Introduction

Type 2 Diabetes (T2D) is a common, complex disease affecting 1 in 10 Americans. It is often a diagnosis of exclusion centered on elevations in a single biomarker: blood glucose or its close derivative, glycated hemoglobin (HbA1c), and it is typically diagnosed in adulthood, after age 451. Independently, the gene GCK is also closely tied to blood glucose levels. GCK encodes for a hexokinase expressed largely in pancreatic beta cells, liver hepatocytes, and glucose-sensitive neurons. GCK gates glucose and insulin regulation through glucose sensing and processing, ultimately playing a key role in setting and maintaining glucose homeostasis for the body and controlling appetite. Both rare loss of function and a subset of damaging missense variants in GCK (pathogenic GCK variants) have been shown to dampen or reduce the enzymatic functions of the protein, resulting in an elevated endogenous glucose set point which, on average, sits just above the clinical diagnosis threshold for T2D2,3,4,5.

Pathogenic GCK variants have near complete penetrance for clinically recognized glucose elevations, although most individuals in the general population harboring these variants are unaware of their genotype and its association with plasma glucose levels6,7,8,9. If elevated glucose is first measured in childhood or early adulthood, a “maturity onset diabetes of the young” (MODY) diagnosis and a genetic etiology, such as GCK, may be considered10. If first measured in adulthood, a clinical T2D diagnosis is more likely to be assigned, as monogenic etiologies are not typically suspected or tested for in this age range11,12. Strong associations have been observed between individuals harboring GCK variants and clinical T2D diagnoses in both T2D-specific cohorts, where up to 1% of enrolled patients harbor pathogenic GCK variants, as well as in gene-disease association studies of all-comer, clinico-genomic cohorts, which often use a broader definition for damaging variants8,12,13,14,15,16,17. Further, penetrance estimates of pathogenic GCK variants for T2D are in line with some of the most well established CDC Tier-1 gene-disease associations, such as BRCA1/BRCA2 with Breast and Ovarian Cancers and LDLR/APOB with Familial Hypercholesterolemia13.

Despite this strong, penetrant association between variant carriers and T2D diagnosis, early work focused on outcomes suggests that those harboring pathogenic GCK variants and standalone mild, stable hyperglycemia (GCK-MODY) are not at increased risk for the typical array of long-range complications associated with T2D, especially microvascular complications14,18,19,20,21,22. In addition, treating patients with GCK-MODY with first-line pharmacological interventions such as insulin or oral hypoglycaemic agents (metformin and glibenclamide) has been shown to have little to no effect on HbA1c levels, and it may be difficult or impossible to reduce glucose levels below standard risk thresholds18,23. Taken together, the excess of T2D diagnoses in individuals harboring pathogenic GCK variants at the population level may represent some degree of misdiagnosis of T2D in adulthood for what is actually asymptomatic, mild, and stable hyperglycemia (GCK-MODY) in these individuals11,18,24.

Untangling benign glucose elevations from accurately assigned T2D diagnoses in those with damaging GCK variants represents a great opportunity for precision medicine12. The deviations in both disease progression and treatment response in studies of GCK-MODY patients compared to T2D patients at large suggest that standalone natural glucose elevations should be considered alongside a clinical T2D diagnosis in those harboring pathogenic GCK variants. These findings, though, stem largely from targeted studies comparing MODY subtypes and focusing specifically on those with isolated mild, stable hyperglycemia, excluding GCK variant carriers with a T2D diagnosis from the analyses9,17,19.

To formulate a population screening strategy for GCK, a retrospective evaluation of T2D rates and T2D-related complications of both pathogenic variant carriers and any additional damaging variant carriers in all-comer cohorts is needed. In this study we: 1) develop and validate a population screening variant annotation strategy, with a focus on missense variants, leveraging the highly penetrant elevations in glucose and Hba1c level as a readout, and 2) characterize differences in T2D diagnosis rates, disease progression in terms of secondary complications, and the alignment between these diagnoses and long-term complications in those carrying pathogenic GCK variants and those carrying any additional GCK variants predicted to elevate glucose in multiple large, unselected, adult population cohorts. We find that GCK variant carriers are not only at increased risk for developing T2D, but also GCK carriers with T2D develop secondary complications of T2D at similar rates to the T2D population at large.

Methods

Subjects and genetic data

All participants gave their informed, written consent prior to participation. All data used for research were deidentified. The UK Biobank (UKB) study was approved by the North West Multicenter Research Ethics Committee, UK. UKB data are available for download (https://www.ukbiobank.ac.uk/) to qualified researchers. Access to the UKB dataset for this study was obtained under Application Number 40436. The Helix cohorts were reviewed by Salus IRB (Reliance on Salus for all sites) and approved (approval number 21143), the WCG IRB for myGenetics (Western Institutional Review Board, WIRB-Copernicus Group) and approved (approval number 20224919), the MUSC Institutional Review Board for Human Research for In Our DNA SC and approved (approval number Pro00129083), and the University of Nevada, Reno Institutional Review Board for Healthy Nevada Project and approved (approval number 7701703417).

We utilized the UKB population level exome OQFE pVCFs for 470,000 individuals (field 23157, with genotypes set to missing when DP < 7 for SNVs and <10 for indels, and variants excluded if there were no homozygotes or the max allelic balance was <0.15 for SNVs or <0.2 for indels as per25) as well as the imputed genotypes from GWAS genotyping (field 22801-22823).

We also utilized 62,406 samples that were sequenced and analyzed at Helix using the Exome+® assay as previously described, recruited from the Healthy Nevada Project (sequenced Jan 2018 to Mar 2023); myGenetics (sequenced May 2022 to Mar 2023); and In Our DNA SC (sequenced Dec 2021 to Mar 2023) (Table 1)26. Genomic data were processed using a custom version of Sentieon and aligned to GRCh38, with variant calling and phasing algorithms following GATK best practices.

