Abstract
Background
Genome sequencing was first offered clinically in the UK through the 100,000 Genomes Project (100KGP). Analysis was restricted to predefined gene panels associated with the patient’s phenotype. However, panels rely on clearly characterised phenotypes and risk missing diagnoses outside of the panel(s) applied. We propose a complementary method to rapidly identify pathogenic variants, including those missed by 100KGP methods.
Methods
The Loss-of-function Observed/Expected Upper-bound Fraction (LOEUF) score quantifies gene constraint, with low scores correlated with haploinsufficiency. We applied DeNovoLOEUF, a filtering strategy to sequencing data from 13,949 rare disease trios in the 100KGP, by filtering for rare, de novo, loss-of-function variants in disease genes with a LOEUF score < 0.2. We compared our findings with the corresponding patient’s diagnostic reports.
Results
324/332 (98%) of the variants identified using DeNovoLOEUF were diagnostic or partially diagnostic (whereby the variant was responsible for some of the phenotype). We identified 39 diagnoses that were “missed” by 100KGP standard analyses, which are now being returned to patients.
Conclusion
We have demonstrated a highly specific and rapid method with a 98% positive predictive value that has good concordance with standard analysis, low false-positive rate, and can identify additional diagnoses. Globally, as more patients are being offered genome sequencing, we anticipate that DeNovoLOEUF will rapidly identify new diagnoses and facilitate iterative analyses when new disease genes are discovered.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
Introduction
With transformative advances in genomic medicine, there has been an exponential rise in the number of individuals undergoing exome and genome sequencing. A shift towards large-scale international sequencing programs is improving affordability and accessibility of such sequencing for diagnostic purposes, where conventional clinical tests have failed to yield a diagnosis (Seaby et al. 2021; Posey et al. 2019). The 100,000 Genomes Project (100KGP) was a research project embedded within the UK National Health Service and the precursor to offering whole genome sequencing (WGS) as a clinical test (Turnbull et al. 2018; Parliament 2015; The 100,000 Genomes Project Pilot Investigators 2021). This pioneering project benefited from sequencing vast patient numbers with rare genetic diseases with improved power to identify multiple patients with overlapping phenotypes and genotypes; however, the number of cases that required clinical assessment for diagnostic reporting versus resources available created a significant bottleneck.
Diagnostic rates for the 100KGP were similar to the international average for rare diseases (Rehm 2022). The flagship 100KGP paper showed that an estimated diagnostic uplift from 15 to 20% could be achieved beyond prior testing, but that the time and level of additional resources required to analyse the full genome was beyond routine diagnostic testing (The 100,000 Genomes Project Pilot Investigators 2021). As a result, the project adopted the use of predefined gene panels (Genomics England PanelApp) (Martin et al. 2019) to target sequencing analysis to the most relevant genes selected from the Human Phenotype Ontology (HPO) (Robinson et al. 2008) terms provided by the referring clinician (Fig. 1) (The 100,000 Genomes Project Pilot Investigators 2021). Whilst this approach restricted the number of variants assessed and improved the efficiency of the variant curation process applied to each patient’s genome for diagnostic reporting, it risked missing variants in genes outside of the gene panel applied. At the latter stages of the project, many NHS England clinical laboratories additionally reviewed all de novo variants and top Exomiser (Smedley et al. 2015) results, although this was never mandatory. This learning has also informed guidance for the evaluation of genome sequencing through the new Genome Medicine Service in the UK.
Accurate phenotyping is essential for gene panel selection, yet there is huge variability in the phenotypes reported by clinicians. For some cases in the 100KGP, only a single HPO term was reported. As WGS becomes more widespread, appropriate matching of HPO terms with optimal panel(s) may be less error prone for experienced geneticists but will represent a challenge for the wider community of clinicians expected to routinely refer patients. Furthermore, whilst the use of HPO terms aids in the standardization of reporting phenotype data, it represents a cross-sectional time-point analysis without resource for re-analysis. Ultimately, using HPO terms lacks the full clinical narrative and challenges gene panel selection. Therefore, there is need to expand genome analysis beyond gene panels to enable a more agnostic and comprehensive genome analysis, yet this needs to be balanced with the number of variants that require manual assessment for diagnostic reporting. Targeting variants with high pathogenic potential across the entire exome provides an opportunity to rapidly identify diagnostic variants and uplift diagnostic rates. Genotype-driven analysis approaches are complementary to the phenotype-drive approach currently utilized by 100KGP.
