The Clinical relevance of Polygenic Risk Scores to Type 2 Diabetes Mellitus in Korean Population

doi:10.21203/rs.3.rs-2998310/v1

The potential clinical utility of type 2 diabetes mellitus (T2DM) polygenic risk scores (PRS) is not thoroughly evaluated in the East Asian population. We aimed to assess whether T2DM PRS could have prognostic value and be used as a clinical instrument.

We constructed T2DM PRS for Korean individuals using large East Asian Biobank data with samples of 269,487 and evaluated the PRS in a prospective longitudinal study of Korean with 5490 samples with baseline and additional seven follow-ups.

Our analysis demonstrated that T2DM PRS could predict not only the progress from non-diabetes to T2DM, but also normal glucose tolerance to prediabetes and prediabetes to T2DM. Moreover, T2DM patients in the top decile PRS group were more likely to be treated with insulin with HR = 1.69 (p-value = 2.31E-02) than the remaining PRS groups. T2DM PRS was significantly high in severe diabetic subgroups with insulin resistance and $\beta$-cell dysfunction (p-value = 0.0012). PRS could modestly improve the prediction accuracy of the Harrel’s C-index by 9.88% (p-value < 0.001) in T2DM prediction models.

By utilizing prospective longitudinal study data and extensive clinical risk factors, our analysis provides insights into the clinical utility of the T2DM PRS.

Health sciences/Diseases/Endocrine system and metabolic diseases/Diabetes/Type 2 diabetes mellitus

Biological sciences/Genetics

Type 2 diabetes mellitus (T2DM) is a major public health problem. Over the past three decades, the number of people with T2DM has more than doubled globally¹. The prevalence of T2DM has also increased in Korea². According to the Korean National Health and Nutrition Examination Surveys³, overall T2DM prevalence among adults has increased from 8.9% in 2001 to 16.7% in 2020. Furthermore, American Diabetes Association (ADA) experts estimated the conversion rate from prediabetes to T2DM up to 70%. Therefore, identifying high-risk individuals for prediabetes and T2DM is important as early targeted detection and intervention can prevent T2DM development and its related complications such as renal complications, heart disease, and stroke⁴.

In recent years, Genome-wide association studies (GWAS) have identified a large number of genetic variants associated with the risk of T2DM⁵. Aggregating the information from GWAS, polygenic risk scores (PRS) have been constructed to predict individual genetic susceptibility and is expected to enable enhanced screening and preventive therapies for T2DM and its medical complications⁶. Previous studies have shown that PRS can identify high-risk individuals for T2DM ^6–8. However, the existing PRS research was largely limited to a disease prediction on cross-sectional data. Although earlier studies have evaluated PRS in longitudinal data, only a few risk variants were included in calculating PRS ^9,10. Moreover, T2DM PRS have been constructed and evaluated mostly in the European population. According to the PRS catalog ¹¹, among 23 studies constructed T2DM PRS, only three were evaluated for East Asian^12–14. To fully understand the prognostic value of T2DM PRS in East Asian and use it as a clinical instrument ¹⁵, PRS construction using the large-scale East Asian biobank and evaluation in prospective longitudinal studies are needed.

The purpose of this paper is twofold. We first constructed East Asian T2DM PRS using large biobank data from Korea and Japan. We carried out genome-wide association analysis and meta-analysis to construct GWAS summary statistics for PRS training and applied Lassosum and PRS-CS to construct PRS. Second, we evaluated the performance of the T2DM PRS using prospective cohort data in the Korean Genome and Epidemiology Study (KoGES) ¹⁶ with 16 years of follow-ups. We demonstrated that T2DM PRS could predict not only the T2DM risk, but also the T2DM severity. Also, we verified that homeostasis model assessment – $\beta$-cell functions (HOMA-B) is associated with T2DM PRS. Moreover, when we classified T2DM patients in novel diabetic subgroups using clinical biomarkers, including HOMA – insulin resistance (HOMA-IR) and HOMA-B, PRS was significantly high in the severe diabetic subgroups. Finally, our evaluation of the prediction of T2DM incidence in a series of models, including family history, physical measurements, and clinical risk factors, showed that the inclusion of PRS improved the prediction.

Study overview and PRS construction

The overview of the study is described in Fig. 1. KoGES has three cohorts, KoGES_Ansan and Ansung, KoGES_HEXA, and KoGES _CAVAS. We carried out GWAS of T2DM using KoGES_HEXA and meta-analyzed them with Biobank Japan T2DM GWAS results. A total of 269,487 samples (44,315 Cases and 225,172 controls) were included in the meta-analyzed East Asian T2DM GWAS summary, which was used for PRS training (Supplementary Fig. S1). KoGES_CAVAS (n = 8,105) was used as validation data for hyperparameter selection. We evaluated two genome-wide PRS construction methods, Lassosum¹⁷ and PRS-CS¹². In KoGES_Ansan and Ansung datasets, both methods performed similarly, with slightly better AUC from Lassosum in the PRS-only model (Supplementary Table S1). We used PRS constructed by Lassosum in the rest of the paper.

