Skip to main content
NCBI home page
Wiley Open Access Collection logoLink to Wiley Open Access Collection
. 2024 Aug 4;49(1):e22583. doi: 10.1002/gepi.22583

Exploring pleiotropy in Mendelian randomisation analyses: What are genetic variants associated with ‘cigarette smoking initiation’ really capturing?

Zoe E Reed 1,2,✉, Robyn E Wootton 1,2,3, Jasmine N Khouja 1,2, Tom G Richardson 2, Eleanor Sanderson 2, George Davey Smith 2, Marcus R Munafò 1,2,4
PMCID: PMC7616876  EMSID: EMS201281  PMID: 39099143

Abstract

Genetic variants used as instruments for exposures in Mendelian randomisation (MR) analyses may have horizontal pleiotropic effects (i.e., influence outcomes via pathways other than through the exposure), which can undermine the validity of results. We examined the extent of this using smoking behaviours as an example. We first ran a phenome‐wide association study in UK Biobank, using a smoking initiation genetic instrument. From the most strongly associated phenotypes, we selected those we considered could either plausibly or not plausibly be caused by smoking. We examined associations between genetic instruments for smoking initiation, smoking heaviness and lifetime smoking and these phenotypes in UK Biobank and the Avon Longitudinal Study of Parents and Children (ALSPAC). We conducted negative control analyses among never smokers, including children. We found evidence that smoking‐related genetic instruments were associated with phenotypes not plausibly caused by smoking in UK Biobank and (to a lesser extent) ALSPAC. We observed associations with phenotypes among never smokers. Our results demonstrate that smoking‐related genetic risk scores are associated with unexpected phenotypes that are less plausibly downstream of smoking. This may reflect horizontal pleiotropy in these genetic risk scores, and we would encourage researchers to exercise caution this when using these and genetic risk scores for other complex behavioural exposures. We outline approaches that could be taken to consider this and overcome issues caused by potential horizontal pleiotropy, for example, in genetically informed causal inference analyses (e.g., MR) it is important to consider negative control outcomes and triangulation approaches, to avoid arriving at incorrect conclusions.

Keywords: ALSPAC, genetic risk scores, Mendelian randomisation, pleiotropy, smoking, UK Biobank

1. INTRODUCTION

Mendelian randomisation (MR) is a genetically informed causal inference approach that uses single nucleotide polymorphisms (SNPs) identified in genome‐wide association studies (GWAS) of putative exposures to examine causal effects of those exposures on outcomes (Davey Smith & Ebrahim, 2003; Richmond & Davey Smith, 2022). A key assumption for MR is that these SNPs are associated only with the outcome through the exposure of interest, and not via other pathways. If this assumption is not met then this may indicate the presence of pleiotropy, which may invalidate MR analyses, meaning that resulting causal inferences could be erroneous (Davey Smith et al., 2008; Sanderson et al., 2022; Smith, 2010).

Pleiotropy can take two different forms: vertical pleiotropy and horizontal pleiotropy, with the latter including correlated horizontal pleiotropy. These are shown in Figure 1. Vertical pleiotropy (Figure 1a), whereby the genetic variants operate via an intermediate phenotype, is not problematic for MR. Horizontal pleiotropy, however, undermines a key assumption of MR (Hemani et al., 2018). Horizontal pleiotropy (Figure 1b) can be balanced (i.e., the pleiotropic effects have a net effect of zero due to SNPs acting through different pathways resulting in both positive and negative effects that negate each other) or unbalanced (i.e., where the net effect is not zero). Balanced horizontal pleiotropy will increase the heterogeneity in MR analyses, but does not bias the effect estimates obtained; however, unbalanced horizontal pleiotropy is problematic and biases the results obtained. If such unbalanced pleiotropic pathways exist, then the results of MR studies may not be valid. Horizontal pleiotropy may also be correlated (Figure 1c). In this case, the genetic instrument influences a heritable confounder that influences both the exposure and outcome. In an extreme case, the SNPs may act entirely via a different primary phenotype, meaning that the target phenotype has been mis‐specified (Figure 1d).

Figure 1.

Figure 1

Open in a new tab

Different pleiotropic pathways to consider when estimating the causal effect of an exposure on an outcome using Mendelian randomisation (MR). (a) A vertical pleiotropic pathway where the genetic instrument G is associated with the outcome via an intermediate phenotype. (b) A horizontal pleiotropic pathway, where G is associated with both the exposure and another phenotype on a different pathway. (c) A correlated horizontal pleiotropic pathway, where G is also associated with a heritable confounder that, in turn, influences the exposure as well as the outcome (Morrison et al., 2020). (d) A pathway where the exposure is mis‐specified, where G is related to a heritable confounder only, but this causes both the exposure and the outcome. Scenario (a) does not invalidate MR estimates because the intermediate phenotype is on the causal path from the exposure to the outcome and the primary phenotype (exposure) is specified correctly (Davey Smith & Hemani, 2014), whereas scenarios (b–d) may invalidate the results of MR analyses. G, genetic instrument; U, heritable confounders of exposure‐outcome relationship.

Such horizontal pleiotropic effects may occur as a result of the much larger samples used in GWAS in recent years, identifying weaker associations with SNPs than found in smaller samples and increasing the likelihood that SNPs included in genetic scores are predictive of broader, correlated phenotypes (Mills & Rahal, 2019). As noted above, this is problematic for MR if the SNPs we use to instrument an exposure are in fact influencing other phenotypes via horizontal pleiotropy. Therefore, it is important to ascertain whether the SNPs we use as genetic instruments for an exposure of interest are in fact instrumenting only this exposure, or whether there is evidence of horizontal pleiotropy. While several MR sensitivity methods now exist that can help us to infer the likelihood of bias from horizontal pleiotropy and identify whether correlated pleiotropy is present (Cheng et al., 2022; Morrison et al., 2020; Xue et al., 2021), it is difficult to directly test for pleiotropic effects without understanding more about the functional biological effects of these SNPs.

One example of a complex behavioural exposure where horizontal pleiotropy may be operating is cigarette smoking. Recent GWAS have identified many genetic variants associated with different smoking behaviours (e.g., smoking initiation, heaviness of smoking). However, subsequent studies have suggested that some of these SNPs may also be influencing different phenotypes via independent pathways (i.e., not via smoking). For example, in studies by Khouja et al. (2021) and by Schellhas et al. (2021), SNPs for smoking initiation were also found to be associated with risk‐taking behaviours in young adults, personality traits in adults and adolescents and externalising disorders in children at age 7.

We assessed the extent that horizontal pleiotropy may be operating, using smoking behaviours as an example. Specifically, we investigated whether phenotypes that we considered could either plausibly be caused by smoking, or could not plausibly be caused by smoking, were associated with genetic risk scores for smoking initiation, smoking heaviness and a lifetime smoking index. We examined this across two cohorts, UK Biobank and the Avon Longitudinal Study of Parents and Children (ALSPAC), which are subject to different patterns of selection bias. We also conducted negative control analyses in never smokers, where any genotype‐outcome associations could not be mediated via smoking (Davey Smith et al., 2012; Lawlor et al., 2016). Associations with outcomes that could not plausibly be caused by smoking may provide evidence: (1) of horizontal pleiotropy (either correlated or uncorrelated), or (2) that our exposure phenotypes are mis‐specified (i.e., that the exposure is not actually the primary phenotype and the genetic variants used as the instrument are associated with, e.g., a downstream phenotype).

