Common genetic polymorphisms modify the effect of smoking on absolute risk of bladder cancer

Abstract

Bladder cancer results from the combined effects of environmental and genetic factors, smoking being the strongest risk factor. Evaluating absolute risks resulting from the joint effects of smoking and genetic factors is critical to evaluate the public health relevance of genetic information. Analyses included up to 3,942 cases and 5,680 controls of European background in seven studies. We tested for multiplicative and additive interactions between smoking and 12 susceptibility loci, individually and combined as a polygenic risk score (PRS). Thirty-year absolute risks and risk differences by levels of the PRS were estimated for US-males aged 50-years. Six out of 12 variants showed significant additive gene-environment interactions, most notably NAT2 (P=7×10^-4) and UGT1A6 (P=8×10^-4). The 30-year absolute risk of bladder cancer in US males was 6.2% for all current smokers. This risk ranged from 2.9% for current smokers in the lowest quartile of the PRS to 9.9% for current smokers in the upper quartile. Risk difference estimates indicated that 8,200 cases would be prevented if elimination of smoking occurred in 100,000 men in the upper PRS quartile, compared to 2,000 cases prevented by a similar effort in the lowest PRS quartile (P-additive =1×10^-4). The impact of eliminating smoking the on number of bladder cancer cases prevented is larger for individuals at higher than lower genetic risk. Our findings could have implications for targeted prevention strategies. However, other smoking-related diseases, as well as practical and ethical considerations, need to be considered before any recommendations could be made.

INTRODUCTION

Bladder cancer is a smoking related disease that occurs most frequently in males living in industrialized countries ¹. According to GLOBOCAN (http://globocan.iarc.fr/), an estimated 382,660 new cases of bladder cancer were diagnosed worldwide in 2008, and 68,812 of those cases were diagnosed in the United States. Bladder cancer has high morbidity and represents an important public health problem since most cancers present as “superficial” tumors that recur frequently and require regular follow-up screening and intervention ². Thus, effective bladder cancer prevention strategies could have an important public health impact.

Well-characterized polymorphisms in two carcinogen-metabolizing genes, NAT2 and GSTM1 are associated with bladder cancer risk ^3,4. More recently, genome-wide association studies (GWAS) have identified additional common genetic susceptibility variants ^5-11, which provide important clues into underlying biological pathways. For instance, a recent GWAS identified the UGT1A region as a bladder cancer susceptibility locus ⁶, and further fine mapping and functional work identified a functional variant in the UGT1A6 gene involved in the detoxification of bladder carcinogens ¹².

Studies of gene-environment interactions can provide insights into biological mechanisms of disease and could have public health implications ^13,14. However, there are very few examples of established gene-environment interactions in cancer. A notable example is the interaction between NAT2 acetylation and smoking in bladder cancer, by which subjects with the slow NAT2 acetylation genotype have a higher relative risk from smoking than those with the rapid/intermediate acetylation genotypes ³. In contrast, other susceptibility loci do not appear to modify the relative risk of smoking associated with bladder cancer ^3,6,15,16.

Although conventional assessment of gene-environment interactions focus on multiplicative scale (i.e. evaluating if the relative risk for smoking varies across levels of genetic risk), assessment of interactions on an additive scale (i.e evaluating if the risk difference for smoking varies across levels of genetic risk) is more relevant for assessing public heath effects such as if the number of cancers that could be prevented by an intervention differs for subjects at different levels of genetic risk. In this report, we used data from seven studies in the NCI-GWAS and a novel approach to study additive gene-environment interactions that could be useful to evaluate the potential implications for targeted prevention strategies, as well as to provide additional biological insights.

METHODS

Data Collection and Definitions

Analyses are based on data from seven studies participating the NCI-GWAS ^6,9. These included two case-control studies (Spanish Bladder Cancer Study (SBCS), and the New England Bladder Cancer Study (NEBCS)), and five prospective cohorts (Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial (PLCO), The American Cancer Society Cancer Prevention Study II Nutrition Cohort (CPS II), Alpha-Tocopherol, Beta-Carotene Cancer Prevention Study (ATBC), Nurse’s Health Study (NHS) and Health Professionals Follow up Study (HPFS)). Cases were defined as histologically confirmed primary carcinoma of the urinary bladder including carcinoma in situ (ICD-0-2 topography codes C67.0-C67.9 or ICD9 codes 188.0-188.9 and 2337). Cases and controls were of European background as determined by population sub-structure analyses with STRUCTURE ¹⁷ and principal component analysis for subjects with scan data or by self-report otherwise.