Participant characteristics

A total of 5,490 participants with Korean chip ¹⁸ genotyped individuals in the KoGES_Ansan and Ansung dataset were used to evaluate the T2DM PRS. The cumulative prevalence of T2DM at the baseline and each follow-up can be found in Supplementary Table S2. At the baseline, the mean age of participants was 52 years old, 47.6% were male, and 13.6% of participants had diabetes. Participants were classified into three groups according to a PRS percentile: the top and bottom decile and the middle (10–90%). There was no significant difference in BMI in these three PRS groups. Compared with the middle and bottom decile of the PRS groups, the top decile PRS group had worse lipid and glucose measures, including triglyceride, fasting glucose, and HbA1c at the baseline. There were 8.85 times more T2DM patients in the top decile PRS group than in the bottom decile PRS group. The top decile PRS group also had 2.58 times more family history of T2DM than the bottom decile PRS group. The detail of characteristics at the baseline is shown in Table 1.

Association between cumulative prevalence of T2DM and PRS

To investigate the relationship between the cumulative prevalence of T2DM and PRS, we conducted a survival analysis with age-at-diagnosis as an outcome, including baseline cases. A Kaplan-Meier plot showed that the cumulative prevalence of T2DM was significantly higher in the top decile of the PRS group than in the other two groups (Supplementary Fig. S2). Hazard ratios from the Cox model that compared the top and bottom decile PRS groups with the middle PRS group were 2.29 (top vs middle, 95% CI: 2.02–2.59, p-value < 2.00E-16) and 0.45 (bottom vs middle, 95% CI: 0.36–0.56, p-value = 2.84E-13), respectively. In addition to the categorized PRS, we used the standardized PRS. The hazard ratio with the standardized PRS was 1.59 (95% CI: 1.52–1.67, p-value < 2.00E-16).

To validate our T2DM PRS, we applied our PRS model to 1503 east Asian samples in UK Biobank (UKBB). Supplementary Fig. S3 shows that the cumulative prevalence of T2DM was significantly higher in the top decile PRS group than in the other two groups. The hazard ratios that compared the top decile of the PRS group and bottom decile PRS group with the middle PRS group are 2.167 (top vs. middle, 95% CI: 1.40–3.36, p-value = 0.00054) and 0.36 (bottom vs. middle, 95% CI: 0.15–0.89, p-value = 0.026), respectively.

PRS can predict incident prediabetes and T2DM

We hypothesized that T2DM PRS could predict not only the progress from non-diabetes to T2DM but also NGT to prediabetes and prediabetes to T2DM. We only included NGT individuals and prediabetes at the baseline for the analysis. Figure 2 showed that the incidence of T2DM in both non-diabetes and prediabetic individuals was significantly higher in the top decile PRS group than in the other two groups (p-value $<$2.00E-16). Furthermore, the higher percentile PRS group was associated with a higher incidence of prediabetes. Hazard ratios that compared the top decile PRS group with the middle PRS group were 1.92 for the overall risk of T2DM in non-diabetes participants, 1.38 for the progression to prediabetes from NGT, and 1.65 for the progression to T2DM from prediabetes (Table 2).

PRS can predict progression to insulin prescription

To evaluate whether T2DM PRS can predict T2DM severity, we analyzed the progression to insulin prescription and T2DM complications. The Cox regression with PRS and sex as predictors were used to model insulin prescription. We excluded the insulin-treated participants at the baseline. Similarly, we used both categorized PRS and standardized PRS. We found that T2DM patients in the top decile PRS group were more likely to be treated with insulin, whereas no patient in the bottom decile PRS group was prescribed insulin. We also fit similar models with T2DM complications as an outcome. Neither categorized PRS nor standardized PRS was significant in the model. The detail of the result is shown in Table 3 and Supplementary Table S3.

PRS is associated with HOMA-B

HOMA-IR and HOMA-B are biomarkers for insulin resistance and $\beta$-cell functions. We investigate the changes in HOMA-IR and HOMA-B in the top decile and the rest of the PRS group during the development of T2DM. For T2DM individuals, we examined the retrograde trajectories of HOMA-IR and HOMA-B by setting the diagnosis of T2DM to time zero and tracing back every two years. For non-diabetes, we examined the forward trajectories of HOMA-IR and HOMA-B from the baseline. Supplementary Fig. S4 showed the changes in HOMA-IR and HOMA-B. Those who developed diabetes had an overall high HOMA-IR and low HOMA-B than those who did not develop diabetes. As expected, we observed overall increasing HOMA-IR and decreasing HOMA-B as the time closer to T2DM diagnosis. The confidence intervals of the two groups at each time point overlapped because of the small sample size. However, by performing permutation test, the HOMA-B trajectories between the top decile and the rest of the PRS group were significantly different within T2DM (p-value = 0.011) and non-diabetes (p-value = 0.0074). We conducted the same permutation test for HOMA-IR, but there was no significant difference in HOMA-IR trajectories between two PRS groups for diabetes (p-value = 0.37) and non-diabetes (p-value = 0.18).

