The Association Between Insulin Use and Asthma: An Epidemiological Observational Analysis and Mendelian Randomization Study

Asthma is a common respiratory disease caused by genetic and environmental factors, but the contribution of insulin use to the risk of asthma remains unclear. This study aimed to investigate the association between insulin use and asthma in a large population-based cohort, and further explore their causal relationship by Mendelian randomization (MR) analysis. An epidemiological study including 85,887 participants from the National Health and Nutrition Examination Survey (NHANES) 2001–2018 was performed to evaluate the association between insulin use and asthma. Based on the inverse-variance weighted approach, MR analysis were conducted to estimate the causal effect of insulin use on asthma from the UKB and FinnGen datasets, respectively. In the NHANES cohort, we found that insulin use was associated with an increased risk of asthma [odd ratio (OR) 1.38; 95% CI 1.16–1.64; p < 0.001]. For the MR analysis, we found a causal relationship between insulin use and a higher risk of asthma in both Finn (OR 1.10; p < 0.001) and UK Biobank cohorts (OR 1.18; p < 0.001). Meanwhile, there was no causal association between diabetes and asthma. After multivariable adjustment for diabetes in UKB cohort, the insulin use remained significantly associated with an increased risk of asthma (OR 1.17, p < 0.001). An association between insulin use and an increased risk of asthma was found via the real-world data from the NHANES. In addition, the current study identified a causal effect and provided a genetic evidence of insulin use and asthma. More studies are needed to elucidate the mechanisms underlying the association between insulin use and asthma.


Background
Asthma is a common respiratory disease and a growing public health problem worldwide [1]. The pathogenesis of asthma is complex and still not fully understood [2]. In recent studies, immunological imbalance and metabolic dysfunction have been identified as one of the important contributing factors to asthma [3,4]. In this setting, identifying modifiable factors, possibly related to metabolism and immunity, maybe help in the prevention of asthma.
In recent years, the use of exogenous insulin has been on the rise [5]. A few studies suggest that insulin use may be associated with an increased risk of asthma [6,7]. In addition, studies have reported that insulin modulates immune cell phenotypes through allergic lung inflammation [8].
Further studies have shown that insulin receptors are upregulated during T cell activation [9,10]. Exogenous insulin also promotes naive T cell differentiation into T helper 2 (Th2) cells [11]. Meanwhile, asthma is characterized by airway hyperresponsiveness and the chronic airway inflammation dominated by Th2 cells [8,12,13]. Insulin resistance is common in patients with asthma and is associated with lower lung function and accelerated decline in lung function [14]. Hence, this evidence suggested a possible role of insulin use in the development of asthma, but its true effect on asthma remains unclear.
Asthma is highly heritable and its incidence increases in parallel with the insulin use, which suggests that they may share some common risk factors or mechanisms [15]. Given the observational design of previous studies, the causal relationship between insulin use and asthma cannot be inferred. Mendelian randomization (MR) is a valuable research method that uses single nucleotide polymorphisms (SNPs) as instrumental variables to verify the causal relationship between exposure and outcome. Due to the avoidance of impact from confounding and reverse causation, MR has been widely accepted as a useful approach for examining causal hypotheses in recent epidemiologic studies [16,17]. Previous MR analyses also revealed the causal effects of endocrine and metabolic factors on the development of asthma [17][18][19]. Thus, a deeper knowledge of their shared genes may help understand the causality and directionality in their relationships.
In this study, we first determine the observational association between insulin use and the risk of asthma in a population-based cohort. Then, we further conducted MR analysis to assess the causal relationship between insulin use and the risk of asthma.

