Dyno Therapeutics - AAV Vectors for Gene Therapy
Thanks to all who are reading this week - disagreements and corrections are welcome.
There’s been considerable excitement about both gene therapy treatments and biotech platforms over the last few years. Gene therapy is seen as having promise for the treatment of rare diseases, leading to attention from both pharmaceutical companies and CDMOs. The FDA’s press release on its approval of the drug Luxturna is a good place to start for understanding the treatment method’s potential:
‘Biallelic RPE65 mutation-associated retinal dystrophy affects approximately 1,000 to 2,000 patients in the U.S. Biallelic mutation carriers have a mutation (not necessarily the same mutation) in both copies of a particular gene (a paternal and a maternal mutation). The RPE65 gene provides instructions for making an enzyme (a protein that facilitates chemical reactions) that is essential for normal vision. Mutations in the RPE65 gene lead to reduced or absent levels of RPE65 activity, blocking the visual cycle and resulting in impaired vision. Individuals with biallelic RPE65 mutation-associated retinal dystrophy experience progressive deterioration of vision over time. This loss of vision, often during childhood or adolescence, ultimately progresses to complete blindness.
Luxturna works by delivering a normal copy of the RPE65 gene directly to retinal cells. These retinal cells then produce the normal protein that converts light to an electrical signal in the retina to restore patient’s vision loss. Luxturna uses a naturally occurring adeno-associated virus, which has been modified using recombinant DNA techniques, as a vehicle to deliver the normal human RPE65 gene to the retinal cells to restore vision.’1
Put more briefly, a patient is born with a genetic mutation that results in the gene not functioning properly. In the above case, normal copies of the gene (150 billion of them!) are inserted into viral vectors and then injected into the patient’s eyes.2 The result is substantially improved vision, although not a complete cure.
Additionally, the success of Moderna and Illumina has investors searching for other instantiations of the platform concept in biotech; included in that cohort is a considerable number of those who cut their teeth in software. The basic thesis is that, just as software has become increasingly modular, so will biopharma become increasingly modular.3 Stripe and Twilio have become so successful because it doesn’t make sense for companies to divert attention away from their main products towards building in-house payments or communications systems. DoorDash is far more likely to beat Uber Eats if it focuses effort on improving food-delivery times and driver utilization, not if it focuses on constructing a superior payment processing system. Many venture investors hope a similar dynamic will occur in biopharma. This modularity has already happened to some degree, as evidenced by CROs, CMOs, CDMOs, and the existence of Illumina. The thought going forward is that this outsourcing will only become more common. DNA synthesis will be done by Twist, synthetic biology work will be done by Ginkgo, and lipid nanoparticle design will be done by an external specialist rather than an mRNA therapeutic company.
Dyno Therapeutics seeks to capitalize on this increased interest in platforms and gene therapy. Rather than developing its own therapeutics, it focuses exclusively on designing AAV vectors that would then be used to deliver in vivo gene therapies.4 As mentioned in the FDA press release, gene therapy can be divided into two components: the gene it seeks to deliver, such as a normal RPE65 gene, and the viral vector that carries the gene. These components are essentially two different problems for a biopharma company to solve. Figuring out which gene will cure or partially cure a disease is all well and good, but represents only half of the challenge. Having a copy of the normal gene isn’t all that helpful if you can’t find an appropriate viral vector to deliver said gene. Dyno focuses exclusively on this vector design problem, and for good reason. There are four common issues that surface (no pun intended) with AAV vectors:
1) They can induce an immune response.
2) They don’t get to the right cells, or not enough of them get to the right cells. A normal RPE65 gene is only helpful if can actually get to the retina! Viral capsids that run into these problems are described as having poor transduction efficiency. You’ll also see the word tropism come up frequently in this context, which refers to a virus’ ability to infect a particular cell or organ type.
3) The AAV vector gets into the wrong cells. Having high transduction efficiency but in the wrong place is not a good thing!
4) The AAV genome is small, and so it’s challenging to fit the gene of interest into the capsid (a virus’ shell).
That said, AAV has some notable advantages. The immune response induced by the vector is far less than that induced by the adenovirus equivalent. A substantial number of people have been infected by some sort of adenovirus as a child, resulting in innate immunity that seeks to fight it off when reintroduced as part of a gene therapy. A strong immune response makes it harder for the therapeutic gene to successfully produce a protein within the desired cells, and so a lot of companies are focusing their work on AAVs instead.5 AAVs are not known to cause any human disease, and so the initial immune response is substantially lower (although not nothing; it’s still seen as an intrusion!)6 As one would expect, this doesn’t remain the case over time, and redosing can end up causing an immune response. AAV vectors are also viewed favorably because they’re not able to replicate without the presence of a helper virus. In some early gene therapy studies using an adenovirus vector there were unfortunate cases where the virus successfully replicated in the patient, and so made people sicker rather than better. The adenovirus can be modified to not replicate, but it’s easier to begin with a virus that can’t replicate anyway!