PRS is associated with severe diabetic subgroup

Previous studies have suggested novel diabetic subtyping, which may guide prevention and treatment strategies for T2DM and its complications ¹⁹. Therefore, we classified T2DM patients into four subgroups by data-driven k-means cluster analysis using BMI, age at diagnosis, HOMA-B, HOMA-IR, and HbA1c and observed how PRS differ within those clusters. Cluster 1 includes 105 (7.94%) of 1322 patients and labeled as a severe diabetic group with extremely high HbA1c, early age at diagnosis, relatively high BMI, insulin resistance (high HOMA-IR), and β-cell dysfunction (low HOMA-B). In contrast to cluster 1, cluster 2 (36.4%) was mild diabetic subgroups with relatively low HbA1c, BMI, and HOMA-IR and high HOMA-B. Cluster 3 (30.3%), labeled as mild age-related diabetes (MARD), was diagnosed with T2DM later age than the other subgroups. Cluster 4 (25.4%) had relatively high BMI, HbA1c, and insulin resistance and was labeled as mild obesity-related diabetes (MOD). Supplementary Fig. S5 shows that even within novel diabetic subtyping, PRS was significantly high in the severe diabetic subgroups (p-value = 0.0012).

PRS can improve prospective prediction accuracy

To evaluate the improvement of the prediction accuracy with PRS, we considered a series of models with and without PRS. We excluded T2DM patients at the baseline and used the baseline measured risk factors to predict the future incidence of T2DM. The baseline model included sex and age. We subsequently added family history, physical measurements (BMI and SBP), smoking status, and clinical risk factors (HDL, LDL, and TG) into the model. The model descriptions can be found in materials and methods. As expected, the model with a larger set of risk factors had a higher Harrel’s C-index. We also showed that models with either standardized or categorized PRS significantly increased the Harrel’s C-index than those without PRS. For example, Harrell’s C-index of the model with sex, age, and family history (model2) was 0.586, but it was improved to 0.631 with the standardized PRS. We also verified that standardized PRS was more informative for the incidence prediction than categorized PRS by showing a larger Harrel’s C-index. The detail of the result is shown in Table 4 and Supplementary Table S5.

Identifying high-risk individuals for T2DM is important as early targeted detection and intervention, such as lifestyle modification or medical intervention, can postpone or even prevent T2DM. By aggregating GWAS results, the PRS has emerged as a powerful tool for identifying individual genetic susceptibility. In addition, PRS has the potential to be used to infer disease prognosis and subtyping ^20,21. However, current PRS research is primarily limited to disease prediction, and the clinical utility of T2DM PRS in predicting incident T2DM is not fully evaluated.

Our analysis demonstrated that the top decile PRS group is more likely to progress to T2DM. Even when we applied the more robust criterion to remove Type 1 DM (T1DM) patients from the participants by excluding people who are diagnosed with diabetes aged under 40, it showed a similar result. Hazard ratios from the Cox model that compared the top and bottom decile with the middle PRS group were 2.21 (top vs. middle, p-value < 2E-16) and 0.442 (bottom vs. middle, p-value = 4.11E-13), respectively.

Prediabetes, defined based on glycemic parameters above NGT but below the diabetes threshold, is a high-risk condition for diabetes with an annualized conversion rate of 5–10% ²². Previous studies have shown that T2DM PRS is associated with prediabetes ^23,24. However, no study showed that T2DM PRS predicts the progression from prediabetes to T2DM. Our results showed that T2DM PRS can predict not only the progress from non-diabetes to T2DM but also NGT to prediabetes and prediabetes to T2DM. By identifying high-risk individuals among the prediabetic population and giving them a guideline to maintain optimal lifestyle habits, we can reduce the progression rate of T2DM from prediabetes ²⁵.

A previous study has shown that T2DM PRS could be a useful tool for predicting disease severity that can be measured by the escalation of treatment options and the progression to T2DM complications ²³. T2DM patients’ glucose levels could be controlled by oral diabetes medication with a combination of lifestyle modifications. However, some patients with a longer duration of T2DM or less well-controlled glucose levels should be treated with insulin ²⁶. Also, people with T2DM have an increased risk of developing macrovascular and microvascular complications. Our study found that T2DM patients in the higher percentile PRS group were more likely to be prescribed insulin. However, we could not demonstrate that the T2DM PRS can predict the progression to neither T2DM macrovascular complication nor nephropathy. Even in existing studies, T2DM PRS was significantly associated with increased risk of neuropathy ²³ and cardiovascular disease ²⁷ but not with macrovascular complication nor diabetic nephropathy. To demonstrate the association between T2DM PRS and diabetic complications, we need to further understand the biological pathway or systems that can clarify the specific cause of genetic risk and T2DM complications ^28,29.

Insulin resistance and $\beta$-cell dysfunction are the measurements to understand a pathophysiological mechanism in T2DM ³⁰. The genetic variants linked to T2DM are associated with $\beta$-cell dysfunction ³¹ and insulin secretion ³². Previous studies also found that β-cell function is already impaired prior to the progression of prediabetes ³³. A recent study showed T2DM PRS was primarily related to $\beta$-cell dysfunction in the Korean population ³⁴. However, this study investigated the association between T2DM PRS and HOMA-B at the baseline measurements only. To fully understand the relationship between PRS and HOMA-B, tracing HOMA-B during the progression to T2DM is needed. The present study examined the trajectories of HOMA-B during the development of T2DM and found that HOMA-B in the top decile PRS group is consistently lower than in the remaining group, both in the group who developed diabetes and stayed non-diabetes.

