Overview of cohort results

The OxClinWGS RD cohort comprised a total of 300 genomes from 122 families. One hundred forty-eight male and 152 female participants were recruited, the majority of whom were of European White ancestry, although African, Asian and American families were also included (Additional file 1: Fig. S3), reflecting the population from which the patients were primarily recruited. Overall cohort statistics, including details of family size, gender, disease categories and solved status of individual cases recruited, are shown in Fig. 1B, C, Additional file 3: Tables S4-S6 and Additional file 1: Figs. S6 and S7. The results of WGS for all patients in the cohort, including causative genes and variants (where solved) and associated phenotypes are provided in Additional file 3: Tables S6 and S7, which also include references for some individual cases which have been published previously. More detailed clinical case histories for selected patients are provided in the Additional file 2. Variants identified in this study have been uploaded to ClinVar [89].

Our diagnostic yield in this RD cohort was 43/122 cases (35%). These were cases which had variants with ACMG classifications of pathogenic/likely pathogenic (39/43) or were variants with evidence of causality in known disease genes (4/43) which were clinically accepted and returned, informing diagnosis or treatment of these patients. Across the cohort, we considered 39% of our cases to be solved (47/122) since an additional four cases had variants in novel disease genes that had compelling evidence of causality from additional families with matching phenotypes, or functional data (Fig. 1B, Additional file 3: Tables S7 and S8). A further 12/122 (10%) cases had a variant of uncertain significance in a lead candidate identified from genetic analysis. Two cases with clinically actionable secondary findings were also identified. An overview of cases considered solved by variant type is shown in Fig. 1D and WGS500 classification of genes (see the Methods) is shown in Fig. 1E. Further details of inheritance pattern, de novo status and result class are summarised in Additional file 1: Fig. S8.

Across the cohort, we identified eight novel disease genes. Three of these have been confirmed and previously published as part of collaborative studies; a de novo p.(Gln735*) mutation in POLR2A in a patient with a novel neurodevelopmental syndrome with profound infantile-onset hypotonia [90]; a de novo p.(Tyr1224fs) mutation in KMT2E in a patient with a neurodevelopmental syndrome and epilepsy [91] and biallelic variants (p.(Gly79fs) and c.764 + 5G > A) in MCM10 causing telomere shortening and giving rise to immune dysfunction and cardiomyopathy [92]. Two genes, DOCK7 and SAMD9L, were novel at the time of discovery and we have evidence of causality for a further three novel disease genes (DHRS3, FOXD3, HDLBP). Variants in other lead candidates are also being investigated in functional studies. Additionally, one gene, RMND1, is novel for the phenotype of polymicrogyria whilst BMP4 is a putative novel gene for Kapur-Toriello syndrome which, if confirmed, would extend the phenotypic range of this gene from its current association with microphthalmia and clefting syndrome.

A summary of the outcomes of the project in terms of cases solved and novel candidate disease genes is shown in Fig. 2.

Fig. 2

Overview of the OxClinWGS Study: genetic and clinical results. The OxClinWGS RD cohort included 122 cases, of which 47 were considered solved and a further 12 cases had variants of uncertain significance in lead candidates identified. Two cases had secondary findings. Eight novel disease genes have been identified to date, five of which are confirmed disease genes and three of which have evidence of causality. The asterisk denotes that this group includes novel and putative novel genes. The phenotype for one gene was expanded. Revised clinical diagnoses were provided for six patients, whilst for eight patients, the findings led to changes in their clinical management. Colours denote cases with genes that are considered solved (green), have evidence of causality (light green) or are variants of uncertain significance in lead candidates (brown). Abbreviations: PAPA syndrome (pyogenic sterile arthritis, pyoderma gangrenosum, and acne); CNS (central nervous system)

