Assessing the comorbidity between asthma and depression through polygenic risk scoring and time-to-event models

Abstract

Background

Patients with asthma have an increased risk of developing depression, affecting their quality of life. To date, the processes contributing to this comorbidity remain unclear.

Methods

We integrated two large genome-wide association studies (88,486 patients with asthma and 447,859 controls; 412,024 patients with depression and 1,587,577 controls) with cross-sectional and longitudinal information available from the All of Us Research Program (N = 87,167) through polygenic risk scoring (PRS), Cox proportional-hazards models, one-sample Mendelian randomization (MR), and gene-set and drug-repurposing analyses.

Results

We observed that depression PRS was associated with increased asthma risk (hazard ratio, HR = 1.13, 95% CI = 1.09–1.17), also when accounting for comorbidity status (HR = 1.08, 95% CI = 1.04–1.12). Conversely, the effect of asthma PRS was null after accounting for comorbidity status. One-sample MR analysis showed an effect of depression genetic liability on asthma, ranging from beta = 0.36 ± 0.03 when considering a linear relationship to beta = 3.21 ± 0.31 when considering possible nonlinear relationships. Conversely, the effect of asthma genetic risk on depression was null after accounting for potential confounders. The gene-set analyses showed that asthma and depression polygenic risks share biological processes, molecular functions, and cellular components related to the immune system and the lung-brain axis.

Conclusions

Genetic predisposition contributes to asthma-depression comorbidity through direct effects and shared pathogenic processes. These findings highlight the potential to develop targeted interventions to prevent and treat the co-occurrence of respiratory and neuropsychiatric disorders.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12916-026-04646-w.

Background

Asthma is a life-altering illness that affects approximately 8% of the United States (US) population [1, 2]. The primary symptoms of the disease include shortness of breath, chest tightness, coughing, and wheezing due to asthma pathogenic processes such as inflammation, airway hyperresponsiveness, and airway remodeling. In addition to respiratory symptoms, asthma has also been associated with negative mental health outcomes in children, adolescents, and adults [3–6]. In particular, depressive symptoms have been reported among asthmatic patients more frequently than in the general population [7, 8], reaching up to 30% in severe patients with asthma [9]. Through the analysis of diagnosis records for > 151 million US residents, a psychiatric subgroup mostly related to depression has also been identified among patients with asthma [10]. In addition to reducing the quality of life, asthma-depression comorbidity is also associated with increased healthcare utilization and expenditure [11].

Several hypotheses have been proposed to explain depression-asthma comorbidity. Shared immune-inflammatory pathogenesis in both disorders was proposed because of the altered cytokine profiles observed among affected patients [12]. More recently, large-scale genome-wide association studies (GWAS) permitted investigators to characterize the genetic architectures of asthma and depression and assess possible shared mechanisms [13, 14]. For example, the RERE gene has been identified as genome-wide significant in both disorders [15]. A study focused on identifying shared asthma-depression loci uncovered 10 single-variant pleiotropic associations and one gene associated with both illnesses through genetically regulated transcriptomic variation in blood [16]. In addition to individual loci, genetic correlation between asthma and depression has been reported with estimates ranging from 0.19 to 0.41 [17, 18]. Mendelian randomization (MR) analyses suggested that this genetic overlap may be due to possible cause-effect relationships where the genetic liability to depression increases asthma risk [16–18]. While these previous studies support that pleiotropy contributes to asthma-depression comorbidity, the dynamics underlying the effects reported remain unclear.

To perform a high-resolution analysis of asthma-depression comorbidity, we integrated genome-wide association statistics generated from large cohorts with newly generated individual-level information available from the All of Us (AoU) Research Program [13, 14, 19]. Because we hypothesize that different pleiotropic scenarios contribute to the comorbidity between asthma and depression, we conducted bidirectional analyses to assess whether asthma can cause depression and whether depression has an effect on asthma. Additionally, we investigated biological signatures shared by asthma and depression polygenic risk, also exploring molecular compounds potentially targeting pathogenic processes contributing to their comorbidity.

