Rare Variant Analysis of Obesity-Associated Genes in Young Adults With Severe Obesity From a Consanguineous Population of Pakistan

Pakistan was recently ranked among the 10 top countries with the highest prevalence of obesity worldwide (1). Apart from the contribution of the prevailing obesogenic environment, the role of genetic factors in the exacerbation of obesity risk in this region with its unique ethnic diversity, together with a remarkably high rate of consanguinity (∼60%), can hardly be ignored (2). In this respect, the Pakistani population presents a unique opportunity to identify novel genes and variants causing severe obesity.

The role of animal models to aid elucidation of putative candidate genes for human monogenic obesity cannot be ignored. The sequencing of these genes has resulted in breakthroughs in our understanding of the mechanisms regulating appetite and body weight (3,4) and in innovative treatment strategies (5). Furthermore, large genome-wide association studies (GWAS) identified hundreds of loci associated with BMI, some of these mapping within or near genes linked with monogenic obesity (e.g., LEPR, POMC, MC4R, BDNF, SH2B1, and PCSK1) (6–9).

In our large cohort of unrelated Pakistani children with severe obesity, named Severe Obesity in Pakistani Population (SOPP) – Children cohort, we previously demonstrated genetic causality in 49% of the case subjects (10). Importantly, almost half of these monogenic obesity diagnoses were due to pathogenic, homozygous mutations in the LEP (encoding leptin) gene. The other pathogenic mutations or copy number variants (CNVs) were found in LEPR, MC4R, ADCY3, or in Bardet-Biedel, Alström, and Prader-Willi syndrome regions (10). The exceptionally high prevalence of monogenic obesity in Pakistani children prompted us to assess the prevalence of monogenic obesity in a new cohort of severely obese young adults from Pakistan (SOPP – Young Adults).

The current study includes 128 young adults with severe obesity. The clinical data of these subjects and normal weight- and age-matched control subjects are reported in Supplementary Table 1. The selection criteria included a BMI ≥35 kg/m², early-onset obesity, and hyperphagia within the first 10 years of life. All of the participants were reported to have attained puberty in response to our questionnaire at the time of recruitment. The subjects mainly came from the urban population. The study protocol was approved by the institutional ethical committees. Participants were recruited to the study on a voluntary basis, and a signed written informed consent was obtained in each case. Anthropomorphic measurements and blood sampling were performed at presentation.

All probands were initially screened for mutations in the coding regions of the LEP and MC4R genes by Sanger sequencing using a 3730 DNA analyzer (Life Technologies). Sequence traces were analyzed using the Mutation Surveyor software package (Soft Genetics). The PCR primer sequences and thermocycler conditions are available elsewhere (11). Subsequently, DNA samples were sequenced using conventional whole-exome sequencing (WES via NimbleGen SeqCap EZ MedExome Enrichment) and an in-house developed augmented WES (CoDE-seq). We have previously described the detailed protocols for both technologies in detail (10,12). Target enrichment was performed according to the manufacturer’s protocol (NimbleGen SeqCap EZ) for Illumina sequencing. Briefly, 1 μg DNA was fragmented through sonication (Covaris E220 Focused-ultrasonicator). The fragmented DNA samples were end-repaired and ligated to the Illumina adapters using the KAPA HTP Library Preparation Kits, on the Hamilton Microlab STARlet automated liquid handling platform. The samples were subsequently amplified by PCR using primers complementary to the adapters. After size selection and sample quantification (PerkinElmer LabChip GX), five samples (StarSEQ) or eight samples (MedExome) were combined in a single pool of at least 1 μg and hybridized to the biotin-labeled SeqCap EZ probe pool. After 72 h at 47°C, the captures were purified using the SeqCap Hybridization and Wash Kit on the Agilent Bravo Automated Liquid Handling Platform. Captures were then amplified using the KAPA HiFi HotStart ReadyMix and quantified by both Caliper LifeScience LabChip GX and Qubit Fluorometric Quantitation assays (Thermo Fisher Scientific).