Epidemiological Observation Analyses
The National Health and Nutrition Examination Survey (NHANES) is a cross-sectional national survey in which participants in the study were selected using a stratified multistage probability design and therefore a representative sample of the US population, and the analysis included people who participated in NHANES from 2001 to 2018. Data on sociodemographic characteristics, diseases, and insulin use were obtained through questionnaires. Asthma was defined as the "mcq010" information from the "mcq" questionnaire in the NHANES data, which was labelled as "informed by hospital about asthma/current or former asthma". Diabetes was defined as having had a diagnosis made by a physician or the healthcare center in the past. Populations with missing baseline data and unknown asthma and diabetes status were excluded. A total of 85,557 people were included (Fig. 1). We considered the Primary sampling units (PSU) and stratification of the NHANES design. Sampling weights, stratification, and clustering provided in the NHANES dataset were incorporated into the analysis to obtain correct estimates and standard errors (SE) [20]. NHANES was approved by the Centers for Disease Control and Prevention (CDC) National Center for Health Statistics Institutional Review Board. Informed consent was obtained from all participants.

Genome-Wide Association Studies (GWAS) Summary Data
The GWAS datasets were obtained from the OpenGWAS database (https:// gwasm rcieu. ac. uk/) developed by the MRC Integrative Epidemiology Unit [21] at the University of Bristol. The datasets use the GWAS VCF format to store GWAS summary data to ensure alignment with hg19 reference sequences [22]. To ensure homogeneity of the study population and reliability of the results, the data on insulin use and asthma for each analysis was derived from two independent large cohorts. Summary statistics on insulin use were obtained from a case-control GWAS meta-analysis of 462,933 UK Biobank (UKB) study participants (4537 insulin-using and 458,396 insulin-free controls) [23]. Summary-level data for insulin use was also extracted from the FinnGen Consortium (https:// www. finng en. fi/ en/) [24]. Our asthma cohort was also drawn from the FinnGen consortium and the UK Biobank Consortium. Asthma diagnoses are based on questionnaires in clinical or medical centers and captured and validated by professional clinical staff. Asthma was defined and classified by the ICD-10 disease code (ICD-10: J45). Summary-level data on diabetes was obtained from three sources: the UK Biobank Consortium, which includes 18,228 individuals with diabetes and 444,705 without; the European Bioinformatics Institute database, which includes 61,714 individuals with type 1 diabetes and 593,952 without; and the European Bioinformatics Institute database, which includes 6683 individuals with type 2 diabetes and 12,173 without. In the UK Biobank database, diabetes diagnoses are based on questionnaires in clinical or medical centers and captured and validated by professional clinical staff [23]. Detailed phenotypic definitions for type 2 diabetes was reported by Xue et al. [25] while definitions for type 1 diabetes was described by Onengut-Gumuscu et al. [26]. Summary-level data on glycated hemoglobin levels was obtained from the European Bioinformatics Institute database, which includes 146,806 individuals [27]. Detailed phenotypic definitions were described by Chen et al. Details on phenotypes and consortiums are available in Table 1. All queues are searchable by GWAS ID at https:// gwasm rcieu. ac. uk/ [28]. All data are publicly available and can be used without restrictions.

SNPs for Exposures
We extracted genome-wide significant SNPs from each cohort of insulin use and diabetes, respectively. All SNPs with p-values < 5 × 10 -8 were considered significant variants associated with phenotype and included in further analysis. We set r 2 = 0.001 and kb = 10,000 kb when excluding SNPs using linkage disequilibrium (LD) analysis, which means removing SNPs with an r 2 greater than 0.001 in the 10,000 kb range with the most significant SNP [29]. To alleviate weak instrument bias, we calculated F-statistics for each SNP and excluded SNPs with F-statistics less than 10 [30]. When the F statistic is less than 10, we consider the genetic variation used to be a weak instrumental variable, which may bias the results to some extent.

