These are my notes from a subset of talks at the AGBT2019 Precision Health conference held in La Jolla, CA on September 5-7, 2019.
Dr. Rehm’s slides are here. She opened by telling this story about a misinterpretation of a clinical genome that could have been avoided through better data sharing. In a happier more recent story, a family contacted her for advice on a sequencing result based on her lab’s ClinVar entry, and together they were able to classify it as likely benign. This illustrates that we are making progress in sharing data to enable clinical genome interpretation. But there is still a long way to go, and Dr. Rehm is leading ClinGen, an effort to create a centralized resource that annotates the clinical significance of variants and genes.
Their framework for scoring genetic association evidence is published [Strande 2017]. The level of evidence (from “definitive” down to “refuted/disputed”) that should be required in order to justify acting on it clinically depends on the use context — for predictive testing, only the two most definitive categories of associations should be used, whereas for diagnostic testing, associations with moderate evidence can count too. The proportion of genes that were deemed “definitive” varied by disease area and was sometimes quite low. This has changed clinical practice. For example, InVitae now only reports the one high-confidence gene for Brugada syndrome, while allowing physicians to still order the other tests in case they become informative in the future.
Dr. Lo spoke about testing of cell-free DNA in plasma in two contexts: fetal DNA in maternal plasma, and tumor DNA in cancer patients’ plasma.
He first reported the existence of fetal DNA in maternal plasma 20 years ago [Lo 1997], using Y chromosome DNA as a marker of a male fetus. The proportion of cell-free DNA that is fetus-derived peaked at 15%, enough to support diagnostic testing. Since then, the use of cell-free DNA for fetal testing has exploded [Chitty & Lo 2015], for example, about 20% of all pregnancies in China now have this testing done.
The fetal DNA fragments (these are naturally occurring fragments — the DNA is not sonicated) have a different size distribution than maternal DNA. The size distribution arises from the fact that the mother and the fetus have different preferred fragment end positions, which they have mapped out using informative SNPs [Chan 2016]. Using the ratio of these different fetal-preferred ends to maternal-preferred ends, it is possible to calculate the proportion of fetal DNA. They are now cataloguing preferred end sites in different tissues. The different end positions might be due to different positioning of nucleosomes — they are still mapping out the molecular basis [Cheng 2018, Serpas 2019].
Some of his lab’s latest work has been on discriminating linear vs. circular mitochondrial DNA in plasma and correlating this to tissue of origin [Ma 2019].
Dr. High was an HHMI investigator at UPenn / CHOP until five years ago when she left to spin out a startup, Spark Therapeutics. She described the history of gene therapy and her work in developing therapies for RPE65 and F9, some of which is also covered in a recent review [High & Roncarolo 2019].
She described gene therapy as having a quiet childhood but difficult adolescence. The first clinical gene transfer occurred in 1990, but the death of Jesse Gelsinger in 1999 and development of leukemia in some children in a trial in 2003 dampened enthusiasm for the modality until more recently.
As I described in my gene therapy post, Spark’s voretigene and Avexis’s onasemnogene were the first two gene therapy drugs approved by FDA. But alipogene tiparvovec (Glybera®), back in 2012, was actually the first such drug approved by EMA in Europe. (That drug never got approval in the U.S., only 31 people were ever dosed, and the sponsor has allowed the approval to lapse. European patients with LPL mutations now have access to volanesorsen, an antisense oligo against APOC3). In just the past couple years, gene therapy IND applications to FDA are now increasing sharply. Most recent efforts involve adeno-associated virus (AAV). Each virus weighs in at about 5.1 MDa molecular weight, of which 74% is protein and 26% is DNA.
She transitioned to talking about Spark’s voretigene neparvovec [Maguire 2008, Russell 2017]. There were some big challenges in clinical development of this drug: they had an ultra-small patient population with a dearth of natural history data. Regulators were looking for a clinical endpoint showing improved vision. Because everyone has two eyes, there is a “nearly irresistible temptation” to inject one eye and use the other as a control, but this trial design was considered unacceptable because it is not how anyone would actually use the drug if approved — everyone would want both eyes treated. They therefore had to do a controlled trial with some participants randomized to placebo. They then offered an open label extension after 1 year, in which the placebo patients transitioned to active drug. This made the design more acceptable to the patient community.
