Selection of ABX464
The first functional screening of the new compounds was based on the use of freshly isolated human peripheral blood mononuclear cells (PBMCs) from healthy donors. These PBMCs were infected by the laboratory HIV strain Ada-MR5. Every molecule was tested at 5 μM on at least 8 different blood isolates; for a rigorous selection of candidate molecules, we maintained the cells with or without the drug for 6 days. Five compounds were selected following these assays and they were demonstrated to have an IC50 in the micromolar range. We selected one, ABX464 (Figure 1a) for further analysis. However, ABX464 is hydrophobic and forms aggregates when applied directly to cell cultures, which leads to some toxicity to the cells (data not shown). To avoid this problem, we prepared a soluble fraction after adding the molecule to cell culture media and elimination of aggregates by centrifugation (see Methods). The binding of ABX464 to bovine serum albumin (BSA) retains the molecule in the soluble fraction as determined by mass spectrometry. Figure 1b, shows dose dependent inhibition of HIV-1 replication by soluble ABX464 in stimulated PBMCs from 5 different donors with an IC50 ranging between 0.1 μM and 0.5 μM. We also determined the concentration of ABX464 with minimal side effects on cell viability using an MTS test (Figure 1c) and cell proliferation was measured by the stable incorporation of the intracellular fluorescent dye 5-(and −6)-carboxyfluorescein diacetate succinimidyl ester (CFSE) into lymphocytes (Figure 1d). ABX464 is only toxic at doses higher than 60 μM but has no effect on cell proliferation.
To generalize the effect of ABX464 on HIV-1 replication in other primary cells, the same protocol was repeated using infected macrophages, which act as viral reservoirs. Cells were treated with between 0.01 μM up to 30 μM concentrations of ABX464 and p24 antigen levels were monitored in culture supernatants over a 12 days period (Figure 1e). Interestingly, ABX464 efficiently blocked virus replication in a dose-dependent manner with an IC50 ranging between 0.1 μM and 1 μM. However, cell viability was not decreased under ABX464 treatment (data not shown).
ABX464 does not select for HIV specific mutations and it is not genotoxic
To complement the previous experiments, which were all performed with primary human cells infected with macrophage-tropic (R5) strains (Ada-MR5 and YU2), we shifted to an in vitro system that may be more relevant to the clinical situation. By infecting primary cells with HIV-1 isolates from patients, we showed that ABX464 had a strong inhibitory effect for all HIV-1 subtypes tested including subtype B, C and recombinant viruses (Figure 2a).
A critical aspect of HIV infection is its ability to generate a diverse viral population through high replication rates and high reverse transcriptase mutation rates. Thus, in infected individual there is a broad pool of virus that may survive in the face of pressure exerted from the host as well as non-potent antiretroviral therapy. This same mechanism is responsible for the selection of drug resistance. ABX464 also very efficiently inhibited the replication of viral strains harbouring mutations that confer resistance to different therapeutic agents in vitro (Figure 2b). While the antiviral drug 3TC was not highly active on K65R and M184V mutant strains, both strains were inhibited by ABX464.
Genetic heterogeneity is a characteristic of HIV, which contributes significantly to its ability to generate mutations that overcome the efficacy of drug therapies. The selection of drug resistant mutants in vitro can be readily accomplished by maintaining the virus in a state of sub-optimal growth, regulated by slowly increasing the amount of drug pressure applied. Resistance to ABX464 was tested on human PBMCs and compared to current therapies (Figure 2c). There were no resistance-inducing mutations detected after treatment with ABX464 for at least 24 weeks (Figure 2c).
We also applied a deep sequencing approach for sensitive detection of low-frequency viral variants across the entire HIV-1 genome. Viruses derived from treated and untreated infected primary macrophages of 4 different donors were sequenced and reads not aligning to human genome were aligned to YU2 sequence using GSNAP [19] (Raw data are provided upon request). The majority of low and high frequency mutations were equally present in treated and untreated samples, demonstrating that ABX464 does not select for specific mutations (Additional file 1: Figure S1a). To ascertain that amplification of viruses from treated samples will not mutate when amplified in PBMCs, they were sequenced following amplification with or without drug pressure. Again, no novel mutations were detected other than the ones existing before treatment in the original samples (Additional file 1: Figure S1b). We conclude that ABX464 was unlikely to select for specific viral mutations that might inhibit viral replication.