Table 1 Study population demographics
Full size table

Phenotypes

UKB data were provided from the UKB resource (accessed September 2022). ICD codes (both ICD-9 and ICD-10) and associated dates were collected from inpatient data (category 2000), cancer register (category 100092), and the first occurrences (category 1712), which records the earliest instance of a diagnosis from the Primary Care data, Hospital inpatient data, Death Register records, and self-reported medical conditions mapped to an ICD-10 code at a three character resolution (i.e., E11 instead of E11.0). UKB glucose values (data field 30450) were converted from mmol/L to mg/dL, and HbA1c (data field 30750) was converted from mmol/mol to %. Covariates used include genetic sex (data field 22001), BMI (Kg/m2) at assessment (data field 21001), which was also used for obesity analyses, European ancestry (data field 22006) as well as the date of birth (data field 33), which was used to calculate current age as of 2023. Date of death (data field 40000) along with date of birth was used to calculate age at death, where applicable, for DCSI analyses. Analyses were performed both retrospectively, using all individuals with available data, or prospectively, looking forward from the baseline assessment visit (between years 2006–2010) for those individuals who were both free of a T2D diagnosis as well as any self-reported use of the T2D-related drugs insulin and metformin (any of data fields 6177 “Insulin”, 6153 “Insulin”, 2986, or 20003 “insulin product or metformin”) at this time point.

Helix cohort phenotypes were processed from Epic/Clarity Electronic Health Records (EHR) data as previously described and updated as of January 202326. ICD codes (both ICD-9 and ICD-10 CM) were collected from available diagnosis tables (from problem lists, medical histories, admissions data, surgical case data, account data, claims, and invoices). The data were sourced from EHR data formatted using the OMOP CDM v5.4. To gather records from this data model, ICD source code was mapped to a source concept id, and the source concept id was used to extract the relevant diagnoses from the conditions table (Supplementary Data 1). Lab values were extracted from OMOP measurement tables using the following concept ids: glucose (3033408, 3031266, 3000483, 3004501, 3002666,3035729); HbA1c (3005673,3004410), all values were converted to mg/dL and %, respectively. For each measurement, median values were used for all analyses. Covariates used include genetic sex, median BMI (concept id 3038553), as well as date of birth, which was used to calculate current age as of 2023.

To measure T2D-related complications, we calculated the Diabetes Complications Severity Index (DCSI), a 14-point scale calculated from ICD codes covering major complication categories27,28. For each of the seven DCSI categories, we used a comprehensive list of diagnosis codes including ICD9, ICD10, and ICD10CM codes (Supplementary Data 1)27. For a given DCSI category, we applied the weighted score to each unique instance of the underlying diagnosis code and then transformed the total category score to 0, 1, or 2, where any summed category score over 1 was assigned a maximum score of 2. The overall DCSI score was calculated as the sum of all the transformed category scores. For the time-to-event analysis, we calculated the age at when each DCSI score was attained (for scores 1–5), based on diagnosis records.

Annotation and qualifying GCK variants

Variant annotation was performed with VEP-10429. Coding regions were defined according to Gencode version GENCODE 33, and the MANE / Ensembl canonical transcript ENST00000403799.8 was used to determine variant consequence30,31. Variants from the exome were restricted to CDS (coding sequence) regions plus essential splice sites. Genotype processing for Helix data was performed in Hail 0.2.115-10932c754edb32. Genotype processing for UKB data was done using Plink.

Population variant interpretation method: Variants were classified as LoF if the consequence was start_lost, splice_donor_variant, frameshift_variant, splice_acceptor_variant, or stop_gained and MAF < 0.1% in all ancestry populations in gnomAD and the analyzed cohorts. For coding variants classified as missense_variant, inframe_deletion, or inframe_insertion, stop_loss, non-benign (by Polyphen or SIFT), and MAF < 0.1%, we also used a combination of a functional evidence map (urn:mavedb:00000096-a), with scores for each variant assigned using a yeast complementation assay, and a statistical evidence map created using the Power Window technique (Supplemental Data 2, Supplementary Data 3) to predict variants that likely increased glucose levels33,34,35,36. Briefly, Power Window is a sliding window analysis that groups variants located near each other into one unit and analyzes them together to improve power, much like a gene-based collapsing analysis but at a smaller scale36. We trained the Power Window model for GCK on glucose measurements from ~330k individuals (n = 333,190) from the UKB, used a beta cutoff of 0.5 to build the model, and tested on glucose measures from ~112k (n = 112,015) unrelated UKB individuals (Fig. 1a, b).

Fig. 1: Identifying individuals with glucose-elevating GCK variants in the population.
figure 1

