Genetic Risk Scores Identify People at High Risk of Developing Diabetic Kidney Disease: A Systematic Review (2024)

Associated Data

Data Availability Statement

Abstract

Keywords: diabetic kidney disease, genetic risk score, polygenic risk score

A third of people with diabetes can develop diabetic kidney disease (DKD) (1, 2). It is the leading cause of end-stage renal disease (ESRD) (1, 3-6). Measures to prevent and treat DKD require better patient stratification.

Genome-wide association studies (GWAS) have discovered genetic variants associated with DKD risk, but translation of these findings into clinical benefit is difficult because individual variants explain a relatively small proportion of overall risk (7). A genetic risk score (GRS), however, aggregates the individual effects of variants to increase their predictive power (7). The aim of this systematic review was to summarize the existing evidence of using GRSs to identify people at high risk of developing DKD.

Research Design and Methods

Results

We identified 400 citations from 4 databases (Fig. 1). After the removal of duplicates (n = 84), 316 titles and abstracts were screened, 287 of which were excluded. Following full text review of 33 articles, 15 studies were included in this review on the basis of selection criteria (Table 1). Overall, 7 studies had positive results, 5 had mixed results, and 3 had negative results.

Results of the methodology quality assessments are reported in Supplementary Table 1 (9). All studies were observational in nature. Four studies (3 case-control studies and 1 cohort study) were categorized as excellent. Six studies (1 case-control study, 3 cross-sectional studies, and 2 cluster analyses) were categorized as good. Four studies (3 case-control studies and 1 cohort study) were categorized as fair. One study was categorized as poor, and this was a case-control study. The total quality design score ranged from 6 to 12 (poor to excellent), with a mean score of 9.4 (good). Within domains, mean scores were as follows: 5.9 points (out of 7) for validity, 2.3 points (out of 3) for results, and 1.3 (out of 2) for relevance. In general, studies that received a low score had differences in study groups that may have affected the outcome of interest, had a low odds ratio, or had low statistical significance.

Eleven studies investigated GRS in type 2 diabetes. Three studies investigated GRS in both type 1 diabetes and type 2 diabetes. One study investigated type 1 diabetes only. The number of SNPs that were used to construct the GRS ranged from 5 to 598 between studies. Most studies (9) investigated people with European ancestries and the rest (6) investigated Asian populations.

The single study investigating people with type 1 diabetes only (10) was a case-control study that investigated 32 SNPs associated with obesity in a European population. This study found an association with all 3 outcomes of DKD (broadly defined DKD, macroalbuminuria, and ESRD) with genetically determined elevated body mass index (BMI). Similarly, van Zuydam et al who investigated type 1 diabetes and type 2 diabetes in a UK cohort (11) constructed a weighted GRS for 20 traits related to diabetes, insulin resistance, obesity, hypertension, coronary artery disease, and lipids. These GRSs, constructed from signals identified in previously published GWAS, included between 10 and 96 SNPs per phenotype. A GRS for increased BMI was associated with all DKD phenotypes. Furthermore, a GRS for increased waist-to-hip ratio was associated with ESRD in subjects with type 2 diabetes in this cohort.

Three studies investigating GRS for DKD in type 2 diabetes used SNPs associated with chronic kidney disease (CKD) or DKD itself. Liao et al investigated 7 SNPs associated with DKD in Han Chinese people in previous GWAS (12). When added to a model of significant clinical predictors of DKD (age, obesity, abnormal triglycerides, hypertension, and heart disease), each additional risk allele in the GRS increased the risk of DKD by 1.24-fold [95% confidence interval (CI) 1.17-1.34]. In contrast, Barbieux et al constructed a GRS of 18 SNPs associated with renal function and CKD (5). In a cohort of French people with type 1 or type 2 diabetes, they found that risk of renal events did not differ according to GRS and that GRS did not significantly predict estimated glomerular filtration rate (eGFR) trajectory (13). Mixed results were seen by Zusi et al (14) who investigated a cohort of Italian people with type 2 diabetes. They constructed GRSs using 39 SNPs related to risk of kidney disease and 42 SNPs related to cardiovascular risk. After adjustment for a number of confounders, they found that the top GRS quintile had the lowest eGFR. No statistical significance, however, was detected between renal GRS and urinary album excretion rate.

