Identification of Novel HIV-1 Latency-Reversing Agents from a Library of Marine Natural Products

Natural products originating from marine and plant materials are a rich source of chemical diversity and unique antimicrobials. Using an established in vitro model of HIV-1 latency, we screened 257 pure compounds from a marine natural product library and identified 4 (psammaplin A, aplysiatoxin, debromoaplysiatoxin, and previously-described alotaketal C) that induced expression of latent HIV-1 provirus in both cell line and primary cell models. Notably, aplysiatoxin induced similar levels of HIV-1 expression as prostratin but at up to 900-fold lower concentrations and without substantial effects on cell viability. Psammaplin A enhanced HIV-1 expression synergistically when treated in combination with the protein kinase C (PKC) activator prostratin, but not the histone deacetylase inhibitor (HDACi) panobinostat, suggesting that psammaplin A functions as a latency-reversing agent (LRA) of the HDACi class. Conversely, aplysiatoxin and debromoaplysiatoxin synergized with panobinostat but not prostratin, suggesting that they function as PKC activators. Our study identifies new compounds from previously untested marine natural products and adds to the repertoire of LRAs that can inform therapeutic “shock-and-kill”-based strategies to eliminate latent HIV-infected reservoirs.

Consistent with published data [8,15,27], both control LRAs exhibited dose-dependent expression of GFP across multiple concentrations ( Figure 1B). For example, maximal GFP expression in J-Lat cells was observed after treatment with 1.4 µM panobinostat (17.1 ± 2.7%) or 12 µM prostratin (6.4 ± 0.9%) ( Figure 1B). Using the approach of Hashemi et al. [28] and normalizing results relative to the average GFP response for 12 µM prostratin (the concentration at which maximal prostratin activity was observed), the EC 50 s for panobinostat and prostratin were calculated to be 0.10 ± 0.02 and 7.1 ± 2.8 µM, respectively (Table 1). These results confirm that panobinostat is approximately 70-fold more potent than prostratin, which is also consistent with previous reports [8,15,27]. Table 1. 50% effective concentrations (EC 50 s) of latency-reversing agents (LRAs). EC 50 s were calculated in J-Lat 9.2, 8.4, and 10.6 cells based on the percent of GFP expression relative to controls treated with 12, 38, or 3.8 µM prostratin, respectively, using the approach of Hashemi et al. [28]. n.d., not determined.  [14] To directly assess the impact of control LRAs on cell viability (i.e., in a manner independent of provirus expression), parental Jurkat cells (Clone E6-1, American Type Culture Collection; Manassas, VA, USA), which do not harbor integrated HIV-1 provirus, were prepared and treated with LRAs as described above. Following 24 h incubation, compound toxicity was assessed by measuring surface expression of the early apoptotic marker annexin V by flow cytometry (by staining with annexin V-APC; BioLegend, San Diego, CA, USA). Results were reported as the fold-increase in annexin V-positive cells relative to 0.1% DMSO-treated control cells (mean ± s.e.m.) from at least 3 independent experiments. Representative data are shown in Figure 1C. Here, treatment with 0.1% DMSO resulted in 13.6% of cells with APC-fluorescence above the bulk of the cell population, which is presumed to lack surface expression of annexin V (top). In contrast, treatment with 0.15 µM panobinostat resulted in 58.6% APC-positive cells (i.e., a 4.3-fold increase in apoptosis from DMSO control) (center), while 12 µM prostratin resulted in 24.5% APC-positive cells (i.e., 1.8-fold increase from DMSO control) (bottom). As expected, control LRAs increased cellular apoptosis in a dose-dependent manner ( Figure 1D). For example, treatment with 0.045 µM panobinostat induced a 5.8 ± 1.2-fold increase in apoptosis, indicating poor cellular tolerance at concentrations that induced latency reversal, while at least 10 µM prostratin induced no more than a 2.0 ± 0.3-fold increase in apoptosis ( Figure 1D). Taken together, our control experiments confirm that panobinostat is a more potent, yet also more toxic, LRA than prostratin [8,15,27].
