Genetics of Monogenic Diabetes: Present Clinical Challenges

The original clinical criteria to define MODY included an age of onset below 25 years, not requiring insulin treatment and having a generational family history of diabetes [10]. Since this first description, however, the criteria have proven to be insensitive in many confirmed cases [2•], as well as increasingly non-specific. This is unsurprising with the rising prevalence of young-onset type 2 diabetes that is, at diagnosis, non-insulin requiring like MODY and typically occurs in the setting of a strong family history. Indeed, in some cases of adult-onset type 1 diabetes, the slow progression to beta-cell failure may not require insulin treatment immediately [11]; thus, the lack of a need for insulin is not particularly discriminatory for monogenic diabetes. Additionally, a family history of diabetes is often encountered in many people with any type of diabetes, again, due to the rising prevalence of type 2 diabetes. In addition it is recognised that > 20% of children with confirmed type 1 diabetes have an affected first-degree relative after decades of follow-up [12]. Age at onset below 25 years is also an arbitrary cut-off and, although commoner in younger adults, confirmed MODY mutations can present into the 5th decade of life [13].

In ethnic groups with a higher prevalence of young-onset type 2 diabetes, the discriminatory value of these criteria are even poorer. Indeed, in one study of UK south Asian individuals referred for genetic testing, although the south Asian referrals were more likely to meet the clinical referral criteria than white individuals, the detection rate of MODY in those genetically tested was less than half of the white group [14•].

The MODY probability calculator offers a more standardised approach to select individuals for genetic testing [15]. The online calculator combines broad clinical information to predict the probability of testing positive for MODY. When compared to traditional criteria, the prediction model improved the sensitivity and specificity for identifying MODY. However, the model was developed and validated in the same cohort of white individuals of European descent. There is a need for validation in other cohorts and some early studies in east Asians have shown good sensitivity but reduced specificity [16].

With non-specific clinical criteria, attention has turned to the use of biomarkers that assist the selection of individuals for genetic testing. C-peptide, pancreatic auto-antibodies, lipid profiles and high-sensitivity C-reactive protein (CRP) have variable discriminatory value, but also have several limitations [17‐22]. Pancreatic autoantibodies are positive close to diagnosis in ~ 80% of people with type 1 diabetes if glutamic acid decarboxylase (GAD) and insulinoma antigen-2 (IA2) antibodies are measured [19], although titres may decrease with duration of diabetes, and a negative result does not therefore exclude type 1 diabetes. People from some ethnic groups with type 1 diabetes have been shown not to have detectable auto-antibodies at diagnosis, and so again the absence of pancreatic auto-antibody positivity does not preclude a diagnosis of type 1 diabetes.

C-peptide levels, a marker of endogenous insulin production, have also been used to help segregate people with type 1 diabetes (where C-peptide is low or undetectable) from those with MODY. However, this is not useful at diagnosis of diabetes, where C-peptide may be detectable in early type 1 diabetes, and even in long duration type 1 diabetes approximately 8% of people have stimulated C-peptide levels well above thresholds considered typical for type 1 diabetes [23]. Similarly, individuals with severe hyperglycaemia due to either monogenic or type 2 diabetes may have undetectable C-peptide levels due to transient beta-cell glucotoxicity.

Practically, a combination of clinical and biomarker approaches is likely to yield the highest sensitivity. In the US SEARCH study, youths diagnosed with diabetes below the age of 30 years were selected for genetic testing if pancreatic auto-antibody testing was negative and fasting C-peptide >0.8 ng/ml, and yielded a detection rate of 8% in those tested. In the Young Diabetes in Oxford study, a pick-up rate for MODY of 15% was observed in those with type 2 diabetes diagnosed before 45 years without features of insulin resistance or 10% of those with type 1 diabetes from individuals selected on the basis of antibody negativity and preserved C-peptide [24]. The UK UNITED study demonstrated a MODY prevalence of 3.6% using a biomarker approach in those diagnosed < 30 years with diabetes, a detectable urine C-peptide and negative antibodies. In this study, the MODY probability calculator also missed 55% of cases identified using the biomarker approach [25]. Recently, the Norwegian childhood diabetes registry was used to estimate MODY prevalence in all the antibody-negative children in the registry, which has a high case ascertainment. A total of 4.1% of individuals had likely pathogenic or pathogenic variants in the commonest MODY genes (HNF1A, HNF4A, HNF1B, GCK or INS) [3]. The biomarker approach is likely to result in more individuals being selected for testing. However, since the cases identified in those tested appear to be less likely to meet clinical criteria or are missed by the MODY probability calculator, testing more individuals may be warranted [2•].

