Here we investigate the pivotal question of whether hA3G-induced G-to-A mutation is always lethal to the virus or if it may occur at sub-lethal frequencies.
We examined whether limiting-levels of hA3G activity could result in sub-lethal mutation rates using an in vitro
hA3G titration and sequencing experiment. The resulting in vitro
mutation patterns and per replication cycle rates were similar to mutation levels found in in vivo
hypermutated HIV DNA sequences implying that our experimental data reflected natural infection 
Second, based on the proportions of sequences carrying hypermutation in these datasets, we estimated that the maximum number of editing hA3G molecules packaged in a virion was 13 (95% CI, 6–26), which was only slightly higher than a previous biochemical estimate of 7+/−4 molecules 
. Using our estimate, we calculated that it was highly likely that the hypermutants we observed at the lowest wt-hA3G concentrations in the titration experiments () were caused by the incorporation of just a single hA3G-unit, and this becomes even more likely if the lower biochemical estimate is correct. As the editing observed was extensive and induced inactivating levels of G-to-A mutations, hypermutation typically seems to be an “all or nothing” phenomenon.
It has been hypothesized that a proportion of hypermutated sequences might be degraded by the cellular uracil DNA glycosylases UNG2 and/or SMUG1 and that this may contribute to the antiviral effect of hA3G 
. This hypothesis is however controversial 
as few studies support it 
while several have demonstrated that the absence or inhibition of UNG2 and/or SMUG1 activity neither abrogates hA3G inhibition of infection nor rescues viral cDNA accumulation in infected cells, suggesting that these enzymes are not involved in hA3G restriction of viral replication 
. Without conclusive data demonstrating UNG-mediated degradation, it is impossible to model in a realistic manner. However, we estimate that UNG-mediated degradation, if it destroyed a large proportion of the hypermutated sequences, would increase our estimation of k (the number of hA3G units in a virion). This would however not impact on our analyses of the role of hA3G in viral evolution in vivo
as sequences that are degraded disappear and do not form part of the viral population.
Third, we simulated editing in silico taking viral reading frames into account, to determine how low levels of hA3G-induced G-to-A mutations should be to increase viral diversification through neutral or potentially beneficial mutations while avoiding induction of lethal mutations (i.e. stop codons). We found that due to hA3G tetranucleotide target preferences, which render it efficient at generating stop codons, only a few mutations were generally needed to inactivate progeny viruses. When we compared the estimated LM50 rate with in vitro hypermutation rates, we found that it was ~10 fold less than the very lowest hypermutation frequency, suggesting that even a single hA3G-unit rarely, if ever, causes G-to-A mutations at potentially beneficial low levels.
Examining the role of hA3G in HIV evolution is an area of active research. In vitro
studies have used reporter-genes to extrapolate the effect hA3G editing on HIV diversification 
and the nucleoside analog RT inhibitor 2′,3′-dideoxy-3′-thia-cytidine (3TC or Lamivudine) to assess the effect of hA3G on the appearance of drug resistance mutations in lab-adapted HIV 
. Population sequencing, which only detects polymorphisms present in >20–25% of the viral population 
, was used to identify drug-resistance mutations and as Lamivudine accumulates to different degrees in different cell lines 
and increases intracellular dATP levels 
, which may affect RT misincorporation 
, the relevance of these studies for HIV evolution in natural infection needs further examination.
Studies of patient-derived HIV sequences either directly support our finding that hA3G is unlikely to contribute to viral diversification 
or does not contrast it 
. One report found that about 25% of rapidly diversifying sites in HIV were in sequence motifs that could be mutated by either hA3C, hA3F, hA3G or RT 
. Another study indicated that RT misincorporation was affected by imbalances in dNTP pools, which could explain the observed bias of G-to-A mutations in HIV evolution, and found no sign of hA3F/G editing 
. A third study of plasma virus sequences from HIV-1 infected patients that were either drug-naïve or had failed HAART demonstrated that Vif was highly polymorphic in both groups, but more so in pretreated patients 
. One of the Vif substitutions (K22H) was further analyzed as another substitution (K22E) had previously been demonstrated to partially neutralize hA3F but not hA3G 
. K22H was shown to partially neutralize hA3G whilst the effect on hA3F was not tested. In vitro
culture of mutated virus in MT2 cells that express high levels of hA3F and hA3G 
resulted in a minority of the sequences carrying sub-lethal mutations, which could be caused by either hA3F or hA3G. In contrast to hA3G, hA3C and hA3F are likely to sometimes induce sub-lethal G-to-A mutations as hA3F neutralization is dispensable for spread of HIV-1 in primary lymphocytes 
and hA3C neutralization is not needed for viral spread in SupT1 cells, which does not express hA3F and hA3G 
. A fourth cross-sectional study of patient-derived sequences found no evidence of an evolutionary footprint of hA3F/G 
and studies of thousands of patient-derived sequences have found either no, or very few, hypermutated RNA sequences, suggesting that low-level hypermutation, or recombination between hypermutated and non-hypermutated viruses, very rarely occurs in vivo
. Such a recombination has been found only once in vitro
after co-transfection of 32 hypermutated and non-hypermutated proviruses and 3TC drug selection 
As hA3G activity has such detrimental effects on HIV, strong viral selective pressures must act to optimize Vif's interaction with hA3G. However, as variable levels of hypermutation are observed in many HIV infected patients, other selective pressures may sometimes also affect vif
evolution. Several studies have demonstrated that CD8+ cytotoxic T-cells (CTL) can target Vif 
and we hypothesize that these CTL responses sometimes select for Vif variants that by chance interact less efficiently with hA3G. As hypermutation frequency has been found to correlate inversely with plasma viremia in three large patient cohorts 
, but not in two smaller cohorts 
, increasing hypermutation frequencies in patients through therapeutic measures is potentially beneficial.
In conclusion, our study suggests that hA3G activity is unlikely to increase HIV evolution and that hA3G-activity is highly likely to inactivate HIV-1.