Functional Precision Medicine: Putting Drugs on Patient Cancer Cells and Seeing What Happens

Some 20 years ago, a long-forgotten, nonmedical acquaintance asked me what I did as a cancer researcher: “What do you do, like, put drugs on cancer cells and see if they work?” I suppressed a complacent chuckle at the absurd simplicity of this notion from someone so obviously ignorant of the splendid arcana of cancer biology. “No, no, it is much more complicated than that. Genes … signaling pathways … axes … mutations, very complex.”

And yet, there is something very obvious and compelling in the logic that if you want to know if a particular drug will kill a patient's cancer, put the drug on the patient's cancer cells and see what happens. Why don't we do that? There was a time, roughly from 1980 to 2000, when there was considerably more enthusiasm for this idea. However, the so-called ex vivo chemosensitivity assays investigated two decades ago never provided compelling evidence of improving patient care. For this reason, several influential articles discouraged their routine use in the clinic (1–3). The general response I got when I asked my colleagues about these assays was, “Yeah, people have tried it, and it doesn't work.”

Why was the utility of these assays limited? For one, they were being studied on very few drugs in an era before targeted therapies. For another, ex vivo culture conditions were relatively primitive. Overgrowth of normal cells was a problem, so that more normal cells than cancer cells would sometimes be studied. Technology to study single cells was primordial, so simple bulk assays dominated. Very often, these assays measured bulk metabolic properties that imperfectly correlated with cell death. Finally, prospective clinical trials demonstrating patient benefit were lacking.

A couple of important events drove attention further from ex vivo assays. In a landmark event for cancer biology in 2001, a small-molecule kinase inhibitor, imatinib, was shown to dramatically improve the outcome of patients with chronic myeloid leukemia (CML) by targeting a genetic alteration nearly universal in this disease (4). Also in 2001, the first draft of the human genome was completed, ushering in an era of ever-improving genomic sequencing techniques, bringing genomic analysis of any tumor into practical reality (5). Cancer centers and commercial entities began touting their ability to use these next-generation genomic technologies to match patients to the drugs that would treat their individual tumors. The pathway to precision medicine, identifying the right drugs for the right patient with cancer, seemed clear. Sequence everyone's tumor, identify the genetic abnormalities, and match them with drugs targeting those abnormalities just as in CML.

This strategy has certainly produced some true clinical successes, including targeting EGFR mutations in non–small cell lung cancer, BRAF in melanoma, and TRK mutations in a variety of tumors. However, clinical reality has forced a recalibration in expectations of a purely genomic-based precision oncology strategy. For several reasons, including the lack of targetable mutations and the lack of useful drugs for certain promising targets, most patients with cancer do not benefit from genomic precision medicine. These limitations have contributed to a reexamination of alternative precision medicine strategies.

The past 20 years have seen many technological advances that enhance ex vivo study of patient tumor samples. For one, we have much better and more varied culture systems for the study of patient samples ex vivo in both two-dimensional and three-dimensional formats. We have many more and better assays to measure changes induced by drug perturbations. We have the ability to study single cells in detail, which was impossible two decades ago. We have vastly improved bioinformatic capabilities that allow the combination of complex drug response data with any number of clinical and molecular annotations. Finally, there are simply vastly more cancer drugs and tool compounds to test, greatly increasing the chances of finding at least one active drug.

Important work from Kurtz and colleagues (6) and Tyner and colleagues (7) demonstrated the feasibility of functional precision medicine (FPM) strategies in hematologic malignancies and lent confidence to other related efforts (6, 7). Many studies followed showing the feasibility of such platforms or describing anecdotes of clinical benefit. However, there is a lingering need for prospective clinical trials to estimate the clinical benefit of FPM. Two articles in this issue of Cancer Discovery begin to address this need (8, 9).

