Mendelian randomisation of eosinophils and other cell types in relation to lung function and disease

Abstract

Rationale: Eosinophils are associated with airway inflammation in respiratory disease. Eosinophil production and survival is controlled partly by interleukin-5: anti-interleukin-5 agents reduce asthma and response correlates with baseline eosinophil counts. However, whether raised eosinophils are causally related to chronic obstructive pulmonary disease (COPD) and other respiratory phenotypes is not well understood.

Objectives: We investigated causality between eosinophils and: lung function, acute exacerbations of COPD, asthma-COPD overlap (ACO), moderate-to-severe asthma and respiratory infections.

Methods: We performed Mendelian randomisation (MR) using 151 variants from genome-wide association studies of blood eosinophils in UK Biobank/INTERVAL, and respiratory traits in UK Biobank/SpiroMeta, using methods relying on different assumptions for validity. We performed multivariable analyses using eight cell types where there was possible evidence of causation by eosinophils.

Measurements and main results: Causal estimates derived from individual variants were highly heterogeneous, which may arise from pleiotropy. The average effect of raising eosinophils was to increase risk of ACO (weighted median OR per SD eosinophils, 1.44 (95%CI 1.19 to 1.74)), and moderate-severe asthma (weighted median OR 1.50 (95%CI 1.23 to 1.83)), and to reduce forced expiratory volume in 1 s (FEV₁)/forced vital capacity (FVC) and FEV₁ (weighted median estimator, SD FEV₁/FVC: -0.054 (95% CI -0.078 to -0.029), effect only prominent in individuals with asthma).

Conclusions: Broad consistency across MR methods may suggest causation by eosinophils (although of uncertain magnitude), yet heterogeneity necessitates caution: other important mechanisms may be responsible for the impairment of respiratory health by these eosinophil-raising variants. These results could suggest that anti-IL5 agents (designed to lower eosinophils) may be valuable in treating other respiratory conditions, including people with overlapping features of asthma and COPD.

Keywords: Asthma Epidemiology; Asthma Genetics; Asthma Mechanisms; COPD epidemiology; COPD exacerbations mechanisms; Eosinophil Biology; Respiratory Infection.

Figure 1

Mendelian randomisation (MR): core assumptions Mendelian randomisation may be used to test for causality between an exposure (eg, eosinophils) and outcome (eg, a respiratory outcome such as FEV₁/FVC), if the following core assumptions hold (see 1–3 on the figure): (1) the genetic variation (single nucleotide polymorphisms in this work) used as instrumental variables are associated with the exposure of interest; the genetic variants are not associated with unobserved confounders of the exposure-outcome association (straight dashed arrow). Genetic variants are allocated randomly at conception (Mendel’s law of independent assortment) and so typically should not be associated with these confounding variables; association between the genetic variants and the outcome is via the exposure, and not via an alternate pathway (ie, there is no ‘horizontal pleiotropy’, see curved dashed arrow). While difficult to verify, reassurance that this assumption holds can be provided using biological knowledge of how the SNP functions, and by checking whether multiple MR methods, each relying on different assumptions for validity, give consistent results (known as triangulation). FEV₁, forced expiratory volume in 1 s, FVC, forced vital capacity; SNP, single-nucleotide polymorphisms.

Figure 2

Selection of SNPs for univariable MR analyses of eosinophils and respiratory outcomes flow chart describing the analysis workflow for initial MR analyses of eosinophils. Of 209 SNPs associated with eosinophil count, 167 were available in lung function GWASs (missingness is due to some SpiroMeta studies not being imputed to the HRC panel). LD proxies at R² >0.8 were retrieved for 24/42 missing variants. Of the resulting 191 SNPs, 188 were successfully harmonised between the SNP-eosinophil and SNP-lung function data sets, and 151* remained after LD clumping at an R² threshold of 0.01. These 151 SNPs were used in analyses. *One SNP, rs9974367, was missing in the moderate-severe asthma GWAS. AECOPD, acute exacerbation of COPD; ACO, asthma COPD overlap; COPD, chronic obstructive pulmonary disease; FEV₁, forced expiratory volume in 1 s, FVC, forced vital capacity; GWAS, genome-wide association study; MR, Mendelian randomisation; SNPs, single-nucleotide polymorphisms.

Figure 3

MR analyses of eosinophils (exposure) on three quantitative lung function traits (top) and four respiratory disease phenotypes (bottom), using 151 eosinophil-associated SNPs top: results of MR analyses of eosinophil counts (exposure) on three quantitative lung function traits (outcome), FEV₁, FVC and FEV₁/FVC. A forest plot of three estimates for each traits is shown (IVW, MR Egger, weighted median), along with the maximum sample size in the outcome GWAS (N), the effect size in SD change in outcome trait per SD increase eosinophil count, and 95% CI, values for Cochran’s Q statistic (Q) and the associated df (Q_df), and the p value for the MR Egger intercept (Intercept_P). Boxes of the forest plot represent effect sizes, whiskers are 95% CIs. Bottom: results of MR analyses of eosinophil counts (exposure) on four respiratory disease phenotypes (outcome), moderate-to-severe asthma, acute exacerbations of COPD (AECOPD), asthma-COPD overlap (ACO), and respiratory infection (Resp. IX). A forest plot of three estimates for each traits is shown (IVW, MR Egger, weighted median), along with sample size in the outcome GWAS for cases and controls, respectively (N), the effect size as OR per SD eosinophil count, and 95% CI, values for Cochran’s Q statistic (Q) and the associated df (Q_df), and the p value for the Mr Egger intercept (Intercept_P). Boxes of the forest plot represent ORs, whiskers are 95% CIs. Nb only 150/151 of the eosinophil SNPs were available in the moderate-to-severe asthma GWAS. COPD, chronic obstructive pulmonary disease; FEV₁, forced expiratory volume in 1 s, FVC, forced vital capacity; GWAS, genome-wide association study; IVW, inverse-variance weighted; MR, Mendelian randomisation; SNPs, single-nucleotide polymorphisms.

Figure 4

Multivariable MR analyses of eight cell types and forced expiratory volume in 1 s (FEV₁) and FEV₁/forced vital capacity (FVC) forest plot showing multivariable MR estimating the causal effect of multiple cell types on quantitative lung function outcomes, after conditioning on the effects of the SNPs on other cell types. Models were run for each of FEV₁ and the ratio of FEV₁ to FVC separately, but effect sizes are shown next to one another for comparison. Effect sizes (beta, 95% CI) are in SD change in lung function outcome per SD cell count (adjusted for the effects of other cell types). Points of the forest plot represent effect size estimate; whiskers are 95% CIs. MR, Mendelian randomisation.

Figure 5

Multivariable MR analyses of eight cell types and two respiratory disease outcomes, ACO and asthma forest plot showing multivariable MR estimating the causal effect of multiple cell types on respiratory disease outcomes, after conditioning on the effects of the SNPs on other cell types. Models were run for each of ACO and asthma separately, but effect sizes are shown next to one another for comparison. ORs (95% CI) are per SD cell count (adjusted for the effects of other cell types). Points of the forest plot represent ORs; whiskers are 95% CIs. ACO, asthma-COPD overlap; MR, Mendelian randomisation; SNP, single-nucleotide polymorphisms.