Selective packaging of HIV-1 RNA genome is guided by the stability of 5′ untranslated region polyA stem

Significance

HIV-1 must select and package its RNA genome from an abundant pool of cellular RNAs. To understand this essential replication step, we studied two nearly identical HIV-1 RNAs that are differentially encapsidated. HIV-1 RNA with one 5′ guanosine (1G RNA) is enriched in virions, whereas RNA with three 5′ guanosines (3G RNA) is largely excluded. We observed that 1G RNA, but not 3G RNA, mainly folds into structures that expose elements important for RNA:RNA and RNA:Gag interactions. We also identified mutants in which 1G and 3G RNAs fold into similar structures, resulting in efficient packaging of 3G RNA. Thus, HIV-1 selects its viral genome based on its capacity to adopt structures that facilitate RNA dimerization and Gag binding.

Abstract

To generate infectious virus, HIV-1 must package two copies of its full-length RNA into particles. HIV-1 transcription initiates from multiple, neighboring sites, generating RNA species that only differ by a few nucleotides at the 5′ end, including those with one (1G) or three (3G) 5′ guanosines. Strikingly, 1G RNA is preferentially packaged into virions over 3G RNA. We investigated how HIV-1 distinguishes between these nearly identical RNAs using in-gel chemical probing combined with recently developed computational tools for determining RNA conformational ensembles, as well as cell-based assays to quantify the efficiency of RNA packaging into viral particles. We found that 1G and 3G RNAs fold into distinct structural ensembles. The 1G RNA, but not the 3G RNA, primarily adopts conformations with an intact polyA stem, exposed dimerization initiation site, and multiple, unpaired guanosines known to mediate Gag binding. Furthermore, we identified mutants that exhibited altered genome selectivity and packaged 3G RNA efficiently. In these mutants, both 1G and 3G RNAs fold into similar conformational ensembles, such that they can no longer be distinguished. Our findings demonstrate that polyA stem stability guides RNA-packaging selectivity. These studies also uncover the mechanism by which HIV-1 selects its genome for packaging: 1G RNA is preferentially packaged because it exposes structural elements that promote RNA dimerization and Gag binding.

To infect a new cell, HIV-1 must assemble infectious particles that contain its genome. Like other retroviruses, HIV-1 packages two copies of its full-length RNA genome (referred to as HIV-1 RNA hereafter) as a dimer into viral particles (1). The vast majority of particles (>95%) contain HIV-1 RNA, although viral RNA represents only a small portion of RNA in the cell (2). Thus, RNA packaging is a highly efficient and specific process. The mechanism of genome selection is complex and not well understood; however, RNA dimerization and Gag:RNA interactions are known to be involved (3–5). HIV-1 particle assembly is driven by Gag, which is translated as a polyprotein consisting of matrix, capsid, nucleocapsid (NC), and p6 domains and two spacer peptides: SP1 and SP2 (6). Soon after or concomitant with virus assembly and budding, Gag is processed to yield mature proteins (6). Of the Gag domains, NC is known to play a particularly important role in RNA binding and selective genome packaging (4, 6–9). In addition, the 5′ untranslated region (UTR) of the genome is essential for genome packaging, as it participates in RNA:Gag interactions and RNA:RNA interactions that lead to genome dimerization (9–16). Heterologous RNAs containing the HIV-1 5′ UTR and a portion of the gag gene can be efficiently packaged into viral particles; thus, these sequences are necessary and sufficient for HIV-1 genome packaging (17).

The HIV-1 5′ UTR has essential, regulatory roles in multiple steps of viral replication, including reverse transcription, gene expression, and genome packaging (18). The first ∼100 nucleotides (nt) of the HIV-1 5′ UTR represent the repeat (R) region and include the transactivation response element (TAR) and a stem–loop structure termed polyA, which contains a polyadenylation signal not actively used because of its proximity to the transcription start site (18). In addition to their critical roles in regulating HIV-1 expression, these elements have also been shown to affect 5′ UTR RNA structure and genome packaging (19–21). Other elements known to promote selective packaging include three stem–loop elements (SL1 to SL3) located further downstream in the 5′ UTR (22–26). Multiple Gag and NC binding sites have been identified within or near these stem loops (9, 12, 13, 15, 24, 27–29). Additionally, the loop of SL1 contains the dimerization initiation site (DIS), a 6-nt palindromic sequence that mediates RNA:RNA dimerization (30, 31).

Recent reports have shown that HIV-1 uses several neighboring transcription start sites to generate multiple RNA species that differ by only a few nt (32, 33). In most HIV-1 sequences, the R region starts with three guanosines, and transcription results in RNA species that contain one, two, or three guanosines, referred to as 1G, 2G, or 3G RNA, respectively. Additionally, HIV-1 RNA has a 5′-5′ triphosphate-linked 7-methylguanosine cap (34). Intriguingly, the number of 5′ guanosines affects genome selection; although 3G RNA is the most abundant species in cells, 1G RNA is the most abundant in virions (32, 33, 35). Thus, HIV-1 distinguishes between minor differences in RNA species and preferentially packages 1G RNA. These RNA species have nearly identical primary sequences; thus, they are most likely distinguished by differences in RNA structure. A model has been proposed that 1G RNA and 3G RNA fold into distinct structures that differ in their ability to dimerize and be packaged into viral particles (33). The 1G RNA folds into a conformation in which the DIS is exposed, thereby facilitating dimerization and packaging. In contrast, 3G RNA folds into a conformation in which the DIS engages in long-range, base-pairing interactions with a region in U5, preventing dimerization. Therefore, 3G RNA remains monomeric and is not selected for packaging. In 3G RNA, the extra guanosines at the 5′ end of the RNA were proposed to participate in the formation of additional base pairs, which destabilize the base of the polyA stem but stabilize the U5:DIS long-range interaction (33). In a later study, the U5:DIS interaction in 3G RNA was suggested to extend into the SL1 and R regions as well, further stabilizing this poorly packaged conformation (36).

