De Novo Mutations in EIF2B1 Affecting eIF2 Signaling Cause Neonatal/Early-Onset Diabetes and Transient Hepatic Dysfunction

Permanent neonatal diabetes mellitus (PNDM) is a genetically and clinically heterogeneous condition diagnosed before the age of 6 months. A genetic cause is identified in 82% of cases, resulting in improved treatment in almost 40% (1).

Thirty-nine percent of patients with PNDM have a genetic etiology resulting in development of at least one extrapancreatic feature, alongside diabetes (1). The most common PNDM syndromic subtype is Wolcott-Rallison syndrome, which is caused by autosomal recessive mutations in the EIF2AK3 gene. Individuals with Wolcott-Rallison syndrome usually develop diabetes in the first year of life. They also present with repeated episodes of hepatitis-like liver dysfunction, which eventually results in fatal liver failure, and skeletal dysplasia. The EIF2AK3 mutations in these patients are thought to result in disruption of the unfolded protein response and cell death due to unregulated endoplasmic reticulum (ER) stress (2,3). Approximately 15% of individuals with syndromic PNDM do not have a mutation in one of the known causative genes, suggesting the existence of novel genetic subtypes (4).

Identifying novel genetic causes of PNDM is crucially important to gain insights into human β-cell development, function, and survival and could identify potential targets for novel therapies.

We used a trio-based genome sequencing approach to identify heterozygous de novo EIF2B1 mutations as the cause of a novel genetic syndrome characterized by PNDM/early-onset diabetes and transient liver dysfunction.

Individuals with PNDM and individuals with early-onset diabetes were recruited by their clinicians for genetic analysis in the Exeter Molecular Genetics Laboratory. The study was conducted in accordance with the Declaration of Helsinki, and all subjects or their parents gave informed consent for genetic testing.

Genome sequencing was performed on DNA extracted from peripheral blood leukocytes of 44 probands diagnosed with PNDM before the age of 6 months and their unaffected, unrelated parents. Samples were sequenced on an Illumina HiSeq 2500 with a mean read depth of 30. The sequencing data were analyzed using an approach based on the gatk (Genome Analysis Toolkit) best practices guidelines. Briefly, the reads were aligned to the hg19/GRCh37 human reference genome with BWA-MEM and Picard was used for duplicates removal and gatk IndelRealigner for local realignment. gatk HaplotypeCaller was used to identify variants, which were annotated using Alamut Batch 1.8 (Interactive Biosoftware, Rouen, France). Variants that failed the QD2 VCF filter or had five or fewer reads supporting the variant were excluded. Copy number variations (CNVs) were called by SavvyCNV (5), which uses read depth to predict copy number.

Replication studies were performed in a cohort of 188 patients diagnosed with diabetes before 2 years of age in whom the 25 known genetic causes of neonatal diabetes had been excluded. Patients were analyzed using a targeted next-generation sequencing assay, including the known monogenic diabetes genes and additional candidate genes, such as EIF2B1 (NM_001414.3). Variant confirmation and cosegregation in family members were performed by Sanger sequencing.

EIF2B1 exon 3 (NM_001414.3) was amplified using in-house designed primers (F, TGTTCACTGATGTATCCCTAGCA, and R, TCCTAGGAAGAAAAGAGCAAACT). PCR products were sequenced on an ABI3730 capillary machine (Applied Biosystems) and analyzed using Mutation Surveyor, version 3.98 (SoftGenetics). The bioinformatics tools SIFT, PolyPhen-2, and Align GVGD were accessed through the Alamut Visual software (Interactive Biosoftware) to predict the effect of the variants.

Variants are reported using the Human Genome Variation Society (HGVS) nomenclature guidelines (6).

