Predictable properties of fitness landscapes induced by adaptational tradeoffs

Essential revisions:

1) Our main concern is the generality of the assumptions made in the analysis. We ask the authors to discuss available data that is compatible or that violates the assumptions. Ogbunugafor et al., 2016, for example analyze dose-response curves for Malaria drugs and show that high resistance mutants don't necessarily have lower fitness in the absence of the drug.

Our model rests on three key assumptions: (i) a tradeoff between growth and resistance, (ii) the scaling property of dose-response curves and (iii) the approximate absence of (positive) epistasis in null-fitness and IC₅₀. The available data relevant to (ii) is rather limited, and we believe our discussion in the subsections “Scaling of dose-response curves” and “Limited epistasis in r and m”, is reasonably complete. Assumption (iii) is addressed in detail in the response to Essential revision 2) below. The reviewer’s reference to Ogbunugafor et al., 2016 concerns assumption (i). Here, there appears to be a broad consensus in the literature that tradeoffs are widespread (e.g., most resistance mutations come with a fitness cost) but exceptions in the form of compensatory mutations exist, and the data analyzed by Ogbunugafor et al. (originally obtained in Brown et al., 2010) belong to the latter category. We have added several references to the third paragraph of the Introduction where the tradeoff assumption is introduced, and extended the discussion of compensatory mutations in the third and eighth paragraphs of the Discussion. We also cite Brown et al., 2010 in this context.

2) The multiplicative behavior of IC₅₀ contributions is not obvious and the degree to which it holds might depend on the resistance mechanism. Would one expect similar multiplicative behavior for efflux pumps or drug-degrading enzymes? A general discussion of how the assumptions are compatible with different mechanisms of drug action and drug resistance would be welcome.

We agree that this behaviour is not obvious. We have added two paragraphs to the Discussion that describe in detail possible scenarios leading to a multiplicative behavior of the IC₅₀, and apply them to the system at hand (Discussion, fifth and sixth paragraphs).

3) We encourage the authors to think of ways to provide intuitive insight into the structure of these fitness landscapes and connect to existing models and metaphors.We have added a new Figure 8 that schematically illustrates different paradigmatic fitness landscape structures and a new paragraph (Discussion, seventh paragraph) where this figure is described.