Association between ATG16L1 gene polymorphism and the risk of Crohn's disease.

Objective To perform a meta-analysis to evaluate studies investigating the association between ATG16L1 gene polymorphism and Crohn's disease. Methods PubMed, Embase and Web of Science databases were searched for all studies focusing on the association of ATG16L1 and Crohn's disease. Combined odds ratios with 95% confidence intervals were calculated for four genetic models (allelic model: G allele versus A allele; additive model: GG versus AA; dominant model: GA + GG versus AA; recessive model: GG versus GA + AA) using either a random effects or fixed effects model. Results A total of 47 case-control studies involving 18 638 cases and 30 181 controls were included in the final meta-analysis. There was a significant association between ATG16L1 and Crohn's disease for all four genetic models. Significant associations were also shown in subgroup analyses when stratified by study design (population- or hospital-based). Conclusion In this meta-analysis, the ATG16L1 genotype was significantly associated with the risk of developing Crohn's disease.

Introduction

Crohn’s disease is a type of inflammatory bowel disease associated with chronic relapsing inflammation of the digestive tract anywhere from the mouth to the anus.[1] Although its aetiopathogenesis remains unclear, it is well established that Crohn’s disease is a complex disorder resulting from the interactions of genetic, environmental and microbial factors. Among these, genetic factors may be responsible for a major component of disease susceptibility.[2]

The role of autophagy processes in the development of inflammatory bowel disease is attracting increasing attention.[3] It is possible that genes involved in the autophagy pathway may contribute to the pathogenesis of Crohn’s disease. The autophagy-related 16-like 1 (ATG16L1) gene encodes an important protein involved in the formation of autophagosomes during autophagy.[4] Genome-wide association studies have shown an association between ATG16L1 polymorphism involving an amino acid change at position 300 and increased susceptibility to Crohn’s disease.[5,6] This substitution of threonine with alanine is the result of a single nucleotide polymorphism in which adenine (A) is replaced with guanine (G). This association has been examined in numerous studies, but the results have been inconsistent. The present meta-analysis was designed to evaluate the association between ATG16L1 and Crohn’s disease using the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) criteria.[7]

Materials and methods

Literature search

Two investigators (B.B.Z and B.Y.) systematically searched the databases PubMed (up to June 2016), Embase (1966 to June 2016) and Web of Science (2003 to June 2016), and also references from articles, reviews and abstracts presented at meetings of related scientific societies. The following search terms were used: (“ATG16L1”) AND (“Crohn’s disease” OR “inflammatory bowel diseases”) AND (“polymorphism” OR “mutation” OR “variant” OR “genotype”). Studies were limited to those published in English.

Inclusion criteria and quality assessment

The same two investigators independently screened each of the titles, abstracts and full texts to determine whether the studies met the following criteria: (i) evaluation of the association of Crohn’s disease and ATG16L1 polymorphism; (ii) case–control design; (iii) sufficient data for the estimation of odds ratios (ORs) and 95% confidence intervals (CIs). In addition, a quality assessment was performed on all included studies using the Newcastle–Ottawa Scale (NOS) as described elsewhere.[8]

Data extraction

The following data were collected from each study included in the meta-analysis: first author’s name, publication date, country, total numbers of cases and controls, and frequency of ATG16L1 genotypes in cases and controls.

Statistical analyses

Strength of agreement between the investigators regarding study selection was evaluated using the Kappa statistic. The combined ORs and 95% CIs were calculated for the allelic model (G allele versus A allele), the additive model (GG versus AA), the dominant model (GA + GG versus AA) and the recessive model (GG versus GA + AA) using either the random effects model[9] or the fixed effects model.[10] Galbraith plots were created to graphically assess the source of any heterogeneity. Publication bias was analyzed using Begg’s funnel plots and Egger’s test, with a P-value < 0.05 being considered representative of statistically significant publication bias.[11] Conformity with the Hardy–Weinberg equilibrium amongst the controls was determined using the χ2-square test and was considered to be in agreement when the P-value is ≥ 0.05. All statistical analyses were performed using Stata statistical software version 11.0 (StataCorp, College Station, TX, USA).

Results

Study characteristics

A total of 843 potentially relevant articles were initially identified. After exclusion of duplicate studies and application of the inclusion criteria, a total of 44 articles[12-55] were included in the qualitative synthesis (Figure 1). Büning et al.[13] contained three separate case–control studies and Fowler et al.[19] contained two separate case–control studies; therefore, a total of 47 case–control studies involving 18 638 cases and 30 181 controls were included in the final meta-analysis. The main characteristics of these studies are given in Table 1.

Figure 1.

Flow diagram of the study selection process. CD, Crohn’s disease.