When antibodies are negative and C-peptide levels are preserved, the clinical picture could be consistent with type 1, type 2 or monogenic diabetes. In these cases, assessing probability of type 1 or type 2 diabetes on the basis of polygenic risk, so-called genetic risk scores, may have some additional discriminatory value. A genetic risk score for type 1 diabetes that incorporates key human leukocyte antigen (HLA) susceptibility loci [26] has been shown to facilitate discrimination of type 1 from monogenic diabetes [27] and type 2 diabetes [28]. However, as with other approaches, this was validated in white European populations and evidence suggests ethnic-specific variations will be required [29].

In practice, misclassification of all diabetes types may occur and healthcare practitioners need be alert to this possibility [30, 31]. The application of a systematic approach to individuals who are newly diagnosed, particularly young adults and in those from non-white ethnic groups, could assist classification of common forms of diabetes and identify those in whom molecular investigation would be beneficial. Using biomarkers in this strategy is worthwhile as long as the limitations of these specialist tests are appreciated.

One such database is the Genome Aggregation Database (GnomAD), which contains data from over 100,000 exomes and 15,000 whole-genomes. GnomAD reports minor allele frequency (MAF) depending on ancestry. These databases can be used to assess whether a variant found in a dominant gene might be disease-causing. For example, the presence of a variant in more individuals than would be predicted from the population prevalence of MODY (50–100 per million [9]), assuming complete penetrance, would indicate that the variant is not acting as an autosomal dominant pathogenic mutation, since it would be present in people without disease.

However, knowledge of MAF is not a magic bullet. Each exome in GnomAD contains an average of 7.6 rare variants (MAF < 0.1%) in Mendelian disease genes, which suggests tolerance to genetic variation is higher than previously assumed. Such data have been used to downgrade pathogenicity status in variants that have previously been thought to be disease-causing [52].

The spectrum of low-frequency variation (MAF < 1%) in the seven commonest genes associated with MODY was examined in 4003 individuals from population studies [53]. Although 1.5% of those studied harboured genetic variants that had been previously reported to cause MODY, the majority were still euglycaemic through to middle age. This study was an early indication that careful consideration in assigning pathogenicity to variants in dominant genes, is needed.

Species conservation measures the degree to which amino acid properties at any given position are evolutionarily conserved [51]. Essential sites for protein function are likely to be invariant across species (highly conserved), whereas a variant sites across species can, by definition, accommodate a degree of substitution. Therefore, missense variants encountered at highly conserved sites are more likely to be pathogenic [54‐57].

Using web-based applications, missense variants can be classified based on the predicted effects of the variant on protein function. The prediction obtained from a single method should not be taken in isolation. Broadly speaking, four main approaches exist: those based on sequence conservation methods; those assessing impact on protein sequences; protein structure; and physicochemical properties and splicing-predictions [51, 58].

Software programs such as SIFT (J. Craig Venter Institute, La Jolla CA) [59, 60], Align-GVGD (International Agency for Research on Cancer hosted by the World Health Organization [61, 62]) and PolyPhen-2 [63, 64] can combine a variety of assessments to predict pathogenicity.

Despite these aligned approaches, it is still often difficult to confidently assign pathogenicity. In these cases, further work may be necessary to shed light on variant functionality. These may be clinical studies in the individual or family members, or by use of reliable in vitro functional or biochemical assays that assess the effect of the variant in a cellular system model [48]. Even then, confident assignment of pathogenicity may not always be possible due to the limitations of these techniques.

