Genetic basis of Asian diabetes

ABSTRACT

Recent genomic studies in Asian populations have advanced our understanding of the diagnosis, pathophysiology, risk prediction, and precision care of diabetes. Monogenic diabetes, including maturity‐onset diabetes of the young (MODY) and mitochondrial diabetes, highlights the importance of genetic diagnosis for subtype‐specific care. For type 2 diabetes (T2D), genome‐wide association studies (GWAS) using large‐scale biobank resources have identified numerous risk loci across the genome. These studies have emphasized characteristic features of East Asian T2D, including onset at lower BMI and relatively greater contributions from beta‐cell dysfunction pathway. Genetic studies of gestational diabetes mellitus (GDM) have revealed both T2D‐related and predominantly gestational components. For type 1 diabetes (T1D), population‐specific HLA and non‐HLA immune‐related loci and possible links to infectious exposures may partly explain global variation in incidence, including lower rates in Asian populations. Polygenic risk scores (PRS) can support risk stratification for diabetes onset and complications, but their implementation in Asian populations is limited by reduced cross‐population transferability. Ancestry‐matched and trans‐ancestry approaches are therefore needed to improve prediction and reduce disparities. Subtype‐aware frameworks, including clinical clustering, pathway‐specific PRS, and BMI‐stratified PRS, may further improve mechanistic interpretation and prediction. Expanding genomic resources across East, Southeast, and South Asia and promoting cross‐biobank studies will be essential for improving diabetes genomic research, PRS validation, and equitable precision diabetes care.

INTRODUCTION

The onset of diabetes is shaped by both inherited genetic factors and environmental exposures. ¹ , ² The proportion of influence from environmental factors versus genetic factors in diabetes differs across diabetes groups. Diseases caused by genetic factors are generally categorized into two distinct disease concepts: monogenic diseases and polygenic diseases, also known as complex diseases or common diseases ³ (Figure 1). Monogenic diseases are caused by strong pathogenic variants in a single gene and are inherited in a Mendelian manner, with maturity‐onset diabetes of the young (MODY) ⁴ , ⁵ and mitochondrial diabetes ⁶ , ⁷ serving as canonical examples in diabetes. By contrast, type 2 diabetes (T2D) is a prototypical common disease influenced by numerous common variants with modest effect sizes, which can be identified by genome‐wide association studies (GWAS). ⁸ , ⁹ Gestational diabetes mellitus (GDM)‐associated loci could be classified into two categories: T2D risk and predominantly gestational. ¹⁰ Even for type 1 diabetes (T1D), an autoimmune disorder, GWAS have uncovered robust associations with immune‐related loci. ¹¹ , ¹² Furthermore, polygenic risk scores (PRS), which are calculated from GWAS results, have attracted attention for their potential to predict diabetes onset and prevent diabetes‐related complications. ¹³ , ¹⁴ , ¹⁵ East Asian populations are known to develop T2D at lower BMIs compared to European populations, suggesting underlying genetic differences in beta cell function, body fat distribution, and hepatic lipid metabolism. ¹⁶ For example, missense variants in genes related to pancreatic acinar cells (GP2) and insulin secretion (GLP1R) have shown higher allele frequencies in the Japanese population than in European populations. ¹⁷ In addition, pathway‐based analyses have shown that lipodystrophy‐related polygenic risk is enriched in East Asian populations and partly explains the lower BMI threshold associated with equivalent T2D risk compared with European populations. ¹⁸ Recent advances in human genetics are increasingly explaining the genetic basis of Asian diabetes.

Figure 1.

Genetic architecture of major diabetes subtypes. Schematic illustration of how major diabetes subtypes can be positioned according to the number of risk variants (x axis) and the effect size per risk variant (y axis). Monogenic diabetes is driven by a single pathogenic variant and includes maturity‐onset diabetes of the young (MODY) and maternally inherited diabetes and deafness (MIDD). Type 1 diabetes (T1D) has a prominent contribution from human leukocyte antigen (HLA) variation alongside additional non‐HLA loci. Type 2 diabetes (T2D) and gestational diabetes mellitus (GDM) are prototypical polygenic diseases influenced by numerous variants of modest effect.

In this review, we summarize recent advances in the genetics of diabetes in Asian populations, primarily focusing on East Asian populations where genomic cohort studies are most abundant, while also discussing findings from other Asian populations where relevant. We also discuss their implications for subtype‐aware and cross‐populational polygenic prediction and precision care.