In a different study using short RNAs that mimic the TAR-polyA interface region, it was proposed that the 5′ guanosine content affects the TAR-polyA structure (37). TAR and polyA were suggested to be coaxially stacked in 1G, but not 3G, RNA. Stacking was proposed to contribute to the greater stability of TAR and polyA in 1G RNA compared with 3G RNA, which may influence other parts of the 5′ UTR structure and affect genome packaging.

In this study, we sought to delineate the mechanistic basis of HIV-1 genome selection. Many large, functional RNAs, such as bacterial riboswitches and some 3′ and 5′ UTRs of eukaryotic messenger RNAs (mRNAs), cannot be described by a single three-dimensional (3D) structure or conformation. Instead, they have evolved to adopt two or more functional conformations and, in some cases, adopt ensembles of many different conformations (38, 39). We hypothesized that the HIV-1 5′ UTR is capable of adopting many different conformations and that the 5′ guanosine content of the transcript modulates the ensemble of RNA structures in a manner that favors or disfavors packaging. To understand how the ensemble of RNA structures adopted by the HIV-1 5′ UTR is modulated by transcription start sites, we used recently developed computational tools that, when combined with experimental, structure-probing data, allow secondary structure prediction of ensembles of large RNAs (40). We combined in-gel chemical probing and cell-based virology experiments to study the effects of transcription start site usage on RNA structure and selective genome packaging. These studies demonstrate the intricate regulation of HIV-1 RNA function and provide insights into the genome selection mechanism.

Results

HIV-1 Generates Multiple Species of Full-length RNAs Using Heterogenous Transcription Start Sites.

To examine the species of HIV-1 full-length RNA generated from proviruses in human cells, we used a previously described, near full-length, NL4-3-derived construct called H0 (41) (Fig. 1A). Construct H0 contains intact long terminal repeats (LTRs), 5′ UTR, gag-pol, tat, and rev and has inactivating deletions in vif, vpr, vpu, and env. Additionally, H0 contains a mouse heat stable antigen gene (hsa) in the nef-reading frame. To generate infected cells, we first transfected H0 plasmids into 293T cells along with a helper plasmid expressing vesicular stomatitis virus G protein, harvested viruses, and infected, fresh 293T cells. By monitoring the HSA marker expression level using flow cytometry, we generated cell pools that were infected at a multiplicity of infection (MOI) between 0.5 and 1. These cells produce noninfectious viral particles because H0 encodes a functional gag/gag-pol gene but has an inactivating deletion in env. We isolated RNAs from cells and from viral particles and performed 5′ rapid amplification of cDNA ends (RACE) to identify the transcription start sites of HIV-1 full-length RNAs (Fig. 1B).

Fig. 1.

System used to study HIV-1 transcription start sites. (A) General structures of NL4-3–based near full-length HIV-1 construct H0. Asterisks denote mutations in the 5′ UTR. (B) Experimental outline. To ensure HIV-1 full-length RNAs were examined, a gag-specific primer was used for cDNA synthesis. (C) HIV-1 RNA species with different transcription start sites detected in this analysis. The first of the three guanosines in the R region is designated as nt 1; this residue corresponds to nt 454 of NL4-3. Other transcripts are rare RNA species such as those with internal deletions or lacking caps (generally less than 2% of the total RNA detected). (D) Transcription start sites of HIV-1 RNA from cells or viral particles, as determined by 5′ RACE. The results represent the total of six independent experiments, with sequencing data for 241 and 201 clones for cellular and viral RNA, respectively.

Consistent with published results, most of the full-length HIV-1 transcripts in cells corresponded to 3G RNA, followed by 1G RNA, 2G RNA, and other transcripts that initiated further downstream, such as from +4 (T4), +5 (C5), and +6 (T6) positions (Fig. 1 C and D). In contrast, most of the transcripts in viral particles corresponded to 1G RNA, followed by 3G RNA, other RNAs, and 2G RNA (Fig. 1D). Compared with cells, viral particles contained a significantly higher proportion of 1G RNA and a significantly lower proportion of 3G RNA (P < 0.00001 in both cases; χ² test); these findings are consistent with previous reports (32, 33). The 5′ UTR of HIV-1 RNA forms complex structures and is critical for RNA genome packaging. Thus, we hypothesize that 1G RNA and 3G RNA adopt distinct structures (or ensembles of structures), leading to the preferential packaging of 1G RNA during virus assembly.

In-gel Chemical Probing Reveals Structural Ensemble Differences between 1G and 3G Wild-type 5′ UTRs.

To probe the conformations adopted by 1G and 3G RNAs, we prepared 400- and 402-nt wild-type (WT) 5′ UTRs encoding either two or four 5′ guanosines, which have previously been shown to accurately mimic ^cap1G and ^cap3G RNAs, respectively (33, 36). To ensure that precise 5′ ends were obtained, a hammerhead ribozyme (HHR), designed to cleave between the HHR sequence and the 5′ terminal G during in vitro transcription, was encoded upstream of the 5′ UTR, as described in the Materials and Methods; these RNAs will be referred to as 1G and 3G WT throughout this paper. Native polyacrylamide gel electrophoresis (PAGE) analysis was performed after refolding the 1G and 3G WT 5′ UTRs under the same conditions, revealing the presence of distinct, conformational ensembles for each RNA (Fig. 2A). Three monomer bands (M_bot, M_mid, and M_top) were observed for both 1G and 3G WT. However, in the case of 1G WT, most of the RNA was dimeric with two distinct dimer bands (D_bot and D_top) observed.

Fig. 2.

