Interaction between smoking and ATG16L1^T300A triggers Paneth cell defects in Crohn’s disease

Subtyping of complex immune diseases, such as inflammatory bowel disease (IBD), has traditionally been done by categorizing clinical phenotypes and disease presentations (1, 2). Incorporating key pathogenic elements, namely genetic susceptibility and relevant environmental exposure (G+E) (3), into subclassification schemes may facilitate identification of therapeutic targets in these subtypes. The identification of more than 200 susceptibility SNPs and recently identified prognosis-associated SNPs (4, 5) for Crohn’s disease (CD), one major form of IBD, poses a challenge to this approach, as many of these SNPs have been associated with gene expression and functional changes in various cell types, such as immune (6–9) and epithelial cells (10, 11). While development of gene scores has shown promise in subtyping patients (12), such scores do not account for environmental exposures that are likely to trigger phenotypes and disease. Likewise, environmental factors identified by epidemiologic studies require relevant and functional testing in preclinical models where exposure to these environmental factors is controlled. Conversely, environmental factors identified in preclinical models need to be confirmed in patients.

Development of a surrogate phenotype/biomarker that can integrate the effects from both genetics and environmental factors will facilitate subtyping of IBD. In CD, morphologic patterns of small intestinal Paneth cells (Paneth cell phenotype) are a surrogate phenotype that stratifies CD into prognostically distinct subtypes (13–15). We and others have shown that, in mouse models, knockout of CD-associated genes (Atg16l1, Xbp1, Irgm, Lrrk2) resulted in Paneth cell defects manifested as secretory granule abnormalities (11, 16–18) that are similar to those observed in CD subjects (11, 14, 15), with potential additive effects between genes (14). We previously showed that administration of a chronic strain of murine norovirus (MNV) could induce Paneth cell defects in Atg16l1^HM (hypomorph) mice, which express low levels of Atg16l1 protein (19). In human subjects, Paneth cell defects in CD are associated with microbiota changes (20) and poor clinical outcome (14, 15). Thus, Paneth cell phenotypes are biologically and clinically relevant surrogate phenotypes ideally suited for mechanistic studies and identification of potential therapeutics in CD.

One G+E trigger for Paneth cell defects in mouse models, MNV (19), as yet has no correlate in human subjects (21, 22). Therefore, our goal was to identify an environmental trigger for Paneth cell defects that occurs in both CD subjects and analogous mouse models. Among the known CD environmental risk factors (1, 23), cigarette smoking is one of the most reproducible (23, 24). It is also associated with an aggressive disease course in patients with established CD (25). A recent study suggested potential interactions between genetics and cigarette smoking (26). Based on these findings, we hypothesized that smoking would induce Paneth cell defects in genetically susceptible CD patients. As a proof of concept, we investigated the correlation of smoking exposure, Paneth cell defects, and postoperative recurrence after ileal/ileocolonic resections in CD subjects with ATG16L1^T300A, the most prevalent CD susceptibility SNP in White patients (4). We then performed functional studies using the Atg16l1^T300A mouse model to identify host factors that mediated smoking-induced Paneth cell defects. Finally, we validated rationally designed therapeutic strategies targeting these factors that result in Paneth cell defects.

CD subjects with ATG16L1^T300A were susceptible to smoking-associated Paneth cell defects. We found that in CD subjects (demographics described in Supplemental Table 1; supplemental material available online with this article; https://doi.org/10.1172/JCI120453DS1) who received ileocolonic anastomosis and postoperative immunomodulatory and/or biologics prophylactic therapy (a known confounder for outcome; n = 128), smoking status and Paneth cell phenotype were prognosticators of recurrence (Supplemental Figure 1) and the combination of these factors further stratified patients into prognostically distinct subgroups (Figure 1A). In addition, CD subjects who were of the ATG16L1^T300A genotype and who were also smokers (T300A-smoking group) showed significantly shorter time to recurrence after surgery (Supplemental Figure 2). We therefore hypothesized that cigarette smoking was a trigger for Paneth cell defects in CD subjects. Given that the most common risk allele for CD susceptibility known to be associated with Paneth cell defects was ATG16L1^T300A (11), we further hypothesized that smoking triggers Paneth cell defects preferentially in CD subjects who harbored the ATG16L1^T300A risk allele(s). In support of this hypothesis, the ATG16L1^T300A genotype in CD subjects who were smokers was associated with a lower percentage of normal Paneth cells, whereas subjects with no-risk (NR) allele were not (Figure 1, B and C, and Supplemental Table 2). We have previously described several distinct classes of abnormal Paneth cell morphology (14, 27). We determined the distribution of each subclass of abnormal Paneth cells and found that the majority of the abnormal Paneth cells were of the D2 subclass (decreased granules) (Supplemental Figure 3); this was similar to previous findings in adult CD (14, 15, 27). None of the individual abnormal morphology subclasses showed a significantly different distribution across the groups; rather, the sum percentage of these abnormal classes (or conversely, the percentage of normal Paneth cells) provided the most robust association in the T300A-smoking group (Figure 1C).