We next screened 257 structurally-diverse pure compounds derived from marine natural products at 2.5 µg/mL for latency reversal activity in J-Lat 9.2 cells. Of these, seven (2.7%) compounds resulted in cytolysis and disruption of cell morphology as observed by light microscopy, consistent with widespread cell death, and were not considered further. Of the remaining 250 compounds, four induced GFP expression in at least 4% of cells (Figure 2A). This "hit" rate of 1.6% is in line with previously reported screens of natural product libraries (~0.5-1.1%) [18,19] and supports the notion that pure natural product libraries are enriched for bioactive LRAs compared to synthetic small molecule libraries, where reported hit rates of~0.1% are more frequent [28,35,36]. Structures of the four identified compounds are shown in Figure 2B. Notably, psammaplin A, originally isolated from the two-sponge associate Poecillastra sp. and Jaspis sp. [29], was previously identified as an HDACi with anti-tumor activity [30,31]. In contrast, aplysiatoxin and debromoaplysiatoxin, which differs from aplysiatoxin by the loss of a bromine atom in the phenol ring, are toxins produced by blue-green algae and potent PKC activators [32][33][34]. However, not all HDACis and PKC activators possess potent HIV latency modulatory functions [14,37,38], and none of these compounds have been investigated as HIV-1 LRAs. Finally, alotaketal C, originally isolated from Phorbas sp., is a potent activator of cyclic AMP and PKC signaling that we recently characterized for its HIV-1 latency reversal activity [14,20] and is thus not assessed further here. Structures of the four identified compounds are shown in Figure 2B. Notably, psammaplin A, originally isolated from the two-sponge associate Poecillastra sp. and Jaspis sp. [29], was previously identified as an HDACi with anti-tumor activity [30,31]. In contrast, aplysiatoxin and debromoaplysiatoxin, which differs from aplysiatoxin by the loss of a bromine atom in the phenol ring, are toxins produced by blue-green algae and potent PKC activators [32][33][34]. However, not all HDACis and PKC activators possess potent HIV latency modulatory functions [14,37,38], and none of these compounds have been investigated as HIV-1 LRAs. Finally, alotaketal C, originally isolated from Phorbas sp., is a potent activator of cyclic AMP and PKC signaling that we recently characterized for its HIV-1 latency reversal activity [14,20] and is thus not assessed further here.
Each new LRA induced dose-dependent reversal of HIV-1 latency in J-Lat cells across multiple concentrations ( Figure 2C,D). The average maximum responses observed for each compound ranged from 1.4 to 2.7-fold more than controls treated with 12 µM prostratin. For example, psammaplin A induced GFP expression in up to 17.6 ± 4.0% of cells at 5 µg/mL (3.8 µM). When results were normalized to the average GFP response for 12 µM prostratin [28], psammaplin A's EC 50 was calculated as 1.9 ± 0.3 µM, approximately 19-fold higher than the EC 50 of panobinostat (Table 1). In contrast, aplysiatoxin induced GFP expression in 9.4 ± 0.1% cells with as little as 0.15 µg/mL (0.1 µM) and yielded a calculated EC 50 of 0.045 ± 0.021 µM. Aplysiatoxin is therefore 160-fold more potent than prostratin and 2.2-fold more than panobinostat, identifying it as a particularly potent LRA in J-Lat 9.2 cells. Debromoaplysiatoxin induced GFP expression in 7.2 ± 1.5% of cells at 1.5 µg/mL (1.3 µM) and yielded a calculated EC 50 of 0.92 ± 0.14 µM, or 7.7-fold more potent than prostratin.
In Jurkat cells, we observed that 3.8 µM psammaplin A induced a 3.3 ± 0.6-fold increase in annexin V staining, with extensive cell death observed at higher concentrations. In contrast, no more than 2.0 ± 0.2 and 1.6 ± 0.1-fold increases in annexin V staining were observed for aplysiatoxin and debromoaplysiatoxin, respectively ( Figure 2E,F). Thus, psammaplin A appeared to be toxic at concentrations that induced latency reversal, while both aplysiatoxin and debromoaplysiatoxin were largely well-tolerated across all concentrations.