Table 1.

Main characteristics of studies included in the meta-analysis.

Reference	Source of subjects	Genotype and allele distribution (case/control)				HWE		NOS score
		GG	GA	AA	G	Conforms	Statistical significance
Baldassano et al., 2007[12]	Population-based	58/78	65/136	19/67	181/292	Yes	NS	6
Büning et al., 2007:[13] study 1	Population-based	98/68	149/143	63/74	345/279	Yes	NS	6
Büning et al., 2007:[13] study 2	Population-based	38/49	86/109	23/49	162/207	Yes	NS	6
Büning et al., 2007:[13] study 3	Population-based	60/66	78/102	19/47	198/234	Yes	NS	6
Cummings et al., 2007[14]	Hospital-based	209/196	282/330	81/157	700/722	Yes	NS	6
Prescott et al., 2007[15]	Population-based	435/321	565/626	236/288	1435/1268	Yes	NS	6
Roberts et al., 2007[16]	Population-based	166/130	243/285	87/134	575/545	Yes	NS	7
Yamazaki et al., 2007[17]	Population-based	23/32	184/167	274/238	230/231	Yes	NS	6
Baptista et al., 2008[18]	Population-based	46/42	94/90	40/57	186/174	Yes	NS	8
Fowler et al., 2008:[19] study 1	Population-based	243/339	315/601	111/304	801/1279	Yes	NS	6
Fowler et al., 2008:[19] study 2	Population-based	59/110	73/189	22/121	191/409	No	P = 0.04	6
Gaj et al., 2008[20]	Population-based	24/32	25/70	11/37	73/134	Yes	NS	8
Glas et al., 2008[21]	Population-based	—	—	—	906/1673	N/A	N/A	8
Hancock et al., 2008[22]	Population-based	216/321	288/569	82/266	720/1211	Yes	NS	7
Lakatos et al., 2008[23]	Population-based	92/33	125/83	49/33	309/149	Yes	NS	7
Lappalainen et al., 2008[24]	Population-based	—	—	—	232/179	N/A	N/A	6
Latiano et al., 2008[25]	Population-based	227/214	335/376	105/159	789/804	Yes	NS	7
Okazaki et al., 2008[26]	Population-based	77/88	103/150	28/76	257/326	Yes	NS	8
Perricone et al. 2008[27]	Population-based	33/30	73/76	57/54	139/136	Yes	NS	7
Peterson et al., 2008[28]	Population-based	—	—	—	655/505	N/A	N/A	6
Van Limbergen et al., 2008[29]	Population-based	217/98	294/176	118/71	728/372	Yes	NS	6
Weersma et al., 2008[30]	Population-based	121/280	125/428	40/163	367/988	Yes	NS	7
Amre et al., 2009[31]	Population-based	102/64	137/135	47/91	341/263	Yes	NS	8
Dema et al., 2009[32]	Population-based	178/246	314/407	115/206	670/899	Yes	NS	7
Dusatkova et al., 2009[33]	Population-based	107/132	158/239	68/128	372/503	Yes	NS	7
Lacher et al., 2009[34]	Population-based	60/56	73/128	19/69	193/240	Yes	NS	7
Márquez et al., 2009[35]	Population-based	125/221	156/347	63/177	406/789	Yes	NS	7
Newman et al., 2009[36]	Population-based	159/253	204/415	72/227	522/921	No	P = 0.03	9
Palomino-Morales et al., 2009[37]	Hospital-based	216/183	253/316	75/167	685/682	Yes	NS	7
Cotterill et al., 2010[38]	Population-based	—	—	—	317/840	N/A	N/A	7
Csöngei et al., 2010[39]	Population-based	108/79	151/163	56/72	367/321	Yes	NS	7
Gazouli et al., 2010[40]	Population-based	189/161	222/274	63/104	600/596	Yes	NS	6
Sventoraityte et al., 2010[41]	Population-based	16/44	28/89	11/53	60/177	Yes	NS	8
Fabio et al., 2011[42]	Population-based	94/50	134/97	51/43	322/197	Yes	NS	6
Frank et al., 2011[43]	Hospital-based	25/17	22/19	14/23	72/53	No	P = 0.007	5
Lauriola et al., 2011[44]	Population-based	6/6	9/11	3/3	21/23	Yes	NS	6
Jung et al., 2012[45]	Population-based	—	—	—	638/864	N/A	N/A	6
Wang et al., 2012[46]	Population-based	44/33	164/140	141/179	252/206	Yes	NS	6
Hirano et al., 2013[47]	Population-based	—	—	—	1993/10141	N/A	N/A	6
Dalton et al., 2014[48]	Population-based	22/8	49/33	12/14	93/49	Yes	NS	6
Jakobsen et al., 2014[49]	Population-based	—	—	—	293/566	N/A	N/A	7
Scolaro et al., 2014[50]	Population-based	25/48	53/106	28/84	103/202	Yes	NS	8
Serbati et al., 2014[51]	Population-based	10/9	43/76	16/30	63/94	No	P < 0.001	6
Zhang et al., 2014[52]	Population-based	77/62	134/166	209/272	288/290	No	P < 0.001	7
Na et al., 2015[53]	Population-based	—	—	—	54/51	N/A	N/A	7
Salem et al., 2015[54]	Hospital-based	108/29	78/13	50/15	294/71	No	P < 0.001	6
Yang et al., 2015[55]	Population-based	226/211	838/1033	745/1192	1290/1455	Yes	NS	7

