Article activity feed

  1. Evaluation Summary:

    This paper will be of interest to microbiologists, clinicians, and public health workers with an interest in the possible impact of antibiotic use and regulations. The scope of the study is unusually high, integrating economic and geographical factors as well as genomic data among others. However, reasonable alternative explanations can be identified such that the data do not strongly favor the preferred hypothesis put forward by the authors.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. The reviewers remained anonymous to the authors.)

    Read the original source
    Was this evaluation helpful?
  2. Reviewer #1 (Public Review):

    The authors set out to understand what influences variation in aminoglycoside resistance in bacterial populations using existing genome data. While current policies aim to reduce antimicrobial consumption, the authors show that ecology and human exchanges are actually better predictors than consumption at explaining this variation.

    Strengths:

    This study uses a unique and ambitious comparative approach to understand the drivers of antibiotic resistance persistence within bacterial populations. We require comparative work of this kind to complement the current research focus on epidemiological modelling and smaller, more-focussed, experiments on the maintenance of antimicrobial resistance.

    The authors have collated an impressive number of genomes and used robust data mining techniques to detect aminoglycoside resistance genes in genomes and assemble metadata. In addition, they have developed new approaches to investigate the genomic context of resistance.

    While I am not an expert in this field, the statistical models, which control for spatial correlation, appear well considered and appropriate.

    This work is comprehensive and goes beyond current approaches. The conclusion may be surprising to some researchers, and so will likely have an impact. While the results are correlational, they highlight that other factors will certainly contribute to the long-term maintenance of antimicrobial resistance in bacterial populations. We need future research efforts to investigate these factors in more depth, in order to reduce the antimicrobial resistance burden.

    Weaknesses:

    The main weakness was the lack of consideration of biased sampling on the major conclusions of the study.

    While the biases in the data are openly acknowledged in the results, and considered a limitation in the discussion, there was no serious attempt to understand how the bias may influence the model predictions.

    At the very simplest, the first headline conclusion is that aminoglycoside-resistant bacteria are very common. But it is natural that the most research attention is given to bacteria that are 1) antimicrobial resistant threats in the first place and 2) sequenced specifically because they display phenotypic resistance. This will naturally bias sequenced genomes to have more resistance genes in the first place.

    The major conclusion of the study is that ecology/habitat is a better predictor of the variation in antimicrobial resistance, than antimicrobial consumption (it is the title). It seems logical that research attention in different habitats will be focussed on different bacterial taxa. These will usually be on bacterial taxa that have the greatest impact in these habitats. For example, clinical samples from humans, cyanobacteria from the ocean, carbon and nitrogen fixers from the soil, etc. How much do these sampling biases, in different taxa from different habitats, influence ecology as an explanatory variable for resistance variation?

    In addition, there may well be a good reason to predict that sampling bias may also influence the effect of human exchanges, as scientific collaborative endeavours will also mirror between-country trade and immigration.

    Therefore, it seems difficult to understand how well these conclusions hold up given the analyses as they are currently presented. One might argue that the sheer weight of the dataset might make dependencies on sampling bias lower, but when you consider sample sizes in different habitats, they can be low, due to how data is submitted to public databases. For example, from hundreds of thousands of complete genomes, there are only hundreds of samples that are scored as from the soil. It is not unlikely that these only come from a handful of studies.

    Read the original source
    Was this evaluation helpful?
  3. Reviewer #2 (Public Review):

    Pradier and Bedhomme predicted aminoglycoside modifying enzymes from over 160,000 publicly available bacterial genomes and examined the distribution of these genes across time and with respect to phylogeny, host bacteria, geography, biome, antibiotic use in their country-of-origin, and mobile genetic element association. Sample metadata from publicly available sequences were used to generate inputs for a model that estimated the relative importance of ecology, human movement, and aminoglycoside consumption for predicting the likelihood of sampling a genome with a given aminoglycoside resistance gene family, from a given ecosystem, at a given time. The authors conclude that the ecology of a sample (defined as the intersection between geography and biome) was the best predictor of what resistance genes were present in that sample and that the bulk tonnage of aminoglycosides used in the sample's host country was the least influential of the variables tested.

    This is one of the most comprehensive examinations of aminoglycoside modifying enzymes in public data, which is certainly a strength of the manuscript. The authors are also quite ambitious in the scope of their analyses, which is commendable. Many of the figure panels are intriguing and the analyses thought-provoking. However, inherent weaknesses in the sample metadata or antibiotic use data call into question some of the authors' conclusions. Though the authors are open about these limitations in the discussion, they are, in my mind, significant enough that the authors should either re-do some analyses and/or refrain from making some conclusions (as I outline below). Most problematically, the authors extend their conclusions from this very specific study of aminoglycoside resistance to antibiotic resistance in general. One of these broad conclusions is that antibiotic use practices are relatively unimportant determinants of resistance gene distributions. Even if their analysis was bullet-proof, this point cannot be generalized beyond their aminoglycoside data to all antibiotic resistance. I fear this could have negative impacts on public health outcomes if misconstrued as a universal truth by policymakers.

    Specific points of concern are as follows:

    1. I question whether the data on country-wide aminoglycoside consumption is appropriate for drawing conclusions regarding cross-biome selection pressures. As the authors mention, aminoglycosides were very rarely used in humans during the sample period, which is where most of their bacterial genomes originated. Thus, one may expect their dataset to be minimally impacted by aminoglycoside use practices. The authors attempt to address this concern on lines 390-408, concluding "we can assume that the available antibiotic consumption data are acceptable predictors for the antibiotic concentrations in other biomes". However, I am not convinced. The paragraph in question is quite general and speculative in its defense of this critical point. The cited work (Li et al) discusses a very specific watershed and appears to conclude that antibiotic concentrations act on finer spatial scales than entire countries.

    2. The authors normalize antibiotic usage by a country's land area (line 836), which I fear introduces undue noise to their analysis and could underweight the importance of this variable. Since about 90% of the bacteria with aminoglycoside modifying enzymes came from clinical/human (77.4%) or farm (12.3%) samples, wouldn't it make more sense to normalize by population or livestock numbers (especially because the latter group is where most aminoglycoside exposure would occur during the sampling period)?

    3. On multiple occasions, the authors make general conclusions about the (un)importance of antibiotic use practices on resistance gene distributions. For examples, see section 3.7 and lines 390-391 ("In this study, one of the goals was to ask whether the impact of antibiotic use on antibiotic resistance genes prevalence could be inferred..."). Since the manuscript focuses only on aminoglycosides (which haven't been used extensively as therapeutics in recent history), I believe the authors' broad claims about antibiotic resistance, in general, are outside the scope of their work. These claims are also potentially problematic, as they could be misconstrued by casual readers to suggest that antibiotic stewardship efforts are unhelpful or unnecessary.

    4. As the authors cite in their discussion, the importance of ecology and phylogeny for structuring resistomes has been reported previously. Though it is helpful to substantiate these observations with new analyses, this aspect of the manuscript lacks some novelty.

    5. Figure 4B is intriguing but supports the idea that aminoglycoside use selects for increasing prevalence of aminoglycoside modifying enzymes. This seems to imply antibiotic use alters resistance gene composition and thus runs counter to the rest of the manuscript's narrative. The authors do well to discuss these points (on lines 356-361), but I still have trouble ignoring this contradiction as I read the rest of the manuscript.

    Read the original source
    Was this evaluation helpful?