Article Text

Download PDFPDF
Original research
Whole-genome analysis as a diagnostic tool for patients referred for diagnosis of Silver-Russell syndrome: a real-world study
  1. Ahmed S N Alhendi1,
  2. Derek Lim2,
  3. Shane McKee3,
  4. Meriel McEntagart4,
  5. Katriona Tatton-Brown4,
  6. I Karen Temple1,
  7. Justin H Davies1,
  8. Deborah J G Mackay1,5
  1. 1 Human Genetics and Genomic Medicine, Faculty of Medicine, University Hospital Southampton NHS Foundation Trust, Southampton, UK
  2. 2 Department of Clinical Genetics, Birmingham Women's and Children's Hospital, Birmingham, UK
  3. 3 Department of Genetic Medicine, Belfast City Hospital, Belfast, UK
  4. 4 Department of Clinical Genetics, St George's Healthcare NHS Trust, London, UK
  5. 5 Wessex Regional Genetics Laboratory, Salisbury NHS Foundation Trust, Salisbury, UK
  1. Correspondence to Professor Deborah J G Mackay, University of Southampton, Southampton SO17 1BJ, UK; djgm{at}soton.ac.uk

Abstract

Background Silver-Russell syndrome (SRS) is an imprinting disorder characterised by prenatal and postnatal growth restriction, but its clinical features are non-specific and its differential diagnosis is broad. Known molecular causes of SRS include imprinting disturbance, single nucleotide variant (SNV), CNV or UPD affecting several genes; however, up to 40% of individuals with a clinical diagnosis of SRS currently receive no positive molecular diagnosis.

Methods To determine whether whole-genome sequencing (WGS) could uncover pathogenic variants missed by current molecular testing, we analysed data of 72 participants recruited to the 100,000 Genomes Project within the clinical category of SRS.

Results In 20 participants (27% of the cohort) we identified genetic variants plausibly accounting for SRS. Coding SNVs were identified in genes including CDKN1C, IGF2, IGF1R and ORC1. Maternal-effect variants were found in mothers of five participants, including two participants with imprinting disturbance and one with multilocus imprinting disorder. Two regions of homozygosity were suggestive of UPD involving imprinted regions implicated in SRS and Temple syndrome, and three plausibly pathogenic CNVs were found, including a paternal deletion of PLAGL1. In 48 participants with no plausible pathogenic variant, unbiased analysis of SNVs detected a potential association with STX4.

Conclusion WGS analysis can detect UPD, CNV and SNV and is potentially a valuable addition to diagnosis of SRS and related growth-restricting disorders.

  • genetics
  • medical
  • genomics

Data availability statement

No data are available. WGS data and associated clinical data are held within the 100,000 Genomes Project Research Environment. These data can be accessed by any researcher by application to join a GeCIP domain (www.genomicsengland.co.uk/join-a-gecip-domain/). Informatic scripts are available upon request.

Statistics from Altmetric.com

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Introduction

Growth restriction is a feature of many congenital conditions, and clinical decisions are required to determine which individuals to treat and in what way. Since growth variation is continuous, not categoric, qualitative clinical criteria must be applied to make prognoses and select treatments. Moreover, pinpointing the underlying pathogenesis enables a more accurate prognosis using evidence-based medicine.

Silver-Russell syndrome (SRS) affects approximately 1 in 15 000 children and is characterised by intrauterine and postnatal growth restriction, along with relative macrocephaly and prominent forehead in infancy, body asymmetry and significant feeding difficulties.1 2 These six clinical features form the Netchine-Harbison Clinical Scoring System (NH-CSS)3: individuals meeting three of the six criteria warrant referral for molecular diagnosis, whereas four of six features constitute a clinical diagnosis of SRS irrespective of molecular diagnosis.

