Investigation of HIV-1 transmission events by phylogenetic methods: requirement for scientific rigour

Since the first use of molecular methods to ascertain HIV-1 transmission was published in 1992 [1], controversy in this area has been vivid [2–4]. When approaching the use of HIV-1 sequences to characterize linkage, a key determinant is the choice for the most informative genetic region; in this respect, the use of the pol gene has hitherto been unpopular. We nonetheless demonstrated that sequences of this genetic region, widely available since the onset of drug resistance testing, holds sufficient genetic variation to allow phylogenetic analyses [5]. In an opinion paper published in response to our article, Stürmer and colleagues [6] not only challenged the reliability of phylogenies constructed on the basis of the protease and reverse transcriptase genes, but also recommended the blind use of the V3 region of the envelope (env) gene. In view of the importance of this issue for epidemiological studies, as well as in the clinical and legal arena, we would like to address their line of argument, in order to widen the debate. In essence, we maintain that the key issue is for a dataset to contain sufficient variability as defined by phylogenetic criteria, regardless of whether the sequence is gag, pol or env. There is no such thing as an ultimate gene for evolutionary analyses of HIV-1. Ideally, full-length sequences should be used for the investigation of potential linkages by phylogenetic means; however, practicalities preclude such an approach. Echoing previous and well-established opinions, see for example Leitner et al. [7], Stürmer et al. [6] recommended the use of env gene sequences, the extensive variation of which has made it attractive for such analyses. However, the exploitation of env is far from unproblematical. First, convergent evolution (i.e. identical mutational patterns in unlinked sequences) has repeatedly been observed in the V3 loop of the env gene [8,9]. More importantly, the rapid genetic diversification of this region is likely to compromise the identification of linked sequences in distantly sampled individuals. Both divergence and diversity of the HIV-1 env gene have been shown to increase linearly in the early stages of infection [10]. The latter observations may explain why one of the possible transmission pairs identified in the pol tree published by Stürmer et al. [6] failed to be supported by their env tree. The choice of an appropriate genetic target for such studies must not only be considered in light of the intrinsic variability of the dataset itself, but also of the possible time span separating the samples under comparison. Our own data suggest that the relative stability of the polymerase gene may confer some benefit in this respect. We identified the same clustering patterns in phylogenies independently constructed on the basis of the pol, gag and env genes of the same HIV-1 samples, supporting the idea of sufficient intrinsic sequence variation in the pol region. Stürmer et al. [6] supported their view by identifying sequences that appear linked within a phylogeny based on the pol region, but which fail to fulfil the arbitrary criteria of a bootstrap support greater than 70% within a tree based on the partial env gene. However, the methodological underpinning of this approach is flawed. First, the authors produced a neighbour-joining tree [11], constructed under the ‘Kimura two-parameters’ (K2P) model of nucleotide substitution [12], which is, in our view, unsatisfactory. If the choice of a phylogenetic method for reconstructing transmissions is less sensitive than the choice of a genetic region [7], the misuse of a model of evolution can have severe consequences on the accuracy of the reconstruction. As rates of evolution differ across HIV-1 lineages, populations, or genetic regions, the selection of an optimal model must be a prerequisite when estimating HIV-1 phylogenies. The systematic (and often unjustified) use of the K2P model of evolution is unfortunately frequent in HIV molecular analyses. When using an over-simplistic model, features of importance in the context of HIV-1 transmission, such as branch length, may be underestimated [13]. Moreover, when statistically testing different evolutionary models for various HIV-1 genetic regions, Posada and Crandall [14] demonstrated that the K2P model was suboptimal, whichever gene was under study. Therefore, the criteria used to discriminate between potential linked and unlinked sequences must be adapted to the specificities of one's dataset. These criteria should be empirically determined from dataset to dataset, and not dogmatically implemented. The unique criterion used by Stürmer et al. [6] (i.e. a bootstrap support above 70%, regardless of the tree topology) is clearly insufficient to draw reasonable inferences regarding the true or false linkage of infections. Such a cut-off value must first consider genes evolving at a different pace, because bootstrap evaluation is sensitive to the degree of polymorphism exhibited by the sequences. Second, bootstrap evaluation is not on its own sufficiently discriminatory, and the branch length supporting suspected transmission pairs is another obvious pattern to take into account. A sensible way to determine such criteria would be to incorporate positive controls within the sequence alignment, such as sequences from known transmission pairs or intrapatient follow-up samples. Stürmer et al. [6] failed to present such controls, compromising the pertinence of their findings. As accession numbers for several of the sequences used by the group have not been provided, we have been unable to re-analyse their data according to the above criteria. In conclusion, we agree that close attention is required in dealing with HIV-1 sequences for epidemiological, clinical or forensic purposes. Using a rigorous and reproducible methodology, we recently showed that our dataset of pol sequences hold enough sequence variation to allow phylogenetic reconstruction, despite the conservation of the gene. Our conclusion was based on a stringent comparative analysis of the respective phylogenetic signals held by the three main genes of HIV-1, namely the pol, gag and env genes.