Structural Variability of the Initiation Complex of HIV-1 Reverse Transcription*

HIV-1 reverse transcription is initiated from a \batchmode \documentclass[fleqn,10pt,legalpaper]{article} \usepackage{amssymb} \usepackage{amsfonts} \usepackage{amsmath} \pagestyle{empty} \begin{document} \(\mathrm{tRNA}_{3}^{\mathrm{Lys}}\) \end{document} molecule annealed to the viral RNA at the primer binding site (PBS), but the structure of the initiation complex of reverse transcription remains controversial. Here, we performed in situ structural probing, as well as in vitro structural and functional studies, of the initiation complexes formed by highly divergent isolates (MAL and NL4.3/HXB2). Our results show that the structure of the initiation complex is not conserved. In MAL, and according to sequence analysis in 14% of HIV-1 isolates, formation of the initiation complex is accompanied by complex rearrangements of the viral RNA, and extensive interactions with \batchmode \documentclass[fleqn,10pt,legalpaper]{article} \usepackage{amssymb} \usepackage{amsfonts} \usepackage{amsmath} \pagestyle{empty} \begin{document} \(\mathrm{tRNA}_{3}^{Lys}\) \end{document} are required for efficient initiation of reverse transcription. In NL4.3, HXB2, and most isolates, \batchmode \documentclass[fleqn,10pt,legalpaper]{article} \usepackage{amssymb} \usepackage{amsfonts} \usepackage{amsmath} \pagestyle{empty} \begin{document} \(\mathrm{tRNA}_{3}^{Lys}\) \end{document} annealing minimally affects the viral RNA structure and no interaction outside the PBS is required for optimal initiation of reverse transcription. We suggest that in MAL, extensive interactions with \batchmode \documentclass[fleqn,10pt,legalpaper]{article} \usepackage{amssymb} \usepackage{amsfonts} \usepackage{amsmath} \pagestyle{empty} \begin{document} \(\mathrm{tRNA}_{3}^{Lys}\) \end{document} are required to drive the structural rearrangements generating the structural elements ultimately recognized by reverse transcriptase. In NL4.3 and HXB2, these elements are already present in the viral RNA prior to \batchmode \documentclass[fleqn,10pt,legalpaper]{article} \usepackage{amssymb} \usepackage{amsfonts} \usepackage{amsmath} \pagestyle{empty} \begin{document} \(\mathrm{tRNA}_{3}^{Lys}\) \end{document} annealing, thus explaining that extensive interactions with the primer are not required. Interestingly, such interactions are required in HXB2 mutants designed to use a non-cognate tRNA as primer (tRNAHis). In the latter case, the extended interactions are required to counteract a negative contribution associate with the alternate primer.

Reverse transcriptase (RT), 1 an RNA-and DNA-dependent DNA polymerase that also harbors an RNase H domain, converts the single-stranded genomic RNA of retroviruses into a double-stranded DNA with duplicated long terminal repeats (1). It initiates DNA synthesis from a tRNA, tRNA 3 Lys in the case of the type 1 human immunodeficiency virus (HIV-1), that is selectively encapsidated into the viral particles (2,3). During budding and maturation of the retroviral particles, the 18 3Ј-terminal nucleotides of the primer tRNA are annealed to the primer binding site (PBS) located in the 5Ј-region of the genomic RNA. Additional interactions between the genomic RNA and the primer tRNA have been described in several retroviruses, including avian retroviruses (4,5), HIV-2 (6), and feline immunodeficiency virus (7), but not in murine leukemia virus (8).
In HIV-1, several alternative interactions between the viral RNA (vRNA) and tRNA 3 Lys have been proposed, but no consensus model has been reached. The most detailed in vitro structural studies of the initiation complex of reverse transcription were performed on the HIV-1 MAL isolate (9 -11). They allowed us to build secondary (9) and tertiary (11) structure models of the MAL initiation complex. In these models, parts of the anticodon arm and of the variable loop of tRNA 3 Lys interact with the vRNA, upstream of the PBS (Fig. 1). Of particular importance is the interaction between the anticodon loop of tRNA 3 Lys and a viral A-rich loop (helix 6C in Fig. 1b) that is crucial for efficient initiation of HIV-1 MAL reverse transcription in vitro (12,13). However, no ex vivo data on the MAL reverse transcription is available. MAL is a complex recombinant of HIV-1 subtypes A, D, and I that contains a 23-nucleotide duplication, including part of the PBS and the downstream region (14) (Fig.  1a). Thus, the general validity of the MAL-based models can be questioned.