In many countries a formal diagnosis of SRS is important for accessing treatment, and early, accurate molecular diagnosis optimises clinical management and genetic counselling. However, the genetic causes of SRS are numerous and account for fewer than 60% of clinically diagnosed patients. Approximately half have loss of methylation (LOM) at imprinting centre 1 (IC1, also known as the H19:IGF2 intergenic differentially methylated region or H19:IGF2 IG-DMR) on chr11p15, or rare CNV or UPD involving chr11p15.1 5%–10% of cases show UPD of chromosomes 7, 14, 20, 6 and 16; the (epi)genetic changes on chr14 overlap with Temple syndrome, which shares clinical features with SRS.4–6 Rare SRS cases have pathogenic single nucleotide variant (SNV) of imprinted genes on chr11p15 (CDKN1C and IGF2) or genes in the IGF2 pathway,7–11 or have multilocus imprinting disorder (MLID) potentially associated with trans-acting mutations altering imprinting control.12 In cases with none of these mutations, numerous differential diagnoses may be considered.1

The recent years have seen increasing adoption of genomic technology in diagnosis, including gene panel or exome testing.13 14 There is a general expectation that whole-genome sequencing (WGS)—analysis of unbiased, uncaptured sequence from genomic DNA—will be a useful tool in the diagnosis of short stature syndromes. In the 100,000 Genomes Project (100KGP), National Health Service (NHS) clinicians from a range of specialties worked with genome medicine centres (GMCs) across the UK to recruit families for whom standard NHS testing had not identified a cause of disease. Recruiting clinicians were responsible for choosing best-fit categories for recruited participants based on their clinical features. Initial WGS analysis focused on curated gene lists for the disorders for which participants were referred, and high-confidence pathogenic findings were returned via the GMCs. In parallel, research groups in the Genomics England Clinical Interpretation Partnership studied anonymised WGS data, seeking novel genetic signatures of disease.

Seventy-three participants were recruited to 100KGP with a clinical suspicion of SRS. We analysed WGS data from these participants and their families to establish whether whole-genome testing was a useful way of uncovering plausible pathogenic variants associated with clinical features of SRS.

Materials and methods

Participants

WGS data were available from 72 participants recruited in the SRS category and their family members. One participant had data in Hg37 and was excluded from further analysis. For the participants studied, 61 pedigrees included both parental samples, 7 had the mother’s sample only, 1 had the father’s sample only and 1 proband had neither parental sample. In families 7, 30, 41 and 60, WGS was additionally available from siblings; in each of these families some siblings had reported a reported clinical presentation, which was SRS in all cases but one. For other families, no data on siblings were available.

Participants were recruited by clinicians across UK GMCs, and participants were anonymised. DNA was not available for additional studies including imprinting analysis; data were analysed within the Genomics England Research Environment; clinical data were limited to demographics and Human Phenotype Ontology (HPO) terms; and contact with referring clinicians was mediated through the Research Environment. However, 12 participants were known to us through participation in the research study ‘Imprinting disorders – finding out why’ (IDFOW). IDFOW participants gave informed consent for molecular investigation into the cause of their disease.

WGS data analysis

The data analysis strategy is summarised in figure 1.

Figure 1

Data analysis strategy. The four columns of the figure illustrate the four informatic processes applied to sequence data from all 72 participants and their families. The boxes at the foot of each column summarise the plausible pathogenic variants uncovered by each process. SNVs marked with an asterisk were identified by genome medicine centres and returned to referring clinicians as the likely causative variant. 100KGP, 100,000 Genomes Project; CADD, Combined Annotation-Dependent Depletion; IBD, identity by descent; IC1, imprinting centre 1; LOM, loss of methylation; MAF, minor allele frequency; ROH, region of homozygosity; SKAT-O, optimised Sequence Kernel Association test; SNV, single nucleotide variant; SRS, Silver-Russell syndrome.

Putative genes and regions of interest

A panel of genes and genomic regions of interest were compiled from several sources, including PanelApp, differential diagnoses from the International Clinical Consensus, known imprinted regions associated with growth-restricting imprinting disorders, and genes associated with growth in published reports (online supplemental table S1).1 13 15 16

Supplemental material

SNV and short indels

Variants passing QC metrics (missingness <5%, coverage ≥15, GQ ≥15, allelic fraction ratio ≥0.25) were reanalysed using ANNOVAR V.1.0.17 Potentially pathogenic variants were filtered as minor allele frequency (MAF) <0.01 in GNOMAD and <0.03 in 100KGP; Combined Annotation Dependent Depletion deleteriousness score >15; present in the coding sequence of the canonical transcript; and predicted benign by neither SIFT nor PolyPhen-2. Filtered variants from the panel of genes of interest (online supplemental table S1) were examined to exclude those likely benign in ClinVar and those not segregating with phenotype.

