Full Text
The terms “epileptic encephalopathy” (EE) and “developmental epileptic encephalopathy” (DEE) have been precisely defined by the International League Against Epilepsy (ILAE) in 2017. It is proposed that in epileptic encephalopathies the developmental delay is not caused by the genetic mutation itself but can result as a consequence of the epileptic activity. In contrast, in DEE both the epileptic activity and the developmental impairment occur on the basis of the underlying pathology (Scheffer et al., 2017). Patients are typically severely affected and show variable cognitive, behavioral and psychiatric characteristics.
More than 400 genes (Møller et al., 2019) have been identified as a cause for EE and DEE so far and many other genes are discussed as possible candidates. As more and more genes are declared EE and DEE causing, the genetic workup of epilepsy patients has greatly gained importance (Hebbar & Mefford, 2020). Meta-analysis of the last decade shows a diagnostic yield of 32% (95% CI: 22–44%) for whole exome sequencing (WES), 23% (95% CI: 18–29%) for epilepsy panels and 8% (95% CI: 6–12%) for chromosomal microarray analysis (CMA) (Ferńandez et al., 2019). More recent studies report an even higher diagnostic yield of 30–50% in WES (Demos et al., 2019; Palmer et al., 2018; Papuc et al., 2019; Rochtus et al., 2020; Varesio et al., 2021).
Whole genome sequencing (WGS) is not part of the routine diagnostic approach in EE/DEE patients. Nevertheless, the diagnostic yield of genetic diseases has been reported to be higher with WGS than standard of care genetic workup and great potential is seen in the application of WGS (Lionel et al., 2018; Mattick et al., 2018; Møller et al., 2019; Palmer et al., 2021; Scala et al., 2020). Not only does WGS cover noncoding parts of the genome, but it is also less susceptible to technical distortions and blind spots (Mattick et al., 2018). However, the consequence of changes in noncoding regions detected by WGS is currently still largely unknown. Therefore, the pathogenicity of these changes should be studied more closely before they can be used in diagnostics (Vaz-Drago et al., 2017).
Since methods used for whole exome sequencing (WES) have also improved over time, it remains unclear what benefits WGS currently offers in comparison to WES. Therefore, the aim of this study was to assess the additional diagnostic yield achieved by trio-WGS in EE/DEE patients who remained without diagnosis after trio-WES and chromosomal microarray analysis.
PATIENTS AND METHODS PatientsWe performed a follow-up of undiagnosed index cases for trio-WGS from a cohort of patients originally reported by Papuc et al. (2019). Inclusion criteria for the initial cohort were developmental delay and onset of epilepsy below the age of 4.5 y, pharmacoresistance to antiepileptic drugs, EEG without persistent spike wave focus, no malformations in MRI, and unknown etiology after clinical evaluation including metabolic screening. The index patients selected for trio-WGS neither had a clear genetic diagnosis, nor a strong candidate gene after trio-WES and high-resolution CMA. One index patient had an affected sibling, who was also enrolled in the study.
SequencingLibrary preparation was performed using 1000 ng genomic DNA extracted from peripheral blood and the Illumina TruSeq PCR-Free workflow. The genomic DNA was sheared on the Covaris S220 system resulting in fragment lengths of ~350 base pairs. The fragments were end-repaired and adapters containing unique indices were ligated. The efficiency of the adapter ligation was reviewed by loading a pool of equal volumes of 15 prepared libraries onto the iSeq 100 System (151 cycles, paired-end). According to the output of the iSeq 100 (% identified reads and passing filter value) the libraries were pooled equimolar and loaded onto an S4 flow cell on the NovaSeq 6000 System (151 cycles, paired-end). Sequencing reads were mapped and aligned to the GRCh37/hg19 human genome reference assembly using the NextGENe software package (Softgenetics Inc.) for the first step, as well as the DRAGEN Bio IT Platform (Illumina) for the second step, resulting in an average read depth of 62.6. For the annotation of single nucleotide polymorphisms (SNPs) and small insertions/deletions we used the open-source software NIRVANA (NImble and Robust VAriant aNnotAtor). Copy number variants (CNV) were called and visualized using commercially available software with our in-house custom modifications (Nexus Clinical v6, BioDiscovery Inc.), which was also used for small variant filtering for the second step.
