Identification of novel risk genes associated with type 1 diabetes mellitus using a genome-wide gene-based association analysis.

AIMS/
INTRODUCTION: Type 1 diabetes mellitus is a serious disorder characterized by destruction of pancreatic β-cells, culminating in absolute insulin deficiency. Genetic factors contribute to the susceptibility of type 1 diabetes mellitus. The aim of the present study was to identify more susceptibility genes of type 1 diabetes mellitus.
MATERIALS AND METHODS: We carried out an initial gene-based genome-wide association study in a total of 4,075 type 1 diabetes mellitus cases and 2,604 controls by using the Gene-based Association Test using Extended Simes procedure. Furthermore, we carried out replication studies, differential expression analysis and functional annotation clustering analysis to support the significance of the identified susceptibility genes.
RESULTS: We identified 452 genes associated with type 1 diabetes mellitus, even after adapting the genome-wide threshold for significance (P < 9.05E-04). Among these genes, 171 were newly identified for type 1 diabetes mellitus, which were ignored in single-nucleotide polymorphism-based association analysis and were not previously reported. We found that 53 genes have supportive evidence from replication studies and/or differential expression studies. In particular, seven genes including four non-human leukocyte antigen (HLA) genes (RASIP1, STRN4, BCAR1 and MYL2) are replicated in at least one independent population and also differentially expressed in peripheral blood mononuclear cells or monocytes. Furthermore, the associated genes tend to enrich in immune-related pathways or Gene Ontology project terms.
CONCLUSIONS: The present results suggest the high power of gene-based association analysis in detecting disease-susceptibility genes. Our findings provide more insights into the genetic basis of type 1 diabetes mellitus.

Introduction

Type 1 diabetes mellitus is a serious disorder characterized by destruction of pancreatic β-cells, leading to absolute insulin deficiency. Type 1 diabetes mellitus arises from uncontrolled inflammatory processes, and accounts for 5–10% of total cases of diabetes worldwide1. Diabetes mellitus is a major risk factor for micro- and macrovascular complications, and is associated with endothelial dysfunction, premature atherosclerosis and reduced capability of neovascularization in ischemic conditions. The increasing number of people developing diabetes might be associated with the changing environment in relation to diet and infection, but more with genetic factors2,3. Identification of genes predisposing to type 1 diabetes mellitus will increase our understanding of the genetic pathogenesis of type 1 diabetes mellitus, and contribute to the development of novel prevention and treatment of type 1 diabetes mellitus in the future.

Extensive evidence has shown that genetic factors play important roles in the development of type 1 diabetes mellitus3–5. However, identification of specific responsible genes and their variants has had limited success. The single-nucleotide polymorphism (SNP)-based genome-wide association studies (GWAS) have identified a long list of risk genes for type 1 diabetes mellitus6,7. However, the traditional GWAS ignored a large number of loci with moderate effects, because of the stringent significance thresholds used.

Gene-based analysis takes a gene as a basic unit for association analysis. As this method can combine genetic information given by all the SNPs in a gene to obtain more informative results8, it is being used as a novel method complementing SNP-based GWAS to identify disease susceptibility genes. Notably, this method can increase our chance of finding novel genes, which are usually ignored by SNP-based association analysis. In the present study, we presented a statistically robust gene-based GWAS, focusing on identifying ‘novel’ genes underlying susceptibility to type 1 diabetes mellitus.

Materials and Methods

Discovery Study Sample

The initial discovery sample included 4,075 cases and 2,604 controls. The case data came from ‘UK Genetic Resource Investigating Diabetes’ (available at www.childhood-diabetes.org.uk/grid.shtml). The control participants came from the 1958 British Birth Cohort. Genotyping, data-quality filter and SNP-based association analysis were detailed in the original publication in Nat Genet7, thus not elaborated here.

Replication Study Sample

Replication analyses were carried out in three independent study samples (Replication sample 1 [R1]: a total of 1,879 samples including 935 diabetic nephropathy cases and 944 normoalbuminuric controls; replication sample 2 [R2]: 486 trios including 223 affected trios and 263 unaffected trios; replication sample 3 [R3]: 685 white individuals with type 1 diabetes from the Diabetes Control and Complications Trial [DCCT]/Epidemiology of Diabetes Intervention and Complications [EDIC] study). Basic characteristics of the study participants, as well as the genotyping process, data quality control and SNP-based association analysis, have been detailed previously in the original publications9–11, thus not elaborated here.