Copy number variation

CNVs were identified using Canvas18 and Manta.19 CNVs passing standard Illumina quality metrics were aggregated from participants with SRS and related growth disorders (n=1290 participants), then selected if greater than 10 kb, MAF <0.01 in the study cohort, overlapping exonic regions of protein-coding genes or ‘classical’ imprinted regions (chr 6, 7, 11, 14, 15, 20), and not found in apparently healthy participants. Filtered CNVs were examined to exclude those common in GNOMAD-SV20 and likely benign in DECIPHER.21

Regions of homozygosity

Regions of homozygosity (ROH) consistent with uniparental isodisomy (iUPD) were detected using Illumina ROHcaller as extended runs of homozygous calls in a proband with altered allele fraction from parents and without alteration in read depth. Theoretical proportion of identity by descent (IBD) was calculated from the total number of ROHs per sample as a predictor of consanguinity.22 Copy-neutral ROHs >1 Mb that overlapped growth-related imprinted loci in probands with IBD <0.5% were reported as potential segmental iUPDs.

Analysis of H19 region

To overcome shortcomings in mapping of the IC1 region, reads spanning IC1 and IC2 (chr11:1670000–3000000, GRCh38) were extracted from WGS bam files and realigned using BWA-MEM, followed by variant calling using GATK423 and CNV calling using Manta.19

Ethnicity PCA analysis

To circumvent spurious association, participants with SRS with plausible genetic causes (n=20), high consanguinity (n=1) and known 11p15 LOM (n=3) were excluded. Data from the remaining 48 participants and 8204 unrelated (kinship coefficient 0.04419417) cancer germline samples from 100KGP were subjected to PCA ancestry analysis as implemented in PC-AIR.24 Case and control samples were prepared in PLINK file format using PLINK V.1.9, and PCs and loading for the samples were calculated using 30 000 autosomal SNPs (MAF >5%, coverage >99%, inbreeding coefficient >−0.1, LD pruning R2 <0.1, HWE p>0.01) restricted to variants present in the Genomics England aggregated data set and 1KGP data set.

Gene burden test of protein-coding variants

Filtered variants were collapsed per gene, and subjects with at least one variant were counted per group and tested for association with growth restriction. Fisher’s exact test with Bonferroni correction for multiple testing (p<0.05) was used to determine genome-wide significance.

SKAT-O test of protein-coding variants

The optimised Sequence Kernel Association (SKAT-O) test was applied to rare variants to increase the power of discovery under different inheritance models by combining variance component and burden tests. Filtered variants as specified above were analysed using the implemented SKAT-O function in SKAT R package V.2.0.1 with default parameters for binary SKAT-O.25 The fitting model was corrected for SNP weight using MAF, gender and the first five principal components of the ethnicity PCA described above. Variants were collapsed considering only the protein-coding region in canonical transcripts in GRCh38.

Genomic visualisation

Sequencing reads for SNVs and short indels, and breakpoints of deletions and duplications, were visualised for validation on Integrative Genomics Viewer.26 For segmental UPDs and large CNVs, variant allele fraction and sequencing depth were visualised using R V.3.5.1.

Methylation analysis of MLID cases

Imprinting was assessed by targeted analysis as previously described27 or by methylation-specific multiplex ligation probe-dependent amplification assay (ME30, ME032 and ME034; MRC Holland, Amsterdam, The Netherlands).

