Statistics

Protein and gene expression data was transformed where appropriate to more closely approximate a normal distribution for genetic analysis. Specifically, normalized blood paxgene expression data was log₂ transformed; normalized bronchial brush expression data was log₂ transformed and then square root transformed. Blood sST2 protein levels were natural log (ln) transformed. Genetic association between IL1RL1 SNPs and gene and protein expression levels was tested by linear regression models coding the SNPs in an additive fashion (e.g., 0 = GG, 1 = AG, 2 = AA). UCSF ATB analyses of only white non-Hispanic subjects included asthma status, inhaled steroid use, age, and sex as covariates. The analysis of all subjects included the formerly stated covariates with the addition of 5 dichotomous variables for race/ethnicity based on self-identification, namely (i) white non-Hispanic, (ii) black non-Hispanic, (iii) Asian non-Hispanic and Native Hawaiian and Other Pacific Islanders, (iv) Hispanic any race, (v) and other, which included subjects reporting as mixed race non-Hispanic, white with no ethnicity, or non-Hispanic with American Indian/Alaska native race.

VST normalized gene expression values, from airway epithelial cell cultures, were used to perform probabilistic estimation of expression residuals (PEER) normalization (45) and to generate PEER residuals corrected for plink (version 1.07) (46) determined sex, genome-wide genetic data determined genetic ancestry principle components (pc1, pc2, and pc3), and 64 hidden PEER factors (corresponding to ~25% of sample size). Similarly, GTEx lung VST-normalized gene expression data was used to perform PEER normalization and to generate PEER residuals corrected for age, sex, genome-wide genetic data–determined genetic ancestry principle components (pc1 and pc2), and 59 hidden PEER factors (corresponding to ~25% of sample size) for the sST2 transcript. Both the GTEx and cultured airway epithelial cell sST2 gene expression PEER residuals were analyzed for association with IL1RL1 SNPs by linear regression analysis using an additive genetic model. The airway epithelial bank IL-13–stimulated sST2 peer residuals were ln transformed for analysis. The untransformed control and square root–transformed IL-13–stimulated sST2 protein levels were also tested for association with IL1RL1 SNPs by linear regression analysis using an additive genetic model including sex and genetic principle components as covariates. Logistic regression was used to test the association between the rs1420101 and rs11685480 genotype and type 2-high vs. type 2-low status, including age, sex, steroid use, and racial/ethnic group (as defined above) as covariates. Between-group differences were assessed using a 2-tailed paired t test (cultured epithelial cell protein and gene expression unstimulated vs. IL-13–stimulated; distal vs. proximal promoter gene expression; EC50 IL-33 in whole blood vs. PBMC); 2-tailed unpaired t test (plasma sST2 protein, sST2 gene expression, or sST2 bronchial brush gene expression in healthy vs. asthma); ordinary 1-way ANOVA with Tukey’s multiple correction test (sST2 bronchial brush gene expression in healthy vs. type 2-high vs. type 2-low asthmatics; or ordinary 1-way ANOVA with post-test for linear trend (plasma sST2 protein in subjects carrying 0–5 risk alleles for rs1420101 or rs11685480). Linear regression was used to assess the relationship between whole blood EC50 for IL-33 and plasma sST2 protein LC-MS, as well as the relationship between plasma sST2 measured with LC-MS and critical diagnostics ELISA. A P value less than 0.05 was considered statistically significant. All statistical analyses were either performed in STATA/SE v11.0 or Graphpad Prism v6.0d.