MONOGENIC DIABETES

MODY and mitochondrial diabetes are canonical examples of monogenic diabetes. Monogenic diabetes accounts for only a small proportion of all diabetes cases, and accurate diagnosis requires confirmatory genetic testing. Thus, individuals with monogenic diabetes are frequently misdiagnosed as type 1 or type 2 diabetes. ⁵ Such misclassification can delay appropriate subtype‐specific treatment.

MODY is a unique subset of diabetes caused by a single‐gene defect and characterized by distinct phenotypes, such as early onset, lean body mass, and insulin independence. ⁴ In a survey of 231 Japanese probands with diabetes onset before age 35, BMI < 30 kg/m², and negative islet autoantibodies, 6 had HNF4A variants (MODY1), 25 had GCK variants (MODY2), 22 had HNF1A variants (MODY3), 7 had HNF1B variants (MODY5), 7 had ABCC8 variants (MODY12), and 1 each had a PDX1 variant (MODY4), NEUROD1 variant (MODY6), and INSR variant. ¹⁹ To minimize missed diagnoses of MODY, it is important to consider tandem repeats and copy number variations (CNVs) in diagnostic testing, in addition to single‐nucleotide variants (SNVs) and short insertions and deletions (indels). ²⁰ , ²¹ , ²² , ²³ , ²⁴

A family‐based study of diabetes with hearing loss demonstrated maternal transmission consistent with mitochondrial DNA involvement, ⁷ and investigations of Japanese families helped establish the disease concept of mitochondrial diabetes. ⁶ Mitochondrial diabetes is often referred to as Maternally Inherited Diabetes and Deafness (MIDD). ²⁵ The most common clinical manifestations of mitochondrial diabetes include diabetes mellitus, deafness, ophthalmic disease, cardiac disease, renal disease, gastrointestinal disease, short stature, and myopathies. ²⁵

Even in monogenic diabetes, treatments can be tailored to the genetic subtype to deliver precision care, achieving better outcomes than standard care. ³ , ²⁶ Therefore, it is important to suspect monogenic diabetes based on its clinical features and the pattern of inheritance within a family, and to guide appropriate genetic testing accordingly.

GWAS HAVE ELUCIDATED GENETIC BASIS OF T2D

Individuals with type 2 diabetes often have family histories, and it has been thought that differences in susceptibility to developing the disease arise from differences in genetic background. ¹ , ²⁷ However, the effect size of each risk variant is small, and the causal genetic variants remained largely unknown until the 21st century; indeed, there was even a time when the genetics of T2D was described as a ‘nightmare for geneticists’. ²

Following the determination of the human reference genome in the 2000s, genetic variants, such as single‐nucleotide polymorphisms (SNPs) and short indels, could be catalogued comprehensively. The genome‐wide association study (GWAS) is an approach that comprehensively examines the association between common variants and diseases across the entire genome. ⁸ GWAS of T2D in East Asian populations has been led primarily by BioBank Japan (BBJ). ¹⁷ , ²⁸ , ²⁹ , ³⁰ , ³¹ Several variants associated with T2D in BBJ were not found in European GWAS and exhibited distinct minor allele frequency (MAF) spectra between populations, ¹⁷ highlighting the importance of establishing large biobanks in East Asian populations.

A large‐scale global GWAS meta‐analysis of 2,535,601 individuals, including 428,452 T2D cases, identified 1,289 independent association signals in 611 loci. ⁹ These loci provide a framework for post‐GWAS analyses, including fine‐mapping, PRS analyses, and pathway‐level interpretation. ⁸ The study showed the value of integrating multi‐ancestry GWASs and cardiometabolic traits with single‐cell epigenomics across diverse tissues to disentangle the etiological heterogeneity driving the development and progression of T2D. ⁹

PRS PREDICT T2D BY AGGREGATING VARIANTS IDENTIFIED IN GWAS

As demonstrated by large‐scale GWAS, common diseases such as T2D are driven by the combined effects of thousands of genetic variants across the genome. ¹³ , ³² Although most individual variants identified by GWAS show modest effect sizes for T2D, the PRS approach, which aggregates the risk variant effects according to an individual's genomic profile, has emerged as an effective approach for risk prediction ¹⁴ , ¹⁵ (Figure 2a). When pathway information for each variant is available, PRS can also be partitioned into pathway‐specific components, providing a framework for interpreting the mechanistic heterogeneity of T2D. These profiles may help characterize East Asian T2D, ¹³ , ³³ in which beta‐cell dysfunction and lipodystrophy pathways show greater contributions than the obesity pathway (Figure 2b). As GWAS sample sizes increase, polygenic scores are likely to play a central role in the future of biomedical research and personalized medicine. ¹⁴

Figure 2.