Methods

Study design

The goal of the present study was to understand dynamics contributing to the asthma-depression comorbidity. Electronic health records (EHR) were the primary data sources for deriving cross-sectional and longitudinal information regarding these health outcomes, considering the existing literature supporting the reliability of EHR-based phenotyping for research purposes [20, 21]. Because different scenarios may be present, we used complementary analytic approaches to integrate genetic and observational data. Polygenic risk scoring (PRS), time-to-event analysis, and Mendelian randomization (MR) were conducted to investigate the bidirectional relationship between asthma and depression. Multiple potential confounders (e.g., age, sex, smoking status, and household income) were considered in a model-based asthma-depression association test. Sensitivity analyses were performed to assess the reliability of the putative causal relationships observed. Evidence consistency across methods was used to assess the robustness of the potential direct effects. To investigate potentially shared processes, we examined the biological signatures overlapping between asthma and depression polygenic risk and conducted a drug-repurposing analysis to identify therapeutic interventions targeting their comorbidity.

All of Us Research Program

AoU is a population-based cohort enrolling US participants to develop precision medicine through the analysis of self-reported information, EHRs, physical measurements, wearable devices, and genomic data [19]. The present study was conducted using individual-level information available from the AoU Research Program. This was approved by the AoU Institutional Review Board. The data used in this study were obtained through an approved data use agreement between the AoU Research Program and Yale University. When enrolled in the AoU cohort, participants provided informed consent and consent for the release of EHRs separately.

In the present study, we leveraged AoU data available from the controlled tier v7, which included 413,457 adult participants aged 18 and older at the time of recruitment. From the backend database of EHRs harmonized in the format of the Observational Medical Outcomes Partnership Common Data Model (OMOP CDM) [22], asthma (SNOMED code: 317,009; OMOP concept ID: 317009) and depressive disorder (SNOMED code: 35489007; OMOP concept ID: 440383) were selected from the AoU domain “Conditions.” Similar criteria have been previously used also by other AoU studies to investigate asthma [23, 24] and depression [25, 26]. Information regarding sex at birth (question ID: 1585845), annual household income (question ID: 1585375), and whether participants smoked at least 100 cigarettes in their life (question ID: 1585857, hereafter referred to as smoking status) were obtained from the AoU “Survey” domain. Annual household income ordinal scale was converted to a binary outcome, considering an annual household income greater than or equal to $50,000. Study participants’ age at recruitment was calculated using each participant’s date of birth and enrollment date (survey item: survey_datetime).

In AoU, whole-genome sequencing (WGS) data were available for 245,394 participants. A detailed description of the WGS protocol, quality control, and genetically inferred ancestry classification is available elsewhere [27]. In our analyses, we applied the following quality control criteria: biallelic variants with minor allele frequency > 1%, Hardy–Weinberg equilibrium p < 10⁻⁶, call rate > 95%, and per-individual genotyping rate > 95%.

Considering AoU participants with WGS data (N = 245,394), we excluded individuals who were not of genetically inferred European descent (N = 117,201), did not pass quality control for the WGS data (N = 24,982), or did not have complete phenotype data (N = 16,044). After applying these exclusion criteria, the final sample included 87,167 AoU participants (Table 1), which comprised 5013 patients with asthma-depression comorbidity, 6202 patients with only asthma, 13,550 patients with only depression, and 62,402 controls (AoU participants without asthma and depression diagnoses). Our analyses were restricted to AoU participants of genetically inferred European descent. This is because the limited sample size of existing asthma and depression GWAS to calculate polygenic risk scores for other population groups, and the AoU non-European-descent sample to be used as a target cohort, would not have permitted us to exclude that null results were false negatives.

Table 1.

Characteristics of the All of Us participants investigated in the present study

	Asthma-depression (n = 5013)	Asthma only (n = 6202)	Depression only (n = 13,550)	Controls (n = 62,402)
Age, mean (SD)	54.1 (15.2)	56.9 (16.3)	55.1 (15.9)	55.5 (17.2)
Sex: female, n (%)	3822 (76)	4039 (65.1)	8561 (63.2)	34,493 (55)
Smoking status, n (%)	2454 (49)	2356 (38)	6921 (51)	24,904 (40)
Annual income ≥ $50,000, n (%)	2315 (46)	4299 (69)	7193 (53)	42,663 (68)

Genome-wide association statistics

To assess asthma and depression polygenic risk, we leveraged genome-wide association statistics generated from large GWAS meta-analyses that did not include the AoU cohort. Asthma GWAS meta-analysis included 536,345 individuals (88,486 patients with asthma and 447,859 controls, 97% European descent) available from combining the UK Biobank with 66 cohorts that are part of the Trans-National Asthma Genetic Consortium [13]. To investigate depression polygenic risk, we used the publicly available version of the GWAS meta-analysis of the Psychiatric Genomics Consortium [14], which included 1,999,601 European-descent individuals (412,024 patients with depression and 1,587,577 controls) from 27 cohorts, excluding 23andMe data.