These agents displayed similar dose-response profiles in J-Lat 8.4 cells, indicating that they act on HIV provirus independent of its integration site ( Figure 3A). For example, when results were normalized to the average GFP response for 38 µM prostratin (i.e., the concentration at which maximal activity was observed in J-Lat 8.4 cells), the EC 50 of psammaplin A was calculated as 1.5 ± 0.1 µM, or approximately 20.5-fold higher than the EC 50 of panobinostat (Table 1). Aplysiatoxin induced detectable GFP expression at concentrations as low as 0.0001 µM, with a calculated EC 50 of 0.011 ± 0.003 µM, or 900-fold more potent than prostratin and 6.6-fold more than panobinostat. Similarly, an EC 50 of 0.52 ± 0.02 µM was calculated for debromoaplysiatoxin, or 19.2-fold more potent than prostratin. Each new LRA induced dose-dependent reversal of HIV-1 latency in J-Lat cells across multiple concentrations ( Figure 2C,D). The average maximum responses observed for each compound ranged from 1.4 to 2.7-fold more than controls treated with 12 µM prostratin. For example, psammaplin A induced GFP expression in up to 17.6 ± 4.0% of cells at 5 µg/mL (3.8 µM). When results were normalized to the average GFP response for 12 µM prostratin [28], psammaplin A's EC50 was calculated as 1.9 ± 0.3 µM, approximately 19-fold higher than the EC50 of panobinostat (Table 1). In contrast, aplysiatoxin induced GFP expression in 9.4 ± 0.1% cells with as little as 0.15 µg/mL (0.1 µM) and yielded a calculated EC50 of 0.045 ± 0.021 µM. Aplysiatoxin is therefore 160-fold more potent than prostratin and 2.2-fold more than panobinostat, identifying it as a particularly potent LRA in J-Lat 9.2 cells. Debromoaplysiatoxin induced GFP expression in 7.2 ± 1.5% of cells at 1.5 µg/mL (1.3 µM) and yielded a calculated EC50 of 0.92 ± 0.14 µM, or 7.7-fold more potent than prostratin.
In Jurkat cells, we observed that 3.8 µM psammaplin A induced a 3.3 ± 0.6-fold increase in annexin V staining, with extensive cell death observed at higher concentrations. In contrast, no more than 2.0 ± 0.2 and 1.6 ± 0.1-fold increases in annexin V staining were observed for aplysiatoxin and debromoaplysiatoxin, respectively ( Figure 2E,F). Thus, psammaplin A appeared to be toxic at concentrations that induced latency reversal, while both aplysiatoxin and debromoaplysiatoxin were largely well-tolerated across all concentrations.
These agents displayed similar dose-response profiles in J-Lat 8.4 cells, indicating that they act on HIV provirus independent of its integration site ( Figure 3A). For example, when results were normalized to the average GFP response for 38 µM prostratin (i.e., the concentration at which maximal activity was observed in J-Lat 8.4 cells), the EC50 of psammaplin A was calculated as 1.5 ± 0.1 µM, or approximately 20.5-fold higher than the EC50 of panobinostat (Table 1). Aplysiatoxin induced detectable GFP expression at concentrations as low as 0.0001 µM, with a calculated EC50 of 0.011 ± 0.003 µM, or 900-fold more potent than prostratin and 6.6-fold more than panobinostat. Similarly, an EC50 of 0.52 ± 0.02 µM was calculated for debromoaplysiatoxin, or 19.2-fold more potent than prostratin.   In J-Lat 10.6 cells, we observed that all LRAs were capable of inducing GFP expression in at least two-thirds of cells ( Figure 3B), indicating robust efficacy. However, we also observed an average of 7.5% of J-Lat 10.6 cells spontaneously expressing GFP in the absence of LRAs, consistent with previous reports [39], indicating a lower barrier to HIV latency reversal compared to J-Lat 9.2 and 8.4. Nevertheless, when results were normalized to the average GFP response for 3.8 µM prostratin (one of three concentrations where maximum activity was observed), the EC50 of psammaplin A was again calculated as 1.5 ± 0.1 µM, or approximately 36.6-fold higher than the EC50 of panobinostat (Table 1). Moreover, GFP expression was observed with aplysiatoxin concentrations as low as 3.7 × 10 −5 µM (37 pM) and a calculated EC50 of 0.0033 ± 0.0012 µM, or 540-fold more potent than prostratin and 12.4-fold more than panobinostat. Finally, the calculated EC50 of debromoaplysiatoxin (0.081 ± 0.029 µM) was 22.2-fold more potent than prostratin. Thus, the rank-order of potency for all LRAs was consistent across all cell lines.  Figure 1A,C, respectively.