HWE, Hardy–Weinberg equilibrium; N/A, not available; NOS, Newcastle–Ottawa scale.

NS, not statistically significant (P ≥ 0.05).

Quantitative synthesis

When all the studies were pooled in the meta-analysis, a significant association was seen between ATG16L1 and Crohn’s disease in all four genetic models (allelic model: OR = 1.29, 95% CI = 1.22, 1.37, Figure 2; additive model: OR = 1.80, 95% CI = 1.68, 1.92, Figure 3; dominant model: OR = 1.47, 95% CI = 1.39, 1.55, Figure 4; recessive model: OR = 1.46, 95% CI = 1.39, 1.54, Figure 5). When stratified by study design (population- or hospital-based), a significant association between ATG16L1 and Crohn’s disease was still seen in all four genetic models (Table 2).

Figure 2.

Forest plot of the association between ATG16L1 and Crohn’s disease using the allelic model (G allele versus A allele). The pooled odds ratio (OR) and 95% confidence intervals (CI) are indicated by the diamond. Percentage weights were calculated using a random effects model.

Figure 3.

Forest plot of the association between ATG16L1 and Crohn’s disease using the additive model (GG versus AA). The pooled odds ratio (OR) and 95% confidence intervals (CI) are indicated by the diamond. Percentage weights were calculated using a fixed effects model.

Figure 4.

Forest plot of the association between ATG16L1 and Crohn’s disease using the dominant model (GG + GA versus AA). The pooled odds ratio (OR) and 95% confidence intervals (CI) are indicated by the diamond. Percentage weights were calculated using a fixed effects model.

Figure 5.

Forest plot of the association between ATG16L1 and Crohn’s disease using the recessive model (GG versus GA + AA). The pooled odds ratio (OR) and 95% confidence intervals (CI) are indicated by the diamond. Percentage weights were calculated using a fixed effects model.

Table 2.

Results of meta-analysis and subgroup analysis for the association between ATG16L1 and Crohn’s disease according to the allelic, additive, dominant and recessive models.

Group analysed	Allelic model (G vs A)		Additive model (GG vs AA)		Dominant model (GG + GA vs AA)		Recessive model (GG vs GA + AA)
	OR (95% CI)	Analysis model	OR (95% CI)	Analysis model	OR (95% CI)	Analysis model	OR (95% CI)	Analysis model
All	1.29 (1.22, 1.37)	Random effects	1.80 (1.68, 1.92)	Fixed effects	1.47 (1.39, 1.55)	Fixed effects	1.46 (1.39, 1.54)	Fixed effects
Population-based	1.28 (1.20, 1.37)	Random effects	1.76 (1.64, 1.89)	Fixed effects	1.44 (1.36, 1.52)	Fixed effects	1.46 (1.38, 1.54)	Fixed effects
Hospital-based	1.46 (1.31, 1.62)	Fixed effects	2.18 (1.76, 2.70)	Fixed effects	1.86 (1.54, 2.26)	Fixed effects	1.51 (1.29, 1.76)	Fixed effects
NOS score ≥ 7	1.33 (1.24, 1.43)	Random effects	1.83 (1.68, 1.99)	Fixed effects	1.49 (1.39, 1.59)	Fixed effects	1.47 (1.37, 1.57)	Fixed effects
Conform to HWE	1.32 (1.24, 1.40)	Random effects	1.79 (1.67, 1.92)	Fixed effects	1.46 (1.38, 1.54)	Fixed effects	1.46 (1.38, 1.54)	Fixed effects

OR, odds ratio; CI, confidence intervals; NOS, Newcastle–Ottawa Scale; HWE, Hardy–Weinberg equilibrium.