Age and smoking information was obtained from risk factor questionnaires administered at the time of enrollment into the studies. For case-control studies this corresponded to the time of diagnosis for cases and time of selection for controls ^18,19. However, the time between questionnaire and diagnosis/selection varied in prospective cohorts depending on when subjects were diagnosed/selected after enrollment. Smoking habits were characterized as smoking status (never/ever and subcategories of ever smokers for current/former/occasional), average number of cigarettes smoked and total duration of smoking, as previously described for SBCS¹⁸ and NEBCS¹⁹. These two studies used the same smoking questions and smoking variables derived from other studies were harmonized based on the information reported in their questionnaires. Smoking status was missing from a total of 21 subjects.

Each participating study obtained informed consent from study participants and approval from its IRB for this study.

In this manuscript we define absolute risk as the probability of developing bladder cancer, relative risk as the ratio of the absolute risk in smokers over the absolute risk in non-smokers, and risk difference as the difference between the absolute risk in smokers minus the absolute risk in non-smokers.

Genotyping and Quality Control

This report includes data on 12 SNPs previously identified as susceptibility variants for bladder cancer^3,4,5-11, and genotyped in the NCI Core Genotyping Facility. Samples from five studies included in stage I of the NCI-GWAS (SBCS, the Maine and Vermont components of the NEBCS (NEBCS-ME, VT), PLCO, CPS II, ATBC) had been scanned with Illumina Infinium Arrays as previously described ⁶. Quality control filtering included completion rates lower than 94%–96%, heterozygosity of less than 22% of greater than 35%, discordant gender information⁶. TaqMan custom genotyping assays (ABI, Foster City, CA) for rs9642880, rs710521, rs2294008 and rs401681 were used to genotype samples from the other three studies (the New Hampshire component of the NEBCS (NEBCS-NH), NHS, HPFS) ⁶, and to genotype rs17863783²⁰ (369 controls from PLCO not genotyped for rs17863783). GSTM1 was genotyped in samples from SBCS, NEBCS (NEBCS-ME, VT), NEBCS (NEBCS-NH), PLCO, CPSII and ATBC. For PLCO and CPSII, we genotyped GSTM1 in samples from all cases and subset of controls (638 from PLCO and 486 from CSPII) using a previously described assay³.

Each SNP was coded as the number of risk-alleles a subject carried. To assess gene-smoking interactions, SNPs were coded as dichotomous variables (see explanation under “Test for gene-environment interactions” below). The rs1495741 AA genotype (corresponding to the NAT2 slow acetylation phenotype) was compared to the combined GG and AG genotypes (corresponding to the NAT2 rapid/intermediate acetylation phenotype) ²¹; and the GSTM1 null (-/-) genotype was compared to the combined (+/-) and (+/+) genotypes (GSTM1 “present”).

Statistical analysis

Analyses were based on pooled individual data for subjects with complete smoking and SNP information. Subjects classified as occasional smokers in SBCS (N=131), NEBCS (N= 53), CPSII (N=26) considered ever smokers in analyses of ever/never smoking, but excluded from analyses comparing never/former/current smokers. Smoking status was considered the primary analysis, whereas analyses for smoking dose and duration were secondary. Logistic regression models adjusted for study, age (five-year categories) and gender were used to estimate odds ratios (OR) and 95% confidence intervals (95%CI) for genetic and smoking variables. ORs were used as an approximate measure of relative risk.

PLCO, HPFS and NHS controls were sampled by smoking status (see Supplementary Table 1 for details) and therefore these two studies did not contribute to estimates of smoking main effects. To account for stratified-sampling of controls by smoking status in the analysis, models with smoking variables included three interaction terms for smoking status and indicator variables for PLCO, HPFS and NHS. All participants in ATBC were smoker males (Supplementary Table 1) and therefore data from this studyalso did not contribute to estimate smoking main effects. Differences in estimates of smoking odds ratios by study design (case-control vs cohort) were assessed by including an interaction term for smoking and study design in a logistic regression model.

