Accurate estimates of virus mutation rates are important to understand the evolution of the viruses and to combat them. However, methods of estimation are varied and often complex. Here, we critically review over 40 original studies and establish criteria to facilitate comparative analyses. The mutation rates of 23 viruses are presented as substitutions per nucleotide per cell infection (s/n/c) and corrected for selection bias where necessary, using a new statistical method. The resulting rates range from 10−8 to10−6 s/n/c for DNA viruses and from 10−6 to 10−4 s/n/c for RNA viruses. Similar to what has been shown previously for DNA viruses, there appears to be a negative correlation between mutation rate and genome size among RNA viruses, but this result requires further experimental testing. Contrary to some suggestions, the mutation rate of retroviruses is not lower than that of other RNA viruses. We also show that nucleotide substitutions are on average four times more common than insertions/deletions (indels). Finally, we provide estimates of the mutation rate per nucleotide per strand copying, which tends to be lower than that per cell infection because some viruses undergo several rounds of copying per cell, particularly double-stranded DNA viruses. A regularly updated virus mutation rate data set will be available at
www.uv.es/rsanjuan/virmut.
The mutation rate is a critical parameter for understanding viral evolution and has important practical implications. For instance, the estimate of the mutation rate of HIV-1 demonstrated that any single mutation conferring drug resistance should occur within a single day and that simultaneous treatment with multiple drugs was therefore necessary (
72). Also, in theory, viruses with high mutation rates could be combated by the administration of mutagens (
1,
5,
21,
44,
53,
83). This strategy, called lethal mutagenesis, has proved effective in cell cultures or animal models against several RNA viruses, including enteroviruses (
11,
39,
44), aphtoviruses (
83), vesiculoviruses (
44), hantaviruses (
10), arenaviruses (
40), and lentiviruses (
15,
53), and appears to at least partly contribute to the effectiveness of the combined ribavirin-interferon treatment against hepatitis C virus (HCV) (
13). The viral mutation rate also plays a role in the assessment of possible vaccination strategies (
16), and it has been shown to influence the stability of live attenuated polio vaccines (
91). Finally, at both the epidemiological and evolutionary levels, the mutation rate is one of the factors that can determine the risk of emergent infectious disease, i.e., pathogens crossing the species barrier (
46).
Slight changes of the mutation rate can also determine whether or not some virus infections are cleared by the host immune system and can produce dramatic differences in viral fitness and virulence (
75,
90), clearly stressing the need to have accurate estimates. However, our knowledge of viral mutation rates is somewhat incomplete, partly due to the inherent difficulty of measuring a rare and random event but also due to several sources of bias, inaccuracy, and terminological confusion. One goal of our work is to provide an update of published mutation rate estimates, since the last authoritative reviews on viral mutation rates were published more than a decade ago (
29,
30). We therefore present a comprehensive review of mutation rate estimates from over 40 original studies and 23 different viruses representing all the main virus types. A second, and perhaps more ambitious, goal of our study is to consolidate the published literature by dealing with what we regard as the two main problems in the field: the use of different units of measurement and the bias caused by selection.
The problem of units is linked to the different modes of replication in viruses. Under “stamping machine” or linear replication, multiple copies are made sequentially from the same template and the resulting progeny strands do not become templates until the progeny virions infect another cell. In contrast, under binary replication, progeny strands immediately become templates and hence the number of molecules doubles in each cycle of strand copying, increasing geometrically. This basic distinction leads to two different definitions of the mutation rate: per strand copying or per cell infection. If replication is stamping machine-like, there is only one cycle of strand copying per infected cell and hence the two units are equivalent. However, binary replication means that the virus completes several cycles of strand copying per cell. The actual replication mode of most viruses is probably intermediate between these two idealized cases, and although it is known to be closer to linear in some viruses (
9,
19) and closer to binary in others (
26,
55), it is often unknown. This leads to uncertainties in mutation rate estimates. For instance, in the case of poliovirus 1, the estimated rate per strand copying can vary by 10-fold depending on whether stamping machine or binary replication is assumed (
27). Typically this difference in the unit of measurement has been overlooked in comparative studies. Here, we express published estimates in the same unit.
The other issue that we address is selection. In general, deleterious mutations tend to be eliminated and hence are less likely to be sampled than neutral ones, introducing a bias in mutation rate estimates. To avoid this problem, selective neutrality is sometimes enforced by the experimenter, such that the number of mutations increases linearly with time (
58,
88). The opposite strategy is to focus on lethal mutations, which have necessarily appeared during the last cell infection cycle, thus establishing a direct and time-independent relationship between the observed mutation frequency and the underlying mutation rate (
13,
37). In between these two special cases, an explicit correction for selection is needed. Even if the effect of each individual mutation on viral fitness is unknown, the effect of selection can be statistically accounted for as long as the number of mutations sampled for estimating mutation rates is large. We do this here using empirical information about the distribution of mutational fitness effects previously obtained for several viruses (
6,
23,
73,
80). Importantly, the basic properties of this distribution appear to be well conserved (
78), and hence the proposed method should be applicable to a wide variety of viruses.
Using the resulting mutation rate data, we retest some previously accepted general patterns, suggest new ones, infer the mode of replication of some viruses, and compare the rates of mutation to substitutions with those to insertions/deletions (indels).
N95s on the way...
