Latency Reversing Agents Induce Differential Responses in Distinct Memory CD4 T Cell Subsets in Individuals on Antiretroviral Therapy

Summary

Latent proviruses persist in central (T_CM), transitional (T_TM) and effector (T_EM) memory cells. We measured the levels of cellular factors involved in HIV gene expression in these subsets. Highest levels of acetylated H4, active NF-κB, and active P-TEFb were measured in T_EM, T_CM and T_TM cells, respectively. Vorinostat and romidepsin displayed opposite abilities to induce H4 acetylation across subsets. PKC agonists were more efficient at inducing NF-kB phosphorylation in T_CM cells but were more potent at activating PTEF-b in the T_EM subset. We selected the most efficient LRAs and measured their ability to reverse latency in each subset. While ingenol alone had modest activities in the three subsets, its combination with an HDACi dramatically increased latency reversal in T_CM cells. Altogether, these results indicate that cellular HIV reservoirs are differentially responsive to common LRAs and suggest that combination of compounds will be required to achieve latency reversal in all subsets.

Introduction

More than 30 years after its discovery, there is still no cure for HIV infection and more than 35 million people live with the virus worldwide. Although antiretroviral therapy (ART) drastically reduces plasma viremia and improves the quality of life of people living with HIV (Palella et al., 1998), it does not eradicate the virus from the body (Palmer et al., 2008), and causes rebound after ART cessation in all but exceptional cases (Saez-Cirion et al., 2013). HIV persists in cellular reservoirs which are not sensitive to ART, invisible to the host immune system, and in which the virus can persist through latency and possibly residual replication (Chun et al., 1997; Finzi et al., 1997; Wong et al., 1997). In virally suppressed individuals, integrated HIV genomes primarily persist in memory CD4 T cells (Chun et al., 1998a; Finzi et al., 1999; Siliciano et al., 2003) and particularly in subsets of memory CD4 T cells including central (T_CM), transitional (T_TM) and effector (T_EM) memory CD4 T cells, whereas naïve cells (T_N) minimally contribute to the pool of infected cells (Buzon et al., 2014; Chomont et al., 2009; Jaafoura et al., 2014).

Several therapeutic strategies have been proposed to reduce or eradicate HIV cellular reservoirs (Cillo and Mellors, 2016; Deeks et al., 2016). Inducing latency reversal in latently infected cells is one of the suggested approaches to achieve a functional cure for HIV infection (Chun et al., 1998b; Lehrman et al., 2005; Ylisastigui et al., 2004). This strategy consists in forcing viral gene expression by using latency reversing agents (LRAs) while maintaining ART to avoid new rounds of viral replication. Latently infected cells in which HIV expression is induced would be subsequently cleared by the host immune system and/or through viral cytopathic effects. A variety of LRAs have been tested for their ability to reactivate HIV expression in vitro, ex vivo and in vivo (Archin et al., 2009; Bartholomeeusen et al., 2013; Bartholomeeusen et al., 2012; Budhiraja and Rice, 2013; Bullen et al., 2014; Fujinaga et al., 2015; Jiang et al., 2015; Spina et al., 2013; Tsai et al., 2016; Wei et al., 2014; Williams et al., 2004). Even though these LRAs induced increases in cell-associated HIV RNA or in plasma viremia in vivo, none of these clinical interventions led to a significant reduction in the size of the HIV reservoir (Archin et al., 2010; Archin et al., 2008; Archin et al., 2017; Archin et al., 2012; Delagreverie et al., 2016; Elliott et al., 2014; Gutierrez et al., 2016; Rasmussen et al., 2014; Routy et al., 2012; Sagot-Lerolle et al., 2008; Siliciano et al., 2007; Sogaard et al., 2015). This suggests that LRAs alone might not be potent enough to induce viral epitopes exposure and subsequent killing of latently infected cells. In this context, several studies have shown that combinations of LRAs were more potent at inducing viral expression than LRAs alone in vitro and ex vivo (Darcis et al., 2015; Jiang et al., 2015; Laird et al., 2015; Martinez-Bonet et al., 2015; Spivak and Planelles, 2018). While some combinations have been evaluated in clinical trials for cancer therapy (Suraweera et al., 2018), none of these combinatorial interventions have been tested to reactivate latent HIV in vivo yet. Studies assessing the impact of LRAs combinations on viral reactivation ex vivo should facilitate the implementation of combinatorial interventions in clinical trials.

Proviral latency is a multifactorial phenomenon that involves epigenetic factors such as histone deacetylation (Coull et al., 2000; Imai and Okamoto, 2006; Jiang et al., 2007; Marban et al., 2007; Tyagi and Karn, 2007; Williams et al., 2006) and DNA methylation (Blazkova et al., 2009; Kauder et al., 2009; Trejbalova et al., 2016) as well as non-epigenetic mechanisms such as the cytoplasmic sequestration of inducible host transcription factors involved in viral transcription (e.g NF-κB and NFAT) (Baeuerle and Baltimore, 1988; Rao et al., 1997), low levels of the positive transcription elongation factor b (P-TEFb) and its sequestration in a large inactive complex (Chiang et al., 2012; Nguyen et al., 2001; Ramakrishnan et al., 2009; Tyagi et al., 2010), and the presence of micro-RNAs responsible for HIV silencing (Rice, 2015). Of note, the majority of these mechanisms regulating HIV gene expression and latency were originally characterized in cell lines, which are unlikely to recapitulate the complexity of HIV latency in vivo. Although several blocks to HIV transcription have been confirmed in CD4 T cells from virally suppressed individuals (Yukl et al., 2018), whether proviral latency results from similar mechanisms in phenotypically and functionally distinct cellular reservoirs remains largely unknown.

