Is Charcot-Marie-Tooth disease a good disease prototype for understanding WGS?

Charcot-Marie-Tooth (CMT) disease is an umbrella term for inherited neuropathies but is a clinically and genetically heterogeneous group of diseases. The clinical subtypes of CMT include demyelinating sensory and motor neuropathy (CMT1), axonal sensory and motor neuropathy (CMT2), sensory and motor neuropathy with intermediate conduction slowing (upper limb motor conduction velocity between 25 and 45 m/s, CMTi), hereditary sensory neuropathy (HSN) and hereditary motor neuropathy (HMN).3 4

The diagnostic utility of WGS for an individual lies in its ability to detect vast numbers, and in theory different types, of genetic variant. Figure 2 illustrates the features of a disease group that make it suitable for considering and understanding WGS testing.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Understanding the utility of WGS. CMT, Charcot-Marie-Tooth disease; WGS, whole-genome sequencing.

One downside of CMT as a disease prototype is that functional validation of novel variants/genes is challenging but this underpins how important WGS is in CMT clinical practice. Gold-standard functional evidence would be ex vivo human diseased tissue demonstrating absent, deficient or dysfunctional protein contributing to pathology. This is possible in theory with peripheral nerve biopsies, but this is an invasive procedure requiring technical expertise. Alternatively, RNA sequencing can be used to demonstrate aberrant transcripts in appropriate tissues; Schwann cells are clearly easier to study than dorsal root ganglia or anterior horn cells. Overall, we feel CMT is an excellent disease to demonstrate the lessons and pitfalls of WGS and will explore these herein.

WGS technologies

A basic understanding of the molecular techniques involved in WGS is important to appreciate its potential pitfalls. First, WGS when used in common medical parlance, refers to ‘short-read’ WGS (srWGS). Table 1 highlights some useful terminology. There are other forms of genomic sequencing, and although used currently mostly in the research setting, their use is increasing in diagnostic genetic laboratories worldwide. Long-read WGS (lrWGS), as suggested in the description, continuously sequences long molecules of DNA, typically tens of kbp in length, but up to many hundreds of kbp depending on the sequencing technology used. The major benefit of lrWGS is the ability to detect and size repeat expansions accurately, and to detect complex, balanced structural variants. The drawbacks include the cost and longer sequencing time, and its error rate on an individual nucleotide level which, when combined with low read depth, affects its ability reliably to detect single-nucleotide variants (SNVs) or insertion-deletions (indels).5 Optical genome mapping is another form of genomic interrogation, and more appropriately termed ‘genome imaging’. Its uses have been compared with those previously investigated with karyotyping (ie, large structural variants) but with the benefit of up to 20 000-fold higher resolution. DNA molecules are enzymatically labelled, and the resultant ligated DNA then ‘imaged’ for its pattern of periodically spaced fluorescent signals. Its ability to detect large structural variants (0.5–1 Mbp) is superior to srWGS and lrWGS, and it is less costly to get higher coverage. As with srWGS and lrWGS it cannot detect aneuploidy (an abnormal number of chromosomes), although this is less relevant in the setting of non-developmental disorders. Another potential drawback of optical genome mapping is the requirement for DNA extraction from a fresh blood sample.6 As neither lrWGS nor optical genome mapping are used in standard NHS testing, from this point forward we will not discuss them further, and we will refer to srWGS simply as WGS.

Next-generation sequencing technology

WGS uses next-generation sequencing (NGS) technology, also known as high-throughput or massively parallel sequencing. NGS has been used for many years in clinical diagnostic laboratories for the sequencing of disease-specific gene panels and whole exome sequencing. There are several sequencing platforms,7 but the dominant provider worldwide is Illumina, which is also used by NHS-GMS, and the process described hereafter. A flow diagram of the process involved is shown in figure 3.8

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

WGS workflow. DNA library is prepared through fragmentation of the extracted DNA, followed by amplification on the flow cell. Sequencing then occurs through ‘sequencing by synthesis’ and the subsequent read data is aggregated in a .fastq file. The data are then processed through a bioinformatic pipeline; initially aligned to the reference genome (resulting in a .bam file) and then variants are identified (‘variant calling’) yielding a .vcf file; the common output file of WGS. Virtual panels can then be applied, and the resultant variants interpreted by clinical scientists, before a report is issued. WGS, whole-genome sequencing.

