Crohn's disease (CD) is a common form of inflammatory bowel disease (IBD) characterised by chronic relapsing and remitting episodes of transmural inflammation of the gastrointestinal tract. High rates of CD are found in North America (including Canada), northern Europe, and the United Kingdom, with an estimated 47,000 new cases diagnosed each year in North America and 41,000 new cases diagnosed each year in Europe [1]. The disease burden of CD is high, often associated with complications such as intestinal strictures, fistulas, and granulomas, the need for surgical intervention, and extraintestinal manifestations [2]. It is believed that immune tolerance to normal intestinal bacteria is disturbed in genetically susceptible individuals, leading to a pathogenic inflammatory response [3].

Evidence from twin and familial aggregation studies [4], as well as more recent evidence from genome-wide association studies [5] suggests that genetic factors play an important role in CD aetiology. Among the first and most consistently replicated genetic associations discovered in CD were three variants in NOD2 (nucleotide-binding oligomerisation domain containing 2; previously known as CARD15) [6, 7], which represent the strongest genetic risk factors for CD identified to date [8]. NOD2, an intracellular sensor of bacteria-derived muramyl dipeptide, plays an essential role in regulating host response to intestinal bacteria [9, 10]. The three main reported disease-associated polymorphisms (R702W, G908R, and 1007fs) all lie within or near a leucine rich repeat domain, and it is thought that these three CD-associated variants contribute to disease pathogenesis through interference with bacterial recognition [11, 12]. Associations between CD and the three main susceptibility variants are well-established [11, 13, 14] with a recent meta-analysis revealing disease odds ratios (OR) under a dominant genetic model of 2.2 (95% CI 2.0-2.5) for the R702W polymorphism, 2.6 (95% CI 2.2-2.9) for G908R, and 3.8 (95% CI 3.4-4.3) for 1007fs [15]. When data for all three variants are combined, the disease OR associated with simple heterozygotes is 2.4 (95% CI 2.0-2.8) and for compound heterozygotes is 9.0 (95% CI 6.0-13.5) [15].

While the association of NOD2 variants with CD is confirmed, the relatively small effect size of disease-associated NOD2 alleles even among homozygotes [16] and the observation that NOD2 knock-out mice do not develop spontaneous intestinal inflammation [9] indicate that additional factors are necessary for disease pathogenesis. Several environmental factors such as appendectomy, dietary factors, domestic hygiene, oral contraceptive use, and childhood infections have been implicated in CD [17], but the most consistent epidemiological evidence points to a clear risk associated with cigarette smoking [18, 19], which is the strongest environmental risk factor identified for CD to date [20]. Two meta-analyses reveal that current cigarette smokers are up to twice as likely to develop CD as non-smokers (OR = 1.8-2.0) [21, 22]. Furthermore, cigarette smoking is known to compromise innate and adaptive immune responses, leading to increased susceptibility to microbial infections [23], supporting its potential role in the pathogenesis of CD.

It has been speculated that an interaction between genetic and environmental factors is necessary for the development of complex genetic disorders including CD [24]. Although gene-environment interactions have often been invoked in explaining the aetiology of CD [25], studies of these interactions in CD have thus far received little research attention. Given the established role of NOD2 in regulating innate immune responses and cigarette smoke's impact on innate immunity, it is plausible that cigarette smoking might modulate the functional consequences of CD-associated polymorphisms in NOD2.

In this study, we used a case-only design to investigate potential gene-environment interactions between cigarette smoking and susceptibility variants in NOD2 among patients with CD. The case-only study design quantifies gene-environment interaction using the interaction odds ratio. Using this measure, the observed effect of both variables combined is compared to the hypothetical effect size when the effect of both factors is multiplied. Accordingly, an interaction odds ratio of 1.0 suggests the absence of interaction, whereas an interaction OR significantly greater than 1.0 suggests positive (synergistic) interaction and an interaction OR significantly less than 1.0 suggests negative (antagonistic) interaction. We hypothesised that the interaction between disease-associated NOD2 alleles and history of smoking would deviate from multiplicativity with regard to risk increase for CD (i.e. interaction OR significantly greater than or less than 1.0), thus being potentially indicative of biological interaction. We further hypothesised that this deviation would be more pronounced among patients who were active smokers at the time of CD diagnosis. The current study represents the largest study to date to examine statistical gene-environment interactions in patients with CD and the first to employ a case-only design.

In an investigation of statistical interactions between cigarette smoking and CD-associated NOD2 polymorphisms, we found a significant negative interaction between the NOD2 1007fs polymorphism and cigarette smoking. We did not observe this interaction for the R702W or G908R polymorphisms. For both smoking variables analysed, i.e. history of ever having smoked and smoking status at the time of diagnosis, the observed interaction OR between smoking and the 1007fs allele was lower than expected. These results indicate that the increase in disease risk conferred by both cigarette smoking and the 1007fs allele is significantly lower than would be expected under a multiplicative model. Accordingly, the lower than expected disease risk in individuals who both carry the risk allele and smoke suggests a negative interaction between both factors.

