Host transcriptomic and translational changes induced by HIV-1 infection
To investigate the translational landscape of HIV-1 infected cells we employed human SupT1 cells, a cell line derived from T cell lymphoblastic lymphoma. Cells were infected with VSVG pseudotyped HIV-1 NL4-3 Gag-iGFP ΔEnv virus, which carries a frameshift mutation in the Env gene that allows a single round of infection with HIV-1. Over 80% cells expressed GFP 24 hours post infection (hpi), verified by flow cytometry analysis (Supp. Figure 1A). Cytoplasmic lysates from both mock and HIV-1 infected cells were harvested at 8, 16 and 24 hpi and subjected to RNASeq and RiboSeq analyses, both in duplicate (Fig. 1A). Quality control of the RiboSeq data indicated a successful ribosome profiling experiment 40. First, the length distribution of trimmed and mapped ribosome protected fragments (RPF) were within the expected range peaking at 30 nt (Supp. Figure 1B) 41,42. Second, we used a probabilistic analysis pipeline (PRICE) to predict the most likely P-site of every RPF to determine the reading frame distribution 43. Metagene analysis revealed that the majority of RPF that map to coding sequences (CDS) are in the 0-frame, indicating efficient nuclease digestion (Supp. Figure 1C). Third, a notable enrichment of reads was observed within the annotated coding transcripts, with only a small percentage of the RPFs mapping to 5'UTRs (Supp. Figure 1D).
At 8 hpi, approximately ~ 0.6% of RNASeq reads were derived from HIV-1 mRNAs, and ribosome-associated viral RNA was barely detectable (~ 0.05%). At 16 hpi and 24 hpi, HIV-1 mRNA reads increased to 1.2% and 5.8%, respectively. Similarly, we observed HIV-1 RiboSeq reads rise to ~ 0.5% and ~ 2.5% of the total at 16 hpi and 24 hpi, respectively (Supp. Figure 1E). We first focused on host transcriptional and translational changes induced by HIV-1 infection by plotting log2 fold changes of RiboSeq and RNASeq for each gene at 8, 16 and 24 hpi compared to mock 44 (Fig. 1B). At 8 hpi, out of the 1254 genes found to be differentially expressed in either RNASeq or RiboSeq, 258 genes (20.57%) showed significant changes on the RNA level, yet 787 (62.76%) host genes showed significant changes exclusively in the RiboSeq (with a false discovery rate (FDR) < 0.05). This indicates that the initial response to HIV-1 infection was driven by translational changes. Gene ontology (GO) analysis of the highly enriched pathways included processes associated with cholesterol metabolism, response to stress, lymphocyte activation, translation and rRNA processing (Supp. Table 1). Specifically, 42 translationally upregulated genes were linked to lymphocyte activation (ID: GO:0046649, fold change: 2.05) and 22 to cholesterol metabolism (ID: GO:0008203, fold change: 4.13) at 8 hpi (Fig. 1B). The latter has been shown to have a critical role in viral entry and is likely stimulated by Nef 45–49. At 16 and 24 hpi, we observed prominent changes in RNAseq reads without changes (991 of 1312 genes at 16 hpi, 75.53%, 1662 of 2257 genes at 24 hpi, 73.64%). Fewer genes showed changes exclusively in RiboSeq reads (58 genes at 16 hpi, 262 genes at 24 hpi), indicating that later stages of the host response are dominated by transcriptional changes (Fig. 1B). In addition, genes linked to stress response pathways were enriched in both RiboSeq and/or RNASeq (173 genes at 16 hpi, fold change: 1.41, 263 genes at 24 hpi, fold change: 1.47; ID: GO:0006950). In contrast, at 16 and 24 hpi, genes linked to rRNA processing pathways decreased in RiboSeq and/or RNASeq (42 genes at 16 hpi, fold change: 6.51, 89 genes at 24 hpi, fold change: 6.85; ID: GO:0006364) (Fig. 1B). Interestingly, 14 genes linked to translation (ID: GO:0006412) were downregulated for either RiboSeq and/or RNASeq at all timepoints, suggesting repression of host translation in response to HIV-1 infection.
