The 3-O-(3',3'-dimethylsuccinyl)-betulinic acid (or YK-FH312 [1], or PA-457 [2], or Bevirimat™ [3,4]), has been used as an antiviral which blocks HIV-1 replication via its inhibitory activity on Gag polyprotein maturation [2,5-8]. DSB differs from conventional protease (PR) inhibitors in that it does not bind to PR, but interferes with the PR-mediated Gag processing. The ultimate cleavage of the C-terminal capsid domain CAp25 into CAp24 + SP1 is required for production of fully infectious virions [9]. DSB blocks this step, and decreases or abolishes virus infectivity [2,4,6,10]. Several lines of evidence indicate that the CA-SP1 junction is the preferred target of DSB in HIV-1 Gag precursor [3,4,8,11]. Although there is no available structural data on DSB-Gag complex which could explain its inhibitory activity at the molecular level, data from in vitro experiments [12], as well as the encapsidation of DSB in equimolar ratio to Gag in vivo [13], suggested that the mechanism of inhibitory activity of DSB results from the direct binding of DSB to the Gag polyprotein, or/and to a transient Gag structural intermediate which occurs during virus assembly.

The latter observation incited us to study the possible effect of DSB on assembly of recombinant HIV-1 Gag precursor (Pr55Gag) expressed in heterologous, eukaryotic system. We observed a dose-dependent negative effect of DSB on the process of assembly and release of HIV-1 VLP from recombinant baculovirus AcMNPV-Pr55Gag-infected cells [14]. This effect was not due to a block in Gag synthesis, and was independent of the N-myristoylation of Pr55Gag and its plasma membrane addressing. It did not depend on the presence of the p6 domain at the C-terminus of Gag. The same effect was observed with the Gag precursor of SIVmac (Pr57Gag^SIV), although at significantly higher DSB concentrations, suggesting that the DSB inhibitory activity on Gag assembly was not as strictly sequence-dependent as the negative effect on Gag processing at the CA-SP1 junction [8]. In addition, we found a lower stability of delipidated cores assembled in the presence of DSB, compared to control cores, suggesting a weakening of Gag-Gag interaction occurring in the presence of DSB [14]. Using Gag mutants and a chimeric HIV-MuLV Gag precursor, we mapped the DSB-responsive domain in terms of Gag assembly to the hinge region overlapping the C-terminal end of the CAp24 and the SP1 domain [14].

The DSB concentration at which we observed an inhibitory activity on Gag assembly in insect cells (IC50 ~8–10 μM) was apparently disproportionate compared to the usual doses required for blocking the CAp25 cleavage in HIV-1-infected mammalian cells. However, a wide range of IC-50 values have been reported for the DSB inhibition of virus maturation, varying from nanomolar (0.35 nM [15] and 7.8 nM [2]) to micromolar values (10 μM [12]), depending on the different assays used. In addition, in Pr55Gag-expressing Sf9 cells, the bulk of Gag protein molecules synthesized at 48 h pi has been evaluated to be as high as 5 × 10⁸ per cell [16]. The addition of DSB at 10 μg/ml to 10⁶ cells corresponded to 12 × 10⁹ DSB molecules per cell, i.e. a DSB to Gag stoichiometric ratio of 24: 1 at this DSB concentration. A 24-fold excess of DSB over Gag was therefore compatible with a mechanism of Gag assembly inhibition due to a stoichiometric interaction between the drug and its protein target.