In J-Lat 10.6 cells, we observed that all LRAs were capable of inducing GFP expression in at least two-thirds of cells ( Figure 3B), indicating robust efficacy. However, we also observed an average of 7.5% of J-Lat 10.6 cells spontaneously expressing GFP in the absence of LRAs, consistent with previous reports [39], indicating a lower barrier to HIV latency reversal compared to J-Lat 9.2 and 8.4. Nevertheless, when results were normalized to the average GFP response for 3.8 µM prostratin (one of three concentrations where maximum activity was observed), the EC 50 of psammaplin A was again calculated as 1.5 ± 0.1 µM, or approximately 36.6-fold higher than the EC 50 of panobinostat (Table 1). Moreover, GFP expression was observed with aplysiatoxin concentrations as low as 3.7 × 10 −5 µM (37 pM) and a calculated EC 50 of 0.0033 ± 0.0012 µM, or 540-fold more potent than prostratin and 12.4-fold more than panobinostat. Finally, the calculated EC 50 of debromoaplysiatoxin (0.081 ± 0.029 µM) was 22.2-fold more potent than prostratin. Thus, the rank-order of potency for all LRAs was consistent across all cell lines. To confirm that LRAs induce HIV protein expression in addition to the GFP reporter, J-Lat 10.6 cells were also stained with the HIV-1 p24 Gag antibody KC57-RD1 (Beckman Coulter, Indianapolis, IN, USA) and processed using the Cytofix/Cytoperm Fixation/Permeabilization Kit (BD Biosciences, Mississauga, ON, Canada) prior to flow cytometric analysis. Results were then reported as the foldincrease in p24 Gag -positive cells relative to 0.1% DMSO-treated control cells (mean ± s.e.m.) from at least 3 independent experiments. All LRAs were observed to induce at least 9.7-fold increased p24 Gagpositive cells, with the same rank order as observed for GFP expression ( Figure 3C). This confirms that LRAs also induce viral protein expression.  To confirm that LRAs induce HIV protein expression in addition to the GFP reporter, J-Lat 10.6 cells were also stained with the HIV-1 p24 Gag antibody KC57-RD1 (Beckman Coulter, Indianapolis, IN, USA) and processed using the Cytofix/Cytoperm Fixation/Permeabilization Kit (BD Biosciences, Mississauga, ON, Canada) prior to flow cytometric analysis. Results were then reported as the fold-increase in p24 Gag -positive cells relative to 0.1% DMSO-treated control cells (mean ± s.e.m.) from at least 3 independent experiments. All LRAs were observed to induce at least 9.7-fold increased p24 Gag -positive cells, with the same rank order as observed for GFP expression ( Figure 3C). This confirms that LRAs also induce viral protein expression.
To investigate whether LRAs induce proviral expression in primary human cells, we obtained peripheral blood mononuclear cells (PBMCs) from three HIV-infected donors on stably-suppressive antiretroviral therapy for at least three years (Figure 4). Study protocols were approved by the Institutional Review Boards of Simon Fraser University and the University of British Columbia-Providence Health Care Research Institute (REB: H15-03077, approved 8 March 2016). Written informed consent was obtained from all donors. Frozen PBMC aliquots were thawed and allowed to recover in R10+ medium at 37 • C, 5% CO 2 for 24 h at 2.5 × 10 6 cells/mL. PBMCs were then incubated at 10 6 cells/mL with positive control cell activators PMA (100 ng/mL) plus ionomycin (1 µg/mL), 3.8 µM psammaplin A, 1.1 µM aplysiatoxin, 1.3 µM debromoaplysiatoxin, or 0.1% DMSO vehicle control. All conditions were performed in duplicate. Following 24 h incubation at 37 • C and 5% CO 2 , supernatant p24 Gag protein was quantified by ELISA (Xpress Bio, Frederick, MD, USA), and cell viability was measured by flow cytometry using Guava Viacount, a DNA intercalating dye (EMD Millipore). In most cases, each LRA caused an increase in supernatant p24 Gag above background; however, substantial donor-to-donor variation was observed, consistent with other studies [10][11][12] ( Figure 4A). For example, while treatment with PMA + ionomycin induced an average 62.3 ± 31.1% (mean ± s.e.m.) increase in supernatant p24 Gag relative to untreated cells, psammaplin A resulted in a 85.0 ± 41.6% increase. Similarly, aplysiatoxin and debromoaplysiatoxin induced increases of 56.0 ± 15.0 and 46.9 ± 22.3%, respectively. No major changes in cell viability were observed, with a maximum 15.4 ± 2.2% (mean ± s.e.m.) reduction in viability in the presence of 3.8 µM psammaplin A ( Figure 4B). These results indicate that LRAs have the capacity to activate latent HIV-1 provirus in primary human cells isolated from persons with long-term viremia suppression on antiretroviral therapy.