Given that the three subsets of memory CD4 T cells that harbour the majority of integrated genomes have different proliferative capacity, homing potential and activation status (Barski et al., 2017; Durek et al., 2016; Masopust et al., 2001; Okada et al., 2008; Reinhardt et al., 2001; Riou et al., 2007; Sallusto et al., 1999), we hypothesized that latency could result from different mechanisms in these three reservoirs. In this study, we developed flow cytometry-based assays to measure the expression of cellular factors involved in HIV latency in subsets of CD4 T cells (T_N, T_CM, T_TM, T_EM) from ART-suppressed individuals, including levels of histone acetylation (H3K9, H4K5/8/12/16), active NF-κB (phospho-serine529), and active P-TEFb (phosho-serine175 CDK9). We also measured the relative expression of various miRNA involved in the inhibition of HIV replication in the different subsets. In addition, we measured the capacity of LRAs [histone deacetylases inhibitors: vorinostat, panobinostat, romidepsin; PKC agonists: bryostatin, ingenol, prostratin; and P-TEFb inducers: JQ-1] at modulating these factors. Finally, we assessed the ability of LRAs alone or in combinations at inducing viral reactivation in the different subsets of memory CD4 T cells by using a p24 flow cytometry-based assay.

Results

HDAC inhibitors display variable activities in subsets of memory CD4 T cells

It is well established that histone deacetylation contributes to HIV latency in primary CD4 T cells (Coull et al., 2000; Imai and Okamoto, 2006; Jiang et al., 2007; Marban et al., 2007; Tyagi and Karn, 2007; Williams et al., 2006). However, whether this mechanism represses HIV expression to similar levels in different subsets of CD4 T cells harbouring latent proviruses is not known. We developed a flow cytometry-based assay to measure the levels of H3K9 and H4K5/8/12/16 in subsets of CD4 T cells and evaluated the pharmacological activity of three HDACi (vorinostat, panobinostat and romidepsin) in these cellular reservoirs (T_CM, T_TM and T_EM cells).

PBMCs from ART-suppressed individuals were exposed to increasing doses of HDACi and the levels of acetylation of histones H3 and H4 were assessed by flow cytometry (Figure 1, Figure S1). Based on the dose-response curves obtained in total CD4 T cells (Figure 1A, B), we observed that panobinostat and romidepsin displayed lower EC50s than vorinostat (4nM, 9nM and 510nM, respectively), indicating that panobinostat and romidepsin were more potent at inducing histone H4 acetylation in total CD4 T cells than vorinostat (Figure 1C). Maximal fold inductions in the levels of acetylated H4 (top plateau values) were relatively similar for the three HDACi (2.5, 2.8 and 2.5 fold for vorinostat, panobinostat, and romidepsin, respectively, Figure 1D), demonstrating similar efficacies for the three HDACi in total CD4 T cells. Similar results were observed with acetylated histone H3 (Figure S1).

Figure 1. HDAC inhibitors display variable activities in subsets of memory CD4 T cells.

Levels of acetylated histone H4 were determined by intracellular flow cytometry in CD4 T cells (A-D) and gated memory subsets (E-I) from n=8 ART-suppressed individuals. (A) Representative histograms showing the expression of acetylated histone H4 in response to the indicated HDACi (maximal dose, see Star Methods) compared to the non-stimulated condition (NS). (B) Dose-response curves of acetylated histone H4 in response to the 3 HDACi indicated. (C) EC50 and (D) fold induction at the top plateau in total CD4 T cells were determined from the dose-response curves. (E) Representative histograms from one participant showing the expression of acetylated histone in the absence of stimulation in gated CD4 T cell subsets and (F) expression levels in subsets from 8 participants. (G) Representative dose-response curves of acetylated histone H4 in gated CD4 subsets following exposure to the 3 HDACi indicated. (H) EC50 and (I) fold induction at the top plateau in CD4 subsets were determined from the dose-response curves. Tables on the right indicate the subsets in which each drug exerts maximal potency (low EC50, in green) and maximal efficacy (high fold induction, in green). VOR: vorinostat; PNB: panobinostat, RMD: romidepsin. Horizontal bars represent median values. For statistical analyses, Friedman tests with Dunn’s correction for multiple comparisons were used (*adjusted p value <0.05; ** <0.01; ***<0.001). See also Figures S1, S2, S7 and Table S1.

In the absence of LRA, baseline levels of acetylated histone H4 were significantly higher in T_EM cells compared to the other subsets (median MFIs: T_EM = 4583 >T_CM = 4032 >T_TM=3756 >T_N=3387, Figure 1E, F), suggesting that the chromatin was less tightly compacted in T_EM cells.

We assessed the relative capacity of the three HDACi at inducing histone H4 acetylation in different subsets of CD4 T cells. Exposure to HDACi resulted in a dose-dependent increase in H4 acetylation in all subsets (Figure 1G). EC50 analyses revealed that vorinostat was more potent at inducing H4 acetylation in T_CM than in T_EM (median EC50s: 465nM and 929nM, respectively), whereas romidepsin displayed an opposite profile (median EC50s: 11.3nM and 8.1nM in T_CM and T_EM, respectively), revealing a broad diversity of activities between subsets among LRAs from the same class (Figure 1H). No significant differences in EC50 were observed between subsets when using panobinostat, suggesting that this HDACi had similar activity in all CD4 subsets (Figure 1H). For all three HDACi tested, fold inductions at the top plateau tended to be lower in T_TM cells compared to all other CD4 subsets (Figure 1I). Comparable results were obtained when measuring acetylation of histone H3 (Figure S1).

We next sought to determine if the observed differences between CD4 T cell subsets could be attributed to various levels of HDACi uptake across subsets. We exposed sorted cells from each subset to vorinostat (500 nM), panobinostat (5 nM) or romidepsin (10 nM) and measured the intracellular concentration of each HDACi by LC-MS/MS analysis. For the three HDACi tested, the intracellular concentration of the drug did not significantly differ between the different CD4 T cell subsets (Figure S2), excluding the possibility that the differential pharmacological activities of HDACi could be attributed to various levels of drug penetration between the different subsets.