Statistical Analyses
For the NHANES analysis, all analyses used weighted samples and considered the stratification and clustering of the design to derive estimates that were applicable to the U.S. population. Chi-square tests were used to determine whether demographic characteristics differed between asthmatic and non-asthmatic subjects. The test was also used to compare the demographics of those who use insulin with those who don't. Multifactorial logistic regression was used to determine the association between the predictors "insulin use" and "diabetes" and the outcome "asthma". The analysis was further adjusted to account for the potential impact of gender and ethnicity. MR was used to analyze the causal relationship between insulin use and the occurrence of asthma. MR analysis is based on the following three assumptions: 1. Instrumental variables and exposures are strongly correlated; 2. Instrumental variables and confounding factors are not correlated; 3. Instrumental variables are not directly related to outcomes, and their effects on outcomes can only be reflected through exposure (Fig. 2). The Wald ratio used a single instrument variable (IV) to estimate causality of exposure to outcomes. Inverse variance weighting (IVW) can be used to combine ratio estimates for individual instruments in the same way as used for meta-analysis [31]. We estimate the effect of each SNP exposed on the outcome using the Wald ratio, and then employed IVW approach to combine the effect sizes of each SNP. Leave one out sensitivity analysis was used to show the effect of each SNP on the outcome and the Cochrane's Q value was used to assess heterogeneity. Egger regression is a method to detect publication bias in meta-analyses that can also be used to detect directed pleiotropy from different genetic variants [32]. We used the MR-Egger intercept method to detect horizontal pleiotropy [33]. If horizontal pleiotropy is detected, we remove outliers and re-use the inverse variance weighting (IVW) method to combine the effect sizes of each SNP [34] .
MVMR is an extension of MR that allows estimation of the causal effects of multiple exposures on outcomes [35]. We used multivariable Mendelian randomization (MVMR) to analyze multiple exposures. We performed MVMR analyses in the UKB cohort and FinnGen cohort using diabetes and insulin use as exposure, and asthma as an outcome, respectively. The analysis employed an IVW approach, which incorporated different phenotypes into MR analysis as a single exposure.
Both univariable and multivariable MR analyses were performed using the R package "TwoSampleMR", recognizing outliers with the "MR-PRESSO" package [36]. All results are reported as odd ratios [2], with 95% confidence intervals (CI). All statistical analysis and data visualization were performed in R software 4.2.0 (https:// www.r-proje ct. org/).

Epidemiological Observational Analyses
The 85,887 NHANES participants with valid questionnaire information represented 294.4 million noninstitutional residents of the United States. Among the total study population, 12,641 (15%) reported the presence of asthma, and 1812 (2%) self-reported taking insulin during the survey period. The asthma population demonstrated statistically significant differences (p < 0.05) between gender, race, adults or not, insulin use, and diabetes compared to those without asthma ( Table 2). Among those who used insulin, there was no statistically significant difference between insulin use and noninsulin use only by gender (p > 0.05) (Table S1). Meanwhile, we analyzed the risk effects of insulin use and the presence of diabetes on asthma in the populations with asthma (Table 3). In the univariate analysis, we found a statistically significant effect of insulin use on the occurrence of asthma (OR 1.38; 95% CI 1.16-1.64; p < 0.001). In the multivariate analysis, we found a statistically significant difference between insulin use on the occurrence of asthma (OR 1.30; 95% CI 1.07-1.57; p < 0.05), while diabetes was not significantly associated with asthma (p > 0.05). We further corrected for gender and ethnicity in the multivariate analysis, and the specific results are shown in Table 3.