Looking at variation across people, some genes are extremely depleted or constrained for variation predicted to result in loss-of-function (LoF) (MacArthur et al. 2012). That is to say, there is negative selection against the loss or inactivation of one allele. By comparing the observed over the expected rate of predicted loss-of-function variants in large population databases, it is now possible to compute the degree of constraint a given gene has for inactivation (Gudmundsson et al. 2021; Karczewski et al. 2020). The loss-of-function observed over expected upper bound fraction, or LOEUF score, is a metric that places each gene on a continuous scale of loss-of-function constraint. Low scores are highly correlated with disease genes and gene essentiality, with the first LOEUF decile (< 0.2) being enriched for haploinsufficient disease genes (Fig. 2) and the greatest burden of OMIM disease entries (Karczewski et al. 2020). Loss-of-function variants in extremely LoF constrained genes are therefore prime targets for potential diagnoses.
We aimed to utilise the sequencing and phenotype data generated through 100KGP and apply a transferable and rapid filtering method, we called DeNovoLOEUF, that can screen for putative pathogenic variants. We developed a rapid, agnostic approach to target the highest diagnostic yield variants in rare disease patients whilst enabling clinical curators to focus on the most important findings, regardless of the gene panel applied, improving efficiency for cases where a diagnosis could be rapidly identified.
Materials and methods
Data access
We obtained access to the GEL research environment (RE) and high-performance cluster (HPC) behind a secure firewall following information governance training and with membership of a Genomics England Clinical Interpretation Partnership (GeCIP): Quantitative methods, machine learning, and functional genomics. We had a project approved (RR359—Translational genomics: Optimising novel gene discovery for 100,000 rare disease patients) which permitted access to anonymised 100KGP sequencing and phenotype data. This included an aggregate vcf file comprising 13,949 rare disease trios with de novo variants, called using the Illumina Platypus pipeline (The 100,000 Genomes Project Pilot Investigators 2021).
Phenotype data
Referring clinicians recorded phenotype data as categorical HPO terms. These were accessible in the RE by querying HPO terms stored in mysql tables in a LabKey data management system. Gene panels were selected by GEL based on the phenotype terms provided. A summary of high-level phenotypes of the patient population is available in Supplementary Data 1.
Data analysis
Data analysis was performed in Autumn 2019 (Fig. 3). Bespoke scripts were developed to query the aggregate vcf file. We selected only variants that passed Illumina QC as previously described (The 100,000 Genomes Project Pilot Investigators 2021). We then applied a filtering strategy called DeNovoLOEUF: First, we selected de novo variants with an allele frequency < 0.001 in gnomAD v2.1.1 (all populations). We further restricted this list to predicted loss-of-function (pLoF) variants including nonsense, frameshift and essential (canonical ± 2 base-pairs) splice site variants. We imported LOEUF constraint gene scores, downloaded from the gnomAD browser (https://gnomad.broadinstitute.org), into the research environment. We then restricted rare, de novo, pLoF variants to genes with a LOEUF score of < 0.2 (n = 1044), approximately equivalent to the first LOEUF decile, representing a list of genes most highly constrained for loss-of-function and predicted to cause disease through haploinsufficiency as outlined in the flagship gnomAD paper and visualised in Fig. 2 (Karczewski et al. 2020). We retained only those LOEUF constrained genes with known disease gene associations in the OMIM (Amberger et al. 2015) database (n = 335) accessed and downloaded as a flat.txt file in October 2019. We excluded known autosomal recessive disease genes leaving 293 genes. Variants in novel disease genes represent further potential diagnoses but are beyond the scope of this disease gene focused assessment and have been published elsewhere (Seaby et al. 2022). We applied LOFTEE v1.0 (Karczewski et al. 2020) to flag variants as potential false positives but retained variants in the terminal exon. Variants remaining following DeNovoLOEUF filtering steps were considered putative diagnostic variants.
Clinical outcome data pertaining to diagnostic reports and individual specific phenotype information were extracted querying Labkey using the RLabKey package (v2.9.0) in R (v4.0.3). These phenotype data were computationally merged with filtered putative pathogenic variants for each patient. We further extracted the diagnostic report status for each patient, which included any returned pathogenic variants, by computationally querying the ‘GMC exit questionnaire’ table in LabKey.
Comparative analysis
Putative diagnostic variants were compared with the diagnoses returned to patients recruited to 100KGP at two time points (October 2019 and April 2021) to assess concordance between our DeNovoLOEUF filtering method and the analysis strategy by GEL.
At the first time-point, putative diagnostic variants extracted using DeNovoLOEUF were compared against variants declared in the Genome Medicine Centre exit questionnaire of the RE as being returned to patients in their diagnostic report. This was to assess the positive predictive value (PPV) of the method. It was expected that many patients would not have had a diagnostic report returned in 2019, i.e., their case status was “yet to be assessed”. The comparative analysis was therefore repeated at the second time-point (18 months later in April 2021), to assess whether our method correctly predicted additional diagnoses determined over time as the proportion of closed 100KGP cases increased.