Data analyses

Polygenic risk scoring

Leveraging asthma and depression genome-wide association statistics as a training dataset and the UK Biobank European sample as a linkage disequilibrium (LD) reference, posterior variant-level effect sizes were calculated using the PRS-continuous shrinkage (PRS-CS) approach and its PRS-CS-auto algorithm to determine the optimal parameters for each outcome [28]. PRS-CS approach is a Bayesian regression framework, using priors to allow for marker-specific adaptive shrinkage (i.e., the amount of shrinkage applied to each genetic marker is adaptive to the strength of its association signal in GWAS) [28]. Based on the PRS-CS output, we used PLINK1.9 [29] to calculate asthma and depression PRS in AoU participants.

Asthma and depression PRS were tested with respect to depression and asthma outcomes in the AoU cohort using multivariate logistic regression models. Sex, age, annual income, smoking status, and top 10 within-ancestry principal components (PC) were included as covariates in the regression models to estimate adjusted odds ratios (OR) and 95% confidence intervals (95% CI). Cragg and Uhler’s pseudo-R² (also known as Nagelkerke’s R²) for each PRS was also calculated as the difference in pseudo-R² between models with and without the PRS term. To account for comorbidity status in the cross-disorder PRS analysis, the asthma diagnosis was included as an additional covariate when testing the asthma PRS association with depression. Similarly, the depression diagnosis was included as an additional covariate when testing the depression PRS association with asthma. Variance inflation factor (VIF) was calculated to evaluate the potential collinearity in these models. To further examine the robustness of the PRS associations, we conducted two additional sensitivity analyses. In one, we included body mass index (BMI), use of antidepressants (N06A), and use of asthma medications (R03A, adrenergics, inhalants; R03B—other drugs for obstructive airway diseases, inhalants; R03D—other systemic drugs for obstructive airway diseases) as additional covariates. In the other analysis, the PRS analysis was limited to AoU participants not using antidepressants or asthma medications, with BMI added as an additional covariate.

Time-to-event analysis

To explore the temporal relationship between asthma and depression, we defined two subsamples of the AoU cohort using information regarding the date of onset for asthma and depressive disorder available in AoU “Condition Occurrence” data. In the “Asthma to Depression” subsample, we excluded those who developed depression before asthma. To calculate the duration for participants with asthma, we set the start date as the date of asthma diagnosis and the end date as the date of the first depression diagnosis if reported in their EHRs, or as the end date of the AoU controlled tier v7 dataset (i.e., January 31, 2023), treating the latter as right-censored. For participants without asthma, we used their enrollment date (survey item: survey_datetime) as the start date and the end date of the AoU controlled tier v7 dataset (i.e., January 31, 2023) as the end date. Similarly, in the “Depression to Asthma” sample, we excluded those who developed asthma before depression. For participants with depression, we set the start date as the date of depression diagnosis and the end date as the date of the first asthma diagnosis if reported in their EHRs. If not, we set the end date of the AoU controlled tier v7 dataset (i.e., January 31, 2023) as the cutoff date of the dataset and considered them to be right-censored. For participants without depression, we used their enrollment date (survey item: survey_datetime) as the start date and the end date of the AoU controlled tier v7 dataset (i.e., January 31, 2023) as the end date. These criteria are in line with those used previously by other time-to-event analyses using EHRs [30, 31].

Considering these subsamples, we employed Cox proportional hazards models [32], assessing the contribution of their polygenic risk. Adjusted hazard ratio (HR) and 95% CI were calculated, including sex, age, annual income, smoking status, and top 10 within-ancestry PCs as covariates. We also calculated VIF to evaluate the collinearity in the model.

