Estimating causality between smoking and abdominal obesity by Mendelian randomization

Background and Aims Smokers tend to have a lower body weight than non-smokers, but also more abdominal fat. It remains unclear whether the relationship between smoking and abdominal obesity is causal. Previous Mendelian randomization studies have investigated this relationship by relying on a single genetic variant for smoking heaviness. This approach is sensitive to pleiotropic effects and may produce imprecise causal estimates. We aimed to assess causality between smoking and abdominal obesity using multiple genetic instruments. Methods We used GWAS results for smoking initiation (n=1,232,091), lifetime smoking (n=462,690) and smoking heaviness (n=337,334) as exposure traits, and waist-hip ratio (WHR) and waist and hip circumferences (WC and HC) (n up to 697,734), with and without adjustment for body mass index (adjBMI), as outcome traits. We implemented Mendelian randomization using the CAUSE and LHC-MR methods that instrument smoking using genome-wide data. Results Both CAUSE and LHC-MR indicated a positive causal effect of smoking initiation on WHR (0.13 [95%CI 0.10, 0.16] and 0.49 [0.41, 0.57], respectively) and WHRadjBMI (0.07 [0.03, 0.10] and 0.31 [0.26, 0.37]). Similarly, they indicated a positive causal effect of lifetime smoking on WHR (0.35 [0.29, 0.41] and 0.44 [0.38, 0.51]) and WHRadjBMI (0.18 [0.13, 0.24] and 0.26 [0.20, 0.31]). In follow-up analyses, smoking particularly increased visceral fat. There was no evidence of a mediating role by cortisol or sex hormones. Conclusions Smoking initiation and higher lifetime smoking may lead to abdominal fat distribution. The increase in abdominal fat due to smoking was characterized by an increase in visceral fat. Thus, efforts to prevent and cease smoking can have the added benefit of reducing abdominal fat.


INTRODUCTION
Smoking prevention and cessation are critical for public health efforts to reduce the incidence of several chronic disorders, particularly respiratory and cardiovascular diseases [1]. However, smoking cessation is often associated with weight gain [2,3], which can decrease motivation for sustained cessation and undermine the health benefits. Despite having lower body weight, smokers often have more abdominal fat than non-smokers, which may increase their risk of cardiometabolic diseases [2,[4][5][6][7][8][9][10][11][12][13]. Whether the association between smoking and body fat distribution is causal or explained by confounding or reverse causality is still unclear.
Mendelian randomization uses genetic variants associated with exposure traits as instrumental variables to assess whether their relationship with outcome traits may be causal. As maternal and paternal alleles are randomly allocated during conception, Mendelian randomization is analogous to a naturally randomized controlled trial (Figure 1). Three previous Mendelian randomization studies have examined the causal effect of smoking heaviness on abdominal obesity using a single genetic variant in the CHRNA3/5 smoking heaviness locus [14][15][16]. The first two studies found no evidence of causality, while the third and largest study suggested a causal relationship between a higher number of cigarettes per day and higher WHR, even after adjusting for BMI [15].
Studies based on a single genetic instrument are sensitive to genetic pleiotropy and may produce imprecise causal estimates, leaving the relationship between smoking and abdominal obesity uncertain. Since the publication of the previous Mendelian randomization studies, genomewide association studies (GWAS) of smoking traits have identified >400 novel loci associated with smoking traits [17]. These findings provide new opportunities for increasing the precision and statistical power of Mendelian randomization studies while reducing bias. Furthermore, new and improved Mendelian randomization methods allow leveraging GWAS summary results at genomewide level to better control for pleiotropy and latent heritable confounders and assess bidirectional causal effects [18][19][20].
Here, we report on two-sample Mendelian randomization analyses to estimate the causal effect of smoking initiation, smoking heaviness, and lifetime smoking (which captures smoking heaviness, duration, and time of cessation) on abdominal adiposity.