Next, we focused on translation efficiencies (TEs) of genes within the affected GO pathways (Fig. 1C, Supp. Table 2). The TE metric integrates RiboSeq and RNAseq by dividing normalized ribosome footprint reads by the normalized RNA sequencing reads. Among the genes linked to cholesterol metabolism, we noted the protein kinase PRKAA1, which was reported to facilitate HIV-1 viral replication through several regulatory pathways 50,51. Genes consistently upregulated throughout infection on the level of TE included several genes involved in lymphocyte activation including HELLS, TOP2B and DOCK11 (Fig. 1C, Supp. Table 2). Amongst the stress response genes, we observed upregulation of the SMC family of proteins, namely SMC3, SMC6, SMCDH1, which are known interactors of Tat 52. Interestingly DUSP3, which plays a key role in cell cycle and transcription is downregulated 53. Strikingly, we also observed a reduction of TE of ribosomal proteins such as RPL13A, RPS21, RPS14 at all infection timepoints (Fig. 1C, Supp. Table 2).
Overall, these results indicate reprogramming of the host upon HIV-1 infection, as well as a mounting of cellular stress response to alter translation. Whilst the initial host response following infection is chiefly regulated at the translational level, as the infection progresses, cellular alterations are primarily driven by changes in cytoplasmic transcript abundance, particularly in genes related to cellular stress, rRNA processing and translation.
HIV-1 infection suppresses global host translation at the initiation level
Since our data indicated a disruption of translational processes, we next investigated the effects of infection on global host translation. For this purpose, uninfected and infected cellular lysates were loaded onto sucrose gradients and RNA profiles were generated at each infection time point (Fig. 2A). At 16 and 24 hpi, we observed a reproducible increase in monosomes (80S) with a slight decrease in the number of polysomes (Fig. 2C-D). Quantification of the polysome-to-monosome ratios at different time points also indicated that there was a reduction in this ratio in infected cells at both 16 and 24 hpi suggesting a slight, but measurable inhibition of translation initiation upon infection (Supp. Figure 2A). At 8hpi, we observed a slight decrease in overall polysome profile at 8 hpi, however, this was not reproducible and the polysome-to-monosome ratio does not change as compared to the mock in both replicates, suggesting the inhibition of initiation takes place at later stages of infection (Fig. 2B, Supp. Figure 2A).
We further checked the distribution of specific host and viral RNAs in individual fractions. Interestingly, we observed a difference between host and viral RNAs. Specifically at 8 hpi, mRNAs of host housekeeping genes GAPDH and ACTB were present in all fractions along the gradient and were mostly enriched in the polysomes, indicating these host RNAs were being actively translated (Fig. 2F (left), Supp. Figure 2B). However, these host mRNAs were found to move from actively translating polysomes towards monosomes by 16 and 24 hpi (Fig. 2F (middle, right), Supp. Figure 2B). At the same time, we saw an increase in the amount of GAPDH and ACTB host mRNAs into the heavier fractions beyond polysomes, which could represent stress granules, formed as part of the cellular stress response 54–56 (Fig. 2F, Supp. Figure 2B). Strikingly, the fully- and partially spliced HIV-1 isoforms, which encode for regulatory, accessory and envelope proteins, were actively translated at all stages of infection (Fig. 2G and H, Supp. Figure 2D and E). Interestingly, at 8 and 16 hpi the majority of the unspliced HIV-1 RNAs were predominately located in the polysomes, but at 24 hpi, a proportion of unspliced RNAs repartitioned into the lighter fractions. The increased availability of untranslated unspliced RNA could provide genomes for assembly into viral particles, as active translation of unspliced RNA is proposed to inhibit viral packaging 57 (Fig. 2I, Supp. Figure 2F). To further understand this phenomenon, we compared TEs of viral genes and host genes using our sequencing data. While host genes GAPDH and ACTB showed a decreasing trend in TE over time, both gag and Pol exhibited an increasing trend, peaking at around 16 hpi (Supp. Figure 2C), again indicative of preferential translation of viral over host genes.
Altogether, these analyses demonstrate a gradual decline in host global translation upon HIV-1 infection possibly due to decreased translation as a cellular stress response. However, the virus ensures its own mRNAs evades this suppression, resulting in increased translation efficiencies as the infection progresses.