Cases that were not assessed or reported as negative (i.e., no diagnosis identified) by 2021 were re-curated by NHS Clinical Scientists to standardize curation of any novel diagnoses not originally detected through the 100KGP. This was achieved in two ways. First, we contacted the patient’s Genomic Laboratory Hub (GLH), previously known as the Genome Medicine Centre, and referring clinician to discuss the variant we had found and asked whether the variant was already known about and/or had been returned as a diagnosis. Communication with the GLH and referring clinicians often prompted local multidisciplinary team meetings followed by diagnostic laboratory confirmation of the variant. Second, for the variants that were unknown to the GLHs, or for which we received no response from the centres contacted, we worked with clinical scientists in the Wessex Regional Genetics Laboratory, an established GEL diagnostic reporting centre, who curated the remaining variants alongside the patients’ phenotypes as per the ACMG-AMP guidelines (Richards et al. 2015). We then determined how many of the remaining putative diagnostics variants would meet a partial or full diagnosis.
Testing the method on non-trio data
In June 2022, we filtered for rare (AF < 0.001), pLoF variants, in OMIM disease genes with a LOEUF score < 0.2 in an additional 6101 families with complex family structures whereby de novo analysis was not possible. This filtering strategy mimicked the DeNovoLOEUF strategy, except for removing the de novo filter. We retained variants present in affected individuals only. As before, we compared any putative diagnostic variants with the patient’s GMC exit questionnaire.
Iterative re-analysis
In August 2022, we repeated DeNovoLOEUF on newly discovered disease genes with a LOEUF score < 0.2, published between 2019 and 2022, that were classified as ‘definitive’ or ‘strong’ in GenCC (DiStefano et al. 2022) to assess possible diagnostic uplift. These variants were then curated by a clinician scientist in an NHS diagnostic laboratory.
Results
A total of 380 putative diagnostic variants were identified by DeNovoLOEUF in 372 patients. Of these variants, 339/380 (89%) were in the Exomiser top-ranked results. There were more variants than patients due to some individuals harbouring more than one de novo variant in the same gene (npatients = 2) or having more than one de novo variant in two different genes (npatients = 6). The patients with two de novo variants in the same gene were explored further and these variants did not represent a complex structural event. Results stratified by time-point assessment are shown in Fig. 4.
Comparative analysis of method in 2021
In April 2021, 284/380 (75%) of all variants initially identified were confirmed as either fully diagnostic or partially diagnostic, including 17 patients who had diagnostic reports returned in 2019 prior to being withdrawn from the study. A single discordant variant identified in 2019 (Fig. 4) was reclassified as a pathogenic variant by 2021 and returned to the patient as diagnostic. Twenty-six patients harboring 26 variants identified in 2019 were unavailable for further assessment in 2021 due to either withdrawing from the 100KGP or data being temporarily removed from the trusted research environment, whilst re-consent was sought for child participants reaching adulthood. Only one variant that we identified did not match the variant returned to the patient, meaning GEL had returned an alternative diagnosis; however, the variant we identified is a known pathogenic variant in ClinVar. This patient had only one HPO term recorded “cystinosis”, which was consistent with the biallelic variants reported by GEL. We are attempting to contact the clinician responsible for this patient to gain further clinical information and establish whether the putative diagnostic variant identified may be a missed additional diagnosis. Six variants could not be assessed for concordance as the variant returned by 100KGP was not identifiable in the GEL research environment. Sixty-two variants in 62 individuals identified using DeNovoLOEUF remained unresolved in 2021 and required further scrutiny. All 62 variants were in Tier 3 of the GEL tiering system as they were de novo loss-of-function variants but were not in the original gene panel(s) applied.
Assessment of remaining 62 variants in 62 unique individuals
We successfully established contact with 17 (27%) of the remaining 62 patient’s GLHs and referring clinicians. Following this connection, all 17 patients had their DeNovoLOEUF variants confirmed as disease-causing following independent validation in their local NHS laboratories. We were unable to establish contact with the remaining 45 patients’ referring clinician and/or GLH, and therefore, the outstanding 45 variants were manually curated by two clinical scientists working in an NHS accredited genomic diagnostic laboratory (Supplementary Data 2). Of these variants, 14 were designated diagnostic, 8 were partially diagnostic, and 5 were identified as incidental findings. Eight variants were considered not diagnostic, and ten cases were uncertain with insufficient clinical information to confirm causality for the patient’s phenotype (Table 1).
