Epidemiological and bioinformatics analyses of air pollution and genetic susceptibility in aortic stenosis risk

Abstract

While air pollution is a recognized cardiovascular risk, its specific impact on aortic stenosis (AS) remains poorly characterized. This study investigates the association between air pollution and incident AS, integrating gene–environment interactions and network toxicology. Based on the UK Biobank as the primary cohort, long-term air pollution exposure (per standard deviation increase) is associated with an increased risk of AS, with HRs and 95% CIs of 1.60 (1.55,1.66) for PM_2.5, 1.37 (1.32, 1.41) for PM₁₀, 1.37 (1.32, 1.42) for NO₂, and 1.36 (1.31, 1.41) for NO_x. The observed associations demonstrate high consistency across a suite of advanced internal methodological validations and are further replicated in the Tianshan Community Cohort. There are joint and interactive effects of genetic susceptibility and air pollutants on AS risk. Network toxicology and bioinformatics analyses reveal key PM-AS target genes enriched in the lipid and atherosclerosis, fluid shear stress and atherosclerosis, and IL-17 signaling pathway. In conclusion, long-term exposure to PM_2.5, PM₁₀, NO₂, and NO_x is significantly associated with an increased AS risk, with a potential interaction between environmental exposure and genetic susceptibility.

This study uses the UK Biobank to associate air pollution with an increased risk of aortic stenosis. The risk is amplified by genetic susceptibility and the main associations are replicated in the Tianshan Community Cohort. Network toxicology and bioinformatics analyses suggest target genes are mainly enriched in the lipid and atherosclerosis signaling pathway.

Introduction

Aortic stenosis (AS), one of the valvular heart diseases, is characterized by narrowing of the aortic valve aperture; AS impedes left ventricular outflow, leading to progressive cardiac dysfunction¹. The natural history of AS reveals that asymptomatic patients generally have a milder disease course, but once symptoms develop, the progression accelerates rapidly. Without timely valve replacement surgery, patient survival rates are significantly diminished^2,3. The prevalence of AS and severe AS in the elderly is 12.4% and 3.4% in Western countries, individually^4,5. A large Australian registry reports that 5-year mortality rates for moderate and severe AS are 56% and 67%, respectively, highlighting the high morbidity and mortality associated with these conditions⁶. More seriously, with increasing life expectancy and the intensification of population aging, the epidemiological characteristics of valvular heart disease have undergone significant changes, with AS replacing rheumatic valvular disease as the leading cause of valvular heart disease^7,8. Given that aortic valve stenosis presents a 94% 5-year mortality rate in untreated populations and, even with treatment, is associated with high surgical costs and significant postoperative complications, it has become an imperative global health concern that cannot be overlooked^9–11.

Ambient air pollution is a prominent environmental factor linked to human health problems. According to the latest evidence from the Global Burden of Disease Study, particulate matter was the leading cause of the global disease burden¹². Although some studies have identified positive associations between air pollution and cardiovascular diseases (CVDs), such as coronary artery calcification, other large-scale investigations (including the MESA and SCAPIS studies) have provided inconclusive evidence regarding this association^13–15. Moreover, to our knowledge, no study has yet investigated the impact of long-term air pollution exposure on the risk of incident AS, particularly within large-scale prospective cohorts. Only a few studies have explored air pollution in relation to valve disease (i.e., valvular heart disease and rheumatic heart disease)^7,16,17 and calcification of arterial valves, the latter being a critical mechanism in AS^15,18, suggesting that air pollution is a potential risk factor for AS. Genetic determinants represent another primary driver of AS incidence. Recent large-scale genomic evidence has identified shared genetic features between AS and coronary artery disease, primarily related to inflammation, lipid metabolism, smooth muscle cell proliferation, and calcification. However, AS also possesses a distinct genetic risk profile involving cell differentiation and development, DNA-damage response and apoptosis, and shear stress, among others^19–21. Furthermore, tissue-specific transcriptomic analyses have revealed that a significant portion of genes dysregulated in AS are uniquely expressed within the aortic valve, highlighting localized pathological processes²². To link these localized genetic predispositions with external environmental triggers, network toxicology provides a systematic framework to identify chemical-protein interactions that drive disease. Advancements in toxicogenomics have significantly enhanced our understanding of how pollutants influence biological processes through these interactions^23–25. Therefore, a comprehensive investigation integrating gene–air pollution interactions and toxicological bioinformatics will significantly elucidate the initiation and progression of this insidious and life-threatening disease.

In this work, we aimed to investigate the longitudinal associations between long-term exposure to air pollutants, including particulate matter with an aerodynamic diameter ≤2.5 µm (PM_2.5), particulate matter with an aerodynamic diameter ≤10 µm (PM₁₀), nitrogen dioxide (NO₂), and nitrogen oxides (NO_x), and the risk of AS incidence based on a large-scale prospective cohort. Furthermore, by examining the joint and interactive effects of genetic susceptibility and air pollution on incident AS, while integrating network toxicology and bioinformatic analyses, we aimed to provide mechanistic insights into the biological pathways underlying the environmental risks associated with AS.