Figure 1

CD subjects with ATG16L1^T300A genotype (T300A) were more susceptible to cigarette smoking–associated Paneth cell defects. (A) In a cohort of CD subjects (n = 186) who underwent ileocolectomy, 126 received postoperative prophylaxis. Within this prophylaxis subset, smokers with type I Paneth cell phenotype (<80% Paneth cells with normal granule morphology) showed the shortest time to disease recurrence (P = 0.0183 by log-rank test). (B) Representative HD5 immunofluorescence. Scale bar: 10 μm. Asterisks indicate abnormal Paneth cells. (C) Cigarette smoking was associated with lower percentage of normal Paneth cells in patients with ATG16L1^T300A allele or alleles, while no significant differences in Paneth cell defects were seen between NR patients with or without smoking history (overall P =0.001). NR-nonsmoking, n = 25; NR-smoking, n = 14; T300A-nonsmoking, n = 84; T300A-smoking, n = 62. Data were analyzed by Kruskal-Wallis test followed by Dunn’s multiple comparison tests between groups and represent mean ± SEM. P values for comparisons between groups are shown in Supplemental Table 2. *P < 0.05; **P < 0.01.

Given that NOD2 is the other CD susceptibility gene known to be associated with Paneth cell defects in North American CD cohorts (14), we also examined the correlation among common NOD2 variant (R702W, G908R, and L1007fs) carrier status, smoking status, and Paneth cell phenotype. There were no significant changes in the percentage of normal Paneth cells in subjects carrying NOD2 variants that were smokers (Supplemental Figure 4A). We further correlated the total numbers of ATG16L1^T300A and NOD2 risk alleles, smoking status, and Paneth cell phenotype. There was no significant difference in the genetic burden regarding Paneth cell phenotype and smoking status (Supplemental Figure 4B). Therefore, smoking-induced Paneth cell defect correlated specifically with ATG16L1^T300A alleles in this cohort.

Atg16l1^T300A mice were susceptible to smoking-induced Paneth cell defects. We modeled the 4 patient populations above in a mouse model representative of the G+E interactions by exposing Atg16l1^T300A mice (28) and WT littermates to cigarette smoke for 4 weeks (Figure 2A). Paneth cell defects were triggered only in the Atg16l1^T300A mice (Figure 2B and Supplemental Table 3), recapitulating the findings in CD subjects. We also examined the distribution of each class of abnormal Paneth cells. In observations similar to those in the human cohort, the abnormal Paneth cells were predominantly of the D2 subclass, with a small percentage of D3 subclass (diminished) (Supplemental Figure 5A). Increased percentages of these 2 subclasses of abnormal Paneth cells were largely responsible for the decreased percentage of normal Paneth cells in the T300A-smoking group (Supplemental Figure 5, B–D).

Figure 2

Atg16l1^T300A mice were more susceptible to Paneth cell defects after exposure to cigarette smoking. (A) Schematic illustration of experimental design. Atg16l1^T300A (T300A) mice and WT littermates were treated with or without cigarette smoking for 4 weeks, and Paneth cell morphology was assessed. (B) Smoking induced more Paneth cell defects, specifically in Atg16l1^T300A mice (overall P < 0.0001). (A and B) WT-nonsmoking, n = 12; WT-smoking, n = 21; T300A-nonsmoking, n = 19; T300A-smoking, n = 25. Results are from 6 independent experiments. Data were analyzed by 2-way ANOVA. (C) Two weeks of cigarette smoking was sufficient to induce Paneth cell defects in Atg16l1^T300A mice (P = 0.0054), while no additional Paneth cell defects were seen with longer exposure time, up to 6 weeks (P > 0.9999). Nonexposed, n = 17; 2 weeks, n = 10; 4 weeks, n = 10. (D) After 4 weeks of cessation of cigarette exposure, the percentage of normal Paneth cells of the Atg16l1^T300A mice returned to a level comparable to that of unexposed status (P = 0.0027). Baseline, n = 6; 2 weeks, n = 7; 4 weeks, n = 7. (C and D) Data were analyzed by 1-way ANOVA, followed by Mann-Whitney U tests between groups. *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001. (B–D) Data represent mean ± SEM.

We also performed transmission electron microscopy (TEM) to investigate potential ultrastructural changes in Paneth cells. We found that Paneth cells of the T300A mice contained cytoplasmic vesicles and degenerative mitochondria (Supplemental Figure 6, A–C), which was similar to our previous observations in Atg16l1^HM mice (11). Importantly, these changes were more frequent in T300A mice exposed to smoking compared with those that were not (Supplemental Figure 6D). Paneth cells from CD patients that were of the ATG16L1^T300A genotype and smokers also possessed similar features (Supplemental Figure 6, E and F). To exclude the possibility that the Paneth cell defects in the T300A-smoking mice were the result of stress associated with the physical presence in the smoking chamber rather than exposure to cigarette smoke, T300A mice were placed in the smoking chamber and exposed to normal air pumped through the machine (i.e., no exposure to cigarette smoke). Physical presence in the smoking chamber alone with exposure to normal air did not result in Paneth cell defects in these mice (Supplemental Figure 7).

We next altered the duration of smoking to determine the impact on Paneth cell phenotypes. We found that a 2-week exposure was sufficient to trigger Paneth cell defects in Atg16l1^T300A mice, but that exposure beyond 2 weeks did not increase the percentage of defective Paneth cells (Figure 2C). We also tested the durability of smoking-induced Paneth cell defects. Paneth cell phenotype was examined in Atg16l1^T300A mice at the end of a 4-week smoking period (baseline), followed by a 2- or 4-week washout period where smoking was discontinued. Four weeks of washout (but not 2 weeks) were required to restore normal Paneth cell morphology (Figure 2D). Therefore, Paneth cell defects in Atg16l1^T300A mice after short-term smoking exposure were reversible upon smoking cessation.