As a first step, genomic data was filtered for variants in 2842 epilepsy and ID genes with a MAF ≤2%. Analysis comprised single case evaluation as well as trio-analysis for de novo and recessive (homozygous, hemizygous, loss of heterozygosity and compound heterozygous) variants.
For the remaining undiagnosed patients, we expanded genetic workup without gene filter, including CNV analysis. De novo calls were filtered for variants with MAF < 0.01%, quality score > 30 and similar previous variants from our in-house database <1. Homozygous, compound heterozygous and hemizygous calls were filtered for quality score > 30, MAF < 0.05% and previous similar variants < 5. After manual review, the remaining calls were assessed using ClinVar, DECIPHER, OMIM (
From the original cohort of 63 index patients with EE and DEE, 26 individuals (41%) had been diagnosed with average 212× coverage WES and high-resolution CMA (Papuc et al., 2019). For 16 (25%) of the remaining patients, strong candidate genes were established, of which three (PIK3AP1 (OMIM * 607942), GTF3C3 (OMIM * 604888) and WRAP53 (OMIM * 612661)) were published in the previous paper. In the meantime, in another four of these cases, novel disease genes (CYFIP2 (OMIM * 606323), UFC1 (OMIM * 610554), IRF2BPL (OMIM * 611720) and MED27 (OMIM * 605044)) could be confirmed as underlying cause (Table 1). Further investigation of the other candidate genes is ongoing. Finally, 21 (33%) patients from the original cohort remained without diagnosis and no strong candidate genes were suspected (Figure 1).
TABLE 1 Patients whose original candidate genes were confirmed as disease genes.
| Patient ID | Inheritance | Genomic position (hg19) | Transcript and variant | AA change | Consequence | Gene | Comments |
| 62075 | DN | chr5:156721843 | NM_001037332.3:c.259C>T | p.(Arg87Cys) | Missense | CYFIP2 | CYFIP2 was identified as disease causing gene and multiple patients with p.(Arg87Cys) variants were reported (Zweier et al., 2019) |
| 76366 | HO | chr1:161123855 | NM_016406.4:c.68G>A | p.(Arg23Gln) | Missense | UFC1 | This variant was identified as candidate disease allele and confirmed with further analysis (Nahorski et al., 2018) |
| 59248 | DN | chr14:77493174 | NM_024496.4:c.962del | p.(Ala321Glufs*24) | Frameshift | IRF2BPL | This variant was identified as candidate disease allele and 11 unrelated individuals were reported with de novo heterozygous truncating variants in IRF2BPL causing DEE (Tran Mau-Them et al., 2019) |
| 56302 | HO | chr9:134738526 | NM_001253881.2:c.617T>C | p.(Val206Ala) | Missense | MED27 | This variant was classified as candidate disease allele and biallelic variants in MED27 were identified in 16 patients from 11 families with neurodevelopmental diseases, including seizures in severely affected individuals (Meng et al., 2021) |
Abbreviations: DN, de novo; HO, homozygous.
