Analysis of Transcription Factors Key for Mouse Pancreatic Development Establishes NKX2-2 and MNX1 Mutations as Causes of Neonatal Diabetes in Man

1Knowledge on transcriptional regulators of pancreatic development and beta cell differentiation is essential for the development of beta cell replacement therapies for diabetes. While major work has been undertaken in mice, the extent to which murine pancreatic development is a good model for man is not fully understood.

2Finding inactivating mutations in patients with diabetes can establish the role of a gene in human pancreatic development and function. Comparing the human phenotype with mouse models for the same gene is important, as it validates that the gene function is similar in man and mouse. Concordant phenotypes between mouse models with biallelic gene inactivation and humans with biallelic mutations suggest similar function of the gene product in rodents and man. For example, biallelic GCK mutations in humans cause permanent neonatal diabetes (PNDM), and mice lacking both copies of the gene have severe hyperglycemia at birth. Both patients with recessive EIF2AK3 mutations and the Perk null mouse (Eif2ak3−/−) develop diabetes due to destruction of beta cells as a result of endoplasmic reticulum stress

3The correlation in phenotype between knockout mice and humans with recessive mutations is not always evident. Biallelic loss-of-function ABCC8 mutations are the most common known cause of hyperinsulinemic hypoglycemia in humans, whereas ABCC8 null mice maintain euglycemia. GLUT2 has been shown to be the primary glucose transporter and sensor in rodent pancreatic islets, whereas studies in humans have suggested that GLUT1 plays an important role in facilitating glucose entry into the developing beta cell. An example of a difference at the level of the gene structure is the insulin gene: in humans preproinsulin is encoded by a single gene (INS), while in mice two functional copies of the gene exist (Ins1 and Ins2).

4Studies investigating transcriptional regulation of beta cell development in mice have, however, been successful in identifying pancreatic transcription factor genes important for human pancreatic development. Much of the mouse work has involved the complete or targeted (conditional) knockout of a gene followed by detailed phenotypic studies during embryonic development and, when viable, in the resulting live offspring. To date, mutations in eight different pancreatic transcription factor genes (PDX1, PTF1A, GLIS3, PAX6, RFX6, NEUROD1, NEUROG3, GATA6) have been identified in patients with neonatal diabetes. For some transcription factor genes the genetic etiology in the patients was initially recognized from the similarity to the mouse phenotype. Mutations causing a pancreatic/diabetes phenotype have not been found in man for many pancreatic transcription factors important for rodent pancreatic development, but there have been no large, systematic studies of these genes in appropriate patients to date.

5The aim of our study was to perform a comprehensive search for recessive mutations in genes encoding transcription factors known to be critical for pancreatic development in mice in a large collection of PNDM patients born to consanguineous parents. Patients with consanguineous parents were chosen, as the likelihood of a recessive cause of their PNDM is greatly increased, and all mouse models of transcription factors influencing pancreatic development involve biallelic inactivation. We tested for mutations in homozygous regions encompassing known transcription factor genes independently of the clinical features to avoid the possible bias introduced when clinical features guide candidate gene testing.

6We studied 147 patients from consanguineous pedigrees. Mutations in non-transcription factor genes were identified in 110 probands (75%) (Table 1, Figure S1, and Table S1). Systematic investigation of 29 transcription factor genes important for murine pancreatic development identified homozygous mutations in 7 genes in 12 patients from 11 families (30% of the undiagnosed probands or 7.5% of the entire cohort) (Table 1). Overall, 106 out of 121 (88%) patients with a known genetic etiology had homozygous mutations, confirming that neonatal diabetes in the offspring of consanguineous families is usually recessive.

7Homozygous NKX2-2 nonsense or frameshift mutations were identified in three patients from two families (Table 2). These mutations are highly likely to be pathogenic, as they are null mutations. Cosegregation studies were consistent with recessive inheritance (Figure 1A). No heterozygous or homozygous truncating mutations have been reported in 6,500 individuals sequenced by the NHLBI Exome Sequencing Project (http://evs.gs.washington.edu/EVS/). All three patients have severe defects in insulin secretion, as shown by intrauterine growth retardation (IUGR) and presentation of diabetes at an early age without features of pancreatic exocrine dysfunction (Table 2). This is very similar to mice that are homozygous for a targeted disruption of Nkx2-2 and die shortly after birth with severe hyperglycemia (Table 2). They have normal exocrine function, lack beta cells, and have fewer alpha and pancreatic polypeptide cells. The patients with homozygous null mutations in NKX2-2 also have severe developmental delay, which affects both motor and intellectual function as well as more specific features, such as hypotonia, cortical blindness, impaired visual tracking, and hearing impairment. These are consistent with the severe neurological features seen in the Nkx2-2 knockout mouse (Table 2) and Nkx2-2 being essential for hindbrain development, ventral neuronal patterning, and oligodendrocyte differentiation.