PRS calculation and pathway‐specific PRS. (a) The first step is to obtain genome‐wide association study (GWAS) summary statistics, describing the effect size of each single‐nucleotide polymorphism (SNP) on diabetes, from the discovery dataset. The second step is to reference the genotype data of each individual in the target population against the GWAS results. Here, genotype data for eight SNPs are shown for four target individuals. The third step is to calculate a polygenic risk score (PRS) for each target individual by summing up the effect sizes of risk alleles for each individual. When pathway information for each SNP is available, pathway‐specific PRS (psPRS) can also be calculated. (b) Conceptual illustration of pathway‐specific PRS burden in East Asian T2D. In this schematic, prediction performance is shown as lowest for European‐derived PRS, higher for East Asian ancestry‐matched PRS, and highest for trans‐ancestry PRS. The designs of the figures were inspired by and followed after the referenced studies. ⁸ , ¹³ , ⁴⁸ , ⁴⁹

Beyond prediction, communication of personal disease risk data including PRS motivates positive changes in health behavior and the propensity to seek care. ³⁴ A prospective study of PRS screening detected individuals eligible for treatment missed by traditional clinical testing with limited overdiagnosis. ³⁵ At the population level, Estonia has begun returning genome‐based disease risk information for 20% of the adult population, advancing the national‐scale evaluation of preventive effects. ³⁶

T2D might be particularly well suited for precision prevention and treatment guided by PRS for the following reasons. T2D is a representative disease for which PRS can identify high‐risk groups, ¹⁵ and moreover, prevention methods through lifestyle guidance and treatment medications are widely available. ³⁷ A recent study used causal forest models to estimate individualized treatment effects (ITEs) of modifiable exposures on the outcome, enabling the identification of potentially effective interventions for individuals with high PRS. ³⁸ Furthermore, PRS for T2D‐related complications had independent and incremental prognostic value compared with traditional clinical risk assessment in type 2 diabetes. ³⁹ A recent study further showed that a CAD PRS improved atherosclerotic cardiovascular disease (ASCVD) risk prediction in individuals with T2D, particularly those at borderline or intermediate 10‐year risk. ⁴⁰

HETEROGENEITY IN T2D AND INTERNATIONAL DISPARITIES IN PRS PREDICTION

Among individuals diagnosed with diabetes based on hyperglycemia or elevated HbA1c, more than 90% are classified as T2D when they lack clinical features suggestive of other subtypes, such as T1D. Clinically indistinguishable T2D is increasingly recognized to comprise heterogeneous subtypes with different BMI sensitivity. ⁴¹ , ⁴² , ⁴³ Consistent with this heterogeneity, PRS performance differs across T2D subtypes. ³¹ Nevertheless, PRS prediction has often been conducted by treating T2D as a single disease without considering subtype heterogeneity, which may have restricted predictive performance. ¹³

Cross‐population transferability is another major challenge for implementing PRS in Asian populations (Figure 3). PRSs derived from European GWAS are known to perform poorly when applied to T2D prediction in East Asian target populations. ⁴⁴ , ⁴⁵ Reduced portability reflects differences in allele frequency, linkage disequilibrium, and phenotype definitions, as well as variation in subtype composition across populations. ¹³ , ⁴⁴ , ⁴⁶ , ⁴⁷ In addition to the issue of limited cross‐population transferability, GWAS sample sizes are relatively small in non‐European populations, such as the East Asian population. There are concerns that this problem could widen health disparities. ⁴⁴

Figure 3.

Cross‐population PRS transferability. (a) Conceptual framework for applying ancestry‐matched, different ancestral groups, and trans‐ancestry PRSs to East Asian and European target populations. This schematic is modified from an original figure created by the present authors in (Ojima et al. Nat Genet, 2024). ¹³ (b) Conceptual illustration of prediction performance in East Asian target populations.

Therefore, improving strategies to address both T2D heterogeneity and cross‐population transferability is essential for the clinical implementation of PRS.