Previously, diabetes is classified into type 1 and type 2 diabetes only. However, recent studies have suggested the stratification of populations at risk for diabetes using clinical biomarkers to prevent progression to T2DM and even T2DM complications^19,35. In the present study, we classified T2DM patients into four different subgroups by clustering analysis. Clusters were based on five variables that are measured at the diagnosis of T2DM. T2DM patients were stratified into severe diabetic subgroups with insulin resistance and $\beta$-cell dysfunction, mild diabetic subgroups, MARD, and MOD. We found that PRS was also significantly high in severe diabetic subgroups.

In our study, we found that the model with PRS performs better in predicting the incidence of T2DM. The basic T2DM prediction model with sex, age, and PRS performed better than without PRS. Adding family history, physical measurements, and clinical risk factors to the basic model steadily improved Harrell’s C-index. Moreover, we found evidence that standardized PRS can improve the prediction performance over the categorized PRS.

Our study has multiple strengths. First, we calculated PRS using a recently developed method and genome-wide meta-analysis to improve prediction accuracy further. Second, by utilizing prospective longitudinal study data, we verified that T2DM PRS is a predictor of disease risk and severity and an associated factor with the clinical biomarker HOMA-B. Moreover, we showed T2DM PRS is related to severe diabetic subgroups. Third, we constructed the predictive model of T2DM, including physical measurements, and clinical risk factors, to increase prediction performance. Although our analysis provides insight into the clinical utility of the T2DM PRS, there are some limitations. The information on the type of oral diabetes medication or dosage of insulin prescription was not explicitly described because all questionnaires were self-reported by participants. Also, the participants weren’t asked about the history of T2DM complications but the comprehensive history of the disease. Therefore, even though we excluded the participants whose incident diseases were ahead of T2DM, we cannot be definite that those diseases are T2DM complications.

In conclusion, our analysis of prospective longitudinal study data infers that PRS could potentially have clinical value. Furthermore, implementing PRS in clinical assessment tools can help T2DM screening and prognosis and hope for preventive intervention and strict glycemic control for high-risk individuals.

Study population and survey methodology

KoGES is a consortium project of prospective cohort studies. The cohorts in KoGES include KoGES_Ansan and Ansung, KoGES_heath examinee (HEXA), and KoGES_cardiovascular disease association study (CANVAS), where participants were recruited from the national health examine registry with age 40 at baseline. Due to its extensive follow-ups, we used KoGES_Ansan and Ansung study as the main analysis data. Participants consecutively responded to baseline and seven additional follow-up phases every two years from 2001 to 2016. During each follow-up, identical questionnaires (socio-demographic data, lifestyle, medical history, etc.), physical measurements (height, weight, blood pressure, etc.), and clinical examinations (blood test, urine test, etc.) were conducted. The trained interviewer questioned participants’ disease history, family history of the disease, and medication prescriptions such as insulin.

Genotyping and quality control

The genotypes were measured using Korean Chip ¹⁸. We excluded samples with low call rate (< 97%), gender discrepancy, cryptic first-degree relative, high heterozygosity, and singletons. The genetic variants were excluded with criteria of Hardy-Weinberg equilibrium (HWE) p-value (< 10E-6) or low call rate (< 95%). The genotypes were phased using Eagle v2.3 imputed using IMPUTE4 with 1000 Genomes project phase 3 data, and the Korean reference genome was used as a reference panel. After excluding genetic variants with imputation quality score (IQS) < 0.8 and minor allele frequency < 1%, a total of 8,056,211variants were used for analysis ³⁶.

GWAS summary statistics construction and PRS calculation

We conducted GWAS with 58,622 participants in KoGES_HEXA cohort studies using a linear mixed model implemented in SAIGE ³⁷. We used the top 10 principal components (PCs), age, and sex for a covariate adjustment. Summary statistic of Biobank Japan (BBJ) was downloaded from ³⁸. We carried out the z-score-based meta-analysis for KoGES_HEXA with BBJ using inverse-variance weighting to obtain p-values and effect sizes for risk prediction. The combined had a total of 7,057,567 variants.

For PRS calculation, we considered two PRS construction methods, Lassosum ¹⁷ and PRS-CS ¹². For linkage disequilibrium (LD) reference panel, we used East Asian (EAS) in 1000 Genome project phase ³⁹. KoGES_CANVAS was used as validation data for tuning hyper-parameters in Lassosum.

T2DM and prediabetes

New-diagnosis of T2DM was defined by at least one criterion: self-reported diagnosed diabetes, treatment with hypoglycemic medication, fasting glucose level $\ge$ 126mg/dL, a glucose level of $\ge$ 200mg/dL after oral glucose test, or hemoglobin A1C (HbA1c) $\ge$ 6.5% (48mmol/mol) ⁴⁰. We excluded people who are diagnosed with diabetes aged under 30. According to the ADA guideline, we define prediabetes as either a fasting glucose level of 100-125mg/dL, or 2-hr glucose level of 140mg/dL to 199 mg/dL during 75-g oral glucose tolerance test, or elevated HbA1c level of 5.7–6.4% (39 to 46 mmol/mol) ²⁶. We defined non-diabetes as normal glucose tolerance (NGT) and prediabetic individuals.

T2DM Complications

The participants self-reported disease history of myocardial infarction, coronary artery disease, congestive heart failure, cerebrovascular disease, peripheral artery disease and kidney disease. Kidney disease was defined as self-reported diagnosis of kidney disease or glomerular filtration rate < 60 that is estimated with the equation suggested by the Chronic Kidney Disease Epidemiology Collaboration ⁴¹. To be clear that those diseases are T2DM complications, we excluded the participants whose incident diseases were ahead of T2DM in our analysis. Also, we used complications diagnosis age as the earliest age of those diseases ages.

