A genome-wide association meta-analysis of diarrhoeal disease in young children identifies FUT2 locus and provides plausible biological pathways

Abstract

More than a million childhood diarrhoeal episodes occur worldwide each year, and in developed countries a considerable part of them are caused by viral infections. In this study, we aimed to search for genetic variants associated with diarrhoeal disease in young children by meta-analyzing genome-wide association studies, and to elucidate plausible biological mechanisms. The study was conducted in the context of the Early Genetics and Lifecourse Epidemiology (EAGLE) consortium. Data about diarrhoeal disease in two time windows (around 1 year of age and around 2 years of age) was obtained via parental questionnaires, doctor interviews or medical records. Standard quality control and statistical tests were applied to the 1000 Genomes imputed genotypic data. The meta-analysis (N = 5758) followed by replication (N = 3784) identified a genome-wide significant association between rs8111874 and diarrhoea at age 1 year. Conditional analysis suggested that the causal variant could be rs601338 (W154X) in the FUT2 gene. Children with the A allele, which results in a truncated FUT2 protein, had lower risk of diarrhoea. FUT2 participates in the production of histo-blood group antigens and has previously been implicated in the susceptibility to infections, including Rotavirus and Norovirus. Gene-set enrichment analysis suggested pathways related to the histo-blood group antigen production, and the regulation of ion transport and blood pressure. Among others, the gastrointestinal tract, and the immune and neuro-secretory systems were detected as relevant organs. In summary, this genome-wide association meta-analysis suggests the implication of the FUT2 gene in diarrhoeal disease in young children from the general population.

Introduction

Diarrhoea, defined as three or more loose stools within the previous 24 h, is probably one of the most common symptoms in children, with an estimated 1370 million annual episodes in children younger than 5 years in 2010 (1). Two percent of these episodes progress to severe disease, and 700 000 episodes lead to death, mainly in low-income countries (1). In developed countries, diarrhoeal disease is a common reason for attendance at a general practitioner, especially in children under 5 years of age (2).

Several pathogens can account for infections associated with diarrhoeal disease, including viruses, bacteria and parasites. The GEMS study (Global Enteric Multicenter Study), conducted as a case–control study in seven African and Asian sites, identified Rotavirus, Cryptosporidium, enterotoxigenic Escherichia coli, and Shigella as most responsible attributable pathogens for cases of moderate-to-severe diarrhoea (3). In developed countries, improvements in public health infrastructure (water and sewage management), has caused a shift in the main causes of acute paediatric diarrhoea, moving from bacterial and parasite aetiologies to viruses. Rotavirus is the primary cause of diarrhoeal disease globally, and is responsible for almost half of the gastroenteritis cases requiring hospitalization in Western countries, followed by Norovirus, Adenovirus and Salmonella (4). The peak age for infection with Rotavirus is between 3 months and 2 years, coinciding with reduced protection by transplacental transfer of maternal antibodies (5) and the end of the lactation period (6). At the age of 5, almost all children have been infected with Rotavirus, and progressively develop natural immunity against this virus (7). Enteric pathogens damage small bowel enterocytes and cause impaired intestinal absorption, low grade fever and watery diarrhoea as a result of the deregulation of ion transport and stimulation of the enteric nervous system (8,9). In children, factors other than enteric pathogens can cause diarrhoea, including among others extra-intestinal infections, intolerances or food allergies (10), nutrient deficiencies, antimicrobials, or hereditary diseases, such as cystic fibrosis, but these represent a small proportion of all cases of diarrhoea.

The exposure to infectious agents plays a major role in the acquisition of the pathogen and development of diarrhoeal disease. However, not all individuals are equally susceptible to infection, and if infected, they may differ in their immunological response. Host genetic factors can explain part of the differences in susceptibility and response to infection. In a study on families with adopted children, it was shown that premature death due to infection of the biological parent, but not of the adopting parent, increased the relative risk of death due to infection in the adopted child by 5.8-fold, a higher risk than observed for vascular disease or cancer (11). Twin studies on infectious diseases have shown higher correlation between monozygotic than dizygotic twins (12–14). Heritability for early childhood diarrhoea was estimated to be 54% in a pedigree-based design in Brazil (15).

Candidate gene studies for infectious gastroenteritis have identified genetic variants in genes involved in the innate and acquired immune responses and in genes that participate in the production of histo-blood group antigens (HBGAs), which serve as receptors for numerous pathogens. In particular, the non-secretor phenotype, associated with null or inactivating mutations in FUT2 gene resulting in a lack of certain antigens in secretions and epithelial mucosa, confers strain-specific protection against Norovirus (16–18) and Rotavirus (19). A review of the associations between enteric pathogens and genetic variants can be found elsewhere (20,21).

The aims of the present study were: (1) to identify genetic variants that confer susceptibility to diarrhoeal disease in young children from the general population of developed countries through a genome-wide association meta-analysis, and (2) to elucidate potential biological mechanisms involved in diarrhoeal disease using pathway analysis approaches.

Results

Sample

Four different traits, any diarrhoea and doctor’s confirmed diagnosis of diarrhoea around 1 year of age (D1Y and DD1Y) and around 2 years of age (D2Y and DD2Y), were explored. Samples analyzed for each outcome in the discovery and in the replication phase are shown in Table 1. Information on diarrhoea was collected through questionnaires, doctor interviews or medical records (see Supplementary Material, Annex A).

Table 1.

