Integration of exome databases

The gnomAD version 2.1.1 exome data are available at https://gnomad.broadinstitute.org/downloads; and we downloaded data in the VCF format. The data included 17,209,972 variants from 125,748 samples of eight populations. The VCF file contained the rs-number, AN_POP for Allele Count, and AC_POP for Alt Allele Count for a population POP, where POP is one of the following: afr (African/African American), amr (Latino/Admixed American), asj (Ashkenazi Jewish), eas (East Asian), fin (Finnish), nfe (Non-Finish European), sas (South Asian), and oth (Other). We calculated the number of reference alleles as AN_POP–AC_POP for each variant per population.

Additionally, we downloaded GME Variome autosomal data from http://igm.ucsd.edu/gme/download.shtml. This project provides genotype frequencies of 669,953 variants on the autosomal chromosome for seven regions in the Greater Middle East (Northeast Africa, Northwest Africa, Arabian Peninsula, Israel, Syrian Desert, Turkish Peninsula, and Central Asia). The Arabian Peninsula included the following regions: Kuwait, Oman, Qatar, Kingdom of Saudi Arabia, the United Arab Emirates, and Yemen [9].

Since both databases are based on the human reference genome GRCh37/hg19, we merged the data when the chromosome number, nucleotide position, reference allele, and alternative allele were the same in both databases for each SNP. We included the alleles if the allele frequencies in each population were >0.01 to exclude rare variants. Because gnomAD data are based on populations, and GME Variome are based on their definitions; thus, we use “populations/regions” as a unit for this analysis. We disregarded the differences in allele frequencies among other populations/regions to identify important SNPs that are unique to the Arabian Peninsula population. We summed up the numbers of alleles over the populations/regions except for the Arabian Peninsula to compare allele frequencies between the Arabian Peninsula population and the other populations/regions (which will be referred to as "other regions" hereafter). In this study, we ignored the differences among populations in “other regions.”

Identification of T2D-related SNPs

To obtain T2D-related SNPs, we searched the ClinVar database (https://www.ncbi.nlm.nih.gov/clinvar/) with the keyword ‘”type 2 diabetes”[dis] OR “diabetes mellitus type 2”[dis]’. We extracted SNPs that were common among gnomAD exome, GME Variome, and the result of keyword search at the ClinVar database based on the chromosome, position, reference allele, and alternative allele. ClinVar provides clinical significance, which we used to evaluate the clinical significance of SNPs.

Statistical analyses

To compare the allele frequencies and genotype frequencies between the Arabian Peninsula and other regions, we divided the data into the Arabian Peninsula and the other regions.

The observed number of each allele in the Arabian Peninsula was calculated as follows:

For the other regions, the calculation was as follows:

Then, the expected allele counts in the Arabian Peninsula were calculated as follows:

Additionally, we calculated the expected values for the genotype count using the same method. These expected allele counts represented the allele counts when the allele frequencies of each population were equal.

We performed Fisher’s exact test for each SNP to identify statistically significant SNPs that were not independent for the population, allele counts, or genotype counts. The Benjamini–Hochberg method was used with a q-value of 0.05 for the correction of multiple tests. For SNPs that were statistically significant, we further examined them using a residual analysis to reveal the differences between the observed counts and expected counts.

Integration of gnomAD exome and GME Variome

Before the GME Variome data became available, genotype data for the Middle Eastern populations were scarce. In 2016, an exome project on the Greater Middle Eastern region was finally published [9], which allowed for analyses of worldwide genotype data including Middle Eastern populations by integrating gnomAD exome and GME Variome data.

We downloaded 17,209,972 and 669,953 variants from gnomAD exome and GME Variome, respectively. We merged the genotype information for gnomAD exome and GME Variome and obtained 475,239 variants for 15 populations/regions. After eliminating the rare variants, we obtained 195,718 variants (Fig 1a). Then, we summed up the numbers of alleles for the Arabian Peninsula and other regions and calculated the expected value for each variant. The allele frequency spectra of overlapped variants showed similar tendencies among the Arabian Peninsula and other regions (S1a Fig) and also among each population/region (S1b Fig).

Fig 1

Number of variants.

(a) The intersection between the gnomAD exome and GME Variome data. 475,239 variants commonly exist in two databases. After the eliminating rare variants, 195,718 variants remained. (b) Intersection among the gnomAD exome, GME Variome, and T2D-related variants.

Significant T2D-related SNPs in the Arabian Peninsula

In addition to the genotype data, we integrated clinical information via ClinVar to reveal genetic features related to T2D. We downloaded 176 T2D-related variants from ClinVar as of January 17, 2021. Only 18 SNPs were common among the three databases (Table 1 and Fig 1b). To identify SNPs that were specifically biased in the Arabian Peninsula subpopulations, we performed Fisher’s exact test on these 18 SNPs. Even though the variants were mainly derived from populations outside of the Middle East, we found that six SNPs were statistically significant (Table 2). Two of the six SNPs were clinically significant, as described below. Among the two statistically and clinically significant SNPs, residual analysis revealed that a variant, rs13266634, increased the number of the risk allele (Fig 2).

Fig 2

Standardized residuals of alleles.

White: non-risk allele and diagonal stripes: risk allele. (a) SLC30A8 rs13266634. (b) KCNJ11 rs5219.

