The first in vitro structural data for a tRNA/vRNA duplex were obtained from enzymatic and chemical probing on a heat-annealed complex between tRNA₃^Lys and a fragment of HIV-1 MAL vRNA containing the complete 5’-UTR [66,67]. They revealed intricate intermolecular interactions between tRNA₃^Lys and viral sequences located upstream of the PBS (Figure 4a-c). An identical conformation of the binary complex was observed when the primer was annealed to the vRNA with the nucleocapsid protein [68]. The stability of the proposed extended interactions was found to depend on the post-transcriptional modifications of tRNA₃^Lys[67,69] and on the complementarity between the anticodon loop of the primer and the conserved A-rich loop of the vRNA located in the apex of the U5 stem [66,70].

The NMR structure of a stem-loop containing the A-rich sequence of the MAL isolate revealed that the structure of the loop resembles the one of a tRNA anticodon loop, with a non-canonical G-A pair closing the loop and a U-turn motif leading to the continuous stacking of 164-AAA-166 onto A167 [71]. This pre-stacked structure renders the loop a good candidate for base-pairing, most likely with the U-rich anticodon loop of tRNA₃^Lys, in the same way as any tRNA anticodon would base-pair with messenger RNA. The solution structure of a duplex representative of the A-rich loop/tRNA interaction pointed to the importance of base modifications for stabilization of the vRNA/tRNA complex, in agreement with previously published work [57,67,69]. Indeed, the thio-group of U34 plays a crucial role in the stabilization of the loop-loop interaction that makes helix 6C (Figure 4c), whereas Ψ39 contributes to it [72]. Accordingly, in vitro reverse transcription experiments revealed that the modified nucleotides of tRNA₃^Lys and extended primer-template interactions are required for efficient initiation and transition to elongation of reverse transcription respectively [53,57].

A three-dimensional model of the MAL vRNA/ tRNA₃^Lys /RT ternary complex was proposed, based on probing data and footprinting experiments [73]. Strikingly, none of the extended intermolecular interactions (helices 3E, 5D and 6C in Figure 4c) are directly recognized by RT. Their assumed role is to impose a particular primer/template conformation that correctly places the intermolecular helix containing the PBS (helix 7F) and the 3’-OH group of the primer into the RNA-binding cleft of RT. Mutational analysis pointed to the three-nucleotide junction between helices 2 and 7F and the intermolecular RNA/tRNA interactions forming helix 6C as the most critical elements for efficient initiation of reverse transcription (Figure 3c) [44,53]. These results are in keeping with the three-dimensional model and agree with the idea that the intermolecular interactions impose a precise tertiary structure and/or counteract possible inhibitory effects of the vRNA structure that could otherwise lead to steric clashes between RT and RNA.

Structural probing performed in vitro on vRNA associated to tRNA₃^Lys by thermal annealing [48] led to the same results as in situ probing on vRNA/tRNA complexes isolated from virions [45,48]. In contrast to the subtype A situation, only the interaction involving the PBS could be observed for subtype B, with no structural rearrangement accompanying tRNA₃^Lys annealing [48]. Recently, two studies using the SHAPE technology were performed on genomic vRNA extracted from virions and deproteinized. Although they generally gave very similar results, the secondary structure models derived from these experiments vary in the PBS region [46,47]. Data from the first publication lead to the proposal of two additional primer/template interactions, on top of the PBS interaction: one between the A-rich sequence and the U-rich anticodon loop of tRNA₃^Lys, the other between the 3’ anticodon stem and part of the variable loop and a sequence upstream of the PBS, as suggested by Iwatani and co-workers [74]. The second set of data does not account for any primer/template interactions beside the one involving the PBS. The difference between the two sets of data may lie in a slightly different protocol for virus preparation and deproteinization.