Summary of results
In summary, 324/332 (98%) of the variants identified through DeNovoLOEUF filtering (excluding incidental findings, variants for withdrawn participants, or in patients where there was inadequate phenotype or genotype reporting in GEL) were classified as diagnostic or partially diagnostic (Table 2).
DeNovoLOEUF on non-trio data
Filtering for rare, pLoF variants in known OMIM disease genes with a LOEUF score < 0.2 on 6101 families with complex family structures revealed a further 776 putative diagnostic variants in 757 individuals. 270/757 (36%) of individuals had diagnoses returned that were fully concordant with our method. The number of individuals per pedigree was significantly different between concordant and discordant cases (Wilcoxon test, p = 6.02–16), with discordant patients having a median pedigree structure of one individual (singleton).
Proportion of de novo disease-causing variants detected by DeNovoLOEUF
We sought to explore the total number of de novo pLoF variants detected by GEL and how many were captured by DeNovoLOEUF. A total of 2074 de novo pLoF were identified in 100KGP (Fig. 4). Of these, 480/2074 (23%) were confirmed diagnoses. Of these diagnoses, 380/480 (79.2%) were in LOEUF constrained genes (with a score < 0.2). We further expanded our method to test different cut offs, including a LOEUF score between 0.2 and 0.4 and between 0.4 and 0.6, whereby the positive predictive values were 69% and 45%, respectively.
Iterative re-analysis
Re-running LOEUF on newly discovered genes between 2019 and 2022 that were not present in our original OMIM gene list identified 13 new de novo pLoF variants. 12/13 (92%) have been confirmed as diagnostic and are being returned to patients (Supplementary Data 3).
Discussion
We describe a fast, unbiased filtering strategy, DeNovoLOEUF, to identify potential pathogenic variants with high positive predictive value and specificity. The LOEUF scores for genes are available in Supplementary Data 4. We utilise the LOEUF spectrum of constraint to identify de novo loss-of-function variants with a high potential of pathogenicity. Unlike the approach adopted in the 100KGP with panel-based tiering, our genotype-driven method is agnostic to phenotype and independent of gene panels which often change over time and require the correct one to be selected. Indeed, we identified 39 diagnostic or partially diagnostic variants in 39 known disease-associated genes that had been missed by standard 100KGP diagnostic protocols due to the gene in which the causal variant was identified not being included on the gene panel selected. 35/39 (90%) of these genes were included ‘green’ on different gene panels, meaning that they were recognised “Panel App disease genes”, but the gene panel selection was suboptimal. One of the issues with PanelApp is in selecting the ‘correct’ panel based on the HPO terms provided. For example, the gene CLCN5 is a disease gene for Dent disease 1 (MIM: 300009), a recognised renal tubulopathy. This gene is on the following PanelApp gene panels: “nephrocalcinosis”, “unexplained kidney failure in young people”, “skeletal dysplasia”, and “unexplained paediatric onset end stage renal disease”. However, it is not on the “tubulopathies” panel, which is perhaps the most appropriate panel selection. For most cases where GEL missed the diagnosis, the panel selected was not inappropriate per se, but did not encompass the exact panel required, again highlighting why ‘agnostic to phenotype’ approaches are critical to ensure increased diagnostic yield, particularly in the UK where gene panels are the selected analytical method.
In view of the recognition of diagnoses outside of the selected gene panel(s), some NHS accredited laboratories adopted a policy, when reviewing results from the 100KGP, to assess de novo variation and Exomiser top-ranked results to uplift the resultant diagnoses. This approach to reporting has been informative for the strategy within subsequent large-scale WGS endeavors. When including these results, 26% of all causal variants returned by NHS labs in the 100KGP were not in the initial gene panel applied, exemplifying the issue with gene panel analysis strategies (Rehm 2022). However, re-analyses involve re-visiting sequencing data and clinical cases, something which could be mitigated by screening and prioritizing highly putative diagnostic variants in the first instance. Using DeNovoLOEUF as a screening strategy would have immediately identified 321 pathogenic variants, saving considerable time and money. On average, DeNovoLOEUF added only one extra variant for assessment in ~ 3% of all rare disease probands (0.023 variants per person).