8Two probands have homozygous missense mutations in MNX1 (previously described as HLXB9). The p.F248L and p.F272L missense mutations are highly likely to be pathogenic, as they cosegregate with neonatal diabetes (Figure 1B), affect highly conserved residues within the homeodomain (Table 2, Figure 1C), are not listed in 6,500 individuals sequenced by the NHLBI Exome Sequencing Project, and in silico analysis predicted the mutations to be disease causing. Our results are in line with the recently published paper by, who independently found the same MNX1 p.F272L mutation in proband 2. Both probands in our study had severe intrauterine growth retardation and were diagnosed with diabetes in infancy without evidence of exocrine pancreatic dysfunction. This is similar to the Mnx1 null mice that are born smaller than their wild-type littermates and have small islets and reduced numbers of beta cells and normal exocrine function (Table 2). These mice also have dorsal lobe agenesis of the pancreas; it is not known if this is a feature of our two patients, as imaging and post mortem investigations were not performed. Proband 1 (but not proband 2) had marked extrapancreatic developmental features, including severe neurological complications and hypoplastic lungs (respiratory failure caused death at 10 months) (Table 2). In keeping with these observations, Mnx1 null mice have aberrant motor neuron differentiation but are difficult to study as they die at birth due to respiratory paralysis, which probably results from failure in the development of the phrenic nerve. The variability in phenotype between the two probands could be due to differences in the severity of their mutations, but further studies investigating the functional consequences of these missense mutations are required in order to explore this hypothesis.

9Proband 1 also had sacral agenesis and a high imperforate anus, which are both cardinal features of Currarino syndrome (OMIM #176450), resulting from heterozygous MNX1 mutations. Mice homozygous for a null mutation do not have skeletal defects. Proband 2 and his heterozygous relatives do not have features of Currarino syndrome. The highly variable phenotype is not unexpected, as around 50% of heterozygous carriers are clinically asymptomatic. As diabetes is not a feature of Currarino syndrome, it seems likely that a single copy of MNX1 is sufficient to maintain normal pancreatic development and glucose homeostasis, a hypothesis that is further supported by the observation that heterozygous mice have a normal pancreas.

10Homozygous mutations in five pancreatic transcription factor genes previously reported to cause PNDM were identified in seven probands. The phenotype of these patients is similar to that seen in mice with biallelic inactivation of these genes (Table 3). Our two patients with GLIS3 mutations, the six previous cases, and the Glis3 mouse knockout all have neonatal diabetes diagnosed in the first week of life and congenital hypothyroidism. We identified one patient with a homozygous NEUROD1 mutation. Both this patient and a previously reported case from a nonconsanguineous pedigree have neonatal diabetes, neurodevelopmental delay, cerebellar hypoplasia, and hearing and visual impairment. The Neurod1 null mouse also has neonatal diabetes and severe neurological defects, including impaired coordination and ataxia, cerebellar hypoplasia, sensorineural deafness, and retinal blindness. The proband with a homozygous RFX6 mutation was diagnosed with diabetes at 8 days and had duodenal atresia and gall bladder agenesis. This is consistent with the previous five cases and the Rfx6 knockout mouse, which has neonatal diabetes and small bowel atresia.

11Two probands with homozygous PDX1 (previously described as IPF1) missense mutations had isolated PNDM. In both man and mouse, the severity of the PDX1 mutation determines the pancreatic phenotype. Pancreatic agenesis was seen in the mouse knockout and in patients with two copies of a severe mutation. In mice, hypomorphic mutations are associated with diabetes without evidence of exocrine deficiency; we and others have shown that patients with less-severe homozygous missense mutations have neonatal diabetes with normal or mildly impaired exocrine function.

