Lung cancer.

The MR-IVW results indicate strong evidence to suggest that increased NMR and increased smoking heaviness both increase the risk of developing lung cancer among ever smokers (odds ratio [OR] = 1.20, 95% confidence interval [CI] 1.15 to 1.25; OR = 4.00, 95% CI 3.50 to 4.57 respectively). However, the MVMR-IVW results do not provide clear evidence to suggest an effect of NMR on lung cancer risk (OR = 1.01, 95% CI 0.94 to 1.08), and indicate that only increased smoking heaviness increased lung cancer risk (OR = 2.94, 95% CI 2.38 to 3.62). Although there is considerable evidence of heterogeneity and potential pleiotropy in the smoking heaviness analyses among ever smokers, these results are supported by the MR-Egger results. Furthermore, there is no clear evidence of pleiotropy or bias due to population stratification indicated by the results among never smokers.

Coronary heart disease.

The MR-IVW results provide no clear evidence that NMR affects CHD risk (OR = 1.02, 95% CI 0.99 to 1.04). This is supported by the MR-Egger results, but the weighted median and weighted mode results suggested there may be a weak effect (OR = 1.03, 95% CI 1.01 to 1.05, OR = 1.03, 95% CI 1.01 to 1.05 respectively). The results provide some evidence that increased smoking heaviness increases CHD risk (OR = 1.14, 95% CI 1.01 to 1.29) among ever smokers. The weighted median and weighted mode results support this finding (OR = 1.14, 95% CI 1.02 to 1.29, OR = 1.14, 95% CI 1.02 to 1.27 respectively); however, the MR-Egger analysis does not support this (OR = 0.95, 95% CI 0.79 to 1.14), and there is some evidence of heterogeneity indicated by the Q-statistics and the Egger intercept indicated weak directional pleiotropy which is supported by evidence of a protective effect among never smokers. The MVMR-IVW results provide no clear evidence of an effect of NMR on CHD risk (OR = 1.00, 95% CI 0.98 to 1.03) but some evidence that increased smoking heaviness increases CHD risk (OR = 1.32, 95% CI 1.08 to 1.62). As with the univariable results, the MVMR-Egger analyses do not support this (OR = 0.94, 95% CI 0.68 to 1.28) and there was evidence of heterogeneity and potential directional pleiotropy and some evidence of horizontal pleiotropy or bias due to population stratification in the analysis among never smokers.

Chronic obstructive pulmonary disease.

The MR-IVW results indicate that increased NMR and smoking heaviness increase the risk of developing COPD among ever smokers (OR = 1.16, 95% CI 1.07 to 1.24, OR = 6.57, 95% CI 4.68 to 9.20 respectively). However, the MVMR-IVW results indicate no clear effect of NMR on COPD risk (OR = 1.03, 95% CI 0.95 to 1.12) but suggest that increased smoking heaviness increases COPD risk among ever smokers (OR = 7.24, 95% CI 3.99 to 13.15). These results are supported by the MR sensitivity analyses and MVMR-Egger results and there is no clear evidence of heterogeneity or directional pleiotropy or horizontal pleiotropy or bias due to population stratification among never smokers (where precise null effects were observed).

Forced expiratory volume in 1 second.

The results relating to FEV-1 are presented in the text as changes in millilitres (mL) of expiration per standard deviation increase in the exposure. The MR-IVW results indicate that increased NMR and smoking heaviness both decrease FEV-1 among ever smokers (Fig 4, b = -13.35, 95% CI -18.38 to -8.33, b = -185.49, 95% CI -220.39 to -150.60 respectively). However, there is no clear evidence of an effect of NMR on FEV-1 in the MVMR-IVW analysis (b = 3.82, 95% CI -12.63 to 4.98), whereas there is evidence to suggest increased smoking heaviness decreased FEV-1 among ever smokers (b = -176.78, 95% CI -239.15 to -114.41). These results are supported by the MR sensitivity analyses and MVMR-Egger results and there is no evidence of directional pleiotropy in the analyses, but there is evidence of heterogeneity in all analyses except for the NMR MR analyses. However, there is evidence of horizontal pleiotropy or bias due to population stratification in the NMR MR analyses among never smokers.