From these 21 patients, all but one, for whom the paternal sample was not available, received further analysis. Therefore, 20 patients, 9 females and 11 males were chosen for trio-WGS. 13 patients had EE, 7 had DEE and median age of seizure onset was 9 months (range 1 m to 4 y 3 m). For four of the 20 trio-WGS individuals a diagnosis could be reached (Table 2), corresponding to an additional diagnostic yield of 20%. Two missense variants (SCN2A (OMIM * 182390), NSF (OMIM * 601633)), one frameshift variant (HNRNPU (OMIM * 602869)) and one complex rearrangement (MECP2 (OMIM * 300005)) were found, all of which occurred de novo and are located in known EE/DEE genes. Retrospectively, all of these variants could have been also detected by current WES technology. Nevertheless, the small 547 bp exonic deletion in MECP2 could be called with a higher reliability in WGS data than in WES data (Figure 2). Optical Genome Mapping using the Bionano Inc. Saphr® technique confirmed insertion of a 39 kb fragment into MECP2 (Figure 2e). Overall, 34 patients from the original cohort of 63 are now with a clear genetic diagnosis, adding up to a diagnostic yield of 54%.
TABLE 2 Patients diagnosed with trio-WGS.
Abbreviation: DN, de novo.
Enlarge this image.FIGURE 2. Pathogenic complex rearrangement in patient 70855, consisting of an approximately 39 kb de novo microduplication of parts of TMEM255B inserted into MECP2 and a 0.5 kb deletion in the last exon of MECP2. Each panel A-D shows from top to bottom genomic position (hg19), gene content, copy number calling, BAM depth, and copy number values of calculated probe bins. (a) WGS showing the deletion in MECP2 with sharp boarders (b) WES showing the deletion in MECP2 (c) WGS showing the duplication of parts of TMEM255B on chromosome 13 (d) WES showing the duplication of parts of TMEM255B on chromosome 13. (e) Results of Optical Genome Mapping using the Bionano Inc. Saphyr® device demonstrating a de novo 39 kb insertion into MECP2, likely corresponding to the 39 kb duplicated segment of TMEM255B.
In this study we performed a follow-up of an EE/DEE cohort initially published in 2019 and examined the additional diagnostic yield achieved by follow-up of candidate genes as well as trio-WGS in 20 patients who remained undiagnosed with previous WES and CMA analyses. Four previously undiagnosed individuals (6%) of the original study cohort (n = 63) had received a diagnosis in the meantime by confirming their former candidate genes as disease genes (CYFIP2, UFC1, IRF2BPL and MED27). This exemplifies that constantly new findings arise in the field of EE/DEE.
In the 20 patients who remained without diagnosis or strong candidate gene, we achieved a diagnostic yield of 20% by trio-WGS. All causal variants occurred de novo in coding regions of known EE/DEE genes. Thereof, one is a complex rearrangement involving MECP2. However, all variants could have also been detected if we had reanalyzed WES data (SCN2A, NSF) or conducted new WES analysis with the current sequencing technology (HNRNPU, MECP2). Nevertheless, small copy number variant calling appears more reliable from WGS data than from WES data. Consequently, the additional diagnostic yield obtained in our cohort should mainly be credited to reanalysis and technical advances rather than to an advantage of WGS in comparison to WES.
In this regard, it is noteworthy that considering our original cohort, systematic reanalysis of already sequenced data after four years would have increased the diagnostic yield by 10% (CYFIP2, UFC1, IRF2BPL, MED27, NSF and SCN2A). Similarly, the review by Ji et al. (2021) reports that exome reanalysis increases the diagnostic yield of approximately 12%, ranging from 5 to 26% between studies. These studies mostly examined intellectual disability, epilepsy and other neurological phenotypes and mainly performed reanalysis 1–2 years after the first analysis. Li et al. (2019) also report an additional diagnostic yield with exome reanalysis after one year of 15% in patients with epilepsy and ID.
Furthermore, technical advances of high throughput sequencing (Ross et al., 2020) allow to identify more variants by WES today compared to the first WES done in 2016. In our cohort technical improvement would have accounted for an additional diagnostic yield of 3% (HNRNPU and MECP2).