Gene-Based Association Analysis

Raw data used in the present gene-based GWAS analysis and replication studies are P-values from genome-wide SNP-based GWAS. The data were downloaded from the publicly available dbGaP database (accession number: phs000180, phs000018, phs000086 and phs000088). Gene-based association analysis was carried out using Gene-based Association Test using Extended Simes procedure12, and the resultant gene-based P-value is a measure of statistical significance. The Gene-based Association Test using Extended Simes procedure was modeled in KGG, a systematic biological Knowledge-based mining system for Genome-wide Genetic studies (available at http://bioinfo1.hku.hk:13080/kggweb). The defined length of the extended gene region is from 5-kb upstream to 5-kb downstream of each gene.

Differential Expression Analyses of Type 1 Diabetes Mellitus-Associated Genes

Based on the normalized data available in the public databases (www.ncbi.nlm.nih.gov/geo), we tested differential expression of the aforementioned identified ‘novel’ candidate genes by comparing mean gene expression signals in peripheral blood mononuclear cells (PBMCs) or monocytes between type 1 diabetes mellitus cases and controls. Specifically, three gene expression datasets were downloaded from GEO Datasets (GSE number: GSE35725, GSE33440, GSE29142). Patients with type 1 diabetes mellitus show functional abnormalities of monocytes and monocyte-derived cells, which are assumed to promote the immunogenic potential of the cells13. Therefore, transcriptional signatures of patients' PBMCs might reflect inflammatory mechanisms14. The original studies were carried out to identify transcriptional signatures as a disease-specific and predictive inflammatory biomarker for type 1 diabetes mellitus14,15. Details on sample quality control and experimental procedures are previously described in the original publications14–16.

Functional Annotation Clustering Analysis

To gain insights into the functions of the identified genes, we tested the probability of the 452 identified genes clustering into a specific Gene Ontology (GO) project or a particular biological pathway that is defined by GO term or Kyoto Encyclopedia of Genes and Genomes (KEGG) database. Specifically, a functional annotation clustering analysis was carried out by using the Database for Annotation, Visualization and Integrated Discovery (DAVID) integrated database query tools (http://david.abcc.ncifcrf.gov/)17. The enrichment can be quantitatively measured by using Fisher's exact test. The Bonferroni correction was used for multiple testing18.

Results

Discovery of Novel Genes Associated With Type 1 Diabetes Mellitus

A total of 24,984 genes were analyzed in the initial gene-based GWAS. Three quantile–quantile plots for gene-based P-values, SNP-based P-values inside genes and SNP-based P-values outside genes are shown in Figure1. We observed dramatic deviations at the tails of the distributions for the three plots. The deviation was much stronger for the plot of gene-based P-values than the other two plots, suggesting relatively higher power for gene-based association analysis.

Figure 1

Q–Q plot of P-values. There are three quantile–quantile plots of the observed P-value distributions. As compared with the expected null P-value distributions, the tail of the distribution for gene-based P-values is the most significant deviation. SNPs, single-nucleotide polymorphisms.

The Manhattan plot of gene-level P-values across chromosomes is showed in Figure2. As expected, the majority of these associations were mapped to the HLA region, whose physical location lies in chromosome 6p21. To adjust for multiple testing, we used the Benjamini–Hochberg19 procedure to control false discovery rate (FDR) of the genome-wide association tests. To obtain a FDR of 0.05 across the whole genome, the significance level for a gene-based test is 9.05E-04. Accordingly, 452 genes were statistically significant (Table S1).

Figure 2

Manhattan plot of gene-based P-values on chromosomes.