Native gel, Rsample analysis results, and major centroid structures for WT 5′-UTR RNAs. (A) Representative nondenaturing 6% polyacrylamide gel of the WT 1G (400 nt) and 3G (402 nt) 5′-UTRs used for in-gel SHAPE probing. Each individual band, which was cut out and probed separately, is indicated. Bands were visualized by ethidium bromide staining. (B) Results of Rsample analysis of the WT 5′-UTRs using in-gel SHAPE data (average of three trials) as constraints. The height of the bars represents the population of each cluster within the structural ensembles. Each color indicates a unique cluster. (C and D) Representative 1G exposed.1 (C) and 3G sequestered.1 (D) conformations. Lines along the RNA backbone represent the location of the polyA loop (black line), the primer binding site (PBS; blue line), and the DIS (red line). Gray nt indicate sequences that could not be analyzed. The 5′ guanosines are shown in red.

To determine the secondary structure of each distinct RNA species observed on the native gel, we performed in-gel chemical probing using selective 2'-hydroxyl acylation analyzed by primer extension (SHAPE) (42). The eight individual monomeric and dimeric RNA bands were excised from native gels, exposed to SHAPE reagent, extracted from gel pieces, and processed as described in the Materials and Methods. The normalized SHAPE reactivities of the most abundant 1G D_top and 3G M_top bands were first compared. Overall, SHAPE reactivity profiles were similar, with most differences located in the polyA and SL1 regions (SI Appendix, Fig. S1). To obtain structural ensemble information for all the probed bands (Fig. 2A), we applied the Rsample tool, using the SHAPE reactivity data as constraints to predict RNA conformational clusters (40). We also determined the abundance of each cluster in the population and a centroid (representative) structure for each cluster (40). For dimer bands, high-SHAPE reactivity constraints were applied to the DIS sequence, which ensured DIS exposure because of known kissing loop interactions between dimeric RNAs (Materials and Methods). The results of the Rsample analysis for 1G and 3G WT RNAs are summarized in Fig. 2B. The Rsample-derived clusters could be classified into one of three groups: 1) “exposed,” wherein the polyA and SL1 hairpins are both present; 2) “DIS only,” wherein the SL1 hairpin is present but polyA is refolded; and 3) “sequestered,” wherein both polyA and SL1 hairpins are refolded. The major 1G dimer band (1G, D_top) was predicted to adopt three distinct, conformational clusters, all of which had intact polyA and SL1 hairpins. In the case of the minor 1G dimer band (1G, D_bot), only ∼40% of predicted structures adopted the exposed conformation, while ∼60% adopted a DIS-only conformation. In contrast, all 1G and 3G monomer bands adopted sequestered conformations lacking intact polyA and SL1 hairpins.

The centroid secondary structure models of the most abundant 1G and 3G conformational clusters (1G exposed.1 and 3G sequestered.1) are shown in Fig. 2 C and D. All other centroid secondary structure models are shown in SI Appendix, Fig. S2, and a complete list of all conformational clusters can be found in SI Appendix, Tables S1 and S2. The major differences between exposed and sequestered conformations are as follows: Exposed conformations have intact polyA and SL1 hairpin structures with an exposed DIS loop, consistent with these conformations being adopted only by dimeric RNA bands. Sequestered conformations do not contain intact polyA or SL1 hairpins because of alternate, base-pairing interactions and were only adopted by monomeric RNAs. As a result, these conformations exhibit sequestered DIS loops. Overall, we observed clear differences in the structural ensembles adopted by 1G and 3G WT 5′ UTR RNAs.

Preferential 1G RNA Packaging Is Not Determined by U5:DIS Interaction.

A previously proposed model suggests that genome selection is based on the different stabilities of the U5:DIS interactions between 1G and 3G RNA (33). According to this model, in 1G RNA, the U5:DIS interaction is unstable, thereby favoring the folding of 1G RNA into a conformation that promotes packaging. In 3G RNA, the guanosine at the 5′ end can base pair with the cytosine (C59) at the 5′ end of the polyA stem (Fig. 3A). This base pairing would release the guanosine (G105) at the 3′ end of the polyA stem, allowing it to base pair with the DIS and stabilize the U5:DIS interaction, thereby preventing RNA dimerization and genome packaging. More recently, it was proposed that the U5:DIS interaction extends into the polyA and SL1 stems, forming a longer, base-paired region (36) (shown in SI Appendix, Fig. S3A). However, the proposed U5:DIS interaction was not observed in any of the major clusters revealed by our structural ensemble analysis (Fig. 2 and SI Appendix, Fig. S2).

Fig. 3.

Examining the impact of the U5:DIS interaction on preferential 1G RNA packaging. (A) The U5:DIS interactions in 1G and 3G WT RNA as previously proposed (33). (B) 5′ UTR mutants examined. Mutated bases are shown in red; potential U5:DIS base-pairings are shown in yellow highlight. (C) Summary of 5′ RACE results. WT results from Fig. 1D are shown for comparison. Each mutant was tested in at least five independent experiments, and results are based on sequencing the following numbers of clones: 218 (cell, LS), 179 (virus, LS), 183 (cell, SS), 171 (virus, SS), 213 (AA, cell), 170 (AA, virus), 157 (GAGA, cell), and 141 (GAGA, virus).

To better understand the genome selection mechanism, we generated two mutants in which 1G and 3G RNAs have the same U5:DIS interaction: lower-stem swap (LS) and stem swap (SS) (Fig. 3B). In the LS mutant, we switched the first two and last two nt of the polyA stem to maintain the stability of the polyA stem but prevent base pairing between DIS and the last nt of the polyA stem in 3G RNA. The SS mutant is similar to LS but with six nt on each side of the stem switched instead of two. In both mutants, G105 is mutated to cytosine to block potential base pairing with the DIS; thus, the U5:DIS interaction in 1G and 3G RNA should have the same stability. If U5:DIS stability is a major determinant of selective 1G RNA packaging, these mutants should lose the preferential packaging of 1G RNA. In addition, in the SS mutant, the proposed extended base pairing between polyA and SL1 should be disrupted (SI Appendix, Fig. S3B), which should lead to more efficient packaging of 3G RNA. However, using 5′ RACE, we found that both LS and SS mutants maintained preferential packaging of 1G RNA into viral particles (Fig. 3C). Compared with cells, viral particles contained significantly more 1G RNA and less 3G RNA (P < 0.00001 in all cases, χ² test). These results indicate that the differential packaging of 1G and 3G RNAs is not determined by the stability of the U5:DIS interaction.