Whatever the molecular mechanism, our observation raised the question of the difference between Pr55Gag-expressing Sf9 cells, in which DSB inhibited VLP assembly [14], versus HIV-1-infected human cells, in which DSB was found to block the CA-SP1 (CAp25) to CAp24 maturation cleavage [3,4,8,11], and to have limited effects on virus assembly [1]. In our experimental model of baculovirus-infected cells [14], assembly of Pr55Gag was analyzed in a context devoid of PR and of glycoproteins (Gp) SUgp120 and TMgp41, three viral components which have been identified as directly or indirectly involved in the antiviral effects of betulinic acid derivatives [8,17,18]. In the aim to reconcile the different antiviral activities of DSB, we explored cellular and viral determinants of the DSB response, and their possible role in modulating the degree of susceptibility to DSB of the VLP assembly process. Among the viral candidates, we analyzed EnvGp160, the precursor to the envelope glycoproteins (reviewed in [19]), and two inner core components, the Vpr and Vif proteins. Vpr is packaged into the virion in substoichiometric amounts with Gag [20-23], and Vif, which is also coencapsidated with Gag, has been found to exert a control on proteolytic processing of Gag in insect cells [24] and human cells [25].

We found that coexpression of wild-type Vif protein (Vif^wt) with Pr55Gag restored the VLP assembly in DSB-treated Sf9 cells at levels observed in the absence of the drug, suggesting an antagonistic effect of Vif towards DSB. Data obtained with Vif mutants indicated that the anti-DSB function of Vif required the integrity of the zinc binding domain (ZBD) recently identified in the Vif protein [26-28], but was independent of the Vif packaging function. Electron microscopic analysis showed that coexpression of Pr55Gag and Vif^wt, in the presence or absence of DSB, resulted in a major change in the VLP egress pathway: the majority of VLP budded in intracytoplasmic vesicles and were released by exocytosis, instead of budding at the plasma membrane as in cells expressing Pr55Gag alone. With ZBD mutants of Vif however, the VLP budding pathway was similar to that observed in cells expressing Pr55Gag alone. Our data suggested that the anti-DSB effect of Vif, a novel function associated with its ZBD, was the indirect consequence of its effect on the cellular pathway of VLP assembly and budding.

The intracellular Gag content was found to remain constant throughout the period of DSB treatment in both Sf9 and 5BD.1 cells (Fig. 1Ai and 1Bi), which confirmed that DSB had no significant effect on the level of Gag protein synthesis [14]. However, a drastic decrease in the yields of extracellular VLP was observed at DSB doses superior to 4 μg/ml in Pr55Gag-expressing Sf9 cells (Fig. 1Aii; and refer to [14]). By contrast, only a moderate decrease in VLP production (20–30%) was detected for DSB-treated 5BD.1 cells at high DSB concentrations (12 to 16 μg/ml; Fig. 1Bii). Protein analysis of VLP showed that their Gag protein content mainly consisted of Pr55Gag and CAp24 proteins, with other minor species migrating at the expected position for intermediate cleavage products, e.g. Pr47 to Pr41 (Fig. 1Bii). Prolonged exposure of autoradiograms of immunoblots reacted with radiolabelled secondary antibody revealed a discrete alteration of the Gag processing at high DSB concentrations: there was a progressive increase in the amount of uncleaved CAp25 versus the CAp24 species (Fig. 1Biii), as expected from previous studies [3,4,8,11].

Figure 1 Effects of DSB on HIV-1 VLP production by (A) insect cells and (B) mammalian cells. (A), Sf9 cells infected with AcMNPV-Pr55Gag were treated with increasing concentrations of DSB in DMSO-aliquots for 30h at 18h pi, as indicated on top of the panels. Cells (more ...)

VLP assembly and release were therefore less sensitive to DSB inhibitor in 5BD.1 cells compared to Gag-expressing Sf9 cells. This suggested that the DSB sensitivity of the VLP assembly pathway might be modulated by the cellular context in which the HIV-1 Gag precursor was expressed, or/and by viral proteins present in 5BD.1 cells and absent from Sf9 cells. The following experiments were designed to address this issue, and to determine which factor(s) possibly interfered with DSB inhibitory activity and accounted for the difference in DSB response between Sf9 and 5BD.1 cells, as well as other mammalian cells.