The potential genotoxicity of ABX464 was also assessed in GLP-compliant in vitro and in vivo studies and in neither case was ABX464 is genotoxic (see below, Additional file 2: Table S1). This is making clear improvement to fulfill criteria for clinical development, unlike IDC16.
ABX464 increases the levels of spliced HIV RNA
Since the parent drug IDC16 has a specific action on HIV-1 pre-mRNA splicing, we assessed the efficiency of ABX464 using the p∆PSP plasmid. p∆PSP containing the HIV-1 proviral genome which is deleted between nucleotides 1,511 and 4,550 (Additional file 3: Figure S2a) and this plasmid recapitulates all splicing events of HIV-1 pre-mRNA in transfected HeLa cells [13]. The mRNAs produced by splicing were then analysed by RT-PCR using forward and reverse primers that amplify several differentially spliced isoforms encoding the viral proteins Nef, Rev, and Tat. To assess whether ABX464 induced preferential selection of a few splice sites and will favour the production of a specific viral RNA, amplified products were visualized and analysed on the LabChip HT DNA assay on an automated microfluidic station (Caliper, Hopkinton, MA). Each band corresponding to specific viral RNA was compared to total RNA amplified using Caliper software. Unlike IDC16 which was previously shown to completely block the production of spliced viral RNA isoforms [13], ABX464 did not alter the splicing profile when used at concentrations of 5 μM or 10 μM (Additional file 3: Figure S2b). Besides, there was no significant variation in the levels of each specific splice variants between treated and untreated samples (Additional file 3: Figure S2c).
In order to verify that ABX464 did not significantly or globally affect the splicing events of endogenous genes, which could potentially lead to some adverse consequences, the effect of ABX464 was tested by RT-PCR analysis on a pre-existing panel of 382 alternative splicing events (ASEs). These 382 ASEs represent a high-throughput (HT) random snapshot of global alterations of alternative splicing. We performed HT-PCR analysis of these (essentially random) 382 ASEs on multiple PBMC samples, with nine of our drug derivatives, including ABX464, and with various controls: either untreated or treated with DMSO or the control antiviral drug (Darunavir) [20,21]. Analysis of the data allowed further stringent quality controls; ASEs were only considered if >75% of the products ran at the expected mobilities and if total expected PCR concentration was higher than 20 nM. This filter ensured the use of high quality PCRs on well-expressed genes and led to 264 remaining ASEs in our analysis. The splicing profiles of the 12 PBMC samples are shown in Figure 3a compared with stem cells and their derived fibroblasts (a previous treatment that had previously been shown to result in widespread splicing changes) [22]. This shows that there was very little difference between the splicing profiles of the drug-treated PBMC samples and controls, as they formed one of three separate poles with the stem cells and their derived fibroblasts. Consistent with this, the untreated cells and ABX464 treated cells per cent spliced in values for these 264 ASEs had a high correlation of R = 0.89, whereas stem cells and derived fibroblasts correlated poorly at R = 0.59 (Figure 3b). Taken together these data show that ABX464 had a minimal or no global effect on pre-mRNA splicing.
