Causal associations of tobacco smoking with cardiovascular risk factors: a Mendelian randomization analysis of the HUNT Study in Norway

Abstract

Background: Tobacco smoking has been associated with cardiovascular risk factors including adverse serum lipid levels, central obesity and higher resting heart rate, but lower blood pressure and body mass index (BMI). We used a Mendelian randomization approach to study whether these associations may be causal. If smoking affects cardiovascular risk factors then rs1051730 T alleles, predictors of increased smoking quantity, should be associated with cardiovascular risk factors among smokers, but not among never smokers.

Methods: Among 56 625 participants of a population-based study, we estimated associations of rs1051730 T alleles with cardiovascular risk factors and examined whether the associations differed by smoking status.

Results: Rs1051730 T alleles were associated with lower BMI and waist and hip circumferences and higher resting heart rate and estimated glomerular filtration rate (eGFR), and the associations were strongest among current smokers (P interaction 5 × 10⁻⁹ to 0.01). Rs1051730 T alleles were associated with lower systolic blood pressure and pulse pressure and higher HDL cholesterol concentrations, but these associations did not robustly differ by smoking status. There were no convincing associations of rs1051730 T alleles with waist-hip ratio, diastolic blood pressure and non-fasting serum concentrations of non-HDL cholesterol, triglycerides, glucose and C-reactive protein.

Conclusions: This Mendelian randomization analysis provides evidence that smoking may cause lower BMI and waist and hip circumferences and higher resting heart rate and eGFR. The findings further suggest that smoking is not a major determinant of waist-hip ratio or adverse blood pressure, serum lipid or glucose levels.

Smoking, body mass index, blood pressure, heart rate, cholesterol, glomerular filtration rate

Key Messages

• In observational studies, tobacco smoking has been associated with adverse serum lipid levels, central obesity and higher resting heart rate and estimated glomerular filtration rate (eGFR), but lower blood pressure and body mass index (BMI).

• We used genetic variation in rs1051730 as a measure of smoking exposure to examine whether these associations are causal.

• The results strongly suggest that smoking may cause lower BMI and waist and hip circumferences, and higher resting heart rate and eGFR.

• Although smoking may cause lower body mass, it does not appear to cause a favourable fat distribution, as indicated by waist-hip ratio.

• Smoking does not appear to be a major determinant of adverse blood pressure, serum lipid or glucose levels.

Introduction

In observational studies, tobacco smoking has been associated with several cardiovascular risk factors, including higher serum concentrations of total and low-density lipoprotein (LDL) cholesterol and triglycerides, lower concentrations of high-density lipoprotein (HDL) cholesterol,^1,–4 central obesity,^5,–7 higher resting heart rate,^3,8,–10 and higher serum concentrations of C-reactive protein (CRP).¹¹ On the other hand, smokers tend to have lower blood pressure^3,9,10,12 and body mass index (BMI).^3,5,13,-15 For impaired glucose homeostasis and type 2 diabetes, observational evidence suggests that smokers may be at increased risk,^16,–19 whereas the results of a randomized controlled trial suggested that smoking cessation may increase the risk of type 2 diabetes, possibly due to weight gain.²⁰ Further, smoking has been associated with increased risk of chronic kidney disease^21,22 and also with increased glomerular filtration rate (GFR),^23,24 suggesting glomerular hyperfiltration that may occur early in the development of chronic kidney disease.²³ It is not known whether the observational associations of smoking with these cardiovascular risk factors are causal or confounded by factors such as adverse lifestyle, socio-economic position or ill health.

Recently, genome-wide association studies have provided strong evidence that common genetic variants in the nicotinic acetylcholine receptor gene cluster (CHRNA5-CHRNA3-CHRNB4) on chromosome 15q25 are associated with higher smoking quantity.^25,–28 Among smokers, each additional T allele of the single nucleotide polymorphism (SNP) rs1051730 in the CHRNA3 gene is associated with increased intensity of smoking corresponding to approximately one additional cigarette per day.^28,29 This genetic variant may serve as an unconfounded measure of smoking exposure because of Mendelian randomization: During gamete formation and conception, allele pairs segregate independently of environment and genes are essentially randomly assorted from parents to offspring. Therefore, genotypes associated with smoking are not likely to be associated with environmental factors that may confound conventional observational studies of smoking.^30,–32 If smoking affects cardiovascular risk factors then rs1051730 T alleles should be associated with cardiovascular risk factor levels among smokers (in whom the T alleles predict higher smoking quantity), but not among never smokers (in whom rs1051730 variants cannot be associated with smoking quantity).^30,33,CHRNA5-CHRNA3-CHRNB4 variants predicting increased smoking quantity have recently been associated with lower BMI^33,–35 and systolic blood pressure,³⁴ and the association with BMI may be restricted to smokers.³³ Associations of these genetic variants with other cardiovascular risk factors are not known. Using genetic variation in rs1051730 as a measure of smoking exposure, we examined the causal associations of smoking with adiposity, blood pressure, resting heart rate, estimated glomerular filtration rate (eGFR), and serum concentrations of lipids, glucose, and CRP in a large Norwegian population study.

