Harvard geneticist George Church: targeting one aging pathway produces an effect of “maybe 2 to 20%,” so his lab is assembling gene combinations and aims to test up to a million combinations in a single animal

Harvard geneticist George Church: targeting one aging pathway produces an effect of “maybe 2 to 20%,” so his lab is assembling gene combinations and aims to test up to a million combinations in a single animal

On October 2 at the ARDD conference on aging, Church discussed gene combinations, viral capsids designed through machine learning to deliver them, and pig organ transplantation. The recording of the talk was released on October 19.

Longevity, Church says, requires “getting everything right.” He describes searching one gene at a time as wishful thinking: “We committed this crime too.”

In a 2019 paper, his lab tested genes encoding three proteins secreted into the bloodstream, individually and in combinations, in mouse models of obesity, diabetes, kidney injury, and heart failure. The researchers chose these proteins for delivery reasons: AAV8, the virus carrying the genes, efficiently infects the liver, which produces most blood proteins. The authors reported an “unexpected negative interaction” between two genes: they worked less well together than individually. The authors excluded genes encoding intracellular proteins because of the difficulty of delivering them together; they suggested that better viral capsids could make combinations of those genes feasible too.

OSK belongs to this group: it consists of three of the four Yamanaka factors, whose full set converts an adult cell into a stem cell. Noah Davidsohn, an author of that paper, tested OSK at Rejuvenate Bio in male mice, starting at week 124, when control mice had a median remaining lifespan of 8,86 weeks. In the 2024 paper, mice given OSK had a median remaining lifespan of 18,5 weeks, and their median total lifespan was 142,5 weeks versus 133 in controls, an increase of about 7%. In Church’s own lab, the SRSF1 gene increased median lifespan in old mice by 14%.

Church sees a strong combination in machine learning and simultaneous testing of up to a million viral capsid variants: “AI will not predict the perfect drug just by looking at a large body of literature.” He explains that the AAV capsid recognizes target cells through a key region, and random substitutions in that region usually disrupt its function. In a 2021 study involving Church, a neural network designed more than 200 thousand variants of this region, and more than half retained the ability to package DNA. According to Church, Dyno Therapeutics, where he is a scientific cofounder, administers millions of these variants to a single primate and identifies which cells each variant reaches. The Dyno-ahq capsid delivers the virus to primate neurons 280 times more efficiently than AAV9, according to company data.

Church considers the three OSK genes one of the largest payloads that the small AAV can carry, so the lab is testing herpesvirus vectors to deliver dozens of genes. For gene combinations, he describes the following plan: collect all studies in which a gene extended lifespan or produced another promising effect, devise combinations with a biological rationale, and test “as many as we can, hopefully a million per animal.” The results will train models for the next round.

Organ transplantation also requires combinations of edits. The eGenesis donor pig carries 69 genome edits: three sugar antigens that trigger rejection have been eliminated, seven human genes have been added, and porcine retroviruses that had raised FDA concerns have been inactivated. In monkey experiments, transplanted kidneys with only the antigens eliminated had poor survival; those with the human genes survived longer. According to eGenesis data released on September 3, five patients had received kidneys from such donors: one had gone more than nine months without dialysis, and two had subsequently received human kidneys. The FDA authorized a trial in 33 patients.