We also determined whether administration of nicotine, a major component in cigarettes, would elicit similar effects on Paneth cells. We administered a daily dose of nicotine (0.7 mg/mouse/d) to mice instead of cigarette smoke. This dose is greater than the predicted absorbed nicotine dose (approximately 0.42 mg/mouse/d) achieved in the cigarette-smoking experiments based on the nicotine content of the cigarettes and known absorption kinetics (29, 30). Surprisingly, nicotine administration did not induce Paneth cell defects or crypt base apoptosis in Atg16l1^T300A mice (Supplemental Figure 8).

Gut microbiota did not alter smoking-induced Paneth cell defects. Paneth cell function is important for maintaining the homeostasis of the gut microbial community (31–33), and dysbiosis can develop as a downstream effect of Paneth cell defects or loss (20, 34, 35). Thus, we examined whether microbiota changes occurred upstream of or as part of a feedback loop with Paneth cell defects in the context of G+E. We first compared the microbial compositions in Atg16l1^T300A mice and littermates with and without smoking. There was no significant difference in microbial composition between Atg16l1^T300A mice and littermates without smoking (Supplemental Figure 9A). Smoking did not result in significant changes in α (Supplemental Figure 9, B and C) or β diversity (Supplemental Figure 9, D and E) in either Atg16l1^T300A mice or littermates. Deeper examination of specific microbial taxa showed only limited differences between the groups of mice. For example, smoking induced relatively increased abundance of Lactobacillales and Turicibacterales and reduced the abundance of Alphaproteobacteria and Betaproteobacteria in Atg16l1^T300A mice, whereas it only increased the abundance of Coriobacterales and Turicibacterales in the WT littermates (Supplemental Figure 9, F and G). Therefore, smoking only modestly altered the composition of the gut microbiota, regardless of genotype.

Because we did detect small differences in the microbiota composition that depended on smoking in Atg16l1^T300A mice, we functionally tested the microbiota for its ability to induce Paneth cell defects. We cohoused Atg16l1^T300A mice and WT littermates exposed to smoking (microbiota donors) with mice of the same genotypes not exposed to smoking (microbiota recipients) (Figure 3A). Recipients were pretreated with antibiotics (36) to allow successful colonization of donor microbiota. The recipients showed microbiota compositions indistinguishable from those of their respective donors after 4 weeks (Supplemental Figure 10). Cohousing of donors and recipients did not induce Paneth cell defects in Atg16l1^T300A recipients (Figure 3B). Therefore, the limited differences in microbial composition observed with smoking in the Atg16l1^T300A mice did not contribute to Paneth cell defects.

Figure 3

Smoking-associated Paneth cell defects in Atg16l1^T300A mice were not horizontally transmissible by cohousing. (A) Schematic illustration of experimental design. Smoking Atg16l1^T300A mice and WT littermates were used as microbiota donors and cohoused with nonsmoking, antibiotic-pretreated mice of the same genotypes as recipients. Cohousing lasted 4 weeks, during which the donors continued to receive exposure to cigarette smoking. (B) The Paneth cell defects of the Atg16l1^T300A microbiota donor mice did not transfer to recipient mice. WT donors, n = 5; WT recipients, n = 15; T300A donors, n = 5; T300A recipients, n = 17. Data analyzed by Kruskal-Wallis test followed by Dunn’s multiple comparison tests between groups. *P < 0.05; **P < 0.01. Data represent mean ± SEM.

Smoking-induced Paneth cell phenotype did not correlate with lung or systemic inflammation. To determine whether Paneth cell defects could be due to secondary changes of lung and/or systemic inflammation, we further examined the lungs for histopathology and serum for inflammatory markers. No overt inflammation was seen in the lungs in any of the mice (Supplemental Figure 11), consistent with a previous report that longer smoking exposure may be required to elicit lung inflammation (37). Likewise, none of the samples showed detectable TNF-α in serum (Supplemental Figure 12A). In addition, there was no significant difference in serum myeloperoxidase, RAGE, CXCL1, CXCL2, IL-6, or IL-1β levels among the groups (Supplemental Figure 12, B–G). Therefore, there was no correlation between lung or systemic inflammation and G+E-associated Paneth cell defects.

Natural MNV infection was an unlikely cause for smoking-induced Paneth cell defects. We previously showed that administration of a chronic strain of MNV could induce Paneth cell defects in Atg16l1^HM mice (19). To exclude the possibility that natural MNV infection, not infrequently encountered in animal facilities (19), could result in Paneth cell defects in Atg16l1^T300A mice exposed to smoking, we also determined the MNV titers in fecal samples. We found that 16% of mice were indeed infected with MNV. However, among the Atg16l1^T300A mice exposed to smoking, there was no significant difference in the percentages of normal Paneth cells between MNV-uninfected and MNV-infected mice (Supplemental Figure 13). Therefore, natural MNV infection was an unlikely cause for the Paneth cell defects observed in this study.

Smoking and Atg16l1^T300A genotype interaction led to unique host transcriptomic changes. The lack of a causative link among the microbiota, systemic inflammatory markers, and Paneth cell defects indicates that the underlying mechanisms most likely stemmed from the host intestine per se. To comprehensively analyze the effect of G+E in all ileal cell types, we performed global RNA sequencing (RNA-seq) using mRNAs isolated from full-thickness ileal sections from Atg16l1^T300A mice and WT littermates with or without smoking. Identified transcriptomic differences were categorized as associated with either genetics (G patterns), smoking (E patterns), or combinatorial effects of genetics and smoking (G+E patterns) (Supplemental Figure 14A).

