Cigarette smoking and personality: Investigating causality using Mendelian randomization

Background Despite the well-documented association between smoking and personality traits such as neuroticism and extraversion, little is known about the potential causal nature of these findings. If it were possible to unpick the association between personality and smoking, it may be possible to develop more targeted smoking cessation programmes that could lead to both improved uptake and efficacy. Methods Recent genome-wide association studies (GWAS) have identified variants robustly associated with both smoking phenotypes and personality traits. Here we use publicly available GWAS summary statistics in addition to data from UK Biobank to investigate the link between smoking and personality. We first estimated genetic overlap between traits using LD score regression and then applied both one- and two-sample Mendelian randomization methods to unpick the nature of this relationship. Results We found clear evidence of a modest genetic correlation between smoking behaviours and both neuroticism and extraversion, suggesting shared genetic aetiology. We found some evidence to suggest an association between neuroticism and increased smoking initiation. We also found some evidence that personality traits appear to be causally linked to certain smoking phenotypes: higher neuroticism and heavier cigarette consumption, and higher extraversion and increased odds of smoking initiation. The latter finding could lead to more targeted smoking prevention programmes. Conclusion The association between neuroticism and cigarette consumption lends support to the self-medication hypothesis, while the association between extraversion and smoking initiation could lead to more targeted smoking prevention programmes.


Introduction 50
There is a well-documented association between smoking and personality traits such 51 as neuroticism and extraversion (Terracciano & Costa Jr. 2004;Malouff et al. 2006;Munafò 52 et al. 2007;Hakulinen et al. 2015), and with associated mental health outcomes such as 53 smoking cessation. For each phenotype, the effect allele was that which corresponded to an 131 increase in the relevant smoking behaviour. 132 Personality. Eleven independent variants associated with neuroticism were reported 133 by Okbay et al. (2016) Of the original variants, 5 of these were unavailable in the TAG 134 smoking data. Proxies were identified for 4 of these SNPs using SNIPA (Arnold et al. 2015) 135 (r 2 >0.85). A complete list of variants used in the neuroticism instrument can be found in 136 Table S1. Five independent variants associated with extraversion were identified by the GPC 137 (van den Berg et al. 2016). Of these variants, 3 variants were unavailable in the TAG 138 summary statistics. Using SNIPA, we identified a proxy for one of these variants (Table S1). Smoking status was defined as ever (consisting of current and former smokers) or 159 never smoker according to responses given at the initial assessment visit in UKB. Smoking 160 heaviness was derived for former and current smokers based on responses to 'number of 161 cigarettes currently smoked daily' at the initial assessment. For former smokers, this 162 question related to number of cigarettes previously smoked daily. 163 Neuroticism scores were derived from a number of neurotic behaviour domains 164 measured at the initial assessment visit. These scores were externally derived by Smith et 165 al. (2013) and are available for use by researchers accessing the UKB resource. Scores 166 range from 0 to 12 with a higher score corresponding to a greater number of neurotic 167 behaviours. There was no direct measure of extraversion in UKB, so analyses of this trait 168 were restricted to those based on the genetic instrument for extraversion. 169 Polygenic risk scores for neuroticism and extraversion were calculated for each 170 individual in UKB. The neuroticism risk score ranged from 1 to 19 and corresponded to the 171 number of neuroticism increasing alleles per individual. A risk score was calculated for 172 extraversion and ranged from 0 to 6. Although weighted scores can give a more precise we restricted analyses using the neuroticism risk score to participants who were not included 178 in the interim release of the genetic data, and who were therefore not included in the 179 discovery sample. 180 181

Statistical analysis 182
Genetic correlation. In a first step, GWAS summary statistics were used to estimate 183 the genetic correlation of smoking initiation with both neuroticism and extraversion. LD score 184 regression was performed (without constraining the intercept) using the GWAS summary genetic correlation between personality measures and additional smoking phenotypes of 187 smoking heaviness and cessation, summary statistics for the personality GWAS would need 188 to be stratified by smoking status. Although the original GWAS summary statistics were not 189 available stratified by smoking status, it was possible to estimate these genetic correlations genetic correlation between the two traits (rG=0.124, SE=0.05). There was also evidence of a 216 larger genetic correlation between extraversion and smoking initiation (rG=0.288, SE=0.01; 217 Table 1). We used GCTA software to calculate genetic correlation using individual level data 218 from UKB. There was evidence of genetic correlation between neuroticism and smoking 219 heaviness among both current (rG=0.248, SE=0.12) and former smokers (rG=0.220, 220 SE=0.06). We found evidence of a negative genetic correlation between smoking cessation 221 and neuroticism (rG=-0.314, SE=0.15; Table 1). 222 initiation risk allele associated with a decrease in neuroticism score (β=-0.023, 95% CI: -242 0.045, -0.001, p=0.037; Table 2). 243 The genetic variant rs16969968 was used as a proxy for smoking heaviness in UKB. 244 Despite strong evidence of an observational association between smoking heaviness and 245 neuroticism, we found no robust evidence of a causal association from the rs16969968 246 variant for smoking heaviness to neuroticism among either former or current smokers (Table  247 2). Using the rs3025343 variant as a proxy for smoking cessation found no strong evidence 248 of an association from smoking cessation to neuroticism (β=0.029, 95% CI: -0.01, 0.07, 249 p=0.161; Table 2) in UKB. 250 When looking at the association between extraversion and smoking, we found weak 251 evidence of an association between smoking initiation and increased extraversion (β=0.198, 252 95% CI: -0.03, 0.42; Table 2) using two-sample MR. There was no relevant measure of 253 extraversion in UKB, so we were unable to look at the association from smoking initiation to 254 extraversion using individual level data in UKB. 255 256