Methods

Study population

All adults aged 20 years or older in Nord-Trøndelag county in Norway were invited to participate in the second survey of the HUNT Study between 1995 and 1997. In total, 93 898 people were invited, and 65 215 (69%) participated. The study has been described in detail elsewhere.^36,37 Briefly, the participants completed a self-administered questionnaire that covered a wide range of health topics, including smoking habits, physical activity, alcohol consumption, socio-economic characteristics and history of cardiovascular disease, hypertension and diabetes. Clinical measurements included height, weight, waist and hip circumferences, blood pressure and resting heart rate, and a non-fasting blood sample was drawn from each participant. The population is predominantly (>97%) Caucasian.³⁶

Genotyping

DNA was extracted from blood samples and stored at HUNT Biobank, Levanger. The rs1051730 polymorphism was genotyped at HUNT Biobank using a TaqMan assay (Assay ID: C_9510307_20, Applied Biosystems) on an Applied Biosystems 7900HT Fast Real-Time PCR System, as previously described.³⁸ The call rate cut-off was set to 90%. In total, DNA was available for 57 082 (87.5%) of 65 215 participants and, among them, rs1051730 was successfully genotyped in 56 664 (99.3%) participants.

Cardiovascular risk factors

Height and weight were measured with the participants wearing light clothes without shoes, and BMI was calculated as weight in kilograms divided by the squared value of height in metres. Waist and hip circumferences were measured using a steel band with the participant standing and the arms hanging relaxed. Waist circumference was measured horizontally at the height of the umbilicus, and hip circumference was measured at the thickest part of the hip. As a measure of central obesity, we calculated the ratio of waist to hip circumferences (waist-hip ratio).

Blood pressure and resting heart rate were measured three times at 1-min intervals using an automated noninvasive blood pressure monitor based on oscillometry (Dinamap 845XT; Critikon, Tampa, Florida, USA). The mean values of the second and third measurements of systolic and diastolic blood pressure and resting heart rate were used in the analyses. We calculated pulse pressure as the difference between systolic and diastolic blood pressures.

Non-fasting serum concentrations of total and HDL cholesterol, triglycerides, glucose and creatinine were measured in fresh serum samples at Levanger Hospital, Nord-Trøndelag Hospital Trust on a Hitachi 911 Autoanalyzer (Hitachi, Mito, Japan) with reagents from Boehringer Mannheim (Mannheim, Germany). Lipids were measured using enzymatic colourimetric methods, and glucose was measured using an enzymatic hexokinase method.³⁶ As a measure of the cholesterol content of atherogenic lipoproteins, we calculated non-HDL cholesterol as the difference between total and HDL cholesterol concentrations. Creatinine was measured using the Jaffe method and subsequently recalibrated to the Roche enzymatic method.³⁹ From the recalibrated values, we calculated eGFR using the re-expressed four-variable Modification of Diet in Renal Disease (MDRD) formula,⁴⁰ which estimates GFR based on serum creatinine, age, sex and ethnicity.

All participants from four of 24 municipalities were selected for serum measurement of high-sensitivity CRP, which was performed after approximately 2 years of serum storage at −80°C, using the CRP (Latex) US assay (Hoffman–La Roche, Switzerland) with a quantification range of 0.1–20 mg/l.⁴¹ Seven percent of samples had undetectable CRP and were assigned the value 0.03 mg/l, corresponding to the detection value of the assay.

Statistical analyses

Among 56 664 participants successfully genotyped for rs1051730, 56 625 (99.9%) reported smoking habits and were included in the analyses.

To illustrate the associations of self-reported smoking quantity with cardiovascular risk factors, we used linear regression analysis to estimate mean levels [with 95% confidence intervals (CI)] of each risk factor by self-reported number of cigarettes currently smoked per day. Number of cigarettes was expressed by a restricted cubic spline with knots at 1, 5, 10, 15, 20 and 30 cigarettes per day. Never and former smokers were assumed to smoke zero cigarettes a day. We adjusted for sex, age and age squared. HDL cholesterol, triglyceride, glucose and CRP concentrations and eGFR were log-transformed due to non-normal distribution and, for these variables, we report geometric means. In separate analyses, we: (i) examined whether statistical adjustment for education (primary/lower secondary, upper secondary or college/university), marital status (married/cohabitant or not), current working status (paid work/self-employed or not), alcohol consumption [no, moderate or heavy (≥30 units or ≥10 times per month)] and physical activity (no, light, moderate, vigorous 1–2 h/week or vigorous ≥3 h/week) influenced the observed associations; and (ii) excluded former smokers to avoid the influence of weight gain that may occur following smoking cessation.⁴²