Among current smokers (Fig 5), the MR-IVW results indicate that increased NMR and smoking heaviness decreased FEV-1 (b = -33.33, 95% CI -41.76 to -24.90, b = -374.28, 95% CI -428.21 to -320.35 respectively). However, the MVMR-IVW results indicate a protective effect of nicotine exposure (i.e., lower nicotine exposure, due to higher NMR, lowers lung function) when smoking heaviness is accounted for among current smokers (NMR b = -17.77, 95% CI -29.92 to -5.62). The MVMR-IVW analyses support the MR-IVW results for smoking heaviness (b = -272.36, 95% CI -359.31 to -185.42). These results are supported by the MR sensitivity analyses and MVMR-Egger analyses. There is some evidence of heterogeneity in all analyses except the NMR MR analyses, directional pleiotropy in the MR analysis of smoking heaviness, and horizontal pleiotropy or bias due to population stratification in the NMR MR analyses among never smokers.

Forced vital capacity.

The results relating to FVC are presented in the text as changes in mL of capacity per standard deviation increase in the exposure. The MR-IVW results indicate that increased NMR and smoking heaviness decreased FVC among ever smokers (Fig 4, b = -14.34, 95% CI -19.42 to -9.26, b = -129.63, 95% CI -174.27 to -85.00 respectively). However, there is no clear evidence of an effect of NMR on FVC in the MVMR-IVW analysis (b = -6.32, 95% CI -17.41 to 4.77), whereas there is evidence to suggest increased smoking heaviness decreased FVC among ever smokers, after accounting for NMR (b = -149.55, 95% CI -228.07 to -71.03). These results are supported by the MR sensitivity analyses and MVMR-Egger results, although there is less clear evidence of an effect in the MVMR-Egger results for smoking heaviness than in the MR-Egger results. There is some evidence of heterogeneity in all analyses except the NMR MR analyses, but no clear evidence of directional pleiotropy. There is weak evidence to suggest potential horizontal pleiotropy / bias due to population stratification in the NMR MR analyses and the smoking heaviness MR-IVW analysis among never smokers.

Among current smokers (Fig 5), the MR-IVW results suggest that increased NMR and smoking heaviness decreased FVC (b = -24.50, 95% CI -34.61 to -14.39, b = -246.13, 95% CI -304.36 to -187.89 respectively). MVMR-IVW results support the evidence for an effect of smoking cessation on FVC but indicate weak evidence of a protective effect of nicotine exposure when smoking heaviness is accounted for (NMR b = -13.50, 95% CI -27.39 to 0.39). The MR median and mode weighted analyses support these findings, and the MR-Egger and MVMR-Egger results are in the same direction but provide weaker support for the NMR results. There is evidence in the MVMR-IVW analysis to suggest that increased smoking heaviness decreases FVC (b -195.31, 95% CI -294.72 to -95.90), which is supported by the MVMR-Egger results. There is some evidence of heterogeneity in all analyses except the NMR MR analyses, but no clear evidence of directional pleiotropy. There is weak evidence to suggest potential horizontal pleiotropy or bias due to population stratification in the NMR MR analyses and the smoking heaviness MR-IVW analysis among never smokers.

Heart rate.

The results relating to heart rate are presented in the text as changes in beats per minute per standard deviation increase in the exposure. Among ever smokers (Fig 4), the MR-IVW results suggest that increased NMR does not affect heart rate (b = 0.06, 95% CI -0.04 to 0.16), but increased smoking heaviness does increase heart rate (b = 1.61, 95% CI 1.12 to 2.10). The MVMR-IVW results support the results for both NMR and smoking heaviness (b = -0.04, 95% CI -0.16 to 0.09, b = 1.63, 95% CI 0.73 to 2.54 respectively). The MR sensitivity analyses and MVMR-Egger results also support these findings. There is some evidence of heterogeneity in all analyses, but no clear evidence of directional pleiotropy. There is weak evidence to suggest potential horizontal pleiotropy or bias due to population stratification in the MR-IVW NMR, MR-IVW and MR-Egger smoking heaviness analyses among never smokers.

Among current smokers (Fig 5), the MR-IVW results indicate no clear effect of NMR on heart rate (b = -0.05, 95% CI -0.20 to 0.09) but indicate that increased smoking heaviness increases heart rate (b = 4.58, 95% CI 3.65 to 5.51). However, the MVMR-IVW results suggest that decreased NMR when accounting for smoking heaviness (i.e., increased nicotine exposure per cigarette) increases heart rate among current smokers (b = -0.30, 95% CI -0.50 to -0.10) and increased smoking heaviness accounting for nicotine exposure increases heart rate (b = 4.22, 95% CI 2.77 to 5.68). The MR sensitivity and MVMR-Egger results support the MR-IVW and MVMR-IVW findings. There is some evidence of heterogeneity in all analyses except the MR NMR analyses, but no clear evidence of directional pleiotropy. There is weak evidence to suggest potential horizontal pleiotropy / bias due to population stratification in the MR-IVW NMR, MR-IVW and MR-Egger smoking heaviness analyses among never smokers.

