Within-host virus evolution during the extended treatment of RSV infection with mutagenic drugs

Article activity feed

1. 

   eLife

   Feb 24, 2023

   Author Response

   Reviewer #1 (Public Review):

   This paper describes the accrual of RSV mutations in a severely immunocompromised child with persistent infection and demonstrates that ribavirin increases the observed mutation rate with base pair changes (C to U and G to A) compatible with its known mechanism. The paper utilizes a mathematical model to explain the counterintuitive finding that viral load does not decrease despite loss of viral fitness and clinical improvement. Positive selection is observed but does not keep pace with deleterious mutations induced by ribavirin. Overall, though the data is restricted and limited to a single person, the analysis is rigorous and supports the paper's interesting conclusions.

   The paper is fascinating, but its generalizability is somewhat limited by the single study participant. …

   Author Response

   Reviewer #1 (Public Review):

   This paper describes the accrual of RSV mutations in a severely immunocompromised child with persistent infection and demonstrates that ribavirin increases the observed mutation rate with base pair changes (C to U and G to A) compatible with its known mechanism. The paper utilizes a mathematical model to explain the counterintuitive finding that viral load does not decrease despite loss of viral fitness and clinical improvement. Positive selection is observed but does not keep pace with deleterious mutations induced by ribavirin. Overall, though the data is restricted and limited to a single person, the analysis is rigorous and supports the paper's interesting conclusions.

   The paper is fascinating, but its generalizability is somewhat limited by the single study participant. Nevertheless, comparisons of therapy-induced deleterious mutations versus adaptive mutations over time is potentially important for multiple viruses.

   We thank the reviewer for their comments. Although we acknowledge that this is only a single case of infection, we believe that it is an interesting case, and we are keen to share our findings with the broader scientific community.

   Reviewer #2 (Public Review):

   In this work, Illingworth et al. investigate the effectiveness of ribavirin and favipiravir on the treatment of a paediatric patient with chronic RSV. These drugs cause mutations and the authors tested whether they could observe this effect through deep sequencing viruses from nasal aspirates over the course of treatment. They found an increase in mutations caused by ribavirin but favipiravir appeared to have no additional mutagenic effect. Despite the lack of change in viral load, the authors suggest that the ribavirin reduced viral fitness and did not lead to adaptive escape mutations. The authors modelled how generation time and fitness interacted with mutational load. They also estimated fitness for different haplotypes generated from the mutational data.

   Strengths of the paper:

   Using mutagenic drugs to treat viruses is generally accepted but results have been mixed with severe viral infections and specific evidence of the precise effects of the drugs is often lacking. This paper is especially valuable for demonstrating that despite in vitro evidence that favipiravir had some effect against RSV, there was no evidence for favipiravir having an effect in a patient. This differs from the authors previous work showing a clear clinical benefit to favipiravir in treating influenza. This paper also appears to be the first to sequence RSV from a patient having been exposed to ribavirin which is important for demonstrating that the drug is having a measurable effect.

   Weaknesses in the paper:

   I think there is a conceptual problem with the paper. Ribavirin is supposed to increase the mutational rate of the virus which would increase the mutational load. Mutational load has been calculated by summing up the frequencies of minor alleles. However, if a particular mutation rises in frequency, it does not mean that ribavirin has caused additional mutations at the same site but rather viruses containing the mutation have risen in frequency. If a subpopulation containing mutations rises through drift or selection to a relatively high percentage that will bias the mutational load. The authors provide ~75 mutations which were at significant percentages across multiple different timepoints. It seems that these mutations contribute significantly to the mutational load but changes in mutation percentages between samples do not reflect changes in mutational events but changes in viral haplotypes/subpopulations. In a previous study Lumby et al. 2020, the authors removed mutations at >5% from their analysis but there is no indication that they performed this step similarly here. Summing many small changes will give an indication of background mutational rate (though counting only a single mutation at each locus is perhaps the only method to remove the effect of viral clonal expansion).

   The mutational load is defined as the mean number of mutations per virus with respect to the consensus, equal to the sum of minor allele frequencies across the genome. We filter variant frequencies prior to calculating mutational load to compensate for noise arising from genome sequencing.