We next generated two mutants, AA and GAGA, in which the proposed U5:DIS interaction is destabilized by eliminating two or three base pairs, respectively (Fig. 3B). Our 5′ RACE results showed that these mutants maintained preferential 1G RNA packaging (Fig. 3C), with more 1G RNA and less 3G RNA in virions than in cells (P < 0.00001 in all cases, χ² test). Therefore, these results demonstrate that the previously proposed U5:DIS base pairing does not contribute to the differential packaging of 1G and 3G RNAs.

Identification of 5′ UTR Mutants with Enhanced 3G RNA Packaging.

During our studies, we identified two classes of mutants that package 3G RNA efficiently. The first class of mutants was designed to destabilize the polyA hairpin. The first and last two nt of the polyA stem are predicted to form C:G and A:U base pairs in 1G WT RNA. In the AC mutant, we changed the last two nt to AC, and in the GU mutant, we changed the first two nt to GU (Fig. 4A); thus, in both mutants, the base pairing at the base of the polyA stem is disrupted. For both mutants, we found that the level of 3G RNA in cells was not significantly different from that in virus particles (Fig. 4B, P = 0.45 and 0.44 for AC and GU mutants, respectively; χ² test). Thus, both mutants efficiently packaged 3G RNA. Furthermore, the LS mutant described in Fig. 3 contains both the AC and GU changes and preferentially packages 1G RNA; thus, it is not the change of the nt sequence but the loss of base pairing at the bottom of the polyA stem that altered the packaging preference.

Fig. 4.

Identification of mutants with enhanced 3G RNA packaging. (A) Design of 5′ UTR mutants. (B) Summary of 5′ RACE results. WT results from Fig. 1D are shown for comparison. Each mutant was tested in at least five independent experiments, and results are based on sequencing the following numbers of clones: 212 (AC, cell), 188 (AC virus), 207 (GU, cell), 192 (GU, virus), 198 (addCC, cell), 235 (addCC, virus), 180 (addAA, cell), 177 (addAA, virus), 175 (addGG, cell), 181 (addGG, virus), 184 (addUU, cell), and 153 (addUU, virus).

We identified a second class of mutants that lack 1G RNA–packaging selectivity; these mutants contained 2-nt insertions between the TAR and polyA stems and were designed to insulate (or separate) these two elements (Fig. 4A). We first examined the addCC mutant, in which two cytosines were inserted between the TAR and polyA stem, and found that the proportion of 3G RNA in virus particles was actually higher than in cells (Fig. 4B, P < 0.00001, χ² test). One possibility is that the inserted cytosines can base pair with the capG and the first guanosine of the 3G transcript, stabilizing the TAR/polyA region and promoting efficient packaging of 3G RNA. If so, then efficient packaging of the 3G RNA would also occur in the presence of two additional uridines between the TAR and polyA stems but not upon insertion of two adenosines or guanosines. To test this possibility, we generated addUU, addAA, and addGG mutants and examined the transcription start sites of HIV-1 RNA in cells and virus particles. We found that the proportions of 3G RNA were not significantly different between cells and viral particles in all three mutants (P = 0.06, P = 0.79, and P = 0.14 for addUU, addAA, and addGG mutants, respectively, χ² test). Therefore, the insertion of two nt between the TAR and polyA stems leads to efficient packaging of 3G RNA regardless of whether these nt can base pair with guanosines at the 5′ end of the transcripts.

The 5′ UTR Mutants Disrupt 1G RNA–Packaging Selectivity through Distinct Mechanisms.

The results described in Fig. 4 revealed the distributions of HIV-1 RNA species but not the efficiency of mutant RNA encapsidation into viral particles. To establish whether the enhanced ability to package 3G RNA led to a general defect in packaging, we determined the packaging efficiency of each mutant in competition with WT RNA. For this purpose, we used a construct, Cdis-WT-B7, to express the WT RNA; this construct is nearly identical to H0 but contains the typical subtype C DIS (GTGCAC) and a B7 marker gene in the nef-reading frame (Fig. 5A). The H0 construct (Fig. 1A, WT or mutant) contains the typical subtype B DIS (GCGCGC) and encodes an hsa marker gene. These two viruses were introduced into cells by sequential infection at low MOI (∼0.1), and dually infected cells were enriched by multiple rounds of cell sorting until most (∼94%) of the cells expressed both B7 and HSA markers. Both viruses encode functional gag/gag-pol, such that infected cells produce virus particles containing RNA. We isolated RNA from cells and virus particles and determined the proportion of full-length RNA derived from each virus using RT-PCR sequencing of the polymorphic DIS region, as previously described (43). Genome-packaging efficiencies were determined by comparing the proportion of H0 RNA in virus particles with that in cells (percent of H0 in viral particles/percent of H0 in cells).

Fig. 5.

Determination of mutant RNA genome–packaging efficiencies by a competition assay. (A) General structure of HIV-1 construct Cdis-WT-B7. (B) Experimental outline for the competition assay. Asterisks denote mutations in LTRs; the polymorphic DIS sequences are shown. (C) RNA-packaging efficiency of H0-derived constructs (WT or mutant) when competing with WT RNA (from Cdis-WT-B7). Results shown represent the average from three experiments, with error bars representing SD. RNA-packaging efficiencies were compared by paired two-tailed t test. Asterisks indicate P values < 0.05.

