No Clear Signal of Cancer Risk With Gene Therapies

Theoretical concerns about the cancer-causing potential of a type of viral delivery system for gene therapy have yet to translate into safety issues in the clinical setting, according to evidence presented to an FDA advisory committee.

Almost all animal-model data on the oncogenic potential of adeno-associated virus (AAV) vectors have come from studies involving mice. No data exist in species outside of rodents, according to an FDA briefing document prepared for the Cellular, Tissue, and Gene Therapies Advisory Committee (CTGTAC). The relevance of that data to oncogenesis in humans sparked considerable discussion during the first of four CTGTAC sessions on AAV safety, spread over 2 days.

AAV-mediated oncogenesis in mice occurs via integration into the Rian locus, which is found only in rodents, the briefing document continued. However, similar genomic regions exist in humans and non-rodent species, and the impact of integration events at these sites remains unclear.

Studies in the neonatal mouse model have shown formation of hepatocellular carcinoma (HCC) after therapeutic administration of AAVs, said Charles P. Venditti, MD, PhD, of the National Human Genome Research Institute in Bethesda, Maryland. Examination of that data must take into account a number of caveats. The caveats begin with the recognition that the neonatal mouse is equivalent to a “very premature infant,” immediately raising questions about the relevance of the model to human diseases.

Early animal model studies were not specifically designed to address pathogenicity and HCC risk but instead the durability of therapy. Additionally, long-term complications, such as HCC, occur after at least a year of follow-up and more likely after 16 to 18 months. Studies that have shorter follow-up periods might not be relevant to oncogenesis.

“I think this is a very important consideration,” said Venditti, a temporary voting member of CTGTAC. “Someone does a study in neonatal mice and they sacrifice the animals after 6 months, 8 months, 9 months, and they conclude that they didn’t see any pathology. Well, they may just not have waited long enough to see whether there was a signal from that AAV that could be something to be concerned about.”

Issues related to dosing, mode of administration, and greater consistency in the design of studies evaluating vector integration all figure into discussions about the relevance of animal-model data.

The issue of relevance and translatability of animal data to human disease does not have a single answer, said panelist Christopher K. Breuer, MD, of Nationwide Children’s Hospital in Columbus, Ohio. Instead, the answers should be disease specific and tailored to the severity of the disease and currently available treatment options.

“I would like to point out that any animal study could potentially be irrelevant,” said Breuer. “You will only find that out after the fact, when you compare your preclinical and clinical data. So far, based on the data presented today, while there is strong animal evidence pointing to the possibility of formation of liver tumors … the signal in the clinic doesn’t seem to be very strong.”

The data presented and the panel’s response suggested “general agreement that animal studies are, in general, problematic,” said CTGTAC chair Lisa Butterfield, PhD, of the University of California San Francisco, during her recap of the discussion of animal models.

“Sometimes they show us what is possible, and other times, they are not sufficiently representative of the human situation,” she added.

Panelists and speakers noted a need for uniform approaches to assessment and integration analyses.

“There are very complex, multiple types of sequence analysis, multiples of interpretation of that sequence analysis, which can give us different levels of samples and different levels of confidence that we’re looking at all of the important sites of integration and learning everything there is to learn,” Butterfield continued. “Long reads is a new technology that will need to be assessed before we know what the best approaches are, and those may also be model dependent.”

Discussions about the appropriateness of specific animal models may be premature in the absence of standardization for the conduct of preclinical studies, said Taby Ahsan, PhD, of the University of Texas MD Anderson Cancer Center in Houston.

“We do need to start to drive toward standards; otherwise we’re really not going to be able to generate the database of information that we need to make informed decisions as we move forward, especially because some of these technologies have unintentional data bias based on the method. I think we are starting to get a sense of the scientific issues that are out there, but we need to start to drive towards some type of standardization and understanding that will help us with preclinical study design as we move forward,” she noted.