Science, discussed.

Fitness landscape analysis illuminates drug treatments [review]

An interesting manuscript in bioRxiv describes the computational analysis of fitness peaks in characterised drug-resistant bacteria to iterate the importance of ‘mixed drugs’ treatments. The use of more than one antibiotic is an already established practice as it allows to: i) kill a wider spectrum of microbes by targeting different biological processes and, in a more advance approach, ii) decrease the likelihood of the appearance of drug-resistance mutations in individual proteins.

Exploiting evolutionary non-commutativity to prevent the emergence of bacterial antibiotic resistance
Daniel Nichol, Peter Jeavons, Alexander G Fletcher, Robert A Bonomo, Philip K Maini, Jerome L Paul, Robert A Gatneby, Alexander RA Anderson, Jacob G Scott

The Authors of this manuscript offer a third reason for this procedure: a single drug might lead to a very ‘specialised’ response at molecular level that could result in an evolutionary bottleneck, i.e. the protein is unable (less likely) to further evolve to cope with a second drug. So, in addition to choosing a pair of effective drugs, the timing of their use also becomes critical.

A computational analysis that, in the words of the authors, ‘in the absence of experimentally determined landscapes‘ cannot answer all concerns about the validity of this approach, but a intriguing and valid perspective on the development of (multi) drug-resistant bacteria nevertheless.


About Pietro Gatti

Interested in discussing (good) Science Lover of coffee & good films. Ideas all & only my own.

5 comments on “Fitness landscape analysis illuminates drug treatments [review]

  1. Artem Kaznatcheev

    I am really excited about this work. Unfortunately fitness landscapes are notoriously difficult to characterize especially when you want to test for special properties like non-commutivity of the transition matrices under variation in environments, so we might have to get creative in how we look for empirical justification. Here, rigorous qualitative analysis using the tools of theoretical computer science might help us narrow down the space of reasonable fitness landscapes. I’m looking forward to see where Dan, Jacob, and the rest of the team take this.

    • @p_gl

      Thank you for your comment Artem, and for linking two very interesting posts on your blog.

  2. Michael Manhart

    I think this is a really neat study, but I’m pretty concerned that two major assumptions would break down if someone actually tested the predictions with evolution experiments on bacteria. The first is that the predictions are based on a very limited sample of mutations (~4). The mutation rate and population size of bacteria are large enough that real populations are probably sampling a large number of mutations simultaneously, and so even if you block the pathway through a few of the best mutations as they propose, the population is likely to just find others, even if they are a little less beneficial. For the same reason SSWM is probably not a very good approximation for bacterial population dynamics. I’m a bit of a hypocrite for complaining about this — I have used this approximation a lot, because obviously it makes models a lot more tractable — but I do suspect it would be a major problem in actually predicting the outcome of a specific experiment. The authors do acknowledge these problems themselves, but I think they are worth emphasizing in considering the practical value of these results.

    • Dan Nichol

      I’m inclined to agree with you here, we’ve chosen the model for its tractability and for that reason the predictive power suffers. However, how much it suffers is not something we know without some experiments (which we’re trying to pull together).

      The question we’d like to ask is if the model is predictive enough to be used as a heuristic to guide treatment scheduling. If it turns that is then that’s great. If it turns out not to be then there is still some value to our results. The qualitative result, that the order drugs are given in matters, at least poses a new question – how can be optimise drug orderings to avoid resistance? If our model is incapable of answering this question then we’ll need to move towards models which match the mechanisms at work more closely, but at least we have a new question to answer.

      Finally, the very limited sample of mutations is something we’re not particularly pleased with ourselves. As occurs often in biology we’ve had to work with the data we can find instead of the data we want. We’re currently excited about work (for example: which explores how we might infer large fitness landscapes from fewer data points. If you know of any larger datasets we could explore with our model then we’d love to hear about it.

      Cheers for you feedback!

      • Michael Manhart

        Thanks for the response, Dan, I’m very curious to see where your work leads. I think it is a very cool idea overall. Also thanks for the response on the bioRxiv comment, I now understand what you meant there.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s


Creative Commons License
This blog is licensed under a Creative Commons Attribution 3.0 License.


Enter your email address to follow this blog and receive notifications of new posts by email.

Join 1,560 other followers

All Categories

%d bloggers like this: