Brain tumors called glioblastomas are among the most vicious of cancers. So when Harvard University pediatrician Evan Snyder's best friend, Jim Galambos, developed a glioblastoma in 1996, Snyder set out to attack the disease, using an unusual tool: neural stem cells, special cells capable of developing into new neurons and other brain structures.

Based at Children's Hospital, Boston, Snyder had already used these cells to treat such inherited diseases as Tay-Sachs in mice. Stem cells tend to migrate throughout the brain, settling in damaged areas and initiating repair. Maybe, Snyder mused, they could find the areas damaged by glioblastomas, which also spread widely, and deliver drugs to the cancer cells. Galambos's illness was a strong motivator. "I promised the kids and...his wife that I would dedicate my efforts to helping him," says Snyder.

Galambos didn't make it, but Snyder's work has since borne fruit. In recently reported mouse experiments, Snyder showed that genetically altered neural stem cells could hunt down brain tumors, delivering a protein that activates an anticancer drug--and dramatically shrinking the tumors. The experiments "succeeded beyond my wildest dreams," says Snyder. Glioblastomas have historically been impossible to eradicate because they diffuse wildly in the brain.

But the therapy still faces serious obstacles. In humans, the stem cells themselves could theoretically generate harmful masses; conversely, they might fail to reproduce and thus allow the tumor to evade treatment. The immune system might attack the stem cells as invaders. Snyder says there's no evidence of any of these problems in mice, but mice are notoriously poor predictors of results in humans.