I caught up with the eminently quotable Alexander Hoischen at ASHG. After we spoke for about 15 minutes, I realized I should have been taking notes - and Hoischen was generous enough to go back through some key points with my pen scribbling. He can also be counted on for some gastronomy - in this case some pastries he had purchased at a French bakery elsewhere in Montreal.
Hoischen's group at Radboud University in Nijmegen works on rare diseases, with a particular focus on neurodevelopmental disorders, vision, deafness, and coagulation.
Radboud has long been a leading edge site for testing PacBio technology for rare disease sequencing. They currently have six Revios, with five used for production and the other used for research and development work, non-WGS sequencing such as Kinnex RNA-Seq or Fiber-Seq, as well as trialing chemistry updates. They've been testing out the SPRQnext chemistry and finding it capable of generating two genomes of data from each Revio flowcell.
The group has been at a pack of 5K genomes per year and expects the SPRQnext to deliver 10K next year with no required new hardware. This is on a path to to 25K genomes per year pace within two to three years - Hoischen declared "if it takes me more than three I'll resign!". That bold pledge is made expecting PacBio to start trialing a higher throughput instrument next year - Hoischen says Radboud otherwise would require 10 more Revios. In my separate interview with new PacBio CEO Mark Van Oene, the timeline of a new instrument in 2027 was confirmed.
Feeding that pipeline is Hamilton liquid handlers for DNA extraction and for library prep. Extraction uses the Chemagen magnetic rod system. For shearing, Radboud finds "pipeteering" reliable - a defined program of repeated aspiration and delivery. They see very uniform sizes and less than 5% sample loss, though small fragment removal is required subsequently. The DNA extraction and library generation components appear to have sufficient capacity to handle the expected scaling up.
Hoischen has been an advocate of HiFi sequencing enabling the retirement of a fleet (15 in his lab) of older technologies. In Radboud's experience, there's only a few aberrations which are still troublesome. The HiFi data is getting so much better that Radboud in one case found a G-homopolymer indel mutation which was missed by Sanger. AG-rich simple repeat expansions/contractions remain tricky to call, but Hoischen believes this will be solved as well. Short reads remain the method of choice for formalin fixed, paraffin embedded pathology samples, non-invasive prenatal testing, and minimal residual disease detection - all regimes where the input DNA is highly fragmented.
I asked him about Robertsonian translocations, and he said this is still tricky but there is informatics work to detect subtle copy number losses near the
telomeres in these cases. He also noted that cases with Robertsonian translocations are strongly foreshadowed by a handful of clinical aspects, such as prior Robertsonian translocation diagnoses in the same family or recurrent spontaneous abortions.
Hoischen is an advocate for "nearly perfect genomes". He pointed out that with true long read genomes, truly contiguous haplotypes can be obtained over very long distances. So even with low coverage, an analyst can be confident that a given genomic site is homozygous because reads of both haplotypes can be observed spanning the site. This can lead to what he referred to as "hot VUS" - if you can have extreme confidence that all alleles have been ascertained in genes of interest, then seeing a Variant of Unknown Significance as the only candidate variant leads to a much stronger argument that the VUS is probably the underlying cause. Conversely, his group has seen cases in which a VUS is deprioritized as a candidate cause because a structural variant is detected by long reads which is a much stronger candidate.
I asked about using methylation information. Hoischen feels these can be very important, but that the tools for interpreting methylation are still embryonic. For example, ideally there would be large databases of methylation patterns ("episignatures") which could be compared against, but these are just starting to emerge.
Radboud will continue to be a pathfinder in long read rare disease work - I look forward to my next conversation with Alex Hoischen.
No comments:
Post a Comment