The G+E patterns could be further classified into T300A-smoking and WT-smoking patterns (Figure 4, A and B, and Supplemental Figure 13, B–E). The T300A-smoking pattern was striking, as it included signatures associated with the promotion of apoptosis and the downmodulation of insulin signaling, predominantly through Ppara/g regulation (Figure 4A, Supplemental Figure 14, B–E, and Supplemental Table 4). Interestingly, Pparg activity has been shown to modulate Paneth cell function during high-fat–diet exposure (38).

Figure 4

G+E interactions resulted in unexpected transcriptomic findings. Unique transcriptomic patterns associated with (A and B) combination from both genetics and smoking interactions (G+E patterns), (C) genetics alone (G patterns), or (D) smoking exposure alone (E patterns). Full-thickness ileum from Atg16l1^T300A mice and WT littermates with or without cigarette smoking exposure was analyzed by RNA-seq. The x axes in all panels represent log P values. Significantly enriched pathways are shown.

The WT-smoking pattern included induction of genes associated with Hnf4α and Ppar regulation in addition to lipid and amino acid metabolism (Figure 4B, Supplemental Figure 14F, and Supplemental Table 5). These molecular responses to smoking were either attenuated or not activated in the Atg16l1^T300A mice.

The G patterns specific to the Atg16l1 T300A ileum included genes that function in metabolism and complement activation (Figure 4C, Supplemental Figure 14, G and H, and Supplemental Table 6). We had previously found that deletion of Atg16l1 in mouse intestinal epithelial cells targets metabolism and acute-phase reactants such as complement (11). The E patterns enriched with smoking included genes that function in cell death, IFN signaling, cyclooxygenase pathway, and gluconeogenesis (Figure 4D, Supplemental Figure 14, I and J, and Supplemental Tables 7 and 8). Of note, smoking can induce apoptosis in a mouse model of emphysema (39).

Apoptosis was a central mechanism behind smoking-associated Paneth cell defects in CD subjects and mice with ATG16L1^T300A. The analysis of the transcriptomics data suggested that apoptosis-associated cell death might be a critical mediator of G+E-induced Paneth cell defects. We found that in CD subjects, the highest level of crypt base apoptosis was present in Atg16l1^T300A-smoking as compared with all other groups (Figure 5A, Supplemental Figure 15A, and Supplemental Table 2). Further analysis showed that Paneth cells themselves were sensitive to apoptosis in ATG16L1^T300A-smoking through defensin 5 (HD5)/TUNEL colocalization (Figure 5B and Supplemental Table 2). In addition, ATG16L1^T300A-smoking, but not NR-smoking, also had lower Paneth cell numbers/crypt (Figure 5C and Supplemental Table 2). The effects of smoking and genotype showed no detectable effect on crypt proliferation, the other major function of epithelial cells in the crypt base (Supplemental Figure 16, A and B, and Supplemental Table 2).

Figure 5

Paneth cell defects were mediated by apoptosis. (A) Smoking was associated with more crypt base apoptosis in CD subjects with the Atg16l1^T300A genotype (P < 0.0001), compared with NR subjects (P > 0.9999). (B) Smoking was associated with more apoptotic Paneth cells in CD subjects with ATG16L1^T300A (P = 0.01) compared with NR subjects (P > 0.9999). (C) ATG16L1^T300A subjects who were smokers had reduced Paneth cell numbers/crypt (P = 0.0103) compared with NR subjects (P > 0.9999). (A–C) Sample sizes and data analysis were as in Figure 1. (D) In mice, smoking induced more profound crypt base apoptosis specifically in Atg16l1^T300A mice (P < 0.0001). (E) More Paneth cells were undergoing apoptosis in Atg16l1^T300A mice exposed to cigarette smoking (P = 0.0018). (F) Smoking did not induce significant alterations in Paneth cell numbers/crypt, irrespective of genotype (P = 0.0948). (G) Smoking did not induce increased apoptosis in the villi, irrespective of genotype (P = 0.5058). (D–G) Sample sizes and data analysis were as in Figure 2. Pan-caspase inhibitor Z-VAD-FMK administration prevented smoking-induced (H) Paneth cell defects and (I) crypt base apoptosis in Atg16l1^T300A mice. (H and I) Control, n = 8; pan-caspase inhibitor, n = 10; nonsmoking, n = 7. Data were analyzed by Kruskal-Wallis tests followed by Dunn’s multiple comparison tests between groups. *P < 0.05; **P < 0.01, ***P < 0.001; ****P < 0.0001. Data represent mean ± SEM.

Atg16l1^T300A mice exposed to smoking also showed increased crypt apoptosis (Figure 5D, Supplemental Figure 15B, and Supplemental Table 3) and specifically increased Paneth cell apoptosis (Figure 5E and Supplemental Table 3), confirming that apoptosis-associated cell death in the crypt base compartment was directly linked to smoking-induced Paneth cell defects. Of note, the Paneth cells that coexpressed cleaved caspase-3 and lysozyme were exclusively of the abnormal morphology (human: 92% D2, 8% D3; mouse: 97% D2, 3% D3). Atg16l1^T300A mice exposed to smoking tended to have fewer Paneth cells as compared with smoking WT littermate controls (WT-smoking), but this was not significant (Figure 5F and Supplemental Table 3). We also found that smoking did not induce apoptosis in villus epithelial cells (Figure 5G and Supplemental Figure 15C), further demonstrating that smoking and genotype specifically affect Paneth cells. Finally, crypt proliferation was not altered by exposing mice of either genotype to smoke (Supplemental Figure 16, C and D, and Supplemental Table 3). Therefore, crypt base apoptosis was a specific response to G+E, and the process did not elicit compensatory alterations in proliferation. To determine whether apoptosis mediated the smoking-induced Paneth cell defects in Atg16l1^T300A mice, we administered pan-caspase inhibitor carbobenzoxy-valyl-alanyl-aspartyl-[O-methyl]-fluoromethylketone (Z-VAD-FMK) to the Atg16l1^T300A mice. Z-VAD-FMK prevented the Paneth cell defects (Figure 5H) and crypt apoptosis (Figure 5I) induced by smoking, confirming that apoptosis is upstream of Paneth cell defects. In addition, we also determined the potential role of necroptosis in mediating Paneth cell defects (40), as a recent report has suggested a link between Atg16l1 and necroptosis (41). Administration of the necroptosis inhibitor nec-1 did not prevent the Paneth cell defect (Supplemental Figure 17A) nor crypt base apoptosis (Supplemental Figure 17B) phenotypes in T300A-smoking mice, confirming the lack of association with necroptosis in this experimental design.