a GCK statistical evidence map for random blood glucose, created using Power Window (PW), with PW+ regions marked in pink and PW- in green. Variants are plotted by position and colored according to their functional score annotation (Screen+ in pink (functional score <0.66), Screen- in green (functional score >1.18), Screen= in yellow (functional score between 0.66 and 1.18), and ScreenX in gray (no results available from functional screen)). See Methods for details on the Power Window technique. b Comparison of random blood glucose levels in individuals with and without GCK variants. Individuals are categorized according to whether their GCK variant was predicted by the functional evidence map to cause high glucose levels as in panel a. Individuals with LoF variants or no variants are shown separately. The dashed lines indicate the median values for those without GCK variants (89 mg/dL) or those with LoF variants (122 mg/dL). The colors indicate Power Window results: pink (PW + ) is those with variants predicted to cause high glucose levels, green (PW-) for low glucose, gray (PW=) for no clear signal, and yellow (PWX) for variants that did not qualify for PW analysis, meaning variants that were MAF > 0.1% or were not predicted by Sift or Polyphen to be damaging. The dataset shown is restricted to the 117k UKB individuals used to test PW. For display purposes, the plot y-axis was cut off at 300 mg/dL, which excludes 163 individuals from the ‘no variant’ category and 1 from the ‘Screen = ’ category. c Assessment and median HbA1c levels by GCK variant group, for the subset of individuals with these readings in UKB (n = 432,826) and Helix (n = 21,744) cohorts, respectively. The first two panels focus on participants from the UKB cohort. The first panel shows glucose-elevating coding (pink, n = 330) and LoF (blue, n = 35) carriers separately, the second panel separates glucose-elevating coding carriers by pathogenicity annotations; with the HbA1c levels of those with P/LP variants (blue, n = 84) in line with levels in those with LoF variants (n = 35). The final panel displays HbA1c levels of genotypic groups used for analyses in this work. The median HbA1c levels for these groups in UKB were 5.4% for no variant (green, n = 432,515), 5.9% only glucose-elevating risk (pink, n = 187), and 6.4% for pathogenic (P/LP) group (blue, n = 124); in Helix cohorts they were 5.6% for no variant (green, n = 21,717), 6.0% for only glucose-elevating risk group (pink, n = 14), and 6.4% for pathogenic group (blue, n = 13). For brevity, the following individuals were excluded from their respective plots: 995 UKB and 207 Helix participants with HbA1c values >=10%. Gray lines for prediabetic (5.7%) and diabetic (6.5%) clinical HbA1c thresholds are shown for reference. The white dot in the middle of the violin represents the median of the data, the thick black bar in the center of the violin indicates the interquartile range, and the body represents the distribution of data points in each group. See Supplementary Fig. 1 for glucose measures by genotype.

Full size image

Traditional variant interpretation method: To determine which variants have enough evidence to garner a likely pathogenic (LP) or pathogenic (P) classification via the ACMG/AMP interpretation guidelines, we developed a curation pipeline to gather and score available variant information via this rubric, updated with specific recommendations from the ClinGen Monogenic Diabetes Expert Panel Specifications(VCEP)(Version 1.3.0) for GCK-MODY37. The detailed interpretations, including the assessment of each of the criteria for each variant, are available in Supplementary Data 3 and the detailed step by step process taken for the interpretation is in Supplementary Data 4. In brief, for a first pass, all variants matching those previously curated as LP or P by the VCEP as well those matching “high-confidence” LP and P in Clinvar (multiple submitters, no conflicts including one interpretation from a well-established diagnostic lab, accessed: 02/15/24) were automatically assigned a P/LP annotation. For all other variants, an automated application of the ACMG interpretation rubric was applied. PVS1 was assigned to LoF variants per VCEP specifications. Grantham distances were calculated for all variants and used to assign PM5 (supporting or strong) criteria to novel missense, using VCEP and high confidence Clinvar P/LPs as the list of established P/LP sites for the interpretations. PS1 was applied to any variant resulting in the same amino acid change at the same site as an established P/LP variant. PM2 was assigned for variants absent from Gnomad as well as those below the popmax FAF threshold of 0.0003%. Variants falling in established mutational hotspots of the Glucose and ATP binding domain were assigned PM1 (Moderate). Variants with REVEL scores > 0.7 and splice variants with maximum SpliceAI scores >0.2 were assigned PP3. Finally, functional evidence PS3 (Moderate) was assigned to variants with activity scores >0.66 from functional evidence map urn:mavedb:00000096-a. No variant-by-variant literature search was performed and criteria specific to phenotype were not considered. Criteria were summed per guidelines to determine LP/P variants.

Statistical analysis

Statistical analyses were run using the statsmodel package in Python 3.7.3. For binary variables, logistic regression or Firth logistic regression was used; for quantitative variables, linear regression was used after rank-based inverse normal transformation; and for count data, negative binomial regression was used. The lifelines package was used for time-to-event analyses including cumulative incidence plots and cox proportional hazard calculations. For time-to-event analyses, the earliest age at relevant diagnosis (T2D or DSCI phenotype) or current age (in 2023) was determined for each participant38. The main analyses were performed on all ancestries together because we are analyzing very rare variants (MAF < 0.1%) collapsed together. Results are similar in the subset of the UKB with genetic similarity to Europeans, these analyses are documented in Supplementary Fig. 313,26,39. Code to reproduce main analyses in this work can be found in Zenodo (code: https://doi.org/10.5281/zenodo.12774460)40.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Results

Identification of individuals with glucose-elevating GCK variants

Annotation strategy and carrier prevalence by cohort

At the population level, 733 individuals in UKB and 131 individuals in HRN harbor rare, damaging (MAF < 0.1%) coding and LoF variants in GCK. To identify the subset of carriers with variants with pathogenic properties, we first applied a traditional variant interpretation strategy. We started with Pathogenic and Likely Pathogenic (P/LP) variant curations from the Monogenic Diabetes Variant Curation Expert Panel (VCEP) and high-confidence P/LP variant curations from the Clinvar database37. For the remaining variants without previously-established high confidence annotations, we applied the ACMG/AMP rubric tailored for GCK-MODY from the VCEP (see Methods, Supplementary Data 2, 3 for variant curations and detailed interpretation plan). Beyond the standard VCEP recommendations, for PS3-Moderate points, we also leveraged a recently published functional evidence map, which evaluates the functional consequence of all missense and nonsense single nucleotide variants at the GCK locus via a yeast complementation assay35. In this work, the authors show alignment in their scores with results from the well-accepted in vitro measures of catalytic efficiency calculations currently used for PS3 determinations. Using this interpretation strategy, 57 variants (seen in 126 individuals) in the UKB and 18 variants (seen in 24 individuals) in Helix receive a P/LP classification, but the majority of individuals harbor rare coding variants that remain unclassified.