We then used the number of rs1051730 T alleles as an unconfounded measure of smoking exposure to investigate the causal associations of smoking with other cardiovascular risk factors. If smoking affects other cardiovascular risk factors, we would: (i) expect the number of T alleles to be associated with cardiovascular risk factor levels; and (ii) expect the associations of T alleles with cardiovascular risk factors to differ by smoking status, being strongest among current smokers and absent among never smokers.^30,33 For each cardiovascular risk factor, we therefore used linear regression analysis to estimate the mean difference (with 95% CI) in risk factor level per rs1051730 T allele both in the entire study population and, separately, among current, former and never smokers. We adjusted for sex, age and age squared. We used likelihood ratio tests to examine whether associations of rs1051730 T alleles with cardiovascular risk factors differed by smoking status, as expressed by P values for interaction. We assumed an additive genetic model because a linear association of the number of rs1051730 T alleles with smoking quantity was observed in a previous analysis of this cohort.³⁸ To evaluate whether the associations of number of T alleles with cardiovascular risk factor levels were linear, we also estimated mean levels of each cardiovascular risk factor by number of T alleles.

Smoking could influence cardiovascular risk factors both through and independently of effects on adiposity. To explore that possibility, we examined whether associations of rs1051730 T alleles were altered after statistical adjustment for BMI (expressed by a restricted cubic spline).

Our stratifying on current and former smoking could introduce collider bias⁴³ in the associations of rs1051730 variants with cardiovascular risk factors, because smokers who carry rs1051730 T alleles may be less likely to quit smoking.⁴⁴ Therefore, we repeated all analyses using the broad strata of ever and never smokers. Although likely to underestimate the true associations of rs1051730 variants with cardiovascular risk factors among smokers, the analysis of ever smokers is unlikely to be substantially influenced by collider bias, because rs1051730 variants are little associated with smoking initiation. Additionally, we examined whether rs1051730 T alleles were associated with lifestyle, socio-economic or clinical characteristics in any of the smoking strata, as the presence of such associations would indicate a potential for collider bias.

To assess whether associations of rs1051730 T alleles among smokers may differ by smoking quantity, we dichotomized current smokers at the median number of cigarettes smoked per day (10 cigarettes) and examined associations of rs1051730 T alleles separately among light (<10 cigarettes) and heavy smokers (≥10 cigarettes per day).

The statistical analyses were carried out using Stata version 12.1 for Windows (Stata Corporation, College Station, TX, USA).

The study was approved by the regional committee for medical research ethics.

Results

Among 56 625 participants, 17 528 (31.0%) were current smokers, 14 350 (25.3%) were former smokers and 24 747 (43.7%) were never smokers. Among current smokers, rs1051730 T alleles were associated with a higher number of cigarettes smoked per day. Carriers of rs1051730 T alleles were more likely to be current smokers than to be former and never smokers, and they were also slightly more likely to be ever than never smokers. Further, they tended to be slightly younger, but rs1051730 T alleles were not associated with other characteristics (Table 1).

Table 1.

Characteristics of the 56 625 study participants by number of rs1051730 T alleles

	No. of rs1051730 T alleles
Characteristic	0	1	2	P-valuea
No. of participantsb	25 125	25 092	6408
Age (years), mean	50.0	49.9	49.5	0.07
Women, %	52.1	52.4	52.8	0.53
Smoking, %
Never	44.3	43.3	42.8
Former	26.0	25.0	23.9
Current	29.7	31.7	33.3	5 × 10−9c
No. of cigarettes currently smoked per day,d mean	10.7	11.4	12.1	1 × 10−24e
Education, %
Primary/lower secondary	36.2	37.1	36.7
Upper secondary	43.8	43.5	43.3
College/university	20.0	19.4	20.0	0.23
Currently not paid work or self-employed, %	37.6	37.7	37.4	0.85
Not married or cohabitant, %	26.3	26.3	27.1	0.40
Heavy alcohol consumption,f %	4.9	5.1	4.9	0.44
Vigorous physical activity <1 h/week, %	72.9	73.2	73.7	0.45
History of cardiovascular disease,g %	8.0	8.0	7.6	0.51
Currently treated hypertensionh	11.0	11.1	10.9	0.84
Diabetes mellitush	3.1	3.0	2.9	0.77

^aχ² test for categorical variables and t test for linear associations according to no. of T alleles.

^bFor each variable, the proportion of participants with available information ranges from 90% to 100%.

^cP-value for the association of rs1051730 T alleles with ever smoking, 0.02.

^dAmong current smokers.

^eF = 105.68.

^f≥30 units or ≥10 times per month.

^gMyocardial infarction, angina pectoris or stroke, by self-report.