In the original SNP-based GWAS analysis, according to the genome-wide P-value threshold of statistical significance (Bonferroni correction P < 9.94E-08), a total of 699 SNPs showed significant associations. These 699 SNPs corresponded to 269 genes. Comparatively, the current gene-based study detected 183 additional candidate genes for type 1 diabetes mellitus that were undetected by the SNP-based association analysis. To discover whether any of the 183 genes had been reported in other previous association studies, we searched the Phenotype-Genotype Integrator (PheGenI; www.ncbi.nlm.nih.gov/gap/PheGenI/), a database archiving previous association results. We compared the 183 genes and the list of genes with significant SNP-based P-values (P < 1.0E-07; Table S1). This comparison showed that just 12 among the total 183 genes were previously reported for an association with type 1 diabetes mellitus7,20–22. The rest of the 171 ‘novel’ genes were first detected for type 1 diabetes mellitus by the present study.

Confirmation of Type 1-Associated Genes by Replication Studies

The results of the replication analyses are summarized in Table S1 (P < 5.0E-02). For the 171 ‘novel’ genes, only one gene was replicated for associations with type 1 diabetes mellitus in study R1. In addition, eight genes were replicated in study R2 and 14 genes were replicated in study R3. In total, 23 of the 171 ‘novel’ genes (Table1) were confirmed for their association with type 1 diabetes mellitus. For the replication studies, the significance level of P < 5.0E-02 was used.

Table 1

‘Novel’ type 1 diabetes mellitus-associated genes with supplementary evidence

Gene symbol	Chr.	Start position	Length	SNP†	Gene P-value	Replication P-value		Differential expression P-value
RASIP1	19	49,218,842	30,128	6	4.83E –04	2.94E –02	R2	3.18E –03	,S1
STRN4	19	47,217,768	36,952	5	1.97E –04	2.67E –02	R2	1.63E –02	,S1
BCAR1	16	75,257,928	49,023	8	9.32E –05	2.27E –02	R2	4.17E –02	,S1
FYN	6	111,977,485	222,142	48	3.08E –04	4.69E –03	R2	2.01E –03	,S1
MYL2	12	111,343,623	19,781	5	4.61E –04	8.53E –03	R3	7.58E –03	,S1
HLA-J	6	29,968,748	13,985	1	3.49E –06	3.89E –02	R3	1.63E –02	,S3
PPP1R11	6	30,029,932	13,178	4	5.50E –06	2.99E –02	R3	2.33E –031.69E –02	,S2,S3
ITPR3	6	33,584,161	85,190	29	4.36E –05	NS‡		4.29E –03	,S1
PLEKHA1	10	124,129,220	67,646	11	9.92E –05	NS‡		1.04E –02	,S1
ATF7IP	12	14,513,611	143,086	13	8.83E –05	NS‡		2.10E –03	,S1
OR2B6	6	27,920,019	10,941	2	2.04E –06	NS‡		2.65E –03	,S1