Codon resolved analysis of stalling events on host transcriptome during infection
Given the observed suppression of host translation by HIV-1, we further investigated the potential mechanisms underlying this effect, including possible alterations of the codon usage and cellular tRNA pools induced by HIV-1 infection. To accurately identify and quantify stalling events on individual codons, we implemented a new algorithm as illustrated in Supp. Figure 3A and described in Methods. Briefly, we sorted the P-site coverage per codon derived from PRICE (peaks) into individual bins and selected peaks in each bin exceeding the mean coverage in that bin by more than twice the standard deviation. Based on these, P-site- and A-site codons of the putative stalling sites were mapped and counted. The same analysis was conducted on randomly sampled peaks in the same RiboSeq dataset and statistical significance of stalling on each codon was calculated based on the enrichment scores.
When examining the A-site stalling of ribosomes from uninfected samples, we noted stalling on Isoleucine (Ile, L), Alanine (Ala, A), and Asparagine (Asn, N) encoding codons (Supp. Figure 3B). The presence of these stalling events can be linked to the presence of rare codons or to limitations in the amino acid pools and tRNA availability 58,59. Throughout infection, stalling profiles of Ala and Asn remained consistent, however we observed slight (0.25–0.34 log2 fold change) increases in A-site stalling on Ile encoding codons at later time points of infection (Supp. Figure 3B). Conversely, the codon dependent P-site stalling events were overall more enriched as compared to the A-site stalling, yet no time dependent changes were observed upon infection. P-site stalling was mostly observed at Asp (D) codons throughout infection, which was reported to be a common stalling site across species, including humans (Supp. Figure 3C) 60. Overall, these data indicate that infection does not increase P-site stalling events, at least in single round infection assays performed here. However, the increase in A-site stalling at the Ile codon as infection progresses is interesting and awaits further investigation.
Changes in HIV-1 gene expression patterns during infection
Having characterized the impact of HIV-1 infection on host translation and ribosome stalling, we proceeded to investigate translation patterns of viral specific ORFs based on codon and frame-resolved RPF signatures. As described earlier, viral RNA was consistently detected throughout the infection process (Supp. Figure 1E, Fig. 3A), and ribosome footprints were seen on all canonical viral coding sequences at later timepoints (Fig. 3A). The density of RPFs was highest at the HIV-1 5'UTR, probably because this region is common among all HIV-1 splice isoforms. Alternatively, it could also be due to increased ribosome pausing within the highly structured 5’UTRs of the HIV-1 transcripts, which was also shown across species in RiboSeq datasets 61 (Fig. 3A). Additionally, there was dense clustering of RPFs on the overlapping regions of HIV-1 genes. To identify which protein was most likely to be translated from the overlapping regions, we used PRICE to identify the correct reading frame. At 8 hpi, read numbers were insufficient to map to each HIV-1 gene. By 16 and 24 hpi, translation patterns of individual HIV-1 genes could be discerned. At 16 hpi, RiboSeq reads mostly corresponded to ‘early’ HIV-1 regulatory proteins Tat and Rev, which are translated from fully-spliced HIV-1 transcripts (Fig. 3A). Amongst the partially spliced mRNAs, we detected a high number of Vpu reads at 16 and 24 hpi, indicating its preferential expression (Fig. 3B). HIV-1 Vpu enhances the release of progeny virions from infected cells and our results suggest that HIV-1 ensures Vpu is produced at greater abundance at later points of infection 62. Unspliced HIV-1 transcripts encode for the Gag and Gag-Pol polyproteins, with Gag-Pol polyprotein produced through a − 1FS event. From 16 to 24 hours, we marked the largest increase (almost 10 -fold) in the HIV-1 reads corresponding to Gag, which is in support of the notion that the Gag protein is needed for the assembly of the virus at later stages of infection.
Identification of hitherto unknown HIV-iORFs in Pol and Vif
Given that the majority of the HIV-1 RPFs originate from the 5’UTR, we next investigated sites of translation initiation. For that, we performed RiboSeq experiments in the presence of harringtonine, which leads to an accumulation of ribosomes at canonical and alternative translation initiation sites. Quality control showed predominantly reads of length 29–30 and a majority of RPF mapping to host coding sequences in the 0-frame, similar to the cycloheximide dataset, confirming a successful experiment (Supp. Figure 4A-C). For harringtonine-treated samples, within the HIV-1 genome, we observed the most prominent peaks in the 5’UTR and upon a closer look at the annotated start codons of virus ORFs (Fig. 4A). Interestingly, most of the peaks we identified in the 5’UTR were located at near-cognate AUG codons suggestive of non-canonical initiation events that give rise to putative short uORFs ranging in length from 3 to 42 amino acids (Fig. 4B).