HOMA-IR and HOMA-B

The HOMA-IR index is the product of basal glucose and insulin levels divided by 22.5, and the HOMA-B is computed as the product of 20 and basal insulin levels divided by the value of basal glucose minus 3.5 ⁴².

T2DM novel subgroups

For T2DM patients’ classification, first, we excluded T2DM cases at the baseline who self-reported T2DM diagnosis. Using BMI, age at diagnosis, HOMA-B, HOMA-IR, and HbA1c that are measured at the diagnosis of T2DM, following ¹⁹, we conducted the data-driven k-means cluster analysis for 1322 T2DM patients. Before clustering, all variables were converted to a mean value of 0 and a standard deviation (SD) of 1. All extreme outliers greater than 5 SD from the mean were excluded. We used the elbow method for the number of clusters k to capture the point at which Within cluster Sum of Squares rapidly decreased. We used scikit-learn package in python version 3.9.7 to conduct k-means cluster analysis.

Construction of the prediction models

We determined the relationship between PRS and T2DM based on a multivariate Cox regression model with sex and age as Eq. (1).

$${model}_{1}: T2DM \sim PRS+ sex+age$$

1

We calculated the time from the baseline age to the diagnosis age for incident T2DM (cases) or to the last follow-up age for each person without T2DM (censored). To increase the performance accuracy, we also considered traditional risk factors for T2DM such as family history, physical measurement, and clinical risk factors that are measured or answered at the baseline. Characteristics at the study population’s baseline were described as means $\pm$ SD or percentages (Table 1 and Supplementary Table S4). Four different prediction models are represented as,

$${model}_{2}: T2DM\sim PRS+ sex+age+Family History$$

2

$${model}_{3}: T2DM \sim PRS+sex+age+Family History+$$

$$BMI+ SBP+Smoking Status$$

3

$${model}_{4}: T2DM \sim PRS+sex+age+Family History+$$

$$BMI+ SBP+Smoking Status+HDL+LDL+TG$$

4

We corrected the LDL level for using the lipid-lowering drug by dividing LDL concentration by 0.7 ⁴³ and adjusted the SBP level for treated individuals with blood pressure-lowering medication by adding 15mmHg to measurements ⁴⁴. We observed Harrel’s C-index as the prediction performance of the model without PRS to observe the contribution of PRS in predicting T2DM. We excluded the fasting glucose and HbA1c in the prediction model because those measures are used to diagnose T2DM.

Statistical analysis

For all the Cox regression analyses, we used sex as a predictor. We classified participants into three groups according to a percentile of PRS: the top (> 90%) and bottom decile (< 10%), and the middle (10–90%) and used the largest 10–90% bin as the reference. We also evaluated with standardized PRS, where standardized PRS is continuous scale PRS with mean zero and variance one. The age of the T2DM diagnosis is defined as the earlier age of either responded T2DM diagnosis age in the survey or the age of the first follow-up with meeting diagnosis criteria. Schoenfeld residual test has been used as statistical tests to test proportional hazards assumption.

T2DM cumulative prevalence

The Cox regression was used to model the time-to-diagnosis T2DM, where the time is defined as the T2DM diagnosis age for T2DM (cases) or the last follow-up age for each participant without T2DM (censored).

T2DM risk progression analysis

Cox regression was used to model development from non-diabetes to T2DM, NGT to prediabetes, and prediabetes to T2DM. We calculated the time from the baseline age to the diagnosis age for incident prediabetes/T2DM (cases) or to the last follow-up age for each person without incident prediabetes/T2DM (censored).

T2DM severity progression analysis

We used the same Cox regression model for insulin prescription and T2DM complications. The time was calculated from T2DM diagnosis age to age answered incident insulin prescription/T2DM complications (cases) or last follow-up age (censored). Because none of the individuals in the bottom decile PRS group were prescribed insulin, we classified participants into two groups: the top (> 90%) and the remaining (90%) PRS group. We used the remaining PRS group as the reference.

HOMA analysis

We calculated the median and the confidence interval of HOMA at each time point for the top decile PRS group and the remaining group. To evaluate whether the HOMA retrograde trajectories between the top decile of the PRS group and the rest are statistically different, first, we obtained the test statistic as the summation of the median difference between the two groups at each time point. To obtain the p-value, we performed a permutation test. We randomly sampled PRS group index 10,000 times and calculated the same test statistic for each permuted sample. Permutation p-values were calculated as the proportion of test statistics from permuted samples that were more extreme than the observed test statistics.

All the above statistical analyses were conducted using R version 4.0.3 software and considered a 2-sided p-value < 0.05 statistically significant.

Ethics Declaration

The study was approved by the Institutional Review Board of Seoul National University (approval number: IRB No. E2012/002-001). This research was conducted following the Declaration of Helsinki. Informed consents were obtained from all the study participants.

Acknowledgements

This study was supported by Brain Pool Plus (Brain Pool+) Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Science and ICT [2020H1D3A2A03100666]. This research was also supported by the New Faculty Startup Fund from Seoul National University.