2. METHODS

Our study consists of two parts, a discovery phase (a phenome‐wide association study in UK Biobank), and an analysis phase (polygenic risk score [PRS] analyses in UK Biobank and ALSPAC). The steps of this study are outlined in Figure 2 and referred to in this section.

Figure 2.

Figure 2

Open in a new tab

Study flowchart. Flowchart of the main steps in analyses. Avon Longitudinal Study of Parents and Children (ALSPAC), Avon Longitudinal Study of Parents and Children; PheWAS, phenome‐wide association study; PRS, polygenic risk score; UKBB, UK Biobank.

2.1. Preregistration

We preregistered our analysis plan on the Open Science Framework (https://doi.org/10.17605/OSF.IO/37XYN). We stated that we would examine evidence of pleiotropy in two follow‐up studies using the Million Veteran Programme and the ALSPAC. However, in a deviation from our preregistered analyses, we only conducted analyses in ALSPAC due to data availability, and because we could also use child data in ALSPAC before smoking commencing as negative control analyses. Additionally, we did not conduct analyses using different p value thresholds, as stated in our preregistration, as we decided it was more relevant to see whether genome‐wide significant SNPs specifically (i.e., p < 5 × 10−08), often used as instruments in approaches such as MR, show evidence of horizontal pleiotropy.

2.2. Study cohorts

2.2.1. UK Biobank

UK Biobank is a large population‐based prospective health research resource with around 500,000 participants, aged 38–73 years at recruitment (between 2006 and 2010), from across the UK (Sudlow et al., 2015). A range of data have been collected including sociodemographic data, lifestyle, cognitive function, self‐reported measures and physical and mental health measures, with the aim of identifying determinants of human disease. Data have been collected via several methods, including paper‐ and web‐based questionnaires, computer‐assisted interviews, clinic visits and data linkage. Baseline assessment took place across 22 assessment centres to enable recruitment from a range of locations, but further data collection is ongoing. Further information can be found on the UK Biobank website (www.ukbiobank.ac.uk). UK Biobank received ethics approval from the Research Ethics Committee (REC reference for UK Biobank is 11/NW/0382). We excluded participants who withdrew their consent using the latest withdrawal lists for this project (project number: 16729). We restricted analyses to individuals who self‐reported as ‘White’ and ‘British’ and who had very similar genetic ancestry based on a principal components analysis of genotypes, which aims to minimise variation in non‐genetic and genetic factors. The self‐reported responses were from questions in the touchscreen questionnaire asking, ‘What is your ethnic group?’ with the options of White; Mixed; Asian or Asian British; Black or Black British, Chinese, Other ethnic group, Do not know; Prefer not to answer. If they selected ‘White’ then they were asked ‘What is your ethnic background?’ with the options of British; Irish; Any other White background; Prefer not to answer. We note that ethnicity is a complex social construct that can have different meanings across different contexts, and is distinct to, although often overlapping with, genetic ancestry (Birney et al., 2021). We also removed related individuals or those with mismatched sex.

2.2.2. ALSPAC

Pregnant women resident in Avon, UK with expected dates of delivery 1 April 1991 to 31 December 1992 were invited to take part in the study. The initial number of pregnancies enroled was 14,541. Of these, there was a total of 13,988 children alive at age 1. When children were age 7, additional eligible cases who had failed to join the study originally were recruited, resulting in a total sample size of 14,901 children. There are 14,833 mothers enroled in ALSPAC and 3807 partners (Northstone et al., 2023). ALSPAC is described in more detail in the cohort profile papers (Boyd et al., 2013; Fraser et al., 2013). The study website contains details of the available data through a data dictionary and variable search tool (http://www.bristol.ac.uk/alspac/researchers/our-data/). We used data from children, mothers and fathers/partners (not all were male) in our analyses. Ethics approval was obtained from the ALSPAC Ethics and Law Committee and the Local Research Ethics Committees (http://www.bristol.ac.uk/alspac/researchers/research-ethics/). Consent for biological samples was collected in accordance with the Human Tissue Act (2004). Informed consent for the use of data collected via questionnaires and clinics was obtained from participants following the recommendations of the ALSPAC Ethics and Law Committee at the time.

2.3. Phenotypic measures

2.3.1. Smoking‐related exposures

We used three smoking‐related exposures in our analyses to examine the extent of pleiotropy across these: smoking initiation (i.e., ever vs. never smoked), smoking heaviness (as measured by cigarettes smoked per day) and a lifetime smoking index (only available in UK Biobank). Further details of these measures for UK Biobank and ALSPAC can be found in Supporting Information S1: Section 1.

2.3.2. Discovery phase (Step 1)

We ran a phenome‐wide association study (PheWAS) (Denny et al., 2010) for smoking initiation using a polygenic risk score (PRS) of smoking initiation as the exposure (p < 5 × 10−08), constructed in UK Biobank from publicly available GWAS data (excluding UK Biobank) (Liu et al., 2019). Further details on the PheWAS can be found in the Supporting Information S1: Section 2. We used smoking initiation only for the PheWAS, as we hypothesised that this behaviour would be most likely to be associated with phenotypes not plausibly caused by smoking.

2.3.3. Phenotype selection (Step 2)

From the most strongly associated phenotypes (i.e., the top 100 results, based on p values, see Supporting Information S1: Table S1), we selected several phenotypes that could plausibly be caused by smoking, and those that could not plausibly be caused by smoking, in subsequent analyses. Phenotypes were selected on the basis that they were (1) likely to be caused by smoking (e.g., chronic obstructive pulmonary disease [COPD]), or (2) that they may be associated with, but were not likely to be caused by, smoking (e.g., age of mother at time of questionnaire). These were selected from consensus between authors, based on their expert knowledge and previous studies (see Supporting Information S1: Section 3). The phenotypes selected for use in analyses in both UK Biobank and ALSPAC are described below.

2.3.4. Phenotypes in UK Biobank plausibly caused by smoking for analysis phase

Details of the 13 phenotypes plausibly caused by smoking are shown in Supporting Information S1: Table S2. These were body mass index (BMI); body fat percentage; wheeze; C‐reactive protein (CRP); ever reported COPD; had dentures; overall health rating; gamma glutamyl transferase (GGT); white blood cell count; mean sphered cell volume; seen GP for nerves, anxiety or depression; numbers of medications taken; and alcohol consumption. We transformed continuous variables which were highly positively skewed to be normally distributed using an inverse normal rank transformation (INRT) (as indicated in Supporting Information S1: Table S2).

2.3.5. Phenotypes in UK Biobank not plausibly caused by smoking for analysis phase

Details of the 13 phenotypes not plausibly caused by smoking are shown in Supporting Information S1: Table S3. These were lifetime number of sexual partners; younger age at first live birth; Townsend deprivation index; takes part in a religious group; cereal intake; risk‐taking; time spent watching television; liking for cabbage; mobile phone usage; ease of skin tanning; mother's age at time of questionnaire; back pain and had an operation on the left‐side of the body. Variables were transformed using INRT are indicated in Supporting Information S1: Table S3.