Single nucleotide polymorphisms associated with other micro- and macrovascular diseases were also investigated in a further 3 studies. Vujkovic et al investigated a large cohort of European people with type 2 diabetes (15). They constructed a GRS from 558 SNPs associated with vascular outcomes in type 2 diabetes and a further 21 SNPs seen in European people only. The risk of DKD increased significantly with increased GRS. Similarly, Tremblay et al investigated a cohort of European people aged >65 years with type 2 diabetes (16). They identified 26 factors and outcomes that were grouped into 10 groups of risk/outcomes with 598 SNPs. Combined with sex, age at onset of diabetes, and diabetes duration, the GRS model allowed prediction of both microvascular and macrovascular endpoints of type 2 diabetes. Hsieh et al investigated whether SNPs for diabetic retinopathy had pleiotropic effects on DKD in Han Chinese people in Taiwan with type 2 diabetes and retinopathy (17). Their results suggested that the 33 SNPs that were investigated exerted no cumulative effect on the eGFR status, DKD risk, or ESRD risk among people with type 2 diabetes.

Four studies investigated SNPs related to insulin secretion, sensitivity, and/or resistance (11, 18-20). Wang et al examined the independent and joint effects of 35 SNPs associated with insulin secretion and 20 SNPs associated with insulin sensitivity on CKD in a prospective cohort of Chinese people with type 2 diabetes (20). After controlling for baseline confounding variables, each additional unit of a weighted GRS was associated with a 37.2% (P = 5.23 ×10⁷) increase in CKD risk. Compared to participants carrying 0 to 2 risk alleles or in the lowest quartile of the GRS, the adjusted risk for CKD was 3.2 (95% CI 1.97-5.07, P = 2.03 ×10⁶) and 2.5 (95% CI 1.56-3.85, P = 1.03 × 10⁴) carrying 5 or more risk alleles or in the highest quartile of the GRS.

Three of the studies investigating insulin secretion, sensitivity, and/or resistance were cluster analyses (18, 19, 21). Ahlqvist et al examined a large cohort of Swedish people with type 1 and type 2 diabetes. They classified people into subgroups of diabetes based on 5 variables: BMI, age of onset of diabetes, presence of glutamic acid decarboxylase antibodies, hom*oeostasis model assessment (HOMA) 2 estimates of β-cell function (HOMA2-B), and insulin resistance (HOMA2-IR) based on C-peptide concentrations. They described 5 “clusters” using these variables: severe autoimmune diabetes, severe insulin-deficient diabetes, severe insulin-resistance diabetes (SIRD), mild obesity-related diabetes (MOD), and mild age-related diabetes. Their analyses showed that SIRD was the cluster with the highest incidence of CKD, macroalbuminuria, and ESRD. They constructed GRS from 65 SNPs associated with type 2 diabetes and found this was associated with all clusters (P < .0008) except SIRD (P = .16). An insulin secretion risk score constructed from 16 SNPs was significantly associated with MOD (P = .0002) and mild age-related diabetes (P < .0001) but again showed no evidence of association with SIRD (P = .65).

Similarly, Wang et al subtyped Southeast Asian people with type 2 diabetes by de novo cluster analysis (19). They identified 3 novel subgroups of diabetes: MOD, mild age-related diabetes with insulin insufficiency (MARD-II), and severe insulin-resistant diabetes with relative insulin insufficiency (SIRD-RII). Over a median of 7.3 years’ follow-up, the SIRD-RII subgroup had the highest risks for progressive kidney disease, while the MARD-II subgroup had moderately elevated risk for kidney progression. They created a GRS for beta cell dysfunction and insulin resistance based on 35 SNPs associated with insulin secretion and 20 SNPs associated with insulin sensitivity, respectively, in Asian populations. A GRS for type 1 diabetes was constructed using 9 SNPs. They found that, compared to the MOD subgroup, the participants in the MARD-II subgroup had a significantly higher GRS for beta cell dysfunction. There was no significant difference in the GRS for beta cell dysfunction between the SIRD-II and MOD subgroups and no significant difference in the GRS for insulin resistance among the 3 subgroups.