Sensitivity analyses

Sensitivity analyses were conducted to determine whether modification of the inclusion criteria of the meta-analysis affected the final results. When the included studies were limited to those conforming to the Hardy–Weinberg equilibrium (P ≥ 0.05), the pooled ORs of these 33 studies were not materially different from those of the full meta-analysis (Table 2). Likewise, when the included studies were limited to those with a high NOS score (≥7), the pooled ORs of these 22 studies were not materially different from those of the full meta-analysis (Table 2).

Analysis of heterogeneity

Significant heterogeneity existed in the allelic model (I2 = 75.4%). A Galbraith plot was created to graphically assess the source of heterogeneity (Figure 6). The studies by Yamazaki et al.,[17] Fowler et al.[19] (study 1), Latiano et al.,[25] Amre et al.,[31] Lacher et al.,[34] Palomino-Morales et al.,[37] Jung et al.[45] and Hirano et al.[47] were identified as contributors to the heterogeneity. When these eight studies were excluded, the I2 was 0.0% and the OR (95% CI) was 1.33 (1.28, 1.37).

Figure 6.

Galbraith plot of the allelic model. The outliers were the studies by Yamazaki et al.,[17] Fowler et al.[19] (study 1), Latiano et al.,[25] Amre et al.,[31] Lacher et al.,[34] Palomino-Morales et al.,[37] Jung et al.[45] and Hirano et al.[47] b, effect estimate; se, standard error.

Publication bias

The shapes of the Begg’s funnel plots did not reveal any evidence of obvious asymmetry (Figure 7). No statistical evidence of publication bias was found using Egger’s regression test (P = 0.09 for the allelic model; P = 0.62 for the additive model; P = 0.08 for the dominant model; and P = 0.83 for the recessive model).

Figure 7.

Begg’s funnel plots with pseudo 95% confidence limits of all studies in the meta-analysis using the four model types: (a) allelic model (G allele versus A allele); (b) additive model (GG versus AA); (c) dominant model (GG + GA versus AA); (d) recessive model (GG versus GA + AA). SE, standard error; OR, odds ratio.

Discussion

Since Hampe et al.[5] reported in 2007 that ATG16L1 gene polymorphism was associated with Crohn’s disease, many studies have evaluated the relationship between ATG16L1 and the risk of Crohn’s disease.[56] However, the results are inconsistent. As the strength of results from a single case–control study is weak due to small sample sizes, the combination of many studies in a meta-analysis has the benefit of overcoming this limitation by increasing the sample size and generating more robust results. Meta-analysis has been widely used in genetic association studies.[57,58] The present meta-analysis was performed to assess whether the combined evidence supports an association between ATG16L1 and Crohn’s disease.

The present meta-analysis examined ATG16L1 gene polymorphism and its relationship with the risk of Crohn’s disease based on data from 47 case–control studies involving 18 638 cases and 30 181 controls. Most of these studies reported that ATG16L1 was associated with the risk of Crohn’s disease, but not all. The results of the meta-analyses demonstrated that overall there was evidence of a significant association between ATG16L1 gene polymorphism and Crohn’s disease. This significant association remained in all four genetic models when subgroup analyses were performed based on study design (population-based or hospital-based).

When considering the potential mechanisms linking ATG16L1 polymorphism with an increased risk of Crohn’s disease, it has been shown that ATG16L1 polymorphism impairs the autophagy processing of pathogenic bacteria and the function of intestinal Paneth cells.[59,60] In addition, it has been shown that ATG16L1 polymorphism is associated with increased susceptibility to Helicobacter pylori infection.[61] In patients with Crohn’s disease, it has been reported that homozygosity of the ATG16L1 risk allele (GG) was associated with a reduced ability to clear pathosymbionts.[62] Paneth cells in ATG16L1-deficient mice have been shown to be dysfunctional and to demonstrate increased expression of pro-inflammatory cytokines.[63,64]

When interpreting the results of this meta-analysis, a number of limitations should be acknowledged. First, it is well known that both environmental factors and individual genetic predisposition contribute to the development of Crohn’s disease. Due to the lack of original data, however, potential interactions between these two types of influence has not been evaluated. Secondly, ATG16L1 seems to exert a close functional correlation with other genes in regulating autophagy. For example, the interaction of ATG16L1 and NOD2 has been implicated in the pathogenesis of Crohn’s disease.[63] Potential gene–gene interactions require further evaluation. Thirdly, the ATG16L1 genotype has been reported to be associated with disease phenotype,[65] which has clinical significance. Further combined analyses are needed to clarify the association between the ATG16L1 genotype and Crohn’s disease phenotype.

In conclusion, the present meta-analysis of robust data and unbiased results demonstrated an association between ATG16L1 genotype and the development of Crohn’s disease. These findings will be helpful in understanding the aetiology of Crohn’s disease and indicate that the ATG16L1 gene might have potential as a therapeutic or diagnostic target.