12An intronic PTF1A mutation (c.784 + 4A > G) was identified in a proband with pancreatic agenesis. This mutation is predicted to result in an alternative splice donor site (Alamut Interactive Biosoftware, version 2.1) and the inclusion of an additional 4 bases in the mRNA, causing a frameshift and premature termination codon at residue 271. The pancreatic phenotype is similar to the three previously reported patients with nonsense or frameshift mutations and the Ptf1a null mouse. Our patient has mild learning difficulties at 15 years old, which is in contrast to the other patients who had severe neurological features, causing death before 4 months of age. These patients and the knockout mouse had cerebellar aplasia. In a mouse model, reduced Ptf1a gene dosage results in pancreatic hypoplasia and glucose intolerance in a dose-dependent manner. We hypothesize that the absence of cerebellar agenesis in this patient may reflect a reduced functional severity of the splicing mutation compared to previously reported null mutations.

13The finding that mutations in NKX2-2 and MNX1 can cause neonatal diabetes takes the number of genes in which mutations are known to cause PNDM from 18 to 20, of which 10 encode pancreatic transcription factors. Unlike previous studies, which have identified mutations in humans because of similarities in phenotype with the null mouse, our systematic approach guided by areas of linkage has allowed us to screen all transcription factor genes independently of clinical features. Our results confirm that the consequence of inactivation of pancreatic transcription factor genes in humans is similar to the phenotype observed in knockout mice. This is in contrast to heterozygous loss-of-function mutations where differences in phenotype have been observed. For example, monoallelic mutations in GATA6 or HNF1B cause pancreatic agenesis with congenital heart defects or renal cysts with diabetes, respectively, while mice heterozygous for a null mutation in Gata6 or Hnf1b show no obvious phenotype.

14There are some limitations to our study. The minimum prevalence of transcription factor mutations in our cohort of patients with consanguineous PNDM was 7.5%, but the true prevalence may be slightly higher since heterozygous or compound heterozygous mutations would have been missed as a result of the study design. However, as all of the patients were known to be consanguineous, the prior probability is that they have recessively inherited disease due to a homozygous mutation. While the identification of patients with homozygous inactivating mutations in pancreatic transcription factor genes has confirmed their role in human organogenesis, these findings can only provide limited information on pancreatic development, as detailed physiological studies in man are not always possible. This is especially true for the patients within our cohort with NKX2-2 and MNX1 mutations that are all currently under 13 years of age and, for those with NKX2-2 mutations, have severe developmental delay, which means that in-depth studies on pancreatic and neurological function are not possible.

15In conclusion, we have shown that mutations in NKX2-2 and MNX1 cause neonatal diabetes. This confirms a key role for NKX2-2 and MNX1 in human pancreatic development. Comparisons of phenotypes observed in humans and mice with inactivation of transcription factor genes supports common steps critical for pancreatic development. This study validates the use of knockout mice for understanding beta cell development in humans.

16We recruited 147 patients from consanguineous pedigrees (parents of the proband were second cousins or more closely related) who were diagnosed with permanent diabetes before 6 months of age or were diagnosed with permanent diabetes before 9 months and had any additional clinical features that made a diagnosis of type 1 diabetes less likely. Informed consent was obtained from all participants or their parents, and institutional review board approval was received for this study.

17To detect mutations in non-transcription factor genes that cause PNDM, we used Sanger sequencing (ABCC8, KCNJ11, INS) in all patients, followed by homozygosity mapping with sequencing of known genes (EIF2AK3, IER3IP1, GCK, SLC19A2, and SLC2A2) within homozygous regions >3 Mb.

18From published literature, we identified 29 transcription factor genes crucial for murine pancreatic development (Table 1). We identified homozygous regions through genome-wide single nucleotide polymorphism (SNP) genotyping and used Sanger sequencing to detect mutations within homozygous regions encompassing these genes.

19Genotyping was undertaken on the Affymetrix 10K, 500K, or v6.0 SNP chips by Medical Solutions, ALMAC Diagnostics, or AROS Applied Biotechnology. Processing of genomic DNA was performed as per the Affymetrix protocol, and in-house Perl scripts were developed to identify regions of significant homozygosity (>3 Mb) as previously described.

20When a large homozygous region encompassing one of the pancreatic transcription factor genes was identified, the coding regions and intron and exon boundaries of the gene were sequenced (Table 1, Figure S1, and Tables S2 and S3). Details of PCR primers and sequencing conditions are available on request. Pathogenicity was assessed on the basis of the characteristics of the mutation, the conservation of the affected amino acid, the absence from variant databases (accessed using Alamut v.2 Interactive Biosoftware), and cosegregation within families.

21For patients with homozygous mutations, their phenotype (both pancreatic and extrapancreatic) was compared with the phenotype of the knockout mouse for the same gene.