Samples included in the study at age 1 year and at age 2 years by diarrhoeal disease definition and study phase

	Age 1 year
	Diarrhoea (D1Y)						Doctor diagnosis of diarrhoea (DD1Y)
	Cohort	Assessment (period in months)a	N total	N cases	% cases		Cohort	Assessment (period in months)a	N total	N cases	% cases
Discovery	ALSPAC (Disc)d	Questionnaire (6–18)	3363	2001	59.5		ALSPAC (Disc)d	Questionnaire (6–18)	3363	893	26.6
	Generation R	Questionnaire (6–12)c	2033	469	23.1		Generation R	Questionnaire (6–12)c	2033	179	8.8
	INMA_SAB	Questionnaire (6–14)	362	223	61.6		COPSAC2000	Doctor interview (6–12)	345	81	23.5
							LISAplus	Questionnaire (6–12)	662	186	28.1
	TOTAL		5758	2693	46.8		TOTAL		6403	1339	21.8
Replication	ALSPAC (Repl)	Questionnaire (6–18)	3361	2047	60.9		ALSPAC (Repl)	Questionnaire (6–18)	3361	871	25.9
	MoBa	Questionnaire (6–18)c	407	255	62.6		COPSAC2010	Doctor interview (6–12)	547	244	44.6
							INMA_VAL	Questionnaire (0–12)	334	149	44.6
							CHOPb	Medical records (6–18)	3223	147	4.6
	TOTAL		3768	2302	61.1		TOTAL		7465	1411	18.9

	Age 2 years
	Diarrhoea (D2Y)						Doctor diagnosis of diarrhoea (DD2Y)
	Cohort	Assessment (period in months)a	N total	N cases	% cases		Cohort	Assessment (period in months)a	N total	N cases	% cases
Discovery	ALSPAC (Disc)d	Questionnaire (18–30)	3189	1746	54.8		ALSPAC (Disc)	Questionnaire (18–30)	3189	514	16.1
	Generation R	Questionnaire (18–24)c	2058	943	45.8		Generation R	Questionnaire (18–24)c	2058	190	9.2
	INMA_SAB	Questionnaire (12–24)	361	166	46		COPSAC2000	Doctor interview (18–24)	319	86	27
							LISAplus	Questionnaire (18–24)	667	200	30
	TOTAL		5608	2855	50.9		TOTAL		6233	990	20.6
Replication	ALSPAC (Repl)d	Questionnaire (18–30)	3187	1759	55.2		ALSPAC (Repl)	Questionnaire (18–30)	3187	485	15.2
	INMA_VAL	Questionnaire (12–14)	329	213	64.7		COPSAC2010	Doctor interview (18–24)	518	271	52.3
							GINIplusd	Questionnaire (12–24)	794	328	41.3
							CHOPb	Medical records (18–30)	3223	190	5.9
	Total		3516	1972	56.1		Total		7722	1274	16.5

^aPeriod referred in the questionnaire, medical record or doctor interview.

^bOnly children not vaccinated against Rotavirus were included.

^cIt refers specifically to gastric flu or gastroenteritis.

^dStatistically significant differences in the proportions of diarrhoeal disease among males and females (P-value < 0.05) [ALSPAC-D1Y-Discovery: 61.2% males versus 57.7% females; ALSPAC–DD1Y-Discovery: 29.2% versus 23.8%; ALSPAC-D2Y-Discovery: 56.6% versus 52.8%; ASLPAC-D2Y-Replication: 58.5% versus 51.7%; and GINIplus-DD2Y-Replication: 46.7% versus 36.0%].

In the discovery phase, 46.8% of the children had had at least one diarrhoeal episode around the age of 1 year (D1Y), while this proportion decreased to 21.8% for the cases diagnosed by a doctor (DD1Y). The proportions for children around the age of two years were 50.9% for diarrhoeal disease (D2Y) and 20.6% for doctor’s diagnosis (DD2Y). Similar frequencies of diarrhoea were observed in the replication samples. The number of diarrhoea cases in Generation R and CHOP was lower than in other cohorts. For some of the diarrhoeal definitions, ALSPAC and GINIplus showed higher proportion of cases among males than among females.

Discovery phase

The Q–Q plots and the Manhattan plots for each outcome are shown in Figure 1 (D1Y) and in the Supplementary Material, Figures S1 (DD1Y), S2 (D2Y) and S3 (DD2Y). Genomic inflation factor, lambda (λ), ranged from 0.9952 to 1.0031.

Figure 1.

(A) Quantile-quantile (Q-Q) plots showing the probability values from GWAS meta-analysis of diarrhoeal disease at age 1 year (D1Y). The line indicates the distribution under the null hypothesis. Lambda value is shown. (B) Manhattan plot of the GWAS meta-analysis of diarrhoeal disease at age 1 year (D1Y). The x-axis represents the autosomal chromosomes and the y-axis represents –log10(P). The dotted line indicates genome-wide significance (P = 5.00E−08), and the dashed line indicates suggestive genome-wide significance (P = 1.00E−05).

For D1Y, the meta-analysis of 5,758 samples revealed a genome-wide significant variant, rs8111874, at 19q13.33 [odds ratio (OR) (95% confidence interval [CI]) = 1.32 (1.21–1.44); P-value = 1.06E−09] (Table 2 and Fig. 2)]. This lead SNP was also nominally associated with DD1Y [OR (95%CI) = 1.26 (1.14–1.39); P-value = 1.05E−05] and with DD2Y [OR (95%CI) = 1.32 (1.17–1.48); P-value = 4.6E−07] and with D2Y [OR (95%CI) = 1.08 (1.00–1.18); P-value = 6.56E−02]. In D1Y, the G allele increased the risk of diarrhoeal disease in all the studies (P-value for heterogeneity = 2.99E−02, Table 2).

Table 2.

Results from the fixed effect meta-analysis for diarrhoeal disease at age 1 year (D1Y) by discovery and replication phase.