Mendelian randomization methods
The Causal Analysis Using the Summary Effect estimates (CAUSE) and the Latent Heritable Confounder MR (LHC-MR) methods utilize genome-wide association data to assess causal relationships rather than genome-wide significant loci only, to correct for sample overlap and better control for correlated and uncorrelated horizontal pleiotropy (Figure 1, B) [19,20]. The CAUSE method assesses the unidirectional causal relationship (Figure 1, C) between an exposure and an outcome trait by calculating the posterior probabilities of a causal effect and a shared effect. The causal effect reflects the effect of the variants on the outcome trait through the exposure, while the shared effect reflects the effect of the variants on the outcome trait through confounders (correlated horizontal pleiotropy, Figure 1, B). The LHC-MR method assesses bidirectional causal relationships (Figure 1, C) by dividing the association between an exposure and an outcome trait into four different effects: the causal effect of the exposure on the outcome, the causal effect of the outcome on the exposure, the effect of confounders that affect the outcome through the exposure (vertical pleiotropy), and the effect of confounders that affect the outcome independently of the exposure (correlated horizontal pleiotropy) [20]. More details on the CAUSE and LHC-MR methods are provided in the Supplementary Note.
In addition to the CAUSE (v1.0.0) and LHC-MR (v.1.0.0) methods, we implemented Mendelian randomization using inverse variance weighted, MR-Egger, weighted median, and weighted mode methods, instrumenting the smoking exposure trait using genome-wide significant loci (P < 5x10 -8 ). We selected lead variants that showed a pairwise linkage disequilibrium (LD) r 2 < 0.001 and a distance of 10.000kb with all other lead variants associated with the same trait.
Variants not available in the outcome trait GWAS were substituted by their LD proxies (r 2 > 0.8).
Ambiguous palindromic variants (A/T, G/C) were excluded. We applied Steiger filtering to remove variants that are likely to affect the outcome trait through other traits than the exposure trait. The strength of the genetic instruments was assessed using the F statistic. The analyses were performed using the TwoSampleMR (v.0.5.6)  RadialMR, we re-ran the Mendelian randomization and sensitivity tests to ensure that horizontal pleiotropy introduced by the outlier variants had been removed.
There is a considerable genetic correlation between smoking status and socioeconomic status [26]. Therefore, we conducted a multivariable Mendelian randomization (MVMR,v. 0.3) [27] analysis to assess the causal effects of smoking initiation and lifetime smoking on abdominal adiposity while controlling for the effect of socioeconomic status (educational attainment, n=458,079) [28]. We followed the MVMR steps described previously by Sanderson et al [29]. First, we identified independent, genome-wide significant variants associated with smoking and education (r 2 < 0.001 within 10.000kb distance) and replaced them with proxies (r 2 > 0.8) when not available in the outcome trait GWAS. We calculated the conditional mean F-statistic to ensure that the instrumental variable was robustly associated with the exposure trait. We obtained IVW estimates before and after outlier removal for genetic instruments with a mean F statistic > 10. We We also utilized the largest published GWAS summary level data of European ancestry for measures of body fat distribution (Figure 1, D), including WHR (n=697,734) and WHR adjusted for body mass index (WHR adjBMI , n=694,649) from a meta-analysis of the Genetic Investigation of Anthropometric Traits (GIANT) consortium and the UK Biobank [32]; waist and hip circumferences (WC and HC) from the UK Biobank (n=462,166 and n=462,177, respectively); and WC adjBMI and HC adjBMI from the GIANT consortium (n=231,355 and n=211,117, respectively) [33]. In the analyses of the causal relation between cigarettes smoked per day and abdominal adiposity in current smokers, we utilized the GIANT consortium results for WHR AdjBMI and WC AdjBMI in current smokers only (n=40,543 and n=43,226, respectively) [33]. In the Mendelian randomization analyses using the IVW, MR Egger, weighted median and weighted mode methods, which are sensitive to bias when there is a sample overlap between the exposure and outcome traits, we used the GIANT consortium data without the UK Biobank for each of the body fat distribution traits (n between 40,543 and 232,101) [34] (Supplementary Table 1). The present study used public GWAS summary-level data and did not have direct contact with study participants. Therefore, no direct ethical approval was needed. The code and curated data for the current analysis are available at https://github.com/MarioGuCBMR/MR_smoking_abdominal_adiposity.

Effects of smoking initiation and lifetime smoking on abdominal adiposity
In Mendelian randomization analyses by the CAUSE and LHC-MR methods, we found evidence of a positive causal effect of smoking initiation on WHR (  There is a considerable genetic correlation between smoking initiation and markers of socioeconomic status, such as educational attainment (r g ~ -0.4) [26]. To determine whether the causal effects of smoking initiation and lifetime smoking on abdominal obesity are independent of genetic effects on socioeconomic status, we performed a multivariable Mendelian randomization analysis. We found that smoking initiation and lifetime smoking are causally associated with WHR, WHR adjBMI , WC and WC adjBMI , even after controlling for genetic effects on educational attainment (Supplementary Note, Supplementary 8-9). This suggests that the causal effects of smoking initiation and lifetime smoking on abdominal obesity are independent of socioeconomic status.