In line with our study, Hamdan et al. (2017) and Martin et al. (2014) report that all pathologies found with WGS in cohorts of 197 and six patients with neurodevelopmental disorders, respectively, could have also been detected by WES analysis. Likewise, Ewans et al. (2022) report an additional diagnostic yield with WGS for six out auf 31 patients with Mendelian diseases, who remained undiagnosed after both WES and a following WES reanalysis two years later. WGS detected a heterozygous dominant pathogenic variant in the 5′UTR of ANKRD26 (OMIM * 610855), which was not covered with WES. WGS also found a partial deletion of exon 1, which may have been missed due to issues with exon 1 calling in WES. A third pathogenic variant was not seen due to a failure in the bioinformatic pipeline. For the remaining three pathogenic variants, WES had reduced coverage, explaining why these variants were missed. The authors, however, postulate that with an improved WES platform, this could have been avoided. Therefore, with better technological conditions (e.g., WES including 5′UTR sequences) and improved bioinformatics most additionally detected variants with WGS may have been found with newest WES methods. Nevertheless, these findings also show that WGS may be more reliable than current standard WES, especially regarding coverage and CNV detection. This is supported by three further studies describing the detection of certain structural variants with WGS that could not be seen with WES. van der Sanden et al. (2022) compared trio-WGS to trio-WES based standard of care genetic testing (SOC) in 150 patients with neurodevelopmental disease. While the higher diagnostic yield was not statistically significant, WGS did generate an additional two conclusive diagnoses, a 5 kb deletion covering exons 14 and 15 of CHD2 (OMIM * 602119) and a 36 kb deletion including the 3′UTR exon 9 of AHDC1 (OMIM * 615790), as well as four possible diagnoses not detected with SOC, all six representing CNVs. The authors put this down to limitations of sensitivity and specificity of CNV variant calling in WES and the absence of targets for WES enrichment. Moreover, Palmer et al. (2021) report three structural variants only detected with WGS in a cohort of 30 patients. Two of these are copy neutral inversions with breakpoints in noncoding regions. Also, Ostrander et al. (2018) report a copy neutral translocation between two chromosomes with breakpoints in noncoding regions that was only seen by WGS in a cohort of 14 patients. They furthermore report a 63 kb de novo duplication containing exons 5–15 of CDKL5 (OMIM * 300203), which was only seen in WGS. Nevertheless, we postulate that WES with today's technological possibilities should show most CNVs involving at least two coding exons, and mainly copy neutral variants and small CNVs or noncoding variants could be missed. Our study did not reveal such abnormalities; however, we did not specifically investigate copy number neutral variants and thus the potential of WGS was not fully leveraged. Moreover, while the MECP2 partial exon deletion is visible in current WES data, it might have been missed in a blinded setting due to the high rate of false-positive results of single exon CNV calling. We, therefore, acknowledge that WGS might still outperform WES for CNV calling due to a higher specificity resulting in less required validation efforts and coverage of noncoding regions.
While studies from 2014 and 2018 clearly described WGS being superior to WES regarding coverage and sensitivity (Gilissen et al., 2014; Mattick et al., 2018; Meienberg et al., 2016), current studies show a more diverse picture. Barbitoff et al. (2020) systematically compared the coverage of the coding region between 70 and 120× coverage WES and 30× coverage WGS data and state that genome sequencing shows a more efficient coverage in only 1% of the exome compared with current WES technologies. They further conclude that with technical advances and specific improvements (e.g., larger insert sizes) WES can eventually catch up to the standards of WGS. Another feature for comparing the performance between WGS and WES is what sizes of CNVs can be detected with either method. In our study, 62.6× coverage WGS detected a pathogenic CNV of 547 bp. Hamdan et al. (2017) reported CNVs ranging from 1304 bp to 8′063′680 bp in their 30× coverage WGS study. The WES benchmark study of Gordeeva et al. (2021) comparing the performance of 16 different CNV calling tools on 10 exomes with average 104× coverage found that most of them focused on detection of CNVs one to seven exons long with a false-positive rate below 50% and a low concordance between tools. Likewise, Sun et al. (2021) found that 40× coverage WGS (MGIEasy PCR-Free DNA Library Prep Set) compared to 120× WES (MGIEasy Exome FS Library Prep Set) has a better sensitivity and positive predictive value for SNV and indels as well as for CNVs. Thereby the CNV detection rate by WGS was greatly influenced by the sequencing depth especially when the CNV size was less than 1 kb with plateauing at about 40× coverage. Thus, especially ≥40× coverage WGS is more suitable than WES for the detection of smaller CNVs due to its higher sensitivity and specificity.
