Faced with the failure of a key clinical trial, a California-based biotech is pressing ahead with plans to test its experimental Alzheimer’s drug in another study, convinced that signs of benefit observed in some patients are proof its treatment works.

Last month, the company, Cortexyme, reported the drug missed both main goals of a late-stage trial, showing no significant difference versus placebo on tests that measured participants’ cognition and physical function. Yet Cortexyme claimed the negative results had a silver lining, noting that, among a smaller subgroup, the drug appeared to slow cognitive decline by a substantial margin.

The new study, which is still being designed, would test whether that effect was a statistical mirage or representative of a real treatment benefit for mild-to-moderate Alzheimer’s patients. Cortexyme will most likely pick the lower of two doses Cortexyme used in the failed trial, as the higher dose was more frequently associated with side effects, particularly those indicating liver damage.

Cortexyme disclosed its plan’s preliminary details on Thursday, alongside a presentation of study results by the company’s chief medical officer at the Clinical Trials on Alzheimer’s Disease conference. The presentation offered the fullest accounting yet of the trial; last month’s disclosure came via a company press release.

The added detail showed that, overall, participants given either dose of Cortexyme’s drug declined at virtually the same rate as those on placebo when measured using a cognitive testing scale known as ADAS-Cog11. On another scale, designed to evaluate daily living activities, patients on Cortexyme’s drug appeared to decline slightly more than those on placebo over the 48-week treatment period.

“As it stands, the trial was negative as both primary endpoints failed to show a benefit of the drug,” said David Knopman, a neurologist at the Mayo Clinic who specializes in Alzheimer’s and dementia, in an email.

Last month and again on Thursday, Cortexyme focused instead on a subset of roughly one-third of the study’s 643 participants that had detectable DNA in their saliva of the bacteria P. gingivalis, which causes gum disease. Cortexyme’s hypothesis, which breaks with much of the Alzheimer’s field but is supported by intriguing research, is that infection with P. gingivalis can cause the disease in some cases.

While all participants in the study had blood antibodies that indicated prior exposure to the bacteria, Cortexyme also planned ahead of time to analyze results based on several different tests for P. gingivalis infection. Looking only at those with DNA in their saliva, treatment with Cortexyme’s drug slowed the rate of cognitive decline by 42% versus placebo at the low 40 milligram dose and by 57% at the higher 80 milligram dose.

Cortexyme has heavily emphasized these findings, touting them as validation of the role played by P. gingivalis infection and the potential for its drug to block its harmful effects.

“These results open a new door to our understanding of Alzheimer’s,” said Michael Detke, Cortexyme’s top doctor, on Thursday. Marwan Sabbagh, a professor of neurology at the Barrow Neurological Institute and the lead investigator for Cortexyme’s study, offered an even more optimistic take: “We are confident that these findings are a true positive,” he said in the same presentation.

But other Alzheimer’s experts aren’t convinced the findings offer more than a starting point for the trial Cortexyme now intends to run.

The results represent an “interesting new direction,” said Constantine Lyketsos, a professor of psychiatry and an Alzheimer’s specialist at Johns Hopkins University, in an email. But there are “still many challenges and uncertainties,” he added.

Knopman, of the Mayo Clinic, took a similar view: “This is an interesting result that unfortunately, like all claims from post hoc analyses, requires validation in a new cohort.” (Cortexyme said its analysis was pre-specified.)

“The data support following this study with a new one in which only patients with detectable P. gingivalis would be enrolled,” he added.

Cortexyme still needs to the meet with the FDA and other regulators to firm up its plans for a confirmatory trial. “Details of future study designs are to be determined, based on regulatory interactions,” a spokesperson said in an email, but noted that a study “will almost certain focus” on the lower 40 milligram dose and only include people with detectable P. gingivalis in saliva.

That the 40 milligram dose is Cortexyme’s focus is likely a reflection of the drug’s safety results, which raise some concerns.

Fifteen percent of participants given the higher 80 milligram dose experienced elevations in liver enzymes that were three times the upper limit of normal — a warning sign for liver toxicity. Two also had elevations in a liver compound known as bilirubin. Together, these spikes can sometimes suggest patients are at risk of drug-induced liver injury.

Cortexyme said “virtually all” study participants with liver enzyme elevations were asymptomatic and that an outside panel of experts indicated the issue could be addressed by adjusting the dose through a process known as titration.

Sizable numbers of patients given Cortexyme’s drug also dropped out of the study, however: roughly 40% in each drug arm, compared to 25% in the placebo arm. “Some” of these discontinuations were due to abnormal laboratory values, Detke said in his presentation.

Five patients on the higher dose and one patient on the lower dose died, although investigators determined none were related to Cortexyme’s drug. Two died of worsening Alzheimer’s disease, and one each of COVID-19, heart attack, sepsis and lung cancer.