The first step is library preparation (figure 3A); the genomic DNA library is a series of short fragments ready for sequencing. The DNA (typically extracted from leukocytes in blood; purple EDTA tube) is fragmented and then each fragment amplified. Fragments are then sequenced in a process called ‘sequencing by synthesis’, whereby fluorescently tagged nucleotides are added to a linear single strand of DNA complementary to the fragment; the resultant fluorescent DNA strand is known as a ‘read’ and can be sequenced by its characteristic spectral emission (one wavelength for each of the four nucleotides, figure 3B). The fragment is sequenced from both ends forming ‘paired-end reads’, allowing additional information to be gleaned when the reads are aligned. Data are then fed into the bioinformatic pipeline (figure 3C). The millions of reads are aligned to the reference genome, which when visually represented, form piles of overlapping reads. The overall coverage of the WGS describes what proportion of the reference genome is sequenced to a satisfactory read depth. Figure 4A shows in detail how an unmutated fragment is sequenced and aligned to the reference.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4

Bioinformatic methods of detecting variants. (A) Schematic of the normal process of paired-end reads of a single fragment, subsequent read alignment to the reference genome and the resultant read depth. (B) Visual representation of detection of a single nucleotide variant (SNV). One allele of the individual’s DNA contains an SNV. When the aligned reads are analysed, 50% have a nucleotide that is different from the reference and the variant is ‘called’. (C) Structural variant detecting methods: ‘Split-read’ and ‘read-pair’. (i) A deletion in the fragment of DNA means that when the read-pairs are aligned to the reference they will appear too close together (red markers are closer than without the deletion). (ii) An insertion means the aligned reads appear too far apart when aligned to the reference (red markers further apart than without the insertion). (iii) A translocation results in one half of the read-pair aligning to a different part of the genome (iv) A repeat expansion, particularly one longer than the read-pair, will result in one of the pairs sequencing only the repeat region. When alignment is attempted, it may not be able to align anywhere (the other read will be ‘unmatched’) or may align elsewhere in the genome. (D) Structural variant detecting method: ‘Depth of coverage’—aligned reads of a deleted region identify a length of sequencing with a 50% drop in coverage suggesting a heterozygous deletion, that is, one normal allele and one allele containing a deletion. R1, read 1; R2, read 2; together form the read pair

Variant calling is the process of identifying variants, that is, variation in an individual’s genome when compared with the reference. The basic output of a WGS bioinformatic pipeline is the identification of small variants; alteration/insertion/deletion of single nucleotides (SNVs, figure 4B) or a small number of consecutive nucleotides (indels). The universal final output for the millions of variants generated is a .vcf file. Other types of genetic variant can also be detected including structural variants (both copy number variants and balanced rearrangements; the latter where there is no change in dose at a particular locus), repeat expansions and mitochondrial DNA (mtDNA) variants, but their detection and calling is variable (figure 4 and see ‘When WGS might not be the correct test’).

After variant calling, the variants are filtered according to specified criteria (see ‘Filtering and prioritisation’). Application of a virtual panel(s) may yield possible candidate variants, which are interpreted by clinical scientists (figure 3D). If there is ambiguity or uncertainty, results are ideally discussed at a multidisciplinary team (MDT) meeting, following which a genetic report can be issued.

Virtual panels

Although WGS in theory allows analysis of variants from an individual’s entire genome, this is neither desirable (incidental unwanted findings) nor practical (a human genome contains approximately five million SNVs), therefore, virtual panels are essential to refine the search. In the NHS, clinicians are required to select virtual gene panel(s) when requesting WGS. The NHS-GMS PanelApp (https://nhsgms-panelapp.genomicsengland.co.uk/panels) is a publicly available resource that uses genetic expertise through crowdsourcing to curate disease-specific gene panels.9 For a gene to be included it needs to be approved as ‘green’ by a number of verified experts; a green gene is broadly one in which plausible disease-causing variants have been found in three or more unrelated individuals/families. However, the panels can only be as correct and up to date as their reviewers and the current available evidence. For example, SORD was discovered as a common, and potentially treatable, cause of CMT in 202010 but was not approved as a green gene until November 2022. Panels are periodically updated, and previous iterations can be found on PanelApp. Genes that cause a complex phenotype which include the disease group of interest, for example, ABHD12 causing polyneuropathy, hearing loss, ataxia, retinitis pigmentosa and cataract (PHARC) syndrome, are often not included if the panel specifies an isolated phenotype; it is not a green gene on the current ‘Hereditary Neuropathy or pain disorder’ panel (R78 V.3.24). Similarly, novel or rare genes may not meet green inclusion criteria. It is, therefore, important to understand which genes are tested in a specific panel, and if there is a particular gene of interest in a clinical case, this should be discussed with the genetic laboratory. It is currently recommended that broad rather than narrow use of panels is applied to maximise chances of identifying causative variants.

Filtering and prioritisation

Refining the vast number of variants detected through WGS requires filtering strategies. The two most powerful tools are the allele frequency of the variant in reference databases (the most commonly used is gnomAD; https://gnomad.broadinstitute.org/, box 1) and in family studies, the inheritance pattern, as defined by relative disease status.