In order to test this possibility, Sf9 were coinfected with AcMNPV-Pr55Gag and AcMNPV-Gp160, and subjected to increasing doses of DSB for 30 h, at 18 h pi. Culture medium samples were collected at 48 h pi and assayed for production of extracellular VLP. Results were compared with VLP yields from Sf9 cells infected with AcMNPV-Pr55Gag alone and treated in parallel with DSB at the same doses. No significant difference in the DSB effect on VLP assembly was detectable with or without coexpression of EnvGp160 (data not shown). This excluded the direct or indirect participation of HIV-1 envelope glycoproteins in the level of susceptibility to DSB of assembly and extracellular release of VLP by Sf9 cells.

Nef in its processed form, called Nef core, has been shown to be a bona fide component of the virion inner core [37-40]. In 5BD.1 cells, which do not express Nef but express Vif [29,30], we observed a significantly lesser inhibitory effect of DSB on VLP assembly, compared to Gag-expressing Sf9 cells (refer to Fig. 1Bii). Considering that Nef protein was absent from both Sf9 and 5BD.1 cells, the difference in DSB response between these two cell types apparently excluded Nef as a possible modulator of the DSB sensitivity of VLP assembly.

Vpr is coencapsidated with Gag via interaction of the N-terminal alpha-helical domain encompassing residues 17–33 in Vpr [41-44] with the LXXLFG motif in the p6 domain of Gag [21,22,45-48]. In Sf9 coinfected with AcMNPV-Pr55Gag and AcMNPV-Vpr, the same DSB sensitivity of VLP assembly was observed as in cells solely expressing AcMNPV-Pr55Gag: both Pr55Gag and Vpr protein signals decreased in parallel and in DSB dose-dependent manner in the extracellular medium of DSB-treated cells, although their intracellular content remained unchanged (Fig. 2). This implied that Vpr did not significantly interfere with the inhibitory effect of DSB on Gag assembly.

Figure 2 Absence of counteracting effect of Vpr on DSB inhibition of HIV-1 VLP assembly and release. Sf9 cells were coinfected with two baculoviruses at equal MOI each (5 PFU/cell), one expressing Pr55Gag, the other expressing His-tagged Vpr. Cells were treated (more ...)

Figure 3 Influence of Vif on the DSB susceptibility of HIV-1 VLP assembly in Sf9 cells. Sf9 cells were coinfected with equal MOI (5 PFU/cell) of two baculoviruses expressing Pr55Gag and Vif, respectively. Cells were treated with increasing concentrations of DSB (more ...)

Figure 4 Genotype and expression of recombinant Vif mutants in Sf9 cells. (A), Sequence alignment of the central and C-terminal domains of HIV-1 Vif proteins, WT and mutants. The zinc binding domain (ZBD) and its three constitutive loops are boxed: loops 1 and (more ...)

We then tested the anti-DSB activity of Vif mutants with different encapsidation phenotypes. With VifsubC, the production of extracellular VLP remained virtually unchanged throughout the DSB concentration range, with less than 15% decrease in VLP production at high DSB doses (Fig. 5). As observed with Vif^wt (refer to Fig. 3b, c), there was a DSB-dependent, progressive decrease of VifsubC mutant protein content in VLP, relative to the Pr55Gag content, with 20–30% lesser Vif protein incorporated at high DSB doses, compared to control samples (Fig. 5b, c, samples 16–20). A similar DSB resistance pattern as with Vif^wt and VifsubC was observed with the other packaging-defective mutants VifsubA, VifsubB, and VifsubCΔ170 (not shown). Likewise, the packaging-efficient mutant VifKRA8 showed the same phenotype as Vif^wt and the packaging-defecting mutants in terms of anti-DSB activity (not shown). These results suggested that the DSB-counteracting function of Vif was independent from the packaging function of Vif.

Figure 5 Counteracting effect of packaging-defective mutant Vif subC on the DSB inhibition of HIV-1 VLP assembly. Sf9 cells were coinfected with two baculoviruses at equal MOI of each (5 PFU/cell), one expressing Pr55Gag, the other expressing the double substitution, (more ...)