2.3.6. Potential confounders in UK Biobank for analysis phase

Potential confounders of our genetic instrument and outcomes were age at assessment centre attendance, self‐reported sex and the first 10 principal components (PCs) of population structure from genotype data.

2.3.7. Phenotypes in ALSPAC plausibly and not plausibly caused by smoking for analysis phase

Where the phenotypes plausibly or not plausibly caused by smoking (or sufficiently similar phenotypes) were available in ALSPAC, we used these to examine whether UK Biobank results could be due to selection biases specific to UK Biobank, or whether these associations are observed in other cohorts. We used data from mothers, fathers/partners and children at age 10, the latter to examine whether associations existed before smoking onset, as this would provide evidence that pathways are not via the exposure of own smoking. We note that there may be some children who did smoke before age 10, but this number is likely to be very small. In addition, we also acknowledge the possible effect of parents sharing a genetic propensity to smoke with their child which could impact these findings, through possible exposure to smoking in utero and to passive smoking in the home. However, we would expect these effects, if present, to be very small. Therefore, any results from child data should be interpreted accounting for this, but if we observe effects in children this would likely still likely suggest horizontal pleiotropy.

Details on the phenotypes used can be found in Supporting Information S1: Tables S4 and S5. The measures used, and timepoints are summarised in Supporting Information S1: Figure S1. We generally used the earliest time point (T1) where data was available. However, there were some cases where this was less appropriate (e.g., due to pregnancy or early child age), in which case we ran additional analyses with the next time point (T2) to assess consistency (see Supporting Information S1: Tables S4 and S5).

For continuous phenotypes, we typically removed outliers by excluding individuals from analyses where data was three times the interquartile range above the upper quartile or below the lower quartile. However, we did not remove outliers for number of medications as values all seemed plausible.

2.4. Genetic data for analysis phase

2.4.1. UK Biobank

Of the 488,377 participants with genotyped samples, 336,988 were included in analyses (quality control steps are described in Supporting Information S1: Section 4).

2.4.2. ALSPAC

Genetic data were obtained from a combination of blood and buccal samples (see Supporting Information S1: Section 5). After quality control and removing those who had withdrawn consent, there were 7961 children, 7912 mothers and 1722 fathers/partners with genotype data available.

2.5. PRS construction for analysis phase

2.5.1. UK Biobank (Step 3)

To construct PRS we used a 10‐fold cross‐validation approach to reduce potential overfitting bias due to sample overlap as UK Biobank data was used for the discovery of GWAS and PRS construction (Burgess et al., 2017). This involved randomly splitting the UK Biobank population into 10 subsamples and then running 10 GWAS for each of the smoking‐related exposures in turn on 90% of the UK Biobank sample (see Supporting Information S1: Section 6), with a different 10% of the sample removed each time. PRS were then constructed for this remaining 10% of the sample to avoid sample overlap (see Supporting Information S1: Section 7). After 10 iterations were complete, PRS from each of the 10% subsamples were brought together such that cross‐validated scores for each of the smoking‐related exposures were available in the full sample. Further details on this 10‐fold cross‐validation methodology and related simulations are reported elsewhere (Fang et al., 2022).

2.5.2. ALSPAC (Step 4)

To construct PRS in ALSPAC, we used SNPs and weights from GWAS of smoking heaviness and initiation we conducted in UK Biobank (without the 10‐fold cross‐validation approach, see Supporting Information S1: Section 8) and from a published GWAS of lifetime smoking also conducted in UK Biobank (Wootton et al., 2019). This allowed us to replicate our UK Biobank analyses.

2.6. Statistical analysis in the analysis phase

All analyses were conducted in R version 3.5.1.

2.6.1. PRS analyses in UK Biobank (Step 3)

We tested the association of PRS for smoking initiation, smoking heaviness and lifetime smoking index (constructed using our cross‐validation approach) with the phenotypes plausibly and not plausibly caused by smoking. We used linear regression models for continuous outcomes, logistic regression models for binary outcomes and ordinal logistic regression for ordered factor outcomes. All models were adjusted for age, sex and the first 10 PCs. We stratified analyses using the smoking heaviness PRS by smoking status (never, former and current smokers), where analyses in never smokers acted as a type of negative control analysis, because individuals have not been exposed to their own smoking.

2.6.2. PRS analyses in ALSPAC (Step 4)

Similar to UK Biobank analyses, we tested the association of the PRS for smoking initiation, smoking heaviness and lifetime smoking on the phenotypes plausibly and not plausibly caused by smoking available in ALSPAC, separately for mothers, fathers/partners and children (due to phenotypic differences). Here we used GWAS from UK Biobank to replicate our analyses in UK Biobank. All models were adjusted for age, sex (in children only) and the first 10 PCs. We stratified analyses with the smoking heaviness PRS by smoking status in adults (ever or never smokers). Stratifying by smoking status in these analyses could introduce collider bias if a confounder affects both the outcome and smoking status—making smoking status a collider. Therefore, as well as conducting the analyses on adults, we additionally conducted the analyses in children (who will not have started smoking) where collider bias would not be introduced.

3. RESULTS

3.1. UK Biobank sample characteristics

Sample characteristics for participants included in this study (N = 101,397–336,988) are shown in Supporting Information S1: Table S6. The mean age was 57 years (SD = 8) and 54% were female. There were 45% participants who had ever smoked, with an average of five (SD = 10) cigarettes per day and a lifetime smoking score average of 0.34 (SD = 0.68). A 1 SD increase in this score is, for example, equivalent to being a current smoker who has smoked five cigarettes per day for 12 years, or a former smoker who smoked five cigarettes per day for 21 years but stopped smoking 10 years ago, compared to a never smoker.

3.2. PRS associations in UK Biobank (analysis phase)

3.2.1. Phenotypes plausibly caused by smoking

Results from analyses between each of the three smoking‐related PRS and the phenotypes plausibly caused by smoking are shown in Figure 3 and Supporting Information S1: Table S7. Results for smoking heaviness are presented for each of the three categories (never, former and current smokers). We found evidence of associations between the PRS for lifetime smoking and PRS for smoking initiation and all of our phenotypes plausibly caused by smoking (p = 1.06 × 10−29 to 2.77 × 10−03 for lifetime smoking and p = 1.82 × 10−18 to 4.48 × 10−04 for smoking initiation), with the direction of effect consistent for both PRS. For the smoking heaviness PRS, we found associations with some but not all of the phenotypes plausibly caused by smoking. In some cases, the effect estimates were attenuated compared to the other smoking PRS associations; however, confidence intervals were also generally wider for smoking heaviness so we may have had lower power to detect these effects. For former smokers, we found positive associations with the smoking heaviness PRS and CRP (b = 0.007; 95% confidence interval [CI] to: 0.001–0.01; p = 0.02), COPD (odds ratio [OR] = 1.04; 95% CI: 1.007–1.07; p = 0.01) and poorer health (OR = 1.01; 95% CI: 1.00–1.03; p = 0.02) only. For current smokers, we found negative associations with the smoking heaviness PRS and BMI (b = −0.10; 95% CI: −0.15 to −0.05; p = 2.19 × 10−04) and body fat percentage (b = −0.15; 95% CI: −0.22 to −0.08; p = 4.31 × 10−05). This is the opposite direction to the associations with the lifetime smoking and smoking initiation PRS. We also found a positive association with COPD (OR = 1.06; 95% CI: 1.03–1.10; p = 5.06 × 10−04). For never smokers, we did not find evidence of any associations between the phenotypes plausibly caused by smoking and the smoking heaviness PRS.