Sequencing was performed on the Illumina NovaSeq 6000 system. On average, a mean sequencing depth of 162× was reached for each individual using 150 base pair paired end reads. Bioinformatic analyses for detection and annotation of variants have previously been described in detail (12). Briefly, the demultiplexing of sequence data was performed using bcl2fastq Conversion Software (version 2.19.1; Illumina). The sequence reads from FASTQ files were mapped to the human genome (hg38/GRCh38) using Burrows-Wheeler Aligner (version 0.7.15) (13). The variant calling was performed using Genome Analysis ToolKit (version 3.7) (14). The annotation of point variants, including missense variants, nonsense variants, small insertions and deletions, and splice-site variants, was performed using the Ensemble Perl Application Program Interfaces (version 89) and other Perl scripts to include data from both dbSNP (version 135) and dbNSFP (version 3.4) databases (15,16). For CoDE-seq samples, the detection of CNVs was performed using the eXome Hidden Markov Model (version 1.0) program (17).

We focused on point mutations and CNVs with an overall minor allele frequency (MAF) <0.005 for homozygous variants and <0.001 for heterozygous variants in the Genome Aggregation Database (gnomAD; version 2.1.1). In the first step of the analyses, we considered point mutations and CNVs identified in known genes linked with monogenic obesity (Supplementary Table 2). Criteria of the American College of Medical Genetics and Genomics (ACMG) were used to determine the pathogenicity of the identified variants (Supplementary Tables 3 and Table 4). Subsequently, we extended our analysis to identify rare/novel point variants and CNVs in the gene/region associated with obesity. To achieve this, our analysis was focused on homozygous variants with overall MAF <0.005 in gnomAD version 2.1.1 or rare CNVs with an overall MAF <0.001 in gnomAD version 3.1. In case of nonsynonymous variants, we considered only those where at least one of the in silico software programs (i.e., MutationTaster, sorting intolerant from tolerant [SIFT]) displayed deleterious effect.

The subset of genes carrying the variants obtained after these steps were checked for association with obesity in humans and mice, by consulting the GWAS catalog, the International Mouse Phenotyping Consortium (https://www.mousephenotype.org), and the Mouse Genome Informatics (http://www.informatics.jax.org), an international database resource for the laboratory mouse. Moreover, we performed a detailed literature survey about each gene through PubMed.

Population structure was inferred by principal component analysis (PCA). The set of individuals from the publicly available database 1000 Genomes Project (1KG) in phase 3 (Genome Reference Consortium Human Build 38) was used as a reference (18). Data from the SOPP cohort were firstly thinned down using the following criteria: samples’ call rate >95%, variants’ call rate >90%, MAF >1%, Hardy-Weinberg equilibrium test P value >1 × 10⁻⁴. Then, SOPP – Young Adults (n = 128) and SOPP – Children (n = 225) were merged with 1KG data (n = 2,504). Palindromic variants and variants located in X and Y chromosomes were removed. The final data set comprised 2,857 samples and 121,660 single nucleotide polymorphisms (SNPs). Genotype data were processed using PLINK software (version 1.9) (19), PCA was performed using GCTA (Genome-wide Complex Trait Analysis) software (version 1.93.2 beta) (20).

For heterozygosity rate estimation, SOPP samples were retained if the call rate of the sample was >95%. We kept only those autosomal variants that were in high linkage disequilibrium regions (r² > 0.5 within a window of 50 kilobase [kb]). The SOPP samples were merged with four South Asian (SAS) populations from the 1KG project (i.e., ITU: Indian Telugu from the U.K.; GIH: Gujarati Indian from Houston, TX; PJL: Punjabi from Lahore, Pakistan; and STU: Sri Lankan Tamil from the U.K.), resulting in a total of 756 samples (353 from the SOPP cohort and 403 from the 1KG SAS cohort) and 158,885 SNPs in common. The heterozygosity rate was calculated as the ratio between the observed number of heterozygous genotypes and total number of nonmissing genotypes. Observed heterozygosity was obtained using the –het option in PLINK software. The heterozygosity rate between the two cohorts was compared using the Kruskal-Wallis test.