				Discoverya												Replicationb					Combined
Marker	rs_number	Chr	Pos	EA	NEA	EAF	N	OR	lCI	uCI	P-value	P-value het	Effects	N variants/ locus	Gene	N	OR	lCI	UCI	p-value	N	OR	lCI	UCI	p-value	p-value het
19:49168942:SNP	rs8111874	19	49168942	G	A	0.57	5758	1.32	1.21	1.44	1.06E−09	2.99 E − 02	+++	71	NTN5 and SEC1P (intronic)	3768	1.25	1.13	1.39	1.69 E − 05	9526	1.29	1.21	1.38	8.05 E − 14	9.01 E − 02
15:50562847:SNP	rs62020330	15	50562847	A	G	0.97	5396	2.16	1.61	2.88	2.02 E − 07	5.05 E − 02	++?	2	4.7kb 5' of HDC	3195	0.99	0.69	1.43	9.72 E − 01	8591	1.60	1.27	2.00	4.97 E − 05	1.35 E − 03
7:2930941:SNP	rs1713926	7	2930941	C	T	0.69	5758	1.37	1.21	1.55	8.06 E − 07	5.68 E − 01	+++	1	15kb 3' of CARD11	1956	1.07	0.92	1.25	3.58 E − 01	7714	1.24	1.13	1.37	1.04 E − 05	1.23 E − 01
7:63621349:SNP	rs139755348	7	63621349	C	T	0.73	5758	0.79	0.71	0.87	1.14 E − 06	9.29 E − 01	—	89	46kb 5' of ZNF735	3768	1.06	0.95	1.19	2.96 E − 01	9526	0.89	0.83	0.96	2.59 E − 03	2.60 E − 03
7:1599067:SNP	rs112411182	7	1599067	T	C	0.95	5396	1.7	1.36	2.13	4.10 E − 06	8.75 E − 01	++?	1	3kb 5' of TMEM184A	3768	0.78	0.60	1.02	7.26 E − 02	9164	1.23	1.04	1.46	1.76 E − 02	1.38 E − 04
12:130325960:SNP	rs34180477	12	130325960	G	A	0.93	5396	0.58	0.46	0.73	4.30 E − 06	4.93 E − 01	–?	1	TMEM132D (intronic)	3009	1.21	0.96	1.53	1.04 E − 01	8405	0.84	0.71	0.98	3.24 E − 02	1.55 E − 05
14:38210286:SNP	rs74731421	14	38210286	G	A	0.97	5396	1.81	1.4	2.33	4.87 E − 06	8.07 E − 01	++?	1	146kb 5' of FOXA1	3768	0.91	0.69	1.22	5.42 E − 01	9164	1.34	1.11	1.63	2.32 E − 03	6.59 E − 03
12:106438211:INDELc		12	106438211	R	D	0.52	5758	1.26	1.14	1.39	4.95 E − 06	8.07 E − 01	+++	1	19kb of 3' NUAK1	407	1.20	0.85	1.68	3.01 E − 01	6165	1.26	1.14	1.38	2.73 E − 06	9.15 E − 01
7:96366342:SNP	rs12704876	7	96366342	T	C	0.51	5758	0.83	0.77	0.9	5.04 E − 06	8.45 E − 01	—	1	27kb 5' of SHFM1	3768	1.10	1.00	1.20	4.94 E − 02	9526	0.93	0.88	0.99	2.74 E − 02	2.32 E − 04
5:1695532:SNP	rs79411306	5	1695532	C	T	0.95	5396	0.55	0.42	0.71	6.17 E − 06	4.86 E − 01	–?	1	13kb 3' of MIR4277	3136	0.89	0.71	1.13	3.55 E − 01	8532	0.72	0.60	0.85	1.79 E − 04	3.61 E − 02
5:40474267:SNP	rs116560909	5	40474267	C	T	0.97	5396	1.9	1.44	2.52	6.90 E − 06	5.52 E − 01	++?	2	206kb 5' of PTGER4	3768	1.08	0.78	1.49	6.63 E − 01	9164	1.50	1.21	1.85	2.03 E − 04	1.74 E − 02
17:52598239:INDELc		17	52598239	R	D	0.94	5396	1.57	1.29	1.92	6.90 E − 06	6.76 E − 01	++?	1	380kb 5' of TOM1L1	407	1.29	0.59	2.80	5.20 E − 01	5803	1.56	1.28	1.88	6.00 E − 06	8.13 E − 01
2:116776614:SNP	rs12615869	2	116776614	A	G	0.94	5396	1.54	1.28	1.87	7.02 E − 06	5.62 E − 01	++?	2	174kb 3' of DPP10	3768	1.11	0.90	1.37	3.38 E − 01	9164	1.33	1.16	1.53	6.57 E − 05	1.25 E − 01
4:184501675:SNP	rs7662749	4	184501675	C	T	0.72	5758	0.81	0.74	0.89	7.24 E − 06	8.53 E − 01	—	1	59kb 3' of RWDD4	3768	1.02	0.92	1.14	6.55 E − 01	9526	0.90	0.84	0.96	1.69 E − 03	2.69 E − 02
22:49732481:SNP	rs5770255	22	49732481	G	A	0.76	5758	1.23	1.12	1.35	8.77 E − 06	7.32 E − 01	+++	2	281kb 3' of C22orf34	3768	0.94	0.84	1.05	2.83 E − 01	9526	1.10	1.03	1.19	5.95 E − 03	4.69 E − 03

Variants with a p-value < 1. E − 05 in the discovery phase are shown.

Cohort designs: ALSPAC, COPSAC2010, Generation R, INMA_VAL, INMA_GIP, LISAplus, GINIplus and MOBA are unselected population-based birth cohorts. COPSAC2000 is a prospective clinical study of a birth cohort of infants born to mothers with a history of asthma. CHOP is a random collection of paediatric patients from a hospital centre.