Taking the latter observation into account, we substituted the serine residue to a valine at position 116. We assumed that the bulky side chain of valine would introduce local disorganization in the 3D structure of the ZBD domain, as did the S116 deletion, and would be detrimental to the anti-DSB effect of Vif. We found that the VifS116V mutant was coencapsidated with Gag at the same levels as Vif^wt (Fig. 6Bi, lane 0). However, the assembly and extracellular release of VLP from Sf9 cells coexpressing Pr55Gag and VifS116V showed the same degree of DSB susceptibility as the one observed when Pr55Gag was expressed alone (Fig. 6Bi, and Fig. 6C). Thus, the lack of antagonistic effect against DSB of the packageable mutant VifS116V confirmed that anti-DSB function and packaging into VLP were separate functions in the Vif protein.

Figure 6 Absence of anti-DSB effect of zinc-binding domain mutants of Vif. Sf9 cells were coinfected with two baculoviruses at equal MOI of each (5 PFU/cell), one expressing Pr55Gag, the other expressing VifS116V (A and B, (i)) or VifC133S (A and B, (ii)). Cells (more ...)

To further analyze the role of the ZBD structure in the Vif anti-DSB activity, we constructed another mutant of recombinant Vif protein. Cysteine at position 133 in Vif is a residue essential for virus infectivity [62,63], for Zn coordinate formation and ZBD-associated functions in Vif [27,28]. We therefore generated mutation C133S in recombinant Vif, and tested mutant VifC133S in co-expression with Pr55Gag in control or DSB-treated Sf9 cells, as above. In untreated cells, VifC133S behaved as VifS116V mutant, and was coencapsidated with Pr55Gag into VLP at levels equivalent to Vif^wt (Fig. 6Bii, lane 0). In DSB-treated cell samples, VifC133S had the same phenotype as VifS116V in terms of lack of anti-DSB effect: assembly and release of VLP from Sf9 cells coexpressing Pr55Gag and VifC133S showed the same degree of DSB sensitivity as from Sf9 cells expressing Pr55Gag alone (Fig. 6Bii, and Fig. 6C).

These results suggested that the antagonistic activity of Vif against the DSB inhibition of Gag assembly, absent from VifS116V and VifC133S mutants, was associated with the ZBD and more precisely involved residues located on the N-terminal side of loop 2. Thus, the phenotype of our Vif mutants with respect to their packaging and anti-DSB properties showed that the integrity of the ZBD structure was not required for the packaging of Vif into VLP produced by Sf9 cells, but was crucial for its DSB counteracting effect.

Figure 7 EM and immuno-EM analysis of Pr55Gag-expressing Sf9 cells, with or without Vifwt coexpression. Sf9 cells were infected with AcMNPV-Pr55Gag alone or coinfected with another baculovirus expressing Vif (AcMNPV-Vifwt) at equal MOI of each (5 PFU/cell), harvested (more ...)

The proportion of VLP following the intravesicular budding and exocytosis pathway compared to the ones using the plasma membrane pathway was estimated under the EM, by counting several hundreds of VLP in subcellular compartments of more than 20 different cells. In control Sf9 cells expressing Pr55Gag alone, less than 5% VLP were found within the vesicular compartment, whereas in Gag+Vif^wt-coexpressing cells, the proportion increased to 30 to 50%, viz. a 5- to 10-fold increase. Likewise, in cells coexpressing Pr55Gag and Vif^wt and treated with DSB, most VLP used the intravesicular budding and exocytic pathway (Fig. 8). Interestingly, many VLP-containing vesicles showed an electron-dense, heterogenous lumen (Fig. 8), resembling multivesicular bodies (MVBs) observed in mammalian cells. MVBs belong to the late endosomal subcellular compartment, and have been identified as the preferred budding sites for WT HIV-1 particles in primary human macrophages (reviewed in [66]), as well as in human epithelial and T cells for gag mutants altered in the cluster of basic amino acids of the matrix (MAp17) domain [67].