Our method is rapid, having identified 172 variants in 2019 that we were unable to efficiently return to the patient’s clinical teams as the processes for returning research results to the clinic were not supported at scale. As a result of our collaboration with colleagues at Genomics England, a form that enables the submission of multiple potential diagnoses for different participants via a single submission within the RE is now in place. DeNovoLOEUF utilizes an effective screening approach to detect highly penetrant putative diagnostic variants across a large cohort. It should however be noted that whilst 285/333 (86%) of all the variants identified were fully diagnostic, 40/333 (12%) were a partial diagnosis, meaning that the variant was considered pathogenic but did not fully explain the phenotype. Additionally, following manual curation, there were ten variants whereby we were unable to confirm whether the variant explained the phenotype; all these patients lacked sufficient clinical data to determine causality, even though five of the variants were pathogenic by ACMG-AMP guidelines. These patients had a median of 4 HPO terms compared with 18 HPO terms for patients with a diagnosis. This highlights some of the challenges with phenotyping in a large-scale national sequencing project and that using HPO terms are sometimes insufficient to make a diagnosis and post-analysis communication with the clinical care team is a critical component of molecular diagnosis as emphasized by ACMG clinical practice guidelines (Bush et al. 2018). Genomics England is actively supporting improvements at the clinical–research interface to enable collaborations between researchers and clinicians and in patient phenotyping by the provision of Hospital Episode Statistics data within the RE as a longitudinal record of participants’ phenotypes.
With ever-increasing application of genome sequencing and a drive to sequence a further 5 million genomes in the UK, there is clear demand to find efficient analytic strategies. We attest our method to be a suitable adjunct to the current protocols to identify causal variation in 100KGP, the NHS Genomics Medicine Service, or other similar international initiatives. DeNovoLOEUF is capable of prioritizing putative pathogenic variants for diagnostic laboratories, saving time, and resources. Furthermore, one of the drawbacks of applying gene panels is that many are already outdated at the point of use, with new genes being consistently added to the literature. Our method can be easily applied iteratively for re-analysis as new genes are discovered and added to ClinVar (Landrum et al. 2014), HGMDPro, or GenCC (DiStefano et al. 2022), before being indexed in OMIM. However, at the time of method development, OMIM represented the best available repository of disease genes with a standardized method for curating genotype–phenotype relationships. We did however re-run DeNovoLOEUF in 2022 on new disease genes (added to GenCC) after our initial analysis in 2019; this identified a further 13 variants of which 12 have been confirmed as diagnostic.
Whilst the DeNovoLOEUF method has a high positive predictive value, 9 variants (1 from the 2021 analysis and 8 from curation analysis) were discordant with the molecular diagnosis returned. One variant was in an X-linked recessive gene and the patient was female. A further variant passed quality control filtering but was artefactual when the reads were directly visualised. Four variants were in a disease gene inconsistent with the patient’s phenotype and there was an average of 3 HPO terms per patient. One variant was within 6 base-pairs of a full splice rescue. Two variants were pLoF on a non-canonical transcript that was poorly expressed across disease-relevant tissues using the pext score (Cummings et al. 2019) based on GTeX data and intronic on the MANE Select transcript. Whilst one option would be to limit our method to variants on the MANE Select and MANE Plus Clinical transcripts, this is potentially problematic as not all genes have been curated to define additional transcripts to be included in the MANE Plus Clinical resource (Morales et al. 2022).
Limitations and opportunities
Whilst our genotype-first method diagnosed patients missed by the initial 100KGP diagnostic strategy, it does not replace the importance of a phenotype-driven approach. It would be foolhardy not to look at all variants in a gene with a close phenotype match to the patient. Our method is best applied as a complementary screening strategy and will not diagnose the majority of patients, especially those with variants in non-constrained genes, or with pathogenic missense or extended splice site variants or with inherited variants. It also does not negate the need for variant curation, since some pLoF variants may not result in LoF and there is yet to be an automated method capable of replicating full manual curation (Gudmundsson et al. 2021). Furthermore, not all pLoF variants may be disease-causing through a loss-of-function mechanism, however we specifically selected genes constrained for LoF meaning we were enriched for pathogenic LoF variants in this specific subset of genes.
Our screening tool selects de novo variants, meaning that we excluded potential pathogenic variants in patients without trio data and excluded diseases where we may expect disease segregation, e.g., cardiac or immune disease. When applying the same filtering strategy, minus the de novo filter, to complex family structures or singletons, we identified an additional 270 diagnoses. This yielded a PPV of 36% vs 98% for trios. Used prospectively, this means that there is an increased probability of type 1 errors, although some of the unsubstantiated variants may represent real diagnoses. Unsurprisingly, for non-trio cases where returned diagnoses were discordant with our method, the median family size was 1, with 631/678 (93%) of cases being singletons.
We selected genes in the first LOEUF decile to increase the specificity of our method. As shown in the flagship gnomAD paper, a LOEUF score of < 0.2 is the most highly enriched for haploinsufficient disease genes. In total across all genes, there were 2074 de novo pLoF variants called in Genomics England of which 480/2074 (23%) were diagnoses. Of these, 79.2% were in LOEUF constrained genes (score < 0.2). Expanding our approach to de novo pLoF variants in disease genes using a higher LOEUF threshold will inevitably increase the diagnostic yield; however, this must be balanced with increased noise, an increased number of variants for review, and significantly reduced specificity. When expanding the analysis to a LOEUF score between 0.2 and 0.4, the positive predictive value reduced to 69%, and with a LOEUF score between 0.4 and 0.6, this reduced again to 45%. We recommend that this approach could be adopted for downstream analyses.
DeNovoLOEUF also risks identifying incidental findings including those in the ACMG secondary findings v3 list, which are a consequence of WGS and a tradeoff for increased diagnostic yield. Applying a LOEUF cut off < 0.2 identified 10 genes in the ACMG list, although these genes could be manually removed from DeNovoLOEUF if preferred, or if the patient has not consented for secondary findings (Miller et al. 2021).
Whilst DeNovoLOEUF has a high PPV, there are opportunities to refine our method to increase sensitivity at the expense of specificity. We propose a revised screening pipeline that will not only identify de novo variants in LOEUF constrained genes, but also screen for all known rare ClinVar pathogenic or likely pathogenic variants regardless of the gene panel applied (Fig. 5). Our suggestion is to place de novo variants, ClinVar variants and novel coding variants into a new tier for assessment by NHS clinical laboratories to complement the current tiering system (The 100,000 Genomes Project Pilot Investigators 2021). Adding this additional review approach, along with a phenotype-driven variant analysis, is consistent with recently released best practices in genome analysis released from the Medical Genome Initiative (Austin-Tse et al. 2022). From reviewing 20 genomes, we estimate that this approach would yield 3–9 potential additional variants per trio, using a LOEUF cut off < 0.35, a missense constraint z-score > 3 (Samocha et al. 2014), and likely pathogenic/pathogenic ClinVar entries with 2 stars or above. Our hope is to achieve a higher diagnostic yield per number of variants assessed by diagnostic labs, whereby we prioritize the most salient variants first.
Conclusion
We present a targeted screening tool, DeNovoLOEUF, that can be applied at scale to rapidly identify putative pathogenic variants with a 98% positive predictive value. Our method complements current family-based analyses and can add value by identifying diagnostic variants missed by filtering strategies that adopt predefined disease-targeted gene panels. We have identified 39 pathogenic variants missed by the initial 100KGP variant prioritization strategy. With 5 million more genomes being sequenced on the NHS, and many other international sequencing studies underway, we believe that our method alongside the new GEL initiative to report on Exomiser top-ranked variants can help rapidly and effectively improve diagnostic efficiency and uplift diagnostic rates for the benefit of rare disease patients and their families.
Availability of data and materials
Access to the 100KGP dataset analysed in this study is only available as a registered GeCIP member in the Genomics England Research Environment, but restrictions apply to the availability of these data due to data protection and are not publicly available. Information regarding how to apply for data access is available at the following URL: https://www.genomicsengland.co.uk/about-gecip/for-gecip-members/data-and-data-access/. All data shared in this manuscript were approved for export by Genomics England. The datasets and code supporting the current study can be accessed within the Genomics England Research Environment. DeNovoLOEUF is available here: https://github.com/lecb/DeNovoLOEUF.
References
Amberger JS, Bocchini CA, Schiettecatte F, Scott AF, Hamosh A (2015) OMIM. org: Online Mendelian Inheritance in Man (OMIM®), an online catalog of human genes and genetic disorders. Nucleic Acids Res 43:D789–D798
Austin-Tse CA, Jobanputra V, Perry DL, Bick D, Taft RJ, Venner E, Gibbs RA, Young T, Barnett S, Belmont JW (2022) Best practices for the interpretation and reporting of clinical whole genome sequencing. NPJ Genom Med 7:1–13
Bush LW, Beck AE, Biesecker LG, Evans JP, Hamosh A, Holm IA, Martin CL, Richards CS, Rehm HL (2018) Professional responsibilities regarding the provision, publication, and dissemination of patient phenotypes in the context of clinical genetic and genomic testing: points to consider—a statement of the American College of Medical Genetics and Genomics (ACMG). Genet Med 20:169–171
Cummings BB, Karczewski KJ, Kosmicki JA, Seaby EG, Watts NA, Singer-Berk M, Mudge JM, Karjalainen J, Kyle Satterstrom F, O’Donnell-Luria A et al (2019) Transcript expression-aware annotation improves rare variant discovery and interpretation. bioRxiv 554444