In an attempt to more comprehensively classify rare, damaging variants seen at the population level, we built a population screening variant annotation strategy that is more universal and more feasible to automate than the ACMG/AMP criteria. This method focuses on resources that can annotate all potential coding variants to identify individuals with variants in GCK that are predicted to elevate glucose levels, a highly penetrant characteristic of the MODY phenotype. Aside from variants with a molecular consequence consistent with LoF, we combined the output from the aforementioned variant-level functional evidence map with a statistical evidence map for regions in the gene associated with glucose elevations, akin to the PM1 hotspot criteria. We built the GCK statistical evidence map with clinicogenomic data from 333,190 randomly selected UKB participants, assessing statistical power-based windows of rare variants across the GCK locus against random blood glucose (i.e., not necessarily measured after fasting) collected as part of a baseline assessment visit (Fig. 1a). We formed our population screening variant annotation strategy by assessing the combination of these functional and statistical annotations against the random blood glucose levels in the remaining UKB participants (n = 112,015, “replication subset”). We found that coding variants that are statistically significantly associated with increased glucose levels compared to those without GCK variants could be reliably identified if: (i) the statistical evidence map predicted the variant to increase glucose levels AND (ii) the functional evidence map did not predict the variant to decrease glucose levels. Those with GCK variants that did not meet this criteria, on average, displayed glucose levels in line with those without GCK variants (Fig. 1b)36,41.

Next, we applied this population screening annotation strategy across all individuals in the UKB and Helix cohorts, identifying 337 individuals in the UKB cohort and 62 in the Helix cohort who were heterozygous and no homozygotes for glucose-elevating coding variants in GCK. This represents ~0.07% (1/1400) of UKB and ~0.1% (1/960) of Helix cohorts, respectively, or ~50% (337/696 in UKB and 62/128 in Helix) of all individuals with rare, damaging GCK coding variants present in each cohort. Additionally, there were 37 individuals in UKB and 3 in Helix cohorts with GCK LoF variants. In total, we identified 374 individuals in UKB and 65 in Helix cohorts with LoF or glucose-elevating coding variants in GCK (0.082% or 1/1200 in UKB and 0.10% or 1/960 in Helix). This is higher than the previously reported population prevalence for UKB (0.042%) and another American cohort (0.024%), where coding variant interpretation was limited to a more traditional variant interpretation strategy8. Indeed, while the vast majority of variants with criteria for a pathogenic or likely pathogenic (P/LP) classification were called glucose-elevating variants using our population screening interpretation strategy (54/58 in UKB and 15/18 in Helix), carriers of this group of variants only represent ~35% of all glucose-elevating variant carriers found using our annotation strategy (126/374 in UKB and 24/68 in Helix).

Functional and statistical data used for population screening annotation as well as data for ACMG/AMP variant interpretations for all variants in both cohorts, including all well-established pathogenic and benign variants, are available in Supplementary Data 3. The current age, average age of T2D diagnosis, and a breakdown by sex and genetic similarity for individuals with or without a glucose-elevating coding or LoF variant are presented in Table 1.

Individuals with glucose-elevating variants in GCK have elevated HbA1C levels, often into clinically relevant ranges

Across the UKB cohort, we find HbA1c levels are higher for those harboring glucose-elevating GCK variants compared to individuals without glucose-elevating GCK variants, (e.g.,the mean HbA1c levels in those with glucose-elevating variants in the UKB was 6.0%, compared to 5.4% for those without variants; see Fig. 1c, Supplementary Fig. 1 for glucose). Further, we used standard clinical cutoffs for random glucose and/or HbA1c to classify individuals as either pre-diabetic (HbA1c > =5.7%) or diabetic (HbA1c > =6.5% or random glucose above 200 mg/dL). For UKB, 68% of those with glucose-elevating coding variants had values in the prediabetes range compared to 21% of those without glucose-elevating GCK variants. Additionally, 19% of those with glucose-elevating variants had values in the diabetes range, compared to 4% of those without glucose-elevating GCK variants (Fig. 1c, Supplementary Fig. 1 for glucose). We further examined a subset of the UKB cohort who did not have a T2D diagnosis and had not reported taking common T2D medications, insulin or metformin, at the baseline assessment (subcohort counts: glucose-elevating=240, no GCK variant= 371,040, Supplementary Fig. 2). Elevated glucose and HbA1c trends for those harboring glucose-elevating GCK variants held within this undiagnosed subcohort, with 63% of those with glucose-elevating coding variants in the pre-diabetic range (HbA1c > =5.7%), compared to 16% of those without glucose-elevating GCK variants.

Next, we subsetted individuals harboring glucose-elevating coding variants by their ACMG/AMP pathogenicity classifications from our traditional variant interpretation assessment (84 with ACMG/AMP pathogenic or likely pathogenic (P/LP) annotations, 193 with only glucose-elevating risk, and 60 with B/LB annotations) and compared mean HbA1c levels of these groups. On average, all three groups of glucose-elevating carriers had significantly higher HbA1c levels compared to HbA1c levels those without glucose-elevating GCK variants (P/LP carriers: HbA1c = 6.4%, only glucose-elevating risk carriers: HbA1c = 5.9%, B/LB carriers: HbA1c = 5.7, p < 0.001 for all groups by linear regression). These trends held when the cohort was limited to those with genetic similarity to Europeans (Supplementary Fig. 3a). Despite elevated levels compared to controls, for the remainder of the study, glucose-elevating carriers with B/LB ACMG/AMP pathogenicity annotations (n = 60) were removed due to conflicting interpretations by ACMG/AMP and screening interpretation methods. ACMG/AMP P/LP and LoF variant carriers were combined to form a ACMG/AMP pathogenic variant carrier group (n = 126) and the remaining carriers with other glucose-elevating variants formed an elevated-glucose risk group (n = 193). For all analyses, these carrier groups are compared to those without a glucose-elevating GCK variant (n = 453,619) (Fig. 1c).