Overview of variant types and HPO information

Our pipeline investigated all variant types, including SNVs, INDELs and SVs. The numbers of all variants by category, their minor allele frequency (MAF), size distribution and predicted impact are shown in Additional file 1: Figs. S9-S12 inclusive. For each variant class, we investigated variants arising de novo per pedigree (Additional file 1: Fig. S13). Fourteen pathogenic/likely pathogenic (ACMG classification) de novo variants were identified, including one secondary finding in FBN1 (see below). HPO terms were integrated into the analysis and helped to prioritise potential disease genes linked to the annotated patient phenotypes. On average, 4.7 HPO terms were recorded per pedigree (range 1–24) with ‘seizures’ being most common (Additional file 1: Fig. S14). We generally observed that solved cases clustered with higher numbers of HPO terms. Heatmap analysis of the HPO profiles (Additional file 1: Fig. S15) demonstrated overlap between different disease groups. For example, our ultra-rare disorder cases aligned with neurological and MSK groups, which can be explained by the fact that this category contains Fine-Lubinsky and Kapur-Toriello syndrome patients, which have some shared features with the craniosynostosis patients in the MSK group. In addition, clinical characteristics shared between vascular, haematological and immunological patients are also reflected in the heatmap.

We used several established and recently published deleteriousness scores to prioritise/filter causative variants and the respective value distributions, including our candidate pathogenic variants, as shown in Additional file 1: Fig. S16. No single score was able to perfectly separate true from false positives and hard filtering, for example, for a CADD PhredScore > 20 as suggested in some studies, would have caused us to miss 14/55 (25%) of our candidate SNVs.

Whilst the majority of our cases were explained by protein-coding SNVs, it is noteworthy that SVs, splice site and deep intronic variants, which have been hitherto under-explored in WGS studies, collectively contributed to 20/47 (43%) of our solved cases. These are described in more detail below.

Structural variants

Structural candidate variants accounted for 4/43 (9%) of our diagnostic yield and 7/47 (15%) of the cases we consider solved (Table 1). Three SVs have led to the identification of two putative novel disease genes. The first, a homozygous 3.9 kb deletion encompassing the promoter and 5′-UTR of DHRS3, was identified in two siblings from a consanguineous Pakistani family with craniosynostosis. A deletion in this gene, which encodes dehydrogenase/reductase-3, would be expected to lead to an increase in the plasma level of the morphogen all-trans retinoic acid, which was confirmed by liquid chromatography multi-stage tandem mass spectrometry.

Table 1 Structural variants found in the OxClinWGS cohort

Two further craniosynostosis families were found to have heterozygous SVs flanking a second, novel RD gene, FOXD3, both segregating with disease in their respective families. FOXD3 encodes a pioneer winged-helix transcription factor (TF) critical for early embryonic development [93] and is therefore a good candidate for craniosynostosis. One of these families, with bicoronal craniosynostosis, harboured a 354 kb deletion downstream of FOXD3, removing a topologically associating domain (TAD) boundary. Another family, with multi-sutural craniosynostosis, had an 11.5 kb duplication upstream of FOXD3 which duplicates a highly conserved enhancer element previously shown to interact with Foxd3 and drive neural crest expression in chick embryos [94, 95]. This SV has been confirmed by modelling in mice, which also develop craniosynostosis [17].

A fourth SV led to a change in clinical diagnosis for a female patient referred with Aicardi syndrome, a rare congenital malformation syndrome found almost exclusively in females and characterised by agenesis of the corpus callosum, seizures and chorioretinal lacunae. No genes have been identified to cause this syndrome. A ~ 3 kb de novo deletion on the X chromosome identified in our patient removes the first exon of ARX. Validation of this deletion by PCR and Sanger sequencing was confounded by the nearby repeats and high GC content of the region but it was instead confirmed by MLPA (Fig. 3A–C). Variants in ARX have been associated with several X-linked intellectual disability (XLID) syndromes, including XL lissencephaly, developmental and epileptic encephalopathy type 1 (DEE1) and Partington syndrome [96], reflecting the central role of this member of the homeobox gene family of TFs in controlling the formation of many brain structures during early embryonic development. As a result of our WGS finding, the clinical characteristics of this patient were reviewed, and since she had developmental and epileptic encephalopathy and corpus callosum agenesis, but not the ophthalmic features typical of Aicardi syndrome, her clinical diagnosis was changed to DEE1 (OMIM #308350).