Results

Clinical and epigenetic features

Participant data included year of birth (range 1981–2017), genetic sex (39 male, 33 female), ethnicity, family members (affected or unaffected) and clinical features reported as HPO terms. A median of 7 clinical features were reported with a range of 0–17 features. Of the participants 50% had ≥3 features of the NH-CSS (online supplemental table S2). Some had features atypical of SRS, most notably 16 participants with reported intellectual or developmental delay. This is a feature of some subtypes of SRS, but it likely reflects a selection bias towards atypical SRS in this real-world cohort.

Two parents of participants had reported phenotypes, one SRS and one intellectual disability. In four families, siblings of participants had reported phenotypes, in one case intellectual disability and in all others SRS. In two participants, GMCs reported likely pathogenic variants.

Although epigenetic aberration was not an exclusion criterion, cases with known imprinting disturbance appeared to be under-represented in recruitment, perhaps because many participating clinicians considered such cases as ‘diagnosed’. Five participants had reported IC1 LOM (4 of 5 had ≥4 NH-CSS features), 2 had reported normal methylation at IC1 and in 65 participants methylation status was not stated.

Findings from WGS using a gene panel of genes and regions implicated in SRS

WGS analysis uncovered 23 potentially pathogenic genetic variants affecting 20 participants (table 1). These variants, and additional SNVs that potentially contribute to the clinical presentation of participants, are listed in online supplemental table S3.

Table 1

Clinical features of participants

Coding SNVs

Initial analysis focused on rare, protein-altering SNV in an extensive list of genes associated with growth and growth restriction (online supplemental table S1). Three participants had variants within the genes on chr11p15 directly associated with SRS (online supplemental figure S1). One had the maternally inherited CDKN1C variant R281I, and two had paternally inherited coding variants in IGF2.

Five mothers of SRS probands had heterozygous variants in NLRP2, NLRP7 and PADI6. Among these, the two with maternal NLRP7 variants were among those with LOM of IC1; two with maternal NLRP2 variants had no detected methylation disturbance, and the participant with a maternal PADI6 variant had MLID, including LOM of the 14q32 IG-DMR, consistent with Temple syndrome (figure 2).

Figure 2

SNV in PADI6 present in a participant with MLID and the mother. (A) Pedigree information for the family. (B) IGV visualisation of the relevant sequence of PADI6 in the participant, mother and father showing the variant in the participant and the mother. (C) Electropherograms of targeted DNA methylation analysis in the participant. The peaks represent amplification products from maternally derived (red bars) and paternally derived (blue bars) sequence. The MEG3 DMR on chr14q32 is paternally methylated and the WRB DMR on chr21q22 is maternally methylated. Compared with DNA from a normal control individual (upper traces), the participant’s DNA (lower traces) is hypomethylated at both loci. DMR, differentially methylated region; IGV, Integrative Genomics Viewer; MLID, multilocus imprinting disorder; SNV, single nucleotide variant.

Compound heterozygosity for pathogenic variants was found in two probands: one involved ORC1, associated with Meier-Gorlin syndrome; another involved IGF1R (online supplemental figures S2 and S3A). One participant inherited an IGF1R SNV from a father also reportedly affected by short stature (online supplemental figure S3B). One participant inherited a variant in LZTR1, associated with Noonan syndrome; another had a variant in PIK3R1, associated with SHORT syndrome, although lack of parental samples prevented determination of inheritance (data not shown).

Two diagnoses were made by the 100KGP GMC. One participant had pathogenic variants in SON and SCN8; another had a pathogenic variant in KDM6A (data not shown). Of these only KDM6A was in our gene panel.

Structural variants

Of 30 SVs found in participants (online supplemental table S4), one was a maternally inherited 410 kb deletion involving IGF1 (figure 3A); another was a mosaic deletion affecting the paternally inherited allele of chr6q24, including PLAGL1 (figure 3B). In the third participant, a maternally inherited deletion of chr17q24.2, including PSMD12, was potentially relevant to clinical presentation (online supplemental figure S4); other SVs were likely benign.

