Forgotten, but not Gone

Why do drug combinations work? Many targeted therapies are combined based on molecular reasoning or evidence of additive or synergistic effects in cell line and animal models – and many clinical trials based on such reasoning have been successful. But what if that isn’t the full story?

While treating cancer cells grown in the laboratory with various anti-cancer drugs, Adam Palmer and Peter Sorger (both researchers at Harvard Medical School) observed that some cancer cell lines were more sensitive to drug A than drug B but, conversely, other cell lines of the same type of cancer were more sensitive to drug B. It occurred to the researchers that the variability in single drug response could partly explain why a population of patients may respond better when treated with two different drugs rather than one – a kind of “bet hedging,” where introducing a second drug boosts the likelihood that a patient will benefit from at least one.

They later found this to be a 50-year old hypothesis, called “independent action,” which had been inexplicably forgotten, and not tested against contemporary clinical trial data. They went and carried out those tests, with some surprising results (1). We spoke with the pair to find out more.

What is “independent action”?

From 1956, the Acute Leukemia Group B (which included many giants of oncology, including Emil Frei III, Emil Freireich, and James Holland) tested combinations of anti-leukemic drugs. They observed in patients how variable cancers were in terms of drug sensitivity. In trials of sequential treatments, an individual patient’s response to one chemotherapy had little relation to their subsequent response to a different chemotherapy, which justified the “independent action” model in which a combination of two drugs will induce a remission if either one of those drugs is able to induce remission by itself (without invoking any synergistic drug interaction). Emil Frei III’s independent action model accurately predicted remission rates in acute leukemia. We suspect that the theory fell out of use because improved methods for survival analysis came to dominate the interpretation of clinical trials in oncology.

What did your study involve?

We set out to distinguish between drug interaction and independence by: i) re-analyzing human clinical trial data in which single and combination therapies are compared, ii) mining a database of drug responses for patient-derived tumor xenografts, and iii) using a computational model of drug responses in a heterogeneous population of tumors.

We found that the independent drug action model (adapted to survival data and accounting for drug cross-resistance) was a sufficient explanation for the entire survival benefit of most of the clinical trials that we analyzed. It was a big surprise – we expected independent action to explain a fraction of the benefits of the combination therapies, but not this much. Combinations of cancer therapies with compelling molecular justifications, and strong evidence of synergy in pre-clinical studies, were not displaying synergy or even additivity in individual human tumors. Conversely, a fraction of combinations were identified as truly synergistic, using this model as the benchmark for the identification of synergy in clinical trial data.

Why do some drugs exhibit synergy in preclinical studies, but not in clinical studies?

We hypothesize that pre-clinical synergy often fails to translate into a detectable clinical benefit because even if drug synergy occurs in human tumors, its effect is overwhelmed by patient-to-patient variability in drug response. For many combination cancer therapies, some patients will be resistant to drug A and some resistant to drug B – resistance to either single drug may exclude the possibility of synergistic interaction. Conversely, some patients may have a durable response to one or the other single drug, and synergy will be not evident in survival data that is within the duration of the most long-lasting single-drug responses.

How could this work affect trial design?

When selecting anti-cancer drugs to include in a combination, this research (and the Acute Leukemia Group’s historical data) suggests that synergistic interaction is unnecessary for benefit: it is sufficient that two drugs each have a good rate of single-agent activity, and critical that they have tolerable toxicity together and non-overlapping mechanisms of resistance.

This research cautions that when a clinical trial shows “drugs A plus B” to be superior to “drug A,” it is not necessarily evidence that the simultaneous combination “A plus B” is also superior to “A, followed by B when needed.” If the toxicity of a drug combination is readily tolerable then the upfront combination may be justified, but when a combination has challenging side-effects (perhaps requiring dose adjustment) there may be value in testing a sequential regime (perhaps without requiring dose adjustment) for non-inferior therapeutic benefit. Whether this is true for a given combination depends on many factors, including possible costs of waiting to see whether drug A was effective. This is likely be relevant to immunotherapies combined with other anti-cancer drugs.

Our research suggests that many cancer therapies are, today, commonly applied with inadequate stratification of patients or tumor subtypes. In the future, clinical trials on more finely stratified patient cohorts may contain less variability in drug response – and therefore might be better able to identify which tumor subtypes, if any, benefit from a clinically impactful synergy.