Causal Effects of Insulin Use on Asthma
In the UKB cohort, we obtained 9 SNPs associated with insulin use (Table S2). No SNPs were excluded due to the palindrome ambiguity, resulting in 9 SNPs as instrumental variables in the analysis of asthma. We found that genetic predisposition to insulin use was significantly associated with an increased risk of asthma (OR:1.18; 95% CI 1.10-1.26; p < 0.001). Causal estimates were broadly consistent among different MR models (Fig. 3A). The MR-PRESSO analysis detected 2 outliers and showed that insulin use was associated with a 1.18-fold increase in odds of developing asthma (95% CI 1.14-1.21; p < 0.001) after outlier correction. The Cochran's Q value showed significant heterogeneity between estimates from individual SNPs in the IVW model (Q = 28.60; p < 0.001) but the MR-Egger regression analysis yielded no indication of potential horizontal pleiotropy (p = 0.33). Furthermore, leave-one-out analysis confirmed the pooled IVW estimate was not reliant on any single SNP ( Figure S1).
The same analysis was done in the Finn cohort. we obtained 56 SNPs associated with insulin use (Table S3). Two SNPs (rs12507026, rs3131640) were removed due to palindromes with intermediate allele frequencies, resulting in 54 SNPs as instrumental variables in the analysis of asthma. We found that genetic predisposition to insulin use was significantly associated with risk of asthma (OR 1.10; 95% CI 1.05-1.16; p < 0.001). Causal estimates were broadly consistent among different MR models (Fig. 3B). The MR-PRESSO analysis detected 2 outliers and showed that insulin use was associated with a 1.08-fold increase in odds of developing asthma (OR 1.08; 95% CI 1.03-1.13; p < 0.001) after outlier correction. The Cochran's Q value showed significant heterogeneity between estimates from individual SNPs in the IVW model (Q = 142.47, p < 0.001) but the MR-Egger regression analysis yielded no indication of potential horizontal pleiotropy (p = 0.18). Furthermore, leave-one-out analysis confirmed the pooled IVW estimate was not reliant on any single SNP ( Figure  S2).
The above results suggest that insulin use is associated with an increased risk of asthma, and showed similar results in two independent cohorts. In addition, also consistent with findings from observational studies, which shows that our results are convincible and robust.

Causal Effects of Diabetes on Asthma
In the UKB cohort, after palindromic SNP exclusion and surrogate substitution, we used 61 diabetes SNPs as instrumental variables to analyze the effect of diabetes on asthma (Table S4). Diabetes was not associated with asthma with ORs close to 1 and p-value > 0.05 in different MR models ( Figure S3). Although in IVW analyses, there is suggestive evidence of an association between diabetes and asthma (OR 1.02; 95% CI 1.01-1.05; p = 0.01), the MR-PRESSO analysis detected 3 outliers and diabetes was not associated with asthma (p = 0.118) after outlier correction. Furthermore, we fit an MVMR model with diabetes and insulin use as exposure and asthma as outcomes (Fig. 4A). No association between diabetes and asthma was detected in sensitivity analysis using the IVW model (p > 0.05). Insulin use is still robustly associated with asthma risk (OR 1.17; 95% CI 1.10-1.24; p < 0.001).
The same analysis was done in the Finn cohort. Diabetes was not associated with asthma with ORs close to 1 and p-value > 0.05 in different MR models ( Figure  S3). After outlier correction, the MR-PRESSO analysis detected 3 outliers and diabetes was not associated with asthma (p = 0.123). MVMR model shown that no association between diabetes and asthma in sensitivity analysis using IVW model (p = 0.278) (Fig. 4B). Insulin use is still significantly associated with asthma risk (OR 1.16; 95% CI 1.02-1.30; p = 0.019).
To further analyze the influence of type 1 diabetes (Table S5) and type 2 diabetes (Table S6) on the results, we fitted two other MVMR models. Type 1 diabetes was not shown to be significantly associated with the risk of developing asthma in a model with type 1 diabetes and insulin use as exposure (p = 0.208). Insulin use was shown to be significantly associated with the risk of asthma (OR 1.25; CI 1.01-1.55; p = 0.043), the results are shown in Figure S4. Similar results were seen in models with type 2 diabetes and insulin use as exposure ( Figure S4). These indicate that association between diabetes and asthma may not mediated through insulin use.  To address potential pleiotropic issues, we performed an MR analysis in the UKB cohort. SNPs associated with diabetes (Table S4) and glycated hemoglobin (Table S7) were removed from SNPs associated with insulin use as exposures and asthma as outcomes. The results suggested that insulin use remained a significant factor of increased asthma risk ( Figure S5). This should help us to a certain extent reduce the correlated pleiotropy.