No detailed structural study of the initiation complex of prototype subtype B isolates, such as HXB2 and the closely related NL4.3, has been performed so far, and functional studies yielded controversial results. Some in vitro experiments suggested that the interaction between the viral A-rich loop and tRNA 3 Lys anticodon loop also exists in HXB2 (15,16), whereas other studies supported different intermolecular interactions (17) (analogous to helices 5D and 3E in Fig. 1b). Long term replication experiments with mutant HXB2 viruses designed to use a variety of non-cognate primer tRNAs indicated a functional role for the interaction between the anticodon and A-rich loops. Indeed, whereas mutant HXB2 viruses in which only the PBS was adapted were unstable and rapidly reverted, mutant HXB2 viruses in which the PBS and the viral A-rich loop were simultaneously adapted either to tRNA His (18 -20), tRNA Met (21), tRNA 1,2 Lys (22), or tRNA Glu (23) could stably use these tRNA species as primer. Finally, in vitro and ex vivo analysis of HXB2 mutated in helix 1 (Fig. 1c) suggested another intermolecular interaction (between the 5Ј-strand of helix 1 and the T⌿C arm of tRNA 3 Lys ), which has been named the PAS (primer activating signal)-anti-PAS interaction (24,25).
Several factors complicate the study of the initiation of reverse transcription and may explain these discrepancies. First, several functional sites overlap in the PBS domain, and mutations may affect several steps in the HIV-1 life cycle (26,27). Second, the HIV-1 RNA structure is versatile (14,28), and mutations may induce aberrant structures, both in vitro and ex vivo, that obscure analysis (29). Third, in vitro systems used to analyze the initiation of reverse transcription might not adequately reflect the in vivo situation. Finally, the possibility remains that the structure of the initiation complex might not be conserved among HIV-1 isolates.
In order to circumvent these problems and to compare the in situ structure of the MAL and NL4.3 initiation complexes of reverse transcription, we performed the first structural probing of the HIV-1 genomic RNA in infected cells (i.e. prior to tRNA 3 Lys annealing) and in viral particles (i.e. after tRNA 3 Lys annealing and virus maturation). These structural data were consistent with the in vitro probing data we collected on the MAL, NL4.3, and HXB2 complexes, and validated the in vitro functional studies we performed on the wild-type (WT) MAL and HXB2 isolates and on HXB2 mutants utilizing tRNA His as primer. Taken together, our results show that in HXB2/NL4.3 and, according to extensive sequence analysis, in 86% of HIV-1 isolates, there is no stable interaction between the anticodon loop of tRNA 3 Lys and the A-rich loop. This loop-loop interaction clusters to the A and G HIV-1 subgroups and recombinants of the former subgroup, including MAL. In HIV-1 MAL, this interaction counteracts a negative contribution of the vRNA. This interaction also exists in the mutant HIV-1 HXB 2 viruses using tRNA His as primer, where it neutralizes a negative contribution of the tRNA His primer. This last result explains why these mutant HXB2 viruses do not behave as the WT HXB2 isolate.

EXPERIMENTAL PROCEDURES
Primers, Templates, and RT Used for in Vitro Studies-Native tRNA 3 Lys was purified from beef liver as described (30). Its sequence and post-transcriptional modifications are identical to those of human tRNA 3

Lys
. Synthesis and purification of tRNA His were previously described (31). ORN 3 Lys and ORN His were obtained from Dharmacon Research Inc. Primers were 32 P-labeled at their 5Ј-end according to published procedures (11).
WT MAL, WT HXB2, HXB2 His, and HXB2 His-AC-GAC RNAs were transcribed in vitro and purified as described (31). The MAL and HXB2derived RNAs correspond to nucleotides 1-311 and 1-295 of the genomic RNA, respectively. A plasmid expressing HXB2 HIV-1 RT was kindly provided to us by Dr. Torsten Unge (Uppsala, Sweden), together with the protocols for protein overexpression and purification.
Annealing of the Primers to the Templates-Viral RNA and tRNA 3 Lys , tRNA His , ORN 3 Lys , or ORN His were denatured in water for 2 min at 90°C and chilled on ice. Annealing was performed at 70°C for 20 min in sodium cacodylate (pH 7.5) 50 mM, KCl 300 mM. Annealing efficiency was routinely checked by native polyacrylamide gel electrophoresis and was found to be higher than 95% in all experiments.