For a large proportion of the Helix cohorts, random glucose or HbA1c measurements were available from the medical records, taken throughout the course of medical care and often at multiple timepoints. We expected that, unlike the UKB cohort where the measurements were taken for every participant at a single point in time regardless of health history, we would observe higher measurement values due to an enrichment for individuals at risk for diabetes with any measurement. Indeed, taking the max value observed for an individual during their medical history, and the same cutoffs as above, 69% and 20% of those with glucose-elevating coding variants and glucose or HbA1c measurements were pre-diabetic or diabetic, respectively, while 44% and 7% of those without variants were in the prediabetes or diabetes range, respectively. Aligning with trends in the UKB cohort, mean HbA1C measures for ACMG/AMP pathogenic (n = 24), glucose-elevating risk (n = 33), and without a glucose-elevating GCK variant (n = 62,349) groups were: 6.3%, 6.2%, and 5.8%, respectively (Fig. 1c, Supplementary Fig. 1 for glucose).

Individuals harboring pathogenic and glucose-elevating GCK variants are at elevated risk for receiving T2D diagnosis at any given time

Overall rates of T2D diagnosis differed by cohort (9% in UKB vs 12% in Helix Cohorts, p = 1.3e−293 by logistic regression after controlling for age, sex, BMI, Table 1). We next compared T2D diagnoses more specifically for those harboring either ACMG/AMP pathogenic or glucose-elevating GCK variants compared to those without a glucose-elevating GCK variant, using a time to event analysis with all available EHR data points. Despite varying population-level rates of T2D, we find similar T2D diagnosis enrichment between variant groups in both cohorts. Compared to those without glucose-elevating GCK variants, individuals harboring glucose-elevating variants are approximately 3-4 times more likely to be diagnosed with T2D at any given time (HR = 3.8, p = 1.38e−19 for UKB and HR = 2.4, p = 0.04 for Helix, by cox proportional hazard) and those harboring ACMG/AMP pathogenic variants are roughly 10 times more likely to be diagnosed with T2D at any given time (HR = 11.1, p = 3.13e−86 for UKB; HR = 9.4, p = 1.36e−13 for Helix, by cox proportional hazard) (Fig. 2a). Trends remain when the UKB cohort is restricted to those with genetic similarity to Europeans (Supplemental Fig. 3b).

Fig. 2: T2D diagnosis prevalence differs by GCK genotype in two cohorts.
figure 2

a Cumulative incidence plot for T2D diagnosis by genotype (No variant in blue, Glucose-elevating GCK in pink, and Pathogenic GCK in green) in UKB (dotted lines) and Helix (solid lines) cohorts. For those with Glucose-elevating variants vs. without, HR = 3.8, p = 1.38e−19 for UKB and HR = 2.5, p = 0.04 for Helix. For those with pathogenic GCK variants vs. without, HR = 11.1, p = 3.13e−86 for UKB; HR = 9.4, p = 1.36e–13 for Helix. Hazard ratios are for the subset of each cohort with both BMI and sex available (n = 452,319 for UKB and n = 45,060 in Helix, using the Cox Proportional Hazard model). b Box plot of the median, interquartile range, and overall range of age at T2D diagnosis made after age 45 by genotype for UKB (light gray, no variant n = 38,870; glucose-elevating n = 44; pathogenic n = 58) and Helix (dark gray, no variant n = 5927; glucose-elevating n = 4; pathogenic n = 9) cohorts. Diagnosis after age 45 makes up 90% of the UKB diagnoses and 60% of the Helix ones. The median diagnosis age is marked as a dotted line, age = 64 for UKB and age = 63 for Helix. Focusing on diagnoses made after age 45, T2D diagnosis happens earlier in those with glucose-elevating and pathogenic variants (UKB: pathogenic 3 years, p = 0.001 and glucose-elevating 2 years, p = 0.05; Helix: pathogenic 2 years, p = 0.05 by linear regression).

Full size image

Focusing on individuals with T2D, both ACMG/AMP pathogenic and glucose-elevating variant carriers are less likely to be obese (defined as a BMI > = 30) compared to those without glucose-elevating variants, which is in alignment with the well-described MODY phenotype (ORs 0.29 and 0.46 for ACMG/AMP pathogenic and glucose-elevating carriers compared to those without glucose-elevating variants, p < 0.005 for both groups by logistic regression controlling for age, sex, and cohort).

While MODY is typically diagnosed in childhood or early adulthood, with glucose elevations present at birth, the majority of T2D diagnoses in individuals harboring ACMG/AMP pathogenic variants (58/67 in UKB and 9/14 in Helix) and those with glucose-elevating variants (44/47 in UKB and 4/8 in Helix) were first made in adulthood, after age 45. After removing all T2D diagnoses made prior to 45 years of age, we observed a small, significant earlier age of diagnosis for those with either ACMG/AMP pathogenic or glucose-elevating GCK variants compared to those without a glucose-elevating GCK variant in UKB and for those with ACMG/AMP pathogenic GCK variants in Helix cohorts (UKB: pathogenic 3 years, p = 0.001 and glucose-elevating: 2 years, p = 0.05; Helix: pathogenic 2 years, p = 0.03 by linear regression controlling for age, sex, BMI; Fig. 2b). Finally, in terms of clinical risk stratification, by age 70, in UKB, 67% of individuals carrying ACMG/AMP pathogenic variants and 30% of individuals carrying glucose-elevating variants have a T2D diagnosis. In Helix, 91% of ACMG/AMP pathogenic carriers and 41% of glucose-elevating variant carriers have a T2D diagnosis. Rates for all carrier groups exceed their relevant population averages (12% for UKB and 20% for Helix by age 70) (Fig. 2a).