Further, within the HIV-1 genome two hitherto unknown HIV-1 iORFs were detected: one in the Pol and the other in the Vif coding region (called by PRICE with a p-value < 0.005) (Fig. 4C and Supp. Figure 4D and 5A-C, Supp. Table 3). The iORF detected in the Pol coding region is located within the non-coding exon 2, 166 nt upstream of the end of the canonical Pol ORF and contains the D2 donor splice site (position 4,962). It starts with a UUG and could generate different peptides ranging from 23–48 amino acids depending on splicing events from the D2 donor site to different acceptor sites namely A2, A3, A4a/b/c, or A5 (Supp. Figure 4D, 5A and B). Indeed, we detected some RiboSeq reads overlapping the D2-A3 and D2-A5 junctions, consistent with translation of this iORF (Supp. Table 4). While the mechanism of initiation at this site is unclear, we want to highlight that if a transcript is D1-A1-spliced (positions 743 to 4,913), the majority of the Gag-Pol ORF is spliced out and the iORF is then localized only 18 nt upstream of the splicing event, so that the alternative UUG initiation codon comes prior to any canonical AUG start codon, with potential ribosome recruitment occurring within the 5’UTR (Supp. Figure 5B).
The other identified iORF is located within the non-coding exon 3 of Vif and starts with a near cognate GUG start codon in the + 1-reading frame relative to Vif. Its start codon is 48 nt downstream of the A2 acceptor site and spans a coding sequence of 9 amino acids. The D3 splice donor site is located within the iORF (Fig. 4C and Supp. Figure 5C). Thus, through splicing of D3-A3, D3-A4a/b or D3-A5 the iORF can generate multiple different fusion peptides (Supp. Figure 5D). Furthermore, alignment of 4903 HIV-1 wild type sequences indicated that the GUG start codon, D3 splice site as well as UAG stop codon for the Vif iORF is universally conserved (Fig. 4D). Next, to assess whether the iORF is expressed in cells we designed a fluorescence reporter assay where mCherry is placed in-frame with the iORF, (+ 1 frame) and stop-codons in the 0-frame, thus mCherry is produced only if translation begins at the GUG start codon of the Vif iORF. As a control, mCherry was placed in the canonical Vif frame, without any 0-frame stop codons (Supp. Figure 4E). Through this set-up, we observed ~ 25% of mCherry expression relative to the control, indicating the Vif iORF can be translated in cells (Fig. 4E).
Intriguingly, Pol or Vif iORF translation across the D2-A4 or D3-A4-splice site respectively would result in translation of a yet undescribed reading frame within exon 4 (Supp. Figure 5E). We were also able to validate translation from this novel ORF via mass spectrometry (Supp. Table 5). Specifically, while we could not detect peptides for the 5’ UTR uORFs nor the iORFs directly, likely due to the short peptide length after trypsin digest, we could indeed detect the novel HIV-1 peptide SSSEQSDSSSFSIK from this undescribed reading frame in exon 4 (Supp. Figure 5F), and in addition a peptide corresponding to the Env ORF (Supp. Figure 5G), for which canonical translation is excluded due to presence of multiple frameshift mutations inserted due to biosafety reasons. To identify the transcripts that contain the mentioned D2/3-A4 splice sites, we next performed long read nanopore amplicon sequencing, which revealed Rev4-5 as the most highly expressed candidates for Pol iORF and Rev7-Rev12 for Vif iORF translation, as for these transcripts the iORFs are located upstream of the canonical Rev start codon (Supp. Table 6). Notably, the only alternative source of the peptide would be the transcript Gp41 2 containing the cryptic D2b-A5 splice junction (Supp. Figure 5H). To differentiate between these possibilities, we explored our RiboSeq data for RPFs spanning the aforementioned D3-A4a/b or D2b-A5 splice sites, detecting 5 and 11 overlapping reads for the D3-A4a and D3-A4b respectively, compared to only a single read overlapping the D2b-A5 splice site (Supp. Table 4), thus supporting the notion that Pol/Vif-iORF translation is the primary source of this transcript. In future work, it will be crucial to identify the mechanism behind translation initiation of near-cognate start codons within HIV-1 transcripts and assess the properties and function of their translation products in infection process.