Altogether, these data indicate that the non-selective pan-HDACi vorinostat and the selective class I HDACi romidepsin display opposite activities in different subsets of CD4 T cells. Panobinostat displayed low and similar EC50 across the three memory subsets. Although similar across subsets, maximal fold inductions in H4 acetylation were consistently higher in T_EM cells.

PKC agonists display variable activities in subsets of memory CD4 T cells

Sequestration of NF-κB in the cytoplasm is known to play a key role in the maintenance of HIV latency (Baeuerle and Baltimore, 1988; Colin and Van Lint, 2009). Activation of the PKC pathway results in NF-κB phosphorylation, which causes its translocation to the nucleus and promotes viral reactivation. We measured the ability of four PKC agonists (PMA, bryostatin-1, ingenol-3-angelate and prostratin) at inducing NF-κB activation.

PBMCs from ART-suppressed individuals were exposed to increasing doses of PKC agonists and levels of phospho-NF-κB (pS529) were assessed by flow cytometry. The four PKC agonists induced a dose-dependent increase of NF-κB phosphorylation in total CD4 T cells (Figure 2A, B). As expected, PMA was the most potent at inducing NF-κB phosphorylation (median EC50s: PMA=14nM > ingenol=36nM > bryostatin=45nM > prostratin=2034nM, Figure 2C). In addition, exposure to bryostatin resulted in a lower maximal fold induction in NF-κB phosphorylation compared to the other PKC agonists, while PMA exhibited the highest efficiency (medians: bryostatin=4.2 < ingenol=5.4 < prostratin=5.6 < PMA=6.7, Figure 2D).

Figure 2. PKC agonists display variable activities in subsets of memory CD4 T cells.

Levels of pNF-κB were determined by intracellular flow cytometry in CD8-CD3+ T cells (A-D) and gated memory subsets (E-I) from n=8 ART-suppressed individuals. (A) Representative histograms showing the expression of pNF-κB in response to the indicated agonists of PKC (maximal dose, see Star Methods) compared to the non-stimulated condition (NS). (B) Dose-response curves of pNF-κB in response to the 4 PKC agonists indicated. (C) EC50 and (D) fold induction at the top plateau in total CD4 T cells were determined from the dose-response curves. (E) Representative histograms from one participant showing the expression of pNF-κB in the absence of stimulation in gated CD4 subsets and (F) expression levels in subsets from 8 participants. (G) Representative dose-response curves of pNF-κB in gated CD4 subsets following exposure to the 4 PKC agonists indicated. (H) EC50 and (I) fold induction at the top plateau in CD4 subsets were determined from the dose-response curves. The tables on the right indicate the subsets in which each drug exerts maximal potency (low EC50, in green) and maximal efficacy (high fold induction, in green). Bryo: bryostatin, Ing: ingenol, Pro: prostratin. Horizontal bars represent median values. For statistical analyses, Friedman tests with Dunn’s correction for multiple comparisons were used (*adjusted p value <0.05; ** <0.01; ***<0.001). See also Figure S7 and Table S1.

In the absence of stimulation, baseline levels of pNF-κB were significantly higher in T_CM cells compared to T_TM and T_EM cells (median MFIs: 1387, 1233 and 1155, respectively, Figure 2E, F).

We next measured the relative ability of these four PKC agonists at inducing NF-κB phosphorylation in the different subsets of CD4 T cells. Stimulation of CD4 T cells with PKC agonists resulted in a dose-dependent increase of NF-κB phosphorylation in all CD4 subsets (Figure 2G). Determination of EC50 revealed that PMA and prostratin were more potent at inducing NF-κB phosphorylation in T_CM than in T_EM cells (median EC50s for PMA: 12.1nM and 16.9nM, respectively; for prostratin: 1.9μM and 2.3μM, respectively). Bryostatin and ingenol displayed similar EC50s across the three memory subsets, whereas significantly higher EC50s were observed in the T_N subset compared to T_TM and T_EM cells (p<0.05, Figure 2H). Interestingly, fold inductions at the top plateau were consistently higher in T_N and T_CM cells compared to T_EM cells for the four PKC agonists tested (p<0.05, T_N>T_CM>T_TM>T_EM)(Figure 2I).

Taken together, these results demonstrate that PMA and prostratin have a higher potency in T_CM cells compared to T_EM cells whereas bryostatin and ingenol display similar potencies across subsets. All the PKC agonists we tested displayed a higher efficacy in T_CM cells compared to more differentiated subsets.

PKC agonists are more potent at activating P-TEFb in differentiated subsets

In latently infected CD4 T cells, HIV transcriptional elongation is restricted by low levels of P-TEFb (composed of CDK9 and Cyclin T1) and its sequestration in an inactive complex (Chiang et al., 2012; Hoque et al., 2011; Nguyen et al., 2001; Ramakrishnan et al., 2009; Tyagi et al., 2010).

We exposed PBMCs from ART-suppressed individuals to increasing doses of the P-TEFb inducer JQ-1 and PKC agonists, and measured levels of pS175 CDK9, a surrogate of P-TEFb activation (Mbonye et al., 2013), by flow cytometry. As expected, JQ-1 displayed no measurable activity on CDK9 phosphorylation in these resting CD4 T cells (data not shown), which are largely devoid of P-TEFb (Chiang et al., 2012; Hoque et al., 2011; Tyagi et al., 2010 ). In contrast, the PKC agonists bryostatin and ingenol induced a dose-dependent increase in CDK9 phosphorylation in total CD4 T cells (Figure 3A, B). Exposure to bryostatin resulted in a lower EC50 compared to ingenol (median EC50s: 8.4nM and 19.2nM, respectively, Figure 3C) and a slightly higher fold induction at the top plateau (medians: 2.6 and 2.0, respectively; Figure 3D).

