Moving Next-Generation Sequencing into the Clinic, part 2

first attempt at live tweets from #ngs2013 http://t.co/SbWrGx8weI

— Paul Denny (@pauldennyuk) March 12, 2013

In my last post, I summarised the first four talks from this symposium:

1st Oxford Workshop and Symposium, 4th Techgene Knowledge Network Meeting,
“NGS2013 Next generation Sequencing: Bioinformatics and Data Analysis”

You could also read the Tweets from the meeting here. But I digress. The afternoon session started impressively with Marcel Nelen, from Radboud University Medical Centre, Nijmegen, Netherlands speaking about:

“Clinical utility of exome sequencing in heterogeneous diseases.”

Dr Nelen described how a strong collaboration between research and diagnostics labs led to the application of exome sequencing in diagnosis of heterogeneous genetic diseases.  He emphasised the importance of a multidisciplinary team effort to define “packages” of genes for which there was evidence of involvement in either intellectual disability (ID), inherited blindness, inherited deafness, movement disorder or oxidative phosphorylation disorders.  The wide diversity of genetic variants that can cause such diseases means that Sanger sequencing, once the gold standard, has become too time consuming and expensive and gives a lower “diagnostic yield” than exome sequencing.

One critical aspect of applying high-throughput sequencing of exomes in a diagnostic setting is gaining appropriate informed consent from patients and their families.  Patients and/or their parents gave informed consent for the entire exome analysis. The sequence data from the exomes of ~550 patients entered a generic annotation pipeline and exome analysis was based on either a ‘de novo’ strategy for ID or a ‘in silico’ targeted strategy for the other diseases.

Analysis is based on a two stage approach:

1.  Analysis of a “package” of genes defined as highly likely to carry pathogenic variants:
   • Search for variants in only these disease-related genes.
   • If pathogenic variant found, end analysis and report.
   • If no pathogenic variant found, move on to the second stage:
2. Whole exome analysis:
   • If no pathogenic variant found, search for mutations in a specific  set of candidate genes.
   • If a pathogenic variant is found and there is solid proof for clinical interpretation, report.
   • Finally, if the earlier analyses fail, the remainder of the exome  is searched in collaboration with researchers and might be reported on if this “new” data holds.

One success story recounted by Dr Nelen concerned a specific case of intellectual disability, in which no causative variant had been identified over several years of investigation by Sanger sequencing of successive candidate genes, whereas whole exome analysis identified a pathogenic variant in the PACS1 gene.

In cases where “incidental” variants are identified, that appear to bear no relationship to the disease under investigation, these are passed to and assessed by an independent team of experts for advice, prior to reporting.

This tiered analysis has gained certification by the Dutch medical authorities as a genetic test and so might act as a model for other EU countries.

We heard next from Michael Mueller, from the NIHR Biomedical Research Centre, Imperial College London, UK, about:

“Rapid whole-genome sequencing: optimising the bioinformatics pipeline for faster turnaround times.”

Using whole-genome sequencing (WGS) for mutation detection can be more powerful than analysis restricted to just the exome. However, the data processing and handling challenges posed by moving WGS into the diagnostic lab are immense.  Dr Mueller described how different hardware and approaches to parallel processing of NGS data could be optimised, presenting some dramatic improvements.  By systematically identifying bottlenecks at each stage, he was able to reduce the time taken to produce  annotated variants, starting with raw Illumina short-read data from a ~ 30x coverage single genome, from ~24 to ~7 hours.  This appeared to be achieved without compromising read-mapping or variant calling quality.

In a follow-up to the talk from Marcel Nelen, Kornelia Neveling, also from  the Radboud University Medical Centre, Nijmegen, spoke about:

“Data analysis for diagnostic exome sequencing”

Dr Neveling began by describing the technical setup at the Radboud University Medical Centre, then sample handling and quality control and finally software tools to help in exome sequence variant filtering and analysis.   The diagnostics lab has access to three Life Technologies 5500 sequencers using the SOLiD platform and also two Ion Torrent PGM sequencers.

A critical step in sample and data handling is quality control (QC) to ensure that final clinical decisions are robust; some of the QC steps outlined by Dr Neveling were:

1. Reliable identification of individuals and relationships (sibling / parent / unrelated)
2. Accurate recording of metadata for each sample e.g. which software version was used for analysis? What was the date of analysis?, etc.
3. That sequence data are highly specific and sensitive e.g. give sufficient, even coverage of all exons.
4. Sequence variant calls meet biological expectations e.g. the ratio of transitions to transversions reflects natural variation.

The Radboud sequencing diagnostics team now have a database of about 1500 individual exomes and have found that a typical exome yields about 40,000 variants, with ~150-200 of those variants being “private” to each sample. Dr Neveling presented a variant filtering tool, with a graphical user interface that looked reminiscent of the software described earlier by Elliot Margulies and described briefly using the tool in a case of hereditary spastic paraplegia to aid in identifying the most likely etiologic variant.

Prof. Anthony J Brookes ended the Symposium, talking about:

“Assigning pathogenicity to NGS-derived variants”

Prof. Brookes was fizzing with ideas, provoking us to think about what we really mean by “pathogenic”.  Focusing on rare diseases, we were reminded that the concept of pathogenicity is a slippery, multi-faceted one.  Inferring pathogenicity can mean some combination of:

• knowing allele frequencies in case and control populations
• whether a variant has been described by others as pathogenic
• whether a variant is absent from databases that are assumed (sometimes wrongly) to consist mainly of “normal” variants e.g. dbSNP
• whether a variant co-segregates with disease in a pedigree
• what the predicted (or known) effect of a variant is on protein structure
• predictions in silico from tools such as PolyPhen or Sift
• functional assays performed in living human cells
• functional or phenotypic assays conducted in model organisms e.g. mouse mutants

Another level of subtlety in the way in which we define pathogenicity is that we can think of two contexts:

1. has a variant ’caused’ a phenotype in a particular patient or family (which relates to expressivity)?
2. can a variant ’cause’ a phenotype in a population (penetrance)?

Prof. Brookes argued that the clinical actionability of a variant should be thought of as a combination of pathogenicity, penetrance and expressivity. He went on to point out that too little is known about the relationship between genotype and phenotype and that we need a number of developments to bridge that gap and improve our ability to recognise pathogenic variants. In particular, he argued that a intermediary database system was required to link together primary resources such as dbSNP or Ensembl with clinical databases, to facilitate data-sharing without compromising confidentiality.  Combining this database  ’ecosystem’ with high data-quality electronic health records should improve our understanding of the genotype-phenotype relationship.

The symposium gave a good overview of the way in which NGS is being taken up in diagnostic genetic labs and used to improve the success rate in identifying causative variants.  In turn, this technological development should lead to better informed choice of therapies or treatments.  It will be intriguing to see how far we have progressed in a year’s time, when a followup meeting is planned.

If anyone would like to get in touch to discuss, correct or update my summaries, please post a comment, below, or send me a tweet:

Tweet to @pauldennyuk