Author contributions

N.Y.K and S.L researched, analyzed the data, and wrote the manuscript. N.Y.K, S.L and S.K conceived and designed the experiments. N.Y.K, S.K and J.L performed the analyses. H.L, S.H.K, Y-H.L, and H.L contributed to the study design and review the manuscript. All authors read and approved the final manuscript.

Competing Interest

All authors declare no competing interests.

Data availability

Data in this study were from the Korean Genome and Epidemiology Study (KoGES; 4851-302), National Research Institute of Health, Centers for Disease Control and Prevention, Ministry for Health and Welfare, Republic of Korea. UK Biobank data were accessed under the accession number UKB: 45227. BBJ summary statistics used in this study were downloaded from https://pheweb.jp/

Chen, L., Magliano, D. J. & Zimmet, P. Z. The worldwide epidemiology of type 2 diabetes mellitus--present and future perspectives. Nat Rev Endocrinol 8, 228-236 (2011). https://doi.org:10.1038/nrendo.2011.183
Nanditha, A. et al. Diabetes in Asia and the Pacific: Implications for the Global Epidemic. Diabetes Care 39, 472-485 (2016). https://doi.org:10.2337/dc15-1536
Bae, J. H. et al. Diabetes Fact Sheet in Korea 2021. Diabetes Metab J 46, 417-426 (2022). https://doi.org:10.4093/dmj.2022.0106
Lall, K., Magi, R., Morris, A., Metspalu, A. & Fischer, K. Personalized risk prediction for type 2 diabetes: the potential of genetic risk scores. Genet Med 19, 322-329 (2017). https://doi.org:10.1038/gim.2016.103
Xue, A. et al. Genome-wide association analyses identify 143 risk variants and putative regulatory mechanisms for type 2 diabetes. Nat Commun 9, 2941 (2018). https://doi.org:10.1038/s41467-018-04951-w
Khera, A. V. et al. Genome-wide polygenic scores for common diseases identify individuals with risk equivalent to monogenic mutations. Nat Genet 50, 1219-1224 (2018). https://doi.org:10.1038/s41588-018-0183-z
Liu, W., Zhuang, Z., Wang, W., Huang, T. & Liu, Z. An Improved Genome-Wide Polygenic Score Model for Predicting the Risk of Type 2 Diabetes. Front Genet 12, 632385 (2021). https://doi.org:10.3389/fgene.2021.632385
Mars, N. et al. Polygenic and clinical risk scores and their impact on age at onset and prediction of cardiometabolic diseases and common cancers. Nat Med 26, 549-557 (2020). https://doi.org:10.1038/s41591-020-0800-0
Lyssenko, V. et al. Clinical risk factors, DNA variants, and the development of type 2 diabetes. N Engl J Med 359, 2220-2232 (2008). https://doi.org:10.1056/NEJMoa0801869
Go, M. J. et al. Genetic-risk assessment of GWAS-derived susceptibility loci for type 2 diabetes in a 10 year follow-up of a population-based cohort study. J Hum Genet 61, 1009-1012 (2016). https://doi.org:10.1038/jhg.2016.93
Lambert, S. A. et al. The Polygenic Score Catalog as an open database for reproducibility and systematic evaluation. Nature Genetics 53, 420-425 (2021). https://doi.org:10.1038/s41588-021-00783-5
Ge, T., Chen, C. Y., Ni, Y., Feng, Y. A. & Smoller, J. W. Polygenic prediction via Bayesian regression and continuous shrinkage priors. Nat Commun 10, 1776 (2019). https://doi.org:10.1038/s41467-019-09718-5
Tanigawa, Y. et al. Significant sparse polygenic risk scores across 813 traits in UK Biobank. PLoS Genet 18, e1010105 (2022). https://doi.org:10.1371/journal.pgen.1010105
Weissbrod, O. et al. Leveraging fine-mapping and multipopulation training data to improve cross-population polygenic risk scores. Nat Genet 54, 450-458 (2022). https://doi.org:10.1038/s41588-022-01036-9
Lewis, C. M. & Vassos, E. Polygenic risk scores: from research tools to clinical instruments. Genome Med 12, 44 (2020). https://doi.org:10.1186/s13073-020-00742-5
Kim, Y., Han, B. G. & Ko, G. E. S. g. Cohort Profile: The Korean Genome and Epidemiology Study (KoGES) Consortium. Int J Epidemiol 46, 1350 (2017). https://doi.org:10.1093/ije/dyx105
Mak, T. S. H., Porsch, R. M., Choi, S. W., Zhou, X. & Sham, P. C. Polygenic scores via penalized regression on summary statistics. Genet Epidemiol 41, 469-480 (2017). https://doi.org:10.1002/gepi.22050
Moon, S. et al. The Korea Biobank Array: Design and Identification of Coding Variants Associated with Blood Biochemical Traits. Sci Rep 9, 1382 (2019). https://doi.org:10.1038/s41598-018-37832-9
Ahlqvist, E. et al. Novel subgroups of adult-onset diabetes and their association with outcomes: a data-driven cluster analysis of six variables. Lancet Diabetes Endocrinol 6, 361-369 (2018). https://doi.org:10.1016/S2213-8587(18)30051-2
Mavaddat, N. et al. Polygenic Risk Scores for Prediction of Breast Cancer and Breast Cancer Subtypes. Am J Hum Genet 104, 21-34 (2019). https://doi.org:10.1016/j.ajhg.2018.11.002
Sharp, S. A. et al. Development and Standardization of an Improved Type 1 Diabetes Genetic Risk Score for Use in Newborn Screening and Incident Diagnosis. Diabetes Care 42, 200-207 (2019). https://doi.org:10.2337/dc18-1785