Minus Strand Strong Stop DNA Synthesis-In a standard experiment, vRNA was annealed with 32 P-labeled primer (10 nM final concentration) and preincubated for 4 min at 37°C with 25 nM RT in 50 mM Tris-HCl, pH 8.0, 50 mM KCl, 6 mM MgCl 2 , 1 mM DTE. Reverse transcription was initiated by adding dNTP (50 M each). At times ranging from 15 s to 60 min, DNA synthesis was stopped by an equal volume of formamide containing 50 mM EDTA, and the products were analyzed on 8 or 15% denaturing polyacrylamide gels and quantified with a Bio-Imager BAS 2000 (Fuji).
Data Analysis-Curve fitting was performed with version 4.08 carbon of the Igor Pro Software (WaveMetrics Inc.). Global curve fitting was performed as described in the software manual. In this procedure, data sets obtained with different primers were simultaneously fit with a bi-exponential function. The amplitude parameters were considered as "local" parameters that could adopt different values for each primer, whereas the fast and slow rate constants were considered as "global" parameters with identical values for the two primers. The global fitting approach was validated by the limited increase of the 2 value divided by the number of data points, as compared with "simple" fitting, despite the fact the number of adjustable parameters was reduced. In all cases, this value was Ͻ6 ϫ 10 Ϫ4 .
Chemical Probing of vRNA in Vitro-After hybridization, the vRNA/ tRNA complexes were incubated at 20°C for 15 min in the annealing buffer supplemented with 5 mM MgCl 2 before probing with dimethyl sulfate (DMS). After addition of 1 g of yeast total tRNA, RNA was modified by addition of 1 l of 10-fold diluted DMS for 5 or 10 min. RNA modification was stopped with 200 l of ethanol and 50 l of sodium acetate (0.3 M, pH 5.3) containing 1 g of glycogen. Modified bases were detected by primer extension with reverse transcriptase as previously described (14).
Infectious Molecular Clones-The HIV-1 NL4.3 molecular clone was used to generate a chimerical construct, pNL4.3-MAL, containing nucleotides 61-274 from the MAL isolate substituted for the homologous NL4.3 region. To obtain this construct, the AflII-BssHII fragment of pJCB (32) was PCR-amplified using primers pS-1/15MAL (5Ј-GTC TCT CTT GTT AGA CCA G-3Ј) and pAS-BssHII (5Ј-CCG CCC CTC GCC TCT TGC CGT GCG CGC TTC AGC AAG CCG AGT CC-3Ј). The underlined nucleotides correspond to changes introduced to restore a subtype B dimerization initiation site (33) and allow subsequent cloning into the pNL4.3 molecular clone. The PCR product was digested with AflII and BssHII and cloned between the same restriction sites of pLTR5Ј (34), generating pLTR5Ј-MAL. Finally, the AatII-SphI fragment of pLTR5Ј-MAL was substituted for the homologous region of pNL4.3.
Cell Culture, Transfection, and Infection-HeLa cells were grown in Dulbecco's modified Eagles's medium. CEMx174 cells were grown in RPMI 1640 medium. All cells were supplemented with 10% heat-inactivated fetal calf serum. HeLa cells were transfected with 10 g of WT and mutant plasmids using the calcium phosphate coprecipitation method. Progeny viruses were quantified by a micro RT assay of the cell culture supernatants (35). Five million CEMx174 cells were infected with equivalent amounts of virus as determined by RT activity. One hour after infection, the cells were diluted in 20 ml of RPMI 1640.
In Vivo RNA Modification-At 72 h after infection, CEMx174 cells were washed twice with phosphate-buffered saline and suspended in 30 l of this buffer. Progeny viruses were clarified by centrifugation and passed through a 0.22-m cellulose acetate filter (Millipore). The culture supernatant was then pelleted through a 20% sucrose cushion by ultracentrifugation with a SW28 rotor at 27,000 rpm for 2 h at 4°C. Virus pellets were suspended in 90 l of phosphate-buffered saline. Cells and viruses were treated with 3 l of DMS for 4 and 8 min at 37°C. Reaction was stopped by adding 1 ml of TriReagent (Molecular Research Center), and RNA was extracted as described by the supplier. Modified bases were detected by primer extension as described (14).
Sequence Alignment-An alignment of the 538 HIV-1 sequences encompassing nucleotides 550 -700 was recovered from the HIV sequence data base (www.hiv.lanl.gov) (36). (Numbering is according to the U3 start of HXB2; it corresponds to nucleotides 95-245 of the HXB2 genomic RNA). This alignment was checked with ClustalX (37) and improved manually in a text editor.

RESULTS
In order to gain further insight into the initiation complex of HIV-1 reverse transcription and to conciliate divergent results, we studied the structure of this complex ex vivo, thus preventing any potential artifact due to in vitro primer annealing and RNA renaturation, and to the absence of viral and/or cellular cofactors. In addition, we compared the initiation complexes formed by the MAL and NL4.3/HXB2 isolates. These isolates significantly differ in their sequence and RNA secondary structure of the PBS domain before tRNA 3 Lys annealing ( Fig. 1, a and c).