A series of assays have also been described to investigate transcription factor variants, such as those in HNF1A, based on published studies [65‐67]. For the HNF1A gene, which has been studied extensively, these include an assessment of transactivation potential (the transcriptional activity of HNF1A), protein production, nuclear-cytosolic localisation and DNA-binding in cellular systems.

In vitro functional studies have also been successfully deployed in cases of neonatal diabetes to support precision-based treatment. In individuals with KCNJ11 mutations causing neonatal diabetes, the success of sulphonylurea treatment is determined by the specific mutation, as not all affected channels are sensitive to blockade. Knowledge of specific mutant effects in vitro can therefore help clinicians decipher whether to try higher doses to achieve full blockade of the channel, or whether insulin is needed [68, 69].

Guidelines recommend collaboration with molecular laboratories in cases where functional effects of variants using established techniques prove inconclusive, but where there is real clinical impact in assigning status. However, this may be challenging for clinicians to access readily as most MODY gene variants have not been studied in this way and accessible pathways to support these efforts routinely may need to be developed.

In practice, for people with monogenic diabetes, treatments can be trialled empirically with or without confirmed mutation status, but the frustration and anxiety in failing to receive a conclusive diagnosis in those with variants of unknown significance is high and the psychological burden remains unexplored.

In many cases, the clues towards pathogenicity come from studying the proband and their family members in more detail. For novel variants, familial co-segregation studies that demonstrate the variant in relatives with diabetes, but absence in those without diabetes, is compelling evidence in favour of pathogenicity. Similarly, demonstrating sulphonylurea sensitivity in those with uncertain variants in HNF1A or HNF4A can also be very helpful.

Aside from the common MODY genes described, a number of other genes have been implicated as causing diabetes. These include the beta-cell transcription factors IPF1 and NEUROD1, and PAX4, KLF11 and BLK. The genetic evidence for these genes causing penetrant autosomal dominant monogenic diabetes is weak and initial findings have not been borne out from large-scale sequencing panels of genes.

Extremely rare mutations in other genes have also been identified. For example, deletions shortening variable number tandem repeats in the CEL gene appear to cause diabetes and pancreatic exocrine dysfunction [70] and specific mutations in POLD1 can result in severe insulin resistance syndromes [71]. Loss of function variants in APPL1 have also been identified in two families from an exome sequencing study. Co-segregation studies are promising, but it will remain to be seen whether APPL1 variants will be reported in other MODY [72].

Understanding the phenotypic spectrum of variants from benign to disease-causing in dominant genes is also key. Common variants in KCNJ11 and GCK can lead to small changes in glucose homeostasis, whilst less common variants leading to a more severe functional impact cause MODY and neonatal diabetes. Another example is the low-frequency E508K variant in HNF1A, which in the Latino population appears to increase risk of type 2 diabetes [73] rather than causing monogenic diabetes. A similar finding has been observed in the Wolfram syndrome gene WFS1, which causes a MODY-like phenotype in a small number of families [74], but common variation in the gene is also associated with risk of type 2 diabetes.

As new data are gathered, revisiting existing genes can also be helpful. Protein truncating variants in the beta-cell transcription factor RFX6 have now been reported to cause MODY with low penetrance [75]. Previously, however, only homozygous mutations in RFX6 had been implicated in neonatal diabetes with gut and gallbladder anomalies and the role as a MODY-gene was questionable due to variable penetrance [76]. Protein truncating variants robustly co-segregate with diabetes, but with an increased age of onset compared to HNF1A-MODY, as only 27% had developed diabetes by age 25, compared to 55% of HNF1A-MODY cases [75].

