To identify potential causative alleles, genome sequencing was undertaken in each family we identified (see Additional file 2, Table S1). For family 1, we interrogated the exome of both parents and the affected child using a custom exon-capture reagent, followed by high-throughput sequencing with the Illumina HiSeq platform. For family 2, we obtained the whole genome sequence of the affected child, his unaffected sibling, and both parents from Complete Genomics. We also performed exome sequencing in this family as part of a pilot study to compare the yield of these methods (see Methods). For families 3 and 4, exome sequencing was performed on the proband only. After sequencing, we identified all coding and near-intronic differences between the subjects and the human reference genome (see Methods). By using the unaffected parents as controls, it was possible to identify de novo mutations in the affected subject.

In each generation, 70 to 175 de novo point mutations are expected [10, 11], and 0 to 3 of these are anticipated to cause protein-coding changes [11]. In subject 1, a single protein-coding de novo mutation was identified (chr18, g.31318578C > T, p.Q404X; hg19). In subject 2, two coding de novo mutations were identified. The first (chr16, g.75258715G > A, p.R248H) occurred in CTRB1, and was considered non-pathogenic, as the variant has been reported in samples analyzed for the Thousand Genomes project (rs191950160), and is identical to the orthologous alleles in the chimpanzee and rhesus macaque. The second occurred in ASXL3 (chr18, g.31318764C > T, p.Q466X) and has not been previously reported. In subject 3, a de novo 4 bp deletion was identified (chr18, g.31319343_31319346delACAG, p.T659fsX41); this frameshift was predicted to generate a premature termination codon (TGA) after an additional 41 amino acids. In subject 4, a de novo 1 bp insertion was identified (chr18, g. 31318789_insT, p.P474fs); this frameshift mutation was predicted to generate a premature termination codon immediately. In all cases, the variant was interrogated in the affected child and parental DNA obtained from peripheral blood by Sanger capillary sequencing. All four de novo ASXL3 mutations generated stop codons, and were predicted, in silico, to generate a truncated ASXL3. Further, the mRNAs containing these premature stop-codon mutations may be degraded by nonsense-mediated decay (NMD) and thus could represent loss-of-function alleles. Neither SNV allele occurred in a CpG dinucleotide; the only known site with an increased propensity for de novo mutations [10] (see Additional file 3, Table S2).

In Drosophila, the additional sex combs (Asx) gene is required to maintain homeotic gene activation and silencing, and in mice, three orthologs (Asxl1, Asxl2, and Axl3) have been identified. Asxl1 acts on the developmentally important Hox genes both as a repressor (HoxA4, HoxA7, and HoxC8) and as an enhancer (HoxC8) [12]; dysregulation of the human HOX genes may account for the developmental phenotype. Little information is available about either embryologic or fetal expression of the ASXL gene family; however, in Drosophila, regulation of the Asx gene is highly variable and tightly controlled in the first 3 hours after fertilization [13]. In humans, ASXL3, like ASXL1 and ASXL2, is a putative polycomb protein and probably acts as a histone methyltransferase in a complex with other proteins [14]. ASXL3 is expressed in similar tissues to ASXL1 including brain, spinal cord, kidney, liver, and bone marrow, but at a lower level [15] (see Additional file 4, Figure S2). Within the brain, ASXL1 has much higher expression in the white matter, whereas ASXL3 has moderately higher expression in the insula, cingulate gyrus, and amygdala, with approximately similar expression elsewhere [16] (see Additional file 5, Figure S3). The high correlation of expression patterns between ASXL1 and ASXL3 may account for some of the shared phenotypic features.

No deleterious ASXL3 mutation was found in a small cohort of patients with BOS without causative ASXL1 mutation (Hoischen, personal communication), consistent with ASXL3 mutations conveying a phenotype distinct from BOS. Using large-scale datasets (Thousand Genomes [17], dbSNP, ESP5400, and Cohorts for Heart and Aging Research in Genomic Epidemiology [18]) we identified four other truncating mutations in ASXL3, which occurred as singletons within each dataset in reportedly phenotypically normal individuals (Figure 2), and thus may represent benign variants. Two of these mutations occur at the extreme 3' of the gene, and thus may escape NMD and retain protein activity. One mutation, p.L902X (rs187354298), identified in a single sample from the Thousand Genomes cohort, occurs more 3' to, but within the same penultimate exon as, the two disease-causing mutations (see Figure 2). More interestingly, however, a high-quality nonsense mutation (R322X) was identified in the exon 9 of the 12-exon ASXL3 gene, and is anticipated to undergo NMD or may be otherwise highly deleterious to protein function.

Although all four disease-associated de novo variants and rs187354298 are predicted to potentially undergo NMD, it is challenging to reconcile such a hypothesis with the observed range of phenotypes for nonsense alleles reported at this locus. However, the current ability to predict NMD is limited, and it has been shown that around 75% of mRNA transcripts that are predicted to undergo NMD escape destruction, and that the nonsense codon-harboring mRNA is expressed at levels similar to wild type in lymphoblastoid cells [19]. Furthermore, the dynamics of mRNA stability and degradation may differ for cells and tissues undergoing rapid developmental changes. Parenthetically, we noted an enrichment of around 50% in gene regions where mRNA would be predicted to escape NMD, or 3' gene bias to the predicted loss-of-function nonsense codon mutations in normal controls from the Thousand Genomes data [19].