In Situ and in Vitro Structural Probing of NL4.3 and HXB2 Initiation Complexes-DMS, which modifies adenines at position N1 and cytosines at N3 when the Watson-Crick side of these nucleotides is not base-paired (38), has been extensively used to probe the structure of retroviral initiation complexes in vitro (8 -11, 13, 24, 29, 31, 39). In order to extend these results to more biologically relevant contexts, we probed the structure of the vRNA directly in infected cells (i.e. prior to tRNA 3 Lys annealing) and in viral particles (i.e. after tRNA 3 Lys annealing and virus maturation). Infected cells and purified virions were treated with DMS, and the RNA was extracted after chemical modification. Modified nucleotides in the genomic RNA were specifically detected by extension of primers complementary to sequences located 3Ј of the major splice donor site, which could not anneal to spliced HIV RNA. The results described below represent the first successful in situ probing of HIV RNA.
We first studied CEMx174 cells acutely infected with NL4.3, as well as their progeny virions (Fig. 2a). In the absence of DMS treatment, a strong pause was observed in the PBS of NL4.3 RNA extracted from cells, but not in the RNA extracted from virions. This specific pausing might be due to RNA degradation by cellular RNases during RNA extraction, or to selective exclusion of the degraded RNA from the virions. This pause decreased further extension of the primer, producing a slightly weaker signal in the cells, as compared with virions. The reactivity profile of NL4.3 RNA in cells is fully compatible with the secondary structure model presented in Fig. 1c: most reactive nucleotides are located in bulges (i.e. A-132, A-133, A-159), internal and apical loops (i.e. C-151, C-152, A-157, A-170, A-171, A-172) and in the single-stranded stretch 3Ј to the PBS; a few are located in unstable helical regions.
When compared with cells, nucleotides A-191, A-192, C-193, A-194, and A-198 are protected in virions, as expected from tRNA 3 Lys annealing to the PBS. However, this protection was incomplete, in keeping with previous studies revealing subop-timal PBS occupancy in virions (40). Beside these changes, the DMS modification pattern of NL4.3 RNA was remarkably similar in cells and in virions, with only minor differences in the band intensities. The most pronounced differences were located at A-216, A-147, and C-151, which were reactive in the cells, and became protected in the virions (Fig. 2a). Remarkably, adenines 168 -170 were not protected in virions, indicating that the tRNA 3 Lys anticodon loop did not interact with the NL4.3 A-rich loop (Fig. 2a). Taken together, our in situ probing experiments on NL4.3 RNA indicated that this RNA underwent little, if any, structural rearrangement during formation of the initiation complex of reverse transcription. In addition, they did not reveal any interaction between tRNA 3 Lys and the vRNA outside the PBS.
In order to check if the same complex was formed in vitro in the absence of cofactors, we probed NL4.3 and HXB2 RNA before and after tRNA 3 Lys annealing (Fig. 2, b and c). For these experiments, the potassium chloride concentration was increased to 300 mM, in order to stabilize and more easily detect weak tertiary interactions (11). In vitro, a strong pause was Lys complex. The tRNA 3 Lys primer is in red, except for the sequences interacting with the PBS and the A-rich loop. Helices 1 and 2, and the junction between the PBS and helix 2 are highlighted in yellow, orange, and purple, respectively. The corresponding sequences are also highlighted in panels a and c. c, secondary structure of nucleotides 125-223 of HIV-1 NL4.3 RNA. The sequence is colored as in panel a. The sequence differences between NL4.3 and HXB2 are indicated in purple.
observed in the PBS of NL4.3 and HXB2 RNA when tRNA 3 Lys was annealed at this site (Fig. 2, b and c). Otherwise, in vitro DMS modification patterns were remarkably similar to those obtained in situ (compare Fig. 2a to Fig. 2, b and c). In vitro, tRNA 3 Lys annealing protected the PBS nucleotides and A-147 in both templates. In addition, the reactivity of C-179 increased upon primer annealing to HXB2 RNA. However, formation of the initiation complex did not induce extensive rearrangements of NL4.3 and HXB2 RNA (Fig. 2, b and c). In addition, the A-rich loop of both RNAs (nucleotides 168 -170) remained accessible to DMS after tRNA 3 Lys annealing, indicating that it did not stably interact with the anticodon loop. The one-nucleotide shift observed around position 215 in the modification patterns of NL4.3 and HXB2 RNA (Fig. 2, b and c) was due to sequence differences between these HIV-1 isolates (Fig. 1c).