A third study investigated the aetiology of the clusters escrybed by Ahlqvist by comparing their molecular signatures (21). This was a cross sectional study of more than 12 828 Scandinavian people with type 2 diabetes. They created a GRS from 394 SNPs related to diabetes-related risks and found that a GRS for beta cell and proinsulin was associated with the SIRD cluster [β-cell, β 1.41 (95% CI −2.21 to −.62); proinsulin, −.28 (95% CI [−.41 to −.15)]. In MOD, the GRS for obesity was significantly higher compared with the other clusters [β .51 (95% CI .34–.68)].

Xu et al investigated a large cohort of Chinese people with type 2 diabetes (22). They created a GRS by using 34 SNPs associated with susceptibility to type 2 diabetes that were previously identified in East Asians. They found that every SD increase in the GRS was associated with a 12% increased risk of eGFR (95% CI 1.04-1.20, P = .001). Compared with the lowest quartile of the GRS, the second, third, and highest quartiles were associated with 15%, 19%, and 34% increased risk of reduced eGFR, respectively (P for trend = .005). Positive results, however, were not found between GRS and increased urine albumin creatinine ratio.

Rattanatham et al conducted a case-control study to investigate the effects of combined gene polymorphisms within TCF7L2, KCNQ1, and KCNJ11 on vascular complications in Thai subjects with type 2 diabetes (23). Among the people with type 2 diabetes, there were no associations for any of the 5 individual SNPs with the complications. They found, however, that a combination of 2 risk alleles in KCNQ1 [rs2237892 (C) and rs2237897 (T)] revealed significant associations. The high-GRS group was associated with increased risks of cumulative nephropathy and/or coronary artery disease [odds ratio (OR), 3.49; 95% CI 1.49-8.15, P = .004]. There was a borderline association with cumulative micro- and macrovascular complications (OR, 2.06; 95% CI .97-4.39) as compared with a group with low GRS. A combination of multiple risk alleles was subsequently investigated, from which a significant association was found only for the combination of TCF7L2 rs7903146 (C), KCNQ1 rs2237892 (C), and KCNQ1 rs223797 (T). Compared with a group with a lower GRS, the high-GRS group revealed significant association with cumulative nephropathy and/or coronary artery disease (OR, 3.92; 95% CI, 1.75 to 8.76; P = .001), and cumulative micro- and/or macrovascular complications (OR, 2.33; 95% CI, 1.13 to 4.79; P = .022).

Finally, Gurung et al studied 2 cohorts of Chinese people to determine the association of a GRS for plasma uric acid and rapid decline in kidney function in people with type 2 diabetes (24). There was no statistically significant association in each individual cohort but when analyzing the cohorts together, the authors found that a higher GRS was associated with increased odds of rapid decline in kidney function (meta-adjusted OR 1.12; 95%CI 1.01-1.24, P = .030).

There were 17 SNPs that were recurrent throughout the studies with positive outcomes (Table 2). These were KCNQ1 rs2237892, CADM2 rs13078807, ETV5 rs9816226, TCF7L2 rs7903146, MTCH2 rs3817334, NRXN3 rs10150332, MAP2K5 rs2241423, SH2B1 rs7359397, SLC39A8 rs13107325, APOB rs1367117, TNNI3K rs1514175, NUDT3 rs206936, BDNF rs10767664, FTO rs1558902, MTIF3 rs4771122, TFAP2B rs987237, and TBL2 rs17145738. The gene whose variants were most commonly used to construct a GRS was KCNQ1. Tremblay et al used 3 variants in KCNQ1, Rattanatham et al used 2, and Gurung et al used 1. Each of the SNPs used in Liao et al's study was unique and did not recur in any of the other positive studies.