APPROACHES FOR CLASSIFYING HETEROGENEOUS T2D

To address the heterogeneity of T2D, data‐driven strategies to subclassify T2D based on clinical features have been developed. ⁴¹ These approaches can be broadly categorized into methods for classifying individuals with diabetes into subtypes ⁴² and methods for classifying variants identified by GWAS into pathways. ⁹ , ⁴³ , ⁴⁸ Furthermore, the palette model captures T2D as the sum of variants classified by pathway, representing the aggregate of multiple pathways within an individual. ⁴⁹ , ⁵⁰

A representative example for classifying individuals is Ahlqvist's classification, which is the most widely replicated method. ⁴² In this approach, individuals with diabetes are clustered into 5 subtypes using k‐means and hierarchical clustering based on six clinical variables: GAD65 autoantibodies, age at diabetes diagnosis, BMI, HbA1c at diagnosis, and homeostatic model assessment estimates of insulin secretion capacity (HOMA2‐B) and insulin resistance (HOMA2‐IR). The clusters were named after their most defining trait.

The subtypes were named after their most defining trait. ⁴¹ , ⁴² Severe autoimmune diabetes (SAID) is defined by GAD65 positivity and includes individuals with type 1 diabetes and latent autoimmune diabetes in adults (LADA). Severe insulin‐deficient diabetes (SIDD) is characterized by insulin deficiency and poor glycemic control and has been linked to a higher risk of early microvascular complications, particularly retinopathy and neuropathy. ⁵¹ Severe insulin‐resistant diabetes (SIRD) is characterized by obesity, severe insulin resistance, high insulin secretion and late onset, but relatively low HbA1c, and shows a higher risk of diabetic kidney disease and non‐alcoholic fatty liver disease. ⁵¹ Mild obesity‐related diabetes (MOD) is characterized by obesity and earlier onset with more modest metabolic derangement. Mild age‐related diabetes (MARD) is characterized primarily by older age at onset and comparatively milder glycemic disturbance. They further demonstrated that PRS prediction accuracy differed across subtypes, showing higher accuracy particularly in SIDD, MOD, and MARD. ³¹

The subtype classification proposed by Ahlqvist based on the ANDIS cohort in Sweden has been replicated in Asian countries such as China, ⁵² Japan, ⁵³ and India. ⁵⁴ , ⁵⁵ The Chinese cohort showed a larger proportion of SIDD individuals as well as generally lower BMI and earlier diabetes onset. ⁴¹ , ⁵²

PATHWAY CLUSTERING APPROACHES FOR T2D‐ASSOCIATED VARIANTS

Representative pathway‐specific PRS (psPRS) approaches that classify variants associated with T2D into pathogenic pathways are Udler's cluster ¹⁸ , ⁴³ and Suzuki's cluster. ⁹ A key advantage of pathway‐based variant clustering is that it offers mechanistic interpretability beyond a single global PRS, potentially aligning genetic risk with targetable biology.

Udler classified 94 variants previously shown to be associated with T2D into five pathogenic pathway clusters using Bayesian non‐negative matrix factorization (bNMF). ⁴³ With the expansion of available datasets, more recent reports have shown that this framework can be extended to yield 12 clusters. ¹⁸ In British Pakistani and British Bangladeshi cohorts, partitioned polygenic scores (pPSs) reflecting insulin deficiency and lipodystrophy showed the strongest associations with T2D risk and earlier age at diagnosis. ⁵⁶ In this study, the Beta Cell 1, Beta Cell 2, and Proinsulin clusters were associated with decreased HOMA‐B, suggesting a primary mechanism of insulin deficiency. The Obesity, Lipodystrophy 1, and Lipodystrophy 2 clusters were associated with increased HOMA‐IR, suggesting a primary mechanism of insulin resistance. Liver‐Lipid and ALP Negative clusters were associated with both increased HOMA‐IR, fasting insulin adjusted for BMI, and decreased HOMA‐B. The Hyper Insulin and Cholesterol clusters were not significantly associated with HOMA‐B or HOMA‐IR, but were associated with other variables suggestive of a mechanism of insulin resistance (fasting insulin, insulin sensitivity index adjusted for BMI). Bilirubin and SHBG‐LpA clusters were not clearly associated with insulin deficiency or resistance.

