Viral vector gene therapy is the clinically validated route to cures for diseases that have no treatment. A recent editorial from Nature Medicine put forward a strong argument that now is the time to push on gene therapies, to gain from the momentum of recent successes and adapt to mitigate the setbacks. Immunogenicity continues to be a challenge and the source of some notable setbacks in viral delivery. In the spirit of momentum building, this blog is laying out the current state of one aspect of viral vector immunogenicity, neutralising antibodies. This is a huge challenge to therapeutic success and one which I see as an open engineering problem to be solved.
Adeno associated virus (AAV) is the most widely used and successful vector for gene therapies; however, innate and adaptive immune responses limit efficacy, exclude large numbers of patients from trials and can be the source of serious adverse events. While there has been a lot of work done on immune responses to AAV already, with a big impressive uptick recently, we've only scratched the surface in our understanding.
Since most widely used AAVs are either wild-type, or engineered vectors based on a common wild-type, the background rate of people with high levels of neutralising antibodies (NAbs) to existing vectors in the population is very high. There are wildly different numbers reported depending on serotype, measurement method and population. In patient cohorts it's not unusual to have over 40% with high enough antibody rates to exclude them from trials both for safety and efficacy (with some populations and more common serotypes even higher). Seropositivity also tends to be much higher in adult populations. As highlighted in a recent blog, sex can have an impact on the efficacy of AAV mediated therapies, with antibody rates being a possible understudied reason but really we don’t know. This is just one in a long list of aspects we don’t have good answers for when it comes to the antibody problem.
In rare diseases exclusion rates are already a problem, but seen as a reasonable trade off for getting to the clinic. Moving into gene therapies for more common diseases and larger patient populations these are unworkable numbers.
Immunosuppression regimes are frequently used in clinical trials to mitigate immune responses. Immunosuppressive drugs, such as steroids, are given before, during and after treatment to obtain improved transgene expression. However, the success of these varies both initially and over time where tissue damage and transgene expression loss are a concern. There's no consensus on what a good immunosuppression protocol looks like. This approach has helped with the viability of some therapies, but is far from an ideal solution and not a small burden on the patients. The alternative is to reduce the response in the first place by changing the capsid.
Some wild type AAVs, for example AAV5 and some found via discovery in animal models, have lower background rates of patients with serum positivity above exclusion levels; however, lower is not none and vary greatly between populations. In a hemophilia cohort rates for AAV5 were reported at 25%, with cross-reactivity with other serotypes between other serotypes being an identified driver in NAbs. Choosing serotypes based on seronegative responses has a clear trade-off in transduction and specificity terms. Finding an existing virus with low rates of NAbs in patients can be a good mitigation, but it's not enough. Current mitigation strategies that have made it to the clinic fall into a category I class as ‘clinical trial practicality’ rather than a solution anyone is happy with.
A lot is still unknown about how differences in capsid affect antibody recognition and in general this part of the field is data poor, but that picture is changing. Relatively similar AAVs can have big differences in antibody recognition patterns. Changing highly exposed regions on the capsid surface has long been shown to alter neutralising antibody recognition, for example by looking at insertional mutants and chimaeric capsids. There have been previous successful attempts to use rounds of evolution in AAV1 to avoid neutralising antibodies and generate capsids which greatly drop their cross reactivity to antibodies for their parent capsid. Highly specific antibody recognition hotspots have been identified as clearly affecting cross-reactivity despite only having relatively small differences present.
More radical solutions to try and get complete antibody avoidance also exist. Obscuring the capsid surface from antibodies through methods such as exosome association and others can reduce antibody recognition but not completely - though it's unclear why this is the case. Aravind Asokan's lab’s utilisation of non-human AAV genomes as part of their capsids have yielded strong results in avoiding mammalian AAV targeting antibodies.
What would it take to engineer very specifically for antibody avoidance without trade offs? My answer is broad data on immune evasion, paired functional data and a platform capable of utilising it all. Does this data exist? Not yet, but it is coming.
For instance an invaluable set of antibodies from Zolgensma patients was recently published providing a much larger set of structures for AAV antibodies than was previously available publicly and importantly showing shifts in the capsid surface during binding. The overall picture here is improving but there are still a lot of open questions and big gaps in data.
To move faster and get to capsids that have low preexisting immunity without compromises on tissue tropism, production and function, we need a broader approach that treats this as an engineering problem.
This is what we've built at Lir - an AI-first lab-in-the-loop that's generating high volumes of data, including on immune evasion. All of this is being done systematically to develop a broad picture about what's possible in this problem space. We've already found novel motifs and target residues for antibody avoidance being further validated in our lab. We have the tools to both dramatically improve our understanding of how antibodies interact with AAVs and take the next steps to engineer capsids better hidden from the population's immune systems, while maintaining control over other aspects of capsid function.
Capsid engineering requires optimising for many problems at once, with no problem existing in a vacuum. A much wider picture of how AAV interacts with the immune system is necessary. In follow ups to this I'm going to be expanding on this further, while also getting into other aspects of AAV interacting with the immune system and how we can design against them.
If any of this sounds like a problem you'd like to solve with us please get in touch!
© 2025 Copyright: Lir Therapeutics