Conclusions

In this systematic review, we identified 15 studies of genetic risk scores investigated worldwide between 2001 and 2022. The GRSs were designed to identify people at a genetically high risk of developing DKD, with the vast majority analyzing people with type 2 diabetes. Most studies with the strongest methodological quality assessment (n = 9) reported statistically significant and favorable findings of a GRS’s association with at least 1 measure of DKD.

This systematic review is the first to support the utility of a GRS to identify people with diabetes that are at high risk of developing DKD. In practice, a robust GRS could be used at the first clinical encounter with a person living with diabetes (Fig. 2). A report could be generated from a routine blood test, allowing the clinician a targeted approach to diabetes medication selection and frequency of surveillance. If this is done at the time of diagnosis, this would be particularly useful and allow high-risk people to be prioritized for review earlier in the course of the disease. Renoprotective agents could then be used with close specialist surveillance in order to reduce DKD progression (25-27). This could ultimately prevent or delay the development of DKD.

From the perspective of genetic architecture, CKD does not represent a single disease but rather a highly heterogenous group of pathophysiological processes. This is reflected in the studies that we reviewed showing the association of DKD with SNPs related to obesity, vascular disease, metabolic syndrome, uric acid, and beta cell dysfunction.

Our systematic review included a significant proportion of studies conducted with Asian participants (7 out of 15 studies) (12, 17, 19, 20, 22-24). In view of the high prevalence of diabetes in these populations, it is encouraging to see an interest in genetic studies that target the early identification and prevention of DKD in people of Asian descent.

Our systematic review has a number of strengths. First, we examined GRSs for DKD using different combinations of SNPs and in many different populations. Second, we used a robust search strategy in consultation with a medical librarian and extracted articles from 4 databases.

This systematic review has several limitations. First, our search was restricted to articles in English, so we might not have fully characterized global efforts at identifying GRSs for DKD. Publication bias also means that our study favors our hypothesis. Second, heterogeneity in sample size, selection of SNPs, population, diabetes type, study design, and stage of DKD existed between all studies. Substantial variability in study characteristics made comparisons between studies difficult. For example, studies that measured more than 1 outcome had a greater chance of being classified as mixed.

Furthermore, the studies varied in their definition of DKD and measurement of renal endpoints. The Kidney Disease Improving Global Outcomes defines DKD as the presence of markers of kidney damage and GFR <60 mL/min/1.73 m² for > 3 months (5). Several studies used progression of DKD (13, 19, 20, 24) or ESRD (10, 11, 13, 17, 18) as endpoints, which made direct comparisons between studies complex. Moreover, the studies did not rely on biopsy-proven diagnoses of DKD. Studies that describe biopsy-proven DKD are rare, and none could qualify for this review. Although this would increase accuracy for future GRS studies, in clinical practice biopsies are only performed when the aetiology of chronic kidney disease is unclear. Hypertension and IgA nephropathy can commonly coexist with long-standing diabetes and may confound the aetiology of DKD. This may be particularly important in certain populations like parts of China where glomerulonephritis remains the leading cause of chronic kidney disease (28). This must be taken into consideration, especially given that 7 out of the 15 studies included in this review were in Asian populations. Future GRS studies can consider renal biopsy for accurate diagnosis or exclusion of other common causes with renal ultrasound or urinalysis.

Our assessment of methodological quality highlights the opportunity for improvement in study rigor. Most studies (n = 5) were rated as good. More methodologically rigorous studies will enable firmer recommendations about the most effective GRS to identify DKD. Improvements could be made in all domains of study quality, including strengthening external validity (eg, demonstrating that a study population is representative of the source population) and ensuring that studies are adequately powered. Furthermore, the Critical Appraisal Skills Programme checklist that was used to assess methodological quality is inherently subjective. We tried to improve objectivity by integrating a score as a quantitative measure into this evaluation. It should be noted, however, that a degree of subjectivity of those reviewing the quality of the studies remains.