Repressed Pparg activation resulted in smoking-induced crypt apoptosis and Paneth cell defects. The unique G+E patterns in the full-thickness ileal transcriptomic analysis demonstrated an attenuation or repression of Ppara/g activation in the Atg16l1^T300A mice exposed to smoking as compared with WT animals (Supplemental Figure 14, C–E, Supplemental Table 9). These data suggested the possibility that Ppara/g activation in WT-smoking mice may be protective of the normal Paneth cell phenotype. In a subsequent analysis of the G+E patterns using Enrichr to probe the GEO drug perturbations data sets (refs. 42, 43, and Supplemental Figure 18A), the WT-smoking pattern genes uniquely matched compounds with multiple highly significant adjusted P value entries (adjusted P < 0.01) (Supplemental Figure 18B). The analysis revealed that selective PPARγ agonists, including rosiglitazone, pioglitazone, and troglitazone (especially among the drug-like molecules) regulate the genes in this potentially protective pattern, whereas fibrates (PPARα agonists) were not detected in this analysis. This suggested the general Ppara/g signature observed in the pathway-level analysis may be more specific to Pparg.

To further justify analysis of this pathway in Paneth cells, we performed global transcriptional analysis of crypt base material (enriched for Paneth cells) from these mice collected by laser capture microdissection (LCM) (Figure 6A). We found that Atg16l1^T300A mice exposed to smoking showed significantly diminished expression of many Pparg-associated genes as compared with the other groups of mice in this experiment (Figure 6B and Supplemental Table 10). We also found a similar enrichment of downregulated PPARγ pathway genes in 2 specific G+E groups from our previous LCM-procured Paneth cell data sets: (a) Atg16l1^HM mice infected with MNV (ref. 19, Figure 6B, and Supplemental Table 11), and importantly, (b) CD subjects who were smokers and of the ATG16L1^T300A genotype (ref. 14, Figure 6B, and Supplemental Table 12). These data collectively suggest that the PPARγ pathway is a central mechanism closely linked to Paneth cell defects in CD subjects and relevant mouse models as a result of G+E interaction.

Figure 6

Pparg-associated metabolism dysregulation resulted in crypt base apoptosis and Paneth cell defects. (A) Workflow for Paneth cell LCM of the mice in the T300A-smoking study. (B) Significantly downregulated Pparg-associated genes were found in all 3 data sets, including the Atg16l1^T300A-smoking mice (n = 108 genes), Atg16l1^HM MNV-infected mice (n = 298 genes), and CD subjects that were ATG16L1^T300A-smokings (n = 166 genes). Adjusted P value for each group (by ChEA, https://omictools.com/chea-tool): ^#P = 0.0044; ^##P = 0.0004; ^###P = 0.038. Ppar-γ agonist rosiglitazone (Rosi) treatment rescued (C) Paneth cell defects (P < 0.0001) and (D) crypt base apoptosis (P < 0.0001) in Atg16l1^T300A mice exposed to cigarette smoking. (C and D) Total n = 9/group for nonsmoking groups; n 0/group for smoking groups. Rosiglitazone administration prevented the (E) Paneth cell defects and (F) crypt apoptosis of PC-Cre⁺ smoking mice (n = 4/group). (C–E) Statistical analysis was performed using 2-way ANOVA. *P < 0.05, ****P < 0.0001. Data represent mean ± SEM.

We next functionally tested the role of the PPARγ pathway in smoking-induced Paneth cell defects. Administration of the PPARγ agonist rosiglitazone rescued the smoking-induced Paneth cell defects (Figure 6C) and crypt base apoptosis (Figure 6D) in the Atg16l1^T300A mice. In parallel, we also generated mice with intestinal epithelium–specific Pparg deletion (Pparg/Villin-Cre mice). These mice showed reduced percentages of normal Paneth cells, reduced Paneth cell numbers/crypt, and increased crypt base apoptosis compared with the Pparg^fl/fl littermate controls (Supplemental Figure 19). Therefore, the PPARγ pathway is a critical mediator of crypt apoptosis and Paneth cell defects.

G+E interactions directly affected Paneth cells and precursors. The crypt base transcriptomic data also suggested that the G+E interaction induced effects directly on Paneth cells themselves. To test this hypothesis in vivo, we utilized a conditional knockout model in which Atg16l1 was deleted exclusively in Paneth cells (α-defensin-4-IRES-Cre Atg16l1^fl/fl mice; herein termed PC-Cre⁺ mice) (44). As seen in Supplemental Figure 20A, the PC-Cre⁺ mice elicited Paneth cell defects when exposed to cigarette smoking, whereas Atg16l1^fl/fl mice (herein termed PC-Cre^– mice) did not, as expected. The PC-Cre⁺ mice also showed increased crypt base apoptosis (Supplemental Figure 20B) and increased Paneth cell apoptosis (Supplemental Figure 20C). Finally, rosiglitazone administration also prevented smoking-induced Paneth cell defects and crypt apoptosis in PC-Cre⁺ mice (Figure 6, E and F). These data further support the notion that the G+E effect can act directly on Paneth cells and their Defa4-expressing precursors, although we cannot exclude Paneth cell–independent mechanisms with certainty.