Effect of smoking heaviness on abdominal adiposity in past and current smokers
Our Mendelian randomization analysis showed no evidence of a causal relationship between a higher number of cigarettes per day or the cumulative pack years of smoking in adulthood and WHR or WHR adjBMI (Figure 2- Figure 4).

Causal effect of abdominal obesity on smoking traits
We used the LHC-MR method to estimate bidirectional causal effects between abdominal obesity and smoking traits. Our results showed a positive causal effect of WHR on lifetime smoking (0.08

Association of smoking variants with fat depots and hormonal levels.
We investigated the effect of smoking variants on VAT, ASAT, and GSAT tissue volumes by genetic risk score analyses. We generated the scores using the smoking initiation and lifetime genetic association results did not support a role of either cortisol or sex hormones in the relationship between smoking and abdominal adiposity.
The strengths of the present study include the use of the largest available GWAS summary-level data on smoking and body fat distribution, and the application of several complementary MR methods and sensitivity analyses to control for pleiotropic effects, sample overlap, latent heritable confounders, reverse causality, and type I error. However, there are also some limitations to our study. Despite performing several sensitivity analyses, we cannot completely rule out the potential influence of residual pleiotropy on our causal estimates.
Furthermore, the sample size for body fat distribution in current smokers was relatively small, limiting the statistical power of our smoking heaviness analysis compared to the larger sample sizes used for smoking initiation and lifetime smoking. Finally, our population was restricted to 1 individuals of European genetic ancestry, and thus the findings may not be generalizable to other populations.
In conclusion, smoking initiation and lifetime smoking may causally increase abdominal and particularly visceral fat. Thus, public health efforts to prevent and reduce smoking may also help reduce abdominal fat and the risk of related chronic diseases.

ACKNOWLEDGEMENTS
The present study was conducted independently without any involvement or influence from any of . The first assumption is that the instrument is associated with the exposure. The second assumption is that the instrument is not associated with the outcome throu confounding pathway (uncorrelated horizontal pleiotropy). The third assumption is that the instrument does not directly influence th outcome but does so indirectly through the exposure (correlated horizontal pleiotropy). Depending on the direction of the causal es being tested, the analysis can be unidirectional or bidirectional (C). Abdominal adipose tissue consists of visceral adipose tissue an abdominal subcutaneous adipose tissue (D). The associations of the genetic risk scores with VAT/GSAT, VAT/ASAT and ASAT/GSAT ratios are also shown. We constructed genetic risk scores using variants from the inverse variance-weighted model for the causal association between smoking initiation and WHR (121 variants), smoking initiation and WHR adjBMI (130 variants), lifetime smoking and WHR (76 variants), and lifetime smoking and WHR adjBMI (83 variants). The smoking variants causally associated with WHR were used to study associations with BMI-unadjusted fat depots, while the smoking variants causally associated with WHR adjBMI were used to study associations with BMI-adjusted fat depots. We computed weighted genetic scores using the beta of the variants for a smoking trait as weights with gtx.package's (v0.0.8) grs.summary function. This function approximates the results of a genetic risk score using GWAS summary statistics by calculating the joint effect of genetic variants on an outcome trait.  a  r  r  o  u  s  L  ,  M  o  u  n  i  e  r  N  ,  K  u  t  a  l  i  k  Z  .  S  i  m  u  l  t  a  n  e  o  u  s  e  s  t  i  m  a  t  i  o  n  o  f  b  i  -d  i  r  e  c  t  i  o  n  a  l  c  a  u  s  a  l  e  f  f  e  c  t  s  a  n  d   h  e  r  i  t  a  b  l  e  c  o  n  f  o  u  n  d  i  n  g  f  r  o  m  G  W  A  S  s  u  m  m  a  r  y  s  t  a  t  i  s  t  i  c T  h  e  a  s  s  o  c  i  a  t  i  o  n  b  e  t  w  e  e  n  a  m  o  u  n  t  o  f  c  i  g  a  r  e  t  t  e  s  s  m  o  k  e  d  a  n  d   o  v  e  r  w  e  i  g  h  t  ,  c  e  n  t  r  a  l  o  b  e  s  i  t  y  a  m  o  n  g  C  h  i  n  e  s  e  a  d  u  l  t  s  i  n  N  a  n  j  i  n  g  ,  C  h