In the largest GWAS report to date, ⁹ Suzuki assigned T2D‐associated variants to eight pathway clusters based on a majority rule across 27 indices. Five of these clusters corresponded to the pathways reported previously, ⁴³ representing beta cell dysfunction with a positive or negative association with proinsulin (PI) and insulin resistance mediated through obesity, lipodystrophy, and liver and lipid metabolism. Body fat cluster was associated with increased abdominal subcutaneous adipose tissue volume, visceral adipose tissue (VAT) volume, and body fat percentage, but not strongly associated with BMI, lipid levels, or basal metabolic rate. Metabolic syndrome cluster was associated with increased fasting glucose, WHR, triglycerides, and blood pressure, and with decreased HDL cholesterol. Residual glycemic cluster was most strongly associated with increased fasting glucose and glycated hemoglobin, but, unlike the two beta cell dysfunction clusters, was not associated with PI or decreased fasting insulin.

In an independent Hong Kong Chinese cohort, Suzuki's cluster pathway‐specific PRS (psPRS) replicated a consistent clinical profile of T2D, suggesting Suzuki's cluster psPRS could guide precision diabetes care. ³³

BMI‐STRATIFIED PRS FOR T2D

T2D heterogeneity limits PRS performance. Ideally, polygenic prediction should be performed separately for T2D subtypes. However, existing large cohorts lack detailed information to classify T2D subtypes; therefore, it is important to evaluate the utility of subtype‐stratified PRSs based on existing traits. ¹³

BMI can be measured easily and is available in most biobanks. Furthermore, the genetic basis of T2D and BMI is complexly interrelated. While obesity and T2D generally show positive genetic correlation, some of the individual risk variants exhibit opposite effects on BMI and T2D. ⁵⁷ Indeed, the Ahlqvist clustering framework identifies subtypes with differing BMI sensitivity, ⁴² suggesting that sample stratification based on BMI may serve as a pragmatic proxy to separate subtypes when detailed information is unavailable.

In large‐scale biobank analyses, stratifying individuals into low BMI and high BMI groups showed improved PRS performance for T2D in low BMI targets ¹³ (Figure 4). PRSs constructed from low‐BMI discovery also outperformed those from high BMI in predicting T2D, but were inferior to PRSs from BMI‐unstratified discovery primarily due to the greater impact of reduced sample size. In this study, the pathogenic pathway corresponding to beta cell function showed the most substantial contribution to improved T2D prediction for low‐BMI targets. ¹³ , ⁴³ Low‐BMI T2D cases showed higher rates of neuropathy and retinopathy. ¹³ Thus, controlling blood glucose levels for low‐BMI T2D cases is essential to prevent complications such as neuropathy and retinopathy, which have higher rates in the low‐BMI group. ⁵⁸ , ⁵⁹ In addition, combining BMI stratification and a method integrating cross‐population effects, T2D predictions showed further improvements. ¹³ , ³²

Figure 4.

Conceptual framework for BMI‐stratified PRS model. In the BMI‐stratified setting, the target dataset is stratified into high or low BMI groups. PRS prediction performance is improved in the low BMI target group. When the discovery GWAS is also stratified by BMI, PRSs derived from low BMI GWAS can show higher predictive performance, but lower performance than PRSs derived from BMI‐unstratified GWAS due to reduced sample sizes. This schematic is modified from an original figure created by the present authors in (Ojima et al. Nat Genet, 2024). ¹³

In the Chinese cohort in Hong Kong, normal‐weight individuals were burdened with higher genetic risks of beta cell dysfunction and lipodystrophy than overweight individuals. ³³ These results suggest that the relationship between BMI and pathway PRS for beta cell function is replicable in East Asian populations.

In addition, building on the BMI‐stratified PRS approach, an extended framework has been proposed in which T2D PRS prediction is stratified jointly by BMI and blood pressure. ⁶⁰ In that study, PRS performance was highest in the subgroup with both low BMI and low blood pressure. ⁶⁰

Some genetic variants exert effects only under specific environmental conditions, a phenomenon referred to as gene–environment interaction (G × E). ⁶¹ , ⁶² An extension of this concept to polygenic scores (PGS), with PRS being a common implementation, is referred to as PGS × E interaction. ⁶³ Notably, PGS × BMI interactions have been shown to influence a wide range of phenotypes, ⁶⁴ not limited to T2D, highlighting the broader potential of this approach.