Tabak, A. G., Herder, C., Rathmann, W., Brunner, E. J. & Kivimaki, M. Prediabetes: a high-risk state for diabetes development. Lancet 379, 2279-2290 (2012). https://doi.org:10.1016/S0140-6736(12)60283-9
Ashenhurst, J. R. et al. A Polygenic Score for Type 2 Diabetes Improves Risk Stratification Beyond Current Clinical Screening Factors in an Ancestrally Diverse Sample. Front Genet 13, 871260 (2022). https://doi.org:10.3389/fgene.2022.871260
Huang, X., Han, Y., Jang, K. & Kim, M. Early Prediction for Prediabetes and Type 2 Diabetes Using the Genetic Risk Score and Oxidative Stress Score. Antioxidants (Basel) 11 (2022). https://doi.org:10.3390/antiox11061196
Glechner, A. et al. Effects of lifestyle changes on adults with prediabetes: A systematic review and meta-analysis. Prim Care Diabetes 12, 393-408 (2018). https://doi.org:10.1016/j.pcd.2018.07.003
American Diabetes, A. 9. Pharmacologic Approaches to Glycemic Treatment: Standards of Medical Care in Diabetes-2021. Diabetes Care 44, S111-S124 (2021). https://doi.org:10.2337/dc21-S009
Yun, J. S. et al. Polygenic risk for type 2 diabetes, lifestyle, metabolic health, and cardiovascular disease: a prospective UK Biobank study. Cardiovasc Diabetol 21, 131 (2022). https://doi.org:10.1186/s12933-022-01560-2
Tremblay, J. et al. Polygenic risk scores predict diabetes complications and their response to intensive blood pressure and glucose control. Diabetologia 64, 2012-2025 (2021). https://doi.org:10.1007/s00125-021-05491-7
Udler, M. S. et al. Type 2 diabetes genetic loci informed by multi-trait associations point to disease mechanisms and subtypes: A soft clustering analysis. PLoS Med 15, e1002654 (2018). https://doi.org:10.1371/journal.pmed.1002654
Laakso, M. Biomarkers for type 2 diabetes. Mol Metab 27S, S139-S146 (2019). https://doi.org:10.1016/j.molmet.2019.06.016
Hur, H. J. et al. Association of Polygenic Variants with Type 2 Diabetes Risk and Their Interaction with Lifestyles in Asians. Nutrients 14 (2022). https://doi.org:10.3390/nu14153222
Suzuki, K. et al. Identification of 28 new susceptibility loci for type 2 diabetes in the Japanese population. Nat Genet 51, 379-386 (2019). https://doi.org:10.1038/s41588-018-0332-4
Saisho, Y. beta-cell dysfunction: Its critical role in prevention and management of type 2 diabetes. World J Diabetes 6, 109-124 (2015). https://doi.org:10.4239/wjd.v6.i1.109
Hahn, S. J., Kim, S., Choi, Y. S., Lee, J. & Kang, J. Prediction of type 2 diabetes using genome-wide polygenic risk score and metabolic profiles: A machine learning analysis of population-based 10-year prospective cohort study. EBioMedicine 86, 104383 (2022). https://doi.org:10.1016/j.ebiom.2022.104383
Wagner, R. et al. Pathophysiology-based subphenotyping of individuals at elevated risk for type 2 diabetes. Nat Med 27, 49-57 (2021). https://doi.org:10.1038/s41591-020-1116-9
Nam, K., Kim, J. & Lee, S. Genome-wide study on 72,298 individuals in Korean biobank data for 76 traits. Cell Genomics 2, 100189 (2022). https://doi.org:https://doi.org/10.1016/j.xgen.2022.100189
Zhou, W. et al. Scalable generalized linear mixed model for region-based association tests in large biobanks and cohorts. Nat Genet 52, 634-639 (2020). https://doi.org:10.1038/s41588-020-0621-6
Ishigaki, K. et al. Large-scale genome-wide association study in a Japanese population identifies novel susceptibility loci across different diseases. Nature Genetics 52, 669-679 (2020). https://doi.org:10.1038/s41588-020-0640-3
Genomes Project, C. et al. A global reference for human genetic variation. Nature 526, 68-74 (2015). https://doi.org:10.1038/nature15393
Yang, S. J., Kwak, S. Y., Jo, G., Song, T. J. & Shin, M. J. Serum metabolite profile associated with incident type 2 diabetes in Koreans: findings from the Korean Genome and Epidemiology Study. Sci Rep 8, 8207 (2018). https://doi.org:10.1038/s41598-018-26320-9
Levey, A. S. et al. A new equation to estimate glomerular filtration rate. Ann Intern Med 150, 604-612 (2009). https://doi.org:10.7326/0003-4819-150-9-200905050-00006
Matthews, D. R. et al. Homeostasis model assessment: insulin resistance and beta-cell function from fasting plasma glucose and insulin concentrations in man. Diabetologia 28, 412-419 (1985). https://doi.org:10.1007/BF00280883
Tobin, M. D., Sheehan, N. A., Scurrah, K. J. & Burton, P. R. Adjusting for treatment effects in studies of quantitative traits: antihypertensive therapy and systolic blood pressure. Stat Med 24, 2911-2935 (2005). https://doi.org:10.1002/sim.2165
Noordam, R. et al. Multi-ancestry sleep-by-SNP interaction analysis in 126,926 individuals reveals lipid loci stratified by sleep duration. Nat Commun 10, 5121 (2019). https://doi.org:10.1038/s41467-019-12958-0