^hBy self-report.

Self-reported smoking quantity and cardiovascular risk factors

We estimated the age- and sex-adjusted associations of self-reported current smoking quantity with measures of adiposity (Figure 1), blood pressure, resting heart rate (Figure 2), non-fasting serum concentrations of lipids, glucose and CRP, and eGFR (Figure 3). Compared with non-smokers, light smokers (approximately five cigarettes a day) had lower BMI, waist and hip circumferences, systolic and diastolic blood pressure, pulse pressure and HDL cholesterol and glucose concentrations, but higher resting heart rate and eGFR. Compared with light smoking, heavier smoking was associated with adverse levels of all cardiovascular risk factors except for pulse pressure (Figures 1–3). The associations of heavy smoking were moderately attenuated after multivariable adjustment, except for the associations with HDL cholesterol, CRP and eGFR (Supplementary Figures 1–3, available as Supplementary data at IJE online). Associations of current smoking quantity with cardiovascular risk factors remained essentially unchanged after exclusion of former smokers (Supplementary Figures 4–6, available as Supplementary data at IJE online).

Figure 1.

Mean (95% CI) body mass index (A), waist circumference (B), hip circumference (C) and waist-hip ratio (D) by self-reported number of cigarettes currently smoked per day, adjusted for sex, age and age squared.

Figure 2.

Mean (95% CI) systolic (A) and diastolic (B) blood pressure, pulse pressure (C) and resting heart rate (D) by self-reported number of cigarettes currently smoked per day, adjusted for sex, age and age squared.

Figure 3.

Mean non-fasting non-HDL cholesterol concentrations (A), geometric mean non-fasting HDL cholesterol (B), triglyceride (C), glucose (D) and CRP concentrations (E), and geometric mean eGFR (F), with 95% CIs, by self-reported number of cigarettes currently smoked per day, adjusted for sex, age and age squared.

Rs1051730 variants and adiposity

Rs1051730 T alleles were associated with lower BMI and waist and hip circumferences. These associations differed by smoking status (P interaction 5 × 10⁻⁹ to 4 × 10⁻⁵) and were strongest among current smokers, in whom each additional T allele was associated with 0.24 kg/m² (95% CI: 0.15, 0.33) lower BMI, 0.46 cm (95% CI: 0.23, 0.68) lower waist circumference and 0.43 cm (95% CI: 0.26, 0.61) lower hip circumference. The associations were weaker or absent in former smokers, whereas among never smokers, T alleles were associated with modestly higher BMI and waist and hip circumferences. Rs1051730 T alleles were not associated with waist-hip ratio (Table 2).

Table 2.

Associations^a of rs1051730 T alleles with measures of adiposity in the total study population and among current, former and never smokers^b

	Mean levels by no. of T allelesc			Per T allele
	0	1	2	Mean difference	(95% CI)	P-value
Body mass index, kg/m2
Total study population	26.7	26.6	26.5	−0.06	(−0.11, −0.01)	0.02
Current smokers	26.1	25.9	25.6	−0.24	(−0.33, −0.15)	7 × 10−8
Former smokers	27.3	27.2	27.1	−0.07	(−0.17, 0.02)	0.14
Never smokers	26.8	26.9	27.1	0.12	(0.05, 0.20)	0.001
P interactiond						5 × 10−9
Waist circumference, cm
Total study population	87.0	87.0	86.7	−0.11	(−0.23, 0.01)	0.08
Current smokers	86.3	86.0	85.3	−0.46	(−0.68, −0.23)	6 × 10−5
Former smokers	88.2	88.2	87.8	−0.13	(−0.37, 0.12)	0.31
Never smokers	87.0	87.2	87.4	0.21	(0.03, 0.40)	0.03
P interactiond						4 × 10−5
Hip circumference, cm
Total study population	102.4	102.3	102.1	−0.12	(−0.22, −0.02)	0.02
Current smokers	101.2	100.8	100.3	−0.43	(−0.61, −0.26)	9 × 10−7
Former smokers	103.4	103.4	103.1	−0.14	(−0.33, 0.05)	0.14
Never smokers	102.9	103.0	103.4	0.20	(0.05, 0.35)	0.01
P interactiond						5 × 10−7
Waist-hip ratio
Total study population	0.848	0.849	0.848	0.0000	(−0.0007, 0.0006)	0.90
Current smokers	0.851	0.851	0.848	−0.0009	(−0.0021, 0.0004)	0.18
Former smokers	0.851	0.852	0.850	0.0000	(−0.0014, 0.0014)	0.96
Never smokers	0.844	0.845	0.844	0.0005	(−0.0005, 0.0015)	0.34
P interactiond						0.27

^aAdjusted for sex, age and age squared.