It is possible that age of disease diagnosis and family history of IBD might confound these results, especially given the earlier age of disease diagnosis associated with the 1007fs variant. However, our final model was adjusted for age at diagnosis and family history, which controls for potential confounding effects. Furthermore, family history of IBD was not associated with smoking status in our cohort. This indicates that age at diagnosis and family history of IBD do not confound our results.

The strong negative interaction between cigarette smoking and the 1007fs variant is intriguing given the conflicting data from previous research. Statistical interactions between CD-associated NOD2 variants and smoking were previously investigated in two case-control studies, which failed to identify any statistical interactions between smoking status and NOD2 genotypes [13, 26]. These studies, however, were limited by small sample sizes and lacked the statistical power to detect interactions of the size identified in our study. However, two small studies have suggested that smokers and non-smokers with CD have differences in their genetic backgrounds, including variants in NOD2, with the 1007fs variant found more commonly among non-smokers [27, 28]. These previous findings are consistent with our results of a negative interaction between cigarette smoking and the 1007fs allele. Taken together, these data are suggestive that cigarette smoking modifies the disease risk associated with the 1007fs variant, and the increase in disease risk for smokers who carry this susceptibility allele compared to non-smokers who carry this allele is significantly less than expected.

To our knowledge the current study represents the largest analysis to date of statistical gene-environment interactions in CD aetiology and the first to employ a case-only design. Previous studies investigating gene-environment interaction in CD [13, 26] did not employ the case-only design, which has been increasingly recognised over the last two decades as an attractive method for assessing the presence of statistical interaction [29–32]. The case-only design offers substantial increases in power as well as greater precision in estimating interactions when compared to traditional case-control designs [31].

Although our findings are promising, the current study does have limitations. The main limitation is that the case-only design only allows for the assessment of statistical interactions under a multiplicative model, which is considered by many epidemiologists to be less meaningful than an additive scale [33]. Furthermore, case-only designs are particularly sensitive to the assumption of independence of the genetic and environmental risk factors in the general population [34]. However, we observed no association between any of the CD-associated risk alleles and history of ever having smoked in a sample of 718 German population-based controls, which supports the assumption that these risk factors are independent. Additionally, a recent meta-analysis suggests that bias in case-only studies is rare, and no differences in OR estimates between traditional case-control and case-only studies of gene-environment interactions were observed [32]. Our study was further limited by the definition of smoking status. In the questionnaire administered to participants, no definition of active smoking was specified. Participants were asked only if they had ever smoked, which may have led to misclassification of the environmental variable among some participants. However, if the rates of misclassification of smoking status and genotypes are non-differential, that is if the classification of smoking status is independent of genotype classification, as one would expect, misclassification would push the interaction OR towards unity rather than generate an inflated estimate of interaction [33].

Despite these limitations, statistical interaction represents a powerful tool to point to potentially meaningful biological interaction. Evidence from functional studies [35] and an animal model of the 1007fs polymorphism [10] suggests that the protein product encoded by this allele exhibits gain of function properties, resulting in an impairment of innate immune tolerance towards commensal bacteria. The three CD-associated NOD2 polymorphisms furthermore have been shown to cause a failure to induce autophagy via ATG16L1, impairing autophagic wrapping and removal of invading bacteria [36]. It has also been observed that cigarette smoke extract delays NOD2 mRNA expression and impairs NOD2 functioning in intestinal epithelial cells [37]. It is reasonable to then speculate that the detrimental gain of function effects associated with the 1007fs associated protein may be modulated by the biological effects of cigarette smoking. The immunosuppressive properties of nicotine, one of the best studied components of cigarette smoke, are well-established [23] and have been shown to persist several weeks after exposure [38].

Our study indicates that smoking may modify the overall disease risk associated with established genetic risk variants. Therefore, it is plausible to suggest that similar gene-environment interactions may exist that influence the clinical manifestations and severity of CD. Indeed it has recently been demonstrated that cigarette smoking interacts with CD risk alleles, including NOD2 variants, to influence disease location in patients with CD [39]. Further research is necessary to elucidate the potential biological mechanisms behind such gene-environment interactions as well as to understand how behaviour modification could potentially benefit patients carrying disease-associated NOD2 variants.

Our finding of a significant negative statistical interaction between cigarette smoking and the 1007fs NOD2 polymorphism in patients with Crohn's disease is intriguing and warrants further investigation in epidemiological and functional studies. Confirmation of a mechanistic interaction by which cigarette smoking modulates the risk associated with disease-associated NOD2 alleles may be important in not only better understanding the biological mechanisms underlying CD pathogenesis, but also in developing appropriate recommendations for individuals at high risk of developing CD.