Smoking heaviness.

Liu and colleagues [14] report summary-level statistics from a GWAS meta-analysis of smoking heaviness (measured by standard deviation change in CPD categories, equivalent to 2–3 additional cigarettes per day) among 337,334 smokers of European ancestry in the GWAS and Sequencing Consortium of Alcohol and Nicotine (GSCAN). Smokers were defined as having ever or currently smoked and smoking heaviness (at the time of smoking) was binned into five categories or included as pre-defined bins from the original study (see supplementary material in Liu and colleagues [14]). Analyses were adjusted for age, age-squared, sex, genetic principal components, and smoking status (current versus former). SNPs associated with CPD explained 4% of variance in CPD. SNPs were reported as independent if they explain additional variance in conditional analyses using a partial correlation-based score statistic [46]. The reported data are available at: https://doi.org/10.13020/3b1n-ff32. To avoid sample overlap, two versions of the GSCAN data were used in our analyses. Where we analysed UK Biobank outcomes, we used a version of the GSCAN data with UK Biobank removed to avoid sample overlap. To maximise sample size, we used GSCAN data including UK Biobank in analyses which did not include UK Biobank in the outcome GWAS. In all analyses, 23andMe were removed from the GSCAN data due to data sharing restrictions.

Nicotine.

Buchwald and colleagues [15] report summary-level statistics from a GWAS meta-analysis of the standard deviation change in NMR among 5,185 current smokers of European descent with cotinine levels ≥10 ng/ml (indicating recent smoking). NMR is a ratio of 3’hydroxycotinine/cotinine which indicates how quickly a person metabolises and clears nicotine. Analyses were adjusted for population substructure, age, sex, BMI, alcohol use, and birth year. The SNPs associated with NMR explained ~38% of the variance in NMR. SNPs were reported as independent if they explained additional variance in a step-wise conditional regression using genome-wide complex trait analysis [47]. Information relating to the nicotine measures used in the supplemental analysis (using cotinine and cotinine plus 3HC data as a proxy for nicotine exposure) can be found S2 Note.

Lung cancer.

McKay and colleagues [43] report summary-level statistics from a case-control GWAS meta-analysis of overall lung cancer risk (including adenocarcinoma, squamous cell carcinoma, and small cell carcinoma) among 50,046 individuals of European descent using the International Lung Cancer Consortium (ILCCO). Analyses were adjusted for age, gender, country (if applicable) and significant principal components. We received the data via request to the authors and we additionally received the data stratified by smoking status (ever versus never smoking). Definitions of ever and never smoking differed by cohort included in the meta-GWAS. Details of the definitions can be found in McKay and colleagues’ [43] supplementary note.

Non-cancer smoking-related health outcomes.

We conducted GWAS of non-cancer health outcomes using data from UK Biobank, a population-based health research resource consisting of approximately 500,000 people, aged between 38 years and 73 years, who were recruited between the years 2006 and 2010 from across the UK [48]. UK Biobank has a particular focus on identifying determinants of human diseases in middle-aged and older individuals who provided a wide range of health information (data available at www.ukbiobank.ac.uk). A full description of the study design, participants, and quality control (QC) methods have been described previously [48,49].

The full data release contains the cohort of successfully genotyped samples (n = 488,377). Further information about genotyping and imputation can be found in S3 Note. Individuals with mismatched sex were excluded and the analyses were restricted to those of ‘European’ ancestry using an in-house k-cluster means method. The GWAS were conducted using the linear mixed model (LMM) association method as implemented in BOLT-LMM (v2.3) [50], adjusting for genotype array, age and sex [44]. BOLT-LMM association statistics are on the linear scale, so test statistics (betas and their corresponding standard errors) relating to binary phenotypes were transformed to log odds ratios and their corresponding 95% confidence intervals (95% CI) on the liability scale using a Taylor transformation expansion series in which we divided the statistics by μ(1 – μ), whereby μ is the case prevalence [51]. Each outcome was stratified by self-reported participant smoking status (ever [further stratified by current or former] and never), resulting in four GWAS per outcome. Smoking status was categorised as never, previous, and current smoking in UK Biobank (field ID 20116). From this variable, we derived an ‘ever smokers’ category which was defined as currently or having previously smoked occasionally, most days or daily (i.e., having smoked more than just once or twice). Current smoking was defined as currently smoking occasionally, most days or daily. Former smoking was defined as not currently smoking but having previously smoked occasionally, most days or daily (i.e., more than just once or twice). Those who had tried smoking once or twice or who had never smoked were categorised as never smokers.