Figure 3

Plausible pathogenic CNVs detected in two participants. (A) Maternally inherited heterozygous deletion affecting IGF1. Upper panel: left, pedigree information for the family; right, ideogram showing the location of deletion. Middle panel: visualisation of read depth around the 411 kb deletion in the participant and the mother. Lower panel: IGV visualisation of breakpoints of the deletion showing discordant reads in the participant and the mother. (B) De novo heterozygous, mosaic deletion affecting PLAGL1. Upper panel: left, pedigree information for the family; right, ideogram showing the location of deletion. Middle panel: visualisation of read depth ratios for the participant and both parents (log scale: 0=diploid read depth; black dots represent reads with depth ratio ≥0; grey dots represent reads with depth ratio <0 indicating loss). Lower panel: AF visualisation from het SNPs in the participant, mother and father, coloured by parental origin (blue: paternal; red: maternal). Weighted regression method was used to detect trends in AF data from paternal and maternal alleles and demonstrate a significant reduction in the average AF of paternal allele compared with maternal allele in the participant. AF, allele fraction; F, father; IGV, Integrative Genomics Viewer; M, mother; MAT, maternal; P, participant; PAT, paternal; SNV, single nucleotide variant.

Regions of homozygosity

ROH was identified as a proxy for UPD. Two participants showed >1% IBD, but their ROH did not overlap genes or regions currently associated with SRS, nor did they overlap one another. Two participants showed extensive regions of maternal homozygosity: one spanned the imprinted region around chr14q32, consistent with Temple syndrome (figure 4); another on chromosome 7 included the imprinted gene PEG10 (online supplemental figure S5), but heterodisomy for the remainder of chr7 could not be assessed due to paternal DNA being unavailable.

Figure 4

Plausible pathogenic region of homozygosity affecting the chr14q32 imprinted region. (A) Ideogram showing location of deletion. (B) Visualisations of allele fraction and depth ratio in the proband (P), mother (M) and father (F) for 15 Mb at 14qter. The allele fraction visualisation of the participant (top panel) shows homozygosity of SNPs in a 10.4 Mb region that encompasses the imprinted 14q32 region, while the read depth ratio (second panel) is normal, showing no loss or gain of genetic material. (C) Representation of the imprinted region on 14q32, indicating genetic location of key features in Hg38. Filled blue rectangle: paternally expressed coding gene, DLK1; red unfilled rectangle, maternally expressed non-coding RNA, MEG3; blue filled circles, the germline imprinting centre MEG3-DLK1 IG-DMR and the somatic MEG3 TSS-DMR. (D) IGV visualisation encompassing the MEG3-DLK1 IG-DMR and MEG3 TSS-DMR showing homozygous SNPs in the participant co-located with heterozygous SNPs in the mother, indicating isodisomy of maternal origin with no paternal contribution. IG-DMR, intergenic differentially methylated region; IGV, Integrative Genomics Viewer.

Non-coding variants affecting IC1

Five participants had IC1 LOM. Since the standard analysis pipeline of 100KGP did not resolve the IC1 region, presumably because of its repetitive structure, the region was remapped, to seek cis-acting variants potentially predisposing to LOM in these participants. No rare CNVs were detected within the reannotated region. One participant had a rare paternally inherited variant within IC1: chr11:2000298G>A (MAF 0.00016), which does not overlap described CTCF-binding or ZFP57-binding motifs (online supplemental figure S6).

Rare variant association tests

In 48 participants with no plausible pathogenic variants identified from our gene panel and with IBD <0.5%, aggregated filtered variants were grouped per gene. In this data set, the aggregated effect of rare variants was assessed using Fisher’s exact test and the SKAT-O test (online supplemental figure S7). SKAT-O combined variance component and burden tests and enabled correction for ethnicity and gender. While uncorrected burden testing identified five genes with p<−0.001, but with no obvious inflation (figure 5A,B), after correction using SKAT-O only one gene, STX4, remained significant (3 of 48 participants, compared with 30 of 8204 controls; p=5.14e-08) (figure 5C,D). Two heterozygous missense variants were found in three participants: V57M, maternally inherited, in participant 6, and M159I in participants 42 and 57, in maternal and paternal inheritance, respectively (table 1, online supplemental table S3). Rare variants in STX4 thus suggested a potentially significant association with SRS.