Table 1

List of T2D-related SNPs.

rs-number	Chr	Position	REF	ALT	Change	gene	protein	Clinical significance
rs41265094	2	227661003	C	G	c.2452G>C (p.Gly818Arg)	IRS1	insulin receptor substrate 1	Likely benign
rs1801276	2	227661921	C	G	c.1534G>C (p.Ala512Pro)	IRS1	insulin receptor substrate 1	Benign
rs1044498	6	132172368	A	C	c.517A>C (p.Lys173Gln)	ENPP1	ectonucleotide pyrophosphatase/phosphodiesterase 1	Benign
rs1799999	7	113518434	C	A	c.2713G>T (p.Asp905Tyr)	PPP1R3A	protein phosphatase 1 regulatory subunit 3A	Uncertain significance
rs13266634	8	118184783	C	T	c.973C>T (p.Arg325Trp)	SLC30A8	solute carrier family 30 member 8	risk factor
rs5219	11	17409572	T	C	c.67A>G (p.Lys23Glu)	KCNJ11	potassium inwardly rectifying channel subfamily J member 11	drug response
rs566325901	12	121177120	A	G	c.1108A>G (p.Met370Val)	ACADS	acyl-CoA dehydrogenase short chain	Conflicting interpretations of pathogenicity
rs121434581	17	7189048	G	A	c.1147G>A (p.Val383Ile)	SLC2A4	solute carrier family 2 member 4	Uncertain significance
rs1801483	17	79767715	G	A	c.118G>A (p.Gly40Ser)	GCGR	glucagon receptor	Benign
rs41309435	19	40741862	C	A	c.1110G>T (p.Pro370 =)	AKT2	AKT serine/threonine kinase 2	Benign
rs776435289	19	40743996	C	G	c.711G>C p.Leu237 =)	AKT2	AKT serine/threonine kinase 2	Likely benign
rs199761368	19	40744854	G	A	c.666C>T (p.His222 =)	AKT2	AKT serine/threonine kinase 2	Benign
rs139125633	19	40761070	G	A	c.282C>T (p.Asp94 =)	AKT2	AKT serine/threonine kinase 2	Likely benign
rs35588791	19	40762915	G	A	c.93C>T (p.Ser31 =)	AKT2	AKT serine/threonine kinase 2	Benign
rs1799816	19	7125518	C	T	c.3034G>A (p.Val1012Met)	INSR	insulin receptor	Conflicting interpretations of pathogenicity
rs142204928	20	43043159	G	A	c.439G>A (p.Val147Ile)	HNF4A	hepatocyte nuclear factor 4 alpha	Conflicting interpretations of pathogenicity
rs147638455	20	43058267	A	G	c.1321A>G (p.Ile441Val)	HNF4A	hepatocyte nuclear factor 4 alpha	Uncertain significance
rs2076026	20	43942676	T	C	c.759T>C (p.Ala253 =)	RBPJL	recombination signal binding protein for immunoglobulin kappa J region like	Benign

Positions are on the Assembly GRCh37. Clinical significance is based on ClinVar. Chr: Chromosome number, REF: Reference allele, ALT: Alternative allele, Change: Substitutions of nucleotide and amino acid.

Table 2

Results of Fisher’s exact test for T2D-related SNPs.

	Allele Count					Genotype Count
rs-number	Observed		Expected		Fisher	Observed			Expected			Fisher
	REF	ALT	REF	ALT		REF/REF	REF/ALT	ALT/ALT	REF/REF	REF/ALT	ALT/ALT
rs41265094	342	0	339	3	0.249	171	0	0	168	3	0	0.248
rs1801276	342	0	337	5	0.062	171	0	0	166	5	0	0.061
rs1044498 #	245	97	275	67	0.009	92	61	18	115	45	11	0.037
rs1799999	285	57	267	75	0.099	117	51	3	109	49	13	0.039
rs13266634 *#§	298	44	244	98	0.000	130	38	3	88	68	15	0.000
rs5219 *#§	70	272	123	219	0.000	12	46	113	23	76	72	0.000
rs566325901	341	1	341	1	1.000	170	1	0	170	1	0	1.000
rs121434581	337	5	341	1	0.217	167	3	1	170	1	0	0.371
rs1801483 #§	225	15	238	2	0.002	106	13	1	118	2	0	0.003
rs41309435	342	0	341	1	1.000	171	0	0	170	1	0	1.000
rs776435289	342	0	342	0	1.000	171	0	0	171	0	0	1.000
rs199761368	341	1	341	1	1.000	170	1	0	170	1	0	1.000
rs139125633	342	0	342	0	1.000	171	0	0	171	0	0	1.000
rs35588791	342	0	341	1	1.000	171	0	0	170	1	0	1.000
rs1799816	335	7	339	3	0.340	164	7	0	168	3	0	0.337
rs142204928	341	1	341	1	1.000	170	1	0	170	1	0	1.000
rs147638455 #§	331	11	342	0	0.001	160	11	0	171	0	0	0.001
rs2076026 #	329	13	311	31	0.007	159	11	1	143	25	3	0.016

Interestingly, for rs13266634, the homozygous risk genotype increased whereas the heterozygous genotype decreased in the Arabian Peninsula (Fig 3). The remaining SNP was rs5219. The number of this risk allele decreased in the Arabian Peninsula (Fig 2).

Fig 3

Standardized residuals of genotypes.

White: homozygote non-risk alleles, dotted: heterozygote, and diagonal stripes: homozygote of risk alleles. (a) SLC30A8 rs13266634. (b) KCNJ11 rs5219.

The numbers of alleles and numbers of genotypes were tested. In the rs-number column, * represents the clinically significant SNP, # represents statistically significant SNP according to Fisher’s exact test the allele count, and § represents statistically significant SNP by Fisher’s exact test using the genotype count. The Benjamini–Hochberg method was used with a q-value of 0.05 for the correction of multiple tests. Observed and expected numbers of alleles and genotypes are shown. REF: reference allele, ALT: alternative allele, REF/REF: homozygote of reference allele, REF/ALT: heterozygote, ALT/ALT: homozygote of alternative allele, Fisher: p-value of Fisher’s exact test.