Figure 3.

Figure 3

Open in a new tab

Associations between the polygenic risk score (PRS) for lifetime smoking score, smoking heaviness and smoking initiation and phenotypes not plausibly caused by smoking in UK Biobank. Associations between PRSs for lifetime smoking index (purple), smoking heaviness (never: blue, former: green and current: orange) and smoking initiation (yellow) and phenotypes plausibly caused by smoking. The figure is split by the type of model used in the analysis; (a) linear regression, (b) logistic regression and (c) ordinal regression. The effect estimate is beta for linear regressions and odds ratios for logistic and ordinal regressions. BMI, body mass index; COPD, chronic obstructive pulmonary disease; CRP, C‐reactive protein; GP, general practitioner; WBC, white blood cell.

3.2.2. Phenotypes not plausibly caused by smoking

Results from analyses between each of the three smoking‐related PRS and the phenotypes not plausibly caused by smoking are shown in Figure 4 and Supporting Information S1: Table S8. We found evidence of associations between the PRS for lifetime smoking and smoking initiation and most of our phenotypes not plausibly caused by smoking (p = 1.06 × 10−29 to 2.77 × 10−03 for lifetime smoking and p = 1.82 × 10−18 to 4.48 × 10−04 for smoking initiation), with the direction of effect consistent for both PRS. However, we did not observe evidence of any association of back pain experienced in the last month with either of the PRS, or of having ever had an operation on the left side of the body with the lifetime smoking PRS. The strongest associations were generally observed with the lifetime smoking PRS and the direction of effect was generally the same, where we found associations, with each of the smoking‐related PRS. For the smoking heaviness PRS, we found positive associations for never smokers with the Townsend deprivation index (b = 0.02; 95% CI: 0.005–0.03; p = 0.007) and liking for cabbage (OR = 1.01; 95% CI: 1.00–1.03; p = 0.02), and for current smokers a negative association with age at first live birth (b = −0.09; 95% CI: −0.19 to −0.003; p = 0.04).

Figure 4.

Figure 4

Open in a new tab

Associations between the polygenic risk score (PRS) for lifetime smoking score, smoking heaviness and smoking initiation and phenotypes not plausibly caused by smoking in UK Biobank. Associations between PRSs for lifetime smoking index (purple), smoking heaviness (never: blue, former: green and current: orange) and smoking initiation (yellow) and phenotypes not plausibly caused by smoking. The figure is split by the type of model used in the analysis; (a) linear regression, (b) logistic regression and (c) ordinal regression. The effect estimate is beta for linear regressions and odds ratios for logistic and ordinal regressions.

3.3. ALSPAC sample characteristics

Sample characteristics for ALSPAC participants included in this study (mother N = 3635–13,093, father/partner N = 1713–9,833, child N = 4703–13,923) are shown in Supporting Information S1: Table S9. Mean age at the first time point was 28 years (SD = 5) for mothers, 31 years (SD = 6) for fathers/partners and 5 weeks (SD = 3.14) for children. In children 43% were female. Overall, 51% of mothers and 55% of fathers/partners had ever smoked. Of those that smoked, the cigarettes per day category with the highest percentage was 10 to 14 for mothers and 20 to 24 for fathers/partners.

3.4. PRS associations in ALSPAC (analysis phase)

3.4.1. Phenotypes plausibly caused by smoking

Results from analyses between each of the three smoking‐related PRS and the phenotypes plausibly caused by smoking are shown in Supporting Information S1: Figures S2–S4 and Table S10. Compared to UK Biobank, where we found associations of PRS with most of our phenotypes plausibly caused by smoking, we found fewer associations of PRS with those phenotypes in ALSPAC. We did find some evidence of associations between PRS and both increased BMI and body fat percentage, similar to our UK Biobank findings. Specifically, for lifetime smoking we found associations with mother's (b = 0.22; 95% CI: 0.04–0.39; p = 0.02) and father's BMI (b = 0.23; 95% CI: 0.007–0.46; p = 0.04), with weaker evidence for child BMI. We found evidence of an association with child's body fat percentage (b = 0.25; 95% CI: 0.02–0.48; p = 0.03), this was not evident for mothers or fathers/partners, but the direction and size of effect were similar. Furthermore, for lifetime smoking we found positive associations with wheeze (OR = 1.07; 95% CI: 1.01–1.15; p = 0.03) and having seen GP for depression in mothers (T2) (OR = 1.09; 95% CI: 1.02–1.16; p = 0.01), which we also observed in UK Biobank. We found negative associations with mother's overall health (T1) (OR = 0.95; 95% CI: 0.90–1.00; p = 0.05), father's overall health (T2) (OR = 0.89; 95% CI: 0.80–1.00; p = 0.05) and child anxiety (OR = 0.94; 95% CI: 0.89–1.00; p = 0.04). This latter finding is the opposite direction to what we observe in UK Biobank.

Unlike in UK Biobank, where we did not find evidence of associations with smoking heaviness PRS in never smokers, in ALSPAC we found a negative association between smoking heaviness PRS in never smokers and number of medications taken in mothers (T2) (b = −0.05; 95% CI: −0.09 to −0.008; p = 0.02). With ever smokers, we found a positive association with CRP (b = 0.07; 95% CI: 0.01–0.14; p = 0.02), similar to in UK Biobank and negative associations with alcohol consumption in mothers (OR = 0.92; 95% CI: 0.86–0.98; p = 0.006) and number of medications in fathers/partners (T1) (b = −0.14; 95% CI: −0.24 to −0.03; p = 0.01). However, we did not replicate our UK Biobank associations with COPD, overall health, BMI and body fat percentage in ALSPAC.

Finally, for the smoking initiation PRS we found positive associations with BMI (b = 0.36; 95% CI: 0.19–0.54; p = 0.00005) and body fat percentage in mothers (b = 0.34; 95% CI: 0.05–0.62; p = 0.02). Again, the direction of effect was the same across all samples for these and the same direction as that in UK Biobank. We also found a negative association with child anxiety (OR = 0.92; 95% CI: 0.87–0.97; p = 0.004), in the opposite direction to UK Biobank and a positive association with alcohol consumption in mothers (OR = 1.06; 95% CI: 1.02–1.11; p = 0.006), similar to in UK Biobank. However, we did not replicate other findings with phenotypes plausibly caused by smoking that we found in UK Biobank.