Only the most significant variant per locus is shown. Variants in the same locus are defined in a 1 Mb window. N variants/locus indicates the number of SNPs in the locus with a P-value < 1E−05.

Only genetic variants with a minimal of 5000 samples are shown.

EA: effect allele; NEA: non effect allele; EAF: effect allele frequency; N: sample size; OR: odds ratio; lCI: 95% lower confidence interval; uCI: 95% upper confidence interval; P-value het: P-value for the heterogeneity test.

^aCohorts in the discovery phase are included in alphabetical order: ALSPAC (Disc), Generation R and INMA_SAB.

^bCohorts included in the replication phase are: ALSPAC (Repl) and MoBa.

^cALSPAC (Repl) has no data for these markers.

Bonferroni correction for 15 variants considered in the replication phase: P-value = 3.33E − 03.

Figure 2.

Forest plots for rs8111874 at 19q13.33 for the four diarrhoeal outcomes: (A) Diarrhoea at age 1 year (D1Y); (B) Doctor diagnosis of diarrhoea at age 1 year (DD1Y); (C) Diarrhoea at age 2 years (D2Y); (D) Doctor diagnosis of diarrhoea at age 2 years (DD2Y). In the vertical panel, the studies participating in the discovery or replication phase are presented. In the horizontal lines, the sizes of the boxes represent precision and the lines the confidence intervals. The diamond shapes represent the pooled effect estimates, for both the fixed- and random-effect models. The horizontal axis shows the scale of the effect estimates. The effect allele is G, and the other allele is A.

Another locus at 4q21.23, was associated with DD1Y at genome-wide significance [N = 6403, OR (95%CI) = 1.31 (1.19–1.44); P-value = 2.92E−08] (Supplementary Material, Table S1). The lead variant, rs1481779, was located in an intron of ARHGAP24 gene. This variant was also nominally associated with D1Y [OR (95%CI) = 1.09 (1.01–1.19); P-value = 3.7E−02], but not with D2Y or DD2Y.

Variants with a P-value < 1E−05 are shown in Table 2 (D1Y) and in Supplementary Material, Tables S1 (DD1Y), S2 (D2Y) and S3 (DD2Y). With the exception of the signal at 19q13.33, no major overlap was observed among the suggestive variants for each outcome (data not shown). A description of the potential function of the genetic variants and genes in each loci detected at a P-value < 1E−05 can be found in Supplementary Material, Tables S4–S7.

Replication phase

Seventy-two loci (P-value < 1E-05 in any of the four diarrhoea definitions) were followed for replication in an independent dataset (Table 1). After multiple-testing adjustment, the SNP rs8111874 at chromosome 19q13.33 was associated with D1Y [replication: OR (95%CI) = 1.25 (1.13–1.39); P-value = 1.69E−05] (Table 2), and it was nominally associated for the other diarrhoea outcomes (Fig. 2). Other variants at the same locus were also associated with DD1Y and DD2Y (Supplementary Material, Tables S1 and S3). None of the variants in other loci replicated, and neither of them reached genome-wide significance in the combined analyses. Results of the replication phase can be found in Table 2 (D1Y) and in Supplementary Material, Tables S1 (DD1Y), S2 (D2Y) and S3 (DD2Y).

Chromosome 19 locus: FUT2 gene

The regional association plot of rs8111874 at 19q13.33 shows a linkage disequilibrium block that overlaps several genes, including FUT2 (Fig. 3A). FUT2 participates in the production of histo-blood group antigens (HBGAs) and contains a stop mutation (rs601338, W154X) known to confer protection against certain infections. We conditioned the top SNP at 19q13.33 (rs8111874) on the stop mutation (rs601338) and vice versa: the odds ratios from these analyses were attenuated (Table 3). After conditioning to rs601338, no secondary signals were observed in the region (window size 1 Mb) (Fig. 3B). The forest plots for rs601338 (W154X) are shown in Supplementary Material, Figure S4.

Figure 3.

(A) Regional association of 19q13.33 locus (top SNP: rs8111874, chr19:49168942) with diarrhoeal disease at age 1 year (D1Y) (N = 5758). The top variant is indicated with a diamond in purple and the flanking variants in circles, coloured according to their linkage disequilibrium (LD). Variant rs601338 is shown in a black circle. The plot was constructed using the 1000 Genomes CEU population (Northern and Western European ancestry). (B) Regional association of 19q13.33 locus (top SNP: rs8111874, chr19:49168942) with diarrhoeal disease at age 1 year (D1Y) conditioned to rs601338 (N = 5758). The associations of the genetic variants in the region were attenuated.

Table 3.

Conditional analysis at 19q13.33 locus (rs8111874 and rs601338)

							Crude results for rs8111874				Results for rs8111874 conditioned to rs601338
Marker	rs_number	EA	NEA	EAF	IMP	Outcome	OR	lCI	uCI	P-value	N	OR	lCI	uCI	P-value
19:49168942:SNP	rs8111874	G	A	0.57	0.82-0.93	D1Y	1.32	1.21	1.44	1.06 E − 09	5214.7	1.08	1.01	1.14	1.40 E − 02
						DD1Y	1.26	1.13	1.39	1.17 E − 05	5576.7	1.05	0.98	1.12	2.00 E − 01
						D2Y	1.08	0.99	1.18	6.56 E − 02	5385.9	1.00	0.95	1.06	9.12 E − 01
						DD2Y	1.32	1.17	1.48	2.74 E − 06	5672.7	1.08	1.00	1.17	4.77 E − 02
							Crude results for rs601338				Results for rs601338 conditioned to rs8111874
Marker	rs_number	EA	NEA	EAF	IMP	Outcome	OR	lCI	uCI	P-value		OR	lCI	uCI	P-value
19:49206674:SNP	rs601338	G (W)	A (X)	0.52	1	D1Y	1.28	1.18	1.39	2.74 E − 09	6224.8	1.06	1.00	1.12	3.93 E − 02
						DD1Y	1.25	1.14	1.37	1.99 E − 06	6872.8	1.07	1.01	1.14	2.48 E − 02
						D2Y	1.10	1.02	1.18	1.84 E − 02	6471.1	1.04	0.99	1.09	1.40 E − 01
						DD2Y	1.27	1.14	1.4	5.90 E − 06	6976.7	1.06	0.99	1.13	1.16 E − 01