Figure 3. PKC agonists are more potent at activating P-TEFb in differentiated subsets.

Levels of pCDK9 (pS175) were determined by intracellular flow cytometry in CD4 T cells (A-D) and gated memory subsets (E-I) from n=8 ART-suppressed individuals. (A) Representative histograms showing the expression of pCDK9 in response to the indicated agonists of PKC (maximal dose, see Star Methods) compared to the non-stimulated condition (NS). (B) Dose-response curves of pCDK9 in response to the 2 PKC agonists indicated. (C) EC50 and (D) fold induction at the top plateau in total CD4 T cells were determined from the dose-response curves. (E) Representative histograms from one participant showing the expression of pCDK9 in the absence of stimulation in gated CD4 subsets and (F) expression levels in subsets from 8 participants. (G) Representative dose-response curves of pCDK9 in response to the 2 PKC agonists indicated in gated CD4 subsets. (H) EC50 and (I) fold induction at the top plateau in CD4 subsets were determined from the dose-response curves. The tables on the right indicate the subsets in which each drug exerts maximal potency (low EC50, in green) and maximal efficacy (high fold induction, in green). Bryo: bryostatin, Ing: ingenol. Horizontal bars represent median values. For statistical analyses, Friedman tests with Dunn’s correction for multiple comparisons were used (*adjusted p value <0.05; ** <0.01; ***<0.001). See also Figure S7 and Table S1.

In the absence of stimulation, pCDK9 was expressed at higher levels in T_TM cells compared to all other subsets (median MFIs: T_TM=1237 > T_CM=1088 > T_EM=842 > T_N=640, Figure 3E, F). Stimulation with all PKC agonists resulted in a dose-dependent increase in CDK9 phosphorylation in all CD4 subsets (Figure 3G). Bryostatin and ingenol tended to be more potent at inducing CDK9 phosphorylation in the T_EM compartment compared to T_TM, T_CM and T_N cells (median EC50s for bryostatin: 2.7nM, 4.3nM, 9.4nM and 14.1nM, respectively; for ingenol: 5.1nM, 6.6nM, 20.2nM, and 50.3nM, respectively; Figure 3H). For both PKC agonists, fold inductions at the top plateau were consistently lower in T_TM cells compared to all other subsets (Figure 3I).

Altogether, these results show that baseline levels of pCDK9 are significantly different between memory subsets, with T_TM cells displaying the highest levels of pCDK9. This resulted in a lower fold induction at the top plateau in T_TM cells compared to the other CD4 subsets. EC50 analyses revealed that the potency of bryostatin and ingenol is maximal in the most differentiated subsets.

miR125b and miR155 are expressed at different levels in CD4 T cell subsets

Several miRNA have the ability to inhibit HIV replication and might consequently be involved in viral latency (Rice, 2015). We measured the relative expression of several cellular miRNA (125b, 150, 155, 196b, 223, 27b, 28, 29b) by RT-qPCR in CD4 T cell subsets. Two miRNAs were expressed at significantly different levels between subsets. miR125b, which was previously described to bind the 3’ end of HIV transcripts and to contribute to HIV latency in resting T cells (Huang et al., 2007), was expressed at higher levels in the less differentiated T_N and T_CM subsets compared to the more differentiated T_EM cells (p<0.01 for the comparison T_N:T_EM). In contrast, miR155, which was shown to reinforce proviral latency (Ruelas et al., 2015), was expressed at higher levels in T_EM cells (p<0.01 for the comparison T_N:T_EM and p<0.05 for T_CM:T_EM)(Figure 4). Although miR196b was expressed at different levels between memory subsets, these differences did not achieve statistical significance. Altogether these data indicate that several miRNAs involved in HIV latency are differentially expressed between subsets.

Figure 4. miR125b and miR155 are expressed at different levels in distinct memory subsets.

The indicated miRNAs were quantified by RT-qPCR in subsets of CD4 T cells obtained by flow cytometry cell sorting from n=5 virally suppressed individuals. Levels of expression of each miRNA were normalized to the levels of GAPDH reference gene. Means and standard deviations are represented. For each miRNA, Friedman tests with Dunn’s correction for multiple comparisons were used (*adjusted p value <0.05; ** <0.01). See also Table S2.

PKC agonists in combination with HDACi have enhanced capacity to reactivate HIV in T_CM cells

We previously developed a flow-cytometry based assay to quantify and characterize latently infected cells (Pardons et al., 2019). We took advantage of the HIV-Flow assay to measure the frequency of p24-producing cells in each subset upon reactivation with a panel of LRAs and their combinations. Among the HDACi, we selected the non-selective pan-HDACi panobinosat, and the selective class I HDACi romidepsin which both displayed low EC50 compared to vorinostat (Figure 1C). Among the different PKC agonists, we selected ingenol because of its high potency compared to bryostatin and prostratin (Figure 2C), and because of its relatively similar activity in each memory subset (Figure 2H).

We used relatively high doses of LRAs to achieve maximal efficacy for histone acetylation, PTEF-b induction and NF-kB activation. Importantly, these concentrations did not induce significant levels of cytotoxicity (Figure S3A). Panobinostat and romidepsin (100nM) had limited to no effect on latent HIV (medians of the % of maximal reactivation by PMA and ionomycin: 1.6%, 0.3%, respectively; Figure 5). In contrast, the PKC agonist ingenol (500nM) induced viral reactivation in samples from all participants (median % of maximal reactivation = 18.5%; Figure 5). Combinations of ingenol with an HDACi further increased the frequencies of p24+ cells (median % of maximal reactivation: 36% for panobinostat+ingenol and 32.6% romidepsin+ingenol; Figure 5). Intriguingly and consistent with our aforementioned observations reporting no effect of JQ-1 on PTEF-b activation, the addition of JQ-1 to the combination panobinostat+ingenol did not increase the frequency of p24+ cells (Figure S3B).