   We use a deterministic model of mutation-selection balance to describe the overall dynamics of mutational load, but are conscious that the dynamics of individual variants are complex. Genetic drift could contribute to these dynamics, as might hidden structure in the viral population, with stochastic observations of viruses from distinct subpopulations. As we make clear, our key assumption regarding mutational load is that all variants from the consensus are at least mildly deleterious; under this assumption calculating the sum of allele frequencies is an appropriate measurement of mutational load. Our model accounts for the possible presence of variants under stronger and weaker selection being observed at lower and higher frequencies respectively.

   We note that, in a case where distinct variants occurred in subpopulations, these variants would be observed in a mixture at lower frequencies than they existed in the subpopulations. This would lead to the observation of more variants overall, with each variant being at a reduced frequency. While stochastic effects would alter the frequencies of mutations in individual samples, if mutational load acted equally on each subpopulation, the total mutational load would be preserved across samples. The existence of subpopulations would not of itself invalidate the calculation of mutational load as we have performed it.

   Our previous study Lumby et al, 2020 considered a case where favipiravir was given for a short period of time in a case of influenza B infection. In that case we did not make an assessment of the total mutational load in a population, although we did remove mutations at >5% when considering the spectrum of mutations i.e. the proportion of mutations of each type C to T, G to A, etc. We are still working on different approaches to measuring mutational load, but we are not convinced that removing high frequency mutations is always a good idea when evaluating the total mutational load. Cutting out higher frequencies is potentially a useful means to study mutational spectra under viral mutagenesis, but in a measurement of mutational load it could exclude deleterious mutations.

   While ribavirin appears to have shown an effect, many questions remain. Why does the mutational load only increase for 3 points before plateauing? The authors would likely argue that this is the new saturation point for mutation load but they don't test it. Sequencing points from after the cessation of treatment would be expected to show lower mutational load but this data was not collected. Furthermore, questions remain over the methodology. It is thought that Ribavirin should only increase transitions and a transition/transversion ratio for the different samples would have been helpful. The absolute numbers of many mutation classes appear to have increased including transversions e.g AU. There isn't a good reason why nucleoside analogues should have caused this effect and perhaps it is an artefact.

   Ribavirin has been shown to increase C to T and G to A mutations; these are both transitions, but T to C and A to G mutations are also transitions; the proportion of these was found to decrease under treatment. We have included a new figure showing Ts/Tv ratios but we do not find a significant pattern of change in these statistics over time.

   The plateauing of the observed mutational load is consistent with the theory of mutationselection balance. Following a change in the mutation rate we would expect a shift to a new equilibrium U/s.

   Sequencing was conducted as part of an investigation that was secondary to treatment of the patient: All of the samples that were collected were sequenced. We agree that upon the cessation of mutagenic drugs we would expect to see a fall in mutational load.

   I don't think that the authors can reasonably determine how many haplotypes there are in the population from short read sequencing data. I think that the sequencing data very clearly shows subpopulations due to the large changes in mutation frequencies between different time points. The authors say that their analysis assumes a well-mixed population which is clearly not the case. Therefore, determining fitness of different haplotypes or mutations is likely not accurate.

   Although we have short read sequencing data, some of the reads we have span more than one locus, providing some information about linkage between variants. As noted in the Methods section our inference approach provides a minimal reconstruction of haplotypes: Our reconstruction details the smallest set of distinct haplotypes necessary to explain the data.

   Where we use a haplotype-based model to reconstruct the within-host evolution of the population, we neglect the potential presence of subpopulations by assuming a well-mixed population, then fully discuss the implications of this assumption for our result.

   Our basic question is whether within-host adaptation leads to a gain in viral fitness in excess of the loss of fitness imposed by an increase in mutational load. In this comparison we make a conservative (i.e. low) estimate for the extent of the loss of fitness through mutational load.

   When we look at within-host evolution our assumption of a well-mixed population attributes all of the systematic change in the viral population to the effects of selection. If some of this change arises through stochastic differences in emissions from a structured population, the influence of selection would be less than our inference. Thus, our estimate of the gain in fitness through within-host adaptation is a high estimate. As our high estimate of within-host fitness gain is less than a low estimate of the fitness lost through mutational load, our result is robust to our assumption.

   The authors construct a model to estimate viral fitness and suggest that viral fitness decreased with the drug. This is somewhat problematic to me as viral load has not changed so it would be reasonable to say that viral fitness was likely unaffected by the drug. The authors define fitness in terms of the number of mutations that each virus likely has and assumes that these mutations are deleterious. The authors then use this to claim that mutagenic drugs reduce fitness. This seems very circular to me. If the drugs reduce fitness, it should be observed as a property of the virus population. As the only measure was viral load, which didn't change, it is difficult to claim ribavirin reduced viral fitness. There are other reasons why there could be an increase in the number of mutations e.g. sequencing more subpopulations which would have nothing to do with fitness.

   We have discussed our assumption that variants in the viral population are deleterious; this lies behind the use of a model of mutation-selection balance. Under this assumption, the accumulation of a greater number of mutations following ribavirin treatment is indicative of a loss of viral fitness, although we cannot precisely quantify the magnitude of this loss. The link between an increased mutation rate and lower viral fitness is intrinsic to the concept of mutagenic drugs; our data show an increase in mutational load coincident with the therapeutic use of ribavirin.

   A change in viral fitness does not necessarily lead to a substantial and clearly observable drop in viral load; we say more about this in the response to comments below.

   At various points, the paper assumes that there is no selection taking place but immunoglobulin was being applied weekly and palivizumab monthly. The timing of when these drugs were given should be included. How did the palivizumab affect selection? The K272E mutation seems to go up and down but it is not clear if this was in response to drug infusion timing or if this mutation was present in a subpopulation.

   Our approach assumes that selection could act at two distinct levels: Firstly, we assume that the observed increase in mutational load correlates to a reduction in viral fitness; the link between viral fitness and mutational load is intrinsic to the equation of Haldane. Secondly we use a haplotype-based model to infer how selection is acting on the level of higherfrequency mutations; under the assumption of a well-mixed model we identify a signal of within-host adaptation.

   We have added details of the timing of palivizumab treatment to Figure 1. Immunoglobulin was given throughout; details of treatment have been given in Supporting Data. As we have now clarified in the Methods, our identification of potentially selected alleles was a two stage process, with the first assessing the level of noise in the data. Our model of noise envisages nonuniformity arising from multiple sources, including a situation whereby the viral population was divided in subpopulations, and in which reads comprised stochastic samples from these subpopulations. Given our model for noise, the observation of the K272E mutation at generally higher frequencies in earlier samples and generally lower frequencies in later samples was sufficient to call this as a potentially selected variant. We did not explore more complex models of drug-dependent selection.

   I think the main impact of the paper will be that favipiravir will not be used in the future to treat RSV. Given that the EC50 of favipiravir against RSC is ~100x that of influenza, favipiravir was unlikely to reach a therapeutic level in the patient. Nucleoside analogues have a mixed record at treating serious viral infections. Hopefully, this work will spur on future studies to precisely measure the effect that ribavirin has on RSV.

   Favipiravir was used in this patient following its successful experimental use against a case of influenza B infection (Lumby et al., 2020). We would be happy if our work inspires future research in this area.

   Read the original source
2. 

   eLife

   Dec 15, 2022

   eLife assessment

   Illingworth et al. investigate the effectiveness of ribavirin and favipiravir on the treatment of a paediatric patient with chronic RSV and identify an increase in mutations caused by ribavirin while favipiravir had no apparent mutagenic effect. Strength of evidence is incomplete for the mathematical model and solid for the mutational load analysis with potential for improvement in both cases with clarification of the methods. Major strengths are an interesting hypothesis and appropriate modeling methodology that will be of interest to virologists, clinicians and evolutionary biologists. Weaknesses in methodology pertain to mutational load measures possibly also capturing clonal expansion of new mutants and lack of clarity about how viral fitness is related to viral load in the mathematical model.

   Read the original source
3. 

   eLife

   Dec 15, 2022

   Reviewer #1 (Public Review):

   This paper describes the accrual of RSV mutations in a severely immunocompromised child with persistent infection and demonstrates that ribavirin increases the observed mutation rate with base pair changes (C to U and G to A) compatible with its known mechanism. The paper utilizes a mathematical model to explain the counterintuitive finding that viral load does not decrease despite loss of viral fitness and clinical improvement. Positive selection is observed but does not keep pace with deleterious mutations induced by ribavirin. Overall, though the data is restricted and limited to a single person, the analysis is rigorous and supports the paper's interesting conclusions.

   The paper is fascinating, but its generalizability is somewhat limited by the single study participant. Nevertheless, comparisons of …

   Reviewer #1 (Public Review):

   This paper describes the accrual of RSV mutations in a severely immunocompromised child with persistent infection and demonstrates that ribavirin increases the observed mutation rate with base pair changes (C to U and G to A) compatible with its known mechanism. The paper utilizes a mathematical model to explain the counterintuitive finding that viral load does not decrease despite loss of viral fitness and clinical improvement. Positive selection is observed but does not keep pace with deleterious mutations induced by ribavirin. Overall, though the data is restricted and limited to a single person, the analysis is rigorous and supports the paper's interesting conclusions.

   The paper is fascinating, but its generalizability is somewhat limited by the single study participant. Nevertheless, comparisons of therapy-induced deleterious mutations versus adaptive mutations over time is potentially important for multiple viruses.

   Read the original source
4. 

   eLife

   Dec 15, 2022

   Reviewer #2 (Public Review):

   In this work, Illingworth et al. investigate the effectiveness of ribavirin and favipiravir on the treatment of a paediatric patient with chronic RSV. These drugs cause mutations and the authors tested whether they could observe this effect through deep sequencing viruses from nasal aspirates over the course of treatment. They found an increase in mutations caused by ribavirin but favipiravir appeared to have no additional mutagenic effect. Despite the lack of change in viral load, the authors suggest that the ribavirin reduced viral fitness and did not lead to adaptive escape mutations. The authors modelled how generation time and fitness interacted with mutational load. They also estimated fitness for different haplotypes generated from the mutational data.

   Strengths of the paper:

   Using mutagenic drugs to …

   Reviewer #2 (Public Review):

   In this work, Illingworth et al. investigate the effectiveness of ribavirin and favipiravir on the treatment of a paediatric patient with chronic RSV. These drugs cause mutations and the authors tested whether they could observe this effect through deep sequencing viruses from nasal aspirates over the course of treatment. They found an increase in mutations caused by ribavirin but favipiravir appeared to have no additional mutagenic effect. Despite the lack of change in viral load, the authors suggest that the ribavirin reduced viral fitness and did not lead to adaptive escape mutations. The authors modelled how generation time and fitness interacted with mutational load. They also estimated fitness for different haplotypes generated from the mutational data.

   Strengths of the paper:

   Using mutagenic drugs to treat viruses is generally accepted but results have been mixed with severe viral infections and specific evidence of the precise effects of the drugs is often lacking. This paper is especially valuable for demonstrating that despite in vitro evidence that favipiravir had some effect against RSV, there was no evidence for favipiravir having an effect in a patient. This differs from the authors previous work showing a clear clinical benefit to favipiravir in treating influenza. This paper also appears to be the first to sequence RSV from a patient having been exposed to ribavirin which is important for demonstrating that the drug is having a measurable effect.

   Weaknesses in the paper:

   I think there is a conceptual problem with the paper. Ribavirin is supposed to increase the mutational rate of the virus which would increase the mutational load. Mutational load has been calculated by summing up the frequencies of minor alleles. However, if a particular mutation rises in frequency, it does not mean that ribavirin has caused additional mutations at the same site but rather viruses containing the mutation have risen in frequency. If a subpopulation containing mutations rises through drift or selection to a relatively high percentage that will bias the mutational load. The authors provide ~75 mutations which were at significant percentages across multiple different timepoints. It seems that these mutations contribute significantly to the mutational load but changes in mutation percentages between samples do not reflect changes in mutational events but changes in viral haplotypes/subpopulations. In a previous study Lumby et al. 2020, the authors removed mutations at >5% from their analysis but there is no indication that they performed this step similarly here. Summing many small changes will give an indication of background mutational rate (though counting only a single mutation at each locus is perhaps the only method to remove the effect of viral clonal expansion).

   While ribavirin appears to have shown an effect, many questions remain. Why does the mutational load only increase for 3 points before plateauing? The authors would likely argue that this is the new saturation point for mutation load but they don't test it. Sequencing points from after the cessation of treatment would be expected to show lower mutational load but this data was not collected. Furthermore, questions remain over the methodology. It is thought that Ribavirin should only increase transitions and a transition/transversion ratio for the different samples would have been helpful. The absolute numbers of many mutation classes appear to have increased including transversions e.g AU. There isn't a good reason why nucleoside analogues should have caused this effect and perhaps it is an artefact.

   I don't think that the authors can reasonably determine how many haplotypes there are in the population from short read sequencing data. I think that the sequencing data very clearly shows subpopulations due to the large changes in mutation frequencies between different time points. The authors say that their analysis assumes a well-mixed population which is clearly not the case. Therefore, determining fitness of different haplotypes or mutations is likely not accurate.

   The authors construct a model to estimate viral fitness and suggest that viral fitness decreased with the drug. This is somewhat problematic to me as viral load has not changed so it would be reasonable to say that viral fitness was likely unaffected by the drug. The authors define fitness in terms of the number of mutations that each virus likely has and assumes that these mutations are deleterious. The authors then use this to claim that mutagenic drugs reduce fitness. This seems very circular to me. If the drugs reduce fitness, it should be observed as a property of the virus population. As the only measure was viral load, which didn't change, it is difficult to claim ribavirin reduced viral fitness. There are other reasons why there could be an increase in the number of mutations e.g. sequencing more subpopulations which would have nothing to do with fitness.

   At various points, the paper assumes that there is no selection taking place but immunoglobulin was being applied weekly and palivizumab monthly. The timing of when these drugs were given should be included. How did the palivizumab affect selection? The K272E mutation seems to go up and down but it is not clear if this was in response to drug infusion timing or if this mutation was present in a subpopulation.
   I think the main impact of the paper will be that favipiravir will not be used in the future to treat RSV. Given that the EC50 of favipiravir against RSC is ~100x that of influenza, favipiravir was unlikely to reach a therapeutic level in the patient. Nucleoside analogues have a mixed record at treating serious viral infections. Hopefully, this work will spur on future studies to precisely measure the effect that ribavirin has on RSV.

   Read the original source
5. 

   eLife

   Dec 15, 2022

   Reviewer #3 (Public Review):

   The use of mutagenic drugs in combating new viral diseases is increasing, so it is imperative to understand how they might impact the evolutionary trajectory of RNA viruses and weigh their potential benefits versus their harms. The authors examined the impact on treatment outcomes and virus populations of treatment with mutagenic drugs (ribavinin and favipiravir) in a child with severe combined immunodeficiency syndrome and RSV pneumonitis. The authors report that despite a three-fold increase in viral mutation within-host evolution was still slow with only minor gain in viral fitness. The patient's clinical status was stable despite virus non-clearance by the drugs.

   Despite looking at only one case, this study illustrates the potential impacts of widespread use of antiviral mutagenic drugs in the event of a …

   Reviewer #3 (Public Review):

   The use of mutagenic drugs in combating new viral diseases is increasing, so it is imperative to understand how they might impact the evolutionary trajectory of RNA viruses and weigh their potential benefits versus their harms. The authors examined the impact on treatment outcomes and virus populations of treatment with mutagenic drugs (ribavinin and favipiravir) in a child with severe combined immunodeficiency syndrome and RSV pneumonitis. The authors report that despite a three-fold increase in viral mutation within-host evolution was still slow with only minor gain in viral fitness. The patient's clinical status was stable despite virus non-clearance by the drugs.

   Despite looking at only one case, this study illustrates the potential impacts of widespread use of antiviral mutagenic drugs in the event of a viral epidemic. The authors warn in the discussion of the study that the results should be interpreted with caution if the same drugs are given to individuals who are immunocompetent, which I agree.

   Read the original source
6. 

   Version published to 10.1101/2022.09.02.22279474 on medRxiv

   Sep 18, 2022