Mendelian randomization

To investigate causal effects underlying asthma-depression comorbidity, we employed the one-sample MR design [33]. This permitted us to use asthma and depression PRS as instrumental variables to test in the AoU cohort whether one disease has a direct effect on the risk of the other, circumventing environmental confounders and reverse causation bias [33]. To ensure the reliability of the MR estimates, we used three different methods that rely on different assumptions: the two-stage least-squares (2SLS) approach [34] implemented in the R package ivreg (available at https://cran.r-project.org/web/packages/ivreg/) and two-stage predictor substitution (2SPS) and two-stage residual inclusion (2SRI) approaches implemented in the OneSampleMR R package (available at https://cran.r-project.org/web/packages/OneSampleMR/) [35]. One-sample MR covariates included age, sex, annual income, smoking status, and the top 10 within-ancestry PCs. To assess the validity of the MR assumptions, we performed multiple sensitivity analyses. Specifically, a weak-instrument test was performed to assess possible violations of the relevance assumption (i.e., the genetic instrument is robustly associated with the exposure of interest). Wu-Hausman diagnostic was performed to assess possible violations of the exogeneity assumption (i.e., the genetic instrument is independent of confounders related to the outcome). Breusch–Pagan test for heteroscedasticity was performed to detect possible violations of the exclusion-restriction assumption (i.e., the genetic instrument is associated with the outcome only through its effect on the exposure). This analysis followed the STROBE-MR (Strengthening the Reporting of Observational Studies in Epidemiology using Mendelian Randomization) reporting checklist [36] (Additional File 1).

Pathway-enrichment and drug-repurposing analyses

We used the PRSet approach [37] to assess the molecular pathways underlying the cross-disorder PRS associations between asthma and depression. Differently from PRS-CS, PRSet is based on the PRSice clumping-thresholding approach [38]. In our study, we considered a p-value threshold = 1, LD r2 = 0.1, and window size = 250 kb. The Molecular Signatures Database was used to define curated gene sets, ontology gene sets, immunologic signature gene sets, and cell-type signature gene sets using the human gene annotation [39]. Considering pathway enrichments surviving Bonferroni correction accounting for the number of tests performed (N = 29,577, p = 1.69 × 10⁻⁶) in both cross-disorder PRS analyses (i.e., asthma PRS with respect to depression and depression PRS with respect to asthma), we conducted a drug-repurposing analysis using the Gene2drug approach [40]. To avoid redundancy among pathway enrichments in this analysis, we employed the REVIGO method, considering Jiang and Conrath’s semantic distance and a similarity degree threshold of 0.5 [41, 42]. The resulting pathways were stratified by gene sets (i.e., curated gene sets, gene ontologies, immunologic signatures, and cell-type signatures) and databases (i.e., biological processes, cellular components, and molecular functions) before being entered into the Gene2drug analysis. Bonferroni correction accounting for the number of tests performed (N = 15,708, p = 3.18 × 10⁻⁶) was applied to the drug-repurposing results.

Results

The sample investigated included 5013 patients with asthma-depression comorbidity, 6202 patients with only asthma, 13,550 patients with only depression, and 62,402 controls (AoU participants without asthma or depression diagnoses). While no major difference was observed in the mean age across these groups, patients with asthma-depression comorbidity included more females (76%) and had a lower annual income (≥ $50,000, 46%) than the other three groups investigated (Table 1). Conversely, the percentage of lifetime smokers was similar in the asthma-depression and depression-only groups (49% and 51%, respectively) and lower in the asthma-only patients and the controls (38% and 40%, respectively; Table 1).

In line with the expected generalizability of previously identified genetic effects, one standard deviation increase in asthma PRS was associated with a 39% increase in the odds of asthma in the AoU cohort (OR = 1.39, 95% CI = 1.36–1.42; Nagelkerke’s R² = 1.83%; Fig. 1; Additional File 2: Table S1). Similarly, depression PRS was associated with an AoU depression diagnosis (OR = 1.24, 95% CI = 1.22–1.26; Nagelkerke’s R² = 1.04%; Fig. 1; Additional File 2: Table S1). Cross-disorder PRS associations were also observed. Specifically, asthma PRS was associated with depression (OR = 1.03, 95% CI = 1.01–1.05; Additional File 2: Table S1), while depression PRS was associated with asthma (OR = 1.12, 95% CI = 1.09–1.14; Additional File 2: Table S1). To further investigate whether cross-disorder PRS associations were fully explained by asthma-depression comorbidity, we also controlled for the co-occurrence of the two disorders. After including comorbidity status as an additional covariate in the model, asthma remained associated with both depression diagnosis and depression PRS (depression-diagnosis OR = 3.51; depression-PRS OR = 1.06, respectively; Additional File 2: Table S1). When testing depression as an outcome and including comorbidity status in the model, we observed a direct association with asthma diagnosis (OR = 3.57; Additional File 2: Table S1) but an inverse relationship with asthma PRS (OR = 0.97; Additional File 2: Table S1). In all multivariate logistic regression models, there was no evidence of collinearity among the variables (VIF < 1.2; Additional File 2: Table S2).

