Background:  Promiscuous G-to-A substitutions catalyzed by APOBEC3G (3G) enzymes are observed in a proportion of HIV-1 sequences in vivo. This “sequence editing” may introduce artifacts into some genetic analyses. The potential impact on genotypically based surveys of transmitted drug resistance was assessed.

Methods:  Rare mutations indicative of 3G-mediated lethal sequence editing were defined using the Stanford Drug Resistance Database and reference sequences representative of pol gene diversity among HIV-1 and other lentiviruses. The sentinel index, a classifier based on these mutations, was assessed in control datasets using receiver-operator curves. Variation among sentinel indices determined for a set of >6000 HIV-1 protease and RT sequences obtained from drug naïve individuals was characterized by using mixture modeling approaches to identify the parameters of distributions and to determine the maximum likelihood classification of sequences as lethally edited as opposed to viable wild type.

Results:  We identified 34 highly conserved 3G target sites in the protease-RT region at which G-to-A substitutions result in rare (<0.1% prevalence in Stanford database) amino acid substitutions or stop codons. The sentinel 3G index was a highly effective classifier for discriminating 3G-mediated sequence editing from naturally occurring and antiretroviral drug-selected sequence variation in control datasets. Analysis of >6000 sequences obtained from drug-naïve individuals indicated that the presence of 3 or more sentinel mutations denotes a >99% probability of lethal editing. Lethal editing in association with spurious drug resistance was detected in ~3% of proviral sequences obtained from whole blood, and 0.2% of samples obtained from plasma.

Conclusions:  Undetected 3G-mediated lethal editing can introduce artifacts into genotypic estimates of resistance. Although lethally edited sequences are only rarely observed in plasma samples, their more frequent detection in proviral DNA may have implications for surveillance programs utilizing dried blood spot filter paper, as these samples typically contain a proportion of proviral genetic material. The sentinel index is a robust classifier of lethal editing, suitable for screening large and diverse HIV-1 protease and RT sequence datasets.