Ribosome stacking upstream of the HIV-1 frameshift site
In addition to the newly identified non-canonical iORFs, HIV-1 is already known to use non-canonical translation events such as the − 1FS, achieved through the means of a FS stimulatory site (FSS) comprising of a slippery sequence and a RNA secondary structure, spanning nucleotides 2084–2130 at the Gag-Pol overlapping gene. Past studies using this frameshift element in vitro indicated ribosomal pausing over the P-site UUA codon of the slippery site (UUUUUA), although where exactly the RNA secondary structure begins is still under debate 28,30,63.
Motivated by previous work highlighting the importance of ribosome collisions for viral frameshifting 64, we complemented our analysis by DisomeSeq on mock and infected samples, by selecting 50–80 nt sized protected fragments 64–66. In DisomeSeq, read lengths were seen to be broader compared to classical Ribo-seq with local peaks of around 54 and 60–63 nt, which is a size range consistent with expected RPF for disomes (Supp. Figure 6A) 65. The first population around 54 nt most likely represents closely stacked ribosomes after a collision event, whereas the second population around 62 nt may correspond to translating ribosomes that slowed down but are still a few nucleotides apart. The DisomeSeq read depth profile mapping to HIV-1 showed distinct stalling peaks compared to the RiboSeq (Fig. 5A). Looking specifically at the vicinity of the canonical frameshift site, we observed a notable accumulation of disomes approximately 70 nucleotides upstream of the slippery site (SS) (Fig. 5A and B). Beyond the GagUAA stop codon, both monosome and disome coverages decreased drastically and only ribosomes that moved to the − 1 frame would produce footprints (Fig. 5C). By using the ratio of ribosome footprints in the Pol and Gag coding sequences, we estimated the frameshifting efficiency of HIV-1 in infected T-cells. Approximately, 12–17% of the Gag-Pol coverage was observed at Pol at both 16 and 24 hpi, which is slightly higher than reported in previous in vitro studies 22. FE also remained constant throughout the course of infection (Fig. 5D).
An alternative RNA fold at the Gag-Pol frameshift site fine-tunes viral gene expression
Next, to explore the stalling phenomena at the Gag-Pol gene upstream of the frameshift site, we performed codon-resolved analysis of the ribosomal P-sites using the RiboSeq data. We noted minor accumulation of ribosomes pausing at the P site UUA (Leu) of the slippery site and the main pause was confirmed to be occurring at the AGG (Arg) codon 74 nucleotides upstream of the Leu codon, similar to the stacking event we observed in disomes (Fig. 6A and Fig. 5B). Alignment of 4903 HIV-1 wild type sequences indicated that both the canonical frameshifting slippery sequence and the amino acid sequence Pro-Arg-Lys-Lys (PRKK), encoded by the -CCU AGG/A AAA AAG/A- nucleotide sequence, at the pause site were universally conserved across HIV-1 variants (Fig. 6A, bottom). In order to validate the existence of the pause on this site, we performed a previously described ribosome pausing assay in vitro in rabbit reticulocyte lysates treated with harringtonine 67,68 (Supp. Figure 6B and C) (see also Methods). In this assay, we employed reporter mRNAs containing a FLAG tag followed by nucleotides 64–2687 (Δ1870–1881) of the HIV-1 genome to best mimic the native genomic context of viral frameshifting (Supp. Figure 6B, up). To accurately mark the position of the predicted paused products, control mRNAs with 2X stop codons (UAA) after the SS (SSpause), FS, no FS and AGG (AGGpause) were designed and employed as size markers. In this assay, we observed a short-lived pause at the Arg codon, indicating ribosome stacking at this position resolves over time (Supp. Figure 6B and C). Intriguingly, although not strongly visible in ribosome profiling data, a persistent translational pause was observed near the canonical slippery site, suggesting there is a potential ribosome drop-off at this position (Supp. Figure 6B and C).