3.4.2. Phenotypes not plausibly caused by smoking

Results from analyses between each of the three smoking‐related PRS and the phenotypes not plausibly caused by smoking are shown in Supporting Information S1: Figures S5–S7 and Table S11. Similarly to in UK Biobank, we found many associations between PRS for lifetime smoking and phenotypes not plausibly caused by smoking. For lifetime smoking PRS we found positive associations with Townsend deprivation index in mothers (both timepoints) (T1: OR = 1.06; 95% CI: 1.01–1.12; p = 0.02 and T2: OR = 1.05; 95% CI: 1.00–1.09; p = 0.05), time spent watching TV in mothers on weekdays (b = 0.05; 95% CI: 0.02–0.08; p = 0.0006) and weekend days (b = 0.03; 95% CI: 0.006–0.06; p = 0.02) and cabbage intake in children (OR = 1.06; 95% CI: 1.01–1.11; p = 0.01), and negative associations with mother's age at first pregnancy (b = −0.19; 95% CI: −0.28 to −0.10; p = 5.85 × 10−05), attending a place of worship in mothers (OR = 0.91; 95% CI: 0.87–0.96; p = 0.0002) and children (OR = 0.90; 95% CI: 0.85–0.95; p = 0.00008), reduced wholegrain cereal consumption (both timepoints) (T1: OR = 0.92; 95% CI: 0.86–0.96; p = 0.0003 and T2: OR = 0.94; 95% CI: 0.89–0.98; p = 0.007) and increased intake of other cereals in mothers (T1) (OR = 1.05; 95% CI: 1.00–1.09; p = 0.04), avoiding risks in children (OR = 0.92; 95% CI: 0.88–0.96; p = 0.0005) and age of the father's mother at time of questionnaire, that is, lower age (b = −0.17; 95% CI: −0.32 to −0.03; p = 0.02).

We did not find any associations with smoking heaviness PRS in never smokers in all of our samples compared to the few we found in UK Biobank—potentially suggesting less pleiotropy for these variants. With ever smokers we found an association between smoking heaviness PRS and attending a place of worship less frequently in mothers (OR = 0.89; 95% CI: 0.83–0.96; p = 0.003), which we did not find in UK Biobank.

Finally, similarly to the many associations we found in UK Biobank between the smoking initiation PRS and phenotypes not plausibly caused by smoking, in ALSPAC for the smoking initiation PRS we found positive associations with cabbage intake in children (OR = 1.05; 95% CI: 1.01–1.11; p = 0.02) and mobile phone usage in mothers (OR = 1.08; 95% CI: 1.00–1.16; p = 0.05), and negative associations with age at first pregnancy in mothers (b = −0.15; 95% CI: −0.24 to −0.06; p = 0.001), attending a place of worship in mothers (OR = 0.92; 95% CI: 0.88–0.96; p = 0.0004) and children (OR = 0.95; 95% CI: 0.90–1.00; p = 0.05), oat cereal intake in mothers (T2) (OR = 0.94; 95% CI: 0.89–0.99; p = 0.01), avoiding risks in children (both timepoints) (T1: OR = 0.93; 95% CI: 0.88–0.97; p = 0.001 and T2: OR = 0.95; 95% CI: 0.91–1.00; p = 0.04) and mother's (b = −0.16; 95% CI: −0.30 to −0.02; p = 0.03) and age of the father's mother at time of questionnaire, that is, lower age (b = −0.35; 95% CI: −0.67 to −0.03; p = 0.03).

4. DISCUSSION

We found evidence of likely horizontal pleiotropy for genetic risk scores intended to capture cigarette smoking phenotypes. This has implications for any study using these exposures within an MR framework, and suggests caution may also be required for studies of other complex behaviour phenotypes.

Specifically, we found evidence that PRS for lifetime smoking index and smoking initiation were associated with most phenotypes not plausibly caused by smoking (as well as all phenotypes plausibly caused by smoking) in UK Biobank. There was less evidence of associations with the smoking heaviness PRS, but the strongest associations were found for current smokers. Some (although not all) of these results were replicated in ALSPAC. Our results are in line with recent studies suggesting associations of increased smoking initiation PRS with increased risk‐taking behaviours and decreased age at first birth, among other phenotypes (Khouja et al., 2021; Liu et al., 2019; Schellhas et al., 2021). However, we also found novel associations with other phenotypes not plausibly caused by smoking, suggesting that these potential pleiotropic effects may occur via a range of pathways.

It is difficult to know how associations with phenotypes not plausibly caused by smoking might arise; they could be a result of correlated pleiotropy, uncorrelated pleiotropy, mis‐specification of the primary phenotype or they may reflect other explanations. For example, the presence of a mis‐specified, unknown, primary phenotype on a vertical pleiotropic pathway may induce correlated pleiotropy. It is unlikely that an unknown phenotype is actually on the causal pathway from exposure to outcome because we consider the outcomes examined to be not plausibly caused by smoking. The presence of pleiotropy is further supported by evidence of associations in never smokers and children, where these phenotypes cannot be downstream of smoking because there has been no first‐hand exposure to smoking.

The fact we find less evidence of associations for the smoking heaviness PRS fits with the theory that smoking initiation measures in particular may be capturing other underlying phenotypes. The smoking heaviness PRS is more likely to capture biological function such as nicotinic pathways, as demonstrated in a previous PheWAS study in UK Biobank, where associations were observed with poorer lung function, higher blood assay levels, COPD, emphysema, cancer and greater facial ageing (Millard et al., 2019) among ever smokers. Whereas the smoking initiation measures are more likely to capture other phenotypes as well, as demonstrated in this study with our phenotypes not plausibly caused by smoking.

It may also be the case that there are alternative explanations for our findings. While it is difficult to know the situation for each of our implausible phenotypes, horizontal pleiotropy is likely to partially play a role. However, other explanations, for example, population stratification could influence some of the associations we find, although we adjusted for PCs to account for this. Furthermore, our findings in children before smoking could also potentially be due to dynastic effects, whereby the parents share a genetic propensity to smoke with their child, and thus effects of second‐hand or intrauterine exposure from their parents could induce an association between the child's genotype and a given outcome. Some of our findings (e.g., for Townsend deprivation index), could reflect upstream effects captured in the original smoking GWAS. This could also result in correlated horizontal pleiotropy, and we would encourage researchers to consider this when using smoking as an exposure in MR, for example by using sensitivity analyses such as MR‐CAUSE (causal analysis using summary effect estimates) (Morrison et al., 2020) which accounts for correlated horizontal pleiotropy. There may be unknown explanations for some of our other findings with phenotypes not caused by smoking as well, and these could be investigated further in future research. For example, liking for cabbage could be impacted by taste changes due to smoking, but this is unlikely to be specific to cabbage and most taste preferences would likely be determined early in life. It is important that findings such as these are considered carefully. One approach for doing so is by creating a directed acyclic graph (DAG) to help consider the casual framework including the likely temporality based on prior literature. Two of our implausible phenotypes are risk taking and age at first birth which we have provided DAGs for in Figure 5. Risk taking is considered an ‘implausible’ phenotype because risk taking is likely to precede and be a contributing factor to smoking initiation, rather than the other way around. Similarly for age at first birth, previous research has shown that smoking is associated with increased time to conceive (Wootton et al., 2023). Our results show the opposite (i.e., smoking leading to younger age at first birth). Therefore, it is an ‘implausible’ phenotype because the direction of effect observed in our study is opposite to what we might expect based on observational analyses.