In Situ and in Vitro Structural Probing of MAL PBS Domain-Next, we performed in situ structural probing of the MAL initiation complex. However, MAL replicates poorly in most cell types (41), and we were unable to obtain enough material for in situ DMS probing by using the MAL molecular clone. Therefore, we substituted nucleotides 61-274 (including the complete PBS domain) of MAL for the homologous region of the NL4.3 molecular clone. The resulting chimerical NL4.3-MAL clone was infectious and replication kinetics of NL4.3 and NL4.3-MAL viruses were similar (data not shown). This clone thus allowed us to probe the MAL PBS domain in cells and in virions.
A very strong RT stop was observed in the PBS of NL4.3-MAL RNA extracted from cells, but not from virions. As a result, the intensity of the bands located 5Ј of this stop strongly decreased (Fig. 3a). In order to correct this bias, we overexposed this part of the gels to equalize the band intensities (Fig.  3b). In sharp contrast with NL4.3 (Fig. 2a), the DMS modification patterns of NL4.3-MAL RNA were very different in infected cells and in virions (Fig. 3, a and b). Remarkably, these patterns were very similar to those obtained prior and after in vitro tRNA 3 Lys annealing (Fig. 3c). The nucleotides in the PBS, as well as A-122, A-134, A-138, A-147, C-158, A-165, A-166, A-211, A-212, and C-213 were more reactive in cells and prior to in vitro tRNA 3 Lys annealing, than in virions and after in vitro tRNA 3 Lys annealing (Fig. 3). Conversely, the reactivity of nucleotides A-130, A-131, A-150, C-151, A-157, A-206, A-207, and A-218 increased in viral particles and after tRNA 3 Lys annealing in vitro (Fig. 3). Hence, our probing data indicated that extensive structural rearrangements took place during formation of the MAL initiation complex of reverse transcription in situ and in vitro. Indeed, as previously shown (9) for the in vitro data, the modification patterns of NL4.3-MAL in cells and virions are compatible with the secondary structure models depicted in Fig. 1, a and b, respectively. The only significant difference between in vitro and in situ data concerned protection of adenines 164 -167, which resulted from the interaction between the viral A-rich loop and the anticodon loop of tRNA 3 Lys (Fig. 1b). This protection was less pronounced in situ than in vitro (Fig. 3, b and c). Several factors might explain this difference. First, the higher potassium chloride concentration and the lower temperature used for in vitro probing experiments are expected to stabilize this interaction (11). Second, nucleocapsid protein, which covers the initiation complex of reverse transcription in virions, might destabilize this labile interaction (42). Quantification indicated that the reactivity of adenines 164 and 167 was 25 and 20% lower in virions than in cells, respectively, while the reactivity of adenines 165 and 166 was 35% lower. Since, adenines 165 and 166 were more strongly protected than A-164 and A-167, both ex vivo and in vitro, the same loop-loop interaction most likely took place in the MAL initiation complex in vitro and inside the viral particles.
Initiation of Reverse Transcription Using tRNA 3 Lys or ORN 3 Lys as Primers-In order to correlate our structural data with the efficiency of reverse transcription of the MAL and HXB2 isolates, we analyzed the rate of primer extension during (Ϫ) strand strong stop DNA synthesis. With both templates, we compared the extension rate of tRNA 3 Lys and ORN 3 Lys , an oligoribonucleotide corresponding to the 18 3Ј-terminal nucleotides of tRNA 3 Lys that does not make any intermolecular interaction outside the PBS.
Little difference was observed between the tRNA 3 Lys -primed reverse transcription of MAL and HXB2 RNA. However, extension of ORN 3 Lys was slower on MAL than on HXB2 RNA (Fig. 4,  a and b). With the MAL template, 50% of ORN 3 Lys were extended in ϳ7.5 min, whereas Ͼ 50% of tRNA 3 Lys were extended in 15 s (Fig. 4b). With the HXB2 RNA, half of the tRNA 3 Lys and ORN 3 Lys primers were extended in ϳ60 and ϳ80 s, respectively. Initiation of reverse transcription of HXB2 and MAL RNA followed bi-exponential kinetics, as previously observed for MAL (13,43). Previous work showed that the fast reaction corresponds to primer extension inside a productive primertemplate:RT complex, whereas the slow reaction is limited by a conformational change of a non-productive primer-template conformation (43). For each template, kinetics of primer extension was analyzed by global fitting (see "Experimental Procedures"). These fits indicated that, for each template, a mixture of "fast" and "slow" initiation complexes co-existed (Fig. 4, b  and c).