The predicted structure of the EIF2B1 stop-loss variant, c.915_916del, p.(*306Thrext*12), was modeled using the Phyre2 Web server in intensive mode 3, with the full-length sequence of variant eIF2Bα as input. The resulting structure was modeled entirely on PDB (Protein Data Bank) entry 3ECS (crystal structure of human eIF2Bα 4), except for the COOH-terminal extension, which was modeled ab initio. All structures were visualized in PyMOL (Molecular Graphics System v2.0; Schrödinger LLC).

EIF2B1 mutation details have been deposited in the Decipher database (https://decipher.sanger.ac.uk/). All other data sets generated and/or analyzed for this study are available from the corresponding author upon reasonable request.

EIF2B1 was the only gene identified with de novo variants in at least three patients. The three missense EIF2B1 variants identified, p.(Gly44Asp), p.(Gly44Val), and p.(Ser77Asn) (Table 1), were not listed in the GnomAD database (7). All affected residues are highly conserved across species (up to zebrafish for p.Gly44, up to chicken for p.Ser77). No deletions or duplications involving EIF2B1 were detected.

Sequencing of EIF2B1 in 188 further patients without a known genetic diagnosis who had developed diabetes before the age of 2 years identified two additional novel heterozygous EIF2B1 variants, p.(Leu34Trp) and p.(*306Thrext*12). Parental testing showed that both variants had arisen de novo.

The clinical features of the five subjects with de novo EIF2B1 mutations are summarized in Table 1. All had been diagnosed with permanent diabetes between the ages of 4 and 56 weeks and were treated with full replacement insulin doses (range 0.46–1.00 units/kg/day). All four patients for whom data were available had low birth weight (<20th centile, SDS range −0.84 to −1.74) consistent with reduced insulin secretion in utero.

Four of the five patients had suffered episodes of hepatic dysfunction in childhood with hepatitis-like derangement of liver enzymes. The severity of these episodes was variable: for case subject 3, the first episode resulted in fatal acute liver failure, while the other three cases resolved with return to normal liver function and enzymes. Case subject 1 presented once with elevated liver enzymes at the age of 38 months. Case subjects 4 and 5 experienced multiple episodes requiring hospitalization between the ages of 5 months and 5.5 years but have not experienced any further episodes (current ages 13 and 18 years).

No severe neurological features were reported; however, case subject 4 has a mild learning disability and case subject 5 has attention deficit disorder.

EIF2B1 encodes for the α-subunit of the heterodecameric eIF2B complex, which modulates the activity of the eukaryotic translation initiation factor 2 (eIF2) to regulate translation initiation and protein synthesis in basal and stress cell conditions. Under cell stress, serine 51 of eIF2α is phosphorylated by kinases of the integrated stress response (ISR) (8). This phosphorylation results in altered interaction with eIF2B to inhibit the rate of translation of most mRNAs, with the exception of stress response gene RNAs, which are able to circumvent this block to allow cell recovery (9,10). The recent publication of the structures of human eIF2B in complex with both phosphorylated and nonphosphorylated eIF2 (11,12) showed that in its nonphosphorylated form, eIF2α makes contact with the eIF2Bβ, γ, and δ subunits—but not with eIF2Bα. However, upon Ser51 phosphorylation, eIF2α binds tightly to eIF2Bα, resulting in a markedly different orientation of eIF2 in the complex with eIF2B (Fig. 1A).

Mapping the heterozygous EIF2B1 variants from our five case subjects shows that these all lie at the same surface as that occupied by phosphorylated eIF2α (Fig. 1B). Moreover, one of these variants (identified in case subject 3) occurs at p.Ser77 of eIF2Bα, which makes both hydrogen-bonded and nonbonded contacts with residues of phosphorylated eIF2α. In the stop-loss variant identified in case subject 5, the COOH-terminal extension to eIF2Bα is likely to sterically hinder the interaction of eIF2α with eIF2Bα p.Leu305 as well as possibly occluding other parts of the interface (Fig. 1C).