Spark also has a program in hemophilia B, which presents different challenges. Here, the endpoints are well-established from other therapeutics in this class of diseases. But whereas voretigene was a small dose into a confined, immune-privileged tissue (the retina), this program requires dosing large amounts of virus into circulation to reach the liver. In this context, host immune response is a more formidable problem, and at first they achieved only transient transgene expression [Manno 2006, Mingozzi 2007]. To address this problem they have encoded a more potent version of Factor IX (with a potentiating missense mutation), which achieves therapeutic effect with lower expression, into an AAV, and trials are ongoing [NCT03587116].
In Q&A, I asked if Dr. High could offer any advice for those of us developing drugs, as to why and how her RPE65 gene therapy succeeded while other very similar programs that had commercial sponsors and entered the clinic ultimately failed. She said that success is the aggregate of getting many small things right, and that there is no one single thing that made the difference. Two things she thinks were important were that 1) in their AAV manufacturing, they added a step to get rid of empty AAVs, whose presence would make it more difficult for active AAV to enter the target cells, and 2) they used a more aggressive immunomodulatory regimen to keep patients from rejecting the vectors.
Sonia Vallabh & Eric Minikel
We had a “fireside chat” interview with Mike Talkowski about our patient-scientist journey, and later in the afternoon we gave a scientific talk (which largely covered the same research from Sonia’s talk at Prion2019 so I won’t re-blog it here). I mostly wanted to note two interesting discussion points from the Q&A.
Dr. Katherine High noted that recessive loss-of-function mutations in RPE65 have been referred to by over 20 different clinical names, and that one important struggle in the effort to get voretigene neparvovec approved as a gene therapy was trying to get a label for RPE65 mutations, rather than for one clinically defined phenotype. Ultimately they prevailed — the FDA-approved label is “for the treatment of patients with confirmed biallelic RPE65 mutation-associated retinal dystrophy”. She noted that it appears the PrP lowering approach in general, and antisense oligos specifically, will be appropriate for anyone with PRNP mutations regardless of clinical subtype. She said therefore one thing we will need to think about is making sure we get the drug approved for (and prescribed for) prion disease and/or PRNP mutations as opposed to just one of the historical clinical names (CJD, FFI, etc.)
Dr. Heidi Rehm asked whether, based on our experience, we think there is a minimum threshold level of patient population size necessary in order for it to be feasible to engage an industry partner to develop a drug. We referred back to our “pistachio analogy”. If technical and regulatory success in drug development were both certain and instant, then basically any disease, no matter how rare, would be profitable (the nut is always delicious); the question is how difficult the nut is to pluck out of the shell, and therefore much work someone else (such as patient groups) needs to do to help make it easy enough. In other words it seems like there is not a binary, but rather a continuum — the rarer the disease, the more that industry will need a target to be de-risked before they can get involved. (Though the nature of the drug development task can certainly change as you get closer to N of 1).
Susan Slaugenhaupt & Nicolais Naryshkin
Dr. Slaugehaupt has devoted her career to the disease caused by recessive partial loss of function of ELP1 (formerly known as IKAP or IKBKAP), which is known clinically as familial dysautonomia (FD), Riley Day syndrome or hereditary sensory and autonomic neuropathy (HSANIII) [Norcliffe-Kaufmann 2017]. Most cases are caused by an extended splice site mutation that excludes exon 20, resulting in a tissue-specific reduction in (but not complete loss) of protein levels. This mutation has a carrier frequency of 1/27 in Ashkenazi Jews. After solving the genetic cause of the disease [Slaugenhaupt 2001], Dr. Slaugenhaupt partnered with a NINDS drug screening consortium and screened a thousand compounds and found one hit: kinetin [Slaugenhaupt 2004], which (after a long effort to develop an faithful mouse model) turned out to work in vivo [Morini 2019]. They tried testing kinetin in humans, but it turned out it wasn’t very potent and required enormous doses that were not well-tolerated. With support from the NINDS Blueprint for Neuroscience Research they launched a three-year-long medicinal chemistry effort to improve upon kinetin, developing and screening 520 analogues in a dual luciferase assay [Salani 2019] and identifying 3 compounds to advance to in vivo testing. At the end of 2015, their efforts to date attracted the interest of PTC Therapeutics and they signed an agreement to collaborate on developing a new splice-modulating small molecle for ELP1 patients.