Results

In this study, a total of 447,216 participants, of whom 44.37% were men, with an average age of 56.25 years (±8.08), were included. During a median follow-up period lasting 12.65 years, 3682 incident cases of AS were detected. Table 1 displays the baseline characteristics of all participants, whereas Table S1 outlines those of white Europeans. In comparison to individuals without AS, those diagnosed with AS tend to be predominantly men, older in age, possessed a higher body mass index (BMI), exhibited a preference for smoking, and lacked adequate physical exercise. Furthermore, within this subgroup, there was a heightened prevalence of hypertension, diabetes, and hyperlipidemia. Table S2 provides a comprehensive overview of the air pollutants distribution, presenting mean values (±SD) of 9.45 (1.53) μg/m³ for PM_2.5, 14.11 (2.11) μg/m³ for PM₁₀, 16.55 (5.09) μg/m³ for NO₂, and 24.33 (8.72) μg/m³ for NO_x.

Table 1.

Baseline characteristics of study participants

Baseline characteristics	Total participants (N = 447216)	Participants without AS (N = 443534)	Participants with AS (N = 3682)	P values
Age (mean (SD), year)	56.25 (8.08)	56.20 (8.07)	62.42 (5.77)	<2.20E-16
Sex (%)				<2.20E-16
Men	198418 (44.37)	196139 (44.13)	2279 (61.90)
Women	248798 (55.63)	247395 (55.78)	1403 (38.10)
Ethnic groups (%)				1.80E-12
White Europeans	422551 (94.48)	418975 (94.46)	3576 (97.12)
Non-white Europeans	22645 (5.06)	22556 (5.09)	89 (2.42)
Missing	2020 (0.45)	2003 (0.45)	17 (0.46)
BMI (mean (SD), kg/m2)	27.32 (4.73)	27.31 (4.73)	29.40 (5.34)	<2.20E-16
Employment (%)				<2.20E-16
Employed	263169 (58.85)	261907 (59.05)	1262 (34.27)
Retired	143826 (32.16)	141706 (31.95)	2120 (57.58)
Unemployed	35595 (7.96)	35323 (7.96)	272 (7.39)
Missing	4626 (1.03)	4598 (1.04)	28 (0.76)
Education (%)				<2.20E-16
University degree	169113 (37.81)	167975 (37.87)	1138 (30.91)
Non-university degree	200339 (44.80)	198867 (44.84)	1472 (39.98)
Missing	77764 (17.39)	76692 (17.29)	1072 (29.11)
Alcohol intake (%)				3.18E-01
Never	19140 (4.28)	18965 (4.28)	175 (4.75)
Current or former alcohol intake	427037 (95.49)	423540 (95.49)	3497 (94.98)
Missing	1039 (0.23)	1029 (0.23)	10 (0.27)
Tobacco smoking (%)				<2.20E-16
Never	249021 (55.68)	247400 (55.78)	1621 (44.02)
Current or former smoking	196041 (43.84)	194002 (43.74)	2039 (55.38)
Missing	2154 (0.48)	2132 (0.48)	22 (0.60)
Diet score (%)				1.73E-04
0–1	64467 (14.42)	63870 (14.40)	597 (16.21)
2–3	240266 (53.72)	238255 (53.72)	2011 (54.62)
4–5	142483 (31.86)	141409 (31.88)	1074 (29.17)
Physical exercise (%)				<2.20E-16
Never	27339 (6.11)	27021 (6.09)	318 (8.64)
Low activity	29776 (6.66)	29428 (6.63)	348 (9.45)
Moderate activity	328741 (73.51)	326229 (73.55)	2512 (68.22)
Vigorous activity	58676 (13.12)	58210 (13.12)	466 (12.66)
Missing	2684 (0.60)	2646 (0.60)	38 (1.03)
TDI (mean (SD))	−1.46 (2.97)	−1.46 (2.97)	−1.35 (3.03)	1.92E-02
Income level (%)				<2.20E-16
<£31000	179966 (40.24)	178024 (40.14)	1942 (52.74)
≥£31000	202736 (45.33)	201620 (45.46)	1116 (30.31)
Missing	64514 (14.43)	63890 (14.40)	624 (16.95)
Hypertension (%)	114727 (25.65)	112836 (25.44)	1891 (51.36)	<2.20E-16
Diabetes (%)	20515 (4.59)	19977 (4.50)	538 (14.61)	<2.20E-16
Hyperlipidemia (%)	37344 (8.35)	36448 (8.22)	896 (24.33)	<2.20E-16
Genetic risk category (%)				<2.20E-16
Low genetic risk	148924 (33.30)	148134 (33.40)	790 (21.46)
Intermediate genetic risk	149369 (33.40)	148211 (33.42)	1158 (31.45)
High genetic risk	148923 (33.30)	147189 (33.19)	1734 (47.09)

Differences between groups were evaluated using the χ² test, Wilcoxon–Mann–Whitney test, or Student’s t-test, as appropriate for the variable type. All statistical tests were two-sided.

AS aortic stenosis, BMI body mass index, TDI Townsend deprivation index.

