The influence of biological, epidemiological, and treatment factors on the establishment and spread of drug-resistant Plasmodium falciparum

Article activity feed

1. 

   Version published to 10.7554/elife.77634 on eLife

   Jul 7, 2022
2. 

   eLife

   Jun 17, 2022

   Author Response

   Reviewer #2: Public review:

   This manuscript reports results from a sensitivity analysis done to assess jointly the contribution of various factors to the spread of P. falciparum malaria parasites that are resistant to antimalarial drugs. It also explores how probable parasite genotypes are to establish as a function of their consequent rate of spread.

   This manuscript's main contribution is its joint consideration of several factors not considered jointly before. The authors achieve their goal of doing a large joint analysis using computer simulations generated under a model framework that includes a model of malaria transmission and a model called an emulator. The malaria model has new features capturing different drug mechanisms and the capacity to track different degrees and types of drug resistance. It is very …

   Author Response

   Reviewer #2: Public review:

   This manuscript reports results from a sensitivity analysis done to assess jointly the contribution of various factors to the spread of P. falciparum malaria parasites that are resistant to antimalarial drugs. It also explores how probable parasite genotypes are to establish as a function of their consequent rate of spread.

   This manuscript's main contribution is its joint consideration of several factors not considered jointly before. The authors achieve their goal of doing a large joint analysis using computer simulations generated under a model framework that includes a model of malaria transmission and a model called an emulator. The malaria model has new features capturing different drug mechanisms and the capacity to track different degrees and types of drug resistance. It is very sophisticated but computationally expensive. The emulator emulates the input to output relationship of the sophisticated malaria model, thereby enabling the authors to do the large joint analysis, which would be computationally prohibitive using the malaria model alone. This is a practical solution to a computationally expensive problem. It could be applied to other computationally expensive models in epidemiology, if not already done so.

   The results are impactful because they reinforce the need for continued surveillance of resistance to so-called partner drugs and they reinforce our understanding of drug properties that best withstand resistance. Three drug profiles were investigated: two monotherapies and a combination therapy that combines the two drugs used as monotherapies. The properties of the drugs mimic the properties of the drugs used in artemisinin-based combination therapies (ACTs). (The drug that is like artemisinin and its derivatives has a short half-life, high maximum kill rate and parasites resistant to can endure longer drug exposure times. The partner-like drug has a short half-life, low maximum killing rate and parasites resistant to it can endure higher drug concentrations.) ACTs are recommended for the treatment of malaria in almost all endemic counties. They include a fast-acting artemisinin derivative and a more slowly acting partner drug to kill residual parasites.

   Supported by their simulated data, the authors conclude that partner drug resistance likely promotes the establishment and spread of artemisinin resistance. They then go on to say that their results support the belief that partner drug resistance precedes the evolution of artemisinin resistance. This belief is consistent with the spread of artemisinin resistance in the Greater Mekong Subregion but not in Africa. It cannot be tested directly in this study because the malaria model does not capture the sequential evolution of resistance, but the arguments the authors use to extrapolate from their results are logical.

   Almost all the results are intuitive, and support previously published epidemiological and laboratory studies. Among the factors that can be acted upon, drug properties play an important role. Longer half-lives of the artemisinin-like drug hinder the spread of artemisinin-like resistance. Longer half-lives of the partner-like drug promote the spread of partner-like drug resistance but protect the artemisinin-like drug. Despite this protective effect, the authors conclude that "reducing the half-life of the partner drug in an ACT regimen could reduce the spread of resistance". Stated thus, this may seem counterintuitive. However, it is logical: longer half-lives of the partner drug likely compromise the artemisinin derivative in the long run by first promoting the emergence, establishment and spread of resistance to the partner drug. Nonetheless, it cannot be tested directly in this study because the malaria model does not capture the sequential nature of the evolution of resistance.