^bAmong 56 625 participants, the no. of participants with available information was 56 251 (99.3%) for body mass index, 56 011 (98.9%) for waist circumference, 56 017 (98.9%) for hip circumference and 56 007 (98.9%) for waist-hip ratio.

^cEstimated fixing age at the median of the total study population (48 years) and sex at 50% women.

^dP-values for interaction between smoking status and no. of rs1051730 T alleles.

Rs1051730 variants and blood pressure and resting heart rate

In the total study population, each additional rs1051730 T allele was associated with 0.27 mmHg (95% CI: 0.04, 0.49) lower systolic blood pressure and 0.20 mmHg (95% CI: 0.04, 0.36) lower pulse pressure. The associations were strongest among former smokers, but did not convincingly differ by smoking status (P interaction 0.63 to 0.65). Rs1051730 T alleles were not associated with diastolic blood pressure (Table 3).

Table 3.

Associations of rs1051730 T alleles with blood pressure and resting heart rate in the total study population and among current, former and never smokers^a

	Adjusted for sex, age and age squared						Adjusted for sex, age, age squared and BMIb
	Mean levels by no. of T allelesc			Per T allele			Per T allele
	0	1	2	Mean difference	(95% CI)	P-value	Mean difference	(95% CI)	P-value
Systolic blood pressure, mmHg
Total study population	134.6	134.4	134.0	−0.27	(−0.49, −0.04)	0.02	−0.22	(−0.44, 0.00)	0.05
Current smokers	133.0	133.1	132.6	−0.14	(−0.53, 0.24)	0.46	0.08	(−0.29, 0.46)	0.66
Former smokers	135.7	135.2	134.8	−0.45	(−0.92, 0.01)	0.05	−0.37	(−0.82, 0.09)	0.11
Never smokers	135.8	135.6	135.3	−0.20	(−0.54, 0.14)	0.24	−0.36	(−0.69, −0.03)	0.03
P interactiond						0.63			0.17
Diastolic blood pressure, mmHg
Total study population	81.6	81.6	81.5	−0.07	(−0.21, 0.06)	0.30	−0.03	(−0.17, 0.10)	0.62
Current smokers	81.3	81.4	81.1	−0.06	(−0.29, 0.17)	0.61	0.08	(−0.15, 0.31)	0.49
Former smokers	82.3	82.1	82.0	−0.14	(−0.41, 0.14)	0.32	−0.08	(−0.35, 0.19)	0.55
Never smokers	81.6	81.5	81.6	−0.03	(−0.23, 0.18)	0.79	−0.10	(−0.30, 0.10)	0.33
P interactiond						0.82			0.45
Pulse pressure, mmHg
Total study population	53.0	52.9	52.5	−0.20	(−0.36, −0.04)	0.02	−0.18	(−0.34, −0.02)	0.03
Current smokers	51.7	51.7	51.4	−0.08	(−0.35, 0.19)	0.55	0.00	(−0.27, 0.27)	0.98
Former smokers	53.4	53.0	52.8	−0.32	(−0.65, 0.02)	0.06	−0.28	(−0.62, 0.05)	0.09
Never smokers	54.1	54.1	53.7	−0.18	(−0.42, 0.07)	0.16	−0.26	(−0.50, −0.01)	0.04
P interactiond						0.65			0.31
Resting heart rate, beats/min
Total study population	72.5	72.9	73.2	0.36	(0.20, 0.51)	5 × 10−6	0.37	(0.22, 0.53)	2 × 10−6
Current smokers	75.0	75.7	76.4	0.67	(0.41, 0.94)	8 × 10−7	0.71	(0.45, 0.98)	1 × 10−7
Former smokers	71.3	71.7	71.8	0.25	(−0.06, 0.57)	0.11	0.28	(−0.03, 0.59)	0.08
Never smokers	71.1	71.1	71.2	0.04	(−0.19, 0.28)	0.70	0.02	(−0.21, 0.25)	0.87
P interactiond						0.002			0.0004

^aAmong 56 625 participants, the no. of participants with available measurements was 56 456 (99.7%) for blood pressure and 56 417 (99.6%) for resting heart rate.

^bThe no. of participants excluded from this analysis due to lack of information on BMI was 367 for blood pressure and 364 for resting heart rate.

^cEstimated fixing age at the median of the total study population (48 years) and sex at 50% women.

^dP-values for interaction between smoking status and no. of rs1051730 T alleles.

Rs1051730 T alleles were associated with higher resting heart rate. The association differed by smoking status (P interaction 0.002) and was strongest among current smokers, in whom each T allele was associated with 0.67 beats/min (95% CI: 0.41, 0.94) higher resting heart rate (Table 3).

Statistical adjustment for BMI did not materially influence the associations of rs1051730 T alleles with blood pressure and resting heart rate (Table 3).