Another distinct interpretation of our observations is that all ASXL3 disease-causing nonsense-encoding mRNAs are translated into prematurely terminated proteins, which act in a dominant-negative fashion. In support of this hypothesis, Sanger sequencing of multiple cDNA extractions derived from a transformed lymphoblast cell line from subject 2 showed that both alleles were expressed (see Additional file 6, Figure S4) although this does not exclude that some degree of NMD may occur or reflect what occurred during development. Further, we observed that in previously reported cases of BOS known disease-causing nonsense mutations in ASXL1[2, 20] occur almost entirely within a very limited region of the protein,. Furthermore, database searches reveal that truncating mutations, in reportedly phenotypically normal individuals, can occur both 5' and 3' of these mutations, just as we now report for ASXL3 (see Additional file 7, Figure S5). This disease-causing mutation hotspot falls between two paralogous regions shared by all ASXL genes (Figure 2) and into a region unique to ASXL1. Interestingly, the presumptive disease-causing mutations we describe here occur within an analogous region in ASXL3, within the first half of the penultimate exon. Further, disease severity may decrease the more 3' the mutation occurs within this region. This region contains a number of predicted phosphorylation sites, an evolutionarily conserved region (residues 420 to 470, approximately), and an evolutionarily conserved serine-rich motif between residues 600 and 800, approximately. We speculate that disruption of these conserved regions may result in dysregulation of post-translational protein modification, resulting in constitutive activation.

Truncating ASXL3 mutations are uncommon, and their de novo nature makes it even less likely that we identified these individuals by chance, which highlights the value of de novo mutation-based methods to find disease-causing loci. To determine the probability of observing multiple de novo truncating mutations in ASXL3, we developed a model [21, 22] accounting for gene size, GC content and de novo rates of SNVs and small insertion/deletions, and of the probability of those mutations causing a truncation of the protein. The probability of developing a de novo nonsense mutation in ASXL3 is 3.35 × 10^-6 per generation, whereas the probability of developing a de novo coding insertion or deletion in ASXL3 is approximately 3.91 × 10^-6. Thus, the total probability of observing three additional individuals with truncating ASXL3 mutation, given the first de novo observation, is around 4.0 × 10^-17. The observation of four de novo truncating mutations occurring in association with a sporadic disease that shares similar phenotypic features is highly unlikely to have occurred by chance; nevertheless, functional studies will be required to show conclusively that truncating mutations in ASXL3 have pathological consequences that cause the observed disease trait.

Although all four subjects shared clinical findings, these characteristics were mostly non-specific. Severe feeding difficulties, present from birth, that required intervention (3/4 subjects). The subjects had small size at birth (3/4), with microcephaly (3/4) and severe psychomotor delay, with missed milestones (4/4) at their most recent evaluation. Deep palmar creases (4/4) and slight ulnar deviation of the hands (3/4), combined with a high arched palate (3/4) were also common. No patient had the typical 'BOS posture' of elbow and wrist flexion, or of myopia or trigonocephaly (0/4).

The phenotype present in the three affected individuals varies in both presentation and severity, a phenomenon that is also reported in subjects with ASXL1 mutation. Several factors may account for this. First, truncating mutations occurring earlier in the gene seem to be associated with a more severe phenotype, with truncating mutations at the extreme 3' end of the gene yielding no observed phenotype. Interestingly, this does not seem to be the case for ASXL1[2]. Additional subjects will be needed for further genotype-phenotype analysis to address a potential polarity hypothesis [23]. Second, because of the importance of the ASXL gene family in very early development, the time at which the mutation arose may also influence the phenotypic outcome; mutations that occurred in the parental gametes could convey a more severe phenotype than those arising post-zygotically or during later embryogenesis [24]. Third, ASXL proteins form complexes with other proteins, and have been shown to influence Trx gene mutations in flies [12]. Mutational load and other epistatic effects may contribute to the observed phenotype, and we could not discern such alleles using our de novo variant approach. Finally, epigenetic factors may contribute to the phenotype. In mice, homozygous Asxl2 mutations can lead to two primary outcomes: around 20% are born very small and die by the age of 2 months, whereas the remaining 80% are smaller at birth but gain weight normally and are successfully weaned [25]. Thus, other factors may contribute to the penetrance and/or expressivity of de novo ASXL3 mutations in humans, and severe phenotypes could be atypical.

The condition defined molecularly in the current study is phenotypically distinct from, but with similar and overlapping features to BOS. This is probably the consequence of functional overlap between the causative candidate genes ASXL1 and ASXL3, which are both developmentally important putative polycomb genes. Differentiating two phenotypically similar syndromes based on clinical presentation alone is challenging, and is further complicated by phenotypic variability. Molecular methods permit an objective means to establish and secure a diagnosis. Moreover, these methods now enable comparative analyses between novel and well-described syndromes to make use of evolutionary genetics in addition to phenotypic features in disease nosology. This allows a distinct molecular diagnosis, and increases diagnostic capabilities for rare syndromes. Interestingly, in this study, the subjects were identified not by establishing phenotypic overlap between them, but rather by identifying that they shared de novo nonsense mutations in the identical genes, and that mutations in a related gene, ASXL1, resulted in a similar phenotype. In particular, subjects 3 and 4 were identified from a small clinical cohort (n = 192) of individuals with psychomotor delay, based upon the presence of rare truncating ASXL3 mutations, which were later determined to be de novo. This is a novel way in which molecular diagnostics can help foster international and inter-institutional collaborations that will be vital to both solving the multitude of very rare diseases and to functionally annotating the human genome.

De novo truncating ASXL3 mutations are very likely the cause of a novel syndrome. The ability to communicate and share genotype findings can elucidate novel disorders that would be challenging to discover based on phenotypic data alone.