DCSI as a tool to quantify T2D outcomes from EHR records

While individuals with ACMG/AMP pathogenic and glucose-elevating GCK variants are enriched for T2D diagnoses, the diagnosis alone is an incomplete representation of the clinical impact of the disease. GCK-MODY research suggests that not all individuals harboring ACMG/AMP pathogenic GCK variants are truly diabetic in the sense of their likelihood to develop microvascular and macrovascular complications12,18,19. In order to assess whether individuals with ACMG/AMP pathogenic or glucose-elevating GCK variants diagnosed with T2D had different outcomes compared to other patients with T2D, we used the adapted Diabetes Complication Severity Index (DCSI), a 14-point scale, derived from the ICD codes obtained from patient medical records, organized into seven disease categories that are considered secondary complications of T2D (see methods for details, Supplementary Data 1 for ICD codes and severity scores for each category)27,28.

As a control, we first assessed DCSI scores in all participants of the UKB and Helix cohorts. We find that DCSI captures T2D-related outcomes from available medical records in both cohorts. Average DCSI scores are higher in Helix cohorts compared to UKB, aligning with the differing population rates of T2D (mean DCSI Helix= 0.98; mean DCSI UKB = 0.69, for T2D rates see: Fig. 2a, Table 1). Despite differences in mean DCSI in each cohort, DCSI scores increase with age across all participants in each cohort, with the highest scores being held almost exclusively by patients with diabetes, as expected for a cumulative score of progressive disease (Fig. 3a, b Supplementary Fig. 4a, b). Visualized as time to event curves, accumulation of substantive DCSI scores (DCSI > = 3) follow with recorded age of onset of T2D diagnosis, as well as with the number of years since diagnosis for those with T2D (Fig. 3C, Supplementary Fig. 4c). Additionally, further validating the clinical relevance of this scale in this population, we find that DCSI scores correlate with mortality within both those with and without a recorded T2D diagnosis (Supplementary Fig. 4d for UKB).

Fig. 3: Exploration of the Diabetic Complications Severity Index (DCSI) in two cohorts.
figure 3

a Fraction of individuals with a T2D diagnosis with 95% confidence intervals, for each possible DCSI value (0 to 14) in the UKB (yellow) and Helix (blue) cohorts b Box plot of the median, interquartile range, and overall range of DCSI values by age group for both the UKB (yellow) and Helix (blue) cohorts (see also Supplementary Fig. 4a, b, separated by T2D). Both panels a and b are calculated across all participants (n = 453,938 in UKB; n = 62,406 in Helix) c Cumulative incidence plots with shaded confidence intervals for substantial DCSI outcomes (DCSI >= 3) as an endpoint in individuals with T2D in UKB (n = 40,414, in yellow) and Helix (n = 7237, in blue) cohorts, bucketed by age at diagnosis ( < 55 (solid line), 55–65 (dashed line), >65 (dotted line)). Substantial DCSI outcomes begin to accumulate earlier for those with earlier diagnosis in both cohorts, when visualized as a cumulative incidence curve.

Full size image

For our analyses, we assess both cumulative DCSI scores for each individual as well as a dichotomous cutoff of a DCSI of three or more for time to event analyses, which can only be obtained once someone is harboring a diagnosis from at least 2 complication categories (see methods for details).

Individuals with pathogenic and glucose-elevating GCK variants and T2D develop T2D complications at rates similar to the T2D population at large

To determine if ACMG/AMP pathogenic and glucose-elevating GCK variant carriers with T2D diagnoses are also at some level of elevated risk for T2D-related complications, we performed a time to event analysis, controlling for sex and BMI, and using age at DCSI of three or more as an endpoint. For both variant groups, those with a T2D diagnosis are significantly more likely to develop more than one category of T2D complications as measured by DCSI than were population controls without a T2D diagnosis (pathogenic UKB (n = 67) HR = 2.4, p = 0.005; pathogenic Helix (n = 11) HR = 7.3, p = 0.00007 and glucose-elevating UKB (n = 46) HR = 2.8, p = 0.002; glucose-elevating Helix (n = 5) HR = 1.5, p = 0.6, by cox proportional hazard) and accumulate complications at a rate similar to that of those without glucose-elevating variants and T2D (non-GCK UKB HR = 2.45; Non-GCK Helix HR = 3.3, by cox proportional hazard) (Fig. 4a, b). Conversely, those with either ACMG/AMP pathogenic or glucose-elevating GCK variants and without a T2D diagnosis had a rate of complications in line with that of population controls without T2D (pathogenic UKB (n = 59) HR = 1.74, p = 0.26, glucose-elevating UKB (n = 146) HR = 0.87, p = 0.75; pathogenic Helix (n = 7) no carriers with endpoint, glucose-elevating Helix (n = 18) HR = 1.6, p = 0.5, by cox proportional hazard) (Supplementary Fig. 5a, b).

Fig. 4: T2D diagnoses are associated with the accumulation of DCSI secondary complications in those with pathogenic and glucose-elevating GCK variants at a rate similar to that of those with T2D in the population at large.
figure 4