Rs1051730 variants and non-fasting serum concentrations of lipids, glucose and CRP

Rs1051730 T alleles were not associated with non-HDL cholesterol concentrations in the age- and sex-adjusted analysis. For HDL cholesterol, each additional rs1051730 T allele was associated with 0.34% (95% CI: 0.02, 0.66) higher concentration. The association was strongest among former smokers, but did not convincingly differ by smoking status (P interaction 0.33). Rs1051730 T alleles were not associated with triglyceride concentrations in the total study population, but among current smokers, each T allele was associated with 1.16% (95% CI: 0.03, 2.28) lower triglyceride concentration (P interaction with smoking status 0.06). We observed no convincing associations of rs1051730 variants with glucose or CRP concentrations, but the association with CRP was estimated with low precision because CRP measurements were only performed in a subsample (Table 4).

Table 4.

Associations of rs1051730 T alleles with non-fasting serum concentrations of lipids, glucose and C-reactive protein (CRP) and estimated glomerular filtration rate (eGFR) in the total study population and among current, former and never smokers^a

	Adjusted for sex, age and age squared						Adjusted for sex, age, age squared and BMIb
	Mean levels by no. of T allelesc			Per T allele			Per T allele
	0	1	2	Mean difference	(95% CI)	P-value	Mean difference	(95% CI)	P-value
Non-HDL cholesterol, mmol/l
Total study population	4.62	4.64	4.62	0.008	(−0.007, 0.022)	0.30	0.012	(−0.001, 0.026)	0.08
Current smokers	4.78	4.78	4.77	−0.003	(−0.029, 0.022)	0.80	0.016	(−0.009, 0.041)	0.20
Former smokers	4.56	4.59	4.59	0.023	(−0.005, 0.051)	0.11	0.029	(0.001, 0.056)	0.04
Never smokers	4.55	4.57	4.53	−0.002	(−0.023, 0.019)	0.87	−0.011	(−0.032, 0.009)	0.27
P interactiond						0.33			0.05
HDL cholesterol,e mmol/l
Total study population	1.33	1.34	1.34	0.34%	(0.02, 0.66)	0.04	0.24%	(−0.06, 0.55)	0.12
Current smokers	1.29	1.30	1.30	0.37%	(−0.20, 0.94)	0.20	−0.11%	(−0.66, 0.43)	0.68
Former smokers	1.35	1.36	1.38	0.79%	(0.14, 1.45)	0.02	0.67%	(0.05, 1.30)	0.03
Never smokers	1.35	1.36	1.36	0.22%	(−0.26, 0.70)	0.36	0.48%	(0.02, 0.94)	0.04
P interactiond						0.32			0.08
Triglycerides,e mmol/l
Total study population	1.53	1.54	1.53	0.11%	(−0.53, 0.75)	0.74	0.38%	(−0.22, 0.98)	0.21
Current smokers	1.63	1.62	1.59	−1.16%	(−2.28, −0.03)	0.04	0.03%	(−1.03, 1.09)	0.96
Former smokers	1.54	1.56	1.53	0.31%	(−0.96, 1.59)	0.64	0.67%	(−0.53, 1.87)	0.28
Never smokers	1.47	1.48	1.48	0.65%	(−0.32, 1.62)	0.19	0.01%	(−0.89, 0.91)	0.99
P interactiond						0.06			0.69
Glucose,e mmol/l
Total study population	5.29	5.29	5.29	0.03%	(−0.22, 0.28)	0.80	0.05%	(−0.19, 0.30)	0.69
Current smokers	5.28	5.30	5.28	0.03%	(−0.39, 0.45)	0.90	0.14%	(−0.27, 0.56)	0.50
Former smokers	5.34	5.31	5.32	−0.33%	(−0.86, 0.19)	0.22	−0.29%	(−0.81, 0.23)	0.28
Never smokers	5.26	5.28	5.28	0.26%	(−0.12, 0.63)	0.18	0.19%	(−0.19, 0.56)	0.33
P interactiond						0.17			0.27
CRP,e mg/l
Total study population	0.82	0.84	0.82	1.09%	(−3.52, 5.93)	0.65	1.55%	(−2.93, 6.24)	0.50
Current smokers	1.03	1.08	1.05	2.12%	(−6.10, 11.06)	0.62	3.88%	(−4.21, 12.66)	0.36
Former smokers	0.84	0.87	0.82	1.07%	(−7.32, 10.22)	0.81	0.76%	(−7.21, 9.42)	0.86
Never smokers	0.65	0.65	0.64	−0.15%	(−7.15, 7.38)	0.97	−1.07%	(−7.78, 6.12)	0.76
P interactiond						0.91			0.63
eGFR,e ml/min per 1.73 m2
Total study population	93.7	94.0	94.7	0.51%	(0.25, 0.76)	0.0001	0.50%	(0.24, 0.75)	0.0001
Current smokers	96.1	97.1	98.0	1.00%	(0.57, 1.44)	7 × 10−6	0.90%	(0.47, 1.34)	5 × 10−5
Former smokers	92.4	92.5	92.7	0.11%	(−0.42, 0.64)	0.70	0.12%	(−0.40, 0.64)	0.66
Never smokers	92.3	92.2	92.9	0.20%	(−0.19, 0.58)	0.32	0.25%	(−0.13, 0.64)	0.19
P interactiond						0.01			0.05