We analyzed 200,619 samples in UK Biobank, with available exome sequencing data using UK Biobank research application no. 67575. The participants were recruited by UK Biobank from across the U.K. between 13 March 2006 and 1 October 2010. Data from the Project Variant Call Format (pVCF, field no. 23156) was used. Variants with a coverage >10 reads and a genotype quality score >20 were retained for further analyses. Annotation of variants was performed using the Ensemble Variant Effect Predictor (VEP) tool version 103 (RefSeq). Subsequently, our analysis was focused on loss-of-function (LoF) variants that were genetically null (i.e., nonsense, frameshift, canonical ±1 or 2 splice sites, start-lost). These variants were also reported rare (with a MAF <0.01) in all ancestries included in the gnomAD. The effect of rare LoF variants per gene on the risk of obesity or overweight was analyzed using the mixed-effects score test (MiST) adjusted for age, sex, BMI, and ancestry. MiST provides a score statistic S(π) for the mean effect (π) of the cluster, and a score statistic S(τ) for the heterogeneous effect (τ) of the cluster. The overall P value is the combined P values P_π and P_τ from the Fisher procedure.

Serum concentrations of leptin, insulin, and cortisol were measured in duplicate using commercially available ELISA kits (Monobind, Lake Forest, CA). All assays were performed according to the manufacturer’s instructions. The intra- and interassay variations were <10%.

The data sets generated during the current study are available upon reasonable request. No applicable resources were generated during the current study.

This study included 128 severely obese young adults (BMI 37.2 ± 0.3 kg/m²; age 18.4 ± 0.3 years; 78 men and 50 women) from families residing in the Punjab province, Pakistan (Supplementary Table 1 and Supplementary Fig. 1). Among the probands, 44% reported a family history of obesity and 30% of diabetes. All subjects were reported hyperphagic and to have attained puberty at the time of recruitment.

We identified 4 pathogenic or likely pathogenic (P/LP) variants and 9 variants of uncertain significance (VUS) in 13 subjects in the genes known to cause monogenic obesity (Table 1). The P/LP mutations included homozygous, start-loss, and missense variation in LEPR (c.2T>C) and PCSK1 (p.Asn127Ile), respectively, and two heterozygous missense mutations in MC4R (p.Ile316Ser and p.Val166Ile) (Table 1). The homozygous start-loss mutation in LEPR gene was identified in an 18-year-old man with a BMI of 39 kg/m². Hyperleptinemia was noticeable in this subject compared with reference values in healthy subjects with normal body weight (Supplementary Table 5). The proband had normal developmental milestones and attained puberty at the normal age. The PCSK1 p.Asn127Ile mutation was identified in a 24-year-old man with a BMI of 41.5 kg/m². This proband started gaining weight and becoming hyperphagic at the age of 8 years. He was reported to present frequent episodes of constipation at an early age (2–3 years). His 30-year-old female sibling with a BMI of 35 kg/m² carried the same homozygous mutation. The parents and other three siblings, who were not obese (BMI 25–27 kg/m²), were heterozygous for the same PCSK1 p.Asn127Ile mutation. The heterozygous pathogenic p.Ile316Ser and p.Val166Ile mutations in the MC4R gene were identified in two 16- and 15-year-old probands, respectively. Whereas hyperleptinemia and hypercortisolemia were observed in both individuals, insulin levels were variable (17 and 27 uIU/mL) (Table 1).