Substituting ORN 3 Lys for tRNA 3 Lys did not changed the extension rate of these particular complexes but affected the ratio of fast and slow initiation complexes, as reflected by changes in the amplitudes of the fast and slow reactions (A fast and A slow ) (Fig. 4c). With the MAL RNA as template, tRNA 3 Lys formed 67% of fast initiation complex, but only 23% of fast complex was formed when the primer was ORN 3 Lys . The situation was dramatically different when the template was HXB2 RNA, since the fraction of fast complex was only slightly reduced (from 62 to 52%) when ORN 3 Lys was substituted for tRNA 3 Lys . Interestingly, tRNA 3 Lys extension was faster when this primer was annealed to HIV-1 MAL RNA, as compared with HXB2 RNA (Fig. 4b), as reflected by the 5-fold difference in k fast (Fig. 4c). On the other hand, only a 2-fold difference in k slow was measured between MAL and HXB2 templates.
The kinetics of the initiation of reverse transcription of MAL and HXB2 RNA were fully consistent with our structural study in showing major differences between these isolates. In HIV-1 MAL, replacing tRNA 3 Lys by ORN 3 Lys had a strong negative effect on the initiation of reverse transcription. At the opposite, substituting ORN 3 Lys for tRNA 3 Lys had only minor consequences on HXB2 reverse transcription, indicating that intermolecular interactions between tRNA 3 Lys and this template were not required outside of the PBS to promote efficient initiation. In addition, the 5-fold difference in k fast for the initiation complexes of these two isolates suggested that they adopt different structures.
Formation of the fast initiation complex was favored by interactions between the primer and MAL RNA outside the PBS, but such interactions were not required with HXB2 RNA. These results suggest that these "additional" interactions were required to counteract a negative effect (such as steric hindrance) that originated from the MAL RNA, rather than from the primer.
Initiation of Reverse Transcription of the HXB2 Mutants Using tRNA His as Primer-The absence of interaction between the A-rich loop of the WT HXB2 RNA and the anticodon loop of tRNA 3 Lys seems to contradict previous ex vivo and in vitro studies on HXB2 mutants that use tRNA His as their reverse transcription primer. In these mutants, an interaction between a mutated A-rich loop and the anticodon loop of tRNA His was required for efficient replication (18 -20) and initiation of reverse transcription (31). Indeed, the results we obtained with the WT HXB2 RNA suggest that the mutant HXB2 viruses designed to utilize tRNA His as primer do not reflect the WT situation.
To solve this contradiction, we studied the initiation of reverse transcription of RNAs derived from the HXB2 His and HXB2 His-AC-GAC molecular clones of HIV-1. HXB2 His is identical to WT HXB2, except that the PBS is complementary to the 18 nucleotides at the 3Ј-end of tRNA His . HXB2 His-AC-GAC contains four additional substitutions introduced by sitedirected mutagenesis into the A-rich loop, and three point mutations selected during replication in tissue culture. The selected mutations are crucial for efficient viral replication (18 -20) and for stable interaction between the mutated A-rich loop and the tRNA His anticodon loop (31).
We followed the kinetics of the initiation of reverse transcription of HXB2 His and HXB2 His-AC-GAC RNA using either tRNA His or ORN His , an 18-mer RNA corresponding to the 3Јend of tRNA His , as primer (Fig. 5a). We performed global fitting of the experimental data with a bi-exponential function as above (Fig. 5a). Fitting indicated that the extension rates of the fast and slow complexes (k fast and k slow ) were the same for the two mutant HXB2 templates (Fig. 5b).
With tRNA His as primer, HXB2 His-AC-GAC RNA, which interacts with the tRNA His anticodon loop (31), was reversetranscribed efficiently, while HXB2 His RNA was not (Fig. 5a). The tRNA His primer formed 82% of fast complex with HXB2 His-AC-GAC RNA, compared with 8% with HXB2 His RNA (Fig. 5b). However, HXB2 His RNA reverse transcription was efficiently primed by ORN His . Indeed, ORN His initiated reverse transcription of HXB2 His and HXB2 His-AC-GAC RNAs at very similar rates (Fig. 5b), and both templates formed Ͼ 80% of fast complex with this primer (Fig. 5b).
Formation of the fast complex required an interaction between the anticodon loop of tRNA His and the vRNA, but no interaction outside of the PBS was required when reverse transcription was initiated by ORN His . Thus, the interaction between the anticodon loop of tRNA His and the vRNA most likely counteracts a negative contribution originating from tRNA His , rather than from the template.
Sequence Alignments-Our experimental data showed that the MAL and NL4.3/HXB2 initiation complexes of reverse transcription are structurally and functionally different. However, MAL is a particular isolate because it is a complex recombinant of HIV-1 subtypes A, D, and I, and it contains a 23-nucleotide duplication, including part of the PBS and the downstream region (14). To check whether the MAL isolate is an oddity, we aligned the 538 sequences from the HIV-1 data base (www.hiv. lanl.gov) that contain the region corresponding to nucleotides 95-245 of the HXB2 genomic RNA (Fig. 6 and data not shown). We found 74 sequences (14%) containing an insertion identical to or very similar to the MAL insertion.
In addition, we found that the insertion 3Ј to the PBS strongly correlated with substitutions 5Ј to the PBS (Fig. 6) In the isolates with an insertion, substitutions 5Ј to the PBS stabilize the hairpin structure with the A-rich sequence in the The continuous and dashed curves correspond to the best fits of the MAL and HXB2 experimental data, respectively, with the function: f ext ϭ A tot ϩA fast ⅐e Ϫkfast•t ϩ A slow ⅐e Ϫkslow•t , where f ext is the fraction of extended primer, A tot , A fast , and A slow correspond to the total amplitude, the fast component, and slow component of the reaction, respectively, and k fast and k slow correspond to the rate constants of these components. c, global fitting of the experimental data. The amplitudes of the fast and slow reactions are represented by gray and white bars, respectively. The values of the rate constants are given above the respective graphs. The error bars correspond to a 95% confidence interval. apical loop, as observed in MAL (Fig. 1a). In the isolates without insertion, the hairpin structure with the A-rich sequence in an internal loop is favored, as observed in HXB2 and NL4.3 (Fig. 1c). Interestingly, stabilizing the "MAL-like structure" (with the A-stretch in the apical loop) in HXB2 has been shown to decrease tRNA 3 Lys annealing and the initiation of reverse transcription (44,45). The MAL and HXB2 sequences 5Ј to the PBS were representative of the isolates with and without an insertion 3Ј to the PBS, respectively ( Fig. 6 and Table I). The only divergences were found at nucleotides 159 and 161 of HIV-1 MAL, which were U-159 and A-161 in most isolates with an insertion (Fig. 6 and Table I). Mfold (46) predicts that in these isolates G-160 and A-161 base pair with C-170 and U-169, and the five adenines of the A-tract are located in the apical loop (Fig. 1a). Thus, it is likely that the MAL and HXB2/NL4.3 isolates are structurally and functionally representative of the HIV-1 isolates with and without an insertion downstream of the PBS, respectively.
Interestingly, the insertion 3Ј to the PBS is not evenly distributed among the HIV-1 groups, subgroups and circulating recombinant forms (CRFs) (36). The insertion is only found in subgroups G (100%) and A (22%), especially in sub-subgroup A2 (100%), of the major (M) group of HIV-1, and in CRFs (Fig.  6). Indeed, the insertion is found in 100% of the CRFs 01_AE, 02_AG, 04_cpx, and 06_cpx that are actively spreading in Africa and Asia. DISCUSSION Previous studies of the HIV-1 initiation of reverse transcription provided divergent results regarding possible interactions between tRNA 3 Lys and the vRNA outside the PBS and their functional roles (9 -13, 15-20, 24, 25). These divergences may originate from different in vitro experimental conditions, the lack of direct ex vivo structural information, and the use of different HIV-1 isolates. To solve these discrepancies, we developed in situ probing of the HIV-1 genomic RNA and performed a detailed comparison of the MAL and NL4.3 isolates. The experiments we reported here represent the first ex vivo probing of HIV-1 RNA. The DMS modification profiles of the genomic RNA in cells and in virions appeared remarkably similar to those obtained prior and after in vitro tRNA 3 Lys annealing, respectively (Figs. 2 and 3). This result is surprising given that in vitro probing was performed in the absence of any protein. In addition, less than 1% of viruses are able to complete their life cycle. However, the majority of virions have been shown to be able to initiate reverse transcription (47). The only significant difference between in vitro and ex vivo DMS probing concerns weak tertiary interactions, which appeared easier to detect in vitro when using a higher ionic strength.