OR5V1	6	29,318,007	11,047	2	2.16E –06	NA§		3.07E –03	,S1
HIST1H4E	6	26,199,873	10,376	2	3.48E –06	NA§		4.29E –03	,S1
HIST1H2BF	6	26,194,787	10,429	1	6.65E –06	NS‡		8.84E –03	,S1
GUSBL1	6	26,834,266	95,067	2	3.48E –07	NA§		1.40E –02	,S1
HMGB1	13	31,027,877	17,204	2	2.07E –04	NA§		1.99E –02	,S1
ZNF192	6	28,104,716	25,520	1	3.65E –06	NS‡		3.92E –02	,S1
RING1	6	33,171,286	14,213	4	1.45E –04	NS‡		4.40E –02	,S1
FAM46B	1	27,326,511	17,822	1	6.36E –05	NA§		4.97E –022.64E –02	,S1,S2
OLFML3	1	114,517,030	12,845	2	2.13E –04	NS‡		7.24E –031.54E –02	,S1,S3
HIPK1	1	114,466,996	58,426	6	5.54E –04	NS‡		3.02E –023.56E –02	,S1,S3
NSL1	1	212,894,495	75,644	12	5.48E –05	NS‡		2.27E –02	,S2
IL10	1	206,935,948	14,891	8	2.28E –04	NS‡		1.05E –02	,S2
TRIM27	6	28,865,779	30,989	3	8.88E –07	NS‡		8.79E –03	,S2
NCAPD2	12	6,598,298	47,834	10	1.36E –04	NS‡		1.40E –04	,S2
PPP1R10	6	30,563,182	26,838	2	4.08E –04	NS‡		3.95E –02	,S2
VPS52	6	33,213,049	31,613	2	6.73E –07	NA§		1.32E –02	,S2
PHF1	6	33,373,773	15,457	2	3.14E –04	NA§		6.98E –03	,S2
BAK1	6	33,535,323	17,747	7	5.18E –05	NS‡		6.11E –031.60E –02	,S2,S3
IKZF3	17	37,916,198	109,243	7	7.54E –05	NS‡		1.87E –04	,S3
ZZEF1	17	3,902,739	148,514	23	3.55E –04	NS‡		1.46E –03	,S3
GNS	12	65,102,222	56,004	4	6.65E –05	NS‡		1.83E –03	,S3
ORMDL3	17	38,072,296	16,558	3	7.22E –04	NS‡		2.96E –02	,S3
BRAP	12	112,074,950	53,840	3	1.16E –04	NS‡		4.45E –02	,S3
CRYZL1	21	34,956,647	62,513	3	3.72E –04	NA§		8.59E –03	,S3
SULT1A1	16	28,611,913	27,953	1	4.74E –04	NA§		2.61E –02	,S3
TMEM129	4	1,712,679	15,405	3	7.99E –04	NS‡		4.31E –02	,S3
IKZF1	7	50,439,231	38,568	15	6.66E –05	4.42E –02	R1	NA§
MICA	6	−1	0	3	7.52E –05	7.52E –05	R2	NA§
PLBD1	12	14,651,597	74,194	11	3.17E –04	3.26E –02	R2	NA§
DEXI	16	11,017,748	23,509	3	8.72E –04	3.03E –03	R2	NA§
SBK1	16	28,298,840	41,330	7	2.28E –05	1.37E –02	R2	NA§
GCA	2	163,195,583	28,566	4	9.67E –06	4.16E –02	R3	NA§
OR2B3	6	29,048,985	11,105	1	5.27E –07	3.31E –02	R3	NA§
HCP5P2	6	29,963,782	12,246	1	3.49E –06	3.89E –02	R3	NA§
HCG4P3	6	29,967,622	10,983	1	3.49E –06	3.89E –02	R3	NA§
OR2U1P	6	29,225,436	11,420	5	1.05E –06	4.73E –02	R3	NA§
VN1R14P	6	26,626,313	10,651	1	6.19E –06	2.94E –02	R3	NA§
HCG2P8	6	29,767,896	13,543	1	6.02E –07	3.80E –02	R3	NA§
HCGVIII-2	6	29,796,425	1,1279	2	4.51E –06	4.26E –02	R3	NA§
RPS10P1	6	26,197,351	10,592	2	3.48E –06	4.93E –02	R3	NA§
CCDC101	16	28,560,249	47,862	6	1.59E –04	2.98E –02	R3	NA§
FUT2	19	49,194,228	19,963	6	3.55E –04	3.28E –02	R3	NA§

