The Common HNF1A Variant I27L Is a Modifier of Age at Diabetes Diagnosis in Individuals With HNF1A-MODY

Mutations in the HNF1A gene cause a common subtype of maturity-onset diabetes of the young (MODY) referred to as HNF1A-MODY (1). Although classically described as being an early-onset form of inherited diabetes (2), the age at which diabetes develops is highly variable, with individuals reported to develop diabetes across eight decades of life (4–74 years) (3).

Sex, mother’s diabetes status during pregnancy, anticipation, and mutation type and position (with respect to isoform or protein domain) have all been reported to be associated with age at diagnosis of HNF1A-MODY (4–7). Furthermore, we have previously shown that a high genetic risk score for type 2 diabetes is associated with diabetes being diagnosed earlier, although no single locus was found to have a significant effect (4). Since that study, larger genome-wide association studies (GWAS) have identified further genetic variants associated with type 2 diabetes risk. The DIAbetes Genetics Replication And Meta-analysis (DIAGRAM) consortium reported, in individuals of European ancestry, multiple independent single nucleotide polymorphisms (SNPs) (rs1169288 [I27L], rs1800574 [A98V], and rs140730081) in/near the HNF1A gene (8). Both 27L and 98V alleles have been reported to result in reduced HNF1A transactivation activity (9,10), suggesting these are the causal variants. Given the genetic and functional evidence, we hypothesized that these SNPs may modify age at diagnosis in individuals with HNF1A-MODY.

All individuals were referred to Exeter or Paris laboratories for routine genetic testing of MODY. Within each family, we genotyped the individual reported as being diagnosed with diabetes at the youngest age. Families were not included if DNA was not available from this individual. To reduce potential population stratification, individuals self-reported as non-Caucasian European were excluded. To avoid misclassification of pathogenic variants, we excluded individuals with mutations that are present in any individual within Genome Aggregation Database (gnomAD) (version 2.0) (11).

Protein-truncating variants (PTVs) were defined as follows: any premature termination codon–introducing variant (nonsense, frameshift) predicted to introduce the premature termination codon at least 50 bp upstream of the penultimate exon/intron boundary and thus predicted to elicit nonsense-mediated decay. Any variant predicted to disrupt splicing and located in the +1, +2, +4, +5, −1, or −2 positions was also defined as a PTV. Finally, any large deletion removing at least one exon was considered a PTV.

We genotyped I27L and A98V either by Sanger sequencing, KASP chemistry (LGC, Middlesex, U.K.), or using our in-house targeted next-generation sequencing panel (12). A subset of 139 samples was genotyped by two methods and demonstrated 100% (278/278) call concordance. In Exeter and Paris cohorts, subdivided by mutation type or not, I27L and A98V genotypes did not deviate from Hardy-Weinberg equilibrium (P > 0.01) (13). We were unable to genotype rs140730081 (referred to as chr12:121440833: D in ref. 8) as this SNP exists in the low-complexity sequence, and there exists no SNP in high linkage disequilibrium (r² > 0.7) with rs140730081 in the Northern and Western European ancestry population (1000 Genomes, phase 3 [14]). Given these difficulties and the lower odds ratio for this variant on type 2 diabetes risk (8), we focused our attention on the two missense SNPs.

Multiple regression analyses with age at diagnosis as the dependent and genotype, sex, and study cohort as independent variables were conducted using SPSS Statistics 24 (IBM, New York, NY). To compare PTV and missense results, equality of regression coefficients was calculated using a Z-test.

For 397 individuals with HNF1A-MODY who had undergone MODY genetic testing in Exeter, initial sex-adjusted analyses did not identify a significant effect for I27L (P = 0.48) or A98V (P = 0.70) on HNF1A-MODY age at diabetes diagnosis. As has been previously reported (4,5) females were diagnosed 2.3 years (95% CI −3.7,−1.0) earlier than males (P = 0.001), and individuals born to mothers who were diagnosed with diabetes before or during pregnancy were found to be diagnosed 4.2 years (95% CI −5.6,−2.8) earlier than those not born to mothers with diabetes during pregnancy (P = 8 × 10⁻⁹). Information regarding mother’s diabetes status during pregnancy was available only for a subset of this cohort (n = 305) and not at all for the Paris replication cohort. For this reason, diabetes status of the mother was not included as a covariate in subsequent analyses.