Figure 5.

Figure 5

Open in a new tab

Directed acyclic graphs showing the theoretical causal framework and our findings for smoking initiation and (a) risk taking and (b) age at first birth.

4.1. Limitations

UK Biobank has some limitations in terms of selection bias. Compared to respondents of the same age range in national surveys, participants in UK Biobank were more likely to be homeowners, with lower BMIs, less likely to be current smokers and to drink alcohol daily, suffer from fewer self‐reported health conditions and have lower mortality in follow‐up (Fry et al., 2017). This indicates a ‘healthy volunteer’ effect in UK Biobank which should be considered when interpreting our results. Therefore, we also conducted analyses in ALSPAC, which may have different patterns of selection due to one being a cross‐sectional study and the other being a birth cohort. We found that some effects were consistent across cohorts. However, for other relationships, effects were not consistent between the cohorts. For example, for smoking heaviness PRS in ever smokers we found that a greater PRS was associated with less frequent attendance at a place of worship in ALSPAC, but we did not find any association in UK Biobank. We also found associations with other outcomes in UK Biobank but not in ALSPAC, although the direction of effect was often consistent. These differences could be due to different patterns in selection, different phenotype definitions, different sample sizes or they could be due to Winner's Curse as UK Biobank was our discovery sample and ALSAPC was our replication sample (Kraft, 2008).

It is also worth noting that we found an association between increased smoking heaviness PRS and fewer medications being taken in mothers who had never smoked. This is unexpected given that number of medications is one of our phenotypes plausibly caused by smoking, although this could reflect low health seeking behaviour. This further highlights that horizontal pleiotropic pathways can exist even for phenotypes plausibly downstream of smoking.

Finally, it is worth discussing our findings in children for our negative control analyses where we also replicated findings; for example, we found a negative association with avoiding taking risks, suggesting that lifetime smoking and smoking initiation PRS may pick up risk‐taking behaviours even in children who are unlikely to have considered smoking. This supports previous findings (Khouja et al., 2021; Schellhas et al., 2021) and further supports our conclusions that there is horizontal pleiotropy in smoking PRS which incorporate smoking initiation. Data from children could be included in additional MR analyses, as these would be conducted in individuals after the age where smoking is likely to have been initiated, to help examine these associations further.

A few limitations should be noted when interpreting our results. First, a subsample of the UK Biobank cohort were part of the UK Biobank Lung Exome Variant Evaluation study (Wain et al., 2015), which oversampled for smokers and for which a different genotyping chip was used compared to the rest of the sample, which could introduce collider bias into our results. It may also be the case that smoking phenotypes are underreported; for example, those identifying as ‘social smokers’ may underestimate their smoking heaviness or may not respond as being a smoker. Similarly, it is possible that stratifying our analyses on smoking status could induce collider bias; for example, if there is smoking cessation after a disease diagnosis or medication initiation. However, we used three different measures of smoking, so these limitations may be countered partially by that, and this is likely to only be the case for a small number of individuals. Another potential limitation is that ALSPAC data were collected in participants of a younger age range than in UK Biobank. Therefore, some of the differences in findings could be due to differences in associations with smoking across age. However, where results are similar across cohorts, this adds strength to our findings as they are consistent despite the different ages. In addition, our analyses were conducted in two European ancestry samples, therefore, our results may be less generalisable to other ancestry groups. Also, some of the phenotypes plausibly caused by smoking we identified have potentially complex relationships with smoking, for example, alcohol consumption and mental health. This may mean that there are bidirectional relationships, for example. However, given the evidence of associations between smoking and these phenotypes, we would argue that they are still valid phenotypes plausibly caused by smoking. Finally, collider bias could also impact our results, although previous simulations suggest that in such a large sample any collider bias from the effect of the smoking heaviness PRS on smoking status is unlikely to have a large impact on the effect estimate (Millard et al., 2019).

We found evidence of potential horizontal pleiotropy in exposures frequently included in MR analyses. Specifically, our results suggest that the SNPs used as genetic instruments for smoking‐related phenotypes may in fact be instrumenting other phenotypes. This has important implications for any study using smoking‐related SNPs or SNPs for other complex traits as instruments in an MR approach, as our inferences from these studies may be incorrect if this is not considered. Our results are particularly important to consider especially given that as GWAS become increasingly larger; for complex traits in particular, it is likely that the problem of pleiotropy will also become more widespread, as discussed elsewhere (Sanderson et al., 2022). Therefore, methods designed to address issues around pleiotropy (e.g., MR‐Egger, MR‐PRESSO [pleiotropy residual sum and outlier] and generalised summary ‐data‐based MR [GSMR] for MR analyses) should be included in genetically‐informed causal inference analyses. While there are many robust MR methods, they all make assumptions about the structure of the pleiotropy and may not perform as well where horizontal pleiotropy is substantial. Therefore, researchers should also consider other approaches to address this potential pleiotropy, such as including outcomes not plausibly caused by an exposure of interest and negative control analyses (e.g., including unexposed groups). Consideration should also be given to the biological mechanisms of the variants used where possible. This could be achieved by using clustering methods such as MR‐Clust (Foley et al., 2021). We found evidence of horizontal pleiotropy to a lesser extent for smoking heaviness, suggesting that for analyses with smoking exposures, SNPs for smoking heaviness may be better suited, especially when including a negative control analysis of nonsmokers or children before smoking onset to assess possible bias from horizontal pleiotropy. However, caution is still advised when using these instruments and consideration of how this may impact results should be given on a study‐by‐study basis. We recommend that similar analyses are also conducted for studies with other complex traits as well.

AUTHOR CONTRIBUTIONS

Marcus R. Munafò, George Davey Smith and Robyn E. Wootton contributed to the study conception and design. Data preparation and analysis were performed by Zoe E. Reed and Tom G. Richardson. The first draft of the manuscript was written by Zoe E. Reed, Robyn E. Wootton and Jasmine N. Khouja. All authors commented on subsequent versions of the manuscript. All authors read and approved the final manuscript.

CONFLICT OF INTEREST STATEMENT

T. G. R. is an employee of GlaxoSmithKline outside of this research. The remaining authors declare no conflict of interest.

CODE AVAILABILITY

Analysis code is available from the University of Bristol's Research Data Repository (http://data.bris.ac.uk/data/), at: https://doi.org/10.5523/bris.3bkvq4bbwzqwc2iovjpmyzffdt.

Supporting information

Supporting information.

