The phenotype of the Rfx6 eGFPcre/eGFPcre
mice is remarkably similar to human patients born with neonatal diabetes and small bowel obstruction due to bowel atresia17, 18
. Despite some reduction in pancreatic size, these cases were not deficient in enzymes of the exocrine pancreas, and autopsies of two cases (proband #1 and case 3 in Ref 17
) revealed normal-appearing exocrine pancreata with clusters of ChromograninA-positive cells but total absence of cells staining for insulin, glucagon, or somatostatin17
. In addition, the syndrome involves hypoplastic gall bladder, and intractable diarrhea unresponsive to pancreatic enzyme replacement.
The disease locus was mapped using overlapping homozygosity in probands #1and #2 (see Supplementary material for pedigree information and references to previous clinical case reports) respectively the offspring of first and second cousins. High-resolution homozygosity mapping identified 10 homozygosity-by-descent (HBD) segments >500 kb in proband #1 (after excluding those that overlapped with her unaffected sibling, Supplementary Table S4
), and 25 HBD segments >500kb in proband #2 (Supplementary Table S5
). Only three HBD regions were common in the two probands, totaling 24 Mb (). Altogether, 194 RefSeq genes map to these regions. Of these genes, only RFX6
, which falls in the largest segment at 6q21-22, had pancreas–enriched expression in the TiGER database 19
(Supplementary Table 6
), and also increased in expression in human pancreas between foetal ages 10 and 20 weeks ( and Supplementary Figs. S13
), concordant with the appearance of endocrine cells 20,21
Regions of homozygosity-by-descent common to probands #1 and #2
Two parallel, independent approaches unequivocally identified mutations in the RFX6 gene in this human syndrome: direct sequencing of the RFX6 gene and unbiased deep sequencing of all exons within the three overlapping HBD regions.
For deep sequencing, exons were captured from DNA obtained from Proband #2 using a tiled oligonucleotide array22
covering 1,309 of 1,322 exons mapping within the HBD regions23
. Amplification and sequencing24
of the captured fragments generated 40,379 sequences of at least 100 bp that aligned within the target regions. Median target coverage depth was 9.2, with 80% of targets having a depth of at least 4. Given that we were searching for a homozygous mutation, this was sufficient for unequivocal detection of exonic variants. Altogether, 30 novel sequence variants were detected (Supplementary Table S7
): 15 in introns, 3 in both introns and untranslated regions (UTRs), 9 in UTRs, and only three in coding sequences, two synonymous. The only non-synonymous variant was 217 Ser>Pro in RFX6
, identifying this gene as the most likely candidate.
In parallel, direct sequencing was performed on the 19 exons and the splicing junctions of RFX6
in all probands. Missense, splicing or frameshift mutations in RFX6
were found in five of the six available probands () with an interesting genotype-phenotype correlation. Probands #1, 4 and 5 all died in the first few months of life and were homozygous for, respectively, a loss of the donor splicing site in intron 2 (IVS2+2 t>c), an out-of-frame deletion in exon 7, and the missense mutation 181 R>Q involving a highly conserved arginine in the DNA-binding domain 8,25
(Supplementary Fig. S15
). Proband #3, still alive at the age of 9 and intermittently off insulin, was a compound heterozygote for donor-site loss in intron 6 (IVS6+2 t>g) and disruption of the acceptor site in intron 1 (IVS1-12 a>g). Proband #2, still alive at age 4.5 years18
, had the homozygous missense mutation 217 Ser>Pro, confirming the unbiased exon sequencing described above. All mutations were inherited from carrier parents.
To determine the significance of the homozygous intron 2 splice donor splicing site mutation in proband #1, we amplified RFX6
mRNA by RT-PCR of high-quality RNA from autopsy pancreas and failed to detect the properly spliced transcript, which was easily amplified from normal foetal pancreas as was the reference gene cyclophilin in the proband’s RNA. We also failed to detect any RNA from exons 1+2, upstream of the splicing mutation, probably due to nonsense-mediated decay26
(Supplementary Fig. S16
We also tested the two missense mutations for their effect on DNA binding by Rfx6. We found that 181R>Q (proband #5), which alters a conserved amino acid in the DNA binding domain, completely abrogated DNA binding, while 217S>P (proband #2), which lies between the DNA-binding domain and dimerization domain of Rfx6, only modestly reduced DNA binding () and did not affect dimer formation (data not shown).
Finally, we failed to identify any mutation in RFX6
in proband #6. In the absence of DNA from the proband, we sequenced both parents and found no point mutation of RFX6
; and long-range PCR did not reveal any deletions of RFX6
(Supplementary Fig. S17
). In the absence of proband DNA we cannot rule out a de novo
mutation, but this case is most likely a phenocopy. We also failed to find RFX6
mutations in a case of the Martinez-Frias syndrome27
. Finally, a search of the RFX6
linkage disequilibrium block in our genome-wide association data 28,29
, combined with those of the WTCC 30
, did not reveal any common variants associated with type 1 or type 2 diabetes (data not shown).