Fig. 1.

Polygenic risk score (PRS) associations with asthma and depression in the All of Us Research Program cohort. Model-1 covariates include sex, age, annual income (i.e., annual household income ≥ $50,000), smoking status (i.e., smoked at least 100 cigarettes in their life), and the top 10 within-ancestry principal components. Model-2 covariates include the Model-1 covariates and disease status (i.e., asthma status when testing the association of asthma PRS with depression, depression status when testing the association of depression PRS with asthma)

When including BMI, antidepressants, and asthma medications as additional covariates, the effect of depression PRS on asthma remained significant (OR = 1.08, p = 6.57 × 10⁻¹¹), also when accounting for comorbidity status (OR = 1.06, p = 2.63 × 10⁻⁷). Similar results were observed when limiting the PRS analysis to AoU participants who were not using antidepressants or asthma medications (Additional File 2: Table S1). Conversely, the effect of asthma PRS on depression became null when BMI, antidepressants, and asthma medications were included as additional covariates, as well as when limiting the PRS analysis to AoU participants who were not using antidepressants or asthma medications (Additional File 2: Table S1).

Considering the dates of onset for asthma and depressive disorder, we identified 3185 patients with incident asthma (i.e., asthma occurred after depressive disorder) and 15,679 patients with incident depression (i.e., depressive disorder occurred after asthma). The median follow-up time was 3.45 years for asthma and 3.23 years for depressive disorder. Fitting Cox proportional hazard models, we explored PRS temporal relationships with asthma and depression diagnoses (Fig. 2). As expected, asthma PRS was associated with an increased risk of asthma (HR = 1.23, 95% CI = 1.19–1.28; Additional File 2: Table S3), and depression PRS was associated with depression (HR = 1.19, 95% CI = 1.16–1.23; Additional File 2: Table S3) in the AoU cohort. In the cross-disorder analysis, asthma PRS was associated with depression risk (HR = 1.04, 95% CI = 1.01–1.07; Additional File 2: Table S3), but the effect became null after accounting for asthma comorbidity (Additional File 2: Table S3). Conversely, depression PRS was associated with asthma risk (HR = 1.13, 95% CI = 1.09–1.17; Additional File 2: Table S3), even after accounting for depression comorbidity (HR = 1.08, 95% CI = 1.04–1.12; Additional File 2: Table S3). Accounting for asthma and depression PRS did not reduce the strong association between asthma and depression: asthma remained associated with depression risk (HR = 3.94, 95% CI = 3.67–4.22), and depression remained associated with asthma risk (HR = 3.93, 95% CI = 3.61–4.27; Additional File 2: Table S3). No collinearity was observed in the Cox proportional hazard models (VIF < 1.2; Additional File 2: Table S4).

Fig. 2.

Temporal associations of polygenic risk scores (PRS) with asthma and depression. Model-1 covariates include sex, age, annual income (i.e., annual household income ≥ $50,000), smoking status (i.e., smoked at least 100 cigarettes in their life), and top 10 within-ancestry principal components. Model-2 covariates include the Model-1 covariates plus disease status (i.e., asthma status when testing the association of asthma PRS with depression, depression status when testing the association of depression PRS with asthma)

To assess possible causal effects underlying the associations observed, we performed a bidirectional one-sample MR analysis (Table 2). Applying three methods based on different assumptions (i.e., 2SLS, 2SPS, 2SRI) [34, 35], we aimed to model different causal relationships that may contribute to asthma-depression comorbidity. The genetic liability to depression had a statistically significant causal effect on asthma with estimates ranging from 3.21 ± 0.31 (2SPS estimate modeling both linear and nonlinear relationships) to 0.36 ± 0.03 (2SLS estimate modeling a linear relationship). Conversely, the genetic liability to asthma appeared to affect depression under the 2SLS and 2SPS assumptions (0.12 ± 0.04 and 0.76 ± 0.25), but the effect was null in the 2SRI analysis (Table 2). Weak-instrument and Wu-Hausman tests supported the strength of the genetic instrument and the robustness of the putative causal effect underlying the effects observed (Additional File 2: Table S5). Conversely, Breusch–Pagan test detected significant heteroskedasticity in the effects detected between depression and asthma (Additional File 2: Table S5). This may affect the reliability of the 2SLS estimate. Applying heteroskedasticity-consistent (HC1) robust standard errors, the 2SLS estimates under heteroskedasticity did not change (depression → asthma 0.36 ± 0.03; asthma → depression 0.12 ± 0.04). Because the 2SPS and 2SRI approaches are based on the generalized method of moments (GMM), their variance estimator is already heteroskedasticity-robust.