In Table 2, the hazard ratios (HRs) and 95% confidence intervals (CIs) for AS incidence associated with each air pollutant were presented, with exposures modeled as both continuous and categorical variables. Participants with chronic exposure to air pollutants were positively associated with an elevated risk of incident AS, with effect size of 1.60 (1.55, 1.66) for PM_2.5 (per 1.53 μg/m³ increase), 1.37 (1.32, 1.41) for PM₁₀ (per 2.11 μg/m³ increase), 1.37 (1.32, 1.42) for NO₂ (per 5.09 μg/m³ increase), and 1.36 (1.31, 1.41) for NO_x (per 8.72 μg/m³ increase). When air pollution exposure was categorized, compared to the reference group (T1 level), the risk of AS incidence in the highest exposure level was 2.58 (2.37, 2.81) for PM_2.5, 2.02 (1.86, 2.20) for PM₁₀, 1.98 (1.82, 2.15) for NO₂, and 2.01 (1.85, 2.19) for NO_x, respectively. The relatively large E-values suggest that the observed relationships between air pollutants and incident AS are unlikely to be fully explained by unmeasured confounding factors. Table S3 presents the associations derived from various causal inference frameworks used to achieve covariate balance. The magnitude and direction of the effect sizes remained highly consistent across these models. For PM_2.5, the estimated HRs (95% CIs) were 1.57 (1.50, 1.64) for the inverse probability weighting (IPW)-weighted model and 1.41 (1.37, 1.46) for the generalized propensity scores (GPS)-adjusted model. Figure 1 illustrates the linear exposure-response relationships between PM_10, NO₂ and NO_x exposure and AS risk (P for non-linearity > 0.05). In contrast, a non-linear exposure-response relationship was observed for PM_2.5 (P for non-linearity = 0.001). Mirroring the trends observed in the exposure–response curves, the associations between air pollutant exposure and the risk of incident AS remained statistically significant even at levels below the World Health Organization (WHO) Air Quality Guidelines (Table S4). Table S5 presents the findings of the external validation conducted within the Tianshan Community Cohort. Consistent with the primary analysis, long-term exposure to air pollutants was significantly associated with an increased risk of incident AS in the Chinese population. The HRs and 95% CIs per SD increase were 1.30 (1.25, 1.34) for PM_2.5 (per 21.83 μg/m³ increase), 1.22 (1.18, 1.27) for PM₁₀ (per 90.88 μg/m³ increase), and 1.15 (1.12, 1.18) for NO₂ (per 7.77 μg/m³ increase), respectively. Furthermore, the exposure-response curves in this validation cohort closely mirrored those observed in the UK Biobank (Fig. S1).

Table 2.

Associations between air pollutants and the risk of incident AS

Air pollutants	HRs (95% CIs)	E values	P values	P for trend
PM2.5
PM2.5-T1	1.00 (Ref.)	--	--	<2.20E-16
PM2.5-T2	1.34 (1.22, 1.47)	2.01	4.90E-10
PM2.5-T3	2.58 (2.37, 2.81)	4.60	<2.20E-16
PM2.5, per SD+ increase	1.60 (1.55, 1.66)	2.58	<2.20E-16	--
PM10
PM10-T1	1.00 (Ref.)	--	--	<2.20E-16
PM10-T2	1.42 (1.30, 1.55)	2.19	3.64E-15
PM10-T3	2.02 (1.86, 2.20)	3.46	<2.20E-16
PM10, per SD+ increase	1.37 (1.32, 1.41)	2.07	<2.20E-16	--
NO2
NO2-T1	1.00 (Ref.)	--	--	<2.20E-16
NO2-T2	1.23 (1.13, 1.34)	1.76	4.00E-06
NO2-T3	1.98 (1.82, 2.15)	3.37	<2.20E-16
NO2, per SD+ increase	1.37 (1.32, 1.42)	2.08	<2.20E-16	--
NOx
NOx-T1	1.00 (Ref.)	--	--	<2.20E-16
NOx-T2	1.25 (1.15, 1.37)	1.82	4.21E-07
NOx-T3	2.01 (1.85, 2.19)	3.43	<2.20E-16
NOx, per SD+ increase	1.36 (1.31, 1.41)	2.06	<2.20E-16	--

HRs and 95% CIs are estimated using Cox proportional hazards models. The models are adjusted for age, sex, ethnic groups, income level, TDI, education, employment, alcohol consumption, tobacco smoking, diet score, physical exercise, BMI, hypertension, diabetes, and hyperlipidemia. All statistical tests are two-sided. ⁺Standard deviation of PM_2.5, PM₁₀, NO₂, and NO_x are 1.53, 2.11, 5.09, and 8.72 μg/m³, respectively.

HRs hazard ratios, CIs confidence intervals, PM_2.5 fine particulate matter with diameter <2.5 μm, PM₁₀ particulate matter with diameter <10 μm, NO₂ nitrogen dioxide, NO_x nitrogen oxides, AS aortic stenosis, BMI body mass index, TDI Townsend deprivation index, T tertile, SD standard deviation.