Discussion
This study provided new evidence on the association between insulin use and the risk of asthma. Specifically, our results showed that insulin use was associated with an increased risk of asthma in univariate analyses. Furthermore, insulin use remained significantly associated with asthma after correcting for diabetes in multivariate analyses. The results of MR analyses based on two independent cohorts were consistent with the epidemiological observational findings from NHANES. To our knowledge, this is the first MR study to assess the causal effect of insulin use on the risk of asthma. Overall, the current findings may support a significant association between insulin use and risk of asthma, which was not affected by diabetes.
Clinically, the widespread use of insulin is closely related to diabetes. Several observational studies have found that insulin use is associated with an increased risk of asthma. A retrospective cohort study of a UK primary care database (N = 894,646 adults) suggests that individuals with T2DM are less likely to develop asthma and that insulin increases the risk of asthma attacks (OR 1.25, 95% CI 1.01-1.56) [7]. Other retrospective studies from Taiwan and Netherlands also found that insulin use increases the risk of asthma in people with diabetes [6,37]. The results of these observational studies are the same as ours in this study. These studies, however, did not explore how the risk of asthma varies by insulin does or type, indicating their dose-response relationship remains unclear. Therefore, it is worthwhile to have an ongoing clinical registry of patients on routine insulin use in preparation for the further studies.
On the other hand, animal studies have found that insulin can modulate immune cell phenotypes in airway inflammation in diabetic mice [8,38]. Exogenous insulin induces a switch of T cells to a T helper type 2 (Th2) type response in vitro [11]. Furthermore, insulin treatment significantly  increased airway hyperresponsiveness in Ovalbumininduced asthma mouse model [8]. Similarly, insulin can cause neuronal M2 receptor dysfunction and airway hyperresponsiveness to vagal stimulation [39]. These preclinical results have revealed that insulin appears to have a regulatory effect on immune responses of airway [8,40]. Moreover, exogenous insulin induces increased secretion of collagen and mucus in the airways [38]. Insulin also increases transcription and protein expression of contractile phenotype markers in airway smooth muscle [41]. These effects may eventually cause structural alterations and bronchoconstriction in the airways [39,42]. For a long time, the development of asthma is mainly thought to be attributed to an overactivated immune response of Th2 cells [43,44]. Meanwhile, the major characteristics of asthma included airway inflammation, hyperresponsiveness, and remodeling. Taken together, this evidence suggested effects of insulin on the airway may promote or aggravate the pathophysiological changes in the development of asthma.
It has been reported that diabetes and asthma may have an antagonistic relationship. Both Type 1 and Type 2 diabetes appear to show a protective effect against asthma. Specifically, low levels of endogenous insulin in patients with advanced type 2 diabetes may contribute to the reduced risk of asthma [45]. Similarly, some studies have found an inverse association between asthma symptoms and type 1 diabetes. These effects may be partly attributed to the be low or absent endogenous secretion of insulin, which also accompanied by a genetic basis [15,46] .
Previous studies have shown that type 1 diabetes is thought to be mediated by Th1 cells [47], whereas asthma is mainly mediated by Th2 cells [12]. The Th1/Th2 imbalance hypothesis was used to explain this relationship [48,49]. This theory has been explained that the expansion of Th1 cells in individuals with T1DM leads to a decrease in Th2 cells, thereby preventing the development of atopic diseases such as asthma. Consistent with this theory, clinical asthma appears to be less severe in the presence of diabetes in many clinical studies [50,51]. Several observational studies have shown reductions in asthma and allergic respiratory symptoms in patients with T1DM [52,53]. This protection was also seen in their unaffected siblings, suggesting that the protection may be due to their shared genetic background [15,50]. Based on the current evidence, it could be speculated that a relative lack of insulin would reduce the risk of asthma in patients with type 1 diabetes. In turn, asthma may manifest itself clinically due to continuous use of the exogenous insulin. These might also explain why diabetes was not associated with asthma in the multivariate model that included insulin use.
For patients who have been using insulin for a long time, attention should be paid to the occurrence of asthma. For high-risk groups of asthma, we need early identification and early intervention. Several studies have found that metformin may reduce the risk of asthma in people with diabetes [7,37,54]. Studies have shown that the use of metformin in asthma patients can significantly reduce asthma exacerbations and hospitalizations [54]. Sulfonylureas showed to play an important protective role in asthma development by relaxing airway muscles in mice [55]. Glyburide inhibits cytokine-mediated eosinophil survival and superoxide production, suggesting it may be a potential treatment for asthma [56]. Notably, there was no relationship between clinical features of diabetes with risk of asthma in these studies, indicating these findings may reflect the protective effect on asthma of the medications themselves. For diabetics with high risk of asthma, the use of anti-diabetic medications and the potential effect of insulin may need to be carefully considered. Overall, the development of individual treatment strategies for diabetics with high risk for asthma is worthy of further exploring.