The management of monogenic diabetes is based predominantly on observational data. The main application of personalised medicine is in the use of sulphonylurea (SU) agents in HNF1A/HNF4A–MODY and in neonatal diabetes caused by mutations in K-ATP channel components. This is supported by a small randomised controlled trial of gliclazide versus metformin in HNF1A-MODY [77] and in observational data collected in cases of neonatal diabetes. Therefore, people with mutations in HNF1A or HNF4A causing MODY can be managed with low-dose sulphonylurea therapy, and those misdiagnosed may therefore potentially stop insulin injections or have tablet regimens rationalised. These precision-based treatments not only achieve good glycaemic control, but there is compelling evidence demonstrating that they are superior to conventional approaches [8]. There are case reports of decades of successful treatment on SU therapy in HNF1A-MODY, although clinicians also, not infrequently, see cases who do not respond well despite possessing an HNF1A variant that is clearly diabetes-causing (personal observation). The reason for this variation in treatment response remains unclear. However, a recent study has suggested that higher HbA1c, higher BMI and longer duration of diabetes at the time of transfer to SU therapy from insulin, may predict failure of SU monotherapy, making the case for early genetic diagnosis [78].

A study of liraglutide use in HNF1A- and GCK-MODY examined both clinical response and mechanistic aspects [79]. This study showed no significant difference in glucose lowering effect between liraglutide and SU treatment, but more hypoglycaemia with SU use, suggesting that the SU dose used was too high. The additional expense and inconvenience of injection therapy does not appear to justify routine GLP1 therapy in HNF1A-MODY, and it is unknown how effective these agents would be in the context of long duration MODY with secondary SU failure. Most clinicians use metformin or a dipeptidyl peptidase-4 inhibitor as second-line treatment although there is no particular evidence base. The effect of sodium-glucose transporter-2 inhibitor agents in HNF1A-MODY, who already have decreased expression of SGLT2 and low renal threshold, was shown in a single dose study to induce greater glycosuria than in type 2 diabetes [80] and it is unknown whether this would lead to greater adverse effects or increased efficacy of the agents.

Permanent neonatal diabetes (PND) caused by mutations in KCNJ11 or ABCC8 can be managed with high-dose sulphonylurea therapy in > 90% of cases, negating the need for insulin therapy and associated complications in affected neonates [68]. For neonatal diabetes, a recent case series of 10-year follow-up of individuals with KCNJ11 or ABCC8 variants treated with SU agents reported that 93% of cases remained on SU therapy alone, achieving glycaemic targets and with a good safety profile [81].

GCK MODY, characterised by lifelong, non-progressive fasting hyperglycaemia, requires no pharmacological therapy and, following diagnosis, affected individuals may be able to stop all treatment. In the largest longitudinal study to date of people with GCK mutations, there is no higher burden of clinically significant microvascular or macrovascular complications, compared to normoglycemic individuals [82].

Most other forms of MODY, neonatal diabetes and mitochondrial diabetes are characterised by insulin deficiency and therefore, in most cases, individuals progress to insulin therapy. Those with mitochondrial diabetes tend to have increased lactate levels, which has raised concerns over metformin use, although there is no evidence that there is an increased prevalence of lactic acidosis, since metformin only appears to cause lactic acidosis in the setting of severe renal dysfunction [83]. The pragmatic consensus is that other oral agents should be used in preference to metformin, particularly in those with neurological features.

The drive to deliver precision-based approaches in diabetes undoubtedly underpin many of the recent advances in monogenic diabetes, although there is still much to achieve. A critical step, as discussed, is to develop and validate strategies for finding cases of monogenic diabetes that have transethnic applicability.

Supporting clinical practitioners to understand and deliver often complex genetic information to people with diabetes, is also a key area that lacks firm guidance. As accessibility to sequencing technologies expands, particularly in understudied populations, there is a potential harm from poor interpretation of variant pathogenicity that needs to be addressed. Close collaboration between clinical practitioners and centralised laboratories that undertake these activities may be helpful in this context.

Moving beyond monogenic diabetes and understanding the impact of common variants on type 2 diabetes risk, phenotype and treatment effects, is likely to be the next clinically meaningful direction for the lessons learnt from monogenic diabetes. Efforts to address this unmet need are underway and are very likely to change the approaches to management of common diabetes types [84].