Given our previous results concerning the effect of mutation type (missense/PTV) on age at diagnosis (6,7), we considered the possibility that allelic heterogeneity might be confounding our analyses. Indeed, significant and distinct effects for I27L emerged when our cohort was stratified into those bearing PTV (n = 230) and missense (n = 155) mutations. In individuals with a PTV, each 27L allele was associated with a 1.5-year (95% CI −2.8, −0.5) earlier diagnosis (P = 0.007), whereas in individuals with a missense mutation each 27L allele was nominally associated with a 1.4-year (95% CI −0.1, 2.9) later diagnosis (P = 0.07) (Fig. 1). No significant effect on age at diagnosis for each 98V allele was found in individuals with either a PTV (β = 3.4 years; 95% CI −0.5, 7.2; P = 0.09) or missense mutation (β = 1.8 years; 95% CI −4.8, 1.1; P = 0.22). In contrast to previous findings (6,7), there was no significant effect for missense mutation position on age at diagnosis, either with respect to presence/absence in transactivation domain (P = 0.46) or isoform structure (P = 0.08).

We sought to replicate our study in an independent cohort of individuals with HNF1A-MODY collected in Paris, France (n = 384). When not taking into account mutation type, we again found no significant associations for I27L (P = 0.52) or A98V (P = 0.79) (Fig. 1). However in the subset of individuals with a PTV mutation (n = 214), we replicated the association between I27L genotype and age at diagnosis found in the Exeter cohort, with each 27L allele being associated with a 1.7-year (95% CI −3.1, −0.2) earlier diagnosis (P = 0.03) (Fig. 1). In the subset of individuals with a missense mutation (n = 162), we found no significant effect of each 27L allele on age at diagnosis (β = 0.9 year; 95% CI −1.4, 3.1; P = 0.44). Last, no significant effect for each 98V allele on age at diagnosis was observed in individuals with a PTV (β = 0.3 year; 95% CI −4.3, 3.7; P = 0.88) or missense mutation (β = 1.0 year; 95% CI −4.2, 6.3; P = 0.72).

We then meta-analyzed the two cohorts including study cohort (Exeter/Paris) as a covariate; individuals in the Paris cohort were diagnosed 2.5 years later than individuals in the Exeter cohort (P = 4 × 10⁻⁶). In individuals with a PTV (n = 444), each 27L allele was associated with a 1.6-year (95% CI −2.6, −0.7) earlier diagnosis (P = 0.001) (Fig. 1). No significant effect for A98V on age at diagnosis was identified (P = 0.38). In individuals with a missense mutation (n = 317), no significant effect on age at diagnosis was found for I27L (P = 0.12) or A98V (P = 0.71). The difference in the effect of 27L on age at diagnosis in individuals with PTVs and missense mutations was significant (P = 0.001).

To completely remove mutational heterogeneity as a confounder, we also looked at the effect of I27L and A98V on age at diagnosis in individuals with a G292fs mutation, the most common HNF1A PTV carried by 27% (212/781) of individuals in these cohorts. In these individuals, each 27L allele was associated with a 2.0-year (95% CI −3.6, −0.4) earlier diagnosis (P = 0.01). No significant effect for A98V was identified (P = 0.78).

Interestingly, in both Exeter and Paris cohorts, there was a weak (β = 0.05 and 0.07, respectively) but significant (P = 0.003 and 0.02, respectively) increase in age at diagnosis with each increasing year of diagnosis. This may be due to the fact that individuals diagnosed many years ago tended also to be recruited when MODY was classically described and suspected in individuals diagnosed with diabetes less than 25 years of age (15). Regression analyses including year of diagnosis had, however, little effect on the results of our study (data not shown).

For complex disease loci identified by GWAS, the presence of a gene mutated in a related monogenic form of the disease is often used to suggest causal mechanisms and infer disease biology. In this study, we hypothesized that GWAS findings for a polygenic disease could explain differences in age-related penetrance for an associated monogenic form of disease. After accounting for allelic heterogeneity, a consistent effect across two independent cohorts for I27L genotype on HNF1A-MODY age at diagnosis was identified.

Studies aiming to identify disease modifier genes often include all mutations, hypothesizing that the larger sample size outweighs the increased variation in the dependent variable (e.g., age at diagnosis) that can occur when including a diverse range of mutations (16,17). However, we have shown that the effect of I27L on modifying age at diagnosis differs significantly for individuals with HNF1A-MODY with PTV and missense mutations. We can think of at least two reasons for this, which are deserving of discussion.