Variant interpretation and reporting

Every candidate variant is classified according to established criteria. UK laboratories use the American College of Medical Genetics and Genomics (ACMG) and Association for Clinical Genomic Service (ACGS) guidelines.16 17 Any given variant, with no supporting data, starts as a ‘variant of uncertain significance’ (VUS). Evidence is combined from different categories (including data on allele frequency, functional studies, segregation and prior literature reports) to upgrade the variant as likely pathogenic or pathogenic, or downgrade to likely benign or benign (figure 5). As with gene panels, variant interpretation relies on the available evidence, and its application, and therefore variant classification may differ between laboratories. Ideally, clinicians will have access to an MDT (with clinical scientists) to discuss WGS results of unsolved cases, cases with unexpected pathogenic variants, or those with a very typical phenotype for a particular gene, in which no variants have been reported. There is a criterion within the ACMG/ACGS criteria (PP4) that uses phenotype specificity to upgrade variants, for example, absence of dystrophin in a muscle biopsy in a male patient with muscular dystrophy phenotype, when considering a variant in DMD. Without the communication of clinical information from clinician to laboratory, the variant might remain a VUS.

• Download figure
• Open in new tab
• Download powerpoint

Figure 5

Variant Classification Schematic based on the the American College of Medical Genetics and Genomics criteria. Every variant under review has evidence for pathogenicity scrutinised under each of the listed categories.

Historically, relevant VUSs were listed as an addendum to genetic reports. However, NHS-GMS has adopted guidance from the ACGS that VUSs should not be reported unless exceptional circumstances apply, after a discussion at an MDT meeting. This change is critical for practising clinicians to be aware of. The rationale is that reporting a VUS may lead to confusion on the part of referring clinician or patient, misinterpretation and potentially misdiagnosis. Even when a VUS is likely to be causative, family screening for the variant would still need careful discussion and counselling, and preimplantation genetic testing or entry into a clinical drug trial would only be considered in exceptional circumstances.

However, we have experience that transparent reporting of VUSs to clinicians with genetic expertise, has been vital in clinching a genetic diagnosis with the passage of time. A ‘warm’ VUS may be upgraded to pathogenic following, for example, a new publication implicating the gene/variant in disease. Without information about VUSs made available on a genetic report, such cases may remain unsolved.

Another example of the need for careful reporting is the presence of a single pathogenic variant in a recessive gene, a so called ‘single hit’. If reported, it should be made clear that the diagnosis is not confirmed, but a single pathogenic variant has been detected. With a suggestive phenotype, a ‘single hit’ will often trigger a discussion with the laboratory to look on the other allele for deep intronic variants (that might affect splicing or create pseudoexons), or structural variants (ie, deletion of a portion of the gene), or explorative analysis of the genome in a research setting.18

When WGS might not be the correct test

The essential first step for genetic testing is ensuring the right test is sent. Jain et al have previously discussed this in detail.1 In the UK, clinicians must consult the NHS Genomic Test Directory https://www.england.nhs.uk/publication/national-genomic-test-directories/. Many neurological diseases, including some that are treatable, have their molecular basis in non-SNV genetic variation. Huntington’s disease (CAG trinucleotide repeat expansion in HTT), genetic motor neurone disease/frontotemporal dementia (GGGGCC hexanucleotide repeat expansion in C9ORF72), spinal muscular atrophy (biallelic deletion of exon 7±8 in SMN1), fragile X syndrome (CGG trinucleotide repeat expansion in FMR1) and Duchenne muscular dystrophy (~60% caused by exon-level deletions in the X-linked DMD) are all caused by either repeat expansion or structural variants. More than 50% of CMT is caused by a duplication of PMP22, and the remainder, a mixture of genetic variant types.

The limitation of WGS to accurately detect structural variants and repeat expansions lies in the read length. Put simply, it is difficult to quantify a variant with genomic size potentially orders of magnitude larger than the unit of measurement. Figure 4 details the use of paired-end reads in the sequencing and alignment process, and how they can be used to detect non-SNV variants. When the DNA fragment is sequenced from both ends, the two paired-end reads contain markers that identify them as a pair. If, when the reads are aligned to the reference genome, they align too far apart or too close together, this can be bioinformatically detected. Similarly, if a read aligns without a ‘mate’ (the other part of the pair cannot satisfactorily align to the reference), this can also be flagged. This approach for detecting non-SNV variants is shown in figure 4C and is known as a ‘paired-end’ (or ‘read-pair’) approach to detecting structural variants. Similarly, the ‘split-read’ approach uses information that a single read is disrupted, or split, by a structural variant. The read depth or ‘depth of coverage’ approach relies on algorithms detecting regions where there is a significant increase or decrease in coverage (figure 4D). All these computational approaches have their limitations for different structural variants, and the best algorithms combine more than one approach.19 Structural variation on a chromosomal level, for example, aneuploidy or ringed chromosomes, will not be detected by WGS and karyotyping should be requested separately.