Issues 1-3 outlined above are actually all interrelated, as explained in a paper by some Dyno team members:
“Among all capsid properties that could be improved, increased tissue-specific transduction is key to enabling safe and effective gene therapies. Improving this attribute would allow for a higher proportion of injected capsids to deliver their payloads to the intended cells, reducing the dose needed for effective treatment. This is turn would make treatment safer by reducing activation of the innate immune response and of B and T cell responses, which increase in magnitude relative to the amount of antigenic stimulus (vector dose) delivered.”7
The better tissue transduction is, the more effective individual capsids are. The more effective each capsid is, the lower the required dose, and the lower the patient immune response ends up being. The follow up question then is how these better capsids can be designed. Here things quickly get complicated. Historically, there have been two predominant methods to design better AAV vectors, whether focused on optimizing the capsid or otherwise:8
1) Random mutagenesis – as the name suggests, this method uses error-prone PCR to introduce mutations into a sequence of interest. The benefit of this method is it doesn’t require much thought, so it’s easier to generate many mutations. The downside is that it’s random and, in the case of AAV vectors, it’s quite easy to introduce a mutation that breaks capsid function.9
2) Rational Design – in this method scientists think carefully about what mutations might make sense to introduce and then alter the virus accordingly. This should lead to more desirable mutations but is much more time and brainpower intensive. It’s also limited in the sense that we just don’t yet understand the AAV virus all that well, making rational design a harder task!
Dyno’s existence is based on the thesis that designing better AAV vectors is fundamentally a data problem. The more data that can be gathered on how individual mutations affect the virus, the more we’ll understand it, and the easier it will be to design a vector that works as intended. Dyno’s team began with generating all single-codon mutations of the AAV2 cap gene, which is responsible for the virus’ capsid. This is not an easy task! The cap gene is composed of 735 codons (a codon is 3 nucleotides that, for all but three of them, codes for an amino acid), and there are 64 different possible codons. This didn’t just involve subbing out one codon for another, a massive project in and of itself, but also included codon insertions, deletions, and swapping out synonymous codons. Dyno then used high-throughput sequencing to map the fitness of each variant as compared to the AAV2 capsid that exists in nature.
It's not difficult to see why this systematic approach to data generation is helpful for understanding AAVs in a way that the random mutagenesis approach is not. The improved comprehension that results from this is no doubt useful for those using the rational design method, but it remains a method that’s time/thought intensive. AI comes in especially handy here, as the Dyno team and others explained through their efforts to create AAV vectors that more effectively target the liver:
“We hypothesized that an additive model built from our data would approximate the fitness of nearby variants with multiple mutations, enabling the design of functional variants with greater throughput than rational design and higher efficiency than random mutagenesis…..using this strategy, we designed 1271 variants in addition to 10,047 randomly generated mutants with 1 to 10 mutations from the WT reference….the designed set contained a much higher fraction of mutants targeted to the liver. The trend was most pronounced when the number of mutations was four or more: 147 designed mutants (25.6% of those tested) were functional, where nearly all of the 4477 randomly generated mutants were not viable or had weaker liver tropism than WT”10
More recently, Dyno has further leaned into leveraging compute capabilities, announcing a deal with Nvidia. It seems reasonable to assume that if designing better AAV vectors is a question of better data generation and compute capabilities, the company’s put itself in a good position to win.
The company’s business model is what you would expect for a biotech platform: upfront payments combined with potential milestone/royalty payments. While these partnerships are often bespoke, Dyno also offers two ‘off the shelf’ products: eCap1, which targets retinal cells, and bCap1, which targets neurons in the CNS.11 Currently, these partnerships give the company a lot of freedom: partnering with Roche on AAV vectors that target the CNS/liver doesn’t mean the company can’t work with other customers on the same problem. I think there’s a question as to how long these flexible agreements will last. We’ve seen Ginkgo have to change its customers contracts as of late, and the reasoning is simple to understand: therapeutics are the lifeblood of most biotech companies, and these companies don’t like the idea of a partner they’re collaborating with sharing any insights with a potential competitor. Presumably what Dyno’s learned from its collaboration with Roche will be at least somewhat applied to its work with other customers targeting the same cells.