Kornauth and colleagues (8) report findings of the EXALT trial, a study deploying their single-cell FPM (scFPM) assay. They used high-content microscopy and automated image analysis to evaluate the effects of 139 drugs on samples from 143 patients with hematologic malignancies, including acute myelogenous leukemia (AML) and B- and T-cell non-Hodgkin lymphoma. These were patients without standard-of-care options remaining, who had a median age of 64 years, and who had a median of three prior therapies. Seventy-six patients were evaluable, of whom 56 were treated according to scFPM results as determined by an expert tumor board. Because this was essentially a collection of N = 1 trials, to evaluate the quality of the therapy assigned by scFPM, each patient's progression-free survival (PFS) was compared with PFS from their prior therapy. Normally, the observation in oncology is that subsequent therapies generally produce shorter and shorter PFS. In this trial, however, 30 of 56 (54%) demonstrated a PFS of at least 1.3 times the duration of that from prior therapy, with a median PFS ratio of 3.4. Perhaps even more impressive was the 21% rate of exceptional response, defined, as by others previously, as a PFS triple the expected median response (10, 11). Some of these were patients who were previously ineligible for stem cell transplant but who subsequently received one after achieving a complete remission. The 20 evaluable patients who were treated according to physician choice rather than scFPM, although not a prespecified control arm, served as a comparator. Compared with physician choice, scFPM-treated patients showed both a PFS and an overall survival benefit.

Malani and colleagues (9) focused on AML samples. Their simpler drug sensitivity and resistance testing (DSRT) assay relied on bulk ATP measurements of mononuclear cell samples from patients with AML that contained blasts but also nonmalignant cells. Although the tumor board had the option to consider genomic and transcriptomic data, these were usually not available by the time a treatment decision was made due to their longer turnaround time compared with DSRT, which took roughly 4 days to complete. They found an actionable drug for a remarkable 97% of patients via these methods. DSRT recommendations were implemented, however, in only 37 of the 186 patients tested. The main reason for this attrition is that most patients were treatment naive and had standard treatment options available. Of the 37 patients treated according to DSRT, all relapsed or refractory, 59% exhibited a clinical response, including 13 of 37 complete responses.

How does the clinical utility of these functional assays compare with that of widely used genomic precision medicine assays? Given the near-universal adoption of genomic precision medicine assays for a variety of cancers, this question is surprisingly hard to answer. Although publications on genomic precision medicine studies often report rates of logistical success or of identification of actionable mutations, studies of clinical response are remarkably difficult to find. Kornauth and colleagues (8) provide a helpful summary in their Table 3. Populations among all these studies vary widely, so comparison is necessarily inexact. However, in the genomic studies, response rates of patients receiving genomic testing were rarely above 10%, so that the functional assays here generally compare favorably. However, there is no need to force -omic and functional approaches into opposition. In the future, a rational use of both classes of assays is likely to be better than either alone.

A decade of clinical experience has made clear that finding the right drugs for individual patients, the central job of precision oncology, requires more than genomics. These studies show that, at least in hematologic malignancies, FPM provides a very credible adjunct to widely accepted genomic methods. To find a place in everyday clinical practice, FPM will benefit from more prospective trials evaluating patient clinical response across a wider variety of tumors. To this end, the EXALT 2 trial (NCT04470947), run by the same people who produced the EXALT trial, will randomize patients with advanced cancer to arms in which therapy is directed by functional means, genomic means, or physician choice.

Both studies demonstrate that “it doesn't work” is now too blithe a response to queries about FPM. That FPM does “work,” at least in the hematologic malignancy context, now seems clear. It now remains to be seen how rapidly and to what extent this exciting approach, simultaneously novel and old-fashioned, takes its place in standard clinical care of the patient with cancer.

A. Letai reports other support from Zentalis, Dialectic, Flash Therapeutics, and Anji Onco outside the submitted work, as well as several patents for BH3 profiling owned by his employer, Dana-Farber Cancer Institute, for which he receives royalties paid by Zentalis and Dialectic.

A. Letai gratefully acknowledges support from NCI grant R35CA242427.