Fig. 1. Association between long-term air pollution exposure and the risk of incident AS.

A PM_2.5; B PM₁₀; C NO₂; D NO_x. A restricted cubic spline regression model with 3 knots (at the 10th, 50th, and 90th percentiles) is used to estimate the exposure-response relationships between air pollutants and the risk of AS among participants. HRs (solid lines) and 95% CIs (shaded areas) are estimated using restricted cubic spline regression models. The models are adjusted for age, sex, ethnic groups, income level, TDI, education, employment, alcohol consumption, tobacco smoking, diet score, physical exercise, BMI, hypertension, diabetes, and hyperlipidemia. All statistical tests are two-sided. HRs hazard ratios, CIs confidence intervals, PM_2.5 fine particulate matter with diameter <2.5 μm, PM₁₀ particulate matter with diameter <10 μm, NO₂ nitrogen dioxide, NO_x nitrogen oxides, AS aortic stenosis, BMI body mass index, TDI Townsend deprivation index.

The association between the genetic predisposition and the risk of AS incidence was shown in Table S6, with the HR and 95% CI of 1.41 (1.36, 1.45) for GRS (per SD increase), 1.42 (1.30, 1.56) for the intermediate genetic risk group vs. the low genetic risk group, and 2.11 (1.94, 2.30) for the high genetic risk group vs. the low genetic risk group. Figure 2 depicts the joint effects of PM_2.5, PM₁₀, NO₂, NO_x, and genetic predisposition on the risk of incident AS. The HRs and 95% CIs of AS risk in the high genetic risk and high pollutants exposure group vs. the low genetic risk and low air pollutants exposure group were 5.86 (4.98, 6.91) for PM_2.5, 4.54 (3.87, 5.34) for PM₁₀, 4.07 (3.48, 4.76) for NO₂, and 4.16 (3.55, 4.87) for NO_x, individually. In Table S7, significant additive interactions were discovered between air pollution and GRS on AS risk. Among individuals with high genetic risk and high pollutant exposure, the relative excess risk due to interaction (RERI) and the attributable proportion due to interaction (AP) were 2.03 (1.47, 2.59) and 35% (27%, 42%) for PM_2.5, 1.30 (0.84, 1.77) and 29% (20%, 38%) for PM₁₀, 0.97 (0.55, 1.39) and 24% (14%, 33%) for NO₂, and 1.02 (0.59, 1.45) and 25% (15%, 34%) for NO_x, respectively.

Fig. 2. The joint association of long-term air pollution exposure and genetic risk with the risk of incident AS.

A PM_2.5; B PM₁₀; C NO₂; D NO_x. HRs and 95% CIs are estimated using Cox proportional hazards models. The models are adjusted for age, sex, income level, TDI, education, employment, alcohol consumption, tobacco smoking, diet score, physical exercise, BMI, hypertension, diabetes, hyperlipidemia, genotyping batch, and the first ten genetic principal components. All statistical tests are two-sided. HRs hazard ratios, CIs confidence intervals, GRS genetic risk score, PM_2.5 fine particulate matter with diameter <2.5 μm, PM₁₀ particulate matter with diameter <10 μm, NO₂ nitrogen dioxide, NO_x nitrogen oxides, AS aortic stenosis, BMI body mass index, TDI Townsend deprivation index, T tertile.

The results of the network toxicology and bioinformatics analyses are presented in Fig. 3. A total of 1308 candidate genes associated with PM were retrieved from the Comparative Toxicogenomics Database (CTD). Additionally, 598 genes highly relevant to AS were extracted from the GeneCards and OMIM databases. A Venn diagram identified 118 target genes linking PM exposure to AS (Fig. 3A). These target genes were predominantly involved in biological processes related to positive regulation of gene expression, cellular components associated with extracellular space, and molecular functions pertaining to cytokine activity (Fig. 3B). Pathway enrichment analysis of the target genes revealed that the lipid and atherosclerosis, fluid shear stress and atherosclerosis, and IL-17 signaling pathway were the top three pathways implicated in the association between PM exposure and AS (Fig. 3C). The protein-protein interaction network of core genes associated with PM exposure and AS is presented in Fig. 3D. Hub genes identified within the protein–protein interaction network (Fig. 3E), including IL6, TNF, and IL1B, represent potential molecular candidates that may be involved in the relationship between air pollution exposure and AS risk. Fig. S2 presents the results of the network toxicology analyses of NO₂ exposure and AS.

Fig. 3. Target genes and potential pathways of PM exposure and AS identified by the network toxicological analyses.

All statistical tests were two-sided. A The Venn diagram; B GO analysis of target genes; C The Sankey diagram of enrichment pathways analyses; D Protein–protein interaction network; E Hub genes of protein–protein interaction network. AS aortic stenosis, PM particulate matter.