Additional potential demonstrated by WGS in comparison to WES concerns complex rearrangements. While some of them could be detected in WES, the full extent of the rearrangement can often only be apprehended with WGS. Our study shows that the exact dimension and location of the duplicated region involved in the complex rearrangement disrupting MECP2 can only be seen with WGS. Likewise, van der Sanden et al. (2022) report that the 32 kb duplication included in a deletion-duplication event could only be detected with WGS. Palmer et al. (2021) describe a pathogenic variant seen in a CMA analysis that could not be interpreted on the basis of this data. Only with WGS the complexity of the rearrangement was understood, and the pathogenicity, therefore, confirmed.
To better assess structural variants optical genome mapping (OGM) has been introduced in recent years. Studies comparing OGM to standard assays such as karyotyping, FISH and CMA report OGM to be a promising detection tool offering the benefit of acquiring breakpoints and, therefore, valuable genetic information (Dremsek et al., 2021) (Mantere et al., 2021). However, although in our case OGM demonstrated a 39 kb insertion into MECP2 (Figure 2e), the origin of the material and the effect on the MECP2 gene remained illusive. Only breakpoint analysis from WGS allowed to determine that the 39 kb duplicated segment corresponds to material of TMEM255B and interrupts the coding region of MECP2. Also, the concomitant partial deletion of MECP2 is not visible by OGM. Likewise, other studies report OGM breakpoint mapping being imprecise and insufficient for breakpoints involving poorly covered regions such as the subtelomeric region of Xp or repetitive materials such as centromers, heterochromatin and telomeres and should, therefore, best be adopted in combination with other diagnostic techniques (Dai et al., 2022; Dremsek et al., 2021).
Nevertheless, the greatest advantage of a whole genome sequencing approach would be to unravel pathogenic variants in noncoding regions, which could also induce genetic diseases (Zhang & Lupski, 2015). The meaning of these changes, though, is difficult to interpret, as there is still little knowledge about the function and pathogenicity of noncoding variants. Therefore, the added clinical value over WES today is rather limited and may, for example, apply for cases with only one pathogenic variant detected in a recessive disease gene. However, great potential lies in these noncoding segments (Vaz-Drago et al., 2017). For instance, Wright et al. (2021) report 5′UTR variants of the MEF2C gene as a cause for developmental disorders, thereby underlining that the analysis of 5′UTRs can increase diagnostic yield. Although no de novo variants in the 5′UTR region of MEF2C were found in this study, further investigations are necessary to evaluate the potential pathogenicity of noncoding variants in our remaining undiagnosed patients.
Various steps to better exhaust the full potential of WGS are currently ongoing. Recommendations concerning the handling of noncoding variants have been made (Ellingford et al., 2022; Kircher & Ludwig, 2022), including the combination of WGS with transcriptomics, metabolomics, proteomics, or methyl profiling to help prioritize pathogenic variants (Marwaha et al., 2022). Furthermore, WGS was recently evaluated using a variety of algorithms for the detection of short tandem repeat (STR) expansions (Rajan-Babu et al., 2021) and to evaluate mitochondrial variants (Davis et al., 2022; Riley et al., 2020; Schon et al., 2021), both of which are increasingly recognized as causative in neurodevelopmental diseases (Chintalaphani et al., 2021; Moreau et al., 2021; Ortiz-González, 2021; Qaiser et al., 2021). Given the abundance of mitochondrial DNA, it's deep sequencing is easily achieved by standard WGS, but the detection of pathogenic mitochondrial variants remains challenging due to the occurrence of heteroplasmy, which may result in a low aberrant read fraction, as well as complex rearrangements, which may only be detectable in muscular tissues (Macken et al., 2021). WES with targeted capture of the mitochondrial DNA has also been used to identify mitochondrial sequence variants and deletions (C. Sun et al., 2022) but a direct comparison to WGS is lacking.