We identified COPD cases as participants who self-reported a doctor’s diagnosis of COPD. Lung function (FEV-1 and FVC, in litres) was measured using a Vitalograph spirometer. CHD diagnosis was determined using linked hospital admission data (ICD codes relating to Ischemic Heart Disease). Further information regarding each health outcome (including UK Biobank field IDs) can be found in the S4 Note.

Selection of genetic variants.

We selected conditionally independent (at the genome-wide significant level, p < 5 × 10⁻⁸) genetic variants identified in the conditional GWAS analyses of NMR and CPD for inclusion in the analysis [14,15]; 55 SNPs associated with smoking heaviness were identified as conditionally independent [14]; 7 SNPs associated with the NMR were identified as conditionally independent. Methods of determining conditional independence varied between the two exposure GWAS, whereby GSCAN reported SNPs as independent if they explain additional variance in conditional analyses using a partial correlation-based score statistic [14] and Buchwald and colleagues reported SNPs as conditionally independent if they explained additional variance in a step-wise conditional regression using genome-wide complex trait analysis [15].

After removing SNPs which were not available in the outcome GWAS (where no available proxy could be identified), harmonising, and clumping the combined exposure datasets with the outcome dataset, the number of SNPs in each analysis varied. S9–S14 Tables detail which SNPs were included and excluded (with reasons) and the total number of SNPs in each analysis. S5 Note gives further details of the proxy search and clumping methods.

We tested instrument strength and validity using the two-sample conditional F-statistic for MVMR and Cochran Q statistic [13,41]. The conditional F-statistic for MVMR indicates instrument strength of each exposure when accounting for the prediction of other exposures in the model (i.e., whether the SNPs jointly predict smoking heaviness after predicting the NMR, and vice versa) [41]. The standard critical values for the F-statistic and Q-statistic can be approximated so F-statistic should be greater than 10 to indicate sufficient instrument strength and Q estimates should be less than the number of SNPs included in the model to indicate no excessive heterogeneity [52,53].

Univariable Mendelian randomisation.

For comparison with the MVMR analysis, we considered the total effect of both the NMR and smoking heaviness on each health outcome using univariable MR. We used four complementary MR methods MR-IVW, MR-Egger, weighted median, weighted mode). Details of the univariable MR analysis methods can be found in S6 Note.

Multivariable Mendelian randomisation.

We explored the direct effects of both the NMR and smoking heaviness on health outcomes using MVMR. We used summary data from Buchwald and colleagues’ [15], GSCAN [14] and ILCCO [43], and UK Biobank GWAS pipeline results. We repeated these analyses using two complementary methods–MVMR-IVW and MVMR-Egger [13,29].

Given that the GWAS of smoking heaviness was restricted to ever smokers and the GWAS of the NMR was restricted to current smokers, the analyses were restricted to ever and current smokers as the summary statistics are not applicable to never smokers. However, the ILCCO lung cancer data have only been stratified by ever and never smoking status; therefore, analyses restricted to current smokers were not possible when exploring lung cancer incidence. In supplementary analyses, we additionally stratified the analysis by never smokers to explore potential horizontal pleiotropy–effects observed among never smokers could indicate horizontally pleiotropic effects (i.e., the included SNPs influencing the outcome directly, or via another phenotype, but not through the measured exposure), misreporting of smoking status, or residual population stratification. Horizontally pleiotropic genetic variants are not valid instruments in MR analyses and their inclusion would result in a violation of the exclusion restriction assumption (further details on the assumptions of MR can be found in S7 Note). Pleiotropy robust methods (e.g., MR-Egger) are robust to horizontal pleiotropy under assumptions of the form that pleiotropy takes. For all outcomes except lung cancer (where data are unavailable), we also included supplementary results restricted to former smokers to explore whether any health effects found among current smokers may be recoverable (i.e., not present among former smokers who have recovered following smoking cessation).