When drug discovery fails: scientists share their frustrations with the process

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript. For scientists who are involved in drug discovery, failure is part of the job. Less than 15% of drug candidates make it to market from the laboratory. Most drugs don’t get through clinical trials because they aren’t effective enough or have unmanageable side effects. Each step in drug development typically takes one to three years and has a high risk of failure1. The probability of moving on from phase I trials, which are often the first test of a substance’s effect on humans, is about 64%. Phase II tests safety, dose and efficacy; just over half of candidates don’t make it past this stage. Phase III, the largest of the three, aims to obtain enough data for drugs to be approved by review boards such as the European Medicines Agency and the US Food and Drug Administration. More than 40% of candidates that reach this stage come to the end of the road here. These failure rates also don’t take preclinical work into account. Before a drug makes it into trials, scientists can spend a good five years identifying and validating its mechanism of action, as well as testing its toxicity, activity and solubility. Most therapeutics don’t clear those hurdles, and even those that make it all the way through the clinical-trial process can still fail because they are too expensive or toxic for real-world use. Three scientists who have faced obstacles during the various clinical-trial phases reflect on the implications of failure and how they have moved forwards. Failing phase I As someone who works on a type of motor neuron disease called amyotrophic lateral sclerosis (ALS), neurologist Jeffrey Rothstein has seen his share of missteps. He chairs the scientific advisory board of the Network of Excellence for ALS, an international group comprised of 160 medical institutions, and provides advice to companies that are developing therapeutics for neurodegenerative diseases. He has therefore come across firms that don’t have the money to do sufficient research ahead of phase I trials, and others that don’t know whether their drugs will reach their intended targets in the brain. In 2021, when Rothstein became principal investigator of a trial for a gene therapy called BIIB078, he felt prepared. The therapy was created by biotechnology company Ionis Pharmaceuticals and licensed to another biotech firm, Biogen, both based in Carlsbad, California, for clinical trials. Several genetic and environmental risk factors have been linked to ALS, and BIIB078, a short, synthetic, single-stranded nucleic acid, was designed to target one such factor, a mutation in a gene called C9orf72. “We had a really complete data package,” Rothstein says. “It was safe in mice; it was not toxic to human brain cells. It did what it’s supposed to do and everything about it was working.” To his surprise, BIIB078 failed in a phase I clinical trial. It increased blood levels of a protein associated with neurodegeneration compared with placebo, and there was no improvement in clinical outcomes such as muscle strength or respiratory and physical function2. “This was 180 degrees opposite of all the preclinical research we had done, and we had not predicted this at all,” Rothstein says. After the trial’s failure was published in 2024, Rothstein was questioned by colleagues about how the drug could fail when all the science was right. He had to tell them that he didn’t know. His only hypothesis was that they had targeted RNA transcripts generated from only one of the two strands of DNA containing the mutation, which had worked well in mice, and “everyone thought that was the right strand to turn off”, he says. Rothstein has learnt to accept failure and not allow his ego to run the show, a sentiment he imparts to his PhD students at Johns Hopkins University in Baltimore, Maryland. “That’s part of what I teach,” he says. “If you’re studying a mechanism of disease, there’s a high likelihood that ten years later, you’ll be wrong about what you think was causing the disease.” Although BIIB078 didn’t make it over the clinical-trial hump, Rothstein is clear-eyed about the drug-discovery process. When he talks to participants at the start of a clinical trial, he tells them there are three possible outcomes. “What you and I want is that it’s going to work,” he tells them. “Second, which is the most common among all drug therapies, is it just fails to do anything. It’s the third option that most participants don’t embrace. They don’t realize that it can actually make things worse.” The failure of BIIB078 hasn’t deterred Rothstein from moving forwards with his research. His team is currently working on suppressing RNA transcripts from the other strand of DNA in the mutation. He wants to know if it has toxic effects and, if so, whether it was somehow activated when the number of RNA transcripts from the original strand was reduced. “We don’t know yet,” he said. “But we do know a lot more now because we ran that trial and collected data.” Failing after phase II Yelena Janjigian, a medical oncologist at Memorial Sloan Kettering Cancer Center in New York City, experienced failure between phase II and phase III trials. A specialist in gastrointestinal conditions, she was working on ‘anti-PD-1’ drugs, a type of