a Cumulative incidence plot with shaded confidence intervals depicting the proportion of individuals with substantial DCSI outcomes (DCSI > = 3) by age for the UKB cohort, grouped by individuals with pathogenic GCK variants and T2D (green line) (HR = 2.4, p = 0.005, n = 67) and glucose-elevating GCK variants with T2D (pink line) (HR = 2.8, p = 0.002, n = 46). Proportions for those with T2D and without a GCK variant are shown in blue. Hazard ratios are calculated using a Cox Proportional Hazard model, with those without glucose-elevating variants and without T2D (yellow line) as a reference group. b) Cumulative incidence plot with shaded confidence intervals depicting the proportion of individuals with substantial DCSI outcomes (DCSI > = 3) by age for the Helix cohort grouped by individuals with pathogenic GCK variants and T2D (green line) (HR = 7.3, p = 0.00007, n = 11). The glucose-elevating group is not plotted due to the low sample size (HR = 1.5, p = 0.6, n = 5). Proportions for those with T2D and without a GCK variant are shown in blue. Hazard ratios are calculated using a Cox Proportional Hazard model with those without glucose-elevating variants and without T2D (yellow line) as a reference group. c) Hazard Ratio (HR as dots; +/− 95%CI as bars) to develop a DCSI score of at least 3 for each cohort, subsetted by GCK genotype and T2D diagnoses, compared to population controls without T2D and Hazard Ratio (HR as dots; +/− 95%CI as bars) for at least 2 in each DCSI secondary complication area, compared to population controls without T2D in either cohort. Hazard ratios are generated from the Cox Proportional Hazard model controlling for the most recent age, sex, BMI and cohort. For display purposes, 3 categories in which there were no cases are filled in with HR of 0.1. As fewer than 5 cases were expected in these groups based on population frequencies of each phenotype and 0 cases were seen, the hazard ratios are in the range of 0.000006–0.00001 and the CIs are too large to plot well. DCSI outcomes are similar regardless of GCK status for the complication categories of peripheral vascular, cardiovascular, and neuropathy and appear depleted in those with GCK variants for ophthalmic and cerebrovascular, and are variable by GCK variant group for nephropathy (too few in the metabolic category to analyze).

Full size image

Combining data from both cohorts, we find that both obese and non-obese individuals with glucose-elevating variants and T2D remain at risk for developing T2D-related complications compared to obese or non-obese controls, respectively (Obese: HR = 3.4, p = 0.00001, Non-Obese: HR = 2.13, p = 0.0045, by cox proportional hazard), suggesting T2D-related secondary outcomes are not solely driven by obesity (Supplementary Fig. 6a, b).

After confirming multi-category T2D complication trends, we next assessed if either carrier group showed enrichment or depletion for each DCSI subcategory with a reasonable case count: cardiovascular, peripheral vascular, cerebrovascular, neuropathy, neurology, and ophthalmic (retinopathy). For this analysis, we combined the clinicogenomic data from both the UKB and Helix cohorts, added an additional cohort covariate to our statistical analysis, and used age at the maximum subcategory score of 2 as the endpoint (for an analysis of the UKB cohort restricted to those with genetic similarity to Europeans, see Supplementary Fig. 3c).

We found that ACMG/AMP pathogenic and glucose-elevating GCK variant carriers with T2D diagnoses appeared just as likely as individuals with T2D and without GCK variants to have complications classified as peripheral vascular (pathogenic HR = 3.7, p = 0.06 and glucose-elevating HR = 4.3, p = 0.04 compared to HR = 4.23 for those without GCK variants), cardiovascular (pathogenic HR = 1.7, p = 0.01 and glucose-elevating HR = 1.6, p = 0.05 compared to HR = 1.75 for non-GCK), and neuropathy (MODY HR = 3.9, p = 0.002 and glucose-elevating HR = 2.2, p = 0.2 compared to HR = 3.4 for non-GCK), and appeared less likely to develop ophthalmic or cerebrovascular complications (Fig. 4c). Surprisingly, nephropathy-related complications varied by genotype, showing similarly significant enrichments in ACMG/AMP pathogenic variant carriers with T2D and those without glucose-elevating variants and T2D (pathogenic HR = 3.3, p = 0.02; non-GCK HR = 3.4, p = 0.0), while glucose-elevating variant carriers with T2D showed no enrichment against population controls without T2D (glucose elevating HR = 1.7, p = 0.5).

Taken together, our retrospective analysis of DCSI outcomes suggests that T2D diagnoses made in those harboring both ACMG/AMP pathogenic and glucose-elevating GCK variants are not merely misdiagnoses related to their natural elevations in blood glucose and HbA1c. Individuals with glucose-elevating GCK variants are at increased risk of a T2D diagnosis, and when diagnosed, develop an array of T2D-related complications throughout the lifecourse, especially in relation to peripheral vascular, cardiovascular, and neuropathy complications, at rates similar to the T2D population at large, even after controlling for age, sex, and BMI.

Discussion

Damaging variants in GCK elevate blood glucose and HbA1c with high penetrance. In this work, we set out to build a population annotation strategy to globally identify all GCK coding variants that produce these elevations. To accomplish this we leveraged the nearly 500,000 UKB participants with exome sequencing data and random glucose measurements taken as part of a baseline health assessment. We evaluated glucose levels with regard to GCK variant information in a sliding window statistical analysis in training and test sets from the UKB as well as variant-level GCK enzymatic functional scores from a yeast-based screening assay. Next, we applied our GCK population genotyping strategy to the entirety of the UKB and an additional American cohort, composed of participants from three health systems (Helix cohorts). The glucose-elevating GCK variants identified included both pathogenic variants by traditional variant interpretation as well as variants that do not reach this interpretation threshold. Combined with LoF variants, our GCK genotyping strategy identified ~1/1000 individuals from two large, unselected population cohorts carrying glucose-elevating GCK variants, a rate higher than previously reported using only a more traditional variant interpretation strategy for coding variants or LoFs for annotation8.

In line with findings from clinical studies focused on pathogenic variants, those harboring glucose-elevating GCK variants using our annotation strategy have elevated HbA1c compared to population averages, often in the prediabetic and diabetic range, with average glucose and HbA1c levels highest in the group of individuals harboring pathogenic GCK variants (Fig. 1c). Aligning with the results from other gene and disease association studies, we find that T2D diagnosis risk is high for glucose-elevating variant carriers, with diagnoses 10 times more likely in those with pathogenic variant and 3 times more likely in those with other glucose-elevating variants compared to population controls and often with earlier onset (Fig. 2).