All other mutations identified in PCSK1, POMC, SIM1, NTRK2, and SH2B1 were heterozygous missense VUS. These mutations were located in PCSK1 (p.Pro663Leu and p.Thr377Met), POMC (p.Pro149Thr and p.Leu253Gln), SIM1 (p.Met164Leu and p.Gly616Arg), NTRK2 (p.Thr411Arg and p.Val486Ile), and SH2B1 (p.Pro643Ala) (Table 1). Whereas the age of onset of obesity was reported soon after birth or at a very early stage of life in probands with VUS in NTRK2 or SIM1, increase in adiposity was noticed between 8 and 10 years of age in most cases.

These data, surprisingly, showed a much lower prevalence of P/LP mutations when compared with children from the same geographical region (3% vs. 49%) (10). We carried out population structure analysis of the present SOPP – Young Adults cohorts and compared it with SOPP – Children, and with the SAS population from the 1KG Project.

The PCA revealed that both SOPP – Young Adults and SOPP – Children are close to the SAS population from 1KG project (Fig. 1A), especially four SAS populations (Fig. 1B). We did not find any distinct clustering of ethnicity-related features among these three groups (Fig. 1). As expected, the heterozygosity rate was significantly lower in both children (0.200 ± 0.0197) and young adults (0.210 ± 0.0106) of the SOPP cohorts compared with the SAS population from the 1KG project (0.232 ± 0.00703) (Fig. 2). However, the average heterozygosity rate of SOPP – Children was slightly lower than the rate of SOPP – Young Adults, suggesting that young adults were less inbred than the children (P = 4.04 × 10⁻⁵).

Next, we analyzed rare homozygous mutations in genes where the knock-out in mice leads to obesity-related phenotypes. We identified a homozygous nonsense mutation (p.Arg423*) in ASNSD1 (encoding the asparagine synthetase domain containing 1) in a 17-year-old severely obese proband (BMI 37 kg/m²) (Table 2). His 22-year-old female sibling with a BMI of 33 kg/m² was also found to carry the same homozygous mutation, whereas his overweight mother (BMI 27 kg/m²) was heterozygous for the nonsense mutation. Both siblings were reported to suffer from muscle weakness, fatigue, and pain in joints and muscles. Furthermore, polycystic ovary syndrome was diagnosed in the female sibling. Recently, Powell and colleagues (21) showed that inactivation of the ASNSD1 results in obesity and muscle atrophy in the mouse model. None of the individuals from gnomAD (version 2.1.1) was reported to carry a homozygous LoF variant in ASNSD1 (i.e., frameshift, splice donor/acceptor or stop-gain, start-loss). Another 17-year-old female proband (BMI 37 kg/m²) was identified with a homozygous nonsense mutation (p.Ser532*) in IFI16 (encoding interferon-γ inducible protein 16) (Table 2). The proband had hyperleptinemia (67 ng/mL) and hypercortisolemia (32 μg/dL). The encoded protein has a predominant expression in the adipose tissue with its role in promoting white adipogenesis and fat storage in the mouse (22,23). This gene has also been linked to obesity-related traits (energy intake - carbohydrate diet, as total grams per day) through GWAS (24). Of note, only one individual from gnomAD (version 2.1.1) was reported to carry a homozygous LoF variant (p.Gln486*) in IFI16.

Subsequently, we analyzed homozygous rare variants in genes that have shown recurrent association with obesity-associated traits through GWAS. This approach resulted in the identification of nine homozygous variants, including seven missense, two nonsense (one of which is IFI16 reported above), and one stop-loss mutation (Table 2). None of these variants have been reported in the homozygous state in gnomAD (version 2.1.1).

The homozygous stop-gain mutation (p.Arg298*) in the ABTB1 gene encoding a protein with an ankyrin repeat region and two BTB/POZ domains was identified in a 20-year-old woman with severe obesity (BMI 35.4 kg/m²). A high waist-to-hip ratio (WHR) of 0.8, elevated insulin levels (31 uIU/mL), and a tendency toward insulin resistance (HOMA-insulin resistance 3.88) were observed in the proband. The leptin levels were relatively low (1.4 ng/mL). A history of hyperphagia, hyperthyroidism, and a delay in menarche were also reported. Both parents of the proband suffered from diabetes, and a history of hypertension was recorded on the maternal side. The ABTB1 gene has been reported to associate with higher WHR adjusted for BMI (25). In the gnomAD database, a homozygous LoF mutation (p.Arg306*) has been reported in three individuals from the European (non-Finnish) and one individual from the SAS population.