Once we validated the occurrence of the pause in vitro, we sought to understand the structural basis for the stalling event. To this end, we performed DMS probing in HIV-1 infected SupT1 cells. This revealed an extended structure of the HIV-1 frameshift site, which broadly agrees with the in vitro probed structure reported by Low et al. 26. According to our DMS probing in cells, this extended RNA structure would similarly fold in a three-way junction with a large central bulge, with the GC-rich stem downstream of the poly-U sequence being highly structured. However, compared to the previously reported structure, we could not detect neither the conventional (nucleotides 183–190 base paired to nucleotides 217–225) nor alternate lower stems, indicating that the alternate lower stem (nucleotides 135–140 base paired to nucleotides 223–228) may not be folded in the majority of molecules in their native, dynamic state in the cell (Fig. 6B).
In order to monitor the dynamics of the RNA fold of the FSE further, we next employed a single-molecule optical tweezers assay 69,70. By periodically applying increasing and decreasing forces we monitored the (un)folding behaviour of the HIV-1 RNA encompassing the putative structure (HIV-1ext) (Fig. 6C). Most of the HIV-1ext curves exhibited two unfolding steps occurring at 10.3 ± 1.7 pN and 17.8 ± 1.5 pN, respectively. The second unfolding step at approximately 18 pN, with a contour length change of 17.2 ± 5.9 nm, indicates the presence of a highly stable ≈ 26 nucleotides long stem loop. Based on our DMS-probing data (Fig. 6C), the most probable structure to linked to this second unfolding step was the canonical 28 nucleotides long frameshift stem loop 71. In agreement, the frameshift stem loop (nucleotides 190–217) was the most stable element regarding base-pair reactivities, and the unfolding length perfectly matches the expected size of the stem loop 72. The unfolding curves also showed an additional low force unfolding event preceding the unwinding of the frameshift hairpin (Fig. 6C). This first step was characterised by significant hysteresis, indicative of complex structures, and showed an unfolding force 10.3 ± 1.7 pN and refolding peak at 5.1 ± 1.0 pN (Supp. Figure 7 and Supp. Table 7). Furthermore, the contour length change of this first step has an average of 64.5 ± 2.7 nm, corresponding to the opening of ≈ 113 nucleotides, excluding the canonical frameshift stem loop, which seem to unfold subsequently. Finally, the presence of the hysteresis during unfolding and refolding can be explained by the existence of this relatively large bulge in our DMS model, which yet seems flexible enough to fold into a more complex structural element.
Altogether, our optical tweezers experiments indicated two major structural elements are present in the HIV-1 FSE: the 3-way junction with a bulge and the frameshifting stem loop (Fig. 6C and Supp. Figures 7A-C). Furthermore, we confirmed that the frameshift stem loop is a stable structure 71,72, even in the context of the extended sequence. Indeed, the addition of 5´and 3´ extremities did not appear to affect the folding of this element. Finally, our data support the existence of an extended structure that undergoes a two-step unfolding, which is in perfect agreement with the stalling profile observed in RiboSeq (Fig. 6C and Supp. Figures 7A-C, Supp. Table 7).
To assess the impact of the extended RNA fold on HIV-1 FE, we employed our well-established dual fluorescence-based frameshift reporter assay in HEK293 cells 73. The extended HIV-1 RNA showed a FE of 5.4 ± 0.1% (Fig. 6D). Next, deleting the extended fold (Δext), which leaves only the canonical SS and frameshift stem loop, resulted in a ~ 30% decrease in FE (3.7 ± 0.5%). Targeting the lower stem of the extended fold with antisense oligonucleotide (ASO1) similarly decreased the FE by ~ 40% compared to non-targeting (Fig. 6B). This suggests the functional importance of the extended fold in HIV-1 frameshifting. The AGG pause site is proximal to nucleotides AAAAAG, which could be a putative slippery site (SS*) (Fig. 6A). Given this site is located upstream of the extended HIV-1 RNA fold, we investigated the occurrence of an alternative frameshift event (FS*). In that case frameshifted ribosomes would encounter a stop codon in the − 1 frame prior to the canonical SS. Mutating SS* (SS*mut) resulted in a slight decrease in FE (4.2 ± 0.8%). Mutating both the canonical SS and the − 1 frame stop codons still allowed for ~ 16% FE relative to WT (FE = 0.9 ± 0.4), corresponding to − 1FS. It is important to note that controls were employed with both SS and SS* mutated to eliminate the background mCherry signal (see also Methods). All in all, our reporter assay confirms the functional relevance of the extended RNA fold, and the existence of a low efficiency alternative frameshift event (FS*). This supports the notion of an additional regulatory layer for Gag-Pol expression to be explored in future studies.