Figure 5

Identification of putative novel genetic associations with SRS through unbiased analysis. (A, B) Results of gene burden testing of rare protein truncating variants and likely deleterious missense variants. (C, D) Results of SKAT-O testing, corrected for principal component analysis of ancestry, gender and upweighting of rare variants. (A, C) Manhattan plots of genes in which variants are enriched in the case cohort (48 individuals with SRS) compared with 8204 aggregated control genomes. The negative decadic logarithm of unadjusted p values is plotted against the chromosomal location of each gene. Genes surpassing the significance threshold are named. (B, D) Q-Q plots of burden and SKAT-O analyses, respectively, clearly showing STX4 as the only gene observed significantly more than expected in the SRS cohort. SKAT-O, optimised Sequence Kernel Association test; SRS, Silver-Russell syndrome.

Discussion

In a group of 72 individuals recruited to 100KGP in the category of SRS, WGS data analysis revealed potentially pathogenic genetic SNV, CNV and UPD crossing every molecular lesion currently associated with SRS, and also variants associated with disorders phenotypically very different from SRS.

The 100KGP was a pragmatic study based on real-world sampling of patients seen within routine clinics who could benefit from WGS. Recruiting clinicians were responsible for choosing best-fit categories for recruited participants, and hence we assume that SRS was a reasonable diagnosis in this cohort. The reported clinical features were insufficient for a provisional diagnosis of SRS in 50% of participants, but it was impossible to ascertain whether other SRS features were present but not recorded. This cohort is representative of clinical practice, where clinical data are often incomplete; however, it was additionally challenged by limited epigenetic data recording and lack of contact with referring clinicians. In a routine clinical scenario, many of the discoveries reported here would be followed up by confirmatory phenotyping and testing, which were not possible with the design of this study, except for participants who were also part of the imprinting study, IDFOW.

Whereas 30%–60% of individuals with clinically diagnosed SRS have LOM of IC1,1 in this cohort only five participants were reported to have LOM. In these participants, remapping and analysis of the IC1 region revealed one paternally inherited rare variant, but this did not overlap any known transcription-factor binding sites28 and thus was not obviously pathogenic. However, two other individuals with IC1 LOM had maternally inherited variants in NLRP7, a maternal-effect gene, and the third participant, whose mother harboured a variant in PADI6, showed MLID with epigenetic changes consistent with Temple syndrome. Therefore, of five participants whose mothers had variants in maternal-effect genes, three had imprinting disturbance detected by targeted testing. We previously observed in a research cohort with MLID that over 50% of their mothers harboured rare coding variants in maternal-effect genes; several reports now associate variants in these genes with a range of reproductive outcomes including infertility, pregnancy loss and imprinting disorders.12 29 30 This finding, in a differently ascertained cohort, makes it reasonable to consider maternal-effect variants for individuals with isolated epigenetic errors, since current epigenetic testing may underestimate the prevalence of MLID. A maternal-effect variant in a family significantly alters recurrence risks and may impact long-term prognosis.

Two participants had regions of maternal homozygosity ≥5 Mb overlying known imprinted loci on chromosomes 7q21 and 14q32. An inclusion criterion for 100KGP was exclusion of alternative diagnoses, including UPD. While the small ROH here might have eluded detection, imprinting analysis of chr7 and chr14 should have detected imprinting disturbance diagnostic for both of these participants. Because of the ethical structure of 100KGP, no confirmatory testing was possible for these participants; however, confirmatory testing is important, particularly for the individual with ROH of 14q32 because this is consistent with a diagnosis of Temple syndrome. Temple syndrome clinically overlaps SRS in early childhood, but has important differences, including very early puberty and metabolic change, which require early diagnosis for appropriate management.4 5 The SGCE/PEG10 locus at 7q21 is one of at least four imprinted regions on chr7, and not currently the prime candidate for its association with SRS.2 31 The apparent maternal isodisomy of 7q21 may be flanked by heterodisomic regions, but without paternal WGS data this could not be proved.

