The resurgence of the gene therapy field has multiple therapies primed to make the transition from bench to bedside. But, experts routinely site delivery as the biggest hurdle standing in the way of getting therapeutic gene cargo to patients.
The current workhorse of choice for delivering to target tissues are the adeno-associated viruses (AAVs). But, today’s natural AAVs need improvements as they do not specifically target diseased cells and tissues, carry small cargo loads, and can be recognized by the immune system.
Scientists at the Wyss Institute and Dyno Therapeutics report an approach to speed up the process of making enhanced AAV capsids. Their work is published in Science in an article titled “Comprehensive AAV capsid fitness landscape reveals a viral gene and enables machine-guided design.” In it, they describe the development of a complete first-order AAV2 capsid fitness landscape, characterizing all single-codon substitutions, insertions, and deletions across multiple functions relevant for in vivo delivery.
Source: Wyss Institute at Harvard University
“These high-throughput technologies paired with machine-guided design lay the foundation for engineering superior and highly tailored AAV variants for future gene therapies,” noted Eric Kelsic, PhD, CEO of Dyno Therapeutics—a company spun out of the lab of George Church, PhD, to apply these methods to enhancing AAV vectors. “Past approaches such as rational design or random mutagenesis each had their drawbacks, either being limited in the library size or being low in quality, respectively. Machine-guided design is a data-driven approach to protein engineering.” Here, Kelsic explained, “we show that even a simple mathematical model, powered by enough data, can successfully generate viable synthetic capsids. This iterative and empirical approach to protein engineering enables us to get the best of both worlds and generate large numbers of high-quality capsid variants.”
The team’s systematic approach to capsid protein-engineering included mutating each of the 735 amino acids within the AAV2 capsid one by one, including all possible codon substitutions, insertions, and deletions at each position. Using next-generation DNA-synthesis, barcoding, and DNA sequencing, they generated a virus library containing about 200,000 variants and identified capsid changes that both maintained AAV2’s viability and improved its tropism to specific organs in mice.
“With the information generated by this library, we were also able to design capsids with more mutations than previous natural or synthetic variants,” explained George Church, PhD, lead of the Wyss Institute’s Synthetic Biology platform and professor of genetics at Harvard Medical School. Furthermore, Church noted, they did it with efficiencies of generating viable capsids that far exceed those of AAV created by random mutagenesis approaches.
The method unexpectedly discovered a new accessory protein hidden within the capsid-encoding DNA sequence that binds to the membrane of target cells. A frameshifted gene in the VP1 region expressed a membrane-associated accessory protein that limits AAV production through competitive exclusion. “Unexpectedly, the high-resolution data we generated enabled us to spot a new protein encoded by a different reading frame within the capsid’s DNA sequence—which had escaped notice despite decades of intense research on the virus,” said Pierce Ogden, PhD, a postdoc in the Church lab. “Membrane-associated accessory protein (MAAP), as we named it, exists in all of the most popular AAV serotypes and we believe that it plays a role in the virus’ natural life cycle. Studying how MAAP functions will be an exciting area for future research and could potentially lead to a better understanding of how to better produce and engineer AAV gene therapies.”
“This reveals the promise of data-driven protein engineering,” noted Sam Sinai, PhD, machine learning scientist at Dyno Therapeutics, “in particular for proteins like the AAV capsid that are difficult to model with current computational approaches.” Sinai asserted that “our results are highly encouraging but also only a first step. Using this data and those from future experiments, we will be building machine learning models to optimize capsids and address a wide variety of gene therapy challenges.”