Somewhat unexpectedly, we also found that individuals with either pathogenic or other glucose-elevating GCK variants and a T2D diagnosis have a rate of T2D-associated complications, quantified using the Diabetes Complication Severity Index (DCSI), similar to that of non-GCK individuals with T2D and nearly three times that of population controls (Fig. 4a, b). The DCSI assesses diabetes complications in seven areas: cardiovascular, peripheral vascular, cerebrovascular, nephropathy, neurology, ophthalmic (retinopathy), and metabolic. Although the rarity of metabolic complications precluded us from completing analyses in this area, we found that certain other DCSI categories were more similar between individuals with T2D with and without glucose-elevating GCK variants than were others (Fig. 4c). In particular, the hazard ratios for developing cardiovascular, peripheral vascular, and neurology complications was statistically significant in those with both T2D and either pathogenic or glucose-elevating GCK variants and similar to those found for individuals with T2D but without GCK variants. In contrast, there appeared to be no enrichment of ophthalmic or cerebrovascular complications in individuals with T2D and either pathogenic or glucose-elevating GCK variants compared to the general population at all. This finding is in line with prior studies showing a lower incidence of retinopathy in those with pathogenic GCK variants compared to those with T2D but without pathogenic GCK variants6,19. Interestingly, the risk for nephropathy-related complications differed by GCK variant group– individuals with T2D and harboring pathogenic variants developed nephropathy complications at a rate similar to non-GCK T2D cases, while individuals with T2D and other glucose-elevating variants showed no enrichment for this phenotype. More studies are needed to confirm these possible allelic trends.

On the surface, it may appear as if our results oppose other studies describing the GCK-MODY phenotype as an isolated, stable hyperglycemia18,19. Instead, it is important to consider the combination of penetrance of both T2D diagnosis and diabetic outcomes in individuals harboring relevant GCK variants. For example, in terms of lifetime risk, 90% of those harboring pathogenic variants have a T2D diagnosis by age 80, but only 50% of those with pathogenic variants who receive a T2D diagnosis will reach a DCSI outcome of at least 3 by age 80. While this represents a substantial enrichment compared to the general population (25% to reach DCSI >= 3 by age 80), it means many individuals with pathogenic GCK variants live their life with high glucose, a T2D diagnosis, and few diabetic complications. This distinction is more obvious at the level of T2D diagnosis in the group of individuals with other glucose-elevating GCK variants, as those without a T2D diagnosis do not appear to be at an increased risk for T2D complications and instead accumulate secondary complications at the same rate as non-diabetic controls (Supplementary Fig. 5). Similar to trends in more complex forms of T2D, obesity appears to be a contributing factor to both diagnosis and secondary outcomes, although it cannot fully explain the risk observed in either variant carrier group. Other factors, such as polygenic risk and lifestyle, also warrant further investigation here.

Taken together, those with glucose-elevating GCK variants and a T2D diagnosis do develop T2D complications as measured by DCSI at rates higher than the population at large and would likely benefit from specific clinical care. Our retrospective results suggest that a prospective population screening strategy for GCK should consider the naturally elevated blood glucose levels across those with glucose-elevating GCK variants while also appreciating that many of the patients maintain a high-risk profile for T2D and secondary complications. For individuals with glucose-elevating GCK variants and a T2D diagnosis, a more nuanced management plan may be required, as one tied to lowering glucose levels may not work for this group of patients. Follow-up studies and controlled clinical trials are needed in this space.

This study was limited by the single baseline assessment for blood labs, BMI, and survey questions administered in the UKB. The amount of follow-up also differs among participants in both the UKB and Helix cohorts. Additionally, not everyone in the Helix cohorts had glucose or HbA1c measurements available. Having measurements at multiple time points that started earlier in life would allow for better classification of individuals and more nuanced risk assessment. T2D diagnoses and secondary outcomes were assessed via diagnosis codes from the EHR. While EHR data may miss some subtleties in phenotypes of individuals, these phenotypes have been shown to be largely accurate in validation studies42 and provide the large benefit of being unbiased and done similarly for both cases and controls.

Finally, our study assessed those with pathogenic coding and LoF GCK variants separately from other glucose-elevating variants. At this time, past these average glucose elevations and their pathogenic interpretation via traditional variant interpretation, the root molecular difference between coding variant carriers in the two groups is unknown, but may, at a high level, reflect different underlying molecular mechanisms of disease of damaging coding variants. For example, GCK-MODY disease pathogenesis is typically understood via a LoF mechanism of disease, in which dampened or lowered enzymatic activity slows down overall glucose utilization and ultimately drives a higher endogenous glucose setpoint5. Additional molecular consequences of coding variants, such as changes that alter dynamics with interacting partners and effects on glucose uptake and glycogen regulation in the liver may also drive glucose elevations or elevated T2D risk43,44. A more thorough assessment of these glucose-elevating variant groups, potential interactions with other genetic risk factors, and follow-up work in additional cohorts may help to better elucidate these patterns and ultimately could offer even more precise diagnoses and understanding of disease progression and long-term complications in those harboring GCK variants. That said, our results show that all glucose-elevating variant carriers are at increased risk for both receiving a T2D diagnosis as well as secondary complications in those with T2D.

GCK has been lauded as a prime candidate for precision medicine. Most of the discussion has centered on using genetics to identify the subset of individuals harboring damaging GCK variants and stand-alone stable, mild hyperglycemia, who may not benefit or require treatment for their condition. Overall, we find that those harboring glucose-elevating GCK variants are at elevated risk for T2D diagnosis, and these individuals develop secondary outcomes at a rate similar to that of individuals without GCK variants who are diagnosed with T2D. Stand-alone glucose elevations in those harboring damaging GCK variants is not equivalent to a T2D diagnosis, as a subset of individuals harboring glucose-elevating variants have clinically elevated glucose without an increase in secondary complications. Population genetic screening for glucose-elevating GCK variants may represent an important tool for assessing future diabetes risk and ultimately making formal diabetes diagnoses, building care plans targeted to specific complications, and providing confidence to providers to eliminate the need for anti-diabetic therapy in those for which it is not warranted.