Navigating the Meandering Path of Oncology Drug Development 

The path from scientific discovery to clinical practice, though full of promise, is often laden with hurdles. With its biological challenges, such as treatment resistance and toxicities, and logistical ones, such as limited funding and inadequate technologies, oncology drug discovery and development have perplexed many scientists hoping to translate their best and brightest ideas to practice-changing therapeutics. This has made it all the more important to learn from those who have navigated this arduous journey. 

“If I’ve learned one thing as an academic entrepreneur, it’s that drug discovery and drug development are not linear,” said AACR President-Elect Keith T. Flaherty, MD, FAACR, during the “Bringing Science to the Clinic: Overcoming Hurdles for Therapeutic Translation” session at the AACR Annual Meeting 2025. 

Flaherty, of Mass General Cancer Center and Harvard Medical School, and Timothy A. Yap, MBBS, PhD, of The University of Texas MD Anderson Cancer Center, organized the session to address the multitude of obstacles that make the path to therapeutic translation so meandering. 

Together with Flaherty and Yap, experts from the research and investment realms offered suggestions for how these challenges could be overcome, emphasizing the importance of predictive biomarkers and discussing how to improve tumor sampling and modeling, learn from negative results, and qualify for investment funding.  

Using Biomarkers to Select Patients and Predict Outcomes 

Nearly all of the session’s presenters extolled the importance of incorporating predictive biomarkers into clinical research, noting that many oncology drugs fail to get past development because they were tested in the wrong patient population. 

“It’s not only about the drug; it’s about selecting the right patient,” said Kurt Schalper, MD, PhD, of Yale School of Medicine. He presented examples of how some drugs—gefitinib, for one—cause opposite effects in different patients, shrinking the tumor in some but promoting its growth in others. Despite this phenomenon, Schalper said that many researchers hesitate to use biomarkers to select patients for phase I clinical trials because of the longstanding principle that these early-stage trials should focus only on safety, tolerability, and dose finding.  

“That is a serious conceptual problem,” said Schalper.  

Beyond predicting treatment responses, biomarkers could also help predict acquired resistance, according to Flaherty. He suggested that patients whose tumors have biomarkers associated with resistance could be prioritized for alternative treatments, and their tumors could be studied to develop new treatment strategies for those tumors. 

“That’s the type of next-generation precision medicine that opens the door for us to do relatively rapid preclinical translation and then ultimately clinical development,” he said. 

Yap added that biomarkers can also inform the design of rational therapeutic combinations, including synthetic lethality approaches. He showed that ATM deficiency was a biomarker for susceptibility to synthetically lethal combinations, such as the combined inhibition of ATR and either PI3K or PARP or the combination of an ATR inhibitor and the chemotherapy drug irinotecan, and that microsatellite instability and mismatch repair deficiency made tumors susceptible to inhibition of the Werner helicase. 

“It’s always important as you move into the clinic, even at a very early stage, to think about better understanding predictive biomarkers of response,” said Yap. 

Improving Tumor Sampling and Modeling to Overcome Treatment Resistance 

But even when tumors respond to an experimental drug, acquired treatment resistance remains a major roadblock in the path from the lab to the clinic. Overcoming resistance will require a deeper understanding of how tumors evolve in response to treatment, but limitations of current tumor sampling and modeling methods hinder this research, noted Samra Turajlic, MBBS, PhD, of The Francis Crick Institute. 

“Sampling technologies have failed to keep pace with advances in molecular testing,” she said, explaining that available approaches do not typically capture specimens that represent the molecular heterogeneity across the tumor and do not provide critical spatial information.  

Turajlic presented data from the VAULT clinical trial that uses a different approach—examining homogenized tumor tissue to better represent the entire tumor. She showed that this approach allowed the researchers to identify novel actionable tumor drivers and to observe patterns of clonal selection that could contribute to treatment resistance.  

“This kind of representative sample can be very informative when it comes to either the choice of initial therapy or combination therapies,” she noted. 

Another strategy to understand and overcome resistance is to better model how combination therapies will perform in the clinic using artificial intelligence-driven digital twins and virtual clinical trials. Turajlic explained that these tools allow researchers to incorporate patient-specific factors, such as immune fitness and the microbiome, to better contextualize each patient and predict drug efficacy and resistance.  

Learning From Negative Clinical Trial Results 

While meeting a study’s endpoint is always the goal, Schalper argued that there is value in analyzing negative clinical trial results and shared one example of how doing so has helped inform future clinical research. 

Based on preclinical findings indicating that MGMT-silenced tumors were sensitive to temozolomide, Schalper and colleagues had conducted a clinical trial to test the therapeutic in patients whose colorectal cancers had high methylation of the MGMT promoter, which is often used as a surrogate for MGMT silencing. Despite the promising preclinical data, temozolomide treatment did not lead to responses in any of the nine treated patients. 

Digging deeper into samples collected from the negative trial, however, revealed something important—increased methylation of the MGMT promoter was not a reliable marker of MGMT status. As a result, many of the patients selected for the trial did not actually have MGMT-silenced tumors and, therefore, should not have been eligible. With this knowledge, the researchers plan to try again, this time selecting patients by measuring MGMT protein level instead. Studying the samples from the negative trial also provided insights into other factors that could influence treatment outcomes, findings that could help the researchers further refine the trial population. 

“I hope I’ve convinced you that early oncology clinical trials represent an amazing opportunity to … expand biological and clinical understanding,” Schalper said. “We should study positive and negative studies, and … we should make every sample count.” 

Offering A Clear-eyed Value Proposition: What Do Oncology Drug Development Investors Look For? 

Alexis Borisy, of Curie.Bio, discussed a different kind of challenge facing clinical translation—procuring investment funding. 

“So you’re doing your work, and you have a new therapeutic insight, a target that no one has appreciated before,” he said. “How do you go from the science to the bedside?” 

Borisy explained that the drug development stage alone requires between $5 million and $50 million, and the next stage of clinical testing needs even more—about $20 million to $200 million.  

“Because of that amount of capital, you often need to have private investment or a pharmaceutical company to underwrite that,” he said. But with only one-third of approved drugs ever making back the investment that went into their development and testing, how do investors decide which investigational agents are worth backing?  

When faced with this decision, Borisy, an investor himself, shared that he categorizes proposed targets into one of five tiers: 

• Tier 1: universally acknowledged to be a compelling and “hot” target that has not been successfully drugged yet 
• Tier 2: strong data to support the target, but the target and supporting data are still new and are not yet widely known or accepted 
• Tier 3: scientifically interesting but cannot be adequately evaluated in preclinical models 
• Tier 4: not a compelling enough target to pursue on its own but might be worth evaluating as part of a platform trial 
• Tier 5: not worth pursuing 

“[The decision] is going to depend on what tier it is and how much money it’s going to require,” said Borisy. Tier 1 targets would easily receive investment, while those in Tier 5 would be a hard pass, he noted. For Tier 2 targets, it would be the researcher’s responsibility to clearly communicate the data and persuade investors that the target could become a Tier 1 target.  

Tier 3 targets that require the upper end of the investment range, several hundred patients, and long-term endpoints would be less likely to procure funding than Tier 3 targets that require less investment and are simpler to test, particularly if they also address an important unmet need, he added. 

“It’s looking at that perspective, of being honest about what is the tiering of the project, what is the capital required on the project, and how do those two things fit together to try to take the wonderful ideas through this very tough gauntlet,” Borisy concluded. “Although we know so much, we have to be humble against this insidious disease [as we] try to go from science to the bedside.”