The National Institutes of Health has approved a gene-engineering experiment with one of the deadliest toxins known to man, prompting an uncomfortable debate among some scientists over the value of such experiments compared with their risks.

The study, as planned by John R. Murphy of the Harvard Medical School, would take the gene for diphtheria toxin and splice it into the genetic material of the common gut bacterium Escherichia coli.

Until recently, the planned experiment would have been prohibited by federal rules, unless special permission were received. The stringent rules stemmed from the hypothetical risk that such an experiment could result in a catastrophic new strain of bacteria that would be as common as E. coli and as lethal as diphtheria. If such a strain escaped from the laboratory, the result could be disastrous.

Scientists generally agree that such a possibility is in the realm of science fiction, but the Murphy experiment, approved under a recent relaxation of the rules, gave nervous quavers to some at NIH.

"It really upset me," said Dr. Mark R. Geier, an assistant professor at Johns Hopkins University and a guest physician at NIH, who called the experiment the riskiest one ever allowed by NIH. "If they don't reject this kind of experiment," Geier said, "what kind would they reject?"

The experiment had to be cleared with both the NIH committee on recombinant DNA and the NIH Biosafety Committee, either of which could forbid it.

The biosafety panel first voted unanimously against the experiment. Later the group reversed itself and voted unanimously to approve it. The committee on recombinant DNA unanimously approved it the first time around.

Murphy's work will be carried out in a high-containment laboratory at Ft. Detrick, Md., and is intended to determine what risks there may be in such experiments, as well as answering two basic questions: whether E. coli will accept the toxin gene and be able to make toxin with it, and whether the toxin would be secreted once made.

Geier acknowledged that the risk of the experiment is extremely small, "one in a zillion or something," but he felt the potential benefit of Murphy's experiments was not great enough to justify it.

Members of the biosafety committee agreed that the risk is minuscule but they disagreed with Geier's assessment of the importance of Murphy's work.

Dr. Carl Merrill of the National Institute of Mental Health, a member of the committee, said the group agreed that the experiment should not be permitted for risk assessment alone.

"But once we realized how important the work was, what the benefit was, I don't think there were any more doubts about it," Merrill said.

The experiment would be the first in a series Murphy plans to do with diphtheria toxin, in a long-term attempt to create new tools of chemotherapy. The hope is that diphtheria toxin, which is a potent cell-killer, can be targeted specifically to attack unwanted cells in the body, such as those in tumors, while leaving others untouched.

In addition, it is possible that the experiments would lead to a new set of tools in a field that might be called "molecular anatomy," physical structure at the level of molecules.

For example, if the toxin were linked to a hormone whose sites of action in the body are unknown, the hormone-toxin pair would go to all those sites in an animal's body and kill the cells, thus providing a map of the hormone's action sites.

Doing something like that in the brain might disrupt only the behavior linked to the hormone or brain chemical being tested, thus giving a map of the chemical's action spots in the brain and their behavioral importance.

The biosafety committee will monitor the experiments, said John Irwin, executive secretary of the committee. If it turns out that the E. coli does make the toxin and kills the laboratory animals, Murphy's experiments will be halted.

Geier acknowledged that the danger he perceives in the experiment is not so much from the bacteria escaping the high-containment lab, but "what I worry about is someone stealing some of it."

"After all, even if the risk is small, this experiment is potentially dangerous, potentially deadly to every mammalian species on the planet," he said. "This means it is more dangerous than a nuclear weapon if the risk realizes the potential of what it could be."

The chances of that are slight. To begin with, the strain of E. coli used in genetic experiments is a crippled strain that cannot live long outside the laboratory. In addition, E. coli has no mechanism to deliver poison to the body. So if the experiment demonstrated that E. coli can make the diphtheria toxin, and that it can devise some mechanism to excrete it into the gut of animal it lives in, it would have to live in the gut for some time and produce enough toxin to cause illness or death in the animals. (In previous risk-assessment experiments, the same strain of E. coli could not survive 72 hours in the gut.)

The bacteria would then have to escape destruction by the biosafety committee and get out of the lab, by theft or some other means. It would then have to alter itself to survive outside the lab and compete with other organisms.

Diphtheria bacteria has been in use in laboratory experiments for decades.

Murphy said he has worked directly with diphtheria for some years, as others have for most of this century, under much less containment, with few serious incidents. If there were any danger, he believes, it would come from the diphtheria bacterium itself rather than a relatively harmless bacterium like the crippled lab strain of E. coli, which has no mechanisms to deliver poison to the body.