Tests for gene-environment interactions

To perform test for gene-environment interaction on relative risk scale (multiplicative interaction) we used an empirical-Bayes (EB) model fitting procedure that can gain power by exploiting the assumption of gene-environment independence in the underlying population and yet is immune to bias when the independence assumption is violated ^{22, 23}. To perform test for gene-environment interaction on the additive scale (i.e. risk difference), we performed a likelihood-ratio (LRT) test comparing an unconstrained and constrained model for joint effects using logistic regression models²⁴. Under the null hypothesis of additive model, the OR for the joint effect of a given SNP and smoking status is constrained so that the risk difference associated with one exposure (e.g., smoking) is constant across levels of other exposure (e.g. SNP), and vice versa. All tests for gene-environment interactions were performed using categorical variables (each SNP was coded as a dichotomous variable indicating the presence of any risk allele) to avoid complex numerical issues in the additive test, and to make the additive and multiplicative tests comparable. The limitations of the additive test are related to non-standard model fitting procedures when using continuous variables, such as log-additive effect of SNP alleles.

We present empirical estimates of joint ORs and compare them with those obtained under multiplicative and additive models. Expected ORs under the multiplicative model were calculated as the product of the observed main effects for smoking and SNP, and expected ORs under the additive model were calculated as the sum of the main effects minus 1²⁵.

Polygenic risk score (PRS)

As a parsimonious way of summarizing the effects of genetic variants across all loci for each subject, we construct a “polygenic risk score” (PRS) variable as the weighted sum of the risk allele counts across all loci where the weight for each individual SNP is determined by the log-odds-ratio of its association with bladder cancer risk, adjusted for study, age, gender and smoking status. PRS was estimated for a subset of 3,211 cases and 3,424 controls from SBCS, NEBCS-ME, VT, CPSII, PLCO, and ATBC with complete genotype data for all 12 SNPs. NEBCS (NEBCS-NH), NHS, HPFS were excluded from PRS analyses because not all 12 SNPs were genotyped (see methods). We derived quartiles of the PRS variable based on its distribution in the control sample. The score was calculated for the all SNPs as well as for the subset of SNPs showing significant additive interactions with smoking status for comparison. Tests for heterogeneity of PRS by tumor grade and stage were performed using logistic regression analyses in cases with tumor subtype as the outcome variable.

Population attributable risk and absolute risk

The population attributable risks (PAR) can be interpreted as the proportion of bladder cancer cases occurring in a population that is attributable to a particular exposure or combination of exposures. We estimated PAR of smoking, PRS, and the combination of both factors for white males in the USA, using data from subjects with these characteristics and complete smoking and PRS data (i.e. 1,385 cases and 1,347 controls from NEBCS-ME, VT, CPSII and PLCO). The combined PARs for smoking and PRS were computed from estimates of the main and joint ORs from logistic regression models and the joint distribution of smoking status and PRS in cases from these studies: PAR=1-Σ(ρ_j/OR_j) for j mutually exclusive strata formed by the cross-classification of smoking and the PRS, where ρ_j is the proportion of all cases in stratum j and OR_j is the OR in stratum j compared to the reference stratum ²⁶. The reference categories were never smokers and the lowest quintile of the PRS. It should be noted that this formula uses frequency distributions from cases only and therefore is not affected by selection factors in the control populations.

To compute absolute risks of bladder cancer for a given starting and end age, we used the formula from Petracci et al. ²⁷. This formula calculates the absolute risk for a combination of risk factors based on estimates of ORs for the combination of risk factors in relation to subjects without these factors, estimates of attributable risk, 5-year age-specific incidence rates for bladder cancer (source: SEER17 urinary bladder cancer age-specific incidence rates from 2005-2007 in white males), and 5-year age-specific mortality rates from causes other than bladder cancer (source:US Mortality 2000-2007, with Kaposi Sarcoma and Mesothelioma in white males). Estimates of attributable risk did not vary substantially with age and therefore a single overall estimate was used in calculations of absolute risk. 95% CIs for estimates of attributable risk and risk differences were obtained from a nonparametric bootstrap ²⁸ with 1000 bootstrap replications. Each bootstrap sample was drawn with replacement separately for cases and controls within each study, with the original number of cases and controls in each replication. For each bootstrap replication, we fitted logistic regression models to obtain estimates of relative risk, attributable risk and risk difference. The 95%CIs were calculated based on the bootstrap distribution of 1,000 estimates obtained from each bootstrap sample.