Fig. 3

Validation data for two patients with X-linked structural variants in OxClinWGS study. A–C Patient with ARX deletion: A screenshot showing 125 bp read alignments supporting a de novo deletion of ARX exon 1. Region shown is chrX:25,003,000–25,019,000 (GRCh38). Visualisation is using IGV v2.11.2, with squished and “view as pairs” options. Alignments are coloured by insert size and transcript shown is NM_139058.3. B UCSC genome browser session showing the position of the deletion in relation to the PCR primers and MLPA probes that were used for validation. Also shown is the GC content which rises to > 80% near the distal breakpoint where the coverage drops also in parental genomes. An interactive version is available at https://genome.ucsc.edu/s/AlistairP/ARX_deletion_v6. C The 3 kb deletion was confirmed by the MLPA validation data visualised using coffalyser and shows a drop in signal only for the proband for the exon 1 probe (red arrow). The grey boxes are reference probes and the orange boxes highlight the 95% confidence range of the reference samples used. D–H Complex rearrangement in patient with X-linked neurodevelopmental disorder. D Read count information from short-read sequencing normalised by ngCGH software (https://github.com/seandavi/ngCGH) showing two X chromosome duplications (red arrows). E Split Illumina read-pairs suggest the two duplications are inter-linked. However, two possible configurations can explain the split read pattern. F Circos plot highlights the only SV identified by the Bionano pipeline above the threshold for SV detection. G Genome browser view of the optical maps robustly detects the ~ 600 kb duplication of Xp22.11p21.3 being inserted into Xq27.1, which is present in the carrier mother and both affected male siblings using the Bionano pipeline excluding Complex Multi-Path Regions (CMPR). The red box highlights the duplication inserted into Xq27.1. H The Bionano pipeline without masking CMPR detects ~ 102 kb tandem duplication (red boxes) flanking either side of the 600 kb insertion from Xp22.11-Xp21.3 (blue box), therefore, supporting conformation 1 as suggested in E

A fifth SV led to an in-frame 219 kb deletion of exons 6–8 of WWOX, leading to loss of 180 amino acids including the mitochondrial targeting sequence. This was in trans with a c.705dup p.(His236fs) variant in this known epilepsy gene and provided a diagnosis for a patient with severe epilepsy. These compound heterozygous variants were previously reported as part of a case series expanding the phenotypic spectrum associated with this gene [97].

Two further SVs represent more complex rearrangements. A large 633 kb duplication of Xp22.11-Xp21.3 had been identified by prior clinical array testing in two brothers with a severe neurodevelopmental syndrome and hypotonia. Short-read WGS data allowed us to confirm the precise breakpoints of this rearrangement, in addition to identifying a second 102 kb duplication of Xq27.1 (Fig. 3D). The larger duplication encompasses PDK3, PCYT1B and POLR1A, while the smaller one did not contain any annotated genes. Although split read-pairs indicated that the two duplications were interlinked, short-read data alone could not resolve which of the two possible configurations was correct (Fig. 3E). However, FISH data combined with optical-mapping, an orthogonal technique (Fig. 3F–H), suggest that the 633 kb segment is inserted within the 102 kb tandem duplication, ~ 200 kb downstream of SOX3. Genomic insertions downstream of SOX3 have been reported to cause a number of variable conditions that include hypoparathyroidism and laryngeal abductor paralysis [23, 98]. Therefore, we postulate a similar positional effect here, involving long-distance regulatory mechanisms.

The second complex rearrangement was found in a patient with hereditary maxillary prognathism. This patient had five segments of chromosome 1 inserted into chromosome 17q24.3, which is hypothesised to disrupt a TAD close to KCNJ2/SOX9. The rearrangement was confirmed by nanopore long-read genome sequencing and has been classified as an example of chromoanasynthesis [99] revealing a new mechanism for this rare craniofacial phenotype.

Although in principle, four of these SVs (WWOX, two FOXD3 and DHRS3 variants) could have been detected by arrays, they were not picked up by standard of care testing prior to WGS referral, because they were inadequately covered by probes, did not meet the thresholds for clinical laboratory reporting or were in novel genes; therefore, their significance was not appreciated (Table 1). We note that for two complex SVs, detection by array only would leave their full complexity under-appreciated and indeed for one of these, WGS analysis to characterise the precise insertion site was the reason for recruitment, as the larger of the duplicated segments had already been identified.