Figure 8 DSB treatment of Sf9 cells coexpressing Pr55Gag and Vifwt. Sf9 coinfected with AcMNPV-Pr55Gag and AcMNPV-Vifwtat equal MOI of each (5 PFU/cell) were treated with DSB at 10 μg/ml for 30 h at 18 h pi. Cells were harvested at 48 h pi, and processed (more ...)

We next examined cells coexpressing Pr55Gag and ZBD mutants of Vif under the EM, and found that, in the presence of Vif116V and VifC133S, the VLP budding pathway was similar to the one observed in Sf9 cells expressing Pr55Gag alone, i.e. a majority of VLP budding at the plasma membrane and rare intravesicular VLP (less than 10%; Fig. 9). The EM pattern of VifS116V and VifC133S mutants was consistent with their phenotype, as both mutants failed to negate the inhibitory effect of DSB on VLP assembly. Taken together, our results suggested that, in the presence of Vif^wt, the VLP assembly and budding process was redirected to the vesicular compartment, and that the VLP egress via exocytosis represented a salvage pathway through which HIV-1 VLP escaped the negative effect of DSB.

Figure 9 EM analysis of Sf9 cells coexpressing Pr55Gag and ZBD mutants of Vif. Sf9 were coinfected with AcMNPV-Pr55Gag and AcMNPV-VifS116V (a) or AcMNPV-VifC133S (b) at equal MOI of each (5 PFU/cell), harvested at 48 h pi, and processed for EM analysis. The vast (more ...)

It is generally accepted that DSB inhibits the cleavage of CAp25 into CAp24 and SP1 by the viral PR, due to its interference with the Gag substrate [8]. However, in recombinant Pr55Gag-expressing Sf9 cells, a cellular context devoid of PR and other viral proteins, DSB showed a dose-dependent inhibitory activity on VLP assembly and release [14]. The aim of the present study was to understand this dual inhibitory activity, and explain the apparent discrepancy between the DSB effects observed in mammalian and non-mammalian, insect cells. We first explored the effect of DSB on VLP production in 5BD.1 cells, a mammalian trans-packaging cell line producing VLP devoid of viral genome, as the VLP produced by AcMNPV-Pr55Gag-infected Sf9 cells. We found that DSB had only a moderate inhibitory effect on VLP yields at high DSB doses (Fig. 1), indicating that VLP assembly in 5BD.1 cells was less sensitive to DSB inhibitor, compared to Pr55Gag-expressing Sf9 cells. This suggested that the DSB negative effect on the VLP assembly process might be modulated by factors depending on the cellular or/and viral context.

We therefore investigated on the possible influence of viral components on the pattern of anti-assembly effect of DSB, and in particular the role of viral partners of Pr55Gag within the capsid. Coexpression of recombinant Pr55Gag with EnvGp160 or Vpr did not modify the level of inhibition of VLP assembly by DSB (Fig. 2), whereas coexpression of Vif^wt restored the production of VLP in DSB-treated cells to levels found in the absence of the drug (Fig. 3). A panel of recombinant Vif mutants (Fig. 4) were then tested for their anti-DSB activity. We found that the DSB-antagonistic effect of Vif was retained in packaging-defective mutants of Vif (Fig. 5), but abolished by a Cys-to-Ser substitution at position 133 (Fig. 6Bii), a mutation which destroyed the zinc finger-like structure or ZBD. A phenotype similar to that of VifC133S was observed for mutant VifS116V (Fig. 6Bi), which carried a mutation on the N-terminal side of the large loop (loop 2) generated by the four HCCH coordinates with the Zn atom (Fig. 4A). Both VifC133S and VifS116V mutants were encapsidated into VLP at levels comparable to Vif^wt (Fig. 6Bi and 6Bii, control lanes 0). Our results therefore suggested that (i) the anti-DSB effect and packaging into VLP were two independent functions in Vif; (ii) the function of Vif which negated the DSB-induced inhibition of VLP assembly depended on the integrity of the zinc-binding domain, and more precisely on a discrete region of loop 2 overlapping residue 116 (Fig. 4A). This region differed from the Vif packaging signals [50].