immunotherapy used to prevent tumour progression and lengthen survival, often in combination with other therapies. A 2024 study3 found that after 3 years of follow-up, the median survival time for patients who had received the anti-PD-1 therapy nivolumab together with chemotherapy was 14 months, compared with 11 months for those receiving chemotherapy alone. It also found that twice as many participants who received the combination were alive as were those who got just chemotherapy. At the time, Janjigian was developing a therapeutic to be used alongside anti-PD-1 drugs to increase their efficacy. “We have that sweet spot where anti-PD-1 therapy works, but not well enough to build on it,” she says. She was focusing on ways to block the TIGIT receptor, a protein on the surface of some white blood cells that is activated by cancer cells and leads to a decreased immune response. She was therefore pleased to be approached by drug developers at Arcus Biosciences, a biopharmaceutical company based in Hayward, California, who wanted to work on a combination trial. She was asked to conduct phase II testing of domvanalimab, an anti-TIGIT therapy, with zimberelimab, an anti-PD-1 therapy, in 41 people with gastro-oesophageal cancer who were receiving chemotherapy. While she was working on her trials, Janjigian heard that other anti-TIGIT combination therapies had been unsuccessful. Tiragolumab, for instance, failed to improve outcomes in SKYSCRAPER-06, a phase III trial of late-stage small-cell lung cancer4. But Janjigian and her colleagues remained optimistic. Unlike tiragolumab, domvanalimab showed “encouraging efficacy”, boosting the median survival time to 26.7 months (patients typically live just over one year from diagnosis with chemotherapy and anti-PD-1 therapy alone)5. “Even when the negative data were coming out from SKYSCRAPER, we were still holding out hope,” Janjigian says. But the phase III trial was halted when early results showed that the combination failed to improve survival rates over the usual treatment of an anti-PD-1 drug with chemotherapy. Janjigian says that the experience had her “reflecting on the resilience of science”, not least because the data from the phase II trial were clean and encouraging. “It makes us realize that we really need to figure out better ways to streamline clinical development,” she says. According to a 2025 article published in Med6, the anti-TIGIT therapies were unsuccessful because researchers don’t have a complete understanding of the drugs’ mechanism of action, making it difficult to determine which patients will respond well. Domvanalimab is still being studied, however, now in people with advanced non-small-cell lung cancer, for which it has shown some promise in early-stage trials. Phase II trials can cost millions of dollars and have the lowest success rate of all the stages of drug development, partly because of their design. They typically have a small number of participants and it’s challenging to gauge efficacy and long-term side effects when the trials only last for a couple of years7. In the case of domvanalimab, Janjigian doesn’t know why the phase II study was so positive, only to be followed by negative results in phase III. “It’s a reminder that we have to be more strategic about the design of trials and bringing more diverse assets.” She suggests that larger phase III studies or skipping phase II could be a solution — but these would expose more participants to a drug before it’s been found out to be safe and effective. Janjigian is focusing on drugs for early-stage tumours as well as for peri-operative (the time before, during and after surgery) use. “One little bump is not going to deter us,” she says. Failing after phase III For any scientist, perhaps the biggest affront is to spend so much time creating a lifesaving drug and moving it through the arduous approval process, only to see it fail after passing all trials. This was the case with alipogene tiparvovec (Glybera), a drug to treat a rare, sometimes fatal, genetic condition known as familial lipoprotein lipase deficiency (LPLD), which causes chronic high cholesterol and severe pancreatitis. Michael Hayden, a physician at the University of British Columbia in Vancouver, Canada, first came across LPLD in 1986 when a patient at the hospital he worked at developed pancreatitis during pregnancy. The woman lost her child and nearly died. At the time, the only treatments were lipid-lowering drugs and severe dietary restriction, particularly of fats and carbohydrates. These weren’t effective and left people sick or dying from pancreatitis. Hayden and his colleagues began studying the disease, the incidence of which was unusually high in a small part of the Canadian province of Quebec. They quickly found mutations, called G188E and P207L, that caused dysfunction in the lipoprotein lipase gene LPL. The team began looking for ways to fix the faulty gene, starting in cats. By using a viral vector to deliver a functional copy of the LPL gene to target cells, the researchers were able eliminate symptoms. They began phase I trials in humans in the Netherlands and Quebec in 2005. Enjoying our latest content? Log in or create an account to continue Access the most recent journalism from Nature's award-winning team Explore the latest features & opinion covering groundbreaking research