All SVs were validated by independent methods, including PCR and Sanger sequencing, MLPA, SNP arrays, nanopore long range sequencing and BioNano (Table 1) and the range of the methods required highlights the challenges of doing this at scale in a routine clinical setting.

Splice site and deep intronic variants

We used three different splicing algorithms to inform our analysis of splice site variants; SpliceAI, MaxEntScan and our novel algorithm, ALTSPLICE. We first validated ALTSPLICE by comparing its performance with that of SpliceAI, using a previously published, manually curated set of clinical splice-altering and control SNVs [100]. The scores from ALTSPLICE and SpliceAI are shown in Additional file 3: Table S9. The area under the precision recall curve was found to be 96.8% for ALTSPLICE and 96.4% for SpliceAI (Additional file 1: Fig. S17), validating the ALTSPLICE algorithm and demonstrating that the performance of the two is similar overall, even though they are independently constructed and trained.

We identified sixteen splice site or deep intronic variants (of which fourteen were unique), which are listed in Table 2. Splice site variants contributed to 12/43 (28%) of our confirmed diagnoses, and to 13/47 (28%) of our solved cases. A further three splice site or deep intronic variants in two cases are variants of uncertain significance. A comparison of the scores from the different splicing algorithms for these fourteen unique variants is shown in Table 2 and Additional file 1: Fig. S18.

Table 2 Splice site and deep intronic variants identified in the OxClinWGS cohort

Three variants involved canonical splice sites in known disease genes. These canonical sites are defined as being the two nucleotide consensus sequences for the 5′ splice donor (GT) and 3′ acceptor sites (AG) and would be expected to be included in flanking regions of exons captured by targeted and exome sequencing so could, in principle, be detected by standard of care testing.

The first canonical splice site variant, a c.1032 + 1G > C variant in CHRNE, was identified in a patient with congenital myasthenia having been missed on clinical testing. This could be observed when the original Sanger sequencing traces were reviewed retrospectively after WGS, demonstrating that review of previous testing results prior to ordering WGS may be useful.

A second canonical splice site variant in a patient with microcephaly resulted from a 3 bp deletion at the end of exon 20 of RTTN, a gene known to be associated with this condition. The SpliceAI score was high (0.91) emphasising the utility of these algorithms for identifying splice site variants created by small INDELs.

A third canonical splice site variant was a de novo splice donor variant (c.2345 + 1G > A), predicted to be pathogenic, in the known microcephaly-associated gene, WDFY3 (OMIM #617520) [101, 102] in a foetus with congenital brain anomalies including small cerebellum seen on a pre-natal scan. This was in addition to a de novo missense p.(Glu237Gly) variant in KIF5C (Additional file 1: Fig. S19), a gene in which pathogenic heterozygous variants cause cortical dysplasia with other brain malformations (CDCBM2, OMIM #615282) [103]. Both SpliceAI and ALTSPLICE predicted the loss of a donor site in WDFY3 on the reverse strand and weakly predicted a donor gain, which would give rise to exon skipping resulting in nonsense-mediated decay (NMD) or an alternative isoform, respectively. The presence of two de novo pathogenic mutations in known disease genes suggests that this patient may have a blended phenotype which may explain the severity of the patient’s microcephaly (see Additional file 2 for further discussion). An early clinical exome study suggested that up to 5% of RD patients may have a phenotype due to two or more single gene defects [104], a value replicated in a more recent study of WES data from 7374 patients [105]. We have not been able to confirm these WDFY3 and KIF5C variants in patient-derived cells, as the original sample was from a termination of pregnancy and no further samples were available, but reporting the additional variant in WDFY3 should be considered given the possibility of gonadal mosaicism [106] and, consequently, the implications for reproductive risk assessment.

Seven non-canonical splice site variants in known genes were identified. The first was just outside the canonical splice site (c.1175-3C > A) of SLC34A1 and validated by minigene assay (Additional file 1: Fig. S20) which, together with a second variant c.241dup p.(Glu81fs), confirmed the diagnosis of nephrocalcinosis in this patient.