This table includes 53 significant type 1 diabetes mellitus-associated genes with supplemental evidence. Among these 53 genes, 37 were differentially expressed, and 23 genes have been replicated. The first seven genes in this table (RASIP1, STRN4, BCAR1, FYN, HLA-J, PPP1R1 and MYL2) have both differential expression and replication association. The numbers before and after ‘,’ are the differential expression P-value and sample number listed in Table2, respectively.

The number of single-nucleotide polymorphisms (SNP) included in a gene.

Not significant in R1, R2 and R3.

Not available in R1, R2 and R3.

Table 2

Differential expression analyses for ‘novel’ genes in type 1 diabetes mellitus-related cells

Sample	S1		S2			S3
Target cells	PBMC		CD14+ Monocyte			PBMC
Sample size	46:44		16:6			9:10
Platform	[HG-U133_Plus_2] Affymetrix Human Genome U133 Plus 2.0 Array		Illumina HumanHT-12 V3.0 expression beadchip			Phalanx Human One Array (version 4.3)
References	14		15			16
GSE NO.	GSE35725		GSE33440			GSE29142
Gene	Probe ID	t-test P-value	Gene	Probe ID	t-test P-value	Gene	Probe ID	t-test P-value
ATF7IP	207728_at	2.10E-03	FYN	ILMN_1781207	2.64E-02	HIPK1	14,914	3.56E-02
BCAR1	223116_at	4.17E-02	BAK1	ILMN_1805990	6.11E-03	OLFML3	21,862	1.54E-02
FAM46B	229518_at	4.97E-02	IL10	ILMN_2073307	1.05E-02	BAK1	22,739	1.60E-02
FYN	1559101_at	2.01E-03	NCAPD2	ILMN_1775008	1.40E-04	PPP1R11	3205	1.69E-02
GUSBL1	1555568_at	1.40E-02	NSL1	ILMN_1739210	2.27E-02	BRAP	12,945	4.45E-02
HIST1H2BF	208490_x_at	8.84E-03	PHF1	ILMN_1746968	6.98E-03	CRYZL1	24,489	8.59E-03
HIST1H4E	206951_at	4.29E-03	PPP1R10	ILMN_1659058	3.95E-02	GNS	8137	1.83E-03
HMGB1	200679_x_at	1.99E-02	PPP1R11	ILMN_1747598	2.33E-03	HLA-J	24,451	1.63E-02
ITPR3	201187_s_at	4.29E-03	TRIM27	ILMN_1655482	8.79E-03	IKZF3	15,319	1.87E-04
MYL2	209742_s_at	7.58E-03	VPS52	ILMN_1666632	1.32E-02	ORMDL3	10,233	2.96E-02
OLFML3	218162_at	7.24E-03				SULT1A1	20,808	2.61E-02
OR2B6	216522_at	2.65E-03				TMEM129	17,978	4.31E-02
OR5V1	234840_s_at	3.07E-03				ZZEF1	16,832	1.46E-03
PLEKHA1	219024_at	1.04E-02
RASIP1	220027_s_at	3.18E-03
RING1	208371_s_at	4.40E-02
STRN4	217903_at	1.63E-02
ZNF192	206579_at	3.92E-02
CRYZL1	1552347_at	2.78E-02

Only the most significant probe was listed, even if more than one probe was tested or detected for a gene.

GSE NO, Gene Expression Omnibus Number (www.ncbi.nlm.nih.gov/geo/); PBMC, peripheral blood mononuclear cells.

Differential Expression Analyses of Type 1 Diabetes Mellitus Associated Genes

For the aforementioned 171 ‘novel’ genes, we used t-test to compare ribonucleic acid expression signals in PBMCs or monocytes between type 1 diabetes mellitus patients and healthy controls. We found that 37 genes, including 21 non-HLA genes (e.g. FAM46B, OLFML3 and HIPK1), were differentially expressed between type 1 diabetes mellitus patients and controls (Table2). For the differential expression study, the significance level of P < 5.0E-02 was used.

In short, through a gene-based association study, we identified 183 type 1 diabetes mellitus-associated genes that were insignificant in the original SNP-based association tests. Among the 183 genes, 171 genes are ‘novel’ genes identified for type 1 diabetes mellitus. Replication studies and/or differential expression studies further supported the significance of 53 genes to type 1 diabetes mellitus. In particular, four non-HLA genes (RASIP1, STRN4, BCAR1 and MYL2) and three HLA genes (FYN, HLA-J and PPP1R11) were validated by both replication and differential expression studies.

Gene ontology analysis showed significant enrichment of 452 identified type 1 diabetes mellitus genes in particular biological terms and pathways. For example, we found a significant clustering (Bonferroni correction P = 1.50E-05) of 33 genes (e.g. HLA-DQB1, HLA-DMB, HLA-DMA) directly involved in immune response (Table S2). These genes tend to enrich in immune-related KEGG pathways (including hsa04940: type I diabetes mellitus, hsa04612: antigen processing and presentation, hsa04672: intestinal immune network for immunoglobulin A production; Table S2).