Figure 5. Combinations of LRAs are more potent at reactivating HIV than LRAs alone.

HIV-Flow was used to measure the frequency of p24+ cells upon reactivation of latent proviruses. CD4 T cells isolated from n=9 ART-suppressed individuals were exposed to LRAs (PNB: 100nM panobinostat, RMD: 100nM romidepsin, ING: 500nM ingenol). Results are represented as the relative percentage of maximal reactivation obtained with PMA/ionomycin (100%). Medians and interquartile ranges are shown. For statistical analyses, non-parametric Wilcoxon tests were used; statistically significant p values are indicated on the graphs. See also Figure S3 and Table S1.

To assess the activity of the LRAs and their combinations in different cells subsets, we analysed the memory phenotype of p24-producing cells following reactivation (Figure 6 and Figures S4). Of note, the proportions of CD4 subsets were preserved when stimulation with LRAs was performed in the presence of the protein transport inhibitor brefeldin A (BFA; Figure S5). As shown previously (Pardons et al., 2019), T_TM and T_EM cells were the main contributors to the pool of p24-producing cells following PMA/ionomycin stimulation (Figure S6A). In contrast T_N and T_TD cells minimally contributed to the translation competent reservoir, which did not allow us to evaluate the effect of LRAs in these subsets. Following reactivation with ingenol, and LRA combinations, the T_TM and T_EM subsets often encompassed higher frequencies of p24+ cells than T_CM cells (Figure 6A). Combinations of LRAs led to higher frequencies of p24-producing cells compared to LRAs alone in all subsets (Figure S6B).

Figure 6. PKC agonists in combination with HDACi have enhanced capacity to reactivate HIV in T_CM cells.

HIV-Flow was used to characterize the phenotype of p24+ cells upon reactivation of latent proviruses. CD4 T cells isolated from n=9 ART-suppressed individuals were exposed to LRAs (PNB: 100nM panobinostat, RMD: 100nM romidepsin, ING: 500nM ingenol). (A) Frequency of p24+ cells per million T_CM, T_TM, T_EM cells. Grey columns depict median values. (B) Percentage of maximal reactivation obtained with PMA/ionomycin in memory CD4 T cell subset. The level of reactivation obtained with PMA/ionomycin (100%) is represented by a dotted horizontal line. Grey columns depict median values. For statistical analyses, non-parametric Wilcoxon tests were used; statistically significant p values are indicated on the graphs. See also Figures S4, S5, S6 and Table S1.

We then expressed the frequencies of p24+ cells from each subset in each condition relative to the frequencies measured following activation with PMA/ionomycin (Figure 6B, Figure S6C). Although ingenol tended to have a slightly higher capacity to reactivate HIV in T_CM cells, no significant differences were observed between subsets (medians = 27.4%, 14.0%, 10.9% for T_CM, T_TM, T_EM, respectively; Figure 6B). Strikingly, the combination of ingenol with an HDACi induced potent HIV reactivation in T_CM cells to levels comparable or higher than those measured with PMA/ionomycin (medians = 104.1% and 73.7% for panobinostat+ingenol and romidepsin+ingenol, respectively; Figure 6B, Figure S6C). Of note, reactivation was more efficient in T_CM cells compared to the other subsets regardless of the contribution of the T_CM subset to the translation-competent reservoir (data not shown). Interestingly, this synergistic effect was restricted to T_CM cells and not observed in the T_TM and T_EM subsets (Figure 6B, Figure S6C).

Altogether, our results indicate that combinations of LRAs are more potent at reactivating HIV than LRAs alone and demonstrate that this phenomenon is in part attributed to a synergistic effect of these combinations in latently infected T_CM cells.

Discussion

The three major cellular compartments in which HIV persists differ in their proliferative capacities, homing potentials and cytokine production profiles (Barski et al., 2017; Durek et al., 2016; Masopust et al., 2001; Okada et al., 2008; Reinhardt et al., 2001; Riou et al., 2007; Sallusto et al., 1999). Given this diversity, we hypothesized that mechanisms driving HIV latency may also differ between these three subsets. We observed that factors involved in HIV latency and reactivation (acetylated histones H3/H4, pNF-κB, pCDK9, miR125b and miR155) are expressed at significantly different levels between these subsets. In addition, and as reported in a recent study (Grau-Exposito et al., 2019), our results show that the majority of commonly used LRAs display different activities in distinct reservoirs.

Multiple studies have shown that histone deacetylation plays an important role in HIV latency. During proviral latency, HDACs are recruited to the HIV promoter by various transcription factors leading to chromatin compaction and transcriptional repression (Coull et al., 2000; Imai and Okamoto, 2006; Jiang et al., 2007; Marban et al., 2007; Tyagi and Karn, 2007; Williams et al., 2006). In this study, we observed that baseline levels of acetylated histones H3/H4 are higher in T_EM cells, which is in agreement with another study showing that memory CD4 T cell differentiation is associated with a progressive loss of DNA methylation and progressive changes in DNA accessibility and transcriptomic profile (Durek et al., 2016). Several HDACi have been shown to reactivate HIV expression in vitro, ex vivo, and in vivo (Archin et al., 2010; Archin et al., 2008; Archin et al., 2009; Archin et al., 2017; Archin et al., 2012; Elliott et al., 2014; Rasmussen et al., 2014; Routy et al., 2012; Sagot-Lerolle et al., 2008; Siliciano et al., 2007; Sogaard et al., 2015; Tsai et al., 2016; Wei et al., 2014). We observed that panobinostat and romidepsin are more potent at inducing histone acetylation than vorinostat, as demonstrated by their relatively low EC50 in CD4 T cells. This is consistent with our phamacodynamic measures showing a higher percentage of HDACi uptake with panobinostat and romidepsin compared to vorinostat (data not shown). Vorinostat and panobinostat are belonging to the class of hydroxamic acids and are acting on class I and II HDACs, whereas romidepsin is a cyclic peptide and is specific to class I HDACs (Xu et al., 2007). Multiple studies have reported different activities between these two classes of HDACi. In fact, vorinostat exhibits little to no inhibition of CD8 cytotoxic functions, while romidepsin has a pronounced inhibitory effect on CTL killing (Jones et al., 2014). Moreover, unlike vorinostat, romidepsin is a substrate for the efflux transporter MDR-1 (multidrug resistance protein 1) (Ni et al., 2015). In our study, we observed that vorinostat is more potent at inducing histone acetylation in T_CM cells than in T_EM cells whereas romidepsin displays the opposite trend. Although these differences between subsets were sometimes modest, they were repeatedly observed across the samples. Future in vivo studies assessing histone acetylation in response to HDACi within distinct CD4 T cell subsets are warranted to confirm the biological relevance of our in vitro data. Of note, the differential activities of vorinostat and romidepsin were not explained by a differential uptake of HDACi between memory subsets. Whether these differential activities may be attributed to differences in the relative proportions of classI/II HDACs between T_CM and T_EM cells remains to be determined.