Data analysis and management was performed with R, and STATA S.E. v.11.1 (College Station TX).

RESULTS

A total of 3,942 cases and 5,680 controls from 7 bladder cancer studies were available for analyses. Characteristics of the study populations are shown in Table 1 and Supplementary Table 1.

Table 1.

Characteristics of study populationsin bladder cancer studies included in the analyses

	Cases										Controls
Study	N	Age at diagnosis			Male		Female		Smoking status		N	Age at selection			Male		Female		Smoking status
		mean	range		N	%	N	%	Never N %			mean	range		N	%	N	%	Never N %
									Former, N %										Former, N %
									Current, N %										Current, N %
Spanish Bladder Cancer Study (SBCS)	1,106	65.9	22	81	970	88%	136	12%	152	14%	1,050	64.6	20	81	925	88%	125	12%	303	31%
									428	41%									389	40%
									471	45%									270	28%
Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial (PLCO)	708	67.2	35	86	570	81%	138	19%	171	24%	1,873	na	na	na	1524	81%	349	19%	488	26%
									425	60%									818	44%
									112	16%									567	30%
The American Cancer Society Cancer Prevention Study II Nutrition Cohort (CPS II)	558	70.4	54	87	446	80%	112	20%	135	25%	729	69.4	52	89	499	68%	230	32%	329	46%
									338	62%									348	49%
									74	14%									32	5%
New England Bladder Cancer Study (NEBCS-ME, VT)	630	64.9	32	79	495	79%	135	21%	98	16%	759	63.9	30	79	578	76%	181	24%	249	34%
									318	52%									373	51%
									197	32%									114	15%
Alpha-Tocopherol, Beta-Carotene Cancer Prevention Study (ATBC)	401	69.8	52	88	401	100%	0	0%	0	0%	707	69.6	60	82	707	100%	0	0%	0	0%
									0	0%									0	0%
									401	100%									707	100%
New England Bladder Cancer Study (NEBCS-NH)	367	65.0	31	79	274	75%	93	25%	58	16%	385	65.3	30	79	277	72%	108	28%	138	37%
									189	53%									185	49%
									113	31%									54	14%
Health Professionals Follow up Study (HPFS)	110	68.4	48	86	110	100%	0	0%	35	32%	116	68.3	49	84	116	100%	0	0%	37	32%
									52	48%									63	54%
									22	20%									16	14%
Nurse’s Health Study (NHS)	62	60.5	37	79	0	0%	62	100%	19	31%	61	60.1	37	78	0	0%	61	100%	21	34%
									14	23%									20	33%
									29	47%									20	33%
All studies combined	3,942	66.9	22	88	3,266	83%	676	17%	668	17%	5,680	66.3	20	89	4,626	81%	1,054	19%	1,565	28%
									1,764	46%									2,196	40%
									1,419	37%									1,780	32%

Differences in the total number of cases and controls and frequencies by smoking status is due to subjects classified as occasional smokers or with missing smoking information. Smoker controls were oversampled in PLCO, HPFS and NHS.

Evaluation of interactions between smoking and individual SNPs

The observed joint relative risk (OR) for the combined effect of ever smoking and the risk allele for each individual SNP were larger than expected under additive effects (Table 2). The test for additive interaction was significant at P<0.05 for 6 out of the 12 SNPs, while the test of multiplicative interaction was significant only for rs149574 in NAT2 (P=0.029). A similar pattern was seen when comparing current and former smokers to never smokers, but weaker evidence for interactions was found when comparing current to former smokers (Supplementary Tables 2A, 2B and 2C). Of note, SLC14A1 showed stronger evidence for additive interactions for current vs never (P=0.002) and former vs never (P=0.015), than for ever vs never (P=0.053). We observed no significant multiplicative or additive interactions when comparing former and current smokers, except for a multiplicative interaction for GSTM1 (P=0.027). Evaluation of multiplicative interactions with smoking dose (average cigarettes per day) and duration (total number of years smoked) for each of the SNPs showed a significant interaction only for rs149574 in NAT2 and smoking dose (EB p-interaction =0.006)).

Table 2.

