mean age of asthma diagnosis was 4 years lower for individuals with the WT TIRAP allele compared to subjects carrying the minor allele. This indicates that rs8177374 is a loss-of-function allele that may delay, but not prevent, development of asthma, and is in itself a novel finding. We then explored relationships between air pollution exposure and genotype combinations by examining asthma exacerbations, responder profiles, and residential distance to nearest road. We hypothesized that distance to roadway and responder status would interactively be associated with likelihood of asthma diagnosis and exacerbations. Consistent with this hypothesis, we found that hypo-responders had significantly lower odds of reporting activity limitations, sleeplessness, and “any” asthma exacerbation compared to neither-responders. Additionally, we found a suggestive interaction between genotype and distance-to-road on odds of asthma diagnosis for hyper-responders. There has been limited research on human innate immune gene SNPs in relation to pollution exposure and airway inflammation. Most studies focus on children: there is a positive association between the TNFα SNP rs1800629 and allergic rhinitis23,24, a protective effect of the TLR4 SNP rs4986791 in the development of hay fever25, and an interaction of the TLR4 SNPs with exposure to particulate matter for prevalence of asthma13. In adults, the role of the TLR4 pathway in association with pollution exposure is much less defined26. Our findings extend the existing literature significantly because our study was conducted in adults and evaluated interaction of pollution with a signaling pathway, an approach that is thought to be more powerful27. Precision medicine should take into account both genetic and exposure profiles. Genetic influences can be assessed with unbiased testing of random genetic and epistatic effects, as happens in large genome-wide association studies. We tested another approach, which analyzes pathways based on strong scientific support for relevance in disease pathogenesis. Such pathway-wide analyses may require smaller sample sizes to detect an effect. Furthermore, by taking into account the additive effect of genetic polymorphisms along the same pathway, the sensitivity of genetic analysis can be amplified. In addition, if we further selectively investigate environmental influences that are known to interact with the genetic pathways assessed, we may be able to create intelligent predictive models of gene-environment interactions that can be used on an individual basis. One can imagine an iterative process of analysis of several such pathways, which would result in an algorithmic approach to treatment of complex environmental disease. Most published reports focus on rare gene variants with strong effects. This approach has great strengths but also limitations, especially in polygenic, complex diseases. For example, previously identified polymorphisms from several large genome-wide association studies only explain a fraction of the variability in lung function28 and are associated with a minority of asthma or chronic obstructive pulmonary disease cases29. Furthermore, it can be argued that the genetic background for many diseases comes from a combination of several, fairly common, weakly active SNPs, which can co-exist in the same individual and additively create a strong pull toward a particular effect30. Our pathway-focused approach served to test this hypothesis. Taking into account environmental exposures, which are specifically linked to the pathway analyzed, enhances the approach. For example, Romieu and London31,32 studied common loss-of-function SNPs in the antioxidant genes GSTM1 and GSTP1, and their effects on the lung function of asthmatic children exposed to high ozone levels. In a series of elegant papers, these investigators showed that asthmatic children with the GSTM1 or GSTP1 null allele were at especially high risk of lung function decrements following exposure to ozone, suggesting a gene-environment interaction. Patients who carried the null allele for both GSTM1 and GSTP1 had even higher risk of ozone-induced lung function effects, suggesting epistatic interactions along the anti-oxidant pathway. Our study has several limitations. We used road classifications rather than traffic volume or measurements of air pollutant concentrations to assign exposure due to data availability. Traffic related air pollution (TRAP) is known to exacerbate existing respiratory diseases. Several studies have validated distance to roads as proxies for TRAP33,34,35 in particular around 200 m from major roads. Our findings would be consistent with these other studies. However, we note that in North Carolina, the location of over 94% of the subjects, the mean annual average daily traffic (AADT) on major roads in 2013 was just above 13,000 vehicles per day, whereas the mean AADT on other small roads where traffic counts were measured was ~3,300 vehicles per day36. Thus, we chose to include only major roads, anticipating that the lower traffic volume on smaller roads would only contribute to background exposure to traffic-related air pollution. Although we used residential location to estimate the effect traffic-related air pollution exposure has on asthma exacerbations, we did not have information on the length of time an individual has resided at the address provided, nor the amount of time they spend at that residential address on a daily, weekly, or monthly basis. Further, our pathway approach of examining responder types in conjunction with environmental exposures presents challenges of sample size and power. Although the model controlled for possible confounders (e.g., race, sex, income, BMI, smoking status), there could be systematic differences between individuals within <250 m and ≥250 m, such as stressful life events and noise levels, that our analysis does not account for. We did not consider land use near residences, although land use may affect asthma symptoms37. Finally, data obtained from the questionnaire (e.g., race, sex, income, asthma exacerbation) is self-reported, with associated limitations. However, in a separate validation of EPR self-reported data in approximately 1,000 individuals, we found a 94% agreement rate between self-reported and on-site medical history and physical examination results (unpublished results), supporting good reliability of our data. Despite limitations, our study provides important