Our findings remained robust across all sensitivity analyses (Table S8), including: (1) excluding participants diagnosed with AS within the initial two years of follow-up; (2) restricting the analysis to individuals with residential tenure exceeding five years; (3) using age as the time scale in the Cox models; (4) adjusting for continuous metabolic markers (systolic blood pressure, blood glucose, and low-density lipoprotein cholesterol) instead of categorical diagnoses; (5) incorporating indoor, commuting, and occupational exposures into the models to control for their potential confounding; and (6) utilizing multiple imputation to account for missing data. Furthermore, exposure-response curves derived from shape-constrained additive models were highly consistent with the primary restricted cubic spline (RCS) analysis (Fig. S3).

Discussion

This study integrates traditional epidemiological analysis, genetic-environmental interaction analysis, as well as network toxicology and bioinformatics analyses to comprehensively investigate the impact of long-term air pollution exposure and genetic susceptibility on the risk of incident AS. The epidemiological analysis identifies a significant association between air pollution and an elevated AS risk, demonstrating remarkable consistency across both the UK Biobank and the external validation cohort. From the perspective of genetic-environmental interactions, there are joint and interactive effects between air pollution and genetic susceptibility. Network toxicology and bioinformatics analyses identify key target genes associated with PM and AS, which are enriched in the lipid and atherosclerosis, fluid shear stress and atherosclerosis, and IL-17 signaling pathways. This study provides valuable epidemiological insights into the influence of air pollution exposure on AS onset and contributes to elucidating the potential mechanisms driving air pollution-associated AS.

While ambient air pollution, notably PM_2.5, is widely acknowledged as a key environmental risk factor for CVDs, and its effects on most CVD subtypes were thoroughly documented, evidence concerning the detrimental influence of inhaling diverse air pollutants on the incidence of valvular heart disease is very scarce. A cross-sectional study in China (n = 4158) reported that long-term residence in high traffic density areas was associated with a higher risk of valvular heart disease incidence (odds ratio and 95% CI: 1.799 (1.221, 2.651))¹⁶. Furthermore, in a study of US Medicare beneficiaries (n = 59,761,494), the exposure-response curve for valvular heart disease exhibited an initial increase followed by a subsequent decline as PM_2.5 levels rose¹⁷. To date, no study has investigated the influence of air pollution on AS. This gap may be attributed to the relatively low prevalence of the disease²⁶, necessitating large-scale prospective cohorts to achieve sufficient sample sizes. Findings from our external validation demonstrated a consistent association between air pollution exposure and incident AS risk, albeit with a discernible discrepancy in effect magnitudes observed between the two cohorts. Specifically, the UK Biobank benefits from relatively mature electronic medical record linkage and a decade-long follow-up, which effectively captures AS cases in a predominantly middle-aged and older population at peak risk. In contrast, the Tianshan Community Cohort includes a larger proportion of younger adults (over 18 years) with a shorter follow-up; the inherent cardiovascular resilience in these individuals may partially offset the adverse effects of air pollutants. Furthermore, whereas air pollution in UK is primarily traffic-derived with high biological toxicity, the air pollution in the Tianshan Community Cohort (Northwest China) may consist of a higher fraction of crustal mineral components, which typically exhibit lower per-unit toxicity despite higher mass concentrations. The use of two large-scale cohort studies to explore and validate the link between air pollution and incident AS risk constitutes a key strength of this research.

For the exposure-response relationship, our study found that the risk of AS increased more substantially at higher concentrations of air pollutants than at lower levels. The harmful effects of air pollution on aortic health were evident even when exposure was limited to the stricter air quality standards recommended by the WHO. These findings align with several large-scale studies showing cardiovascular impacts from low-level exposure to PM_2.5 and NO₂^27–29, which indicate no safe threshold; even minimal exposure correlates with adverse cardiovascular outcomes. Given the asymptomatic yet life-threatening nature of AS, it is essential to prioritize the transition to clean energy, enhance urban greenness, and implement regular clinical screening programs.

Despite the numerous GWAS dedicated to identifying genetic loci associated with AS, individual genetic variations typically exert minor effects and do not fully capture the genetic heritability of the disease^30,31. Thus, this study represents a pioneering effort to investigate the interplay between genetic susceptibility and ambient air pollution exposure on the risk of incident AS, utilizing a GRS that integrates multiple identified genetic loci. Our research indicates that individuals with a high genetic susceptibility who are concurrently exposed to elevated levels of air pollution have a 4.07 to 5.86-fold increased risk of incident AS. Moreover, the additive interaction observed between genetic predisposition and air pollutants is particularly noteworthy, suggesting that these risk factors may share common pathogenic pathways in the onset of AS. For instance, exposure to air pollution exacerbates vascular inflammation and promotes the development of aortic calcification, leading to reduced valve opening and restricted outflow from the left ventricle^32,33. Some evidence suggests that inflammation-related genes, such as IL-6 and TNF, may influence CVD risk factors through their interaction with air pollution exposure³⁴. A multicenter study conducted in Europe found interactions between air pollution exposure and variations in the IL-6 gene region on blood IL-6 concentration³⁵. Another case-control study conducted in Stockholm observed that the relationship between air pollution and blood IL-6 levels was significant in individuals carrying the minor (G) allele of the IL-6 variants rs1800795 and rs1800797³⁶. In the experimental model, air pollution exposure led to increased lipid accumulation in the aortic root and enhanced activation of the CD36 and NLRP3 inflammasomes in ApoE−/− mice³⁷. Meanwhile, the lipoprotein(a) locus (rs10455872) has been identified as having genome-wide significance for aortic valve calcification²⁰.