Our in vitro and in situ probing experiments revealed marked differences between the MAL and NL4.3/HXB2 PBS domains. While the MAL PBS domain undergoes very extensive structural rearrangements upon tRNA 3 Lys annealing, in agreement with previous in vitro studies (9 -11,13), the structure of the NL4.3/HXB2 PBS domain is only marginally affected. In addition, the A-rich loop of the MAL RNA is protected in the initiation complex, most likely by the anticodon loop of tRNA 3 Lys (9, 10), but the NL4.3/HXB2 A-rich loop is not. In line with these results, our in vitro functional studies indicate that interactions between tRNA 3 Lys and the vRNA are required outside the PBS for efficient initiation of reverse tran- scription of HIV-1 MAL RNA, in agreement with previous studies (12,13,39). However, such interactions are not required for efficient initiation of reverse transcription of WT HXB2 RNA (Fig. 4). This conclusion is in agreement with recent measurements of (Ϫ) strand strong stop DNA synthesis in vitro, using mutants of NL4.3 RNA (17).
Why such a difference between HIV-1 isolates? In the MAL vRNA/tRNA 3 Lys complex, the interaction between the anticodon and A-rich loops (i.e. helix 6C) and the three-nucleotide junction between helices 2 and 7F are crucial elements for efficient initiation of reverse transcription, and helices 2 and 1 also play important roles (13) (Fig. 1B). Helices 1 and 2 and the junction between helices 2 and 7F are directly recognized by RT, but helix 6C is not (11). Noteworthy, none of these elements exits in the MAL RNA prior to tRNA 3 Lys annealing (Fig. 1, A and B). At the opposite, helices 1 and 2 pre-exist in HXB2 (and NL4.3) RNA, and the junction between helices 2 and 7F is already single stranded (Fig. 1C). Therefore, we propose that the interaction between the anticodon and A-rich loops drives the rearrangement of the MAL RNA during tRNA 3 Lys annealing, generating the structural elements eventually recognized by RT. As these elements exist in HXB2 RNA prior to tRNA 3 Lys annealing, the interaction between the anticodon and A-rich loops is not required. It has been shown that deleting the A-stretch impaired the initiation of HXB2 reverse transcription (48). However, this effect might be due to RNA misfolding induced by the deletion, and it would be interesting to test nucleotide substitutions at these positions.
Even though MAL is a "particular" isolate, sequence analysis indicates that 14% of all HIV-1 isolates have an insertion 3Ј to the PBS and a MAL-like structure 5Ј to the PBS. Thus, these African and Asian isolates are likely to require an interaction between the anticodon and A-rich loops for efficient initiation of reverse transcription.
It is surprising that this interaction is not conserved in the WT HXB2 isolate, especially as it also plays a crucial role in mutants of HXB2 designed to use tRNA His as primer (18 -20, 49). Indeed, our study reveals that these mutants do not behave as the parent strain they are derived from. The kinetics of reverse transcription of these mutants (Fig. 5) suggests that the interaction between the vRNA and the tRNA His anticodon loop is required to compensate a negative contribution of this primer. Thus, our results suggest that all tRNAs are not equally suited to serve as reverse transcription primers, in agreement with a recent study comparing several non cognate primers (23).
Besides, our results indicate that the interaction between the vRNA and the primer tRNA anticodon loop fulfills different functions in WT MAL and in the HXB2 mutants using tRNA-His . In the former case, it counteracts a negative effect of the vRNA, rather than of the primer. This convergent solution to different problems highlights the adaptability of the HIV-1 initiation complex of reverse transcription.
The interaction involving the tRNA 3 Lys anticodon loop is not the only intermolecular interaction that has been proposed to take place in the HXB2 reverse transcription initiation complex. Berkhout and co-workers (24,25) identified the PAS-antiPAS interaction in vitro, in the absence of nucleocapsid protein. However, neither our DMS probing data, nor the similar kinetics of reverse transcription of HXB2 RNA observed with tRNA 3 Lys and ORN 3 Lys are in keeping with this proposal. One cannot exclude that the mutants designed to test this interaction do not reflect the WT situation (24,25,29), as reported above for the HXB2 mutants utilizing tRNA His as primer.
Alternatively, Iwatani et al. (17) recently proposed that parts of the anticodon stem and variable loop of tRNA 3 Lys interact with nucleotides 143-149 of NL4.3 RNA. Our in vitro data would suggest that this interaction does not take place in NL4.3. However Iwatani et al. (17) only observed this interaction in the presence of nucleocapsid protein. In addition, mutation of the proposed interaction had pronounced effects in a short template that could not form helix 1, but limited effects in a template forming this helix (17). Our in situ probing data only partially support the existence of this interaction. Therefore, additional experiments will be required to conclude about the existence of this interaction in the context of the full-length genomic RNA in vivo.
The present study revealed an unexpected variability of the HIV-1 initiation complex of reverse transcription in vitro and in vivo. We cannot exclude that detailed studies of other HIV-1 isolates will expand the repertoire of structures allowing efficient initiation of reverse transcription.