Shuai Xu

Supplemental Figure 1: Flowchart displaying how we arrived at the final analytic sample. Among the women recruited to the study, 127 were missing breastfeeding duration and 32 were missing values for all of the MBD measures. An additional 18 participants were missing values for one of the confounding variables; 11 women were missing age at menarche, 1 woman was missing BMI and 6 women were missing information about family history of breast cancer.

Supplemental Table 1 shows results from IPW analysis.

Background: Genome‐wide association studies (GWAS) have identified over 80 susceptibility loci for lung cancer risk. However, the genes underlying these associations remain largely unknown. Methods: We conducted a large transcriptome‐wide association study (TWAS) to identify lung cancer susceptibility genes. We leveraged gene expression data from lungs and 48 other tissue types and whole‐genome sequencing data from up to 706 samples of European ancestry in the GTEx (version 8) to build lung‐tissue and joint‐tissue gene expression prediction models. These models were applied to GWAS data, including 29,266 lung cancer cases and 56,450 controls, to assess the associations of genetically predicted gene expression levels with lung cancer risk. Results: A total of 8624 genes were successfully built for single‐tissue models, and 11,341 genes for joint‐tissue models (12,133 unique genes altogether). Among 40 genes whose expression levels were associated with the risk of lung cancer at a Bonferroni‐corrected significance level, ZKSCAN4 was located more than 2 Mb away from the GWAS‐identified variants linked to lung cancer. Among the remaining 39 genes within 2 Mb of GWAS‐identified variants, seven genes were independent of these. Among 53 genes associated with the risk of lung cancer subtypes, 13 genes were beyond 2 Mb of GWAS‐identified variants, and four genes were independent of the GWAS‐identified variants within 2 Mb regions. Conclusion: Our TWAS identified over 50 candidate susceptibility genes for lung cancer, providing new insights into lung cancer genetics.

Supplementary Figure 2 shows spearman correlation between the healthy lifestyle score and metabolic signature.

Supplementary Table 3 shows selected baseline characteristics of study participants by training and testing sets in the UK Biobank.

Supplementary Table 4 shows selected baseline characteristics of study participants in the Rotterdam Study.

Supplementary Table 1 shows ACS guidelines for cancer prevention and lifestyle components of the HLS in the UK Biobank.

Supplementary Table 5 shows association of the metabolic signature of healthy lifestyles with CRC risk in the Rotterdam Study.

Supplementary Figure 3 shows correlation matrix for the top (20 positive and 20 negative) contributing metabolic biomarkers to the metabolic signature.