Table 2.

Bidirectional Mendelian randomization between asthma and depression

Exposure	Outcome	Method	Estimate	95% CI
Depression	Asthma	2SLS	0.36	0.29–0.42
		2SPS	3.21	2.61–3.80
		2SRI	2.99	2.37–3.61
Asthma	Depression	2SLS	0.12	0.04–0.20
		2SPS	0.76	0.27–1.25
		2SRI	0.47	− 0.03–0.98

2SLS, two-stage least-squares; 2SPS, two-stage predictor substitution; 2SRI, two-stage residual inclusion

To assess whether shared biological processes and molecular pathways also contribute to asthma-depression comorbidity, we conducted a bidirectional cross-disorder PRS analysis stratified by gene sets related to gene ontologies, immunologic signatures, and cell type signatures. Applying a Bonferroni correction accounting for the number of tests performed (N = 29,577, p = 1.69 × 10⁻⁶), we identified 1334 gene sets (Fig. 3; Additional File 2: Table S6) that were statistically significant in both the asthma PRS association with depression and the depression PRS association with asthma. Among the cell-type signatures, top findings included lung-ciliated cells (asthma—p = 1.71 × 10⁻²⁴, depression—p = 1.07 × 10⁻²⁶), human midbrain GABAergic neurons (asthma—p = 7.55 × 10⁻¹⁹, depression—p = 9.23 × 10⁻²⁸), and fetal lung cells (asthma—p = 1.49 × 10⁻³⁰, depression—p = 3.18 × 10⁻¹⁸). With respect to the immunologic signatures, the top result was “Genes up-regulated in plasmacytoid dendritic cell 7d vs 0d in young adults (18–50) after exposure to FluMist, time point 7D” (asthma—p = 5.17 × 10⁻²⁶, depression—p = 1.44 × 10⁻¹⁷). Regarding the curated gene sets, “Genes up-regulated in brain from patients with Alzheimer's disease” (asthma—p = 3.95 × 10⁻⁴⁰, depression—p = 1.37 × 10⁻²⁴) was the top association. Among the gene ontologies, strong associations were observed for regulation of immune system process (asthma—p = 4.34 × 10⁻⁷⁵, depression—p = 7.69 × 10⁻²⁴), synapse (asthma—p = 2.78 × 10⁻²⁶, depression—p = 6.57 × 10⁻²⁹), and neurogenesis (asthma—p = 1.64 × 10⁻³⁶, depression—p = × 10⁻²⁵). The drug-repurposing analysis identified 13 molecular compounds (p < 3.18 × 10⁻⁶; Additional File 2: Table S7) associated with transcriptomic perturbations of loci included in the gene sets shared between asthma and depression (Additional File 2: Table S6). Among them, orphenadrine, HC toxin, and albendazole showed convergent significance across the gene sets (i.e., curated gene sets, gene ontologies, immunologic signatures, and cell type signatures) and the databases (i.e., biological processes, cellular components, and molecular functions) investigated (Additional File 2: Table S7).

Fig. 3.

Enrichment z-score distribution among gene sets reaching Bonferroni multiple testing correction in both asthma and depression cross-disorder polygenic risk score analyses. Full results are available in Additional File 2: Table S6

Discussion

Depression is a well-known non-respiratory asthma comorbidity contributing to poor disease control and greater risk of exacerbations [43]. As noted above, the analysis of diagnosis records for > 151 million US residents highlighted depression as the primary outcome characterizing a distinct subgroup of patients with asthma [10]. Previous studies highlighted the contribution of shared genetic predisposition to asthma-depression comorbidity [16], with a potential causal effect of depression genetic liability on asthma [16–18]. These hypotheses were also supported by studies reporting an association between depression PRS and asthma [44, 45]. Building on these previous studies, the findings generated by our study expanded the understanding of how polygenic risk contributes to asthma-depression comorbidity.