Odds ratios¹ for joint associations of smoking status (ever vs never smokers) and 12 susceptibility variants with bladder cancer risk

						Observed ORs (95%CIs) 1			Expected OR joint2		P-values interaction3
Chr location (gene/s in neighbouring region)	rs number [ risk allele]	N4	Cases	Controls	RAF5	OR SNP	OR smoking	OR joint	Additive	Mult.	Additive	Mult.
8p22 (NAT2)	rs1495741 [A]	8	3,927	5,662	0.77	0.97 (0.81, 1.17)	2.02 (1.68, 2.42)	2.48 (2.08, 2.96)	1.99	1.96	6.6×10-4	0.029
1p13.3 (GSTM1)	[del]	6	3,619	3,927	0.71	1.70 (1.38, 2.09)	3.30 (2.71, 4.03)	4.69 (3.86, 5.69)	4.00	5.61	0.008	0.126
8q24.21 (MYC)	rs9642880 [T]	5	3,392	5,103	0.45	1.31 (1.05, 1.64)	2.73 (2.17, 3.44)	3.49 (2.80, 4.35)	3.04	3.58	0.035	0.850
3q28 (TP63)	rs710521 [A]	5	3,386	5,105	0.74	1.37 (0.91, 2.06)	2.89 (1.84, 4.54)	3.68 (2.45, 5.53)	3.26	3.97	0.282	0.747
8q24.3 (PSCA)	rs2294008 [T]	6	3,710	5,433	0.47	1.08 (0.88, 1.34)	2.43 (1.95, 3.02)	2.90 (2.37, 3.56)	2.51	2.63	0.033	0.425
5p15.33 (CLPTM1L)	rs401681 [C]	5	3,393	5,112	0.55	1.06 (0.82, 1.36)	2.50 (1.90, 3.27)	2.87 (2.23, 3.70)	2.55	2.63	0.126	0.544
4p16.3 (TMEM129 TACC3-FGFR3)	rs798766 [T]	8	3,929	5,663	0.20	1.21 (1.00, 1.46)	2.35 (2.02, 2.72)	2.72 (2.33, 3.18)	2.56	2.84	0.327	0.701
22q13.1 (CBX6, APOBEC3A)	rs1014971 [T]	8	3,932	5,638	0.65	1.15 (0.86, 1.55)	2.08 (1.51, 2.87)	2.71 (2.03, 3.64)	2.23	2.40	0.036	0.465
19q12 (CCNE1)	rs8102137 [C]	8	3,934	5,663	0.33	1.31 (1.09, 1.58)	2.55 (2.14, 3.03)	2.85 (2.41, 3.38)	2.86	3.35	0.961	0.133
2q37.1 (UGT1A family)	rs11892031 [A]	8	3,928	5,655	0.92	1.24 (0.95, 1.61)	2.30 (1.74, 3.05)	2.89 (2.23, 3.74)	2.54	2.85	0.106	0.934
2q37.1 (UGT1A6)	rs17863783 [G]	8	3,914	5,299	0.97	1.63 (0.95, 2.78)	1.92 (1.07, 3.46)	3.83 (2.24, 6.54)	2.55	3.12	8.8×10-4	0.492
18q12.3 (SLC14A1)	rs10775480 /rs10853535 [T]	8	3,883	5,635	0.43	1.20 (0.98, 1.47)	2.31 (1.88, 2.82)	2.84 (2.34, 3.44)	2.51	2.77	0.053	0.833

¹ORs estimated from logistic regression models including main effects for SNP and smoking status (ever, never) and an interaction term for SNP*smoking status, and adjusted for study, age, and gender. ORs for each SNP are for the risk allele, assuming dominant effects.

²Expected ORs for the joint association of SNP and smoking status under additive and multiplicative (Mult.) models. The expected ORs under an additive model are calculated as OR SNP+OR smoking-1. The expected ORs under a multiplicative model are calculated as OR SNP*OR smoking

³P-values from 1df tests for additive and multiplicative (Mult.) interactions between smoking status and SNP. P-values <0.05 are bolded.

⁴N: number of studies included in analyses (NEBCS-ME, VT and NEBCS-NH are counted as one study). GSTM1 and rs2294008 data missing in NHS and HPFS. rs9642880, rs710521, rs401681 data missing in NEBCS (NH), NHS and HPFS.

⁶Differences in the number of cases and controls in each study and the total number of cases and controls in Supplementary Table 1 are due to missing genotype or smoking information (see methods section for details).