   Although this study makes an important contribution, it has some weaknesses. Firstly, it does not capture the sequential evolution of resistance. Secondly, it is important to note that monotherapies are non-longer recommended for malaria treatment. Looking forward, the authors discuss briefly how their findings might extrapolate to triple combination therapies (TACTs), arguing that the two long-acting drugs of TACTs should ideally have matching half-lives. Although it seems reasonable to make this point based on extrapolation, a TACT-like drug profile merits full investigation. What would happen, for example, if the two long-acting drugs exert inverse drug pressure, selecting complementary mutations? Of course, it is not possible to consider all factors that might impact the spread of antimalarial drug resistance. Some potentially important factors not discussed presently in this manuscript include sub-quality drugs and additional factors that impact coverage, such as absorption (nutritional status). Recombination, an obligate stage of the malaria parasite lifecycle which does not feature in the malaria model, is mentioned briefly. Under a modified malaria model, recombination could affect some of the results at higher entomological inoculation rates (EIRs) because higher EIRs leads to more effective recombination. For example, resistance to the partner-like drug might not spread preferentially in high EIR settings when access to treatment is high. This is because the phenotypes of parasites resistant to partner drugs are typically encoded for by more than one mutation, so can thus be disrupted by recombination. Recombination could also affect the spread of artemisinin resistance. Although artemisinin resistance is typically encoded for by a single mutation, compensatory mutations elsewhere in the genome may play a role in mitigating the fitness cost. If so, recombination might restore the resistance cost in high EIR settings with low access to treatment. On the contrary, recombination could unite multiple mutations that encode drug resistance. In short, recombination could have a complicated and hard-to-intuit effect. It thus merits further investigation using a model.

   We thank the reviewer for his/her supportive public review and for sharing his/her expert view on factors not captured by our models. We do identify a typo for non-specialist readers: the “partner-like drug” has a long half-life (we call it a “typo” because subsequent comments show the Reviewer is well aware of this fact).

   (1) Sequential resistance and recombination

   We recognise that OpenMalaria does not incorporate recombination is a limitation of our study. We have, in the past, seriously investigated whether to re-code our model to incorporate recombination, but for now, it would require fundamental, far-reaching changes to the mosquito model, parasite transmission model, and code. Additionally, to realistically represent recombination would result in significant increases in both memory use and computational time. We have prioritised using existing functionality rather than committing to resource-intensive code revisions, noting potential impacts of recombination on establishment and spread of resistance are addressed below and in our discussion.

   Practically, the lack of recombination means that we can investigate the spread of resistance of one mutation at a time. For example, we could not simultaneously simulate the spread of a mutation that confers resistance to drug A and the spread of a mutation that confers resistance to drug B in a drug-sensitive parasite population. However, we could assume that resistance arises first to drug B and gets fixed before the emergence of resistance to drug A. Similarly, we assumed that the resistant genotypes had a fixed degree of resistance (i.e., a fixed number of mutations) and could not acquire a new mutation that could confer higher degrees of resistance across the simulation. However, we assessed how the selection pressure and impact of factors vary for different degrees of resistance (from low to high degrees of resistance). Thus, our model can capture the effect of a changing pattern of selection that occurred with the increasing degrees of resistance due to sequential evolution. However, we acknowledge that we did not dynamically model the sequential evolution of resistance. We added this remark to the discussion (L547-550).

   In our paper we highlighted that the consequence of not modelling recombination is that we overestimated the evolution of drug resistance in settings with a high rate of infection when the resistant phenotype involves multiple mutations. The reviewer rightly highlighted that this could also influence the evolution of resistance to artemisinin (despite being caused by one mutation) due to compensatory mutation. We added this point to the revised version of our paper. We have also added that the effect of recombination depends on the frequency of each mutation needed to confer resistance. When these mutations are present at a low frequency (such as during the establishment phase), recombination will have a stronger effect as resistant parasites are more likely to recombine with sensitive parasites. Thus, the resistant phenotype is more likely to be lost. However, the impact of recombination decreases when the frequency of resistant mutations increases because resistant parasites are more likely to recombine with a resistant parasite. Thus, the main consequence of not simulating recombination events was that in high transmission settings, our model overestimated the probability of establishment of resistant parasites that need multiple mutations to be drug-resistant or that require additional mutations to restore the fitness cost. This means that the difference between the probability of establishment in high and low transmission settings is probably larger than reported here (see figure 4). In addition, we overestimated the spread of these resistant parasites when the mutations were present in low frequencies. However, these assumptions likely did not impact the probability of establishment and rate of spread of parasites that only need one mutation to confer resistance or do not have a mutation that reduces the fitness cost associated with resistance.

   (2) Modelling Triple drugs (TACTs)

   As pointed out by the reviewer, monotherapies are no longer recommended for malaria. Here we assessed the impact of factors on the evolution of parasite-resistance to the short-acting drug and the long-acting drug used separately in (non-recommended) monotherapy to identify determinants specific to each drug profile. This allowed us to identify some determinants that would not have presented themselves if we had only examined drug combination therapy. Once we identified which factors drive resistance for each drug profile, we looked at the combination of these two drug profiles and observed how dynamics changed. As suggested by the reviewer, the next step would be to look at the evolution of resistance under triple combination therapy.

   Our study showed that resistance to the partner drug (long-acting, previously referred to as drug B) depends on the length of the selection window. This result supports the evidence that triple artemisinin combination therapies (TACTs) can delay the spread of resistance to partner drugs as it can minimise the selection pressure that occurs during the selection window if the two long-acting drugs have the same half-lives. While we agree future work could focus on selective pressures from different drug profiles in TACT, this would require a very large study to look at different profiles of three drugs etc. and is outside our scope of already a very large study. Even so, we believe no additional analysis is necessarily needed to highlight the points on TACT in the paper as they logically follow from our results.

   However, we agree that other factors are likely to play a role in the evolution of resistance under TACTs, such as the inverse selection pressure generated by some drugs, as highlighted by the reviewer. Understanding the impact of factors on the evolution of resistance to TACTs is an important question and should be further investigated. However, this question is outside the scope of our study, as it would require running many more analyses and considering additional factors (such as the inverse selection pressure generated by some drugs, the synergic effect between drugs, or the fact that some mutation can conferee some degree of resistance to multiple drug, etc.), and could be considered in future work.

   (3) General comments

   As underlined by the reviewer, we did not directly assess the impact of sub-quality drugs and absorption (i.e. from poor nutritional status) on the rate of spread. However, one could extrapolate the impact of sub-quality drugs and absorption by recognising that both lead to a lower Cmax. In our study, we assessed the impact of low Cmax on the rate of spread, and thus one can extrapolate the effect of sub-quality drugs and absorption based on findings from our study. Factors such as poor nutrition could also affect other drug factors such as half-life, which we did not investigate, so we regard this as an important operational point made by the Reviewers which could be addressed in future studies.

   Read the original source
3. 

   eLife

   Jun 17, 2022

   Evaluation Summary:

   This manuscript explores the establishment and spread of antimalarial drug-resistant P. falciparum parasites using a combination of transmission modeling and model emulation. The authors add an important component to the broader understanding by jointly considering multiple factors driving drug resistance.

   (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
4. 

   eLife

   Jun 17, 2022

   Reviewer #1 (Public Review):

   The authors present a very thorough description and comparison of how various drug and epidemiological properties accelerate or decelerate the spread of antimalarial drug resistance, using a combination of transmission modeling and model emulation. The authors relate the selection coefficient, which is easier to measure computationally and in the field, to the probability of establishment, which is more challenging to measure. This was a very interesting paper and I appreciated how comprehensive and rigorous the approach was so that all these factors could be compared using the same framework.

   Read the original source
5. 

   eLife

   Jun 17, 2022

   Reviewer #2 (Public Review):

   This manuscript reports results from a sensitivity analysis done to assess jointly the contribution of various factors to the spread of P. falciparum malaria parasites that are resistant to antimalarial drugs. It also explores how probable parasite genotypes are to establish as a function of their consequent rate of spread.

   This manuscript's main contribution is its joint consideration of several factors not considered jointly before. The authors achieve their goal of doing a large joint analysis using computer simulations generated under a model framework that includes a model of malaria transmission and a model called an emulator. The malaria model has new features capturing different drug mechanisms and the capacity to track different degrees and types of drug resistance. It is very sophisticated but …

   Reviewer #2 (Public Review):

   This manuscript reports results from a sensitivity analysis done to assess jointly the contribution of various factors to the spread of P. falciparum malaria parasites that are resistant to antimalarial drugs. It also explores how probable parasite genotypes are to establish as a function of their consequent rate of spread.

   This manuscript's main contribution is its joint consideration of several factors not considered jointly before. The authors achieve their goal of doing a large joint analysis using computer simulations generated under a model framework that includes a model of malaria transmission and a model called an emulator. The malaria model has new features capturing different drug mechanisms and the capacity to track different degrees and types of drug resistance. It is very sophisticated but computationally expensive. The emulator emulates the input to output relationship of the sophisticated malaria model, thereby enabling the authors to do the large joint analysis, which would be computationally prohibitive using the malaria model alone. This is a practical solution to a computationally expensive problem. It could be applied to other computationally expensive models in epidemiology, if not already done so.

   The results are impactful because they reinforce the need for continued surveillance of resistance to so-called partner drugs and they reinforce our understanding of drug properties that best withstand resistance. Three drug profiles were investigated: two monotherapies and a combination therapy that combines the two drugs used as monotherapies. The properties of the drugs mimic the properties of the drugs used in artemisinin-based combination therapies (ACTs). (The drug that is like artemisinin and its derivatives has a short half-life, high maximum kill rate and parasites resistant to can endure longer drug exposure times. The partner-like drug has a short half-life, low maximum killing rate and parasites resistant to it can endure higher drug concentrations.) ACTs are recommended for the treatment of malaria in almost all endemic counties. They include a fast-acting artemisinin derivative and a more slowly acting partner drug to kill residual parasites.

   Supported by their simulated data, the authors conclude that partner drug resistance likely promotes the establishment and spread of artemisinin resistance. They then go on to say that their results support the belief that partner drug resistance precedes the evolution of artemisinin resistance. This belief is consistent with the spread of artemisinin resistance in the Greater Mekong Subregion but not in Africa. It cannot be tested directly in this study because the malaria model does not capture the sequential evolution of resistance, but the arguments the authors use to extrapolate from their results are logical.

   Almost all the results are intuitive, and support previously published epidemiological and laboratory studies. Among the factors that can be acted upon, drug properties play an important role. Longer half-lives of the artemisinin-like drug hinder the spread of artemisinin-like resistance. Longer half-lives of the partner-like drug promote the spread of partner-like drug resistance but protect the artemisinin-like drug. Despite this protective effect, the authors conclude that "reducing the half-life of the partner drug in an ACT regimen could reduce the spread of resistance". Stated thus, this may seem counterintuitive. However, it is logical: longer half-lives of the partner drug likely compromise the artemisinin derivative in the long run by first promoting the emergence, establishment and spread of resistance to the partner drug. Nonetheless, it cannot be tested directly in this study because the malaria model does not capture the sequential nature of the evolution of resistance.

   Although this study makes an important contribution, it has some weaknesses. Firstly, it does not capture the sequential evolution of resistance. Secondly, it is important to note that monotherapies are non-longer recommended for malaria treatment. Looking forward, the authors discuss briefly how their findings might extrapolate to triple combination therapies (TACTs), arguing that the two long-acting drugs of TACTs should ideally have matching half-lives. Although it seems reasonable to make this point based on extrapolation, a TACT-like drug profile merits full investigation. What would happen, for example, if the two long-acting drugs exert inverse drug pressure, selecting complementary mutations? Of course, it is not possible to consider all factors that might impact the spread of antimalarial drug resistance. Some potentially important factors not discussed presently in this manuscript include sub-quality drugs and additional factors that impact coverage, such as absorption (nutritional status). Recombination, an obligate stage of the malaria parasite lifecycle which does not feature in the malaria model, is mentioned briefly. Under a modified malaria model, recombination could affect some of the results at higher entomological inoculation rates (EIRs) because higher EIRs leads to more effective recombination. For example, resistance to the partner-like drug might not spread preferentially in high EIR settings when access to treatment is high. This is because the phenotypes of parasites resistant to partner drugs are typically encoded for by more than one mutation, so can thus be disrupted by recombination. Recombination could also affect the spread of artemisinin resistance. Although artemisinin resistance is typically encoded for by a single mutation, compensatory mutations elsewhere in the genome may play a role in mitigating the fitness cost. If so, recombination might restore the resistance cost in high EIR settings with low access to treatment. On the contrary, recombination could unite multiple mutations that encode drug resistance. In short, recombination could have a complicated and hard-to-intuit effect. It thus merits further investigation using a model.

   Read the original source
6. 

   Version published to 10.1101/2022.02.05.22270500v1 on medRxiv

   Feb 7, 2022