DiStefano MT, Goehringer S, Babb L, Alkuraya FS, Amberger J, Amin M, Austin-Tse C, Balzotti M, Berg JS, Birney E et al (2022) The Gene Curation Coalition: a global effort to harmonize gene-disease evidence resources. Genet Med
Gudmundsson S, Karczewski KJ, Francioli LC, Tiao G, Cummings BB, Alföldi J, Wang Q, Collins RL, Laricchia KM, Ganna A (2021) Addendum: the mutational constraint spectrum quantified from variation in 141,456 humans. Nature. https://doi.org/10.1038/s41586-021-03758-y
Karczewski KJ, Francioli LC, Tiao G, Cummings BB, Alföldi J, Wang Q, Collins RL, Laricchia KM, Ganna A, Birnbaum DP et al (2020) The mutational constraint spectrum quantified from variation in 141,456 humans. Nature 581:434–443
Landrum MJ, Lee JM, Riley GR, Jang W, Rubinstein WS, Church DM, Maglott DR (2014) ClinVar: public archive of relationships among sequence variation and human phenotype. Nucleic Acids Res 42:D980–D985
MacArthur DG, Balasubramanian S, Frankish A, Huang N, Morris J, Walter K, Jostins L, Habegger L, Pickrell JK, Montgomery SB (2012) A systematic survey of loss-of-function variants in human protein-coding genes. Science 335:823–828
Martin AR, Williams E, Foulger RE, Leigh S, Daugherty LC, Niblock O, Leong IU, Smith KR, Gerasimenko O, Haraldsdottir E (2019) PanelApp crowdsources expert knowledge to establish consensus diagnostic gene panels. Nat Genet 51:1560–1565
Miller DT, Lee K, Chung WK, Gordon AS, Herman GE, Klein TE, Stewart DR, Amendola LM, Adelman K, Bale SJ et al (2021) ACMG SF v3.0 list for reporting of secondary findings in clinical exome and genome sequencing: a policy statement of the American College of Medical Genetics and Genomics (ACMG). Genet Med 23:1381–1390
Morales J, Pujar S, Loveland JE, Astashyn A, Bennett R, Berry A, Cox E, Davidson C, Ermolaeva O, Farrell CM (2022) A joint NCBI and EMBL-EBI transcript set for clinical genomics and research. Nature. https://doi.org/10.1038/s41586-022-04558-8
Parliament HO (2015) The 100,000 Genomes Project. In: (POSTNOTE, Parliamentary Office of Science and Technology
Posey JE, O’Donnell-Luria AH, Chong JX, Harel T, Jhangiani SN, Akdemir ZHC, Buyske S, Pehlivan D, Carvalho CM, Baxter S (2019) Insights into genetics, human biology and disease gleaned from family based genomic studies. Genet Med 21:798–812
Rehm HL (2022) Time to make rare disease diagnosis accessible to all. Nat Med 28:241–242
Richards S, Aziz N, Bale S, Bick D, Das S, Gastier-Foster J, Grody WW, Hegde M, Lyon E, Spector E et al (2015) Standards and guidelines for the interpretation of sequence variants: a joint consensus recommendation of the American College of Medical Genetics and Genomics and the Association for Molecular Pathology. Genet Med 17:405–424
Robinson PN, Köhler S, Bauer S, Seelow D, Horn D, Mundlos S (2008) The Human Phenotype Ontology: a tool for annotating and analyzing human hereditary disease. Am J Hum Genet 83:610–615
Samocha KE, Robinson EB, Sanders SJ, Stevens C, Sabo A, McGrath LM, Kosmicki JA, Rehnström K, Mallick S, Kirby A (2014) A framework for the interpretation of de novo mutation in human disease. Nat Genet 46:944–950
Seaby EG, Ennis S (2020) Challenges in the diagnosis and discovery of rare genetic disorders using contemporary sequencing technologies. Brief Funct Genomics. https://doi.org/10.1093/bfgp/elaa009
Seaby EG, Rehm HL, O’Donnell-Luria A (2021) Strategies to uplift novel mendelian gene discovery for improved clinical outcomes. Front Genet. https://doi.org/10.3389/fgene.2021.674295
Seaby EG, Smedley D, Taylor Tavares AL, Brittain H, van Jaarsveld RH, Baralle D, Rehm HL, O’Donnell-Luria A, Ennis S (2022) A gene-to-patient approach uplifts novel disease gene discovery and identifies 18 putative novel disease genes. Genet Med 24(8):1697–1707. https://doi.org/10.1016/j.gim.2022.04.019
Smedley D, Jacobsen JO, Jäger M, Köhler S, Holtgrewe M, Schubach M, Siragusa E, Zemojtel T, Buske OJ, Washington NL (2015) Next-generation diagnostics and disease-gene discovery with the Exomiser. Nat Protoc 10:2004–2015
The 100,000 Genomes Project Pilot Investigators (2021) 100,000 genomes pilot on rare-disease diagnosis in health care—preliminary report. N Engl J Med 385:1868–1880
Turnbull C, Scott RH, Thomas E, Jones L, Murugaesu N, Pretty FB, Halai D, Baple E, Craig C, Hamblin A (2018) The 100 000 Genomes Project: bringing whole genome sequencing to the NHS. BMJ 361:k1687
Acknowledgements
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 Limited (a wholly owned company of the Department of Health and Social Care). The 100,000 Genomes Project is funded by the National Institute for Health Research and NHS England. The Wellcome Trust, Cancer Research UK, and the Medical Research Council have also funded research infrastructure. The 100,000 Genomes Project uses data provided by patients and collected by the National Health Service as part of their care and support. We would further like to extend our thanks to all the patients and their families for participation in the 100,000 Genomes Project. We are grateful to the Broad Center for Mendelian Genomics and Genome Aggregation Database teams for their helpful discussions in the development and application of constraint metrics in novel gene discovery.