EM analysis of Sf9 cells coexpressing Gag and Vif^wt or Vif mutants gave some insight into the cellular mechanism of anti-DSB activity of Vif. Sf9 cells coexpressing Pr55Gag and Vif^wt, with or without treatment with inhibitory doses of DSB, showed a high proportion of VLP budding into intracytoplasmic vesicles and egressing via exocytosis (Fig. 7b–e and Fig. 8). This contrasted with cells expressing Pr55Gag alone, in which the majority of VLP budded at the plasma membrane (Fig. 7a). When Pr55Gag was coexpressed with one or the other of the ZBD mutants, VifS116V or VifC133S, we observed a drastic change in VLP budding, compared to Vif^wt coexpression, consisting of a reversion to the plasma membrane budding pathway, as in Sf9 cells expressing Pr55Gag alone (Fig. 9). Since both ZBD mutants lacked the anti-DSB activity and failed to redirect VLP to the vesicular compartment, as did Vif^wt, it might be hypothesized that the antagonistic activity of Vif towards DSB would be the indirect effect of a Vif-mediated change in the VLP assembly sites and mode of cellular exit.

It has been shown that the assembly and release of HIV-1 virions proceeds via two pathways, depending upon the cell type [67]: (i) in primary human macrophages, virions preferentially follow the exosomal pathway via MVBs [67-69]; (ii) in HeLa cells and T lymphocytes, the major exgress route consisted of plasma membrane addressing and direct budding at the cell surface, but MA polybasic signal mutants of Gag use the MVB pathway in these cells [67]. Sf9 cells expressing Pr55Gag alone belonged to the second category of cells [16,64,65], but when coexpressed with Vif^wt, the VLP assembly and budding process mimicked the MVB budding and exocytic pathway used by MA polybasic mutants in HeLa and T cells. The hypothesis formulated above implied that the intravesicular budding and exocytic pathway of VLP would be less sensitive to DSB inhibitory activity than the plasma membrane assembly and budding pathway usually observed in insect cells. If confirmed, this would be an example of drug resistance mechanism (DSB, in the present case) which involves the bypass of a drug-sensitive assembly and budding pathway by the virus or virus-like particle progeny.

The results of our study suggested that DSB and other betulinic acid derivatives could be considered not only as antivirals for patients treatment in vivo, but also as chemical probes to analyse the molecular and cellular mechanisms of retroviral Gag assembly in vitro. In the latter context, considering Vif as a determinant of the budding pathway usage in Sf9 cells, and as a modulator of the DSB response in terms of VLP assembly, any evaluation of potential HIV-1 assembly inhibitors using the baculovirus-insect cell system should be carried out in the presence of the Vif protein.

The authors declare that they have no competing interests.

SDF performed the bench work, and BG and PB performed the EM analyses. PC synthesized the DSB. SSH and PB conceived the strategies and designed the experiments. SB contributed to data analysis. PB wrote the manuscript. All the authors read and approved the final manuscript.

This work has been supported by the Agence Nationale de Recherche sur le SIDA (ANRS Grant 2005–2006/003 and DendrAde-2007). SDF was the recipient of an ANRS fellowship. We are grateful to Eric Cohen (University of Montréal, Québec) for supplying us with the baculoviral clone expressing His-tagged Vpr, and to David Rekosh and Mari-Lou Hammarskjöld (University of Virginia at Charlottesville) for their gift of 5BD.1 packaging cells. We acknowledge with thank Elisabeth Errazuriz (Centre Commun d'Imagerie de Laennec) for her significant contribution to specimen processing and EM studies, and Sylvie Fiorini for her expert technical assistance. The efficient secretarial aid of Cathy Berthet is also gratefully acknowledged.