Effects of personality traits on smoking 257
Two-sample MR using summary statistics found no clear evidence of an association 258 from neuroticism to smoking initiation (OR=1.165, 95% CI: 0.71, 1.91, p=0.499; Table 3). 259 Several MR approaches were used to investigate the association from neuroticism to 260 smoking initiation. MR-Egger and weighted median approaches found no robust evidence of 261 an association after adjusting for pleiotropy and allowing for invalid instruments (Table S2). 262 Among the UKB participants, there was no clear evidence of an association from neuroticism 263 to smoking initiation when using an unweighted risk score (OR=1.000, 95% CI: 0.997, 1.003, 264 p=0.980; Table 3). When looking at the association from extraversion to smoking initiation, 265 we found no strong evidence of an effect when using a two-sample approach (OR=1.733, 266 95% CI: 0.37, 8.23, p=0.268). However, this may be due to a lack of power. The direction of 267 effect was consistent within the UKB sample, where there was strong evidence of an association. Each additional extraversion allele was associated with an increase in the odds 269 of being an ever smoker (OR=1.015, 95% CI: 1.01, 1.02, p=9.6x10 -7 ; Table 3). 270 Using an IVW approach, we found no clear evidence of an association from 271 neuroticism to smoking heaviness (β=0.050, 95% CI: -4.11, 4.21, p=0.979; Table 3). 272 However, MR-Egger suggested some evidence of biological pleiotropy (β=-0.500, p=0.026; 273 Table S2) and a bias adjusted estimate suggested some evidence of an association 274 between neuroticism and increased smoking heaviness (β=22.55, p=0.027). We also looked 275 at this association in UKB when stratifying according to smoking status and found some 276 evidence of an association. Among current smokers, the neuroticism risk score was 277 associated with increased smoking heaviness (β=0.068, 95% CI: 0.02, 0.12, p=0.009; Table  278 3). In this analysis, each additional neuroticism risk allele was associated with smoking an 279 extra 0.07 cigarettes per day. We found no robust evidence of an association from 280 extraversion to smoking heaviness when using a two-sample MR approach (β =0.017, 95% 281 CI: -16.33, 16.37, p=0.997; Table 3) or when stratifying on smoking status and investigating 282 this association in UKB. This remained the case among both former (β=-0.021, 95% CI: -283 0.08, 0.04, p=0.519) and current smokers (β=-0.038, 95% CI: -0.13, 0.05, p=0.419; Table 3). 284 When using two-sample MR with summary statistics, we found no robust evidence of 285 association between neuroticism and smoking cessation when using the IVW approach, or 286 when adjusting for bias (Tables 3, S2). This remained the case when looking within UKB 287 (OR=0.997, 95% CI: 0.99, 1.00, p=0.272; Table 3). There was no strong evidence of an 288 effect from extraversion to smoking cessation when using two-sample MR with summary 289 statistics (OR=0.586, 95% CI: 0.07, 4.76, p=0.387; Table 3). When restricting our analyses 290 to current and former smokers within UKB, we found weak evidence of an association. Each 291 additional increase in extraversion risk allele was associated with 1.1% lower odds of Analyses involving neuroticism were also performed restricting to participants whose 296 genetic data was not included in the interim release of UKB data. Full results are reported in 297 Table S3. Results remained largely consistent. In this subset, the strength of evidence for 298 the effect of neuroticism on smoking heaviness was weakened (current smokers: β=0.053, 299 p=0.120). However, the effect size remained consistent, so this may be due to a lack of 300 power in this smaller sample. 301 302