TNF antagonism rescued smoking-induced Paneth cell defects. We previously suggested that TNF-α, a major therapeutic target in CD, may be an important mediator of Paneth cell defects in CD subjects (15). We have also shown that anti–TNF-α treatment can ameliorate the intestinal pathology in Atg16l1^HM mice infected with MNV and treated with dextran sodium sulfate (DSS) (19). Interestingly, it has also been shown that PPARγ antagonism in preadipocytes conferred increased sensitivity to TNF-α–induced apoptosis (45) and that treatment with PPARγ agonist blocked TNF-α–induced apoptosis in vitro (46), suggesting that a defective PPARγ pathway could prime the host tissue to TNF-α–induced apoptosis. In Atg16l1^T300A mice, anti–TNF-α treatment prevented Paneth cell defects (Figure 7A) and crypt apoptosis (Figure 7B). Anti–TNF-α administration did not alter the expression of genes involved in the PPARγ pathway (Supplemental Figure 21), suggesting that TNF-α acts downstream of PPARγ and mediates apoptosis induction. We further tested the role of TNF-α by crossing Atg16l1^T300A mice to TNF receptor 1–deficient (Tnfr1–deficient) mice. Atg16l1^T300A Tnfr1^–/– mice were resistant to smoking-induced Paneth cell defects (Figure 7C) and crypt base apoptosis (Figure 7D). Therefore, the TNF-α–signaling pathway is a key mediator and therapeutic target for smoking-induced crypt apoptosis and Paneth cell defects in Atg16l1^T300A mice.

Figure 7

Anti–TNF-α prevented Paneth cell defects. Administration of anti–TNF-α prevented smoking-induced (A) Paneth cell defects (P < 0.0001) and (B) crypt base apoptosis (P = 0.0011) in Atg16l1^T300A mice. (A and B) Control, n = 9; anti–TNF-α, n = 10. Compared with the Atg16l1^T300A mice, Atg16l1^T300A Tnfr^–/– mice exposed to cigarette smoking showed significantly (C) fewer Paneth cell defects (P = 0.0022) and (D) less crypt base apoptosis (P = 0.0087). (C and D) n = 6/group. Data were analyzed by Mann-Whitney U test. **P < 0.01; ****P < 0.0001. Data represent mean ± SEM.

We previously showed that Paneth cell phenotypes are associated with CD genotypes, microbiota composition, a pathologic hallmark, unique transcriptomic profiles, and clinical outcome (11, 14, 15, 19, 20). However, clinically relevant environmental trigger or triggers and the mechanism or mechanisms driving Paneth cell defects were unclear (Figure 8A). In the current study, we show that relevant environmental stimuli can trigger Paneth cell defects in genetically susceptible hosts, confirming this phenotype as a unique readout to functionally test potential genetic and environmental interactions. We also show that the G+E interactions resulted in previously unpredicted intestinal metabolic dysregulation, leading to crypt base apoptosis and Paneth cell defects mediated by PPARγ that could additionally be blocked by anti–TNF-α (Figure 8B). Finally, we show that the G+E interactions directly affect Paneth cells and their Defa4-expressing precursors, although other cell types may also be affected. The Paneth cells that underwent apoptosis were exclusively of the abnormal morphology patterns. Therefore, the abnormal Paneth cells may undergo apoptosis; however, they may also revert to normal morphology once smoking is discontinued.

Figure 8

Previous and current models of G+E resulting in Paneth cell defects. (A) Previous knowledge suggests G+E interaction may lead to Paneth cell defects, but the mechanisms have been unclear. (B) Proposed mechanistic model based on the current study. Clinically relevant genetic and environmental factors of CD could interact and induce metabolism dysfunction (defective PPARγ signaling), leading to TNF-α–mediated crypt base apoptosis and Paneth cell defects.

We focused on Atg16l1^T300A, as a knockin model that possesses the same polymorphism as CD patients, providing a mechanistic advantage over the whole gene–knockout models in select cell types. Similar mouse models for Nod2 polymorphisms exist, but this allele is much less common in CD cohorts of European ancestry (4). Other CD susceptibility genes associated with abnormal Paneth cells do not yet have mouse models of their respective genetic polymorphisms. In addition, while we have previously shown that Paneth cell defects are induced in hypomorphic Atg16l1 mice after MNV infection, we are unable to yet identify such a link in CD patients (21). Even so, smoking is a clinically relevant environmental trigger (23, 47). We show that the combination of relevant host genetic and environmental factors can provide insight into disease pathogenesis and therapeutic targets, as has been recently demonstrated in studies such as those of nonalcoholic fatty liver disease (48, 49).

