It has been almost five years since I wrote this post about Bionano Genomics and OpGen, and tools to look at structural variation. At that time OpGen was mapping bacterial genomes and Bionano Genomics would do insect genomes (about 100x as large), and the open question at that time was whether these technologies could scale to the entire human genome, or a factor of another five-fold for Bionano.
As of last year’s Advances in Genome Biology #AGBT17 conference, I then shared this update on BioNano Genomics and their new Saphyr instrument, and the applications that Bionano early customers highlighted then for genetic disease and for cancer.
New Technology at #AGBT18
At this year’s Advances in Genome Biology and Technology conference, several vendors were noticeably absent from the array of suites at the venue. Indeed, the sponsor list for Gold, Silver and Bronze at the AGBT website numbers thirteen in total, and there were perhaps four additional suites from the ‘other sponsors’ list. If you have never attended AGBT before, vendors get a suite room to setup, instead of a tradeshow event ‘booth’ that you are undoubtedly familiar with. One omission was Bionano Genomics, however they did have no less than four posters presented.
These posters all highlighted their new Direct Label and Stain (DLS) technology, which cuts in half the amount of imaging, doubles the sample throughput, cuts in half the per-sample price, and gives much longer maps (on the order of 60-fold). It replaces the original nick-and-label method used before, which had to use two different enzymes on the same sample in separate imaging steps.
The weakness of the prior nickase method
By using two different nickase enzymes, the same sample would have random nick sites that would be scattered throughout the genome, thus the maps would be single-molecule restriction-site-like maps that could range up to megabases of unique regions of the genome. They call this the Nicking, Labeling, Repairing and Staining (NLRS) method. The randomness of the nicking, however, could coincidentally have nicks close to each other on opposite strands, thus forming a fragile site.
I’m told by the company that these fragile sites can form when 50 to a few hundred nucleotides separate the nicks, depending on the GC content, which makes sense. The process of uncoiling DNA and the hydrodynamic forces involved at the molecular level puts strain on getting individual molecules in a straight line.
These fragile sites would then break the single long molecule in their nanochannel microfluidics platform, limiting the length of the molecules imaged and thus the length of the overlapping fragments that comprise the overall map.
Advantages of Direct Label and Stain (DLS)
By using a single enzyme, a single DL-Green dye adsorption and cleanup step, and then a subsequent staining step, the overall sample labeling process takes a similar time (about 2 days) compared to the prior NLRS method.
By using a single enzyme/labeling scheme, the DLS then will only require 1/2 the imaging, effectively doubling the overall throughput (and halving the per-sample cost, as you can now use 2x the real estate on the Saphyr chips).
It is the elimination of the aforementioned fragile sites (where the two different nick sites are too close together causing the single long molecules to break) that is the real headline. Below is an image from one of their posters (found online here) – a 2.34 megabase molecular map from human chromosome 3; in a second poster they also show the entire arms of chromosome 3 being assembled.
Another interesting aspect of single molecule human restriction maps is the ability to detect large structural variants (insertions, deletions, transversions and translocations). They show detection sensitivity data of heterozygous insertions and deletions down to 500 bp, inversions down to 50 bp, and translocations to 70 bp.
For those unaware, these large structural variants can be impossible to detect via array Comparative Genomic Hybridization (aCGH) or mate-pair short-read libraries via NGS; the only other approach to-date has been single-molecule long read sequencing (i.e. PacBio or Oxford Nanopore). More about Oxford Nanopore in a future post, as the first PromethION data is coming out and some interesting work on the GridION was presented at #AGBT18 from Dick McCombie at Cold Spring Harbor Laboratory about work in-progress.
A long road to get here
The first time I wrote about Bionano Genomics in 2012 I had doubts about the scalability of the technology; last year about this time the new Saphyr instrument was promising up-scaled technology for both rare disease and for oncology; and this year by halving the price, doubling the throughput and vastly increasing contig lengths this is quite the accomplishment.
And yet, and yet… there needs to be a direct connection where this technology can be uniquely suited for a disease, where no other technology could fit. Looking over their extensive publication list, the predominant types of publications are for different cultivated plant species, whether wheat or tobacco or millet.
Is the road ahead much shorter?
One idea would be using this technology to characterize samples where a Mendelian disorder was suspected, but whole-genome sequencing (WGS) came up negative for any causal mutation using conventional short reads. If memory serves correctly, the diagnostic yields of trio WGS or WES (whole-exome sequencing) hovers in the 25% to 35% range.
Say if 70% of samples were run through the Saphyre using DLS, and 1/3 of them come up with a large structural variant or a transversion or an inversion (believe me, the genome can get mutated in very complex ways) the yield will then rise 23% (1/3 of 70%), going from an overall yield of 30% to 53%.
Another effort is underway in Neonatal Intensive Care Units (NICU), in particular the San Diego CA Rady Children’s Hospital (Dr. Stephen Kingsmore) and in Fairfax VA INOVA Hospital System (Dr. Benjamin Solomon). As a newborn in the NICU has unique challenges, Dr. Kingsmore has shown higher diagnostic yields compared to the 25% to 35% for Mendelian disorders referenced above, but nonetheless the same argument could be made: any increase in the diagnostic yield is a large net positive.
Eric Vilain did publish his work using Bionano Genomics optical maps for pathogenic mutation detection in Duchenne muscular dystrophy (DMD).
For characterization of cancer samples, it may take a bit longer to show the value of somatic structural variation to guide existing therapeutic choices; also the issue of sample processing would need to be tackled. Standard hospital sample handling procedures dictate FFPE preservation of tissue, which will cross-link and damage the DNA so as to be unusable on the BioNano platform. Thus samples would need to be fresh-frozen, which would necessitate a global change in how samples are processed.
This is not impossible; fresh-frozen samples are routine for teaching hospitals attached to major academic centers for different types of clinical research studies. It being not part of standard procedure however, would present an obstacle that would need to be overcome.
And on that note Vanessa Hayes did publish in March 2017 her work on structural rearrangements in prostate cancer here.
I had to page through 50 references on the Bionano website to find these two publications for human disease, and if you add another reference or two on human reference assemblies that’s still less than 10 percent of their publications being applied to human applications.
Sadowski H and Borokin M et al, AGBT 2018 Poster, “Labeling Human DNA with Bionano’s Direct Labeling Enzyme Avoids Nickase-Based Double-Stranded Breaks and Allows for Chromosome-Arm Length Assemblies” PDF here.
Barseghyan H and Vilain E et al, Genome Med 2017 “Next-generation mapping: a novel approach for detection of pathogenic structural variants with a potential utility in clinical diagnosis” PDF here.
Jaratlerdsiri W and Hayes VM et al, Oncotarget 2017 “Next generation mapping reveals novel large genomic rearrangements in prostate cancer” PDF here.