Another mechanism involved in HIV latency is the sequestration of the transcription factor NF-κB in the cytoplasm by its inhibitor IκB (Baeuerle and Baltimore, 1988). At baseline, we measured higher levels of pNF-κB in T_CM cells than in T_EM cells, which is in line with the increased survival potential of T_CM cells compared to T_EM cells (Macallan et al., 2004; Riou et al., 2007; Sallusto et al., 2004). In fact, the NF-κB pathway is known to promote T cell survival by increasing the expression of several genes involved in the inhibition of the apoptotic machinery (Luo et al., 2005). PKC agonists induce HIV expression in vitro and ex vivo (Bullen et al., 2014; Jiang and Dandekar, 2015; Jiang et al., 2015; Mehla et al., 2010; Spina et al., 2013; Williams et al., 2004). Bryostatin has been tested in a clinical trial but failed to increase cell-associated HIV RNA, probably due to low plasma concentrations (Gutierrez et al., 2016). By measuring EC50 of several PKC agonists, we found that ingenol and bryostatin are more potent at activating NF-κB than prostratin, which is consistent with the fact that prostratin has the lowest capacity to bind PKC, whereas bryostatin displays a high binding activity to PKC (Bogi et al., 1998). By analysing the effect of each PKC agonist in distinct memory subsets, we observed that PMA and prostratin are more potent at inducing NF-κB phosphorylation in T_CM than in T_EM cells, while no significant differences in EC50 were observed between subsets for bryostatin and ingenol. These differences between PKC agonists might be explained by a differential selectivity towards PKCs. Indeed, PMA and prostratin are phorbol esters that bind to the diacylglycerol binding domain of conventional (PKCα, PKCβ, and PKCγ) and novel PKCs (PKCδ, PKCε, PKCη, and PKCθ) leading to their activation (Beans et al., 2013). In contrast, bryostatin is a selective activator of the conventional PKCα and the novel PKCδ (Mehla et al., 2010), and ingenol-3-Angelate activates PKCδ while inhibiting PKCα (Benhadji et al., 2008). Interestingly, other studies have reported opposite effects of bryostratin and prostratin on various CD8 and NK cell functions (Clutton and Jones, 2018; Desimio et al., 2018). Future studies will be needed to measure the relative proportions of the different PKC in memory CD4 T cell subsets, which may underlie our observations. In spite of these differences in potency, all PKC agonists displayed a higher efficacy (fold induction at the top plateau) in the T_CM compared to the T_EM subset, which is consistent with the elevated proliferation potential of long-lived T_CM cells compared to other memory subsets (Sallusto et al., 2004; Sallusto et al., 1999).

Another non-epigenetic mechanism contributing to HIV latency is the restricted levels of the elongation factor P-TEFb (composed of CDK9 and Cyclin T1) in resting memory cells (Tyagi et al., 2010) due to high levels of miRNAs targeting Cyclin T1 (Chiang et al., 2012; Hoque et al., 2011) and to the sequestration of P-TEFb in a large inactive complex (Nguyen et al., 2001). Consequently, activation of the P-TEFb pathway requires to increase the quantity of Cyclin T1 and to dissociate P-TEFb from its inactive complex (Bartholomeeusen et al., 2013), which leads to the phosphorylation of CDK9 on Ser175 (Mbonye et al., 2013). Bromodomains inhibitors such as JQ-1 are known to act by releasing P-TEFb from its inactive complex and by limiting the interaction between P-TEFb and the bromodomain BRD4, which favours the binding of the HIV Tat protein to the active P-TEFb (Bartholomeeusen et al., 2012). In our study, P-TEFb inducers had no measurable impact on the levels of activated P-TEFb, which may be explained by the low levels of P-TEFb in resting CD4 T cells (Chiang et al., 2012; Hoque et al., 2011). However, the PKC agonists bryostatin and ingenol rapidly induced the phosphorylation of CDK9, which is in line with previous studies showing that PKC agonists increase the levels of P-TEFb and favour the release of P-TEFb from its inactive complex (Bartholomeeusen et al., 2013; Fujinaga et al., 2012; Pandelo Jose et al., 2014; Sung and Rice, 2006). We observed that baseline levels of pCDK9 were the lowest in naïve cells, which is consistent with their lower transcriptional activity compared to memory subsets (Durek et al., 2016). PKC agonists were more potent at inducing CDK9 phosphorylation in T_EM cells compared to less differentiated subsets, which may contribute to their capacity to mount immediate effector responses following antigen stimulation (Sallusto et al., 2004; Sallusto et al., 1999).