⁶RAF: risk allele frequency in the control populations. The rs1495741 AA genotype (corresponding to the NAT2 slow acetylation phenotype and with a frequency of 0.60 in controls) was compared to the combined GG and AG genotypes (corresponding to the NAT2 rapid/intermediate acetylation phenotype” ²¹ in all analyses. The GSTM1 null (-/-) genotype with a frequency of 0.51 was compared to the combined (+/-) and (+/+) genotypes (GSTM1 “present”) in all analyses. The rare homozygous and heterozygous genotypes for rs11892031 (AC/CC) and rs17863783 (GT/TT) were combined and compared to the common homozygous genotypes carrying two risk alleles (AA for rs11892031 and GG for rs17863783) for all analyses due to the low frequency of the rare homozygous genotypes.

Evaluation of interactions between smoking and PRS

To summarize the genetic risk associated with multiple SNPs, we constructed a PRS based on the 12 SNPs associated with bladder cancer risk. For white males in USA studies, those in the top quartile of the 12-SNP PRS had a 2.94-fold risk of bladder cancer (95%CI 2.32-3.73) compared to subjects in the bottom quartile (Table 3). The OR for smoking did not vary significantly with quartiles of the PRS, although there was a suggestion of larger ORs for those in the highest quartiles of the PRS (Table 3). The estimated relative risks for the PRS did not vary significantly by stage or grade of the tumor (data not shown).

Table 3.

Association between smoking status (never, current, former) and polygenic risk score with bladder cancer risk among white males in US studies¹

	Cases	Controls	OR2	95% CI)
Smoking status
Never	254	464	Ref.
Ever	1130	882	2.57	(2.03,3.15)
Former	849	704	2.23	(1.74,2.85)
Current	264	156	4.79	(3.43,6.70)
Polygenic risk score (12 SNPs)
Score 0 (low risk)	171	325	Ref.
Score 1	326	343	1.87	(1.46,2.39)
Score 2	392	346	2.22	(1.74,2.82)
Score 3 (high risk)	495	332	2.94	(2.32,3.73)

Smoking status association by polygenic risk score (12 SNPs)3
Ever vs Never	never	ever	never	ever
Score 0 (low risk)	37	134	98	227	1.77	(1.11,2.84)
Score 1	61	265	117	226	2.50	(1.68,3.70)
Score 2	72	320	129	217	2.82	(1.94,4.11)
Score 3 (high risk)	84	411	120	212	3.11	(2.17,4.45)
			Test for interaction 4= 0.235
Former vs Never	never	former	never	former
Score 0 (low risk)	37	98	98	174	1.55	(0.95,2.53)
Score 1	61	197	117	185	2.12	(1.40,3.19)
Score 2	72	249	129	174	2.49	(1.69,3.68)
Score 3 (high risk)	84	305	120	171	2.66	(1.83,3.87)
			Test for interaction4= 0.272
Current vs Never	never	current	never	current
Score 0 (low risk)	37	34	98	47	3.31	(1.74,6.27)
Score 1	61	66	117	35	5.29	(3.01,9.29)
Score 2	72	65	129	38	4.66	(2.69,8.08)
Score 3 (high risk)	84	99	120	36	6.15	(3.64,10.42)
			Test for interaction4= 0.373

¹Polygenic risk score (PRS) calculated as quartiles of a linear predictor including 12 SNPs in table 2. The PRS is defined only for subjects with complete data on all 12 SNPs.

²OR from logistic regression models adjusted for study, age, and gender

³Median (range) number of risk alleles for each genetic score in the control population is: 10 (5-14) for score 0; 12 (9-17) for score 1; 13 (11-16) for score 2; and 15 (13-20) for score 3 (the maximum number of risk alleles is 20 because rs11892031 and rs17863783 are considered as dichotomous variables due to the low frequency of the rare homozygous genotypes).