It was suggested that in MAL the extended interactions with tRNA₃^Lys, which incidentally were also confirmed by in situ probing [48], are required to trigger the structural rearrangements generating the three-dimensional elements ultimately recognized by RT. These rearrangements would be necessary to prevent steric clashes. The absence of structural rearrangements in NL4-3/HxB2 was accounted for by the fact that essential structural elements such as helices 1 and 2 and the three single-stranded nucleotides at the junction between the PBS helix and helix 2, found in the MAL RNA/ tRNA₃^Lyscomplex, preexist in subtype B vRNA prior to tRNA₃^Lys annealing [44,48] (Figure 4c, d). In line with this interpretation, an 18-mer RNA oligonucleotide complementary to the PBS efficiently primed reverse transcription of NL4-3/HxB2 RNAs [75,6], but that of MAL RNA [53].

The possible role of the A-rich loop was nevertheless investigated in the two very similar HxB2 and NL4-3 RNAs, leading to unexpectedly divergent observations. The deletion of the AAAA sequence severely reduced initiation of reverse transcription [75,76]. However, the final amount of (-) ssDNA synthesis was found to be only modestly decreased, presumably due to the elimination of strong pausing sites after the addition of 11 to 14 nucleotides [76]. This deletion was also reported to have only a modest effect on (-) ssDNA synthesis in the presence of NCp [77]. By contrast, an AAAA to UUUU substitution on NL4-3 RNA was found to enhance the efficiency of (-) ssDNA synthesis, in the absence and presence of NCp [74]. It is possible that the deletion of the AAAA sequence induces structural perturbation of primer/template complex, thus explaining these differences.

An alternative intermolecular interaction between the TΨC loop in tRNA₃^Lys and a conserved 8-nucleotide sequence within helix 1, downstream of the PBS (123GACUCUGG130) (Figure 4d, e), termed the Primer Activation Signal (PAS), was proposed to be important for regulated reverse transcription [43,50]. Mutation of the PAS sequence (mutant 2L, Figure 4d) strongly reduced the efficiency of the initiation of reverse transcription, while mutation of the opposite strand (mutant 2R, Figure 4d) enhanced reverse transcription, both in vitro and using virion-extracted primer/template complexes. The latter effect is most likely due to disruption of helix 1 and exposure of the PAS sequence [43,50]. The double mutation (2LR), theoretically restoring base-pairing, did not sustain efficient reverse transcription, suggesting that the sequence rather than the helical structure is important for reverse transcription. It was also shown that the efficiency of reverse transcription could be modulated by PAS mutations engineered to strengthen or weaken the interaction with the anti-PAS [78]. However these mutations not only interfere with the PAS/antiPAS interaction, but also affect the stability of helix 1 that is directly recognised by RT [44,48]. Recently, other sequences downstream of the PBS were reported to regulate initiation of reverse transcription in vitro, by affecting the accessibility of the PAS motif [79]. Finally, attempts to complement, in vitro, the 2L mutation in the PAS sequence by a corresponding change in the antiPAS of a synthetic tRNA₃^Lys primer (2L- tRNA₃^Lys) were not successful [78]. The similarity between the PAS/anti-PAS interaction in HIV-1 and the interaction proposed for Rous sarcoma virus genomic RNA and its tRNA^Trp primer [80], together with the conservation of the PAS sequence among several retroviruses, led to the proposal that retroviral reverse transcription could be regulated by a common mechanism [78].

These conclusions were challenged by another study conducted on the same HxB2 vRNA variants [81]. In agreement with the previous studies [43,50], in vitro (-) ssDNA synthesis was severely affected by mutation 2L and was not restored by the double mutation 2LR, whether tRNA₃^Lys or an 18-mer RNA oligoribonucleotide complementary to the PBS were used as primers. However, this reduction was correlated with enhanced pausing, while the initial rate of primer extension was unaffected [81]. On the contrary to previous results [43,50], mutation 2R was found to reduce (-) ssDNA synthesis, with a five-fold decrease of the initial rate of primer extension [79]. This observation is consistent with in vitro data from another group indicating that efficient tRNA₃^Lys-primed (-) ssDNA synthesis required the 24 nucleotides downstream of the PBS (which include the sequence concerned by mutation 2R) [74].

As mentioned above, efficient (-) ssDNA synthesis was shown to require the 24 nucleotides downstream of the PBS when using NL4-3 vRNA template and tRNA₃^Lys as a primer [74]. The same template sequence was also found to be required with an all-RNA 18-mer primer (R18) complementary to the PBS or a chimeric primer containing nine deoxyribonucleotides at the 3’ end, but not with a DNA primer or a chimeric primer containing deoxyribonucleotides at the 5’ end. This was consistent with the fact that (-) ssDNA synthesis is sensitive to the nature of the helical conformation of the primer/template duplex [53-55,82]. Interestingly, the requirement for the downstream sequence is alleviated by the nucleocapsid protein (NCp) when natural or unmodified tRNA₃^Lys, but not the R18 primer, were used as primers [74]. Remarkably, mutation of nucleotides 142-148 complementary to the 3’ strand of the anticodon stem and the variable loop of tRNA₃^Lys (Figure 4d, e) abolished the rescue by NCp in the absence of the downstream sequence [74]. Thus, NCp seems to be able to increase reverse transcription by facilitating extended interactions between tRNA₃^Lys and upstream template sequences. These results address the question of the conditional requirement for highly conserved sequences forming helix 1, depending on the presence or absence of NCp. A possible explanation given by Iwatani et al. [74] is that helix 1 (Figure 4d) might be important for the maintenance of a structure stabilizing the initiation complex in the absence of NCp.

The deletion of the A-rich loop (169AAAA172) in HxB2 RNA resulted in diminished levels of infectivity and reduced synthesis of viral DNA [75]. After long-term culture, 167GG168 were substituted by two As, restoring wild-type reverse transcription and replication levels [73]. These results argued in favour of the importance of the conserved A-rich loop in HIV-1 replication. Besides, an earlier study showed that a 26-nucleotide deletion in the 3’ part of the U5 region (∆153-179) of NL4-3_{vRNA produced a severe defect in infectivity. However, the deletion was found to affect integration but did not impair viral DNA synthesis in acutely infected cells, despite the absence of the A-rich sequence. This study also indicated that deletion of a 25-nucleotide fragment adjacent to the previous one (∆127-152) had no detectable effect on virus replication [81]. Incidentally, this deletion contains seven nucleotides proposed to interact with the 3’ strand of the anticodon stem and variable loop of tRNA3^Lys, and six of the eight nucleotides of the PAS sequence [83] (Figure 4).}

Evidence for a role of the A-rich loop/anticodon interaction was brought by experiments showing that HIV-1HxB2 could replicate by stably utilizing non-cognate tRNAs as primers, provided that the PBS and the A-rich loop were simultaneously mutated to match both the 3’-end and the anticodon loop of the chosen tRNAs. This held true for tRNA^His[84-86], tRNA^Met[87,88], and tRNA_1,2^Lys [89]. These studies, together with the A-loop deletion (see above), pointed to the biological importance of the A-rich sequence. However, the results might be biased by the presence of overlapping sequences important for other functions, like integration [83,90,91]. It was also proposed that mutations of the A-rich loop allowed stable usage of non-cognate tRNAs by disfavouring tRNA₃^Lys, rather than favouring alternate primers [92]. In addition, mutated viruses rapidly acquired additional mutations that improved their replication efficiencies [85-89,93], suggesting that the A-rich sequence/anticodon-loop interaction may not be sufficient for optimal initiation of reverse transcription. The replication defect of these viruses might also be linked to their inability to selectively package their non-self tRNA primers [94].

Nevertheless, a direct correlation between the evolution of mutants using tRNA^His in cell culture [85,86,93], their efficiency in initiating reverse transcription in vitro, and the structure of the primer/template complex was observed [95]. Indeed, viruses that acquired additional adaptive mutations allowing a more stable interaction with the tRNA primer replicated more rapidly and displayed nearly wild-type levels of reverse transcription in vitro [85,86]. In addition, probing experiments indicated that the mutated A-rich loop forms stable interactions with tRNA^His only in the presence of these adaptive mutations [95].

Attempts were also made to switch the tRNA primer usage by mutating the PAS sequence. In short term cultures, viruses containing the PBS-PAS double mutations introduced to enforce the use of non-self primers tRNA^Pro or tRNA_1,2^Lys replicated more poorly than viruses with mutations in the PBS alone [96]. This observation is in line with in vitro experiments indicating that tRNA_1,2^Lys primed reverse transcription of an HxB2 RNA template containing the same PBS-PAS double mutation with only ~5 % efficiency compared to the wild-type situation [78]. However, when using primer/template complexes extracted from virions, the PBS-PAS double mutation restored tRNA_1,2^Lys extension to wild type levels [79]. The origin of the difference in the effects of these mutations when using either in vitro annealed primer/template or primer/templates extracted from virions is unclear.

Multiple independent virus evolution experiments selected one PBS variant that was able to stably utilize tRNA_1,2^Lys and restored high replication rates [96]. This mutant contained a second site mutation in the PAS sequence (U126C) that might stabilize the PAS/anti-PAS interaction by converting a G-U pair into a G-C pair [96]. Another second site mutation was identified within the RT gene (G3600A) that was responsible for a G490E change in the RNase H domain. Both mutations improved the replication capabilities of the mutated virus but only the U126C mutation was able to prevent reversion to the wild-type PBS [96].

Upon prolonged culture of a virus with the wild type PBS and a mutated PAS sequence, revertant viruses that partially overcame the reverse transcription defects were selected [43]. Interestingly, revertant viruses all acquired a single nucleotide substitution (C150U) that did not restore the PAS sequence but stimulated elongation of reverse transcription [43]. This observation fits with the proposal that mutations in the PAS affect elongation of reverse transcription rather than initiation [81].

The inability of HIV-1 to accommodate a non-self tRNA primer by substituting the PBS alone undoubtedly demonstrates the importance of additional interactions for the initiation of reverse transcription: MLV, for which no additional interactions have been described [97] easily utilises different tRNAs as primers [65], while avian sarcoma virus, for which an interaction between U5 and the TΨC loop has been proposed [80], does not [98]. Incidentally, HIV-1 has been shown to infrequently use tRNA_5a^Lys as primer [99]; this tRNA only differs from tRNA₃^Lys by one base-pair in the acceptor stem and has the potential to interacts with both the viral A-rich loop and the PAS sequence.

Although the large collection of findings reviewed above provide evidence that the complementarity between the PBS sequence and the 3’ part of the primer tRNA is not sufficient to determine primer usage, it is difficult to unambiguously rationalize the existence of precise additional interactions. The origin of apparent discrepancies is probably multiple and reflects the high complexity of the process. The most obvious explanation arises from (i) the existence of biases such as the nature of the strains used, (ii) the experimental parameters and tools utilised and (iii) unexpected effects of mutations. In particular, RNA mutations may interfere with a variety of steps other than the targeted one during the replication cycle. In addition the source of the interference can be direct (i.e., directly altering a binding site) or indirect (i.e., by triggering an undesired conformation that becomes inactive or induces a steric conflict). Even though a lot of publications already addressed the question of the structure of the complex mediating initiation of reverse transcription, several additional studies might be envisaged. For instance, whereas many detailed kinetic studies have been performed in vitro on the HIV-1 MAL isolate, in vivo studies of mutations affecting the proposed structure are yet to be performed. Similarly, the effects of mutations affecting the proposed PAS/anti-PAS interaction on the kinetics of reverse transcription have never been studied in infected cells.

The proposed mutually exclusive interactions might proceed transiently and be required at different steps of the replication cycle. Nevertheless, the results collected from experiments designed to probe the structure of subtype A and subtype B vRNA/tRNA complexes undoubtedly reveal an unexpected structural versatility of the HIV-1 initiation complex. The extraordinary plasticity of the vRNA and its capacity to adapt to a variety of constraints, in particular those imposed by mutations, offers the virus several solutions to precisely match RT and the viral RNA/tRNA primer complex in order to efficiently promote reverse transcription. As an example, the absence of interaction between the A-rich loop of NL4-3 and HxB2 vRNAs and the anticodon loop of tRNA₃^Lys is at variance with both in vitro and ex vivo data showing that a double mutation of the PBS and A-rich loop allowed the use of tRNA^His as a primer for reverse transcription. It was proposed that in this case the loop-loop interaction was used to compensate a negative contribution of tRNA^His[95]. This would imply that all tRNAs are not equally suited to prime reverse transcription, in agreement with a study comparing non cognate tRNAs [100].