Technical Comments

Response to Comment on "Evidence for Positive Epistasis in HIV-1"

See allHide authors and affiliations

Science  12 May 2006:
Vol. 312, Issue 5775, pp. 848
DOI: 10.1126/science.1110994

Abstract

Wang et al. analyzed artificially biased data to show that our results can be explained by a bias against sequences with low fitness. We explicitly acknowledged this potential caveat in our original study. Showing that an artificially introduced bias can produce a spurious signal of positive epistasis does not demonstrate that such bias exists in our original data.

Wang et al. (1) argue that our finding of positive epistasis in HIV-1 (2) could be explained by a bias in our data against sequences with low fitness. To illustrate this point, they repeat our analysis by generating two sets of simulated data, one based on an abstract fitness landscape and one based on the original data, in which a certain percentile of the sequences with lowest fitness is discarded. We explicitly mentioned this possible caveat in (2). That an artificially introduced bias can produce a spurious signal of positive epistasis is in our view obvious but does not demonstrate that such bias exists in our original data. The real issue is whether a bias actually exists and whether it undercuts our conclusion that our findings present a challenge for the mutational deterministic hypothesis (3). This an analysis of artificially biased data cannot address.

Our analysis was based on a large number of sequences bearing drug-resistance mutations and their associated fitness values measured in the absence of drugs. As stated in (2), the mutations were captured in our database because of their fitness benefit in the presence of drugs, irrespective of their effect on fitness in the absence of drugs. In most genetic backgrounds, these mutations confer a fitness cost in the absence of drugs (4), or else they would have become fixed in the wild-type virus. Contrary to the statement by Wang et al., our analysis does not assume that fitness in the absence of drugs is completely unrelated to fitness in the presence of drugs. Rather, it was based on the assumption that the mutations observed in our data set were generated by a selection pressure that was absent from the environment in which their fitness was measured.

Wang et al. argue that fitness in the presence of drugs and fitness in the absence of drugs are not completely unrelated because (i) compensatory mutations increase fitness in the absence of drugs, (ii) increased sensitivity to drugs (i.e., hypersusceptibility) correlates with low fitness in the absence of drugs, and (iii) mutations that greatly impair enzyme function reduce fitness both in the absence and in the presence of drugs. Because we did not claim that fitness in the presence and in the absence of drugs are generally unrelated, we discuss these points only with regard to their possible effect of generating a bias against strongly deleterious mutations. With regard to point (i), compensatory mutations increase fitness in the absence of drugs only in the genetic background of the associated primary resistance mutation, but are deleterious otherwise. These mutations enter our database because of their selective benefit in the presence of drugs and not because of their effects on fitness in the absence of drugs. We believe point (ii) is irrelevant, because the observation that increased sensitivity to drugs correlates with low fitness in the absence of drugs does not allow one to conclude that decreased sensitivity to drugs (i.e., resistance) should correlate with high fitness in the absence of drugs. Point (iii) suggests a misunderstanding by Wang et al. We do not claim that our data set contains mutations that are also deleterious in the presence of drugs. Mutations that are deleterious in treated patients are unlikely to go to fixation. Mutations captured in our data set are likely to be beneficial in treated patients and have thus been selected irrespective of their deleterious effects in the absence of drugs. Finally, Wang et al. raise the concern that a considerable fraction of the sequences come from untreated or lightly treated patients. However, we reported in the supporting online material (2) that only 192 out of 9466 sequences (2%) had the drug-sensitive wild-type amino acid at 59 selected sites that show drug-resistance–associated polymorphisms [see figure 2 in (2)].

To put this discussion into the larger context of the evolutionary benefit of recombination, we emphasize that to address this question, one needs to consider the effect of recombination operating on the genetic diversity that exists under natural conditions. The prerequisite for recombination in retroviruses is that a single cell becomes superinfected by two or more viable viral particles. Hence, recombination in retroviruses operates on a pool of viable mutants, not on a pool of mutants that includes some that have too low a fitness to be present in a population under natural conditions. We used mutations that (i) are selected for high fitness (i.e., higher fitness than the unmutated virus) in the presence of drugs, not against low fitness in the absence of drugs, and (ii) are typically viable but confer lower fitness than the wild-type allele in the absence of drugs. We thus studied the effect of recombination on a pool of mutations that are associated with fitness costs in the environment in which fitness is measured but have evolved because of a fitness advantage in the environment from which they were obtained. This may well resemble situations that the virus encounters under natural conditions, such as transmission, where mutations that have conferred a fitness advantage in the donor patient may turn out to confer a fitness cost in the recipient. The mutational deterministic hypothesis (4) tested in (2) requires negative epistasis among deleterious mutations. [Positive epistasis among beneficial mutations also selects against recombination provided there is no other force generating linkage disequilibria (5, 6).] The predominance of positive epistasis among the mutations found in our data set indeed represents a challenge for this hypothesis. In agreement with our findings, studies using approaches different from ours have also reported evidence for positive epistasis in other viruses (7, 8). Finally, we thank Wang et al. for pointing out that figure 1B in our original paper mistakenly showed the standard deviation divided by the number of observations rather than the standard error. Figure 1 shows the correct standard errors, which do not affect our analysis or conclusions.

Fig. 1.

Correction of figure 1B in (2), showing mean and standard error (gray dots and bars) of log fitness as a function of the number of amino acids differing from the reference virus (Hamming distance) for all sequences in the data set. The original figure mistakenly showed the standard deviation divided by the number of observations rather than the standard error. This figure reproduces the graph with the correct standard errors. Our mistake does not affect mean and 95% confidence intervals of the shown nonparametric regression, because these were calculated based on the actual data.

References and Notes

View Abstract

Navigate This Article