A c.1512-16A > G variant in the polypyrimidine tract of SEC23B decreases splicing efficiency leading to skipping of exon 14. The variant was identified in a patient originally thought to have hereditary spherocytosis (HS) but no pathogenic variants had been found in genes associated with HS in this patient. A second pathogenic variant in SEC23B, c.40C > T p.Arg14Trp, was also identified in this patient. SEC23B is not a gene associated with HS but is known to be associated with the recessive disorder, congenital dyserythropoietic anaemia (CDAII), which is often mistaken for HS. The WGS finding required extensive investigations of the patient’s blood cells using electron microscopy to confirm the change in clinical diagnosis to CDAII. This highlights the need for the resources, in terms of clinical expertise and costs, which may be required to validate findings arising from WGS, and can be challenging in a clinical setting and requires research support. The intronic c.1512-16A > G variant would not have been included on the targeted panel used for conventional clinical testing.

A third, non-canonical splice site variant, c.135 + 26A > G, was identified in the recessive gene, ABCB4, in a patient with genetic cholestasis disease and confirmed by a minigene assay (Additional file 1: Fig. S21). The first hit, a c.2200G > T p.(Glu734*) stop gain variant, had been identified by conventional clinical testing, but this second hit from WGS finally provided a genetic diagnosis for this young patient.

Four non-canonical splice site variants were identified in VHL. Whilst biallelic VHL variants are known to cause congenital erythrocytosis, the condition with which these patients were referred, these variants were too deep into the introns (> 100 bp from exon/intron boundary) to have been picked up by conventional testing or exome sequencing. In three of the patients, a known pathogenic variant, p.(Arg200Trp), had already been identified as a first hit prior to WGS. WGS identified the same second hit, c.340 + 770 T > C VHL variant in the three patients, which resulted in dysregulation of splicing and retention of a cryptic exon, and was confirmed by functional studies to be pathogenic [107]. An additional deep intronic homozygous variant in VHL, c.340 + 816A > C, was identified in another patient with congenital erythrocytosis, which was also confirmed to be pathogenic (Betty Gardie, personal communication).

Three splice site variants were identified in genes that are novel or putative novel disease genes, or were novel at the time we identified them, and therefore would not have been investigated by conventional testing, but are canonical splice site variants or are close to intron/exon boundaries. A c.764 + 5G > A variant was identified in MCM10, in trans with a c.236del p.(Gly79fs) in a patient with restrictive cardiomyopathy. The variants were found to have constrained telomerase activity leading to stalled replication forks and telomere shortening, confirming their pathogenicity [92]. A splice site variant identified in DOCK7 (c.5724 + 1G > T), in a patient with seizures and severe ILD associated with microcephaly, was a novel disease gene at time of sequencing our patient, but was subsequently reported in patients with developmental and epileptic encephalopathy [108]. This patient also had a duplication in TP53BP2. A third splice site variant, in HDLBP, leads to loss of an exon in this RNA-binding protein. Whilst a canonical splice site, this gene has not previously been reported as a human disease gene but its likely pathogenicity is supported by segregation in five affected individuals in our consanguineous family, by functional data (loss of RNA binding) and by GeneMatcher hits, including families cited in a previously published report [109].

Three VUS were identified in potential candidate disease genes. A deep intronic, de novo variant, c.370 + 441G > A, was found in a highly conserved region of BMP4 in a patient with Kapur-Toriello syndrome. This extremely rare condition is characterised by facial dysmorphism, severe intellectual deficiency, cleft lip and palate, skeletal anomalies, ophthalmic features, intestinal and cardiac anomalies and growth retardation. Pathogenic variants in this bone morphogenetic protein have previously been associated with anophthalmia-microphthalmia and digital anomalies [110] (OMIM #607932) and cleft-lip palate [111] (OMIM #600625) and indeed the patient presented with several features of this syndrome including anophthalmia and clefting, and hence, BMP4 is considered a good candidate by the referring clinician. Current splice algorithms did not suggest that this variant introduced a novel cryptic splice site and indeed minigene experiments did not support an effect on splicing (Additional file 1: Figs. S22 and S23). The variant is predicted by deepHaem to be located within a weak open chromatin site in embryonic lung fibroblasts, creating a sequence which quite closely matches the consensus for a HOX(A/B/D)13 TF binding site (Additional file 1: Fig. S24), a prediction that is currently being experimentally tested. This case demonstrates the challenges associated with validating deep intronic variants which are not cryptic splice site variants and do not have any clear regulatory annotation.