Several cellular miRNAs have been reported to exert an inhibitory effect on HIV-1 replication by targeting host cell HIV dependency factors (miR27b, 29b, 150, 155, 223) and/or by directly targeting HIV transcripts (miR28, 29b, 125b, 196b, 223) (Rice, 2015). Our results indicate that miR125b is expressed at lower levels in the more differentiated subsets which is in line with previous studies showing that CD4 activation is associated with a downregulation of this miRNA and increased HIV replication (Huang et al., 2007). Therefore, the lower levels of miR125b in T_EM cells may contribute to their heightened susceptibility to HIV infection (Groot et al., 2006). The opposite trend was observed when we measured miR155, which was expressed at higher levels in the more differentiated subsets. miR155 is known to suppress the expression of the HIV activator TRIM32, which activates NF-κB by inducing the degradation of its inhibitor IκB (Ruelas et al., 2015). Therefore, the heightened expression of miR155 in T_EM cells is consistent with the lower levels of active NF-κB we measured in this subset.

LRAs from several classes have been evaluated in vivo for their ability to reactivate HIV from latency. Disappointingly, most of these clinical interventions failed at reducing the size of the HIV reservoir (Archin et al., 2010; Archin et al., 2008; Archin et al., 2017; Archin et al., 2012; Delagreverie et al., 2016; Elliott et al., 2014; Gutierrez et al., 2016; Rasmussen et al., 2014; Routy et al., 2012; Sagot-Lerolle et al., 2008; Siliciano et al., 2007; Sogaard et al., 2015). This suggests that LRAs alone may not be potent enough to reactivate all latently infected cells or to induce sufficiently high levels of viral antigens for subsequent killing of these infected cells. Several studies have shown that combinations of LRAs are more efficient at reactivating HIV from latency than LRAs alone in cell lines and ex vivo cell cultures from ART-treated individuals (Darcis et al., 2015; Laird et al., 2015; Martinez-Bonet et al., 2015). Our results indicate that LRAs from the same class often display different activities in distinct subsets of memory CD4+ T cells, suggesting that using multiple LRAs from the same family may increase the likelihood of targeting multiple reservoirs simultaneously. Moreover, and as shown previously in several studies, our results suggest that LRAs synergize to reverse HIV latency (Darcis et al., 2015; Jiang et al., 2015; Laird et al., 2015; Martinez-Bonet et al., 2015; Spivak and Planelles, 2018). In a recent study, Grau-Expósito et al identified romidepsin+ingenol as a powerful combination to efficiently induce HIV RNA production in CD4 T cells, which is consistent with our observations. In contrast, they reported an antagonist effect of the combination of panobinostat+ingenol on HIV reactivation (Grau-Exposito et al., 2019), which we did not observe in our study. These contrasting data may result from differences in the kinetics of exposure to LRAs, from differences in the concentrations of compounds used in these two studies, or from differences in the characteristics of the participants from which samples were collected.

In this study, we also characterized the cell subsets in which this synergism may be critical for successful latency reversal. Indeed, we took advantage of a flow-cytometry based assay (HIV-Flow) to determine the frequency of p24-producing cells in total CD4 T cells and memory CD4 T cell subsets after reactivation with LRAs alone or in combination. High doses of panobinostat, romidepsin and JQ-1 had minimal impact on HIV reactivation in total CD4 T cells, while the PKC agonist ingenol reactivated HIV in samples from all participants. This is in line with the study by Bullen et al. showing that at clinically relevant concentrations, only PKC agonists are able to increase virus release and intracellular HIV-1 mRNA (Bullen et al., 2014). As reported in a previous study (Spivak et al., 2015), we did not observe an impact of the PKC agonist ingenol on the levels of histone acetylation (data not shown), suggesting that its anti-latency effect is mainly mediated through the activation of the NF-κB and P-TEFb pathways. As expected, combinations of a PKC agonist with an HDACi induced higher levels of viral reactivation than LRAs alone. Darcis et al. (Darcis et al., 2015) and Laird et al. (Laird et al., 2015) showed that the addition of JQ-1 to PKC agonists improved viral production. In our study, the addition of JQ-1 to the combination of panobinostat and ingenol did not increase reactivation. The high doses of panobinostat and ingenol used in our study may not have allowed us to observe the additional effect of JQ-1. Interestingly, even when used at high doses, LRAs combinations only reached 30% of the maximal reactivation obtained with PMA/ionomycin, questioning the possibility of achieving latency reversal in all infected cells with currently available LRAs.

Following reactivation with PMA/ionomycin and ingenol alone or in combination, the T_TM and T_EM subsets often encompassed higher frequencies of p24+ cells than T_CM cells. This is consistent with the data obtained by Lassen et al. in a primary cell model of HIV latency (Lassen et al., 2012) and with the higher frequency of cells harbouring intact HIV proviruses in the T_EM subset (Hiener et al., 2017). Of note, ingenol alone failed to induce p24 production in T_CM cells from 4 out of 9 participants. This is likely attributed to the low contribution of this subset to the pool of latently infected cells in these individuals (<3%, data not shown), which highlights the need to study samples from participants with various contributions of T_CM, T_TM and T_EM cells to the total reservoir to adequately assess latency reversal in all cellular reservoirs.

Finally, our study reveals that combining panobinostat or romidepsin with ingenol dramatically increased latency reversal in T_CM cells to levels similar or even higher to those obtained with PMA/ionomycin. In contrast, these combinations only had a modest effect in the T_TM and T_EM subsets when compared to PMA/ionomycin reactivation. This suggests that histone hypoacetylation may be a major contributor to HIV latency in T_CM cells whereas this mechanism may be less involved in the silencing of HIV genomes in T_TM and T_EM cells. Future studies will be needed to identify a combination of LRAs inducing reactivation of HIV as efficiently as PMA/ionomycin not only in T_CM cells but in all subsets of memory CD4 T cells.