EA: effect allele; NEA: non effect allele; EAF: effect allele frequency; IMP: imputation quality (from–to); D1Y: any diarrhoea at age one year; DD1Y: doctor’s confirmed diagnosis of diarrhoea at age 1 year; D2Y any diarrhoea at age two years; DD2y doctor’s confirmed diagnosis of diarrhoea at age 2 years. Distance between SNPs is 37.7 kb.

Enrichment analysis

Two prediction programs, the Meta-Analysis Gene-set Enrichment of variaNT Associations (MAGENTA) and the Data-driven Expression Prioritized Integration for Complex Traits (DEPICT) were used to investigate gene-sets enriched among the variants with the lowest P-values. Using MAGENTA, the ‘KEGG_GLYCOSPHINGOLIPID_BIOSYNTHESIS_GLOBO_SERIES’ gene-set was identified in DD2Y at a 5% False Discovery Rate (FDR) (Supplementary Material, Table S8). Six out of 14 genes of this pathway had SNPs in the 95th percentile of lowest P-values. This gene-set was also weakly associated with diarrhoea at age 1 year (D1Y). Other gene-sets with nominal evidence are listed in Supplementary Material, Table S8, including the ’KEGG_GLYCOSPHINGOLIPID_BIOSYNTHESIS_GANGLIO_SERIES’ and the blood pressure regulatory gene-set ‘KEGG_RENIN_ANGIOTENSIN_SYSTEM’.

The top 10 gene-sets detected with DEPICT for each outcome are shown in Supplementary Material, Table S9. The gene-sets ‘ENSG00000147955—SIGMAR1 PPI subnetwork’ (involved in ion channels regulation and modulation of neurotransmitter release), ‘MP:0001675—abnormal ectoderm development’, ‘GO:0007492—endoderm development’ and ‘ENSG00000140612—SEC11A PPI subnetwork’ (component of the microsomal signal peptidase complex) remained after multiple-testing for DD1Y. Genetic variants showing suggestive P-values were linked to genes with enriched expression in the gastrointestinal tract (D1Y), in the immune system (DD1Y) and in the neuro-secretory system (D2Y), among others (Supplementary Material, Table S10). Exclusion of genetic variants in 1 Mb around the FUT2 locus gave similar gene-sets and tissue/organs, but P-values were attenuated (data not shown).

Overlap with known variants and genes for related diseases

We compared the results from this study with variants reported in the literature as associated with inflammatory bowel disease (IBD) risk (22) and with viral infection and response to vaccination (23–36). Nine out of the 162 variants identified in IBD were nominally associated with D1Y (Supplementary Material, Table S11). Two of these variants passed the multiple-testing correction: one of them located in the FUT2 gene (P-value = 4.67E−09, opposite effect direction), and the other one in CARD11 locus (P-value = 1.52E−05, same effect direction). In addition, one out of 139 variants associated with infections was also associated with D2Y after correction for multiple-testing (Supplementary Material, Table S12). Specifically, the A allele of rs17793829, located in TTC7B gene, was associated with higher anti Cytomegalovirus IgG titer (31) and higher risk for D2Y (P-value = 2.12E-04). Finally, we evaluated 86 genes retrieved from OMIM (Online Mendelian Inheritance in Man) with the entry ‘diarrhoea’. None of them was associated with childhood diarrhoeal disease after multiple testing correction (Supplementary Material, Table S13).

Discussions

This study suggested the implication of the FUT2 locus in the diarrhoeal risk and provided evidences supporting the role of the histo-blood group antigen (HBGA) production and the regulation of ion transport pathways. The gastrointestinal tract and the immune and neuro-secretory systems were detected as relevant organs.

