Our systematic review demonstrated that the majority
Our systematic review demonstrated that the majority of NOD2-smoking studies were underpowered to detect an interaction; for example, 10 identified studies included fewer than 300 participants (de Diego et al., 2006; Giachino et al., 2004; Laghi et al., 2005; Mardini et al., 2005; Mendoza et al., 2003; Protic et al., 2008; Renda et al., 2008; Sabate et al., 2008; Walker et al., 2004; Nagy et al., 2005). Because of these small samples, the three most common NOD2 variants (1007fs, G908R, and R702W) were often combined. However, this meta-analysis confirms that the NOD2-smoking interaction in Crohn\'s disease is SNP-specific. Thus, combining SNPs to improve power may not be methodologically sound. The biological plausibility of a negative association between smoking and the 1007fs variant of NOD2 is unclear. However, the complex interplay between the different NOD2 variants and cigarette smoking, and their effects on dendritic cells may underlie the negative interaction observed between smoking and the 1007fs variant but not the NOD2 variants (Butler et al., 2007; van Heel et al., 2005; Ueno et al., 2014; Kramer et al., 2006). The 1007fs variant of NOD2 results in a protein that is truncated in the region of NOD2 responsible for intracellular detection of peptidoglycans on bacterial cell walls. The G908R and R702W variants are single aa-utp substitutions that may also alter the structure of NOD2. However, the functional impacts of each variant are not known. Alternatively, the negative relationship between NOD2 and smoking may be explained by a methodological phenomenon. Our case-only study examined the impact of age of onset on the interaction between NOD2 and cigarette smoking in patients with Crohn\'s disease. Twenty percent of NOD2 carriers were diagnosed prior to age 17, whereas only 3% were diagnosed after age 40. In contrast, nearly three-quarters of patients over the age of 40 at diagnosis had a history of smoking compared to only 4% of those diagnosed under age 17. This inverse relationship between NOD2 and smoking, across ages at diagnosis results in minimal overlap of the two factors and may misleadingly appear as a gene-environment interaction. However, the negative interaction between NOD2 and smoking persists after adjusting for age at diagnosis (Helbig et al., 2012). Supplementary Fig. 3 explains the bias introduced by combining patients with varying ages of diagnosis, resulting in a negative interaction. Consistent with prior studies, we analyzed the NOD2-smoking interaction in Crohn\'s disease using a case-only study design (Helbig et al., 2012). The case-only design increases resource and statistical efficiency to detect gene-environment interactions (Thomas, 2010). However, there are some assumptions of case-only designs including: (1) the absence of population stratification; and (2) the independence of the genetic and environmental risk factor in the source population for the cases (Liu et al., 2012; Pierce and Ahsan, 2010; Wang and Lee, 2008). Bias due to population stratification (i.e., confounding by ethnicity) arises when studies combine different ethnic groups with variable allele frequencies and prevalence of exposure (Wang and Lee, 2008). Sensitivity analyses limited to studies where all patients were non-Jewish and White were consistent with our main analyses. However, we were not able to assess the interaction between NOD2 and smoking in other ethnicities (e.g., African Americans) due to the paucity of studies. Secondly, we assumed gene-environment independence due to a prior study that found no association between NOD2 and smoking in their control population (Helbig et al., 2012) and NOD2 has not been identified in genome-wide association studies of smoking behaviors (Furberg et al., 2010). The results of this study point to the importance of evaluating genetic and environmental factors in sub-phenotypes of Crohn\'s disease. We focused on age at diagnosis. However, the findings of NOD2-smoking interaction studies may be influenced by other disease phenotypes such as disease location or disease behavior. Both NOD2 and smoking are associated with ileal Crohn\'s disease. In addition, ileal and colonic Crohn\'s disease have been shown to be genetically distinct phenotypes (Cleynen et al., 2016). Thus, our study provides one example of how phenotypic characteristics may influence gene-environment interactions.