To test whether ABX464 influenced the splicing of HIV RNA in infected cells, we performed an array-based sequence capture using a customized library probes targeting HIV sequences to get rid of cellular RNA. The probes were used to capture cDNAs prepared from infected treated and untreated PBMCs. After double capture, libraries were prepared and sequenced using 454 pyrosequencing (according to GS junior method manual). The average size of the reads around 400 bp allowed unambiguous assembly of viral genome from untreated sample (after 3 and 6 days of infection) using reads that were not mapped to human genome (hg19) (Figure 3c). All sequencing data were analysed using GSNAP [19]. After 3 days post-infection we obtained 32,289 reads for the untreated DMSO sample and 4149 reads for ABX464 treated sample. Strikingly, 17.4% of the reads from treated sample corresponded to splice junctions, against 0.93% in the untreated sample. While the number of reads from treated and untreated samples were similar at 6 days post-infection (20,585 and 27,984, respectively), the fraction corresponding to splice junctions was again larger in treated (13.3%) compared to untreated sample (1.93%). Based on these results we conclude that ABX464 favoured spliced HIV RNA in infected PBMCs, which compromised subsequent synthesis of full-length HIV-1 pre-mRNA and assembly of infectious particles (Figure 3c). Consistently, standard procedures to measure unspliced and mutispliced RNA by quantitative RT-qPCR, demonstrated that ABX464 treatment produced 1.5 times more multispliced RNA than unspliced RNA compared to untreated infected cells (Additional file 3: Figure S2d).
In order to provide further evidence supporting the hypothesis that the anti-HIV activity we observed with ABX464 was the consequence of its inhibitory effect on viral RNA splicing after proviral DNA integration, we examined the effect of the drug on single round NL4.3R E LUC virus containing the entire HIV-1 genome mutated in the envelope gene and harbouring a luciferase marker gene in the Nef position (Additional file 3: Figure S2e). The amount of luciferase activity in cells infected with these virions reflects both the number of integrated proviruses and expression of multiply spliced species encoding Nef/Luc. As expected luciferase activity was strongly compromised with AZT treatment, as this drug blocks the synthesis of proviral DNA. In contrast, ABX464 treatment increased luciferase activity by more than 2 fold compared to untreated infected cells. These results confirmed that ABX464 acted after integration and favours the production of multispliced RNA encoding the luciferase protein.
ABX464 interacts with CBC complex and prevents Rev-mediated export of unspliced viral RNA
To test the effect of ABX464 on splicing and/or export of viral RNA, we used a state of the art system to visualize single HIV RNA molecules in living cells. It is based on an HIV reporter system containing the 5′ and 3′ LTRs that harbour the promoter and polyA sites, respectively, packaging sequences and RRE elements. In addition to this, the construct contains 128 MS2 binding sites inserted between the major donor HIV-1 site (SD1) and the last splice acceptor (SA7) (Figure 4a). The reporter was introduced in HeLa cells stably expressing Tat and MS2-GFP, using the Flp-In system to create cells carrying a single copy of the transgene. Stable expression of MS2-GFP protein allowed excellent visualisation of the transcription site and single pre-mRNA molecules (Figure 4b) and did not alter splicing rate (data not shown). To assay for RNA export, we transfected these cells with constructs expressing Rev protein that will bind to the RRE and facilitate the export of unspliced viral RNA, while protecting it from the splicing machinery [23,24]. Rev expression led to a reduced GFP signal at both the transcription site and in the nucleoplasm (Figure 4b). This was expected since following association of Rev with the high-affinity RRE “nucleation site”, additional Rev molecules can polymerize along the length of the RRE in a step-wise fashion through both protein-protein and protein-RNA interactions [25], thereby removing MS2-GFP from their target sequences. Rev-mediated RNA export will also lead to a reduction of unspliced RNA in the nucleus and a reduction in the intensity of GFP in the nucleus. More nuclear GFP signal was observed in Rev transfected cells treated with ABX464 compared to untreated cells (Figure 4 b and c upper panel). Crucially, ABX464 interfered with both activities of Rev by preserving the GFP signal both at the transcription site (Figure 4 b and c lower left panel) and in the nucleoplasm of cells expressing Rev (Figure 4 b and c lower right panel). However, ABX464 showed no effect on reporter cells in the absence of Rev (Additional file 4: Figure S3).