When both viruses contain unmodified LTRs, RNAs from the two viruses were packaged equally well (Fig. 5C, 0.98; WT-WT). RNA derived from the GU virus was packaged less efficiently when competing with the WT RNA (0.66; WT-GU; P = 0.029 compared with WT-WT, paired two-tailed t test). In contrast, RNAs generated from the addCC and addGG mutants were packaged at higher or similar levels as WT control RNA (P = 0.045 and 0.276 for WT-addCC and WT-addGG, respectively; paired two-tailed t test). Thus, the mutant with a destabilized polyA stem (GU) had a packaging defect, whereas the mutants with 2-nt insertions between TAR and polyA stem (addCC and addGG) did not have packaging defects. These results suggest that these two types of mutants achieve enhanced packaging of 3G RNA through distinct mechanisms: In the GU mutant, 1G RNA likely lost its ability to be preferentially packaged, allowing 3G RNA to compete for packaging and leading to an overall decrease in genome-packaging efficiency. In contrast, in the addCC and addGG mutants, 3G RNA likely gained an enhanced ability to be packaged, thereby maintaining efficient genome packaging.

Dimerization Properties of 5′ UTR Mutants that Efficiently Package 3G RNA.

To further explore the mechanisms responsible for these phenotypes, we next compared the dimerization properties of 1G and 3G RNAs from mutants that efficiently packaged 3G RNA. As shown in Fig. 6A, the WT 1G 5′ UTR dimerized significantly more efficiently than the 3G 5′ UTR. Three distinct monomer species and two distinct dimeric species were observed on native gels (Fig. 6 A, Left). In some experiments, dimeric 3G RNA was not detected (Fig. 2A), and quantification of bands over four trials revealed that the 1G WT 5′ UTR displays a more than threefold increase in dimer formation relative to the 3G WT 5′ UTR (Fig. 6B). We next investigated the dimerization efficiency of the two classes of mutants that displayed efficient packaging of 3G RNA. In the GU and AC mutants, which were designed to disrupt the stability of the polyA stem, 1G and 3G 5′ UTRs dimerized poorly, similar to the 3G WT construct (Fig. 6 A, Middle). In contrast, the addNN mutants, which were designed to insulate TAR and polyA stem elements, 1G and 3G 5′ UTRs dimerized nearly as efficiently as the 1G WT construct (Fig. 6 A, Right). Quantification of all dimerization assays is shown in Fig. 6B. In contrast to the WT RNAs, for most of the mutants, no significant differences in the dimerization propensity were observed between the 1G and 3G RNAs. These results further support our hypothesis that these two classes of mutants disrupt 1G RNA–packaging selectivity through distinct mechanisms: GU and AC mutants lead to impaired dimerization (and hence packaging) of 1G RNA, whereas addNN mutants lead to enhanced dimerization and packaging of 3G RNA.

Fig. 6.

Native gel analysis of WT and mutant HIV-1 1G and 3G 5′ UTR RNAs. (A) Representative native polyacrylamide (6%) gel showing WT (Left), polyA stem (Middle), and addNN 5′ UTR (Right) variants. The number of 5′ guanosines is denoted above each lane, and the major monomer and dimer species are indicated at the left. (B) Individual band densities of the indicated monomer and dimer species were quantified, summed, and used to determine dimerization efficiency. Results shown represent the average of at least three experiments per construct. Dimerization efficiencies were compared between 1G and 3G RNAs by unpaired two-tailed t test. Asterisks indicate P values < 0.05.

Mutants that Efficiently Package 3G RNAs Have Similar 1G and 3G 5′ UTR Structural Ensembles.

We next investigated the impact of one member of each class of mutants with enhanced, 3G-packaging capability, the GU and addGG mutants, on the 1G and 3G 5′ UTR structural ensembles using in-gel SHAPE. Based on this analysis, all monomeric GU mutant 5′ UTR bands probed were predicted to adopt predominantly sequestered conformations regardless of the number of 5′ guanosines, with only minor differences (Fig. 7A). The predicted structure for the most abundant GU band (1G and 3G, M_top) was a unique, sequestered conformation (sequestered.4) not found in the other RNA species probed here (SI Appendix, Fig. S2E). The amount of GU dimer RNA that migrated as a distinct band was insufficient for SHAPE analysis. A complete list of all conformational clusters of the GU mutant is summarized in SI Appendix, Tables S1 and S2.

Fig. 7.

Results of in-gel SHAPE probing and Rsample analysis of mutant 5′-UTR RNAs. (A) Conformational ensembles for monomeric GU 5′-UTR RNAs. (B) Conformational ensembles for monomeric and dimeric addGG 5′-UTR RNAs. The average of three trials of in-gel SHAPE data were used in this analysis. The height of the bars represents the population of each cluster within the structural ensembles. Each color indicates a unique cluster.

We also probed one of the 2-nt insertion mutants, addGG, which exhibited improved dimerization and packaging of 3G RNA. All eight bands observed in the 1G and 3G WT RNAs were also observed in the addGG mutant (Fig. 6A), although the monomer species represented a smaller proportion of the RNA. The results of in-gel SHAPE probing and Rsample analysis are summarized in Fig. 7B. In the addGG mutant, the 1G and 3G 5′ UTRs adopted structural ensembles that were strikingly similar to each other, as well as to the 1G WT 5′ UTR. Dimeric 1G and 3G RNAs primarily adopted one of two exposed conformations, similar to the 1G WT dimers. In contrast, all monomeric addGG RNAs (1G and 3G) adopted sequestered conformations, similar to the 3G WT monomer bands (Fig. 2B).

Discussion

In this report, we probed the molecular mechanism by which HIV-1 selectively packages 1G RNA into particles over 3G RNA. We showed that the 5′ UTR of HIV-1 RNA folds into multiple conformations that vary in their ability to dimerize: 1G RNA primarily adopts DIS-exposed conformations that readily dimerize, whereas 3G RNA primarily folds into DIS-sequestered conformations that do not dimerize efficiently. We also examined the possible role of U5:DIS interactions, which were previously proposed to contribute to the selective packaging of 1G RNA (33). Our mutational studies showed that neither the stability nor the base pairing of the U5:DIS interaction affects 1G RNA packaging. However, we identified two other classes of mutants in which most of the virion RNA is 3G RNA: 1) mutants designed to disrupt the stability of the base of the polyA stem (GU and AC) and 2) mutants containing 2-nt insertions between TAR and polyA (addNN). By studying these two types of mutants, we gained insights into the mechanisms of genome selection and packaging.