The genome-wide association meta-analysis followed by replication identified an association between rs8111874 mapping to the 19q13.33 locus and diarrhoea around 1 year of age (D1Y). Although with different strength, the association was also observed in all the different diarrhoeal outcomes investigated. The rs8111874 variant is located in an intronic region of the NTN5 (Netrin 5) and SEC1P (Secretory Blood Group 1, Pseudogene) genes and close to FUT2, previously associated with susceptibility to infection. FUT2 encodes the Fucosyltransferase 2 enzyme that participates in the production of histo-blood group antigens (HBGAs) by catalyzing the addition of a fucose residue in α1,2 linkage to the galactose of O- or N-glycoproteins and globo-, ganglio- or lacto-series of glycolipids (37). The FUT2 enzymatic activity is polymorphic, exhibiting the non-secretor phenotype (lack of certain antigens in the gut and epithelial mucosa) when inactivating mutations are present in the FUT2 gene. The most common FUT2 inactivating variant in Caucasians (Europeans and Iranians) and in Africans, is the stop mutation W154X (rs601338), while in Asians it is the missense variant A385T (rs1047781) (38). In order to investigate whether the signal observed at 19q13.33 locus could be caused by the stop mutation W154X (rs601338), the lead SNP in the region (rs8111874) was conditioned to the inactivating mutation and vice versa. The magnitudes of the effects of the conditional analyses were substantially attenuated, suggesting the presence of one single association signal. Children with the A allele at rs601338 (W154X), which results in a truncated FUT2 protein and the non-secretor phenotype, had lower risk of diarrhoeal disease during the first years of life. In agreement with our results from a population based design, the non-secretor phenotype has been associated with protection against Rotavirus (19), Norovirus (16–18) and Helicobacter pylori (39,40) in small settings of very well characterized hospitalized subjects. It is known that Norovirus (41) and Rotavirus (42–44) bind to the antigen associated with the FUT2 secretor phenotype to enter the cells. In contrast, the FUT2 non-secretor phenotype has been associated with higher risk of urinary tract infections (45), acute pyelonephritis (46), oral or vaginal Candida infections (12,47,48), Haemophilus influenza (49), Neisseria meningitides and Streptococcus pneumoniae infections (50). In addition, non-secretor individuals are at a higher risk of developing autoimmune diseases such as inflammatory bowel disease (22,51), psoriasis (52,53), and Behcet’s disease (54) and they also have higher vitamin B12 plasma levels (55,56). Recently, the non-secretor phenotype has been associated with gut microbiota at both the compositional and functional level (57). Non-secretors have lower species richness than the secretors (58), and the secretion status is a modifying factor for gut microbiota composition in Chron’s disease (59). Fucose from fucosylated proteins synthesized by FUT2 in response to the activation of the innate immunity can be used by microbes as an energy source, and this has been shown to reduce bacterial infection and downregulate the expression of virulence genes (60,61). Fucosylation appears to be a protective mechanism to maintain host-microbial interactions during pathogen-induced stress, but on the other hand, it facilitates viral entrance into the cells.

Although the non-secretor phenotype related mutation in FUT2 (W154X, rs601338) is the most likely causal variant of childhood diarrhoeal disease at 19q13.33 locus, the functional role of other variants in the region cannot be completely ruled out based on statistical analysis. Another variant, located in an intron of the ARHGAP24 gene (rs1481779), reached genome-wide significance in DD1Y, but it could not be replicated. ARHGAP24 codes a Rho GTPase-activating protein involved in cell polarity, cell morphology and cytoskeletal organization; and SNPs in it have been reported to be associated with blood pressure regulation (62).

In order to gain insight into potential molecular mechanisms underlying diarrhoeal disease in young children, we performed enrichment studies. The analysis using two different programs identified several gene-sets at 5% FDR. The ‘KEGG_GLYCOSPHINGOLIPID_BIOSYNTHESIS-GLOBO_SERIES’ gene-set contains genes related to the production of histo-blood group antigens (HBGAs) from globo-series glycolipids and was at least nominally associated with several definitions of diarrhoea (D1Y and DD2Y). Apart from FUT2, already discussed above, A4GALT (Alpha 1,4-Galactosyltransferase) in the gene-set also showed some evidence of association with D2Y. A similar gene-set, ‘KEGG_GLYCOSPHINGOLIPID_BIOSYNTHESIS-GANGLIO_SERIES’, was also identified.

A second gene-set consisting of a SIGMAR1 protein interaction network (ENSG00000147955—SIGMAR1 PPI subnetwork) was detected for DD1Y. SIGMAR1 is an endoplasmic reticulum-resident two-transmembrane chaperone that regulates voltage-gated ion channels including calcium, sodium and potassium channels (63). We also identified other pathways related to hydro-electrolytic balance such as the ‘KEGG_RENIN_ ANGIOTENSIN_SYSTEM’, the ‘GO:0005227—calcium activated cation channel activity’, the ‘GO:0015299—solute:hydrogen antiporter activity’ and the ‘ENSG00000187446—ENSG00000187446 PPI subnetwork’, related to cell pH regulation by controlling plasma membrane-type Na(+)/H(+) exchange activity (64). Deregulation of transport of ions is a central mechanism in the pathophysiology of enteric pathogens (8,9). Host genetic variants might compensate or accelerate watery stools after enteric infection.

Other gene-sets related to ectoderm and endoderm development were observed for DD1Y: ‘MP:0001675—abnormal ectoderm development’ and ‘GO:0007492—endoderm development’. The endoderm is the inner most germ layer that develops into the gastrointestinal tract, the lungs and associated tissues. Finally, the ‘ENSG00000140612—SEC11A PPI subnetwork’ was also identified. SEC11A is a component of the microsomal signal peptidase complex which removes signal peptides from nascent proteins as they are translocated into the lumen of the endoplasmic reticulum.

As expected, the gastrointestinal tract and the immune system were detected as relevant organs for diarrhoeal disease. The gastrointestinal tract was identified with the general definition of diarrhoea around 1 year of age (D1Y); while doctor diagnosis of diarrhoeal disease (DD1Y), probably more related to severe cases, highlighted the immune system. In addition, the neuro-secretory system was found to be associated with diarrhoea around 2 years of age (D2Y). The central nervous system communicates with the intestine through the hypothalamic–pituitary–adrenal axis. The autonomic enteric nervous system regulates gastrointestinal motility, secretion, barrier function and inflammatory response at the mucosa (65). Indeed, gene-sets related to neurotransmission were also identified: ‘Panther—Nicotinic_ acetylcholine_receptor_signaling_pathway’, ‘GO:0090278—negative regulation of peptide hormone secretion’ or a protein network regulating neurotransmitter release ‘ENSG00000147955—SIGMAR1 PPI subnetwork’. Several organs, such as the urogenital and musculoskeletal systems, not a priori of relevance for diarrhoeal disease, were identified for DD2Y.