Our cross-sectional analysis (through multivariate logistic regression models) and temporal analysis (through Cox proportional hazard models) showed a bidirectional relationship where asthma PRS is associated with depression, while depression PRS is associated with asthma. However, we observed differences between the two effect directions. The effect of asthma PRS on depression (3% increase in disease odds observed in the cross-sectional analysis, 4% increase in disease risk observed in the temporal analysis) was smaller than that of depression PRS on asthma (12% increase in disease odds, 13% increase in disease risk). The latter effect sizes are in line with previous PRS analyses [44, 45]. Conversely, we also provide the first evidence that asthma genetic predisposition can contribute to depression, although with a small effect. However, we also showed that these cross-disorder PRS relationships differ when accounting for comorbidity status. Indeed, when accounting for asthma status, asthma PRS is inversely associated with depression (i.e., higher asthma polygenic risk is associated with reduced depression, also when accounting for asthma comorbidity). This could be interpreted as individuals with a high genetic predisposition to asthma but not affected by asthma are less likely to receive a depression diagnosis. Because asthma polygenic risk can interact with environmental factors [46–48], individuals with high genetic risk but not affected by asthma may be less exposed to environmental triggers. Previous studies highlighted how environmental factors can also play an important role in asthma-depression comorbidity [49–51]. However, the effect of depression PRS on asthma was reduced but did not change direction, suggesting that individuals with high genetic risk but not developing depression are still associated with an increased risk of asthma. Accordingly, because depression also has an important environmental component, we hypothesize that depression’s environmental risk factors may not have a strong effect on asthma, while asthma environmental factors may also play a role in depression risk. Considering additional confounders (i.e., BMI and use of antidepressants and asthma medications), we observed that the effect of asthma PRS on depression became null, but depression PRS associations with asthma remained significant. This supports that asthma polygenic risk may be indirectly related to depression, while depression polygenic risk may be more directly implicated in asthma pathogenesis.

Previous studies applied the two-sample MR framework to investigate asthma-depression causal relationships, observing a possible effect of depression genetic liability on asthma but no reverse relationship [16–18]. Applying one-sample MR approach, we boosted the statistical power of our genetically informed causal inference analysis through the integration of genome-wide association statistics available from large-scale GWAS meta-analyses and individual-level data available from AoU cohort. Additionally, the one-sample MR framework permitted us to explore different types of relationships contributing to asthma-depression comorbidities. Our findings highlighted that the effect of depression genetic liability on asthma is much stronger than the effect of asthma genetic risk on depression. Using three one-sample MR designs, we assessed different dynamics. With respect to the effect of depression genetic liability on asthma, all one-sample MR designs confirmed this relationship. However, we observed differences in the effect sizes, with the 2SLS method indicating an effect (beta = 0.36) much smaller than those observed in the 2SPS and 2SRI analyses (beta = 3.21 and 2.99, respectively). This could be explained by the fact that 2SLS does not adequately model nonlinear relationships as 2SPS and 2SRI do [34, 35]. The small reduction in the effect size observed between 2SPS and 2SRI could be due to the fact that the latter uses residuals to account more accurately for the effect of covariates. With respect to the effect of asthma genetic risk on depression, the differences in the estimates observed across one-sample MR methods were less pronounced, with the 2SLS method showing the smallest effect. Importantly, although its direction was consistent with that of the other methods, the 2SRI effect was not statistically different from the null. This suggests that the effect of asthma genetic predisposition on depression may be explained by other factors, as also supported by the sensitivity analyses related to the PRS associations observed.