Discussion

Elucidation of the genetic basis of type 1 diabetes mellitus remains one of the huge challenges in the field of human genetics, which is largely because of the complex nature of the genetic determination for type 1 diabetes mellitus, including polygenic determinations, gene-by-gene and gene-by-environment interactions. Thus far, most of the published GWAS studies reported the results of single-marker-based analysis, where each SNP was analyzed individually6,21,23. Because of the large number of SNPs tested in a GWAS, stringent P-value thresholds for significance (typically P < 5.0E-08) are used to control false positive findings. Consequently, a large number of SNPs with moderate effects are missed. The gene-based association test is an important supplementary method for the SNP-based association test, which combines genetic information given by all the SNPs in a gene, thus obtaining a more informative result24. As we expected, the present gene-based association study identified more significant type 1 diabetes mellitus-susceptibility genes than the SNP-based test. Specifically, 171 ‘novel’ genes were identified for type 1 diabetes mellitus.

Compared with SNP/variant-based association tests, the gene-based association analyses have several distinct advantages. First, by combining the effects of all SNPs assigned to genes into a statistic analysis while correcting for linkage disequilibrium (LD), the gene-based analysis substantially alleviates the multiple-testing burden by reducing the number of tests in a GWAS. Second, the test unit is a gene, which is highly consistent across populations. Therefore, gene-based association analyses ignore confounding factors that are intrinsic in genetic variants-based association tests, such as allele frequencies, LD structure and heterogeneity across diverse human populations25. Finally, the gene-based tests are also ideally suitable for a network (or pathway) approach to interpret findings from GWAS26.

As previously reported2,3,27,28, more than half of the significant association signals for type 1 diabetes mellitus were identified within chromosome 6p21, which is the most important region in the vertebrate genome with respect to infection and autoimmunity, and is crucial in adaptive and innate immunity. The present study also found that more than 50% of the identified ‘novel’ genes (100 HLA genes/171 novel genes) are located at the HLA region. However, among 53 type 1 diabetes mellitus-associated genes validated by replication studies and/or differential expression studies, 28 genes are non-HLA genes. So far, the roles of these non-HLA genes directly in the pathogenesis of type 1 diabetes mellitus are largely unknown, but some of these genes play important roles in immune response. For example, BCAR1 might mediate its diabetogenic impact through impaired β-cell function29. IKZF1 and IKZF3 are members of the Ikaros family of zinc-finger proteins. The gene product is a transcription factor that is important in regulation of B lymphocyte proliferation and differentiation30,31. PLEKHA1 (also known as TAPP1) encodes a pleckstrin homology domain-containing adapter protein. The interaction of TAPP1 adapter proteins with phosphatidylinositol (3, 4)-bisphosphate could regulate B cell activation and autoantibody production. Several studies suggest that TAPP1 might play roles in B and T cell activation, which are necessary and sufficient conditions for immune response32. MYL2 can regulate the coordinated rearrangements of the actin–myosin cytoskeleton, and facilitate early and late events in T cell activation and signal transduction33. High mobility group box-1 (HMGB1) is an important component of the immune response, which can activate immune cells involved in immune process34. Interleukin-10 (IL-10) is a cytokine with anti-inflammatory and immunomodulatory function, which can regulate the biological functions of B and T cells35. GCA is a causal factor in autoimmune pancreatic β-cell destruction36. The aforementioned evidence supports that BCAR1, IKZF1, IKZF3, PLEKHA1, MYL2, HMGB1, IL-10 and GCA might have functional relevance to diabetes mellitus or immune response. In addition, some previous studies suggested that IL-10, ORMDL3 and FUT2 have been associated with type 1 diabetes mellitus7,37. Further studies are required to dissect the roles of these non-HLA genes in the pathogenesis of type 1 diabetes mellitus.

Our replication studies had a relatively low replication rate for the significant genes detected in the initial study. Small sample size (e.g., 486 trios subjects in R2 and 685 subjects in R3), and difference in demography (e.g., discover sample from the UK, whereas three replication study samples from New England, the USA and Canada) and genetic background could contribute to this. Another more important factor might be the difference in case identification. For R1, the case is s diabetic nephropathy patient. For R3, although the case is an individual with type 1 diabetes, the association analysis was carried out between the gene and type 1 diabetes-associated phenotype (E-selectin level in serum).

In conclusion, the findings presented in our study suggest high power for gene-based association analyses in detecting disease-susceptibility genes across the human genome. Our findings point to the involvement of new pathways in the pathogenesis of type 1 diabetes mellitus, and provide more insights into the genetic basis of type 1 diabetes mellitus.