Nevertheless, it is essential to distinguish between statistical interaction and biological synergism. While our analysis revealed a significant additive interaction between genetic predisposition and air pollutants, and network toxicology identified IL6, TNF, and IL1B as hub genes, these findings remain exploratory in nature. They provide a hypothesis-generating framework that necessitates future experimental research to elucidate the precise biological mechanisms of these associations.

Several significant advantages of our study should be acknowledged. This research constitutes the largest prospective cohort study to date to elucidate the association between long-term air pollution exposure and incident AS risk, with findings rigorously cross-verified through multi-model and external validation. Furthermore, the integration of gene-environment interaction analysis, network toxicology and bioinformatics approaches has significantly deepened our understanding of environmental health risks and the underlying disease mechanisms. Despite its strengths, several limitations of this study should be considered. Firstly, while causal inference methodologies, including IPW and GPS, were employed to balance measured covariates, the observational design of this study limits our ability to confirm a definitive causal effect. The possibility of residual confounding from unmeasured variables cannot be entirely excluded. Nevertheless, results remained robust across multiple modeling approaches, and the temporal precedence inherent in our prospective cohort design further bolsters the inference of a potential association between exposure and outcome. Second, deriving the AS-GRS from European-ancestry data limits its broader generalizability. Such cross-population disparities in linkage disequilibrium architecture and allele frequencies often lead to biased risk estimates^38,39. Furthermore, ancestral backgrounds can modify gene–environment pathways, as population-specific gene expressions may alter sensitivity to pollution-induced oxidative stress and systemic inflammation^40,41. Environmental inequality is another key factor resulting in disproportionate cumulative exposure. When coupled with heightened genetic sensitivity, these environmental disparities further amplify the overall health damage. For instance, several multi-ethnic studies have shown that Black populations have exhibited a higher risk of air pollution-associated incident CVD compared to White populations^42–44. Moreover, a recent study found that the benefits of PM_2.5 reduction have been unevenly distributed, with non-Hispanic Black and Hispanic populations experiencing less pronounced decreases in cardiovascular mortality relative to their non-Hispanic white counterparts³⁷. Thus, the integration of multi-ethnic genomic data is essential for optimizing GRS in future studies, enabling a more precise evaluation of AS risk within a gene-environment interaction framework. Third, the UK Biobank participants are not a random sample of the UK population, introducing potential selection bias. Fourth, although network toxicology databases are comprehensive, they are constrained by the scope of existing literature, potentially omitting targets that remain undocumented in current studies. Moreover, the pathways identified herein are primarily hypothesis-generating and warrant subsequent in vivo and in vitro experimental validation to elucidate their functional roles. Finally, the UK Biobank currently lacks the longitudinal imaging data necessary to quantify the rate of stenotic progression or calcium accumulation. Future studies employing serial imaging are warranted to elucidate the impact of air pollution on the clinical trajectory of AS, thereby identifying specific windows of vulnerability along the disease course.

In conclusion, long-term exposure to ambient PM_2.5, PM₁₀, NO₂, and NO_x is positively associated with an elevated risk of AS incidence in the UK population. Furthermore, our findings underscore a significant gene-environment interplay, suggesting that genetic susceptibility may modify the impact of air pollutants on AS incidence. These findings bolster the existing epidemiological evidence regarding the role of air pollution in incident AS, potentially providing mechanistic insights into the broader health impacts of ambient pollutant exposure.

Methods

UK Biobank received ethical approval from the North West Multi-Centre Research Ethics Committee (REC reference¹⁶:/NW/0274). The present analyses were conducted under UK Biobank application numbers 69741 and 615033. The Tianshan Community Cohort received ethical approval from the Ethics Committee of the People’s Hospital of Xinjiang Uygur Autonomous Region (KY2023042009). Written informed consent was obtained from all participants in both cohorts.

Study design and population

The UK Biobank is a large-scale prospective cohort study encompassing comprehensive phenotypic and genetic data from over 500,000 participants aged 37–73 years across the UK. Participants were recruited between March 2006 and December 2010. Upon enrollment, individuals provided informed consent via electronic signature, completed questionnaires regarding sociodemographic, lifestyle, and health-related factors, and underwent a series of standardized physical and clinical assessments to further enrich the dataset^45,46.

From the initial pool of 502,480 participants, individuals were excluded based on the following criteria: (1) participants with a prior history of AS, ischemic heart disease, stroke, aortic valve disease, or diseases of the arteries (n = 32,192); (2) participants with missing air pollutant exposure data (n = 9147); and (3) participants with incomplete genetic information (n = 13,925). This resulted in a final cohort of 447,216 individuals.