The other mutation of interest identified in this study is a novel homozygous stop-lost mutation (p.*566Cysext*8) in the WSCD2 gene. This mutation was identified in a 19-year-old male proband with a BMI of 38 kg/m² from a first-degree parentage. This proband started gaining weight at the age of 9 years and reported a history of depression, mood swings, and aggressive behavior. One of his brothers was also reported to be obese (data not available). Several GWAS reports have identified the association of variants in this gene associated with body fat percentage (26) and WHR (27,28). Moreover, variants in this gene have also shown recurrent association with type 2 diabetes (adjusted for BMI) (29,30) and anxiety (31).

Finally, we analyzed rare CNVs located in the genes that have shown recurrent association with obesity traits in GWAS or where the knock-out in mice has led to the obesity-related phenotypes. We found four copy-loss CNVs (in heterozygous state) in the genomic regions that have previously been associated with obesity-related traits (BMI, body weight, body fat mass, trunk fat mass) through GWAS (Table 3). Three of these CNVs were novel: 1) a 839-kb deletion included coding exons of CPVL, PRR15, and CHN2 genes, 2) a 59-kb deletion, including an intronic region of DLGAP1, and 3) a 199-kb deletion that included at least one coding exon of RBFOX1. The other 96-kb deletion encompassed at least one coding exon of GRID2 and was reported in gnomAD with an allele frequency of 0.000046 (Table 3). The phenotypic characteristics and hormone profiles of probands carrying variants in genes associated with obesity are also summarized in Supplementary Table 5.

We next assessed genetic associations between LoF variants in our subset of genes of interest and adiposity by using 200,000 exome data from UK Biobank (Table 4). All of the LoF variants that we detected were in the heterozygous state in the UK Biobank population. We found that LoF variants in ABTB1 were significantly associated with higher odds of overweight (P_π = 0.003, with an odds ratio [OR] of 1.14; 95% CI 0.044–0.215; P_τ = 0.008) and obesity (P_π = 0.015, with an OR of 1.14; 95% CI, 0.025–0.234; P_τ = 0.012). MiST here gave rise to a significant P value P_τ due to a significant effect of the cluster of LoF variants on overweight and obesity. Further analysis at variant level focusing on the most frequent variant in ABTB1 (p.Arg306*), unravels that this variant was responsible for driving all the association for the cluster (Table 4).

We also found a significant effect of LoF variants in USP6 on overweight (P_π = 0.021 with an OR of 1.124; 95% CI, 0.018–0.218). Of note, LoF variants in AMZ1 (P_π = 0.0021), DNAJC27 (P_π = 0.014), and IFI16 (P_π = 0.044) were significantly associated with lower risk of obesity with an OR of 0.80 (95% CI, −0.360 to 0.078), OR of 0.48 (95% CI, −1.360 to 0.160), and OR of 0.54 (95% CI, −1.262 to 0.027), respectively. LoF variants in the rest of the six genes did not reach statistically significant levels, although showing an effect on increase in adiposity (Table 4).

Exome sequencing is currently a widely used approach to identify causative variants, especially in monogenic (“Mendelian”) diseases. However, identifying specific causative variants remains challenging and even more so in case of heterogeneous genetic disorders such as obesity. Here, the genetic analysis on subjects with extreme phenotype from a single geographical region and, more importantly, from a consanguineous population resulted in identification of rare and highly penetrant protein coding variants (1,8,32). This study demonstrates the potential of such cohorts in bringing to the fore novel or extremely rare (MAF < 0.001) potentially causative genetic variants underlying human adiposity.