As described previously [8][9][10][11][12][13][14][15], treatment of cells with combinations of LRAs acting through different mechanisms tends to result in synergistic effects on HIV-1 latency reversal, while treatment with compounds acting through similar mechanisms tends to yield at most additive responses. These observations can therefore be used to identify potential functional classes of novel LRAs [14,15]. To demonstrate this, we assessed GFP expression in J-Lat 9.2 cells treated with novel LRAs in combination with control LRAs, including the pro-inflammatory cytokine TNFα, the HDACi panobinostat, and the PKC activator prostratin ( Figure 5). In these studies, synergism was observed in all cases where control LRAs were applied in combination: for example, treatment of J-Lat 9.2 cells separately with 0.01 µg/mL TNFα or 0.15 µM panobinostat induced 22.3 ± 2.5% and 8.9 ± 1.8% GFP-positive cells, respectively, whereas treatment of cells with both compounds simultaneously led to 49.0 ± 1.2% GFP-positive cells, which is~1.6-fold higher than would be expected if the effects of these two compounds were strictly additive (i.e., 31.2%; Figure 5). Similarly, treatment with TNFα plus 12 µM prostratin, or panobinostat plus prostratin, induced responses that were 1.6-and 2.6-fold higher than expected for additive effects, respectively. These levels of synergism were statistically significant (p < 0.05; Student's unpaired t-test with a two-sided, Bonferroni correction) as measured by the Bliss independence model, which was calculated as described previously [10,11].    Treatment of J-Lat 9.2 cells with 2 µM psammaplin A in addition to TNFα or prostratin induced 49.8 ± 1.1 and 44.4 ± 5.1% GFP-positive cells, representing 1.7-and 2.7-fold increases in GFP expression over what would be expected from strictly additive effects, respectively. However, treatment of J-Lat cells with psammaplin A plus panobinostat induced only 9.8 ± 1.7% GFP-positive cells, or 0.6-fold of what would be expected by strictly additive effects ( Figure 5). These results are consistent with the known function of psammaplin A as a HDACi [30,31]. Conversely, treatment of cells with 2 µM aplysiatoxin plus TNFα or panobinostat induced 51.1 ± 0.5 and 48.4 ± 2.2% GFP-positive cells, respectively, or 1.5-and 2.3-fold increases in GFP-positive cells over expected additive effects, while co-treatment with prostratin induced only 12.2 ± 1.8% GFP-positive cells, or 0.6-fold of expected additive effects. These observations are consistent with the known function of aplysiatoxin as a PKC activator [33,34]. Similar results were found using debromoaplysiatoxin: co-treatment with TNFα or panobinostat induced 50.4 ± 1.5 and 47.6 ± 1.5% GFP-positive cells, respectively, or 1.5 or 2.3-fold over expected additive effects, while co-treatment with prostratin induced 12.0 ± 1.8% GFP-positive cells, or 0.6-fold of expected additive effects, indicating that its latency reversal activity is also likely due to activation of PKC. All synergistic effects were statistically significant as measured with the Bliss independence model. Taken together, these results indicate that the latency reversal properties of these pure natural products in J-Lat cells are consistent with their previously reported functional properties.
In summary, we identify four LRAs derived from marine natural products that can be added to the repertoire of known HIV-1 shock-and-kill agents. The likely mechanisms of action for all four compounds, supported here and by prior studies, are consistent with established functional classes, including one HDACi (psammaplin A) and three PKC activators (aplysiatoxin, debromoaplysiatoxin, and previously-described alotaketal C) [14]. Dose response profiles suggest that psammaplin A is a less potent LRA than panobinostat, while aplysiatoxin and debromoaplysiatoxin are more potent than prostratin. These observations were confirmed in two additional J-Lat cell lines and in PBMCs from three HIV-infected donors, indicating that latency reversal occurs independent of proviral integration sites and that these compounds can reverse latency in primary cells. The contributions of these mechanisms to HIV-1 latency reversal were further supported by synergism studies, described both here and elsewhere [14].
Our previous discovery of multiple novel HIV inhibitors from this chemically-diverse library (e.g., [25]) led us to hypothesize that we might also identify new modulators of HIV latency with distinct molecular mechanisms. However, while this screen instead identified only compounds of the HDACi and PKC activation classes, the results described here should not preclude future screening of natural product-derived compound libraries for additional LRAs which may act by novel mechanisms of action. Conversely, as the kinetic properties of both HDACis and PKC activators do not necessarily correspond with latency reversal efficacy [14,37,38], testing of library compounds with both known and unknown molecular functions remains warranted. In support of this, we notably identify one PKC activator, aplysiatoxin, that is particularly potent across multiple cell models, with activity observed in J-Lat 8.4 and 10.6 cells at picomolar concentrations. Finally, we note that additional latency reversal mechanisms by these agents may remain undetected. For example, we previously described alotaketal C as an LRA of the PKC activator class [14], but it is additionally reported to function as an activator of cyclic AMP signaling [20], which may also modulate HIV-1 latency reversal [40,41].
Taken together, this study highlights the benefits of natural product-based screens for LRA discovery. The identification and evaluation of new LRAs will support the development of novel therapeutic combinations and clinical approaches to reduce or eliminate latent HIV-1 reservoirs.