First, given the number of different loss-of-function mechanisms associated with HNF1A missense mutations (loss of DNA binding, reduced transactivation activity, protein instability, defects in subcellular localization [18–20]), it is likely that there is more variation in the functional deficit associated with these mutations compared with PTVs. Genetic and functional studies support this, and although PTVs have been found only to be associated with MODY, missense variants have a spectrum of effects ranging from weak effects predisposing to type 2 diabetes to highly penetrant effects causing MODY (10,21). Our exclusion of any variant found in gnomAD (11) aimed to reduce this heterogeneity and may explain our inability to replicate the effect for missense mutation position on age at diagnosis (6,7). However, substantial variability in functional deficit associated with the missense mutations may still remain, masking any modifying effect for I27L.

Second, the existence of I27L within the dimerization domain of HNF1A and a proposed dimerization defect for L27 monomers (22,23) raises the possibility that inheriting L27 in cis to a stable missense mutation could reduce the amount of mutated HNF1A dimer and potentially increase age of diagnosis in individuals heterozygous for I27L. Conversely, inheriting L27 in trans to a stable missense mutation could increase the amount of mutated dimer, decrease HNF1A activity, and lead to a lower age at diagnosis in individuals heterozygous for I27L. Therefore without knowledge of phase and assuming an equal number of individuals with the mutation in cis and trans to I27L, such a model would predict no effect of I27L on age at diagnosis. Our limited understanding of how each missense mutation affects HNF1A function and whether I27L does affect dimerization means time-consuming efforts to phase mutations and I27L are not currently appealing.

Studies looking at the effect of other variants in HNF1A on MODY phenotype would seem to be an avenue worth pursuing, and on the basis of our findings analyses, stratifying with respect to mutation type would seem prudent. For example, in PTV carriers, one would predict that a significant effect on HNF1A activity will occur only when the variant is in trans to the mutation (as the mutated allele is likely to be degraded by nonsense-mediated decay). To detect a significant modifier effect phasing of variant with mutations is likely to be very important for SNPs rarer than I27L. Indeed, the lack of phasing in this study may partly explain the lack of significant effect for A98V (minor allele frequency = 2.9%, gnomAD, European Non-Finnish [11]). Parental samples were available for only 3 of 23 individuals heterozygous for A98V with a PTV mutation, and we could confirm that the 98V allele was inherited in trans to the PTV mutation in at least one case. Given this and the lack of association between A98V and age at diagnosis in our unphased analysis of PTV carriers, if an effect does exist, we are likely underpowered to detect it. Rarer HNF1A SNPs with much larger effects on type 2 diabetes susceptibility have also recently been identified (10); the prevalence and phase of these variants with mutations in our cohorts could be of considerable interest.

Last, the current study should motivate studies looking into the effect of type 2 diabetes–associated SNPs at other monogenic diabetes loci (e.g., GCK, HNF1B, HNF4A) on respective MODY phenotype. The variable expressivity associated with mutations in these genes means these studies could be of particular clinical relevance. Furthermore, after accounting for genetic variation at monogenic diabetes loci, genome-wide studies will have increased power to uncover other modifier loci that are likely to be of significant interest with respect to understanding disease pathophysiology and avenues of therapy.

Acknowledgments. The authors would like to thank Gwendoline Leroy, Séverine Clauin, and Christelle Vaury, Assistance Publique-Hôpitaux de Paris, for technical support and the sequencing of the Paris cohort.

Funding. This article presents independent research funded by Medical Research Council grants to L.W.H. (MR/J006777/1) and M.N.W. (MR/M00507/1) and supported by the National Institute for Health Research Exeter Clinical Research Facility.

The views expressed are those of the authors and not necessarily those of the Medical Research Council, National Health Service, National Institute for Health Research, or the Department of Health.

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

Author Contributions. J.M.L. conceived, designed, and conducted the experiments; analyzed data; interpreted the results; and wrote the manuscript. C.S.-M., T.W.L., A.R.W., and S.A.S. conceived, designed, and conducted experiments; analyzed data; interpreted the results; and reviewed and edited the manuscript. K.A.P., S.E., C.B.-C., A.T.H., L.W.H., and M.N.W. contributed to the design of the experiments and reviewed and edited the manuscript. J.M.L. 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.

Prior Presentation. This work was previously presented in poster form at the European Association for the Study of Diabetes Study Group on Genetics of Diabetes (EASD-SGGD), Leiden, the Netherlands, 11–13 May 2017.