Discussion 303
We attempted to disentangle the relationship between smoking and the personality 304 traits of neuroticism and extraversion. Although much of the observed association between 305 smoking and personality appears to be non-causal, we found evidence of a modest genetic 306 correlation with both neuroticism and extraversion, suggesting some shared genetic 307 aetiology. Given that available GWAS summary statistics for neuroticism and extraversion 308 are not stratified by smoking status, we initially used two-sample MR approaches to look at 309 the bidirectional association with smoking initiation. This was followed by one-sample MR 310 using individual level data from UKB. When looking at the association from smoking to 311 personality, we found some evidence that the rs6265 variant for smoking initiation was 312 associated with both decreased neuroticism and increased extraversion. When looking in the 313 other direction, we found evidence that the neuroticism risk score was associated with 314 increased smoking heaviness, and that the extraversion risk score was associated with 315 increased smoking initiation and decreased smoking cessation. 316 Both two-sample MR and one-sample analyses in UKB found consistent effects of 317 the smoking initiation instrument (rs6265) on lowering neuroticism scores, although the 318 evidence for this was weak. Although this is a strong instrument for smoking initiation, there 319 could be potential pleiotropic effects. The inclusion of several strongly associated, but 320 independent variants could reduce the potential impact of these effects, as all pleiotropic 321 effects would need to operate in the same direction (Gage et al. 2017). Analyses from 322 personality to smoking behaviours found some evidence of an association between increased genetic liability for neuroticism and greater smoking heaviness using one-sample 324 MR and two-sample MR after adjusting for biological pleiotropy. Both the observational and 325 MR analyses found a stronger effect among current smokers, where at least part of this 326 association appears to be a causal effect. The observed association between smoking 327 initiation and a decrease in neuroticism scores, plus the association between increased 328 levels of neuroticism and heavier smoking would appear to lend support to the self-329 medication hypothesis. 330 We also observed evidence of an association between extraversion and smoking. 331 Unlike neuroticism, extraversion did not show evidence of a causal relationship with smoking 332 heaviness, but we did find an association with smoking initiation. Although there was no 333 strong evidence of an association when using a two-sample approach, this could be due to a 334 lack of power given that the direction of effect was consistent with that observed in UKB. 335 Using UKB data, there was evidence that individuals with a higher genetic liability for 336 extraversion had greater odds of taking up smoking. One potential mechanism for this is that 337 extraversion could lead to more social contacts and greater susceptibility to peer influences, 338 which are known to be important in smoking initiation. These findings could be taken forward 339 to develop novel interventions. If self-medication does contribute to the smoking behaviours 340 of individuals, as suggested by these results, it seems likely that targeting relevant 341 personality traits, in addition to addressing the ensuing smoking behaviours could result in 342 increased efficacy of any intervention. 343 There are a number of limitations to our analysis that should be considered. First, 344 UKB formed a large part of the discovery cohort for the GWAS of neuroticism. We were 345 therefore unable to use weighted risk scores to assess the association between smoking 346 phenotypes and neuroticism in our one sample analysesweights should be identified in 347 independent samples to avoid overfitting the data and introducing bias into effect estimates 348 (Hartwig & Davies 2016). However, we performed sensitivity analyses restricting to 349 individuals who were not included in the discovery samples, and results remained from smoking heaviness and cessation to neuroticism and extraversion because the 352 personality summary statistics were not stratified by smoking status. However, we did 353 investigate the association in both directions for neuroticism when using the UKB data. Both 354 the two-sample and UKB analyses gave consistent results when looking at the neuroticism 355 to smoking initiation relationship. Third, due to the lack of an extraversion phenotype 356 currently available in UKB we were unable to investigate whether there was evidence of an 357 effect from smoking to extraversion. Fourth, MR analyses can often suffer from a lack of 358 power, with large sample sizes and strong instruments required to detect effects. We have 359 identified genetic variants robustly associated with each trait of interest based on results of 360 large recently published GWAS in order to maximise the strength of our instruments. In this 361 analysis, we use a combination of one-and two-sample MR approaches based on GWAS 362 summary statistics in addition to data from UKB in order to maximise our power to detect any 363 effect. Fifth, we stratified on smoking status to investigate the association of smoking 364 heaviness and cessation phenotypes. Although this allows us to investigate pleiotropy, there 365 is the potential to introduce collider bias (Munafò et al. 2017). However, our instruments are 366 principally associated with smoking heaviness and cessation rather than smoking initiation, 367 so that the risk of collider bias is minimised (Gage et al. 2016). 368 In conclusion, we found evidence of modest genetic correlation with both neuroticism 369 and extraversion, suggesting some shared genetic aetiology and implying that much of the 370 observed association between smoking and personality is non-causal. However, we also 371 found some evidence for specific causal pathways between personality and smoking 372 phenotypes -higher neuroticism and heavier cigarette consumption, and higher extraversion 373 and increased odds of smoking initiation. The association between neuroticism and cigarette 374 consumption lends support to the self-medication hypothesis, while the association between 375 extraversion and smoking initiation could lead to more targeted smoking prevention 376 programmes. 377

Conflict of interest 380
None.  0.220 (0.06) 1.8x10 -5 Neuroticism and smoking cessation 2 -0.314 (0.15) 0.002 1 Genetic correlation estimates for smoking initiation were generated using GWAS summary statistics 2 Estimates for smoking heaviness and cessation were generated using individual level data from UK Biobank