Related to this, one participant had mosaic deletion of the paternal allele of the imprinted gene PLAGL1, which due to its low mosaicism was missed by cytogenetic array analysis. While paternal PLAGL1 duplication is associated with transient neonatal diabetes,32 deletion has not been reported; however, maternal uniparental disomy of chromosome 6 is tentatively associated with prenatal and postnatal growth restriction.33 We believe this is the first reported case of a patient with growth restriction due to deletion of PLAGL1, a growth pattern predicted by studies in mouse.34

Several genes implicated in growth restriction were affected by putatively pathogenic coding variants. Aside from genes associated with SRS, CDKN1C and IGF2,7–10 variants affected IGF1R, ORC1 (associated with Meier-Gorlin syndrome) and LZTR1 (associated with Noonan syndrome).35 36 ,37 These findings parallel those recently obtained through exome and gene panel approaches.13 14

Further potentially pathogenic variants were identified, but detailed clinical follow-up was beyond the scope of this project.

GMCs made diagnoses in two participants. One had pathogenic variants in SON and SCN8: SON is associated with ZTTK syndrome, where poor growth is seen alongside motor and developmental delay, and both genes are outside the differential diagnosis of SRS.38 Another participant had a pathogenic variant in KDM6A, associated with Kabuki syndrome, whose clinical features include prenatal and postnatal growth restriction.39 The presence of these diagnoses in this cohort suggests either that the syndromes have broader phenotypes than currently recognised, or that the affected participants, and potentially others, did not fully clinically accord with SRS. In line with the latter suggestion, the recorded clinical features of 50% of the cohort did not meet the NH-CSS criteria for epigenetic testing.

In 48 participants, no plausible pathogenic variants were found using a focus on CNV, ROH or coding SNV in a gene panel for growth restriction. In these participants we attempted to explore beyond a known gene panel, performing gene burden analysis to identify rare coding variants enriched in participants versus a large control data set. This agnostic analysis identified a single significant gene, STX4. Syntaxins are highly conserved proteins with essential, partially overlapping roles in intracellular vesicle trafficking. Notably, STX4 is involved in insulin granule exocytosis from pancreatic beta-cells as well as insulin-stimulated glucose uptake by skeletal muscle40 41; these processes are potentially relevant to the pathophysiology of SRS. However, in all cases heterozygous variants were inherited from a parent with no reported features of SRS, and the presence of both maternal and paternal inheritance did not indicate imprinting of STX4. Further, informatic metrics assessing tolerance of genes to variants, suggest that STX4 is tolerant of mutations (pLI (probability of loss-of-function intolerance) and Z-score 9.66E-05 and 2.014, respectively).42 However, other key growth restriction genes also have metrics suggesting tolerance of variation (such as IGF2: 2.53E-02 and 2.308; GH1: 1.74E-02 and –0.351; CUL7: 2.24E-11 and 0.482), suggesting that such metrics may not be a key guide to variant interpretation in growth restriction disorders. The small size of this cohort and the unavailability of clinical follow-up impede interpretation of STX4 variants, and further molecular studies would be required to determine whether STX4 is involved in SRS, potentially in combination with other genetic or environmental factors. Notwithstanding the equivocal findings from our agnostic approach, we suggest that burden testing in large, well-phenotyped cohorts should be attempted, to go beyond panels of known genes and truly exploit the potential of WGS to identify novel causes of rare disease.

This study shows that WGS analysis is a potentially valuable tool for diagnosis of individuals with clinical features of SRS. It extends the range of genes to be considered, demonstrates the indispensability of clinical phenotyping and illustrates the challenges of incorporating WGS into routine diagnostics. More fundamentally, it shows the challenge of defining a clinical entity like SRS in a way that reflects evolving understanding of the disorder, but remains as practical and helpful as possible for diagnosis and management, for health professionals and for families.