Our review is also limited by the number of studies available for analysis. This may be due to the relatively recent technical advancements in genetics that allows quicker and cheaper genetic typing and association studies of large cohorts of people. Furthermore, although it is relatively well understood that there is a genetic basis for the development and progression of DKD, the utility of GRSs remain elusive. The limited number of studies make interpretation of our findings less robust but also highlight the importance of further research in this area.

Research into GRSs has been fruitful and has led to the development of in vitro diagnostic tests and improved clinical practice. For instance, women in the United Kingdom are invited to start mammographic screening when they turn 47 years old, which corresponds to a 2.4% 10-year risk threshold, as this is the average risk for women at this age. According to a polygenic risk score using 77 SNPs, women in the top 10% of genetic risk reach this risk threshold in their early 30s , whereas women in the bottom 10% of the polygenic risk remain below this threshold throughout their lifetime (29, 30). Thus, information on genetic risk is more effective than an age-based criteria in guiding initiation of mammographic screening. In cardiovascular disease, a genetic risk score can identify individuals with a 4-fold increased risk for coronary artery disease. This risk is similar to monogenic conditions like familial hypercholesterolaemia (31). If a similar genetic risk score can be used to identify high and low risks for DKD, we could target screening and treatment options to those that need it most.

Consideration must also be given to the selection of SNPs for the most robust GRS for identification of DKD risk. For instance, certain studies explored an expansive set of genes (up to 598 SNPs in Tremblay et al's study), while other studies focused on only a small and specific number of genes. Our review highlighted 17 SNPs that were recurrent throughout studies that showed a positive outcome. The gene most often included in the scores was KCNQ1, which encodes the pore-forming α subunit of voltage-gated potassium channels (32) and was initially identified as a type 2 diabetes susceptibility gene (33). Further research then showed that it may also confer susceptibility to diabetic nephropathy, as evidenced by the presence of macroalbuminuria (34, 35). Whether variants in this gene would serve to enhance other GRSs for DKD requires further research.

The most recent literature suggests there is moderate evidence to support the utility of GRSs to identify people with diabetes who are at high risk of developing DKD. Further prospective research is needed to strengthen the evidence for the utility of GRSs to predict DKD in clinical practice. Once a robust GRS for DKD has been established, future research can focus on its integration into clinical workflow and its impact on clinical outcomes.

Contributor Information

Aleena Shujaat Ali, Department of Medicine, The University of Melbourne, Melbourne 3084, Australia. Australian Centre for Accelerating Diabetes Innovations, The University of Melbourne, Melbourne 3000, Australia.

Cecilia Pham, Department of Medicine, The University of Melbourne, Melbourne 3084, Australia. Australian Centre for Accelerating Diabetes Innovations, The University of Melbourne, Melbourne 3000, Australia.

Grant Morahan, Australian Centre for Accelerating Diabetes Innovations, The University of Melbourne, Melbourne 3000, Australia. Diabetes Research Foundation, The University of Western Australia, Perth 6009, Australia.

Elif Ilhan Ekinci, Department of Medicine, The University of Melbourne, Melbourne 3084, Australia. Australian Centre for Accelerating Diabetes Innovations, The University of Melbourne, Melbourne 3000, Australia. Department of Endocrinology, Austin Health, Melbourne 3084, Australia.

Funding

Diabetes Australian Research Programme (Grant ID: Y22G-ALIA).

Author Contributions

A.S.A. conducted the literature search; screened articles; extracted data; and wrote, reviewed, and edited the manuscript. C.P. screened studies for inclusion. G.M. contributed expert knowledge and edited the manuscript. E.I.E. screened studies for inclusion, contributed expert knowledge, and edited the manuscript. All authors approved the final version of the manuscript. A.S.A. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Disclosures

We declare no competing interests.

Data Availability

Original data generated and analyzed during this study are included in this published articles or in the data repositories listed in the References.

PROSPERO URL: https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=402057. PROSPERO ID: CRD42023402057.

References