Repeat expansions, where the number of repeats is critical to the diagnosis, can be challenging to size through WGS; large repeat expansions will be longer than the read, or read-pair (figure 4Civ). ExpansionHunter is a tool that estimates the repeat size at the loci of known expansions, which when paired with visual inspection, was sensitive and specific for correctly sizing expansions in the 100KGP when the expansion size was less than the read length.20

However, there are three important caveats to the above. First, as with virtual panels, if the gene and specifically the expansion (if that is the diagnostic question) is not on the virtual panel, non-SNVs will not be tested. Second, when the expansion is larger than the read length (as seen in FMR1, C9orf72, DMPK (myotonic dystrophy type 1) and FXN (Friedreich’s Ataxia), although an expansion could be identified, it was often significantly underestimated by ExpansionHunter (figure 4Civ). Although RFC1, the gene recently identified as causing cerebellar ataxia, neuropathy and vestibular areflexia syndrome (CANVAS) through biallelic pentanucleotide repeat expansions, was not examined by Ibañez et al, the same would apply; the expansion is typically >1000 repeats (>5000 nucleotides).21 In NHS laboratories, RFC1 is currently tested using non-WGS methods. Third, early iterations of the 100KGP pipeline did not routinely analyse for any non-SNVs, and many were missed and not reported.

MtDNA sequencing is currently requested as a separate test to sequencing of the nuclear genome. Studies have shown that with a satisfactory read depth, WGS can detect mtDNA variants at a heteroplasmy level down to 10%,22 but if there is a significant suspicion for mitochondrial disease, mtDNA sequencing should be requested separately. Other types of genetic mechanism including epigenetic factors such as DNA methylation or imprinting will not be detected using WGS and should have separate testing requested. Lastly, in the NHS, if a rapid result is critical to guide management, the R14 ‘Acutely unwell children with a likely monogenic disorder’ WGS can be requested for critically ill children and adults, with a turnaround time of 2–3 weeks.23

Examples from the clinic

Key to our diagnostic success in the 100KGP was access to the data in the research environment, and regular review of cases at a dedicated clinical-research-genetic MDT. We illustrate with clinical cases practical examples of potential pitfalls discussed above.

Case 1

A woman in her late teens presented with a subacute history of sensory changes in her hands, a few weeks following a viral illness. She developed progressive weakness and wasting of intrinsic hand muscles. At initial assessment she also had mild subclinical distal lower limb weakness (figure 6A–E). There was no family history of neuromuscular illness and parents were non-consanguineous. Initial neurophysiology showed a patchy, widespread, conduction slowing neuropathy. She was treated in her local hospital with intravenous immunoglobulin for presumed chronic inflammatory demyelinating neuropathy. Subsequent CSF examination showed normal constituents, nerve roots were markedly thickened and pathologically enhancing on MRI, and nerve biopsy showed a chronic demyelinating neuropathy without inflammation (figure 6G). She progressed slowly despite treatment; initial genetic testing, including CMT1A with multiplex ligation-dependent probe amplification (MLPA), and a 14 gene panel in 2015, was negative. She was enrolled into the 100KGP with her parents, with no primary findings. Through a research collaboration we identified the variant c.4271C>T p.(Thr1424Met) in ITPR3, a gene only reported in three families and not included in the virtual panel.24 25 Additionally, the variant was maternally inherited (figure 6F). Clinically the mother had no symptoms and a completely normal neurological examination, but neurophysiology showed a clear conduction slowing neuropathy. The diagnosis is CMT, with remarkable variability in severity, due to an ITPR3 variant. This case highlighted the importance of the assigned affected status; segregation was confirmed but only by neurophysiology. Similarly, research access to the 100KGP data was essential to identify a gene not on the virtual panel but in the literature.

• Download figure
• Open in new tab
• Download powerpoint

Figure 6

Clinical cases. (A–E) Distal limb muscle atrophy in proband of case 1. (F) Integrative Genomics Viewer (IGV) showing heterozygous variant in ITPR3 (blue arrow) in proband and mother. (G) Electron microscopy of sural nerve biopsy of case 1 (proband) showing reduced myelin thickness and onion bulb formation. (H) IGV showing 1.5× read depth of aligned reads in PMP22 compared with a region on chromosome 1. (I) IGV shows loss of coverage in intron 2 of RFC1 (red circle) indicating a biallelic repeat expansion in that region. Reads highlighted in red (black arrow) are unmatched pairs. J. IGV shows ~9 kb 50% drop in coverage (read depth) in MME encompassing exons 15 and 16; corresponding to a heterozygous deletion. (K, L) Mild distal lower limb atrophy in proband of Case 5. (M) IGV shows heterozygous variant in PMP2 carried by proband and affected father.