In addition to investigating the relationships between asthma and depression and their genetic risk, we also characterized the shared mechanisms that could contribute to their comorbidity. Across the different gene-set types we investigated (i.e., curated pathways, gene ontologies, immunologic signatures, and cell type signatures), we observed that asthma-associated genes shared with depression were linked to brain processes, while depression-associated genes shared with asthma were observed in relation to lung and immunologic mechanisms. For instance, the cell-type signature related to lung-ciliated cells was linked to asthma PRS in line with the role of airway epithelial barrier dysfunction in asthma development [52]. However, “lung-ciliated cell” signature was also observed in the context of depression PRS. Similarly, we observed the cell-type signature related to GABAergic neurons with respect to depression PRS in line with the potential role of the GABAergic system in depression pathogenesis [53]. GABAergic-neuron signature was also observed with respect to asthma PRS. These findings suggest an asthma-depression overlap at a cellular level. In line with its role in both asthma and depression [54, 55], we also observed immunologic signatures (e.g., genes up-regulated in plasmacytoid dendritic) and several immunologic gene ontologies (e.g., regulation of immune system process) related to both asthma and depression PRS. Additionally, among the curated gene sets, we identified “Genes up-regulated in brain from patients with Alzheimer's disease” as shared between asthma and depression PRS. In support of this finding, previous studies highlighted that both asthma and depression are associated with an increased risk of Alzheimer’s disease [56, 57]. Overall, the biological processes and molecular pathways identified by our asthma-depression PRS analysis contribute to understanding how the lung-brain axis contributes to the comorbidity between respiratory and neuropsychiatric disorders as also hypothesized by previous studies [58–60]. Our drug-repositioning analysis uncovered multiple molecular compounds associated with transcriptomic perturbations related to asthma-depression shared gene sets. In particular, we observed convergent statistical evidence for orphenadrine (a muscarinic antagonist used to treat drug-induced parkinsonism and to relieve pain from muscle spasm), HC toxin (a histone deacetylase inhibitor), and albendazole (a benzimidazole anthelmintic).

We acknowledge several limitations of the present study. First, the data available did not permit us to investigate the comorbidity between asthma and depression symptoms or to investigate heterogeneity among patients affected by both disorders. Second, due to the limited availability of genome-wide data informative of diverse population groups, we investigated only individuals of European descent to avoid interpreting results related to other population groups that may be strongly underpowered. Accordingly, our findings may not be generalizable to other populations because of potential differences in genetic susceptibility, environmental factors, and clinical assessment. Additionally, because AoU participants are not a representative sample of the US population [61], our results will need to be validated in clinical settings representative of diverse population groups to translate them into preventive strategies for identifying individuals vulnerable to asthma-depression comorbidity. Finally, time-to-event analysis may have been affected by the fact that the age at diagnosis reported in the EHRs may have been several years after the age of onset. For instance, left censoring may have affected older participants more strongly, because they may have received their first diagnosis before the use of EHRs.

Conclusions

Our study uncovered how polygenic risk relates to asthma-depression comorbidity through direct effects and shared pathways. In particular, our analyses converged on the potential direct effect of depression’s genetic liability on asthma, which does not appear to be affected by confounders. Conversely, asthma’s genetic liability does not seem to be directly related to depression, but the association may be driven by confounders related to their comorbidity. In this context, we also observed shared genetic mechanisms linking asthma and depression to the immune system and lung-brain axis, with molecular compounds potentially targeting pathogenic processes contributing to their comorbidity. Overall, these findings can contribute to the development of preventive strategies (e.g., respiratory screening in patients with depression) and approaches designed to treat asthma-depression comorbidity.

Abbreviations

2SLS

Two-stage least-squares

95% CI

95% Confidence intervals

AoU

All of Us Research Program

BMI

Body mass index

EHR

Electronic health records

GWAS

Genome-wide association study

HR

Hazard ratio

LD

Linkage disequilibrium

MR

Mendelian randomization

OMOP CDM

Observational Medical Outcomes Partnership Common Data Model

OR

Odds ratio

PC

Principal components

PRS

Polygenic risk score

PRS-CS

PRS-continuous shrinkage

STROBE-MR

Strengthening the Reporting of Observational Studies in Epidemiology using Mendelian Randomization

US

United States

VIF

Variance inflation factor

WGS

Whole-genome sequencing

Authors’ contributions

XW and RP designed the study. XW conducted the primary data analysis. JH, BCM, DQ, and ZM provided analytic support. WG supported phenotyping. YL and JK provided methodological support. All the authors participated in the interpretation of data and critical revision of the manuscript for important intellectual content. All authors read and approved the final manuscript.

Funding

This study was supported by a grant from the National Institutes of Mental Health (RF1 MH132337) and One Mind.

Data availability

The All of Us Research Program data are available on the All of Us Researcher Workbench, (https://www.researchallofus.org/data-tools/workbench/).

Declarations

Ethics approval and consent to participate

Because this research used publicly available de-identified data, it is exempt from IRB approval.

Consent for publication

Not applicable.

Competing interests

Dr. Polimanti is paid for his editorial work on the journal Complex Psychiatry. The other authors declare no competing interests.