Two further variants of uncertain significance were identified in a patient presenting with T-cell negative, B-cell positive and NK-cell positive (T-B + NK +) severe combined immunodeficiency (SCID) and limb malformations. A de novo, missense variant, c.587A > G p.(Asn196Ser) in VDAC2, was predicted by both SpliceAI and ALTSPLICE to lead to creation of an alternative, in-frame splice acceptor site resulting in a new isoform with one less codon, and a reduction in the use of the canonical splice acceptor site, results which have been confirmed by RT-PCR (Additional file 1: Fig. S25). VDAC2 has not yet been recognised as a disease-causing gene, although it has been described to have a central role in determining thymocyte survival through its regulation of the pro-apoptotic protein, BAK2 [112] and is therefore a possible candidate underlying the patient’s immunodeficiency. We also note the presence of a second de novo variant in this patient, a c.1437 + 2 T > C in intron 11 of INPP5D, which is predicted to weaken the canonical splice donor site efficiency and result in usage of an alternative acceptor site leading to a premature stop codon. Again, this gene has not previously been described to be associated with human disease, although in vitro studies have shown that INPP5D (also called SHIP1) is a protein phosphatase which regulates the PI3K signalling pathway and plays a key role in both T cell biology [113] and mammalian skeletal development [114] and it could contribute to one or both elements of the patient’s phenotype. Since the patient had been treated with haematopoietic stem cell transplantation for genetically undefined T-B + NK + SCID during infancy, we could not confirm the splice site variant in blood and, consistent with GTEx predictions, we could not detect it in patient-derived skin fibroblasts. Expression in additional cell types is now being investigated.

Overall, we found the SpliceAI and ALTSPLICE scores to show good correlation (Additional file 1: Figs. S17 and S18) but ALTSPLICE provided additional annotation and experimental hypotheses to test.

Somatic mosaic variants

Although all cases were referred for germline testing, we highlight the importance of considering somatic variants for RD patients with features of overgrowth syndromes. One patient was referred with Klippel-Trenaunay syndrome, a rare disorder, presenting at birth, characterised by vascular and lymphatic anomalies and abnormal veins in association with overgrowth of soft tissue and bone. Some cases of familial inheritance or de novo germline variants have been described [115]. Germline analysis of our patient revealed a de novo c.535 T > G, p.(Lys179Val) variant in RBPJ. Although the variant involves the last base in exon 6, RNA analysis indicated no effect on splicing, consistent with in silico predictions (Additional file 1: Fig. S26), so the variant was not considered pathogenic. Somatic mutations occurring shortly after birth in the primitive cells destined to become the blood and lymphatic vessels have been described to be causative for this condition [115], but often occur at very low frequency making them difficult to detect by standard coverage WGS. Indeed, we identified a somatic, mosaic c.3140A > G p.(His1047Arg) PIK3CA mutation in our patient at low (8%) frequency using targeted high coverage NGS.

Clinical impact

Our results informed the diagnosis of RD patients in this cohort and, additionally, influenced treatments (see Fig. 2).

Impact on clinical diagnosis

For six patients, the genetic diagnosis led to a change in the clinical diagnosis with the identification of pathogenic variants from WGS. The diagnosis of a patient referred with Aicardi syndrome was changed to DEE1 on discovery of a structural variant in ARX (see SV section above). Two patients referred with Fine Lubinsky syndrome and found to have pathogenic variants in POR and SLC39A13, respectively, had their clinical diagnoses changed to Antley-Bixler and spondylocheiro-dysplastic Ehlers-Danlos syndromes, respectively. The clinical diagnosis of two brothers referred with familial juvenile hyperuricemic nephropathy was revised to papillorenal syndrome following identification of a PAX2 pathogenic variant, a diagnosis which was confirmed by ophthalmological investigations [116]. A family originally diagnosed with Majeed syndrome had their diagnosis changed to PAPA syndrome on identifying a PSTPIP1 variant whilst another family received a revised clinical diagnosis of CDAII further to identification of a SEC23B variant, when originally diagnosed with HS (see the “Results” section).

