The death of a patient in a gene therapy clinical trial for inflammatory arthritis in July underscores that gene therapy remains as much art as science. As company officials from Targeted Genetics Corp., which sponsored the trial, and the FDA investigate what went wrong in the latest incident, recent papers continue to unravel the details of what happens to cells during gene therapy.

A series of papers in the Aug. 1, 2007, issue of the Journal of Clinical Investigation presents updates on three different gene therapy trials, including gene therapy's biggest train wreck and what may be gene therapy's biggest mystery: a trial in which three of 11 boys treated at a pediatric hospital in Paris for X-linked severe combined immunodeficiency developed leukemia, and a very similar trial at a children's hospital in London where none of the 10 treated patients has developed serious side effects to date. (See BioWorld Today, Jan. 16, 2003, and March 7, 2005.)

Both X-SCID trials used retroviral vectors, which integrate stably into the host genome. Scientists and clinicians hope that learning more about exactly where such retroviruses integrate may lead to a better understanding of the nuts and bolts of gene therapy - and the ability to predict and prevent adverse events such as the leukemia cases in the French study.

However, for now the mystery remains: "At present, there is no explanation for why adverse events occurred in one SCID-X1 trial and not the other," Frederic Bushman, professor of microbiology at the University of Pennsylvania Medical school, wrote in a commentary that accompanied the JCI papers.

Both papers did find that the viruses did not integrate into random sites; instead, they favored setting up house near the 5-prime end of active genes. However, while previous studies had shown that integration near oncogenes could be partially responsible for the leukemia cases, the integration site pattern did not differ between those boys in the trial who developed cancer and those who did not.

Bushman concluded: "There is not yet any support for the idea that information about integration site distribution contains any data predictive of adverse events."

Bushman also noted that the British trial is not out of the woods yet, for several reasons. "Less time has elapsed since cell infusion for the patients from the second trial, so differences are still hard to evaluate," he said. The boys in the Paris trial all did so in a tight time window roughly 30 months after treatment onset.

Bushman also noted that the numbers of patients are still very low - in fact, the London and Paris trials are not statistically significant at this point.

The X-SCID studies used retroviral vectors, which are different from the AAV vector that Targeted Genetics used. But AAV has had its own cancer scare in peer-reviewed literature recently. A paper in the July 27, 2007, issue of Science reported that mice that received gene therapy with an AAV vector were indeed cured of their disorder, but again at the price of high rates of cancer.

In the research described in Science, one-third of mice with a rare lysosomal storage disease developed liver cancer after being treated with AAV-delivered gene therapy. Healthy control mice treated with AAV gene therapy also developed liver cancer at a higher rate than controls.

The exact cancer risk associated with gene therapy is the source of vigorous debate, with some previous preclinical papers receiving a skeptical welcome from the clinical community. (See BioWorld Today, May 22, 2006.)

But the Science paper contained another surprise: Like in the X-SCID trials, the integration of the viral vector was decidedly not into random sites of the genome.

"From our perspective, that was one of the most surprising findings of the paper," senior author Mark Sands told BioWorld Today. Sands said that the cells in the X-SCID trials have a well-documented growth advantage over the patient's other cells if they produce the therapeutic protein, and so the combination of that growth advantage with even random integration in or near an oncogene would end up producing the appearance of clustering.

But cells producing beta-glucuronidase, which is the gene Sands delivered in his experiments, have no such growth advantage. To see the integration sites congregate within a 6,000-base-pair stretch of the 3-billion-base pair mouse genome "is very unusual."