PTC Therapeutics is committed to developing splice-modulating small molecule drugs, and Dr. Naryshkin has led their flagship program for spinal muscular atrophy [Naryshkin 2014]. ELP1 is their second target, and they also have a program in Huntington disease (HTT). They were attracted to ELP1 as a target by the genetically well-defined target, proof-of-concept compound (kinetin), established assay and medicinal chemistry, leading academic collaborator (Dr. Slaugenhaupt), and oustanding patient advocacy organization. Kinetin has a 10 μM potency (concentration to double ELP1 expression, indicated by E2X). The MGH/Blueprint collaboration yielded a 400 nM compound, and PTC has now achieved 0.3 nM. It has clear dose-responsiveness in patient cell lines and in vivo in the transgenic mouse brain after oral dosing. They anticipate entry into the clinic this year.
Dr. Korf introduced the Alabama Genomic Health Initiative (AGHI) based at University of Alabama Birmingham in collaboration with HudsonAlpha Institute. They have two cohorts: a population cohort (healthy people) that undergoes a genotyping array and get results returned for any actionable variants, and a phenotypic cohort (affected patients) that undergoes whole genome sequencing with return of pathogenic variants.
In the population cohort, about 1.4% of the 5,000+ participants have had an actionable result returned based on ACMG v2.0 guidelines. The genotyping chip does not include all possible medically actionable variants. A majority (about two-thirds) of population cohort participants who had an actionable result had reported a family history that would be classified as moderate or high risk, but surprisingly, only about half the time was that family history suggestive for the phenotype associated to the actionable variant that was identified — the rest of the time it was apparently coincidental. For the phenotypes associated to their actionable variants, most of the people would have been assessed as being at a typical population risk level if not for the genotyping result.
In the phenotypic cohort, they have 200 families, some of which are sequnced as trios so in total 491 whole genomes have been sequenced. They’ve identified putative causal variants in tens of different genes.
They’ve recently had a few requests for return of raw data, which concerned them because they then have no control over how results are interpreted. They therefore submitted an IRB protocol specifically to educate people on the cautions and caveats of using their own raw data.
Looking forward they are aiming to engage with more clinical practices throughout the state, and start implementing polygenic risk scoring.
Dr. Geschwind is based at UCLA’s Geffen School of Medicine. UC Health has 14-15 million patients, so it presents a big opportunity for genomic data aggregation. He is speaking today about the California Center for Rare Diseases and the UCLA Precision Health project ATLAS, and specifically how they are moving from whole exome to whole genome to RNA-seq for diagnosis. Early on they anticipated that personnel time to consent patients could be a major bottleneck so they developed a video-based consent system, translated into 8 languages. To date, over 50,000 patients have consented, of which 30,000 have opted in and 20,000 have been genotyped. Consistent with Los Angeles demographics, there is no majority ethnic or racial group in the cohort; the plurality are Latino (46%).
They began offering exome sequencing for rare disease patients in 2012. Since then, they have had 337 physicians from 67 institutions order tests, totaling thousands of patients. The most common type of phenotype is neurological or neurodevelopmental though there are also a lot of neuromuscular diseases. After exome sequencing, they reach a conclusive diagnosis in 26% of cases, potential diagnosis in 22%, and identify a strong novel gene in 3% of cases. “Potential diagnosis” means the variant or gene does not perfectly match the phenotype, so that further phenotyping or segregation analysis is needed in order to decide whether it’s causal. The novel disease genes discovered through this program include KAT6A [Arboleda 2015] and a number of autism genes. A number of different mutational mechanisms can contribute to autism including CNVs [Luo 2012], exonic SNPs or short indels, and intronic mutations that affect the transcriptome but may be missed on exome and uninterpretable on whole genome sequencing.
They are finding that moving from exomes to whole genomes and ultimately RNA-seq can improve diagnostic yield. For example, out of a cohort of unsolved patients, they were able to diagnose 11 using exomes alone, and another 7 using whole genomes. Out of 48 that then underwent RNA-seq, they were able to reach a diagnosis in 7. In one example, exome sequencing identified a single paternally-inherited hit in a known recessive disease gene that matched the phenotype; only through RNA-seq were they able to figure out that the second hit was a rare variant inherited from the mother, which was synonymous but turned out to affect splicing. In most cases like this, no deleterious effect on gene expression or function would be predicted by any bioinformatics software — only RNA-seq will solve it.