We report the identification of EIF2B1 heterozygous de novo variants in five patients with PNDM/early-onset diabetes and transient liver dysfunction. The variants, four missense and one stop-loss, are all predicted to map to the same protein surface, and in silico modeling suggests that they are likely to disrupt the interaction of the EIF2Bα subunit (encoded by EIF2B1) with phosphorylated EIF2α during the cell ISR.

Our patients’ clinical features are markedly different from those reported in patients with leukoencephalopathy with vanishing white matter (VWM), a rare pediatric neurological disease caused by recessive loss-of-function EIF2B1 mutations (13). VWM is a progressive condition, usually fatal in childhood. Extraneurological features are generally not present, although diabetic ketoacidosis at 8 months was reported in one patient with a homozygous p.(Leu49Arg) variant in EIF2B1 (14). None of the five patients we report had severe neurological features; however, case subject 4 has a mild learning disability and case subject 5 has attention deficit disorder. The parents of individuals with VWM, who are heterozygous carriers for EIF2B1 loss-of-function mutations, are unaffected, thereby supporting our hypothesis that the mutations identified in our patients do not result in complete loss of function of the eIF2B complex.

eIF2B regulates translation initiation and protein synthesis by modulating eIF2 activity. During protein synthesis under normal conditions, eIF2B acts to recycle initiation factor eIF2 for further rounds of translation by promoting its dissociation from GDP, allowing replacement by GTP and subsequent association of GTP-bound eIF2 with the initiator methionyl-tRNA. Under stress conditions, specific kinases phosphorylate the serine 51 of eIF2α (8) resulting in inhibition of GDP dissociation, attenuating recycling of eIF2 to its active GTP-bound form through altered interaction with eIF2B and consequently inhibiting the rate of translation of most mRNAs (9,10) (Fig. 2A).

Mapping the heterozygous EIF2B1 variants identified in the five case subjects we report shows that these all localize at the same surface as that occupied by phosphorylated eIF2α (Fig. 1B). Moreover one variant, p.(Ser77Asn), identified in case subject 3, occurs at a residue that directly interacts with phosphorylated eIF2α. Two mutations, identified in case subjects 1 and 2, involve eIF2Bα p.Gly44. Previous experiments have shown that mutation of the equivalent residue in GCN3, the yeast ortholog of EIF2B1, results in full cell viability under normal conditions but failure to respond to eIF2α phosphorylation under stress (15). In contrast, most EIF2B1 variants that have been reported in patients with VWM occur at interfaces with other eIF2B subunits (Fig. 1D), consistent with their reported effects on the formation and/or stability of the eIF2B complex (16). Taken together, this evidence strongly suggests that the EIF2B1 variants identified in our patients are likely to result in formation of the eIF2B complex and support translation under normal conditions but impair eIF2B binding to phosphorylated eIF2α under stress conditions.

Impaired cellular stress response is a known β-cell dysfunction mechanism and is involved in other genetic subtypes of PNDM, including Wolcott-Rallison syndrome, the most common syndromic form of the disease (17). Wolcott-Rallison syndrome is characterized by PNDM/early-onset diabetes, severe liver dysfunction (presenting with transient episodes of acute liver failure), and skeletal dysplasia and is caused by recessive mutations in the EIF2AK3 gene, encoding the pancreatic eIF2-α kinase (PERK) (18). EIF2AK3 loss-of-function variants in Wolcott-Rallison syndrome disrupt the ISR, thereby limiting the response to ER stress, impairing the cell’s ability to recover, and leading to cell death (Fig. 2B).