GEPI-49-0-s001.docx (2.1MB, docx)

ACKNOWLEDGEMENTS

This research has been conducted using data from UK Biobank (project ID: 16729), a major biomedical database (http://www.ukbiobank.ac.uk/). We are extremely grateful to all the families who took part in the ALSPAC study, the midwives for their help in recruiting them and the whole ALSPAC team, which includes interviewers, computer and laboratory technicians, clerical workers, research scientists, volunteers, managers, receptionists and nurses. We thank all the contributors to the consortia we have used GWAS results from in our analyses. We would like to thank the research participants and employees of 23andMe Inc. for making this work possible. This work was supported by the UK Medical Research Council Integrative Epidemiology Unit at the University of Bristol (Grant ref: MC_UU_00011/1, MC_UU_00011/7). The UK Medical Research Council and Wellcome (Grant ref: 217065/Z/19/Z) and the University of Bristol provide core support for ALSPAC. This publication is the work of the authors and Z. E. R. and M. R. M. will serve as guarantors for the contents of this paper. A comprehensive list of grants funding is available on the ALSPAC website (http://www.bristol.ac.uk/alspac/external/documents/grant-acknowledgements.pdf); this research was specifically funded by the British Heart Foundation for mother's clinic data (Grant ref: SP/07/008/24066), the Wellcome Trust and MRC for father's clinic data (Grant ref: 092731), Wellcome Trust for mother's genetic data (Grant ref: WT088806) and the Wellcome Trust and MRC for father's genetic data (Grant ref: 102215/2/13/2). GWAS data was generated by Sample Logistics and Genotyping Facilities at Wellcome Sanger Institute and LabCorp (Laboratory Corporation of America) using support from 23andMe. This work was also supported by the National Institute for Health Research Bristol Biomedical Research Centre (M. R. M). This work was also supported by Cancer Research UK (Grant ref: C18281/A29019).

Reed, Z. E. , Wootton, R. E. , Khouja, J. N. , Richardson, T. G. , Sanderson, E. , Davey Smith, G. , & Munafò, M. R. (2025). Exploring pleiotropy in Mendelian randomisation analyses: What are genetic variants associated with ‘cigarette smoking initiation’ really capturing? Genetic Epidemiology, 49, e22583. 10.1002/gepi.22583

DATA AVAILABILITY STATEMENT

GWAS data for smoking initiation with UK Biobank and 23andMe removed can be found here: https://conservancy.umn.edu/handle/11299/201564. Full GWAS summary statistics for the 23andMe discovery data set (which we combined with the publicly available smoking initiation data) will be made available through 23andMe to qualified researchers under an agreement with 23andMe that protects the privacy of the 23andMe participants. Please visit https://research.23andme.com/collaborate/#dataset-access/ for more information and to apply to access the data.

ALSPAC data access is through a system of managed open access (http://www.bristol.ac.uk/alspac/researchers/access/).

UK Biobank data are available through a procedure described at http://www.ukbiobank.ac.uk/using-the-resource/.