Finally, given the known opposite effect of the FUT2 locus in infectious diseases compared to autoimmune diseases, we decided to compare our GWAS results with the results reported in the largest GWAS of IBD, for which an infectious component has been suspected (22). Besides the FUT2 locus, where we confirmed the opposite direction of the association, the same allele in a SNP at CARD11 locus, was associated with higher risk of IBD and higher risk of D1Y. CARD11 (Caspase Recruitment Domain Family, Member 11) is involved in the co-stimulatory signal essential for T-cell receptor (TCR)-mediated T-cell activation. We also compared our results with several GWAS of viral infections and response to vaccination (23–36). The TTC7B (Tetratricopeptide Repeat Domain 7B) locus, associated with higher anti Cytomegalovirus IgG titer (31), a marker of either new infection or frequent viral reactivation, was associated with higher risk for D2Y. Other studies have reported the participation of TTC7B in Chikungunya virus infection (66), and a paralog of TTC7B, TTC7A, has been implicated in a rare hereditary disease characterized by intestinal obstructions and profound immune defects (67).

The main limitations of the study are the low specificity of the phenotype definition, and the lack of underlying etiological information for the defined episodes. Information on diarrhoeal disease was retrieved using standard epidemiological tools for population-based birth cohort studies: questionnaires (ALSPAC, Generation R, GINIplus and LISAplus, INMA_VAL and INMA_SAB and MoBa), medical interviews (COSPAC) and medical records (CHOP). Apart from medical record definitions, the most specific assessment of infectious gastroenteritis was in Generation R and in MoBa studies, where the questionnaires included the following statement: ‘infections of the stomach/intestine or gastric flu’. In fact, Generation R showed a low number of diarrhoeal disease cases and a high effect of the variant at FUT2 locus, suggesting less misclassification problems. We acknowledge that misclassification and heterogeneity (i.e. infection by different underlying aetiologies, bacterial species or viral strains or seasonality) may have decreased the statistical power in the discovery phase and might explain the lack of replication of suggestive hits besides the FUT2 locus. Furthermore, cohorts with different study designs participated in the analyses: mainly population-based birth cohorts, but also a random collection of paediatric children from a hospital centre (CHOP, replication), and a population-based birth cohort of infants born to mothers with a history of asthma (COPSAC2000, discovery). The potential relationship between asthma and infection diseases might have affected the associations. To the best of our knowledge, there are no population-based studies of this sample size with molecular diagnosis of infectious gastroenteritis. Although we analyzed all available samples from the EArly Genetics and Life course Epidemiology (EAGLE) consortium following a flexible inclusion criteria the sample size is still intermediate for genome-wide scale studies. Therefore, increasing the specificity of the outcome and the sample size in population-based designs might reveal novel loci for childhood diarrhoeal disease and confirm the role of host genetics in infectious diseases during the first years of life. Finally, although diarrhoeal disease in young children is mainly caused by Rotavirus, which is a wide-spread virus producing seasonal break-outs in all the countries included in this study, we cannot completely exclude the possibility that the identified variants reflect different levels of exposure to the virus, rather than a higher susceptibility to infection.

In summary, the genome-wide association meta-analysis of diarrhoeal disease in children suggested the implication of the FUT2 locus at the population level, and has pointed to W154X (rs601338) as the most likely causal variant. The histo-blood group antigen (HBGA) production and the regulation of ion transport were plausible underlying biological mechanisms accounting for part of the host genetic variability of diarrhoeal disease.

Materials and Methods

Sample and diarrhoeal disease definition

This study was performed within the framework of the Early Genetics and Life course Epidemiology (EAGLE) Consortium (http://www.wikigenes.org/e/art/e/348.html) and it was divided in two phases: discovery and replication. The following population-based birth cohorts or studies settled up in developed countries participated in the discovery and/or replication phases (Table 1): the Avon Longitudinal Study of Parents And Children (ALSPAC), the Children's Hospital of Philadelphia (CHOP) study, the Copenhagen prospective studies on asthma in childhood (COPSAC2000 and COPSAC2010), the Generation R study, the Influence of Life-style related factors on the development of the Immune System and Allergies in East and West Germany plus the influence of traffic emissions and genetics (LISAplus) study, the Study on the influence of Nutrition Intervention plus Air pollution and Genetic on Allergy development (GINIplus), the Infancia y Medio Ambiente (INMA) project and the national Norwegian Mother and Child Cohort Study (MoBa). Cohorts were allocated in the discovery or in the replication set with the aim of making both sets comparable. Diarrhoeal disease was defined in two different time windows: around age 1 year (from 0 months to 18 months) and around age 2 years (from 12 to 30 months). At each time point diarrhoea (D1Y and D2Y) and doctor diagnosis of diarrhoea (DD1Y and DD2Y) were studied. Data were collected from parental questionnaires, doctor interviews or medical records. A detailed description of diarrhoeal disease definitions in each cohort at each time point is described in the Supplementary Material, Annex A. A comparison of the year of initiation of the cohort vs. the year of introduction of Rotavirus vaccination in each country can be seen in Supplementary Material, Annex B. The vaccine was introduced in the USA at the time when children from CHOP study were enrolled, and thus we excluded vaccinated children, identified through medical records, from the analysis.

Each cohort obtained the ethical approval from the respective Ethical Committees and a written consent including permission to perform GWAS analyses was obtained from participating parents.

Genotyping, quality control and imputation

Genotypes within each cohort were collected using high-density SNP arrays on Illumina (ALSPAC, CHOP, COPSAC2000, COPSAC2010, Generation R, INMA_SAB, INMA_VAL, MoBa) and Affymetrix (GINIplus and LISAplus) platforms. Each cohort imputed up to ∼30 M variants using MACH (68) or IMPUTE2 (69) considering the 1000 Genomes Project CEU release March 2012 (http://mathgen.stats.ox.ac.uk/impute/ALL_1000G_phase1integrated_v3_impute.tgz) as the reference population panel. More details on the process followed by each cohort are described in Supplementary Material, Annex A.