Importantly, the concepts of BMI‐stratified PRS and PGS × BMI are not restricted to BMI and can be applied to stratification based on other readily available traits. ⁶³ More generally, target stratification based on existing traits can improve polygenic prediction across diverse phenotypes. ¹³ , ⁶⁰

GENETIC BASIS OF GDM

Gestational diabetes mellitus (GDM) is a common disorder of pregnancy, and women with a history of GDM are at increased risk of developing T2D. ¹⁰ , ⁶⁵ , ⁶⁶ A GWAS of GDM in 12,332 cases and 131,109 parous female controls identified 13 GDM‐associated loci and found they could be classified into two categories: T2D risk and predominantly gestational. ¹⁰ The eight loci with GDM‐predominant effects are located near genes that are involved in plausible cellular processes such as signal transduction and hormone processing. ¹⁰ Analyses integrating single‐cell RNA expression datasets indicated that both GDM and T2D are significantly associated with pancreatic beta cells, but only GDM had significant associations with brain tissues. ¹⁰ In cohorts of GDM women, PRS for T2D was associated with increased risk of future development of T2D after GDM onset. ⁶⁵

In South Asian population, pathway‐specific PRSs other than bilirubin cluster, especially beta cell and lipodystrophy clusters, were associated with GDM cases and individuals with incident T2D after GDM, compared to non‐diabetic controls. ⁵⁶ In contrast, the association of GDM with the obesity psPRS in the South Asian population was relatively weaker, despite obesity psPRS being highlighted as a major etiological pathway in European and Turkish mothers with GDM. ¹⁰ , ⁵⁶ , ⁶⁷

GENETIC BASIS OF T1D

Even for T1D, an autoimmune disorder, GWAS have uncovered robust associations with immune‐related loci. ¹¹ , ¹² Recent multi‐ancestry GWAS have further refined ancestry‐specific HLA effects and identified seven loci that were associated with T1D risk at genome‐wide significance: PTPN22, HLA‐DQA1, IL2RA, RNLS, INS, IKZF4‐RPS26‐ERBB3, and SH2B3, with four associated with T1D age at onset (PTPN22, HLA‐DQB1, INS, and ERBB3). ⁶⁸

A genetic risk score (GRS), which is a type of PRS generated using fewer genome‐wide significant variants, predicted progression from islet autoimmunity to clinical T1D in at‐risk individuals. ⁶⁹ In a Chinese cohort, a GRS for T1D was highly discriminative of T1D risk in the Chinese population and could aid in discriminating between T1D and T2D. ⁷⁰ GRS for T1D can also differentiate between monogenic diabetes and T1D, ⁷¹ and may serve as a biomarker for distinguishing early‐onset diabetes. ⁷²

The incidence of T1D varies across global regions, with higher rates reported in regions with predominantly European‐origin populations, such as Northern Europe, Northern America, and Australia/New Zealand, whereas lower rates have been reported in parts of Africa, South America, and Asia. ⁷³ , ⁷⁴ The lower incidence of T1D in Asian populations than in European populations may partly reflect differences in genetic susceptibility and the spread of historical infectious diseases. In East Asian populations, T1D risk is strongly influenced by HLA haplotypes that differ from those commonly observed in Europeans, ⁷⁵ and studies in Japanese and Chinese populations have also implicated non‐HLA loci, including INS, IL2RA, ERBB3, CLEC16A, and IL7R. ⁷⁶ , ⁷⁷ In addition, given evidence linking autoimmune disease‐associated loci to historical pathogen‐driven selection and individual infectious past history, ⁷⁸ , ⁷⁹ , ⁸⁰ genes involved in anti‐infectious responses, particularly against enteroviruses, may associate genetic susceptibility with infectious triggers. ⁸¹ , ⁸² , ⁸³ Differences in infectious exposures across global regions may also contribute to populational differences in T1D incidence, but direct evidence regarding gene–infection interactions remains limited and should be addressed in future studies.

CROSS‐BIOBANK STUDIES

Large‐scale biobank studies of T2D in East Asian populations have been led primarily by BBJ. Using GWAS summary statistics of T2D from BBJ, post‐GWAS analyses including PRS prediction have increasingly been conducted in recently established biobanks across East Asia. For example, improvements in PRS performance for T2D using a cross‐populational PRS construction method were demonstrated in Taiwan Biobank. ⁸⁴ , ⁸⁵

