Kibur Science Spotlight: A discussion with Dr. Antonio Fojo, Professor of Medicine at Columbia University
We recently read the article, Journeys to Failure that Litter the Path to Developing New Cancer Therapeutics, an invited commentary in JAMA Network Open penned by Dr. Antonio Fojo. He comments on the paper by Jentzsch et al that discusses failed cancer drugs. Dr. Fojo, kindly agreed to an interview to discuss the issues of the current state of cancer pharmaceutical development.
How can we do a better job with pharmaceutical development from a biotech/ early industry research perspective?
Unfortunately, we are still doing an unsatisfactory job of developing effective cancer therapeutics. Early research on these therapeutics now largely occurs in small pharma and biotechnology companies, rather than large pharma as it did over a decade ago. This model enables large pharma to cherry pick winners that unfortunately often lack robust scientific support. The result is a loss of institutional knowledge, whereas if large pharma was developing these therapies in-house, more thoughtful decisions might be made. There are too many ‘me too’ drugs (reviewed by Dr. Fojo in 2014) and IRBs are approving trials where similar therapies have already failed. While the focus of IRBs is first on patient safety and ethics, greater focus is needed on review of the science. These ‘me too’ drug failures are costing industry billions of dollars annually and these expenditures are then in turn being transferred eventually to patients.
Would consideration of alternative clinical models be a solution? Are there better models that translate to clinical activity in humans rather than those currently in use?
There really are no effective animal models for immunotherapy, and this continues a tradition of pre-clinical models poorly emulating the situation in humans. While the industry continues to look for better models, unfortunately all are flawed although every researcher feels their model is the best. A glaring example was the development of mitotic spindle poisons that many thought would render existing microtubule targeting agents obsolete. However, tumors in preclinical models divide much faster than tumors in humans – nearly every 24 hours compared to every 30 to hundreds of days for tumors in humans. Thus, “classical” mitosis inhibitors such as vincristine and vinblastine, as well as all mitotic spindle poisons that were developed, cause mitotic arrest in murine models quickly, and thus don’t accurately mimic the mitotic arrest in human tumors. ]Sincethey divide so much slower, they are not killed due to inhibition of mitosis, but rather due to impaired trafficking of critical proteins on microtubules, making predictions of success and extent of side effects challenging. Thus, it’s very difficult to obtain results in mice that appropriately mimic human outcomes.
AI is suggested as a potential solution for targeted drug development with plenty of associated skepticism. How do you think AI will influence new therapeutics in the near term? Ten years from now?
Cancer is extremely complicated, there is still so much unknown. Ten years ago, we all thought cancer would be nearly cured due to genomics – but of course that hasn’t happened. The problem is that while AI can be used to do analyses in a variety of different ways and provide some insights it will not provide us with information that will lead to cures. A more realistic expectation from the use of AI is that it will provide us incremental information, which will gradually lead us to better treatment outcomes. Keep in mind that the information AI will have has been extensively mined in incredibly diverse ways by exceptionally intelligent humans who understood what they were doing, recognized their own limitations and the limitations of their analyses and often considered innumerable variables. AI is built on the shoulders of these amazing scientists and personally I find it highly unlikely it will find us the path to cures. Highly unlikely that so much was missed. And this does not even consider all that we don’t know, and that AI cannot intuit. For example, it’s not possible to give AI an intracellular concentration of a drug nor can it begin to surmise it – there are innumerable unknown variables that we lack. Can AI integrate all the binding affinities of drugs for their targets?
We are interested in the prospect of microdosing Phase 0 clinical trials, with the thought that failures can be identified earlier in the drug development process. Do you think this is a rational approach to de-risk earlier? Would it save time and money?
The unwritten oncologist mantra is that if a little is good, more is likely to be better, and we’ve come to that conclusion for a reason – because it’s very often true. Pharmaceutical companies know this and consequently often want to go to higher concentrations, basically identify the maximum tolerated dose. Side effects may be difficult to identify with low concentrations. The Phase 0 approach can provide valuable information in a quick turnaround and provide details on varying concentrations. This can be very valuable information, but we must acknowledge tumors tend to have different kinetics, and there may be multiple targets of one drug in a cell, and the most important ones, or the more “downstream ones” may not be what you are measuring, something important to keep in mind when using this approach. However, it may help identify likely failures earlier and identifying failed strategies early can help go/no go decisions.
For additional information, please see the articles below.
https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2807715
https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2807710
https://pubmed.ncbi.nlm.nih.gov/25068501/