Only sparse information is available relating to genetic etiology of severe obesity in the South Asian region. Here, we have analyzed 128 severely obese young adults from a consanguineous population of Pakistan in an attempt to complement our previous study in Pakistani children.

As a first step, we assessed potentially causing mutations in 24 genes already known to be linked with monogenic obesity and identified 13 rare variants in 7 of these genes. Whereas, 11 of these point mutations were predicted deleterious by SIFT, only 4 probands were identified with P/LP mutations in LEPR, PCSK1, and MC4R genes according to ACMG criteria. Previously, in our relatively large cohort of severely obese children (n = 225) from this population, homozygous pathogenic mutations in LEP and LEPR accounted for 47 and 15% of elucidated cases, respectively (10). In the present cohort of young adults, none of the probands was identified with LEP deficiency, and cases of LEPR deficiency were also underrepresented. This unexpected discrepancy may possibly be explained on account of various factors, including a possible high early mortality rate and/or by severe disabilities of these children, thus preventing them to enter the mainstream of the young adult population. In a previous genetic investigation on rather small cohort of 25 severely obese subjects from the same geographical region, the authors reported only two homozygous mutations in the LEPR gene and the absence of mutations in LEP and MC4R (33). Interestingly, both probands with LEPR deficiency, aged 1 and 12 years, were reported to have diabetes.

In the present investigation, we also identified two heterozygous P/LP mutations in MC4R in two unrelated young adults. In Western populations, MC4R deficiency following a codominant inheritance is the most prevalent form of monogenic obesity that accounts for 5–7% of severely obese subjects (34). In the SOPP – Children, heterozygous siblings and parents of probands with homozygous MC4R mutation did not present obesity (10,11,35). The discrepancy may not be so surprising, as penetrance of MC4R mutations in other populations has also been reported to be very low or variable in heterozygous carriers (34). In a Greek population, the estimated penetrance of MC4R mutations was reported to be very low in heterozygous carriers, and in an inbred Bedouin Israeli population, subjects with homozygous MC4R mutations expressed early-onset severe obesity, whereas heterozygous mutation carriers were either moderately overweight or had a normal phenotype (36,37). In SOPP – Young Adults, the observed penetrance of heterozygous MC4R deficiency in the two subjects could be linked to a differential exposure to a more intense obesogenic environment due to a higher socioeconomic status of the affected families. However, we failed to find appreciable differences in ancestry and consanguinity between SOPP – Children and – Young Adults (Figs. 1 and 2). Alternatively, the possibility that as yet an unidentified genetic factor or factors associated with an exacerbation of adiposity in these probands with MC4R haploinsufficiency remains. Progress in obesity genetics has been greatly aided by focusing on mouse models of obesity and GWAS analyses (8,38). In the current study, ASNSD1 knockout mice that show obesity and muscle weakness (21) prompted us to consider ASNSD1 with a homozygous stop gain mutation (p.Arg423*) as a putative candidate gene. Thus, our novel homozygous null mutation in the ASNSD1 suggests for the first-time a causative role of ASNSD1 in human obesity and sarcopenia. Additionally, a homozygous nonsense mutation p.(Ser532X) was identified in the interferon-γ inducible protein 16 (IFI16) that has been suggested to promote adipogenesis in mice and human (15,16). Furthermore, IFI16 has also been reported to be a susceptible gene for obesity via GWAS (17).

We also identified rare and/or novel variants in genes associated with obesity in GWAS. We found homozygous point mutations (seven missenses, one nonsense, and one stop-loss) (Table 2) in nine genes and in five genes located in the copy-loss CNV regions (Table 3). To further establish the causality of these genes, we performed association studies to examine whether these genes affect body weight and BMI by referring to the UK Biobank record. We found a significant association between LoF variants in two genes, ABTB1 and USP6, and a higher risk of obesity. Here it is worth mentioning that two other genes, AMZI and DNAJC27, where probands were found to carry missense mutations, and IFI16, where a proband had identified with stop-gain mutation, have shown a negative correlation with obesity. Nevertheless, a careful analysis of LoF variants in these genes associated with obesity and higher BMI identified in our study, in relation to UK Biobank data, further suggests that these variants significantly contribute to the obesity burden.