HIV-1 genome packaging is mediated by RNA:Gag interactions. The NC domain of Gag plays a major role in genome selection. In a previous reverse foot-printing study, wherein NC-RNA binding was disrupted in virio using aldrithiol-2, seven high-affinity binding sites for the mature NC protein were identified in the HIV-1 5′ UTR (13) (SI Appendix, Fig. S4D). All seven NC binding sites contained exposed, single-stranded guanosines. Mutation of these guanosines has been shown to impair Gag binding (12, 28) and genome packaging (11). Mutating individual sites led to only mild defects in genome packaging, whereas mutating several sites led to synergistic defects in packaging that, in some cases, were quite severe (11). These findings showed that the packaging signal is comprised of multiple, redundant NC binding sites, although some sites are more important than others (11, 28). Given this, we examined whether these guanosines in the NC binding sites are exposed in the different 5′ UTR conformations predicted for 1G and 3G RNAs (SI Appendix, Fig. S4). In WT dimeric 1G RNA (DIS-exposed conformers), most of the NC binding sites are unpaired. However, in WT monomeric 1G or 3G RNA (DIS-sequestered conformers), most of the NC binding sites are paired (SI Appendix, Fig. S4A). Interestingly, for the GU mutant, most of the NC binding sites are paired in both monomeric 1G and 3G RNAs, consistent with the packaging defect observed for this mutant (SI Appendix, Fig. S4B). In contrast, for the addGG mutant, most of the NC binding sites are unpaired in both dimeric 1G and 3G RNAs (SI Appendix, Fig. S4C). This is consistent with our observation that the addGG mutant efficiently packages both 1G and 3G RNAs into viral particles. These results suggest that RNA conformation affects genome packaging by altering both the ability of the RNA to dimerize and the availability of unpaired guanosines to serve as Gag binding sites. Furthermore, RNA dimerization will effectively double the number of NC binding sites in the complex and provide an additive, if not synergistic, difference between dimeric and monomeric RNA.

Based on our studies, we propose a “dynamic zip and pack” model for HIV-1 genome selection (Fig. 8A). We hypothesize that the 5′ UTR of HIV-1 RNA is capable of folding into multiple conformations, with structural ensemble differences between 1G and 3G RNAs. During the folding process, the nt that comprise the polyA stem play a critical role in determining RNA structure. When a stable polyA stem is formed, the rest of the 5′ UTR is more likely to fold into conformations with exposed DIS and Gag binding sites; these features facilitate events required for genome packaging, including Gag binding and RNA dimerization. In contrast, when a polyA stem is not formed, these nt interact with other regions, and the 5′ UTR is more likely to fold into structures with sequestered DIS and Gag binding sites. RNAs that fold into these conformations are poor substrates for Gag binding and RNA dimerization and, therefore, are not packaged efficiently (Fig. 8A).

Fig. 8.

The “dynamic zip and pack” model for HIV-1 genome packaging. (A) Proposed impact of polyA stem stability on the folding of HIV-1 5′ UTR and genome packaging. (B) Proposed effects of polyA stem stability on the distributions of HIV-1 RNA in cells and viral particles. The predicted structures of the partial TAR and polyA stems are shown on the left. Blue and red open circles denote 3G and 1G RNA, respectively, folded into conformations that disfavor genome packaging. Blue and red filled circles denote 3G and 1G RNA, respectively, folded into conformations that favor genome packaging.

Previous small-angle, X-ray–scattering analysis revealed that the TAR/polyA domain formed an extended, dumbbell-like helical structure stabilized by coaxial stacking between the stems (44). This conclusion is supported by recent experimental and computational modeling studies in the context of the ^cap1G 5′ UTR (36, 37, 45). In contrast, experimental studies suggest that the coaxial stack is disrupted in ^cap3G RNA (36, 37). We hypothesize that the stable coaxial stacking interaction between TAR and polyA stems found in the 1G RNA increases the stability of the polyA stem and leads to RNA conformations with an exposed DIS and Gag binding sites. In the WT 3G 5′ UTR, the additional guanosines and the cap at the 5′ end of the RNA interfere not only with coaxial stacking between TAR and polyA but also disrupt the stability of the polyA stem (Fig. 8 B, Top). Conflicting conclusions have been reported on the role of the polyA stem–loop structure on HIV-1 RNA packaging. Some studies suggested that the polyA structure is important for genome packing, although the mechanistic basis was unknown (21, 46, 47). Other studies suggest that deletion of this RNA element has no effect on RNA packaging (27). Here, we definitively show that the stability of the polyA stem plays a critical role in selective 1G RNA packaging; destabilization of the polyA stem in 3G RNA leads to downstream RNA conformational changes that result in a sequestered DIS and few exposed Gag binding sites; consequently, 1G RNA is selectively packaged into HIV-1 virions.

Our model also explains the enhanced 3G packaging observed in the two classes of mutants that impact TAR/PolyA structure and stacking. In the GU mutant, the unpaired nt destabilize the polyA stem in both 1G and 3G RNAs (Fig. 8 B, Middle), such that both RNAs primarily fold into conformations with a sequestered DIS and few exposed Gag binding sites. Therefore, the GU mutant has impaired dimerization, less efficient genome packaging, and reduced packaging selectivity, with high levels of 3G RNA in cells and virions. In the addGG mutant, two guanosines are inserted between the TAR and polyA stems (Fig. 8 B, Bottom). In this mutant, the additional guanosines and cap in 3G RNA no longer destabilize the base of the polyA stem because of the greater separation of the TAR and polyA stem. Therefore, both 3G and 1G RNAs fold into conformations favorable for dimerization and genome packaging. Thus, like the GU mutant, the addGG mutant also packages high levels of 3G RNA into virions. However, unlike the GU mutant, the addGG mutant does not have a packaging defect when competed against WT RNA (Fig. 5C). In the addGG mutant, 1G and 3G RNAs are not expected to have coaxial stacking between TAR and the polyA stem because of the 2-nt insertion. Nevertheless, these RNAs are efficiently packaged, implying that the stability of the polyA stem has a greater role than coaxial stacking in downstream RNA-folding events and genome selection. Intriguingly, the addCC 3G RNA is enriched in the viral particles beyond its cellular abundance (Fig. 4). In this mutant, the two cytosines inserted between TAR and polyA likely facilitate coaxial stacking between the stems. It is possible that the potential coaxial stacking of the two RNA elements provides the addCC 3G RNA the advantage to be enriched in particles.