Both, RNA export and RNA splicing are controlled by the cap binding complex (CBC) which interacts directly with either Rev or the transcription/export (TREX) complex, a multi-protein complex, required for transcription and export of bulk mRNAs [26]. These interactions are thought to recruit Rev and TREX to a region near the 5′-terminal cap structure of mRNA [27] and thereby connect the transcription and export of newly transcribed RNAs. Since ABX464 interfered with Rev-mediated functions, it was important to test whether ABX464 bounds to either Rev or the CBC complex. Using a derivative of ABX464 that has a photoactivatable moiety and competition with ABX464 on purified recombinant CBC20 and CBC80 (CBC) [28], we discovered that ABX464, itself, was able to induce dose-dependent covalent bridging between CBC20 and CBC80, after UV irradiation and this complex could be resolved by SDS-PAGE (Figure 4d and Additional file 5: Figure S4). Mass spectrometry analysis of gel-purified CBC20, CBC80 and the putative complex CBC (80 and 20), showed that the trypsin digestion of CBC (80 and 20) complex gave rise to all predicted peptides except the peptide corresponding to the position 37–66 of CBC20 which was reproducibly under-represented or absent (Additional file 6: Table S2). However, individual digestion with trypsin of either CBC20 or CBC80 from the same sample aroused all predicted peptides. Remarkably, the peptide 37–66 in the crystal structure of the CBC [29] corresponded to the interface between CBC20 and CBC80, which could be the site of interaction between ABX464 and the CBC [30] (Additional file 5: Figure S4 a and b). The same results were obtained with ABX464-N-glucuronide, a more soluble derivative of ABX464 that is produced as unique metabolite using human hepatocytes (see below).
However, neither ABX464 nor its metabolite ABX464-N-glucuronide affected the binding of CBC complex (Figure 4e) to capped RNA probe in a gel mobility shift assay [31]. While the complex between CBC and capped RNA was competed by the m7GpppG, no competition was observed with ABX464 or ABX464-N-glucuronide at any of the concentrations tested, confirming that ABX464 did not interact with the cap binding site of CBC20 (Figure 4e). Our results support the idea that ABX464 bound directly to CBC to specifically prevent Rev-mediated export of viral RNA without interfering with cap binding or export of cellular transcripts.
Efficacy of ABX464 in humanized mouse models
Humanized mice reconstituted with human lymphoid cells provide rapid, reliable, reproducible experimental systems for testing the efficacy of ABX464 in vivo [32,33]. In the initial setting, SCID mice were reconstituted with PBMCs and then infected with the HIV-1 strain JR-CSF [33,34]. Mice were treated twice a day (b.i.d) for 15 days by oral gavage with 20 mg/kg of ABX464. Measures of viral RNA showed that the oral treatment with ABX464 was able to significantly reduce the viral load over a period of 15 days of treatment (Figure 5a). FACS analysis of blood samples showed that treatment with ABX464 prevented depletion of CD4+ cells following infection of reconstituted mice and thereby restored the CD8+/CD4+ ratio back to that of non-infected mice (Figure 5b).
To test the long term effect of ABX464 on the immune system and viral replication in infected humanized mice (hu mouse), newborn NOG mice were transplanted with CD34+ haematopoietic progenitor cells isolated from the umbilical cord blood [35]. This hu mouse model has previously been shown to be accurate for exploring the antiviral potency of new compounds targeting the latent HIV reservoirs [35]. Treatment of NOG hu mice for one month with 20 mg/kg or 40 mg/kg of ABX464 neither altered engraftment values of CD45+ cells nor the ratio of CD8+/CD4+ compared to controls without treatment (Figure 5c). In this study NOG hu mice were infected with the YU2 HIV-1 virus and fed daily for 30 days with 40 mg/kg of ABX464 (Figure 5d left panel) or with HAART (3TC-Tenofovir-Raltegravir) (Figure 5d middle panel) and viral loads were measured as explicated before. Most significantly, unlike HAART, ABX464 leads to a sustained reduction of viral levels in humanized mice after treatment cessation (Figure 5d right panel). Compared to HAART, ABX464 treatment induced a slower kinetic of viral load reduction (Figure 5d left panel). Whereas HAART was very efficient