In other cases, the genes identified were novel at time of discovery in our WGS programme, including DOCK7 for ILD and SAMD9L for autosomal dominant ataxia-pancytopenia syndrome; therefore, they would not have been picked up by prior testing and genetic diagnoses could be provided for the first time for these patients. Phasing of the de novo variants in SAMD9L was undertaken using nanopore long read sequencing [83, 117].

Expansion of phenotypic spectrum

For some cases, variant identification expanded the phenotypic spectrum associated with a given gene. Pathogenic variants in RMND1 were originally described for a patient with combined oxidative phosphorylation deficiency 11 (COXPD11), characterised by neonatal hypotonia and lactic acidosis as well as infantile onset renal failure, hearing loss and multi-organ defects. More recently the genotype–phenotype spectrum of this mitochondrial disorder has been expanded [118] but polymicrogyria, which was confirmed on neuropathological examination of the foetus’ brain, has not previously been reported in RMND1-related disorder and this, in the context of arthrogryposis, may be an early indicator of an RMND1 disorder (Additional file 1: Fig. S27).

Impact on treatment

Provision of a genetic diagnosis can have a profound impact on the treatment of patients, and in five of our patients, it was considered life-saving; a family diagnosed with the newly described arrhythmia syndrome, cardiac ryanodine receptor (RyR2) calcium release deficiency syndrome [119], was prescribed flecainide for protective effect against ventricular arrhythmia and sudden cardiac death [120].

A 32-year-old woman who had been admitted to an emergency department with presumed meningoencephalitis, and who had had several prior episodes of coma, was found to have biallelic variants in CFI, indicating a non-classical presentation of Complement Factor I deficiency. Her sister had died of fulminant haemorrhagic leukencephalopathy at the age of 16 years, demonstrating the severity of this condition if left undiagnosed and untreated. The clinical management of this patient’s condition involves optimisation of vaccination strategies and prophylactic use of antibiotics [121].

The identification of a variant in SLC5A7 in a patient with congenital myasthenic syndrome (CMS) indicated that the patient had the very rare and severe CMS Type 20 [122], characterised by life-threatening respiratory episodes, which benefit from cholinesterase inhibitors, and potentially, salbutamol.

An adult patient with congenital neutropenia and inflammatory bowel disease, a condition which can lead to fatal infections, showed clinical remission of his G6PC3 deficiency further to haematopoietic stem cell transplantation (HSCT) [123].

Identification of compound heterozygous variants in NPHP3 provided a diagnosis of nephronopthisis type 3 in a family with fibrotic kidney disease. Confirmation of genetic status in a clinically unaffected sibling enabled a successful kidney donation to his affected brother, who would otherwise have had to wait for a deceased donor kidney.

Variants in genes linked to perturbations in metabolism provided readily accessible treatments; one patient’s congenital erythrocytosis was found to be a consequence of hypermanganesaemia caused by a missense variant in the manganese transporter SLC30A10. This gene would not have been routinely screened when erythrocytosis is the primary referral condition, although erythrocytosis is a known feature of these transporter defects due to inhibition of the iron centre of the hypoxia-inducible factor prolyl hydroxylase enzymes by the retained transition metals. Early diagnosis allows patients to be treated before the accumulation of manganese deposits causes irreversible damage to the central nervous system and liver. As a result of the WGS finding, our patient was treated with manganese-chelating drugs, sparing her the need for phlebotomy and, potentially, any longer-term organ damage. A variant in SLC4A1 giving rise to pseudohypokalaemia was the cause of another patient’s apparent potassium deficiency, resulting in cessation of supplements.

Finally, the patient diagnosed with Klippel-Trenaunay syndrome, with splenic and hepatic haemangiomas and telangiectatic lesions of the right hindquarter (discussed above), was found to have a somatic mosaic mutation in PIK3CA. This patient is now being recruited to EPIK, a randomised controlled study of the PI3K inhibitor, alpelisib, for treatment of PIK3CA overgrowth syndromes.

Secondary findings

We identified two patients with secondary findings in this cohort. The first was a variant in FBN1, a gene known to be associated with Marfan syndrome. Using the framework for secondary findings that we had established [34], the patient was re-consented for her interest in receiving secondary findings and referred to the cardiovascular genetics clinic. She was found to have mild aortic root dilatation (41 mm at Sinus of Valsalva) o