Furthermore, a number of these genes have established connections with obesity. For example, DnaJ heat shock protein family (Hsp40), a member of C27 (DNAJC27) is known to encode a protein associated with increased adiposity in several global GWAS (39). Another gene, DLG associated protein 1 (DLGAP1), has been shown to negatively regulate the browning of white adipocytes (40). For other rare/novel homozygous mutations and CNVs, we found a lack of evidence of their role in energy homeostasis, but five of these genes (WSCD2, RBFOX, GRID2, CHN2, and DLGAP) are predominantly expressed in the brain (Genotype-Tissue Expression), which is an established key region in the regulation of energy homeostasis and food intake.

In conclusion, the present investigation identified four rare pathogenic or likely pathogenic variants in established monogenic obesity genes in a cohort of young adults from a consanguineous population. We also found 11 rare, potentially deleterious point mutations and four CNVs in genes associated with obesity through GWAS or rodent models. Further functional analyses are needed to determine the pathogenicity of these newly identified genes and variants potentially involved in human obesity.

A focus on monogenic obesity and its genetic classification has already led to the discovery of certain pharmacologic targets for development of precision medicine and shall continue to do so in future. In addition to recombinant leptin tailored for the treatment of leptin deficiency, the newly developed drug, setmelanotide, an α-melanocyte-stimulating hormone agonist, has been approved for the treatment of obesity due to LEPR, POMC, and PCSK1 deficiencies and is likely to be efficacious in some other forms of obesity such as congenital leptin deficiency and possibly in some forms of syndromic and common obesity (5,41).

The present study has some limitations. Since we did not have access to DNA samples from the family members, except for two families, the segregation of the mutations with the disease could not be assessed. Moreover, most mutations were singletons. Therefore, the causality of genetic variants associated with obesity in these families is still questionable. This urges the importance of analyzing other consanguineous populations for the presence of putative pathogenic variants in these genes.

Acknowledgments. The authors thank the patients and their families for participation in the study. The authors are thankful to Frédéric Allegaert and Nicolas Larcher (Inserm UMR 1283, CNRS UMR 8199, European Genomic Institute for Diabetes [EGID], Institut Pasteur de Lille, Lille, France) for DNA extraction and storage.

Funding. This work was supported by funding from the Medical Research Council (MRC) MR/S026193/1 (P.F.) and the Pakistan Academy of Sciences (MA-03). The work related to association analysis was conducted using the UK Biobank Application no. 67575. The work was also funded by grants from the French National Research Agency (ANR-10-LABX-46 to the European Genomics Institute for Diabetes and ANR-10-EQPX-07-01 to Lille Integrated Genomics Network for Advanced Personnalized Medicine [LIGAN-PM]), from the European Research Council (ERC GEPIDIAB – 294785 to P.F and ERC Reg-Seq 7155575 to AB), from the National Center for Precision Diabetic Medicine – PreciDIAB, which is jointly supported by the French National Agency for Research (ANR-18-IBHU-0001), by the European Union (FEDER), by the Hauts-de-France Regional Council, and by the European Metropolis of Lille (MEL).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. S.S., Q.M.J., A.H., R.K., and M.A. collected samples and performed biochemical analysis. S.S., E.D., E.V., A.Ba., L.B., S.A., M.D., and A.Bo. performed whole-exome sequencing and analyzed the genetic data. SS, A.Bo., M.A., and P.F. designed the study and wrote the first draft of the paper. L.N., M.B., M.C., and A.Bo. performed statistical analysis. All authors contributed to the final version of the manuscript. S.S. and P.F. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in abstract form at the European Human Genetics Conference, Virtual Conference, 28–31 August 2021.