Causes and costs in biological vs clinical resistance

This Wednesday, on These few lines, Rob Noble warned of the two different ways in which the term de novo resistance is used by biologists and clinicians. The biologist sees de novo resistance as new genetic resistance arising after treatment has started. The clinician sees de novo resistance as a tumour that is not responsive to treatment from the start. To make matters even more confusing, Hitesh Mistry points to a further interpretation among pharmocologists: they refer to the tumour remaining after a partial but incomplete response to treatment as de novo resistant. Clearly this is a mess!

But I think this is an informative mess. I don’t think it is a matter of people accidentally overloading the same word. Instead, I think it reflects a conceptual difference in how biologists and clinicians think about resistance. A difference that is a bit akin to the difference between reductive and effective theories. It is also a difference that I had to deal with during the revisions of our recent work on measuring the games played by treatment sensitive and treatment resistance non-small cell lung cancer (Kaznatcheev et al., 2018).

In our experiments, we observed that even in the absence of drug, resistant cells tend to have a higher growth rate than parental cells in the same environment (i.e. proportion of parental cells in the co-culture). A reductionist could rationalize our observations by saying that we actually selected for two different qualities in our resistant line: (i) a general growth advantage, and (ii) resistance to Alectinib. In fact, it is one of our reviewers that raised this concerned about if our resistant cell lines were resistant or something more:

I think you are glossing over a very important assumption: are we really looking at resistance to alectinib only? What you call “resistance” has clearly some other positive effect on proliferation. Without clarifying what these mutant cells actually do it seems a stretch to say that it is resistance to alectinib that leads to the effect you observe, and to stress the difference with current common wisdom. If you had a mutant differing from the parental cells only for the ability to resist alectinib, then you could make a clear case that we are just looking at resistance. But here, since you do not know what happened to these cells during selection in alectinib, I don’t believe you can state clearly that we are observing the effect of resistance only. You are comparing a parental line with a mutant line that happens to be resistant and that probably also has a lot of other mutations.

We originally replied to this in the main text, but due to length constraints it has now been moved to the appendix, section A.3.

The reductive view is a reasonable hypothesis, but it faces a few challenges. First, both parental and resistant cells were evolved for the same length of time, with escalating dosages of DMSO for the former and Alectinib for the latter. Thus, (i) cannot be due to just subculturing, but is somehow linked to drug. Second, there is no growth rate advantage of resistant cells in monoculture; the advantage is only revealed when parental and resistant cells are cultured with a common proportion of parental cells.

Finally, to even make the distinction between (i) and (ii), one has to implicitly assume that resistance has to be neutral or costly by definition. For an oncologist, however, both (i) and (ii) would constitute clinical resistance if they led to a tumour escaping therapeutic control. By using a definition of clinical resistance that is broad enough to capture both aspects, we observe resistance that is neither neutral nor costly in DMSO co-culture.

It is this final point that I want to focus on. How do we give reductive definitions of resistance that don’t define away the possibility of non-costly (or even negative cost; i.e. benefit in DMSO) resistance? This might or might not be a common form of therapy resistance, but if we definite it away then we can’t even try to estimate if it is important.

To circumvent this problem, I want to turn to some history on the cost of resistance presented by Lenormand, Harmond & Gallet (2018). The most reductive view of a mutation is: one mutation, one fitness effect. Every mutation is either beneficial, neutral, or deleterious. From this view, any mutation that grants resistance is by definition beneficial. And it is also context-independent. This is a view we can still encounter in some introductory discussions, or on smooth fitness landscapes.

The cost of resistance idea was developed in the 80s and 90s to overcome this overly reductive view. Instead of having a single scalar value associated with a mutation, people associated two: a cost and a benefit. More importantly, these two values were context dependent. Under the treatment condition, the resistance mutation provided a benefit, and under the non-treatment condition, the resistant mutation caries a cost.

It was essential to the cost of resistance view that the signs of these two contributions were opposite. If the signs were the same then it was simply a purely beneficial or purely deleterious mutation, and not a resistance mutation. This was denominational. And I think that this is the perspective that our reviewer was coming from. Form them, what we called resistant cell lines, actually carried at least two mutations: one was a trade-off and thus a resistance, and the other was just a beneficial mutation.

Given that we’re dealing with definitions, that is certainly an acceptable decomposition. It is one based on effects. And it is a definition that sits nicely with the biologist. But I don’t think that it sits nicely with the clinician.

For the clinician, it seems like both effect and cause matter. The effect is the diagnostic that isn’t under the doctor’s direct control: under treatment the tumour either responds, responds temporarily, responds partially, or doesn’t respond. The cause is the variable that the doctor controls: treatment is applied or not. It is important to note, that this operational view of cause is an essential difference from the biologist. The biologist is much more an observer, and is fine with viewing mutation or other aspects outside his control as a cause. The clinician, however, is an actor, and as a person of action, she operationalizes her causes as something she controls — the independent control variable.

The biologist has an unknown hidden reductive cause — mutation — to be discovered. The clinician has a clear effective cause — application of treatment — to be controlled. Hence the different views of resistance.

To bring it back to Rob’s warning: For the biologist, de novo resistance is something that he couldn’t have controlled: it arose not because of how he set up the experiment but from a chance mutation during the experiment. For the clinician, a de novo resistance is also something that she couldn’t have controlled: it is present from the start of therapy and in a counter-factual world would have been there even if the therapy wasn’t applied. The biologist would have controlled for this aspect: he would have purified his initial population to eliminate the resistant clone. But a clinician can’t do on the patient the magic a biologist does on his Petri dish.

Since the biologist and clinicians have different methods of control, and different views of cause, they arrive at different definitions of resistance. The biologist sees causes as reductive, to be discovered, and outside of his control — and so arrives at a reductive view of resistance. The clinician sees causes as effective, to be controlled and optimized — and so arrives at an effective view of resistance.

Personally, I am biased towards the clinician’s view. Mostly because we’re doing all this research for her; or for her patient — to be more exact.

Of course, maybe this is a matter of field-endogenous vs field-exogenous questions. And so I should resist the potential applications and focus on the basic biology of the reductive definition.

But, if we are going to give reductive definitions for resistance then there is no reason to stop at just two contexts. We can consider a wider range of contexts, not just in drug or in DMSO. At the very least, there is drug concentration. Then there is other factors like non-drug resource availability. Finally, there are aspects of the population like frequency dependence on the sensitive vs ‘resistant’ types. This last aspect is what our work focused on.

As I mentioned above, if we looked at just monocultures of sensitive cells vs monocultures of resistant cells then it would have looked like a standard case of neutral resistance — not surprising since our resistant cells were produced gradually and thus would have had time to eliminate pleiotropic costs. This would be a confirmation of the standard model of resistance. The wrinkle, however, comes in when you widen the context to include not just drug-on vs drug-off but also type frequency. Now, once the two cell types are in the same micro-environment: i.e. same drug-state and same proportion of sensitive cells around them, the simple story disappears. The resistant cells start to outperform or match the sensitive cells in every context (at least of Fibroblasts aren’t present).

How are we to make sense of this in the language of cost of resistance? The context matters.

References

Kaznatcheev, A., Peacock, J., Basanta, D., Marusyk, A., & Scott, J. G. (2018). Fibroblasts and Alectinib switch the evolutionary games played by non-small cell lung cancer. bioRxiv: 179259.

Lenormand, T., Harmand, N., & Gallet, R. (2018). Cost of resistance: an unreasonably expensive concept. bioRxiv: 276675.