The majority of the microbiome studies on the effect of smoking have centered on the oral cavity microbiota (50). A recent population-based microbiome study showed that smoking status and history showed modest effect on Bray-Curtis distance without significant associations for individual species or pathways (51). A small cross-sectional study of CD patients showed that smoking is associated with reduced microbiota diversity, with reduced abundance of limited taxa at the genera level (52). Our in vivo study was consistent with these findings. Along with the failure of horizontal transmission of Paneth cell defects in cohousing experiments, our data strongly suggest that microbiota changes were not a cause of Paneth cell defects. Overall, this supports our working model (53) showing that Paneth cell defects promote dysbiosis only in the presence of active inflammation. In addition, recent studies have suggested that necroptosis modulates Paneth cell function (40, 41). In our study, smoking-induced Paneth cell defects in T300A mice were only rescued by apoptosis inhibition, but not necroptosis inhibition, suggesting that different injuries may elicit different predominant cell death responses.

The value of the unbiased global transcriptomics approach is highlighted by the identification of Pparg as a central mediator in the T300A-smoking patterns, a finding that was not deducible from examining the effect of each single factor alone. PPARg has been shown to be downregulated in smoking-associated emphysema (54), suggesting that the modulation of this pathway by smoking is likely a general tissue response. In addition, Pparg is linked to reduced Paneth cell numbers in mice fed a high-fat diet (38). Given that agents targeting this pathway (e.g., rosiglitazone) are readily available for routine clinical use, further clinical studies using these agents to treat CD patients who are smokers with Paneth cell defects will validate the importance of this pathway in CD. Furthermore, metabolic dysregulation is tightly connected with the TNF-α–associated apoptosis pathway. Therefore, previously recognized important genetic factors, an environmental factor, and inflammatory pathways converged to affect Paneth cell health and clinical prognosis. Of note, our complementary approaches (crypt base LCM, PC-Cre⁺ mice) support the transcriptomics analysis from full-thickness ileum indicating that Paneth cells are the main target of the G+E effect.

Our data also suggest that smoking cessation may be beneficial for ATG16L1^T300A CD subjects with smoking-associated Paneth cell defects. Other potential intervention approaches include nicotine patch, PPARγ agonists, and anti–TNF-α. Anti–TNF-α is a major treatment modality for CD (1), and rosiglitazone has been shown to be efficacious in ulcerative colitis (another major form of IBD) (55). Our data indicate that the Paneth cell phenotype may be used to stratify CD patients who may benefit from these therapies. One limitation of our study is that, due to the physical restraint of the smoking chamber, the cigarette-smoking experiments could not exceed 6 weeks. Therefore, the effect of long-term cigarette smoking on Paneth cell defects, in particular the reversibility of the approaches described above, is unclear. In addition, while neither nicotine, lung pathology, nor systemic inflammatory signals were shown to affect Paneth cells in this model, it is possible that the changes in gut transcriptomics and subsequent Paneth cell defects are the results of processes initiated external to the diseased/target organ (gut), similar to what occurs in rheumatoid arthritis (56), in which cigarette smoking has been shown to alter transcriptomic changes of the joints (57). One such possibility could be the lung-gut axis, such that cigarette smoking affects the lung on a molecular level (potentially through lung microbiome and/or metabolites) (58, 59), which may in turn affect gut transcriptomics. In addition, the PPARγ pathway has best been studied in liver, skeletal muscle, and adipocytes in the context of metabolism (60). For example, based on the known PPARγ upstream regulatory mechanisms, we also speculate that the combination of G+E could affect either fatty acid–binding proteins or fatty acid transporters, which would potentially involve liver-gut crosstalk (61).

In summary, we show that genetic and environmental factors synthesize to trigger unique biologic processes, resulting in a clinically relevant phenotype. Our data also provide complementary mechanistic insights into the role of Paneth cells in mediating CD pathogenesis (16, 62) and identification of actionable therapeutic targets.

Study design. The overall objective of our study was to determine the associations between the ATG16L1^T300A genotype and exposure to cigarette smoking in triggering Paneth cell defects. For human subjects, based on the prevalence of adult CD subjects harboring type I Paneth cell phenotype (14, 15) and the natural history of CD after resection (1), 90 subjects were required to achieve a power of 80%. For in vivo experiments, we used a previously described mouse strain (Atg16l1^T300A) that is known to possess Paneth cell defects (28). Cigarette smoking was performed following a previously described protocol (63, 64), with the cigarette filters removed. Paneth cell analysis was performed using immunofluorescence (14, 15) on distal ileum. All the experiments were performed in several replicates over the course of 2 years. At least 3 to 6 biological replicates were used for each group/experiment. The mice were randomized, and the investigator performing the histologic analysis was blinded to the sample identity. The design for microbiome studies included proper littermate controls and cohousing (65), and microbiome composition was analyzed using 16S rRNA-seq. Transcriptomic analysis was performed using RNA-seq. All data were included (no outliers were excluded). Additional details, including the total numbers per study group, are included in the respective figure legends.

CD subjects. CD subjects who underwent ileocolectomy between 1999 and 2010 at Washington University or Cedars-Sinai Medical Center were previously described (14). A second CD cohort from Washington University composed of consecutive CD patients who underwent ileocolectomy between 2011 and 2013 were additionally included. Deidentified tissue samples from ileal resection margins that were free of acute inflammation were used for Paneth cell phenotype analysis.

The following information was retrieved from the medical record: sex, age at operation, smoking history (never smoker vs. active/exsmoker), medication history (including immunomodulators and biologics), and endoscopic findings at the first visit 6 to 12 months after surgery. Recurrence was defined by endoscopy (Rutgeerts score ≥ i₂). The genotypes of the patients were obtained using ImmunoChip (14) or through TaqMan genotyping assay (Thermo Fisher Scientific), with genomic DNA extracted from formalin-fixed, paraffin-embedded tissue based on the manufacturer’s instructions.