⁴LRT (3df)

Based on data from white males in USA studies, the estimated PAR for ever smoking was 50% (95%CI 41%-58%), the PAR for the upper 3 quartiles of the PRS was 42% (95%CI 33%-51%), while the combined PAR for smoking and PRS was 67% (95%CI 51%-78%). The 30-year cumulative risks of bladder cancer for a white male 50 years of age by increasing quartiles of the PRS were 1.3%, 2.5%, 2.9% and 3.8%. Figure 1 shows the 30-year cumulative risk of bladder cancer by smoking status and the 12-SNP PRS. Estimates of 30-year absolute risk associated with smoking status ranged between 0.9% to 2.9% for subjects in the lowest quartile of the PRS and between 1.7% and 9.9% for subjects in the highest quartile of the PRS. Similarly, the absolute risk associated with quartiles of the PRS ranged between 0.9% and 1.7% for never smokers and between 2.9% and 9.9% for current smokers.

Figure 1.

Cumulative 30-year absolute risk for bladder cancer in a 50 year old male in the USA, overall and by quartiles of a polygenetic genetic score. Risk differences (RD) are differences in absolute risk for current vs never smokers (also shown in Table 4)

Table 4 shows estimates of risk differences associated with smoking status by quartiles of the PRS. Tests for additive interactions indicate that there are strongly significant differences in risk differences for smoking status across levels of the PRS (Table 4). The p-value for additive interaction testing for differences in smoking Risk differences between the top and bottom quartiles of the risk score was 1.2×10^-7 when smoking status was defined as ever/never smoking (1df test), and it was 3.3×10^-7 when smoking status was defined as never/former/current smoking (2df test). The risk difference for current versus never smokers was four times larger for subjects in the top quartile (8.2% with 95%CI 4.7%-13.1%) compared to the bottom quartiles (2.0% with 95%CI 0.8%-3.8%); P=1.1×10^-4. Estimates of absolute risk and risk differences based on a PRS using the 7 SNPs with significant additive interactions (Table 2) yielded similar results as the 12 SNP PRS (data not shown). We chose to highlight findings from the 12 SNP PRS score since the observed joint effects were super-additive for all 12 SNPs, albeit statistically significant at alpha=0.05 for only 7 SNPs (Table 2).

Table 4.

Differences in 30-year absolute risk¹ of developing bladder cancer in USA white males aged 50 years according to smoking status and polygenic risk score (PRS).

	Cases1	Controls1	Current vs Never		Current vs Former		P-add
	N/F/C	N/F/C	RD2 (95%CI)	P-add 3	RD2 (95%CI)	P-add4	1df5	2df6
Smoking status	245/849/264	464/704/156	4.9% (3.7%-6.9%)	N/A	3.3% (2.0%-5.2%)	N/A	N/A	N/A
Smoking status by polygenic risk score7
Score 0 (low risk)	37/98/34	98/174/47	2.0% (0.8%-3.8%)	Ref.	1.5% (0.3%-3.2%)	Ref.	Ref.	Ref.
Score 1	61/197/66	117/185/35	5.3% (2.8%-8.6%)	0.020	3.8% (1.4%-7.3%)	0.084	1.9×10-3	6.4×10-3
Score 2	72/249/65	129/174/38	4.9% (2.5%-8.5%)	0.029	2.8% (0.4%-6.4%)	0.441	3.9×10-4	2.0×10-3
Score 3 (high risk)	84/305/99	120/171/36	8.2% (4.7%-13.1%)	1.1×10-4	5.4% (2.0%-10.4%)	0.063	1.2×10-7	3.3×10-7

¹Number of never/former/current (N/F/C) cases and controls

²Differences in absolute risks (RDs) are calculated based on case frequencies and ORs (adjusted for study, age, and gender) for smoking status and PRS in white males in USA studies with complete smoking (excluding occasional smokers) and PRS data (1,324 cases and 1,367 controls from PLCO, NEBCS(ME,VT), CPSII), SEER17 urinary bladder cancer age-specific incidence rates from 2005-2007 in white males, mortality rates from causes other than urinary bladder cancer from US Mortality 2000-2007, with Kaposi Sarcoma and Mesothelioma in white males.

³P-values from 1degree of freedom (df) additive interaction tests comparing RDs for current versus never smokers in different polygenic risk scores.

⁴P-values from 1df additive interaction tests comparing RDs for current and former smokers in different polygenic risk scores.

⁵P-values from 1df additive interaction tests comparing RDs for ever and never smokers in different polygenic risk scores. The ever smoking category includes former and current smokers, plus additional subjects from NEBCS classified as occasional smokers.

⁶P-values from 2df additive interaction tests comparing RDs for current, former and never smokers in different polygenic risk scores.