Our results are consistent with the hypothesis that the EIF2AK3 variants in Wolcott-Rallison syndrome and the novel EIF2B1 variants identified in our cases affect different steps within the same ER stress response pathway (Fig. 2C), highlighting the importance of this pathway in pancreatic β-cell and liver cell function and integrity (19). The convergence of both EIF2B1 and EIF2AK3 on the ISR provides an explanation for the phenotypic overlap between Wolcott-Rallison syndrome and the disease caused by dominant EIF2B1 variants, consisting of neonatal/early-onset diabetes and liver dysfunction. However, whereas liver symptoms appear to resolve with age in the patients with EIF2B1 variants, prognosis in Wolcott-Rallison syndrome is poor and most patients die of liver failure. Additionally, individuals with Wolcott-Rallison syndrome have skeletal dysplasia, a feature not observed in the case subjects with EIF2B1 variants we report. It has been proposed that the skeletal manifestations in Wolcott-Rallison syndrome arise specifically from a developmental defect in osteoblasts due to absence of PERK activity (2,3) rather than from the ER stress regulation defect that is thought to cause PNDM and liver failure. Another possible explanation for the differences in the extrapancreatic phenotypes in case subjects with Wolcott-Rallison syndrome and the individuals with EIF2B1 de novo variants we report could be the existence of quantitative differences in strength of signal or in its specific relationship to ER stress. It is possible that the level of attenuation of the ISR and the context for the attenuation caused by heterozygous mutations in EIF2B1 are enough to cause β-cell dysfunction and diabetes but not enough to result in the severe liver and skeletal phenotype.

In summary, we report a novel genetic syndrome of neonatal/early-onset diabetes and transient liver dysfunction caused by dominant de novo EIF2B1 mutations that disrupt the ability of the eIF2B complex to sense eIF2 phosphorylation and likely result in severe ER stress. These results further highlight the importance of ER stress regulation for β-cell function, adding another key gene to the list of ER stress regulators that, when mutated, cause syndromic forms of diabetes (4,20–27). This knowledge is crucial to inform current efforts aimed at developing targeted therapies for patients with syndromic forms of diabetes caused by ER stress dysregulation (28,29) but also for patients with type 2 (30) and type 1 (31) diabetes, as ER stress is known to play a key role in the pathogenesis of these more common diabetes subtypes.

Acknowledgments. The authors thank the families for participating in the study. The authors are also grateful to Rebecca Ward (Exeter Genomics Laboratory, Royal Devon and Exeter Hospital) for technical assistance, Dr. Thomas Laver (University of Exeter Medical School) for critical review of the manuscript, and Dr. Ethel Codner (Institute for Mother and Child Research, University of Chile) for assistance with clinical data collection. Whole-genome data analysis was run on the University of Exeter ISCA high-performance computing facility.

Funding. E.D.F. is a Diabetes UK RD Lawrence Fellow (19/005971) and the recipient of a European Foundation for the Study of Diabetes Rising Star Fellowship (2018). E.D.F. has also received funding from the DRWF (Diabetes Research & Wellness Foundation). A.T.H. and S.E. are the recipients of a Wellcome Trust Senior Investigator award (grant WT098395/Z/12/Z). A.T.H. is employed as a core member of staff within the National Institute for Health Research (NIHR)-funded Exeter Clinical Research Facility and is an NIHR senior investigator. S.E.F. has a Sir Henry Dale Fellowship jointly funded by the Wellcome Trust and the Royal Society (grant number 105636/Z/14/Z). D.R.’s research is supported by a Wellcome Trust Principal Research Fellowship (Wellcome Trust 200848/Z/16/Z).

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

Author Contributions. E.D.F., A.T.H., S.E., and S.E.F. participated in study conception and design. E.D.F., S.E.F., M.B.J., and S.E. performed the genetic analysis. A.T.H. analyzed the clinical data. M.N.W. and R.C. performed bioinformatics analysis. D.R. participated in data analysis and interpretation. E.D.F. wrote the first draft of the manuscript. R.C., A.T.H., and D.R. participated in manuscript improvement. A.Z., V.C.D., C.T.B.N., R.G., M.V.J., and M.E.-K. collected patients’ samples and clinical data. All authors reviewed the manuscript. E.D.F. 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.