Previous MR studies have also investigated the causal relationship between several endocrine/metabolic factors and risk of asthma. In recent years, many studies have confirmed that obesity-related traits are significant genetic predisposition factors for asthma with causal evidence [57,58]. In addition, the complex association between obesity and asthma involves various pathways and cytokines [59]. Of these, the common pathways such as TNF and IL-6 inflammatory pathways in both obesity and asthma may also play an important role in mediating hyperinsulinemia and insulin resistance [59][60][61]. The association between obesity and severe asthma may be mediated by low-grade systemic inflammation, where fat cells and inflammatory macrophages secrete various pro-inflammatory cytokines, including IL-6, which is linked to the development of insulin resistance and other conditions [62]. The study by Michael C Peters et al., has demonstrated a possible correlation between insulin resistance, reduced lung function, and accelerated loss of lung function in asthmatic patients [14]. These evidences suggest that there may be shared genetic relationships and underlying mechanisms among insulin, obesity, and asthma. Hence, it is worth to further study the effects of improved management of insulin resistance and obesity on individuals with asthma, which would require conducting more clinical trials in patients with both asthma and insulin resistance. Moreover, levels of vitamin D and adiponectin have been considered useful predictors and may have a regulative effect on airway inflammation and hyperresponsiveness for asthma [63,64]. Interestingly, vitamin D and adiponectin were associated with insulin secretion and sensitivity, respectively [21,65] . These also suggest the possibility of co-regulation of asthma between different metabolic substances, although this needs to be further elucidated in the future [66]. Overall, these findings may provide new insights into the genetic role of endocrine/metabolic factors in the predisposition and development of asthma.
Our study has several major advantages. First, in traditional observational studies on the effect of insulin use on asthma, the results are prone to bias due to reverse causality, as obesity may be present in diabetes and thus contribute to asthma. We used MR study design, thus avoiding reverse causality and minimizing residual confounding. Moreover, two highly representative and independent cohorts from FinnGen and UKB were used for our analyses. These consistent results supported a positive association between insulin use and asthma. The estimated effects derived from MR methods performed in both cohorts were similar to those derived from epidemiological observational analyses. We also have attempted to remove instrumental variants associated with both diabetes and glycosylated hemoglobin from the MR analysis of insulin use, and the results remain stable. The F-statistics for genetic variants were all greater than 10 in the MR analyses and different sensitivity analyses were applied, giving sufficient strength to the SNPs. Despite this, the potential effect of diabetes and related conditions may not be completely ruled out.
Our study has several limitations. First, the study population was limited to Europe and the Americas, which may limit generalizability of the results. In future studies, it's worth to expand our analysis to the other population to overcome this difficulty. Second, insulin use may reflect the presence of more severe diabetes, and it is important to consider the broader context of the relationship between insulin, diabetes, and other co-occurring conditions. It is reassuring that sensitivity analyses based on various assumptions failed to detect any horizontal pleiotropy, suggesting robust effect estimates in different MR models. Additionally, future work will be needed to confirm our results when larger GWAS statistics become available. Finally, the current data make it difficult to further clarify the complexity of the relationship through subgroup and population-specific analysis. It's difficult to completely rule out the potential effect of severe diabetes and related conditions. The absence of individual-level data and population stratification should be considered, as they may help avoid spurious causality by excluding confounding. Further research could clarify this question by exploring differences in the genetic effects of insulin use on asthma based on uniform diabetes levels and related conditions.

Conclusions
Our study further reveals the causal relationship between genetic susceptibility to insulin use and increased asthma risk. However, more studies are still needed to elucidate the mechanisms underlying the association between insulin use and asthma.