SRS is currently defined as a clinical entity, using clinical nosology.1 As with other growth disorders, it is defined by qualitative clinical thresholds applied to continuously variable metrics of growth. Besides anthropometric criteria it recognises other features; however, all are ‘soft’, all evolve over time and none are unique. This clinical nosology was essential when SRS was defined as an entity since its molecular aetiology was unknown. As molecular causes of SRS have been defined, individuals with a positive molecular diagnosis have increasingly defined the cohort whose features define the clinical diagnostic criteria. The molecular diagnostic rate of SRS has risen to ~60%; however, the corollary of this is that SRS is now both pleiotropic—the same molecular change associated with a range of phenotypes—and heterogeneous—similar phenotypes associated with different genetic diagnoses. Concurrently, genetic diagnosis has become technologically more agnostic, since comprehensive approaches like WGS do not require a prior clinical hypothesis, and more democratic, since a wider range of health professionals may refer individuals for genetic testing. This approach, carried to an extreme, would be a genomic survey applied in the virtual absence of clinical definition, the opposite of a clinical nosology. This study arguably shows both the potential and pitfalls of such an approach, in that a group of patients collected under a relatively broad clinical definition of SRS turned out to have a broad range of genetic changes, including some scarcely related to SRS.

What is a pragmatic way forward? ‘Nosologists in all fields tend to be either ‘lumpers’ or ‘splitters’’.43 ‘Lumping’ represents clinical nosology: grouping patients in terms of their clinical presentation plainly guides their treatment, but this may include people whose genetics are contraindicated (eg, growth hormone treatment in Bloom syndrome) or exclude those who do not meet heuristic criteria (eg, higher birth weight in Temple syndrome preventing the appropriate use of growth hormone in some countries). ‘Splitting’ represents a genetic nosology, defining patients by their molecular disorder, and this guides precision medicine to the same extent as it subdivides patients—to the point where evidence-based management may become impracticable. Both approaches have evident strengths and weaknesses in a growth restriction disorder like SRS.

A useful compromise may be reached by combining a shared clinical term—SRS—to lump individuals with similar medical needs, with a degree of genetic splitting to stratify care. Such an approach, involving a dyadic (gene:phenotype) taxonomy for genetic disorders, has recently been proposed.44 If the taxonomy of SRS evolves, this may reflect an evolving definition of the clinical entity. Fresh discussion among clinicians, geneticists, healthcare providers and families may be warranted to reassess the clinical criteria that prompt testing, the genetic and epigenetic changes included and excluded in its definition, and whether it constitutes a syndrome, a spectrum or other grouping within the matrix of growth restriction conditions.

In conclusion, our study has shown that WGS has a place in diagnosing individuals with features of SRS; however, harnessing its full potential will require excellent clinical characterisation, long-term follow-up, and nuanced molecular investigation including epigenetics, mosaicism, maternal-effect mutations and imprinted inheritance patterns.

Data availability statement

No data are available. WGS data and associated clinical data are held within the 100,000 Genomes Project Research Environment. These data can be accessed by any researcher by application to join a GeCIP domain (www.genomicsengland.co.uk/join-a-gecip-domain/). Informatic scripts are available upon request.

Ethics statements

Patient consent for publication

Ethics approval

The IDFOW study received Southampton and South West Hampshire research ethics approval (07/H0502/85).

Acknowledgments

The authors gratefully acknowledge the families that participated in the 100,000 Genomes Project and the clinical colleagues who recruited participants to the project. This research was made possible through access to the data and findings generated by the 100,000 Genomes Project. The 100,000 Genomes Project is managed by Genomics England (a wholly owned company of the Department of Health and Social Care). The 100,000 Genomes Project uses data provided by patients and collected by the National Health Service as part of their care and support.

References

Footnotes

  • Contributors ASNA performed bioinformatic analysis and cowrote the manuscript. DL, SM, MM, KT-B and IKT recruited patients and contributed to the manuscript. IKT, JHD and DJGM secured funding, supervised the informatic and laboratory studies, and cowrote the manuscript. DJGM coordinated the study.

  • Funding ASNA was funded by the Child Growth Foundation, UK. IKT is supported by the NIHR Biomedical Research Centre (BRC), Southampton. The Wellcome Trust, Cancer Research UK and the Medical Research Council funded the research infrastructure. The 100,000 Genomes Project is funded by the National Institute for Health Research and NHS England.

  • Competing interests None declared.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.