insights about the relationships between genetic profile, air pollution, and asthma exacerbations. Individual residential addresses allowed us to estimate residential proximity (air pollution exposure) at the individual level instead of in aggregate over a spatial unit (e.g., block group, county). Using distance to roadway to estimate exposure allowed us to include individuals who do not reside near an air quality monitor, which tend to be located in more urban areas38,39, effectively limiting the geographic scope and increasing exposure measurement error. Our findings suggest that traffic-related air pollution exposure and specific genotype combinations may affect likelihood of asthma diagnosis and exacerbations, including serious exacerbations (e.g., ER visits). Further studies need to be done to show if avoidance of major roadways based on responder status will decrease asthma exacerbations. However, short-term associations between primary pollutants from traffic sources and pediatric asthma ER visits has been noted40. This work supports that targeted association analysis of specific environmentally responsive pathways at a population level may allow the detection of pathway-environment interactions, which underlie asthma development and pathogenesis. Personalized predictions for the activity of a complex disease will require an amalgamation of extensive genotype and environmental data, an era we are approaching with the advent of less expensive whole genome sequencing and deep biomonitoring. Ultimately, multiple such associations may be compiled into a predictive algorithm that can be utilized for individualized asthma management. Methods Asthma diagnosis and exacerbations The Environmental Polymorphisms Registry (EPR)41 is a repository of DNA samples from over 18,000 individuals. The EPR database contains individual-level self-reported data from 8,843 individuals on physician-diagnosed asthma and presence of asthma exacerbations. Asthma exacerbations included asthma-related activity limitations and sleeplessness in the previous 14-days and asthma-related ER visits in the previous 12-months. Asthma diagnosis and exacerbations were categorized as present/absent. DNA genotyping Blood samples were collected from 2,996 EPR participants for whom health data were available, and total genomic DNA was isolated using Qiagen’s Autopure LS system. DNA was genotyped for four genes in the TLR4 pathway. Samples were genotyped using Illumina multiplexing and/or TaqMan SNP Genotyping Assays (Applied Biosystems): rs8177374, TIRAP; rs4986791, TLR4; rs2569190, CD14; and rs1800629, TNFa. Our analyses defined and used individuals with one (heterozygous) or both (homozygous minor) copies of the minor allele as carriers of the SNP. Individuals with both major alleles are WT. Based on genotyping results, subjects were divided into three responder-types: 1. (1) Hyper-responders: [carriers of CD14 and TNFa minor SNPs] and [WT for TLR4 and TIRAP SNPs]; 2. (2) Hypo-responders: [carriers of TLR4 and/or TIRAP minor SNPs] and [WT for CD14 and TNFa SNPs]; 3. (3) Neither-responders: All others. Hyper- and hypo-responders were anticipated to have over- and under-reactive responses to air pollution exposure, respectively. Neither-responders were expected to have intermediate responses to pollution. Exposure: proximity to a major road For the 2,996 genotyped individuals, geocoding (assignment of latitude and longitude based on residential address) was performed based on the contact address provided, using ArcGIS 10 software. Of genotyped subjects, 2,830 (94.5%) were successfully geocoded. Individuals with missing covariate data (n = 146) or responder type (n = 16) were removed from the overall analysis dataset, for a sample size of 2,668 individuals. Geocoded individual residential addresses were overlaid with the Streetmap Premium 201442 streets layer. We determined the linear distance between geocoded residential address and the centerline of the nearest major road. The nearest major road included primary [major highways with and without limited access (freeways)] and secondary (primarily state and county highways) roads, based on the Topologically Integrated Geographic Encoding and Referencing database22. Previous studies of asthma have used road proximity as a proxy for traffic-related air pollution exposure43,44,45,46,47, but have not categorized distance to road consistently. Data suggest that air pollution levels are elevated near major roads, with pollution levels returning to background concentrations at approximately 300 m48. Individuals residing closer to a major road are expected to have systematically higher exposures to traffic-related air pollution than individuals farther from a major road. Thus, we dichotomized distance from each geocoded address to the centerline of the nearest primary or secondary road into <250 m and ≥250 m. Statistical analysis Individual SNP genotype associations with asthma diagnosis and exacerbations Cohort statistics, numbers and percentages, were determined for each SNP for individuals with and without asthma, including sex, race, ethnicity, smoker status, BMI, and annual household income. We performed t-tests on age at asthma diagnosis for each SNP, comparing WT to carrier. In the subset of participants with asthma, we examined the prevalence of asthma-related exacerbations including activity limitations and sleeplessness during the previous 14-days and asthma-related ER visits in the previous 12-months. Unadjusted odds ratios were generated from logistic regression analyses independently for each SNP. Logistic regression was performed for each SNP with adjustments for sex, race, smoking status, BMI, and income. Smoking status was operationalized as a binary variable. Individuals who reported previously or currently smoking >100 cigarettes in their lifetime were classified as ever smokers, while those who did not report previous or current tobacco smoking were classified as never smokers. Underweight was classified as a BMI of <18.5, normal weight was a BMI of 18.5 to <25, overweight was a BMI of 25 to <30 and obese was a BMI of 30 or higher. We collapsed income into four categories (<$20,000; $20,000–$39,999; $40,000–$59,999; and ≥\$60,000). Due to small cell sizes in the Asian, Native American, Native Hawaiian/Pacific Islander, multiple races, and unknown/not reported race categories, these individuals (n = 144) were combined into an “other races” category.