Exposure assessment

This study obtained near-surface high-resolution (1 × 1 km²) PM_2.5, PM₁₀, NO₂, and NO_x pollution data generated using an air dispersion model from the Department of Environment, Food, and Rural Affairs (https://uk-air.defra.gov.uk). By synergizing a multitude of air pollution data sources, such as the Emission inventory, Energy projections, Road traffic counts, Road transport activity projections, Road transport emission factors, Measurement data, and Meteorological data, the air dispersion model produced data of annual mean PM_2.5, PM₁₀, NO₂, and NO_x concentrations. These data underwent a rigorous validation process to ascertain the model’s robustness. Detailed metrics of the model’s performance were documented in a technical report (https://uk-air.defra.gov.uk/library/reports?section_id=9). Like many previous studies^47–50, we computed individual air pollutants exposure, based on their home location, over a defined period that spanned the time of enrollment to the time of censoring events.

Genetic data

The genetic dataset in the UK Biobank comprises an extensive array of genotype information. Detailed descriptions of the genotyping process, data estimation, and quality control procedures can be found in the published literature⁵¹. Single-nucleotide polymorphisms (SNPs) associated with AS at a genome-wide significance threshold (P ≤ 5 × 10⁻⁸) were predominantly selected from two genome-wide association studies (GWASs)^31,52. A total of 40 independent (r² < 0.05) AS-related SNPs (minor allele frequency > 0.05) were incorporated into the process of genetic risk score (GRS) construction (Table S9). The GRS for AS was computed utilizing the subsequent equation:

$$GRS=\sum \beta \times SNP\times \left(\frac{N}{Sum\left(\beta \right)}\right)$$ 1

Where β represents the effect sizes of risk alleles estimated by the referred GWASs, and $N$ denotes the total number of SNPs^49,53. The calculated GRS was divided into tertiles, categorized as low, intermediate, and high genetic risk.

Outcome definition

Health-related outcome data in the UK Biobank were compiled by establishing linkages with records from hospital admissions and death registrations. AS cases in this study were identified by referencing previous research and confirmed based on the International Classification of Diseases, 10th Revision (ICD-10), as well as the Office of Population Censuses and Surveys Classification of Interventions and Procedures, 4th Revision (OPCS-4)^54,55. Detailed ICD-10 and OPCS-4 codes were presented in Table S10. To focus on incident AS, we excluded participants with pre-existing AS at baseline. Incident AS was defined as the initial diagnosis of AS during the follow-up period, identified through linkage to hospital inpatient and death registry records. Participants were followed until the earliest of the following censoring events: the diagnosis of incident AS, death, or the end of follow-up (September 22, 2021).

Covariate assessment

This study adjusted for several potential confounders, including sociodemographic characteristics, health behaviors, and baseline comorbidities. Covariates comprised age (years), sex (men or women), BMI (kg/m²), income level (<£31,000 or ≥£31,000)⁵⁶, ethnic groups (white European or non-white European), education (university degree or non-university degree), and employment (employed, retired, or unemployed). Health behaviors, such as tobacco smoking and alcohol consumption (never, former, or current), dietary habits, and physical activity, were self-reported and categorized into various tiers. Specifically, dietary scores ranged from 0 to 5, and physical exercise was classified as never, low, moderate, or vigorous, consistent with a prior study⁵⁷. Additionally, we accounted for pre-existing comorbidities, including hypertension, diabetes, and hyperlipidemia.

Network toxicology and bioinformatics analyses

The molecular mechanisms underlying air pollution exposure and AS were explored through toxicogenomic bioinformatics analyses. Air pollution–related genes were retrieved from the Chemical–Gene Interactions module of the CTD^58,59, with results restricted to human-specific interactions identified using the keywords “Particulate Matter” or “Nitrogen Dioxide”. AS–related genes were collected from GeneCards⁶⁰ (https://www.genecards.org) and OMIM⁶¹ (https://omim.org) using “Aortic Stenosis” as the keyword. To ensure the biological relevance and comprehensiveness of the results, all targets were cross-referenced and standardized through the UniProt database⁶² (https://www.uniprot.org/). The intersection of air pollution–related and AS–related genes was identified using a Venn diagram. Protein–protein interaction networks were constructed using the STRING database⁶³ (https://www.string-db.org/) (Homo sapiens, interaction score > 0.4) and analyzed in Cytoscape (v3.10.3)⁶⁴. For each gene, degree, betweenness, and closeness centralities were computed, and those exceeding the median in all three metrics were identified as core genes. Hub genes were identified using the CytoHubba plugin with the Maximal Clique Centrality algorithm. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were performed using the DAVID database^65,66(https://david.ncifcrf.gov/) (Homo sapiens). GO terms included biological process, cellular component, and molecular function.

External validation

External validation was performed using the Tianshan Community Cohort (https://clinicaltrials.gov/study/NCT07207863), a large-scale prospective study tracking longitudinal health in Northwest China. Data were integrated from a multi-tier network of 7,523 primary care centers and 226 hospitals, encompassing both health examinations and clinical records. Detailed cohort characteristics have been documented in prior literature^67–69. From an initial pool of 9,254,018 screened individuals aged > 18 years who underwent baseline examinations in 2017, we applied exclusion criteria identical to those utilized in the UK Biobank analysis. The final validation subset comprised 8,329,679 participants, among whom 6,256 incident AS cases were identified during the follow-up period (2017–2024).