Analysis, meta-analysis and replication

The study included only at term Caucasian singletons and children with congenital anomalies were excluded. The association between diarrhoeal disease and the variant dose was assessed in each study using logistic regression analyses assuming an additive genetic model. Sex and principal components accounting for genetic sub-stratification were added as covariates. Chromosome X was analysed under the same statistical model but without sex adjustment. More details on the programs used by each study to perform the analysis can be found in Supplementary Material, Annex A.

Only variants with a Minor Allele Frequency (MAF) ≥0.01 and with a quality of imputation ≥ 0.4 (INFO) or ≥ 0.3 (R2) were considered. Due to the limited sample size of some cohorts, an additional filtering based on expected minor allele counts (EMAC) was performed. This parameter is related to both the sample size and the quality of imputation (2*N*MAF*quality of imputation). Variants that did not reach an EMAC ≥ 50 were excluded. After quality control, from 5.4 to 8.7 million variants were kept for the analysis in each cohort. The genomic inflation factor lambda (λ) was calculated for each study. A summary of the quality control procedure is shown in Supplementary Material, Table S14. Marker names and alleles were harmonized among studies.

A fixed effect meta-analysis weighted by inverse variance was conducted using GWAMA (70). The genomic control approach was applied to the meta-analysis results. Only variants with data for at least 5000 samples were considered. Genome-wide level of significance was defined at P-value ≤ 5E−08 and suggestive associations at P-value ≤1E−05. Quantile–quantile (Q–Q) plots, calculation of lambda (λ) and Manhattan plots were performed in R software environment version 3.2.3 (71). Regional association plots were performed with Locus Zoom (72).

In total 72 variants were followed for replication in an independent dataset. They were selected among the four outcomes based on the statistical significance (P-value < 1E−05). Association P-values from the replication phase were corrected for multiple testing using Bonferroni correction (for each trait independently). Exclusion of CHOP cohort, as it comprises potentially vaccinated children, from the replication phase did not revealed any new replicated genetic variant, and the association of replicated variants was maintained (data not shown).

Conditional analysis

We conditioned the analysis of the leading SNP identified at 19q13.33 (rs8111874) to a stop mutation situated 37.7 kb apart (rs601338, W154X) using the GCTA program (73). As reference, we used the INMA 1000 Genomes imputation [restricted to variants with MAF > 0.01 and imputation quality (INFO) >0.8].

A similar analysis was performed to search for secondary signals in ±500 kb surrounding the stop mutation (rs601338) or the top SNP at 19q13.33 (rs8111874). In this case, the significance threshold was calculated by Bonferroni correction, where the number of independent tests was the effective number of variants in this region estimated using Nyholt’s procedure and the 1000G reference data for Europeans (74).

Annotation and enrichment analysis

Genetic variants annotation (nearest gene, eQTLs, protein binding, and regulatory features) was done with the HaploReg v4.1 program (75). In addition, a second search for eQTLs and for expression levels in tissues was performed with Genotype-Tissue Expression (GTEx) data (http://www.gtexportal.org/; date last accessed July 2016). GeneCard (http://www.genecards.org; date last accessed July 2016) and the USC genome browser (http://genome.ucsc.edu/; date last accessed July 2016) were used to search for gene functions and GWAS signals, respectively.

Two different tools were used to explore gene-set enrichment analysis. We performed an analysis with MAGENTA software that uses genome-wide summary statistics (76). Briefly, first, the program links variants to genes considering flanking regions and then computes the gene-set enrichment comparing the variants with the lowest P-values (95th percentile) versus the rest. Gene-set databases evaluated in MAGENTA were Panther, KEGG (Kyoto Encyclopaedia of Genes and Genomes) and Ingenuity. In addition, we used DEPICT to identify enriched genes-sets as well as tissues/cell types where genes from associated loci are highly expressed (77). In this case, only variants associated with diarrhoea with a P-value ≤  1E−05 and a sample size >5000 were considered. Both programs estimate adjusted P-values using the FDR method. In the case of MAGENTA, FDR was applied within each database.

Overlap with known variants and genes for related diseases

We investigated the association between diarrhoea and known variants for inflammatory bowel disease (IBD) (22). We also looked at variants associated with viral infection susceptibility, disease progression and response to vaccination against different viruses (23–36). Variants were selected from GWAS retrieved from the GWAS catalog (http://www.ebi.ac.uk/gwas/, April 2015). All variants reported in European populations, regardless of their statistical significance, and their replication status, were evaluated. Finally, we evaluated 86 genes retrieved from OMIM (Online Mendelian Inheritance in Man; July 2016) with the entry ‘diarrhoea’. To test the association of these ‘candidate’ genes, we performed a gene-based analysis using VEGAS2 (Versatile Gene based Association Study, http://vegas2.qimrberghofer.edu.au/) (78), considering linkage disequilibrium patterns described in European populations and a flanking region of ±50 kb around the gene. The corrected statistical significance level was calculated using Bonferroni correction accounting for the number of variants/genes within each analysis (inflammatory bowel disease, viral infection, or hereditary diarrhoeal diseases) as independent tests.

Genome-wide summarized results of the discovery phase can be found at INMA’s web page (Infancia and Medio Ambiente project, http://www.proyectoinma.org/) and at the EAGLE consortium web page (http://www.wikigenes.org/e/art/e/348.html).

Supplementary Material

Supplementary Material is available at HMG online.