Statistical analyses were performed using MedCalc for Windows (MedCalc Software, Ostend, Belgium), SAS versions 9.3 and 9.4 (Cary, NC), and R 3.2.249.

Responder type, distance to road, and asthma diagnosis and exacerbations

We performed descriptive analyses using proportions for categorical variables and means and medians for continuous variables (e.g., distance to major road prior to categorization into <250 m and ≥250 m). Tests for homogeneity in individual characteristics across the three responder-types and two distance to roadway groups were evaluated using χ2 contingency table statistics for categorical variables. To evaluate our hypothesis that asthma diagnosis and exacerbations are associated with proximity to a major road and responder type, we examined asthma diagnosis using logistic regression to model presence/absence of an asthma diagnosis. We also examined presence/absence of asthma exacerbations using self-reported data on asthma-related ER visits, sleeplessness, and activity limitations, in both unadjusted and adjusted models. A third model specification was fit to examine our hypothesis of an interaction between responder type and distance to roadway on asthma diagnosis and exacerbations.

Adjusted models controlled for covariates available in the EPR database and suspected to be potential confounders in the existing literature50,51,52,53: sex, race, smoking status, BMI, and income.

We used likelihood ratio tests to compare nested models, testing the hypothesis that adjusted models are an improved fit over unadjusted models. As a sensitivity analysis, we re-analyzed data using the same methodology but included tertiary roads, classifying proximity to road as <250 m and ≥250 m from the nearest primary, secondary, or tertiary road. We also applied different classifications for near versus far distance from road, including 300 m, 400 m, and 500 m, in addition to the 250 m presented in the main analysis. A 2-sided 95% significance level was used for all statistical inference (α < 0.05), unless stated otherwise.

This research was conducted under NHGRI, NIH Institutional Review Board approved protocol #14-E-0053 and all participants have provided written informed consent. All methods were performed in accordance with the relevant guidelines and regulations.