at reducing the viral load to undetectable level after 2–3 weeks of treatment in all infected mice (Figure 5d middle panel), with ABX464 only 2 out of 6 mice had undetectable levels of viral load, 2 mice were under 1000 copies and 2 mice are still have a viral load higher than 6000 copies after 1 month of treatment (Figure 5d left panel). While we cannot exclude the possibility that the slow kinetics of viral reduction could be inherent to ABX464’s mode of action, it is also possible that in this mouse model ABX464 used as monotherapy, is inefficient at reducing the viral load to undetectable levels. For instance mice put on 3TC treatment alone for 42 days all had a viral load above ABX464 (unpublished data). The most striking result is when the HAART treatment was stopped, 6 out 6 mice had a high viral load of above 17,000 RNA copies, with some reaching 200,000 copies (Figure 5d right panel). In contrast, only 2 mice out of 6 treated with ABX464 demonstrated any rebound and that was lower than the rebound after HAART (Figure 5d right panel). Furthermore, these two mice had a viral load of 1,000 copies when we stopped the ABX464 treatment, meaning that ABX464 had controlled most of the cells contributing to viral rebound, at least during the 52 days of treatment cessation. Moreover, 2 mice had undetectable levels of viral load during the whole period of ABX464 cessation and two mice had low viral loads of under 1,000 copies (Figure 5d right panel). These results mean that even though we can detect viruses in the blood, most infected cells in the body that could contribute to viral rebound, mostly latently infected resting CD4+ T cells and macrophages with integrated viral DNA [36,37], are not producing new viruses after ABX464 treatment cessation.
ABX464 is readily metabolized and is non-toxic
Pharmacokinetic studies showed that ABX464 was rapidly absorbed, reaching a maximal plasma concentration within a few hours of dosing in mice (Additional file 7: Figure S5a). These results showed that the long lasting effect in hu mice was due to a reduction of infected cells and not due to a sustained accumulation of ABX464 in the mouse. We tested the potential formation of ABX464 metabolites using cryopreserved hepatocyte primary cultures from various species. The only ABX464 metabolite detected upon incubation with human hepatocytes was the ABX464-N-glucuronide (Figure 1a, Additional file 8: Figure S6 a and b). This metabolite was also formed by mouse and non-human primates, but not by rat, minipig or dog hepatocytes (Additional file 8: Figure S6 a and b). In the mouse at Tmax there was two times more ABX464-N-glucuronide than ABX464 (Additional file 7: Figure S5a). Interestingly, in vitro ABX464-N-glucuronide was as efficient as ABX464 in inhibiting replication of virus strain YU2 in primary macrophages, without inducing any toxicity (data not shown). Thus both ABX464 and ABX464-N-glucuronide could be expected to induce inhibition of viral replication in humans. However, neither glucuronide nor ABX464 were still detectable after 24 hours post-treatment in mice (Additional file 7: Figure S5b right panel). Moreover, cumulating dosing of ABX464 did not sustain the presence of neither ABX464 nor ABX464-N-glucuronide over 6 days after treatment cessation (Additional file 7: Figure S5b left panel). Therefore, we concluded that the long lasting effect, at least in mice, cannot be attributed to glucuronide accumulation.
Furthermore, pharmacokinetic and toxicology studies performed in non-human primates showed that ABX464 was rapidly absorbed with a maximal plasma concentration reached after 2 to 4 hours in marmoset monkeys (Additional file 7: Figure S5c). ABX464-N-glucuronide was formed rapidly with comparable Tmax values but Cmax and AUC values were much higher for the metabolite than for the parent compound in non-human primates (Additional file 7: Figure S5c).
In rat, the no observed adverse effect level (NOAEL) of ABX464 was achieved at a dose of 55 mg/kg b.i.d (Additional file 2: Table S1). In non-human primates, the main target organ of ABX464 toxicity was found to be the gastro-intestinal tract. The NOAEL was considered to be 250 mg/kg b.i.d (Additional file 2: Table S1). Rough estimate of the concentration in the animal indicate that these NOEL dose are 50 times above the efficient dose required to inhibit viral replication in vitro.