There’s a natural question that arises about why biopharma companies won’t design their own vectors internally. There are a few reasons to think it would make sense to use Dyno instead. The first reason, which was mentioned above and that Century of Bio has noted in the past, is that designing the therapeutic gene and designing the delivery vector are two separate problems within gene therapy. Learnings from designing one don’t feed into learnings about the other. Rather than diverting researcher attention to delivery, companies are better off focusing only on the therapeutic and letting Dyno figure out how to optimize for tropism, high transduction efficiency, a minimal immune response, etc. The second reason is that Dyno sits in a superior position to solve the problem. In theory, the team already has a data advantage, and one that will only compound as their research continues. Your ROIC is just likely to be better if you give the capital to Dyno instead of an internal team. Sarepta being a customer provides some evidence that this is true: the company already has an FDA approved AAV-based gene therapy, but management clearly thinks Dyno offers some value that Sarepta itself either doesn’t have or doesn’t want to devote its own resources to.
Reason (1) holds true, and reason (2) most likely holds true. Recent history, however, may give one pause in being overly confident that this means Dyno sits in a position to win. Notably, delivery mechanisms also pose a challenge in the world of mRNA vaccines. Designing the correct mRNA strand is one problem, successfully getting that mRNA strand into cells is quite another, and required decades of experimentation with lipid nanoparticles (LNPs). As with gene therapy, learnings from designing the therapeutic don’t feed into learnings on designing the right LNPs. In some cases, most notably that of Pfizer/BioNTech’s Covid vaccine, this has led to pharma companies outsourcing the LNP work to specialists. Interestingly, however, Moderna didn’t take this route, and after working with external partners eventually opted to formulate a LNP in-house. Perhaps even more interestingly, Moderna has continued to do LNP research going forward as the company develops additional therapeutics. I don’t have a good explanation for why Moderna decided to develop LNPs in-house other than management thought they could do a better job than a specialist like Acuitas.12 This is a poor attempt at an answer, but it illuminates a noteworthy distinction between biotech platforms and their API analogues. It’s difficult to envision a world where it makes sense for DoorDash to build out their own payments capabilities; this remains true even if Tony Xu thinks his team has an insight into payments that Stripe doesn’t have. Building a marginally better payments system than Uber Eats just won’t lead to DoorDash winning more market share. This is not the case with therapeutics companies and lipid nanoparticle design or AAV capsid design. Having a better AAV vector than a competitor can be the difference between having an FDA approved, patent-protected drug and not having one, or between getting to market first and getting to market second. If you think your internal research department might have an insight into AAV vector design that Dyno doesn’t have, it makes complete sense to ensure your team has the funds and bandwidth to test that idea. Designing gene therapy delivery mechanisms might be a very different problem to finding and testing a therapeutic gene, but it’s a problem that’s vital to a company’s success. None of this is to say Dyno is a poor venture investment, only that biotech platforms operate in quite a different environment, with a very different customer base, than their software equivalents.
Disclaimer: The information in this post is not intended to be and does not constitute investment or financial advice. You should not make any decision based on the information presented without conducting independent due diligence.
An in vivo gene therapy is one where the functional gene is administered directly to the patient. This is different from something like CAR-T therapy, an ex-vivo approach where white blood cells are taken out of the patient, modified in a lab, and then infused back into the patient.
That's not to say adenovirus vectors have been completely tossed aside. They’re very useful in instances when you actually want to induce an immune response, such as when making vaccines.
That’s not to say one can’t have pre-existing immunity to AAV vectors ahead of a gene therapy treatment, as the product page for Roctavian points out.
From the paper ‘Overcoming Immunological Challenges Limiting Capsid-Mediated Gene Therapy With Machine Learning.’
There’s more to improved AAV vector design that just improving the viral capsid, such as designing tissue-specific promoters to avoid protein expression occurring within the wrong cell types.
For more info on the challenges of AAV vector design, I’d recommend this article, which also has Dyno’s footprint on it.
While these products are ‘off the shelf’, ordering them still involves partnering with Dyno and negotiating the downstream value payments.
One clue on Moderna’s decision can be found in their Q12019 earnings call: “Obviously we published last year in a paper on our website, Tasset et al, some of the deficiencies with the legacy lipid nanoparticle formulations in terms of local tolerability in the vaccine context. That underpinned a lot of our efforts too -- several years ago to develop our own proprietary lipid nanoparticle delivery technologies that are actually already in programs like CMV and hMPV PIV.”