^aAmong 56 625 participants, the no. of participants with available measurements was 56 570 (99.9%) for non-HDL and for HDL cholesterol, 56 586 (99.9%) for triglycerides, 56 567 (99.9%) for glucose, 8833 (15.6%) for CRP and 56 581 (99.9%) for eGFR.

^bThe no. of participants excluded from this analysis due to lack of information on BMI was 34 for CRP and 374 for the other variables.

^cEstimated fixing age at the median of the total study population (48 years) and sex at 50% women.

^dP-values for interaction between smoking status and no. of rs1051730 T alleles.

^eGeometric means by no. of rs1051730 T alleles and percentage difference in geometric means per T allele.

Smoking could influence serum lipids through effects on adiposity, and this could countervail effects that smoking may have on serum lipids through other pathways. To explore that possibility, we additionally adjusted for BMI in the statistical analyses. After such adjustment, each rs1051730 T allele was associated with 0.029 mmol/l (95% CI: 0.001, 0.056) higher non-HDL cholesterol concentration among former smokers, but there was no clear association among current or never smokers (P interaction with smoking status 0.05). Adjustment for BMI did not substantially change the association of rs1051730 T alleles with HDL cholesterol, but the association with triglycerides among current smokers considerably attenuated after adjustment for BMI (Table 4).

Rs1051730 variants and eGFR

Rs1051730 T alleles were associated with higher eGFR. The association differed by smoking status (P interaction 0.01) and was strongest among current smokers, in whom each T allele was associated with 1.00% (95% CI: 0.57, 1.44) higher eGFR. The association was essentially unchanged after statistical adjustment for BMI (Table 4).

Additional analyses

We repeated all analyses using the broad strata of ever and never smokers, but the associations of rs1051730 T alleles with cardiovascular risk factors, including interactions between T alleles and smoking status, were essentially similar in the analyses of ever vs never smokers to those in the original analyses of current, former and never smokers (Supplementary Table 1, available as Supplementary data at IJE online). To evaluate the potential for collider bias, we also examined whether rs1051730 T alleles were associated with lifestyle, socio-economic or clinical characteristics relevant for cardiovascular health, but we did not find evidence of substantial associations in any of the smoking strata (Supplementary Table 2, available as Supplementary data at IJE online).

Among current smokers, we examined associations of rs1051730 T alleles separately among light (<10 cigarettes) and heavy smokers (≥10 cigarettes per day). In both groups, rs1051730 T alleles were associated with a higher number of cigarettes smoked per day, but rs1051730 variants were not associated with other lifestyle or socio-economic characteristics in either group (Supplementary Table 3, available as Supplementary data at IJE online). Associations of rs1051730 T alleles with cardiovascular risk factors were similar in light and heavy smokers (Supplementary Table 4, available as Supplementary data at IJE online), except that the positive association with eGFR was stronger among light smokers (P interaction 0.04).

Discussion

In this Mendelian randomization analysis, we aimed to study the causal associations of tobacco smoking with other cardiovascular risk factors. Rs1051730 T alleles, which predict increased smoking quantity, were associated with lower BMI and waist and hip circumferences and higher resting heart rate and eGFR among current smokers, and there was strong statistical evidence that the associations differed by smoking status. Rs1051730 T alleles were also associated with lower systolic blood pressure and higher HDL cholesterol concentrations, but these associations did not convincingly differ by smoking status. Rs1051730 T alleles were not associated with adverse levels of any of the measures of adiposity, blood pressure, serum lipids, glucose or CRP.

For several reasons, it is plausible that the observational associations of heavy smoking with adverse levels of some cardiovascular risk factors in this and other studies^{1–10,16–19} may be confounded. First, heavy smoking may be associated with adverse lifestyle, socio-economic and clinical characteristics that cannot be fully controlled for in observational studies. Second, we observed that light smokers had favourable levels of several cardiovascular risk factors, including adiposity and blood pressure, compared with non-smokers. If the observational associations were causal, this would imply that smoking has dual effects, with light smoking having favourable effects and heavy smoking having adverse effects. Such dual effects appear biologically less plausible and are not supported by animal evidence suggesting that increasing nicotine doses may lead to lower weight without any sign of dual effects.^45,46

Genetic variants that predict increased smoking exposure provide an opportunity to study the causal effects of smoking, because associations of the genetic variants are not likely to be confounded by environmental factors or influenced by reverse causality.^31–,33 Such Mendelian randomization analyses require large sample sizes because the association of genetic variants with smoking quantity is weak. The large study population is therefore an important strength of our study. Nonetheless, we could not estimate weak associations and their interaction with smoking status with optimal precision, especially in the analysis of CRP, which was only measured in a subsample. Another limitation is that the non-fasting state of the blood sampling may have reduced our ability to detect associations with triglyceride and glucose concentrations.