Genomics England Research Consortium
J. C. Ambrose1, P. Arumugam1, R. Bevers1, M. Bleda1, F. Boardman-Pretty1,2, C. R. Boustred1, H. Brittain1, M. J. Caulfield1,2, G. C. Chan1, T. Fowler1, A. Giess1, A. Hamblin1, S. Henderson1,2, T. J. P. Hubbard1, R. Jackson1, L. J. Jones1,2, D. Kasperaviciute1,2, M. Kayikci1, A. Kousathanas1, L. Lahnstein1, S. E. A. Leigh1, I. U. S. Leong1, F. J. Lopez1, F. Maleady-Crowe1, M. McEntagart1, F. Minneci1, L. Moutsianas1,2, M. Mueller1,2, N. Murugaesu1, A. C. Need1,2, P. O‘Donovan1, C. A. Odhams1, C. Patch1,2, D. Perez-Gil1, M. B. Pereira1, J. Pullinger1, T. Rahim1, A. Rendon1, T. Rogers1, K. Savage1, K. Sawant1, R. H. Scott1, A. Siddiq1, A. Sieghart1, S. C. Smith1, A. Sosinsky1,2, A. Stuckey1, M. Tanguy1, A. L. Taylor Tavares1, E. R. A. Thomas1,2, S. R. Thompson1, A. Tucci1,2, M. J. Welland1, E. Williams1, K. Witkowska1,2, S. M. Wood1,2
1Genomics England, London, UK
2William Harvey Research Institute, Queen Mary University of London, London, EC1M 6BQ, UK
Funding
EGS was supported by the Kerkut Charitable Trust, Foulkes Foundation, and University of Southampton’s Presidential Scholarship Award; HR by the NHGRI U24 HG011450 and U41 HG006834; and AO’D-L by the National Institute of Mental Health U01 MH119689 and Manton Center for Orphan Disease Research Scholar Award. EGS, HLR, and AO’D-L were supported by the National Human Genome Research Institute (NHGRI), the National Eye Institute, and the National Heart, Lung and Blood Institute under Grant UM1 HG008900. DB was generously supported by a National Institute of Health Research (NIHR) Research Professorship under Grant RP-2016-07-011.
Author information
Authors and Affiliations
Consortia
Contributions
EGS conceived the idea for the study, conducted data analysis, data interpretation, and wrote the first draft of the manuscript. NST and AW assisted with data analysis. ALTT and HB assisted with data interpretation. SE, AO’D-L, HLR, and DB supervised the project, including overseeing data analysis and interpretation. All authors read and approved the final manuscript.
Corresponding author
Ethics declarations
Conflict of interest
The authors declare no conflicts of interest.
Ethics approval and consent to participate
All patients included in this study consented to participate in the 100,000 Genomes Project—ethics approval by the Health Research Authority (NRES Committee East of England) REC: 14/EE/1112; IRAS: 166046. The ethical approval letter is available upon request.
Consent for publication
Not applicable.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Members of the Genomics England Consortium as well as their affiliations are listed in the Acknowledgement section.
Supplementary Information
Below is the link to the electronic supplementary material.
439_2022_2509_MOESM3_ESM.xlsx
Supplementary file3 Supplementary Data 3: Curation of 13 variants identified following iterative re-analysis of DeNovoLOEUF on new disease genes (XLSX 16 KB)
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Seaby, E.G., Thomas, N.S., Webb, A. et al. Targeting de novo loss-of-function variants in constrained disease genes improves diagnostic rates in the 100,000 Genomes Project. Hum Genet 142, 351–362 (2023). https://doi.org/10.1007/s00439-022-02509-x
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00439-022-02509-x