Table 1. Characteristics at the baseline of the KoGES_Ansan and Ansung dataset

	PRS Bottom Decile (n = 549)	Middle (10-90%) (n = 4392)	PRS Top Decile (n = 549)	P-value
Basic Information
Male (%)	267 (48.63)	2069 (47.11)	278 (50.64)	0.260
Age (year)	51.63	51.50	51.77	0.707
Physical Measurements
BMI (kg/m²)	24.51 3.09	24.64	24.70	0.550
SBP (mmHg)	119.85	120.83	122.26	0.078
DBP (mmHg)	80.01	80.10	80.58	0.620
Clinical Risk Factors
HDL (mg/dL)	50.01	49.32	48.39	0.063
LDL (mg/dL)	117.61	119.78	119.72	0.348
TG (mg/dL)	140.93	152.93	164.64	1.72E-03
Fasting Glucose (mg/dL)	86.58	91.65	102.32	< 2.00E-16
Fasting Insulin (IU/mL)	7.23	7.57	7.63	0.212
HbA1c (%)	5.47 (36 mmol/mol)	5.75 (39 mmol/mol)	6.25 (45 mmol/mol)	< 2.00E-16
Others
Current Smoking (%)	134 (24.42)	1025 (23.38)	136 (24.77)	0.727
Family History (%)	40 (7.29)	508 (11.57)	103 (18.76)	1.20E-08
Type 2 Diabetes Mellitus
Case (%)	20 (3.64)	549 (12.5)	177 (32.24)	< 2.00E-16

BMI: Body Mass Index; WC: Waist Circumference; SBP: Systolic Blood Pressure; DBP: Diastolic Blood Pressure; HDL: High-Density Lipoprotein; LDL: Low-Density Lipoprotein; TG: Triglyceride. Obtained P-value using one-way ANOVA analysis.

Table 2. Results of Cox regression analysis for predicting progression from non-diabetes to T2DM, from NGT to prediabetes, and from prediabetes to T2DM

Progression Prediction	PRS	Hazard Ratio	95% CI	P-value
Non-diabetes to T2DM	Categorized PRS
	Bottom Decile	0.5290	0.4161 – 0.6725	1.99E-07
	Middle (10-90%)	1.00 (reference)
	Top Decile	1.911	1.604 – 2.276	4.05E-13
	Standardized PRS	1.430	1.346 – 1.519	< 2.00 E-16
NGT to prediabetes	Categorized PRS
	Bottom Decile	0.8352	0.7224 – 0.9656	0.0150
	Middle (10-90%)	1.00 (reference)
	Top Decile	1.370	1.128 – 1.664	0.00151
	Standardized PRS	1.094	1.042 – 1.148	0.000286
Prediabetes to T2DM	Categorized PRS
	Bottom Decile	0.6181	0.4598 – 0.8309	0.00144
	Middle (10-90%)	1.00 (reference)
	Top Decile	1.642	1.352 – 2.000	6.06E-07
	Standardized PRS	1.280	1.192 – 1.374	1.07E-11

To evaluate the incidence cases, individuals with diabetes at the baseline were excluded. We evaluated Cox regression model with sex as a predictor. NGT: normal glucose tolerance; CI: confidence interval.

Table 3. Results of Cox regression analysis for predicting insulin prescription and T2DM complications

Prediction	PRS	Hazard Ratio	95% CI	P-value
Insulin Prescription	Categorized PRS
	Remaining	1.00 (reference)
	Top Decile	1.692	1.075 – 2.662	0.0231
	Standardized PRS	1.663	1.331 – 2.077	7.61E-06
T2DM Complications	Categorized PRS
	Bottom Decile	0.7841	0.4290 – 1.433	0.429
	Middle (10-90%)	1.00 (reference)
	Top Decile	0.8992	0.6900 – 1.172	0.432
	Standardized PRS	0.9878	0.8910 – 1.095	0.816

Cox regression model was used with sex and time from T2DM diagnosis to insulin prescription or T2DM complications in a year. For the insulin prescription prediction, participants were classified into two groups according to a percentile of PRS: top decile and remaining (90%). T2DM complications include myocardial infarction, coronary artery disease, congestive heart failure, cerebrovascular disease, peripheral artery disease, and kidney disease. T2DM: type 2 diabetes mellitus.

Table 4. Prediction performance evaluation using Harrel’s C-Index

Model 1: T2DM ~ sex + age; Model 2: T2DM ~ sex + age + family history; Model 3: T2DM ~ sex + age + family history + BMI + SBP + smoking status; Model 4: T2DM ~ sex + age + family history + BMI + SBP + smoking status + HDL + LDL + TG.

No competing interests reported.

The Clinical relevance of Polygenic Risk Scores to Type 2 Diabetes Mellitus in Korean Population

Status:

Journal Publication

Version 1

Abstract

Figures

Introduction

Results