Mice. Atg16l1^T300A mice have been described before (28) and were a gift from Ramnik Xavier (Harvard Medical School, Boston, Massachusetts, USA). Heterozygotes were used to breed Atg16l1^T300A and littermate controls. Atg16l1^T300A mice were crossed with Tnfrsf1^atm1/lmxmice (Tnfr1 KO; The Jackson Laboratory, 3244) to generate Atg16l1^T300A Tnfr1^–/– mice. PC-Cre mice were generated by introducing Cre recombinase gene driven by the α-defensin-4 promoter in the embryonic stem (ES) cells (44). The PC-Cre mice were subsequently crossed with Atg16l1^fl/fl mice to generate PC-Cre⁺ mice. Mice with intestinal epithelium–specific knockout of Pparg were generated by crossing Pparg^fl/fl mice (The Jackson Laboratory, catalog 004584) with Villin-Cre mice (The Jackson Laboratory, catalog 4586). All mice were on a C57BL/6 genetic background. The distal ileum of the mice (distal, 3 cm) was used for analysis in this study.

Mouse treatments. Four- to six-week-old mice were exposed to cigarette smoking at 4 cigarettes per day for 5 days per week using Kentucky research cigarette 3R4F (with filters removed) (University of Kentucky, Lexington, Kentucky, USA) and a previously described protocol (64, 66). Mice were exposed to 2 weeks of smoking unless otherwise indicated and then sacrificed for tissue collection. For cohousing experiments, Atg16l1^T300A and WT littermates designated as microbial recipients were exposed to an antibiotic cocktail of vancomycin, neomycin, ampicillin, and metronidazole for 2 weeks (36), followed by cohousing with mice of the same genotype that were exposed to cigarette smoking (microbial donors). Cohousing lasted 4 weeks, and the microbial donors continued to be exposed to cigarette smoking during this period. For the nicotine experiment, nicotine was added into the drinking water at a concentration of 0.1 mg/ml for 4 weeks. In other experiments to study effects of pharmacological agents in preventing cigarette smoking–induced Paneth cell defect, mice with the Atg16l1^T300A genotype were exposed to smoking for 2 weeks. During this period, mice were administered either pan-caspase inhibitor Z-VAD-FMK (10 mg/kg/d) intraperitoneally (ApexBio) or Ultra-LEAF anti-mouse TNF-α antibody (0.5 mg/mouse/injection, 2 injections/wk) (BioLegend). For nec-1 inhibitor, mice received intraperitoneal administration of necrostatin (Sigma-Aldrich) at 4 mg/kg/d for 2 weeks. For rosiglitazone administration, mice received daily oral gavage with either PBS or rosiglitazone (Sigma-Aldrich) at a dose of 20 mg/kg/d for 2 weeks.

Paneth cell phenotype analysis. Lysozyme and HD5 immunofluorescence was interpreted (by T.-C. Liu), as described previously (11, 14, 19, 20, 27). For both human and mouse samples, each Paneth cell was classified into normal or 1 of the 5 abnormal categories, including the following: disordered (abnormal distribution and size of the granules), diminished (≤10 granules), diffuse (smear of lysozyme or defensin within the cytoplasm with no recognizable granules), excluded (majority of the granules do not contain stainable material), and enlarged (rare, megagranules) (14, 27). The last 2 categories were only observed in human samples. The Paneth cell phenotypes of CD subjects (used for outcome correlation) were then defined by the percentage of total abnormal Paneth cells in the sample. The type I Paneth cell phenotype was defined as 20% or more of total Paneth cells showing abnormal morphology patterns, whereas the type II Paneth cell phenotype was defined as less than 20% of total Paneth cells showing morphologic defects (14).

Statistics. Clinical outcome correlation was performed using log-rank test. For analysis between different genotype and smoking exposure combinations, Kruskal-Wallis tests followed by Dunn’s tests between groups were performed. For Paneth cell phenotype and various cellular readout comparisons in mouse experiments, 2-way ANOVA followed by Tukey’s multiple comparison adjustment was used. For microbiome studies, principle coordinate analysis was performed using analysis of similarities (ANOSIM) with 999 permutations. Relative operational taxonomic unit (OTU) abundance data were input into linear discriminant analysis effect size (LEfSe) to determine biomarkers with significant linear discriminant analysis effect size (67). The determination of sample size and data analysis for animal studies followed the general guideline of Festing and Altman (68). Based on the law of diminishing returns, Mead recommended that a degree of freedom (DF) of 10 to 20 associated with error term in an ANOVA would be adequate to estimate preliminary information (69). All tests were 2 tailed, and a P value of less than 0.05 was considered significant. Data were plotted and analyzed using GraphPad Prism (version 6.05) and SAS version 9.4 (SAS Institute). Data represent mean ± SEM.

Study approval. The study was approved by the institutional review boards of Washington University School of Medicine and Cedars-Sinai Medical Center. Subjects provided written, informed consent. The animal studies were approved by the ethical committee at Washington University School of Medicine.

Data and materials availability. All Atg16l1^T300A mouse study full-thickness ileal RNA-seq data were deposited in the EMBL-EBI’s ArrayExpress database (E-MTAB-5707). All Atg16l1^T300A mouse study LCM-procured Paneth cell microarray data were deposited in the ArrayExpress database (E-MTAB-6168). All 16s rRNA-seq data for Atg16l1^T300A mice were deposited in the ArrayExpress database (E-MTAB-5717). All 16s rRNA-seq data for the cohousing experiment were deposited in the ArrayExpress database (E-MTAB-5720).