Although not specifically done in this study, it should be acknowledged that STR detection could be improved by optimized bioinformatic pipelines and focused STR analysis for both WES and WGS (Rajan-Babu et al., 2021; van der Sanden et al., 2021). However, while some clinically relevant STRs may be detected from WES data, many are located in noncoding regions (Chintalaphani et al., 2021). In addition, calling of dinucleotide repeats in general and repeats 3–6 bp in length in the homopolymeric state from short-read sequences proves challenging (Halman & Oshlack, 2020). In this regard, it is expected that long-read sequencing will overcome the limitations of established short-read sequencing platforms in genotyping large and/or complex repeat expansions (Chintalaphani et al., 2021). Next to the advantages of more reliable STR calling, first clinical studies indicate that the main advantage of long-read WGS lies in unambiguous calling of variants in the 98% of the so called short-read sequencing dead zones, superior performance in identifying structural variants and its capacity to determine genomic methylation defects in native DNA (Sanford Kobayashi et al., 2022).
In conclusion, this study shows that with current routine methods the additional diagnostic yield of WGS in comparison to WES is still limited. Taking higher cost and data processing requirements into consideration, the application of 40-50× WGS today might make most sense after nondiagnostic WES and systematic reanalysis, to potentially find 5′UTR variants, STRs, small CNVs and copy neutral structural variants not detected in WES, and fully understand complex structural variants. However, as we expect new knowledge about the relevance of noncoding variants to arise, great potential lies in the application of WGS.
AUTHOR CONTRIBUTIONSAR, PJ and II designed and supervised the project. AG, MR, AB, KS, LA, MP, MZ and BO acquired and analyzed the data. AG wrote the initial draft of the manuscript. All authors reviewed and revised the manuscript.
ACKNOWLEDGMENTSWe thank the families for their participation in the study.
CONFLICT OF INTEREST STATEMENTThe authors declare that they have no conflict of interest.
DATA AVAILABILITY STATEMENTRaw data or samples are not publically available due to consent restrictions.
ETHICAL COMPLIANCEThe study was approved by the ethical committee of the Kanton of Zurich (StV 11/09 and PB_2016-02520) and informed consent of the participating individuals or their legal guardians was obtained.
© 2023. This work is published under http://creativecommons.org/licenses/by-nc/4.0/ (the “License”). Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.
Details
; Steindl, Katharina 1 ; Abela, Lucia 2 ; Papik, Michael 1 ; Zweier, Markus 1 ; Oneda, Beatrice 1 ; Joset, Pascal 3 ; Rauch, Anita 4
1 Institute of Medical Genetics, University of Zurich, Zurich, Switzerland
2 Division of Child Neurology, University Children's Hospital Zurich, Zurich, Switzerland
3 Medical Genetics, Institute of Medical Genetics and Pathology, University Hospital Basel, Basel, Switzerland
4 Institute of Medical Genetics, University of Zurich, Zurich, Switzerland; University Children's Hospital Zurich, Zurich, Switzerland; University of Zurich Clinical Research Priority Program (CRPP) Praeclare – Personalized prenatal and reproductive medicine, Zurich, Switzerland; University of Zurich Research Priority Program (URPP) AdaBD: Adaptive Brain Circuits in Development and Learning, Zurich, Switzerland; University of Zurich Research Priority Program (URPP) ITINERARE: Innovative Therapies in Rare Diseases, Zurich, Switzerland