Smokers that carry rs1051730 T alleles may be less likely to quit smoking,⁴⁴ and the stratified results among current and former smokers could therefore be influenced by collider bias.⁴³ However, two observations suggest that such bias will not have substantially distorted our results. First, rs1051730 T alleles were not associated with known socio-economic or lifestyle confounders within any of the smoking strata. Second, associations of rs1051730 T alleles with cardiovascular risk factors remained similar when we used the broad strata of ever and never smokers, and those results are unlikely to be substantially influenced by collider bias, because rs1051730 T alleles were only weakly associated with ever smoking.

One assumption of Mendelian randomization analyses is that the genotype should influence the outcome only through the exposure of interest. The positive association of rs1051730 T alleles with BMI and waist and hip circumferences among never smokers in our data suggests that this assumption may be violated. Possibly variants of nicotinic acetylcholine receptors, which are expressed in neurons involved in energy homeostasis,⁴⁷ may influence body mass through mechanisms other than smoking. Nonetheless, it seems unlikely that such pleiotropic effects of rs1051730 may have created the interactions of rs1051730 variants and smoking status with cardiovascular risk factors that we observed. Collider bias could possibly explain the positive association of rs1051730 T alleles with measures of adiposity among never smokers, as rs1051730 T alleles were associated, albeit weakly, with ever smoking.

If smoking affects other cardiovascular risk factors, we would expect to find associations of rs1051730 variants with cardiovascular risk factors among smokers. Conversely, we would not expect similar associations among never smokers, in whom rs1051730 variants cannot be associated with smoking quantity. Such gene-environment interaction, which we observed for measures of adiposity, resting heart rate and eGFR, has been observed previously for BMI³³ and provides evidence that smoking may cause lower BMI and waist and hip circumferences and higher resting heart rate and eGFR.⁴⁸ In our cohort, similar gene-environment interactions were observed in a study of lung cancer risk and loss of lung function.⁴⁹ The association of rs1051730 T alleles with eGFR suggests that smoking may cause glomerular hyperfiltration,²³ but catabolic effects of smoking on skeletal muscle⁵⁰ could create a similar association of rs1051730 variants with creatinine-based estimates of GFR. Nonetheless, an association of smoking with glomerular hyperfiltration has also been observed when using ⁵¹Cr-EDTA clearance to estimate GFR,⁵¹ and this method does not depend on creatinine levels.

By contrast, the associations of rs1051730 variants with systolic blood pressure and HDL cholesterol concentrations did not convincingly differ by smoking status, and we can be less confident that smoking causes lower systolic blood pressure and higher HDL cholesterol concentrations. Alternatively, rs1051730 variants may be associated with blood pressure and HDL cholesterol through mechanisms other than smoking. The CHRNA5-CHRNA3-CHRNB4 gene cluster encodes subunits of nicotinic acetylcholine receptors in the central nervous system and peripheral autonomic ganglia, and it has been suggested that variants in this gene cluster may influence blood pressure through the autonomic regulation of blood pressure.^34,52 Such an effect would expectedly be similar in smokers and non-smokers. In one randomized controlled trial, smoking cessation was followed by an increase in HDL cholesterol concentrations that appears to contradict the positive association between rs1051730 T alleles and HDL cholesterol in our data.⁵³

Smoking is a major cause of cardiovascular disease,⁵⁴ and mortality risk among people with coronary heart disease is substantially reduced after smoking cessation.⁵⁵ Despite its health hazards, some people may consider smoking as a way to reduce body weight. However, the lack of association of rs1051730 T alleles with waist-hip ratio in our data suggests that smoking does not cause a favourable distribution of body fat.

In summary, the results of this Mendelian randomization analysis strongly suggest that tobacco smoking may cause lower BMI and waist and hip circumferences and higher resting heart rate and eGFR. The findings further suggest that smoking is not a major determinant of waist-hip ratio or adverse levels of blood pressure, serum lipids or serum glucose.

Funding

This work was supported by the Research Council of Norway (grant no. 193277).

Acknowledgments

The HUNT study is a collaborative effort of the HUNT Research Centre (Faculty of Medicine, Norwegian University of Science and Technology), Nord-Trøndelag County Council, Central Norway Health Authority and the Norwegian Institute of Public Health. The HUNT Research Centre provided the data.

Conflict of interest: None declared.