Previous studies proposed that the 5′ UTR can adopt at least two mutually exclusive conformations with the DIS exposed or sequestered (36, 48–52). The details of the DIS-sequestered conformations differ between the published studies, and in most cases, the role of 5′ end guanosine content was not explored. The observed differences in the previously reported structures are at least in part due to the different methods, conditions, and HIV-1 sequences studied. Our native PAGE analysis revealed that another likely explanation for the differences is that the HIV-1 5′ UTR RNA (whether 1G or 3G) can adopt an ensemble of conformations, and solution studies yield the average structure. The proposed U5:DIS interaction recently reported for a 3G^cap 5′ UTR derived from the HIV-1 MAL isolate based on NMR studies (36) was not observed in any of the major clusters revealed by our analysis of the NL4-3 5′ UTR. Interestingly, the centroid structure of one of our major clusters observed for the WT 3G monomer top bands (sequestered.3, see SI Appendix, Fig. S2) is remarkably similar to the long-distance interaction conformation predicted for the HIV-1 LAI 5′ UTR (50, 53). Recent SimRNA 3D modeling of the ^cap1G 5′ UTR predicted that polyA nt pairing/unpairing may propagate structural rearrangements throughout the 5′ UTR (45), consistent with conclusions from our experiments. Our study demonstrated that minor changes at or near the 5′ end of HIV-1 RNA can have global effects on the 5′ UTR structure. As HIV-1 genomes contain multiple RNA regulatory elements, it is intriguing to contemplate whether minor changes in other regions of the HIV-1 genome can also cause large conformational changes that impact RNA function.

In summary, our studies underscore how HIV-1 exploits RNA structure to distinguish RNAs with minor differences at the 5′ ends. This ability provides additional complexity to the regulation of HIV-1 replication. Despite containing nearly identical primary sequences, 1G and 3G RNAs each fold into multiple conformations that differ in the exposure of the DIS and Gag binding sites, which directly correlates with the ability of the HIV-1 RNA species to dimerize and be packaged. Thus, both RNA dimerization and the binding of multiple Gag proteins to high-affinity sites are required for efficient genome packaging. These prerequisites are consistent with our current model of viral RNA packaging, which proposes that HIV-1 RNA and multiple Gag proteins form complexes at the plasma membrane, the major site of virus assembly (5, 54). Two RNA:Gag complexes merge during the RNA dimerization step, and Gag accumulates using the RNA dimer as a scaffold to form a complete, immature virus particle (54).

Materials and Methods

Generation of HIV-1 Constructs.

The previously described NL4-3–based pON-H0 is referred to as H0 in this report (41). Sequences known to be important for RNA packaging and viral replication are unmodified in H0, including LTRs, 5′ UTR, and gag-pol; portions of NL4-3 genome deleted in H0 have been shown to be unimportant for packaging (17, 23–25, 27). The nef gene of H0 contains a mouse has gene, internal ribosomal entry site, and a nonfunctional green fluorescence protein gene; for simplicity, only the hsa marker gene is shown. The Cdis-WT-B7 is almost identical to H0, except it has a typical subtype C DIS (GTGCAC) and a B7 marker gene instead of hsa gene. To generate various mutants, mutated regions were synthesized (IDT) and used to replace the WT sequence using standard, molecular cloning techniques. The general structures of newly generated plasmids were verified by restriction enzyme mapping; additionally, regions that were synthesized or amplified by PCR were verified by DNA sequencing.

Cell Culture and Flow Cytometry.

Human embryonic kidney 293T cells were maintained as previously described (11). To generate cell lines, 293T cells were transfected with HIV-1 constructs along with helper plasmids pCMVd8.2 and pHCMV-G (55, 56). Supernatants were harvested 24 to 48 h after transfection and clarified through 0.45-μm filters. To generate pools of infected cells for 5′ RACE, 1 to 2 million 293T cells were infected at MOIs of 0.3 to 1.5. To determine the MOI, flow cytometry was performed 3 d after virus infection to measure the proportion of cells expressing the HSA marker.

To determine RNA-packaging efficiencies, dually infected cell lines were generated by sequentially infecting 293T cells first with Cdis-WT-B7 virus at MOI < 0.1, then with H0 virus (WT or mutant) at MOI < 0.1. Doubly infected cells expressing both B7 and HSA markers were enriched by repeated rounds of cell sorting until >94% of the cells expressed both markers. Each cell pool consists of more than 42,000 independent infection events.

To perform flow cytometry analysis, cells were processed as previously described (11), stained with 0.4 µg/mL phycoerythrin (PE)- or allophycocyanin-conjugated anti-HSA (BioLegend) and/or 2.0 µg/mL PE-conjugated anti-B7 antibodies (Miltenyi Biotec). Flow cytometry was performed using an LSR II system (BD Biosciences), while cell sorting was performed on an FACSAria II system (BD Biosciences). Flow cytometry data were analyzed using FlowJo software (TreeStar, LLC).

RNA Isolation, 5′ RACE, and RT-PCR Sequencing.