⁷Polygenic risk score calculated as quartiles of a linear predictor including the 12 SNPs in table 2. Median (range) number of risk alleles for each genetic score in the control population is: 10 (5-14) for score 0; 12 (9-17) for score 1; 13 (11-16) for score 2; and 15 (13-20) for score 3 (the maximum number of risk alleles is 20 because 2 SNPs (rs11892031 and rs17863783) are considered as dichotomous variables

DISCUSSION

Our analyses provide strong evidence for additive gene-environment interactions between smoking and known susceptibility loci for bladder cancer. The degree of risk stratification obtained from the combined effects of smoking and the PRS could be of significance for development of targeted risk reduction strategies in the population. Specifically, our results suggest that the projected number of cases that could be avoided by smoking prevention efforts is larger for subjects at higher than lower genetic risk. For instance, according to our estimates of smoking risk difference by PRS, if elimination of smoking occurred in 100,000 men in the lower PRS quartile, 2,000 cases would be prevented, and if elimination of smoking occurred among 100,000 men in the upper PRS quartile, 8,200 cases could be prevented.

Smoking behavior results from the complex interplay of social, cultural, behavioral and genetic factors, and recent discoveries on the genetic basis of smoking behavior ²⁹ have opened a debate on whether individualized strategies based on genetic information could improve the effectiveness of smoking prevention ³⁰. Our findings provide “proof of principle” for the potential use of information on genetic risk for smoking-related diseases in targeted smoking-cessation programs. However, additional data on gene-smoking interactions for all major smoking related diseases, e.g. lung and other smoking related cancers, respiratory tract and cardiovascular diseases, as well as considerations on the acceptability and ethical aspects of using genetic information in public health interventions, is needed before any recommendations can be made.

This report focuses on smoking since this is the strongest and most common established risk factor for bladder cancer. However, accounting for other known risk factors such as occupational exposures and arsenic ¹ and their potential interactions with genetic factors could result in larger discrimination in risk stratification. A bladder cancer risk assessment model based on case-control data and including smoking, occupational, environmental exposures and a biomarker of susceptibility, showed good risk discriminatory power ³¹. Further studies are needed to evaluate risk stratification of bladder cancer by combined effects of all known environmental factors and the emerging susceptibility SNPs from GWASs.

Given the link between biological interactions between two exposures (i.e. when the two exposures need to be present to cause the disease through a particular pathway) and super-additive effects at the population level ³², it is noteworthy that among the 12 individual SNPs studied, the strongest evidence for additive interactions were seen for NAT2 and UGT1A6. These loci code for enzymes that play a key role in Phase II detoxification and specifically enhance excretion of aromatic amines ^4,12, which are considered the most potent bladder carcinogens in tobacco smoke. In addition, our data provides support for an additive interaction between another detoxifying enzyme (GSTM1) and smoking that was previously reported in a meta-analysis of studies not included in this report³³. Thus, our analyses support the presence of biological interactions between these loci and cigarette smoking on bladder cancer risk.

To our knowledge, this is one for the first studies to evaluate gene-environment interactions on the risk difference rather than relative risk of bladder cancer or any disease, and provides a model for future studies in the post-GWAS era. Although this is one of the largest studies of bladder cancer, missing data on genotypes resulted in reduced sample size to evaluate effects of the PRS. Missing genotype data should be non-differential with respect to disease status and exposures, and thus will result in a loss of power but no bias in absolute risk, relative risk or risk difference estimates. Although the sample size was adequate to obtain risk estimates for overall bladder cancer risk, larger studies are needed to obtain subtype specific estimates, particularly low-grade/non-invasive papillary tumours and the muscle-invasive tumours. Different reference time for smoking assessment in case-control and cohort studies can lead to heterogeneity of smoking effects across studies. However, we observed no significant differences in estimates by study design within USA studies (P-heterogeneity=0.691 for ever vs never OR, 0.813 for current vs never OR and 0.734 for former vs never OR). Of note, our estimate of the proportion of bladder cancer cases that could be attributable to smoking (PAR) based on white males participating in the three USA studies agreed with the estimate recently reported in a prospective cohort study in the USA (PAR of 50% for both men and women)³⁴.

In conclusion, our analysis provides strong evidence for additive gene-smoking interactions on the risk of bladder cancer, suggesting the potential value of risk stratification in targeted prevention strategies.