There is an increasing number of studies incorporating BBJ and Tohoku Medical Megabank data, ⁸⁶ , ⁸⁷ and the BMI‐stratified PRS study for T2D replicated its findings in TMM cohort. ¹³ In another study, the findings in BBJ indicating an East Asian‐specific negative genetic correlation between metabolic diseases (such as T2D and dyslipidemia) and respiratory diseases (such as asthma and chronic obstructive pulmonary disease [COPD]) were replicated by PRS analysis in TMM. ⁸⁸ Age‐related loss of the Y chromosome (LOY) in a subset of male cells was associated with an increased risk of T2D in BBJ, and a consistent effect was observed in the TMM study. ⁸⁹

Recently, large‐scale genetic studies are increasingly advancing across East Asian biobanks, including the China Kadoorie Biobank (CKB), ⁹⁰ Korean Genome and Epidemiology Study (KoGES), ⁹¹ Korean Cancer Prevention Study‐II (KCPS2), ⁹² Taiwan Precision Medicine Initiative (TPMI), ⁹³ , ⁹⁴ Japan Multi‐Institutional Collaborative Cohort (J‐MICC), ⁹⁵ Japan Public Health Center‐based prospective Study (JPHC), ⁹⁶ Japan Environment and Children's Study (JECS), ⁹⁷ Okinawa Bioinformation Bank (OBi), ⁹⁸ and the Nagahama Study. ⁹⁹ In addition, BBJ has expanded its resources through a second, independent cohort, which recruited participants independently from the original BBJ. Whereas the original BBJ enrolled participants between 2003 and 2007, BBJ‐2nd collected data between 2013 and 2017.

A key consideration for the genetic basis for “Asian diabetes” is that Asian populations are genetically diverse and structured, spanning East, Southeast, and South Asia with distinct demographic histories, allele frequency, and LD patterns. ¹⁰⁰ , ¹⁰¹ Methodologically, improving trans‐ancestry PRS portability requires both better statistical approaches and expanded ancestry‐matched discovery data. ¹³ , ³² , ¹⁰² Therefore, developing and expanding cohorts across Asia is essential for robust PRS validation, and cross‐biobank studies in Asia are expected to further develop and accelerate progress in this field.

CONCLUSIONS AND FUTURE PERSPECTIVES

Diabetes comprises a continuum of disorders ranging from monogenic forms to highly polygenic common forms, with substantial heterogeneity in pathophysiology, treatment response, and complication risk. Recent advances in human genetics, such as the molecular diagnosis of monogenic and mitochondrial diabetes as well as GWAS and PRS for T2D, T1D, and GDM, are changing how diabetes can be classified and managed on an individual basis.

For rare monogenic diabetes, improving clinical recognition and access to appropriate genetic testing remains essential to avoid missed diagnoses and to enable subtype‐specific therapy. ⁵ , ²⁶ For T2D, future work should move beyond treating T2D as a single disorder and incorporate subtype‐aware frameworks, including subtype clustering and pathway‐specific polygenic scores. ⁴¹ Using readily available traits, such as BMI and blood pressure or other clinical traits, offers a practical route to partially capture etiological heterogeneity and improve prediction. ¹³ , ⁶⁴

Many challenges remain to be addressed before PRS can be implemented in routine clinical practice. For example, as discussed in this review, directly applying GWAS results derived from European populations to non‐European individuals leads to reduced predictive accuracy and raises concerns about international health disparities. ⁴⁴ Although cross‐population PRS construction methods have been developed and are narrowing performance gaps between populations, these disparities have not yet been fully eliminated. ¹³ , ³² Therefore, expanding genomic resources across East, Southeast, and South Asia, and integrating multi‐ancestry methods with functional annotation will be key to improving both predictive performance and equity. ¹³ , ⁴⁴

In addition, ethical and practical concerns have been raised about the clinical use of PRS, including inappropriate use and misinterpretation of results. ¹⁰² , ¹⁰³ , ¹⁰⁴ Thus, physicians and healthcare professionals who can appropriately explain genetic risks based on an understanding of PRS and recommend suitable prevention and interventions are desired. ¹⁰³ It is also essential to provide appropriate and understandable explanations and science communication from genetic scientists, enabling a broad range of people, including individuals with diabetes, to correctly understand the characteristics and limitations of PRS. ¹⁰³

DISCLOSURE

The authors declare no conflict of interest.

Approval of the research protocol: N/A.

Informed consent: N/A.

Registry and the registration no. of the study/trial: N/A.

Animal studies: N/A.

DATA AVAILABILITY STATEMENT

Data sharing not applicable to this article as no datasets were generated or analysed during the current study.