Statistical analysis

To compare the distribution of characteristics between the AS and non-AS groups during follow-up, we selectively employed statistical methods such as the χ² test, Wilcoxon-Mann-Whitney test, or Student’s t test based on the respective variable types. Covariates with missing values were handled using the missing-indicator method. We applied a multivariable-adjusted Cox proportional hazards model to explore the associations between air pollutant exposure and the risk of incident AS. The model adjusted for a set of underlying confounders, such as socio-demographic factors, lifestyle factors, and disease history, as detailed in the “Covariate assessment” section. To achieve a balanced distribution of confounding variables, this study utilized IPW and GPS as robust measures to minimize residual confounding and strengthen the basis for causal interpretation of the observed associations^70,71. GPS is an extension of the propensity score approach, representing the probability of an individual being exposed to the observed level given the covariates⁴⁷. IPW is an advanced method aimed at achieving optimal covariate balance and simulating a randomized assignment of exposure⁷². In implementing IPW, we fulfilled the core criteria of consistency, positivity, and exchangeability (Figs. S4 and S5). In addition, E-values were utilized to estimate the minimum association required for an unmeasured confounder to nullify the observed effects⁷³. To characterize the exposure-response relationship between long-term air pollution exposure and AS risk, this study employed an RCS model. The optimal number and placement of knots were determined by minimizing the Bayesian Information Criterion (Table S11)⁷⁴. Additionally, we aimed to explore the associations between air pollutant concentrations below the 2005 WHO Air Quality Guideline levels (PM_2.5 < 10 μg/m³, PM₁₀ < 20 μg/m³, and NO₂ < 40 μg/m³) and the risk of incident AS. Low-level NO_x exposure was defined as an exposure level below the median. To provide external validation, we independently replicated the aforementioned analysis within the Tianshan Community Cohort. The proportional hazards assumption was verified using Schoenfeld residual tests. The effect sizes of the model were presented as HRs along with their corresponding 95% CIs.

Focusing on gene–environment interactions, this study investigated the combined effects of long-term air pollutant exposure and genetic susceptibility among participants of European ancestry. We subsequently evaluated additive interactions between air pollution exposure and genetic predisposition in relation to incident AS, quantified via the RERI and AP⁷⁵. In these analyses, models were further adjusted for the top ten genetic principal components and genotyping batches. The 95% CIs for additive interaction measures were calculated using the bootstrapping method⁷⁶.

Several sensitivity analyses were conducted to evaluate the robustness of our findings, including: (1) excluding participants diagnosed with AS within the initial two years of follow-up to minimize potential reverse causation; (2) restricting the analysis to individuals with a residential tenure of at least five years at their baseline address to ensure exposure stability; (3) employing age as the underlying time scale in the Cox models; (4) replacing categorical comorbidities (hypertension, diabetes, and hyperlipidemia) with their continuous counterparts (systolic blood pressure, blood glucose, and low-density lipoprotein cholesterol); (5) incorporating indoor, commuting, and occupational exposures into the models to control for their potential confounding influence on the associations; and (6) addressing missing data via Multiple Imputation by Chained Equations under the Missing at Random assumption⁷⁷. The specific details of the imputation procedure and its quality are detailed in Text S1, Fig. S6 and S7. (7) To validate the functional form of the exposure-response curves, a shape-constrained additive model was implemented to complement the RCS analysis. All statistical analyses were performed using the R programming language (4.2.0).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

Y.H.T., J.Y., Y.N.Y. and Y.D.Y.M. contributed to the conception and design of the study. Y.H.T., Y.D.Y.M., Q.J. and J.Z. advised on all statistical aspects and interpreted the data. Y.D.Y.M., Q.J. and J.Z. performed the literature search and the analyses. All authors critically reviewed this and previous drafts. All authors approved the final draft for submission, with final responsibility for publication. All authors approved the final version of the manuscript. The corresponding author attests that all the listed authors meet authorship criteria and that no others meeting the criteria have been omitted.

Peer review

Peer review information

Nature Communications thanks Yichen Wang, Philippe Pibarot, and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

Data from the UK Biobank are available upon application. The UK Biobank is an open research resource, available after submission of a research plan via the website (https://www.ukbiobank.ac.uk). The Tianshan Community Cohort data are not publicly available due to participant privacy and ethical restrictions. Researchers may request access by contacting the corresponding author (yaohua_tian@hust.edu.cn). Access requires a formal research proposal and institutional ethical approval. We will respond to requests within 30 days. Approved users must sign a Data Use Agreement to ensure data is used only for the specified research. Air pollution data (2006–2021) are provided by the Department for Environment, Food and Rural Affairs (https://uk-air.defra.gov.uk). Source data are provided with this paper.

Code availability

The custom R scripts used for statistical analysis and data visualization in this study have been deposited in the GitHub repository and are publicly available at https://github.com/MoooDylan/Ma_AS_air-pollution.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-73026-y.