To isolate RNA from viral particles, supernatants from infected cells were harvested, clarified through 0.45-μm filters, and concentrated by centrifugation at 25,000× g for 90 min through a 20% sucrose cushion. RNA was then isolated using the QIAamp Viral RNA Mini Kit (Qiagen). Total cellular RNA was isolated using the RNeasy Plus Mini Kit (Qiagen). Isolated RNA was converted into cDNA using the SMARTer RACE 5′/3′ Kit (Takara Bio) and a primer that anneals to gag (5′-GGTGGCTCCTTCTGATAATG-3′). The cDNA was then amplified by PCR using a gag-specific reverse primer (5′-GATTACGCCAAGCTTTCGTTCTAGCTCCCTGCTTGCCCATAC-3′) and the universal forward primer from the SMARTer RACE 5′/3′ Kit. PCR products were gel purified, cloned into pRACE plasmid, and sequenced. The 5′ ends of cDNA were determined for at least 140 clones in each sample; the numbers of clones analyzed for each sample are indicated in the figure legends.

To determine RNA-packaging efficiencies, we used a previously described competition assay based on RT-PCR sequencing of the polymorphic DIS region (43). The packaging efficiencies of mutants were calculated by dividing the proportion of mutant RNA in viral particles by the proportion of mutant RNA in cells.

RNA Preparation for in vitro Studies.

To ensure homogeneous 5′ ends in all in vitro–transcribed RNAs, 5′ HHRs were encoded downstream of the T7 RNA polymerase promoter. HHRs were designed to cotranscriptionally cleave the transcript leaving the desired number of 5′ guanosines (57). Plasmids for in vitro transcription–encoding 400- or 402-nt WT HIV-1 5′ UTR and part of the gag gene were purchased from IDT (IDTSmart plasmids). In vitro transcription was carried out using homemade T7 RNA polymerase (58, 59). The resulting RNAs have either two 5′ guanosines to mimic the ^cap1G RNA (referred to as 1G), or four 5′ guanosines to mimic the ^cap3G RNA (referred to as 3G). All RNA constructs were purified by elution from a 6% denaturing polyacrylamide gel (58).

The following mutations were made within the 5′ UTR constructs: four mutants containing 2-nt insertions between the TAR/polyA hairpins (addAA, addUU, addCC, and addGG) and two mutants that disrupt base-pairing interactions at the base of polyA (CA_58,59GU and UG_108,109AC, named GU and AC, respectively). All mutagenesis was carried out using site-directed, ligase-independent mutagenesis (60).

Native Gel Electrophoresis to Monitor HIV-1 5′ UTR Dimerization.

The RNAs (0.4 µM) were folded in 50 mM HEPES, pH 7.5 by heating at 80 °C for 2 min, at 60 °C for 4 min, addition of MgCl₂ (1 mM), and incubation at 37 °C for 6 min and on ice for 30 min. RNAs were mixed with native loading dye and run on 6% nondenaturing polyacrylamide gels containing 1 mM MgCl₂. Gels were stained with ethidium bromide and RNA bands were visualized using an Amersham Imager 680 (GE Healthcare). Band intensity was determined via densitometry using the ImageQuant TL software (GE Healthcare). Dimerization efficiencies were calculated by dividing the total intensity of dimer bands by the total intensity of all monomer and dimer bands. Results are based on at least four independent trials.

In-gel SHAPE Probing.

The protocol for in-gel SHAPE probing was modified from the previously described protocol (48). RNAs were folded as described in the Native Gel Electrophoresis section, and seven aliquots of 80 µL folding mixtures were mixed with native loading dye and run on separate lanes of a 20 × 20 cm nondenaturing 6% polyacrylamide gel containing 1 mM MgCl₂. For each RNA, one lane was cut out, stained using ethidium bromide, and visualized under ultraviolet (UV) light. The stained gel piece and the unstained gel were aligned, and the bands of interest from the remaining six lanes were cut out. The excised gel pieces were cut into small pieces and soaked in 45 mM Tris-borate, pH 8.3, 1mM MgCl₂ (TBM) buffer containing 10% of fresh 100 mM N-methylisatoic anhydride (NMIA, Sigma-Aldrich) in DMSO (Sigma-Aldrich) and incubated at 37 °C for 45 min. The NMIA reaction buffer was then removed by pipetting, and 1.2 mL elution buffer (500 mM ammonium acetate, pH 8.0, 1 mM EDTA) was added to the gel pieces followed by incubation of the RNA at 37 °C for 5 h with shaking. The eluted RNAs were divided into three aliquots and recovered by ethanol precipitation. Control reactions were performed by treating folded RNA with neat DMSO in the absence of NMIA [(−)-reactions]. All primer extension products were generated via SuperScript II reverse transcriptase (Thermo Fisher Scientific) using a 5′ NED-labeled primer (Applied Biosystems, Life Technologies) complementary to the 3′ end of the SHAPE-modified or control reaction RNAs. Capillary electrophoresis was performed by the Shared Genomics Research Facility (Ohio State University) and analyzed using the RiboCat SHAPE analysis software (61).

Determination of HIV-1 Secondary Structure Ensembles.

The Rsample algorithm (40), a component of the RNAstructure (62) software package, was used to calculate the RNA structural ensembles. Briefly, the RNA sequences and the SHAPE reactivity data were used as the input in Rsample, allowing an SHAPE-guided partition function file to be generated. In the case of the dimer bands, the reactivity of the DIS sequence was set to a value consistent with high reactivity prior to Rsample analysis. Even though these nt are strongly protected from SHAPE reagent, this was done to ensure the DIS exposure expected for the kissing loop interaction between dimeric RNAs. The “stochastic” command was used to create a Boltzmann ensemble of 1,000 structures based on the SHAPE-guided partition function. Default parameters were used for both the Rsample and stochastic commands. The structures were clustered by running an R language script (as described in https://rna.urmc.rochester.edu/Text/Rsample.html). The output of this analysis provided the population of each cluster, as well as centroid structures for each cluster. A cluster is defined as a group of structures with similar base pairs, and a centroid structure includes all base pairs with base-pairing probability >0.5 within the cluster (40). The centroid structures were then visualized in RNAstructure (62) and drawn using XRNA (http://rna.ucsc.edu/rnacenter/xrna/xrna.html).

Data Availability

All study data are included in the article and/or SI Appendix.