REFERENCES

  1. Birney, E. , Inouye, M. , Raff, J. , Rutherford, A. , & Scally, A. (2021). The language of race, ethnicity, and ancestry in human genetic research. ArXiv. [Preprint] Available from: 10.48550/arXiv.2106.10041 [DOI]
  2. Boyd, A. , Golding, J. , Macleod, J. , Lawlor, D. A. , Fraser, A. , Henderson, J. , Molloy, L. , Ness, A. , Ring, S. , & Davey Smith, G. (2013). Cohort profile: The 'children of the 90s’—The index offspring of the Avon Longitudinal Study of parents and children. International Journal of Epidemiology, 42(1), 111–127. 10.1093/ije/dys064 [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Burgess, S. , Small, D. S. , & Thompson, S. G. (2017). A review of instrumental variable estimators for Mendelian randomization. Statistical Methods in Medical Research, 26(5), 2333–2355. 10.1177/0962280215597579 [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Cheng, Q. , Zhang, X. , Chen, L. S. , & Liu, J. (2022). Mendelian randomization accounting for complex correlated horizontal pleiotropy while elucidating shared genetic etiology. Nature Communications, 13(1), 1–13. 10.1038/s41467-022-34164-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Davey Smith, G. , & Ebrahim, S. (2003). Mendelian randomization’: Can genetic epidemiology contribute to understanding environmental determinants of disease? International Journal of Epidemiology, 32(1), 1–22. 10.1093/IJE/DYG070 [DOI] [PubMed] [Google Scholar]
  6. Davey Smith, G. , & Hemani, G. (2014). Mendelian randomization: Genetic anchors for causal inference in epidemiological studies. Human Molecular Genetics, 23(R1), R89–R98. 10.1093/HMG/DDU328 [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Davey Smith, G. , Lipsitch, M. , Tchetgen, E. T. , & Cohen, T. (2012). Negative control exposures in epidemiologic studies. Source: Epidemiology, 23(2), 350–352. [DOI] [PubMed] [Google Scholar]
  8. Denny, J. C. , Ritchie, M. D. , Basford, M. A. , Pulley, J. M. , Bastarache, L. , Brown‐Gentry, K. , Wang, D. , Masys, D. R. , Roden, D. M. , & Crawford, D. C. (2010). PheWAS: Demonstrating the feasibility of a phenome‐wide scan to discover gene‐disease associations. Bioinformatics, 26(9), 1205–1210. 10.1093/bioinformatics/btq126 [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Fang, S. , Hemani, G. , Richardson, T. G. , Gaunt, T. R. , & Davey Smith, G. (2023). Evaluating and implementing block jackknife resampling Mendelian randomization to mitigate bias induced by overlapping samples. Human Molecular Genetics, 32, 192–203. 10.1093/HMG/DDAC186 [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Foley, C. N. , Mason, A. M. , Kirk, P. D. W. , & Burgess, S. (2021). MR‐Clust: Clustering of genetic variants in Mendelian randomization with similar causal estimates. Bioinformatics, 37(4), 531–541. 10.1093/bioinformatics/btaa778 [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Fraser, A. , Macdonald‐Wallis, C. , Tilling, K. , Boyd, A. , Golding, J. , Davey Smith, G. , Henderson, J. , Macleod, J. , Molloy, L. , Ness, A. , Ring, S. , Nelson, S. M. , & Lawlor, D. A. (2013). Cohort profile: The avon longitudinal study of parents and children: ALSPAC mothers cohort. International Journal of Epidemiology, 42(1), 97–110. 10.1093/ije/dys066 [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Fry, A. , Littlejohns, T. J. , Sudlow, C. , Doherty, N. , Adamska, L. , Sprosen, T. , Collins, R. , & Allen, N. E. (2017). Comparison of sociodemographic and health‐related characteristics of UK biobank participants with those of the general population. American Journal of Epidemiology, 186(9), 1026–1034. 10.1093/aje/kwx246 [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Hemani, G. , Bowden, J. , & Davey Smith, G. (2018). Evaluating the potential role of pleiotropy in Mendelian randomization studies. Human Molecular Genetics, 27(R2), R195–R208. 10.1093/HMG/DDY163 [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Khouja, J. N. , Wootton, R. E. , Taylor, A. E. , Davey Smith, G. , & Munafò, M. R. (2021). Association of genetic liability to smoking initiation with e‐cigarette use in young adults: A cohort study. PLoS Medicine, 18(3), e1003555. 10.1371/journal.pmed.1003555 [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Kraft, P. (2008). Curses—Winner's and otherwise—In genetic epidemiology. Epidemiology, 19(5), 649–651. 10.1097/EDE.0B013E318181B865 [DOI] [PubMed] [Google Scholar]
  16. Lawlor, D. A. , Tilling, K. , & Davey Smith, G. (2016). Triangulation in aetiological epidemiology. International Journal of Epidemiology, 45(6), 1866–1886. 10.1093/IJE/DYW314 [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Liu, M. , Jiang, Y. , Wedow, R. , Li, Y. , Brazel, D. M. , Chen, F. , Datta, G. , Davila‐Velderrain, J. , McGuire, D. , Tian, C. , Zhan, X. , Choquet, H. , Docherty, A. R. , Faul, J. D. , Foerster, J. R. , Fritsche, L. G. , Gabrielsen, M. E. , Gordon, S. D. , Haessler, J. , … Vrieze, S. (2019). Association studies of up to 1.2 million individuals yield new insights into the genetic etiology of tobacco and alcohol use. Nature Genetics, 51(2), 237–244. 10.1038/s41588-018-0307-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Millard, L. A. C. , Munafò, M. R. , Tilling, K. , Wootton, R. E. , & Davey Smith, G. (2019). MR‐pheWAS with stratification and interaction: Searching for the causal effects of smoking heaviness identified an effect on facial aging. PLoS Genetics, 15(10), e1008353. 10.1371/JOURNAL.PGEN.1008353 [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Mills, M. C. , & Rahal, C. (2019). A scientometric review of genome‐wide association studies. Communications Biology, 2(1), 9. 10.1038/s42003-018-0261-x [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Morrison, J. , Knoblauch, N. , Marcus, J. H. , Stephens, M. , & He, X. (2020). Mendelian randomization accounting for correlated and uncorrelated pleiotropic effects using genome‐wide summary statistics. Nature Genetics, 52(7), 740–747. 10.1038/s41588-020-0631-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Northstone, K. , Ben Shlomo, Y. , Teyhan, A. , Hill, A. , Groom, A. , Mumme, M. , Timpson, N. , & Golding, J. (2023). The Avon Longitudinal Study of Parents and children ALSPAC G0 partners: A cohort profile. Wellcome Open Research, 8, 37. 10.12688/WELLCOMEOPENRES.18782.1 [DOI] [Google Scholar]
  22. Richmond, R. C. , & Davey Smith, G. (2022). Mendelian randomization: Concepts and scope. Cold Spring Harbor Perspectives in Medicine, 12(1), a040501. 10.1101/CSHPERSPECT.A040501 [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Sanderson, E. , Glymour, M. M. , Holmes, M. V. , Kang, H. , Morrison, J. , Munafò, M. R. , Palmer, T. , Schooling, C. M. , Wallace, C. , Zhao, Q. , & Davey Smith, G. (2022). Mendelian randomization. Nature Reviews Methods Primers, 2(1), 6. 10.1038/s43586-021-00092-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Schellhas, L. , Haan, E. , Easey, K. E. , Wootton, R. E. , Sallis, H. M. , Sharp, G. C. , Munafò, M. R. , & Zuccolo, L. (2021). Maternal and child genetic liability for smoking and caffeine consumption and child mental health: An intergenerational genetic risk score analysis in the ALSPAC cohort. Addiction, 116(11), 3153–3166. 10.1111/ADD.15521 [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Smith, G. D. (2010). Mendelian randomization for strengthening causal inference in observational studies: Application to gene × environment interactions. Perspectives on Psychological Science, 5(5), 527–545. 10.1177/1745691610383505 [DOI] [PubMed] [Google Scholar]
  26. Smith, G. D. , Timpson, N. , & Ebrahim, S. (2008). Strengthening causal inference in cardiovascular epidemiology through Mendelian randomization. Annals of Medicine, 40(7), 524–541. 10.1080/07853890802010709 [DOI] [PubMed] [Google Scholar]
  27. Sudlow, C. , Gallacher, J. , Allen, N. , Beral, V. , Burton, P. , Danesh, J. , Downey, P. , Elliott, P. , Green, J. , Landray, M. , Liu, B. , Matthews, P. , Ong, G. , Pell, J. , Silman, A. , Young, A. , Sprosen, T. , Peakman, T. , & Collins, R. (2015). UK biobank: An open access resource for identifying the causes of a wide range of complex diseases of middle and old age. PLoS Medicine, 12(3), e1001779. 10.1371/journal.pmed.1001779 [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Wain, L. V. , Shrine, N. , Miller, S. , Jackson, V. E. , Ntalla, I. , Artigas, M. S. , Billington, C. K. , Kheirallah, A. K. , Allen, R. , Cook, J. P. , Probert, K. , Obeidat, M. , Bossé, Y. , Hao, K. , Postma, D. S. , Paré, P. D. , Ramasamy, A. , Mägi, R. , Mihailov, E. , … Hall, I. P. (2015). Novel insights into the genetics of smoking behaviour, lung function, and chronic obstructive pulmonary disease (UK BiLEVE): A genetic association study in UK Biobank. The Lancet Respiratory Medicine, 3(10), 769–781. 10.1016/S2213-2600(15)00283-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Wootton, R. E. , Lawn, R. B. , Magnus, M. C. , Treur, J. L. , Corfield, E. C. , Njølstad, P. R. , Andreassen, O. A. , Lawlor, D. A. , Munafò, M. R. , Håberg, S. E. , Davey Smith, G. , Reichborn‐Kjennerud, T. , Magnus, P. , & Havdahl, A. (2023). Associations between health behaviours, fertility and reproductive outcomes: Triangulation of evidence in the Norwegian Mother, Father and Child Cohort Study (MoBa). BMC Medicine, 21(1), 125. 10.1186/S12916-023-02831-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Wootton, R. E. , Richmond, R. C. , Stuijfzand, B. G. , Lawn, R. B. , Sallis, H. M. , Taylor, G. , Hemani, G. , Jones, H. J. , Zammit, S. , Davey Smith, G. , & Munafò, M. R. (2020). Evidence for causal effects of lifetime smoking on risk for depression and schizophrenia: A Mendelian randomisation study. Psychological Medicine, 50, 2435–2443. 10.1017/s0033291719002678 [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Xue, H. , Shen, X. , & Pan, W. (2021). Constrained maximum likelihood‐based Mendelian randomization robust to both correlated and uncorrelated pleiotropic effects. The American Journal of Human Genetics, 108(7), 1251–1269. 10.1016/J.AJHG.2021.05.014 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supporting information.

GEPI-49-0-s001.docx (2.1MB, docx)

Data Availability Statement

GWAS data for smoking initiation with UK Biobank and 23andMe removed can be found here: https://conservancy.umn.edu/handle/11299/201564. Full GWAS summary statistics for the 23andMe discovery data set (which we combined with the publicly available smoking initiation data) will be made available through 23andMe to qualified researchers under an agreement with 23andMe that protects the privacy of the 23andMe participants. Please visit https://research.23andme.com/collaborate/#dataset-access/ for more information and to apply to access the data.

ALSPAC data access is through a system of managed open access (http://www.bristol.ac.uk/alspac/researchers/access/).

UK Biobank data are available through a procedure described at http://www.ukbiobank.ac.uk/using-the-resource/.

Articles from Genetic Epidemiology are provided here courtesy of Wiley

RESOURCES