In conclusion, this study highlights the importance of considering the HIV reservoir as multiple and heterogeneous pools of latently infected cells that respond differently to LRA and paves the way for the development of a combination of LRAs that would reactivate latent HIV in all subsets.

STAR Methods

LEAD CONTACT AND MATERIALS AVAILIBILITY

Further information and requests for reagents may be directed to and will be fulfilled by the Lead Contact, Nicolas Chomont (nicolas.chomont@umontreal.ca).

This study did not generate new unique reagents.

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Samples from 18 HIV-infected individuals on stably suppressive ART for a median time of 9.3 years were used in this study. All participants signed informed consent forms approved by the McGill University Health Centre, the Centre Hospitalier de l’Université de Montréal and the Martin Memorial Health Systems review boards. Participant’s characteristics are summarized in Table 1.

Table 1.

Characteristics of study participants

ID	Age (years)	Gendera	CD4 (cells/μL)	CD8 (cells/μL)	CD4/CD8 Ratio	Viral Load (HIV RNA copies/mL)	Time to ART initiation (years)	Estimated time of infection (years)b	Time on ART (years)
ST114	40	M	1145	1410	0.8	< 20	0.4	6.2	5.9
ST116	47	M	772	921	0.8	< 20	2.6	9.4	6.8
ST120	48	M	686	792	0.9	< 20	6.0	15.4	9.4
ST135	32	M	734	703	1.0	< 20	4.9	7.4	2.5
ST140	36	M	1102	875	1.3	< 20	N.Ac	N.A	N.A
ST142	46	F	1030	1192	0.9	< 20	N.A	18.6	N.A
ST144	56	M	715	709	1.0	140	N.A	27.6	N.A
ST145	46	M	1897	830	2.3	< 20	N.A	19.7	N.A
ST105C	60	M	677	788	0.9	< 20	N.A	N.A	N.A
ST137B	41	M	267	518	0.5	< 20	N.A	N.A	N.A
CHO-17	48	M	662	658	1.0	˂ 40	0.2	16.9	16.7
CHO-18	36	M	882	1177	0.7	˂ 40	4.7	13.8	9.1
CHO-20	55	M	847	613	1.4	˂ 40	15.1	29.3	14.1
CHO-25	60	M	803	422	1.9	˂ 40	4.8	25.1	20.4
CHO-32	47	M	625	511	1.2	˂ 40	0.3	15.1	14.8
CHO-34	36	M	471	322	1.5	˂ 40	7.8	12.0	4.2
CHO-37	58	M	602	349	1.7	< 20	1.0	9.2	7.9
CHO-38	57	M	1173	1209	1.0	< 20	0.6	20.7	20.1
Median	47	-	753	749	1.0	-	3.7	15.4	9.3
IQ range	[40–56]	-	[666–993]	[542–910]	[0.9–1.4]	-	[0.5–5.2]	[10.7–20.2]	[6.5–15.3]

^aF = female, M = male

^bTime of infection was calculated from the date of diagnosis

^cN.A = not available

METHOD DETAILS

Cell isolation

All participants underwent leukapheresis to collect large numbers of PBMCs. PBMCs were isolated by Ficoll density gradient centrifugation and were cryopreserved in liquid nitrogen.

Latency reversing agents

Panobinostat (S1030) and romidepsin (S3020) were obtained from Selleckchem. Phorbol 12-myristate 13-acetate (PMA, P8139), ionomycin (I9657), ingenol-3-angelate (SML-1318) and prostratin (P0077) were obtained from Sigma-Aldrich. Bryostatin was obtained from Enzo Life Sciences (BML-ST103–0010). Vorinostat was provided by Merck. JQ-1 was obtained from BioVision (2070). All compounds were resuspended in DMSO and stored at −20°C or −80°C until use. Importantly, we ensured that the vehicular control DMSO had no measurable effect on any of the readouts used in this study (histone acetylation, pNF-kB, pCDK9 and HIV reactivation, data not shown).

Antibodies

Live/Dead Aqua Cell Stain (L/D) (405nm) was obtained from Thermo Fisher Scientific (L34957). Antibodies to cell surface receptors (CD3, CD4, CD8, CD45RA, CD27, CCR7) are listed in Supplementary Table 1. Anti-acetyl histone H3 was obtained from Cell Signaling Technology (9683S) and anti-acetyl H4 (05–1355) was obtained from EMD Millipore. Anti-pS529 p65 NF-kB was obtained from BD Bioscience (558423). Anti-pS175 CDK9 was kindly provided by Jonathan Karn. KC57-PE was obtained from Beckman Coulter (6604667) and p24 28B7-APC was obtained from MediMabs (MM-0289-APC).

Antibody labeling

Anti-acetyl H4 and anti-pS175 CDK9 were labeled with the Mix-n-Stain CF647 Antibody Labeling kit from Sigma-Aldrich (MX647S100) according to the manufacturer’s instructions.

Gating strategy for flow cytometry analysis

The gating strategy is shown in Figure S7. Live CD4+ T cells were defined as (CD3+/L/D-/CD8-/CD4+). CD4 T cell subsets were defined using the following gating strategy: T_N (CD45RA+/CD27+/CCR7+); T_CM (CD45RA-/CD27+/CCR7+); T_TM (CD45RA-/CD27+/CCR7-); T_EM (CD45RA-/CD27-/CCR7-) and T_TD (CD45RA+/CD27-/CCR7-). Of note, due to the low numbers of T_TD cells acquired by flow cytometry in most of the samples, this subset was excluded from all analyses.

Measure of histones H3 and H4 acetylation

The flow-cytometry based assay used to detect H3/H4 acetylation was adapted from pre-existing methods (Archin et al., 2012; Rigby et al., 2012; Sogaard et al., 2015). PBMCs were exposed to 4nM-62.5μM vorinostat, 0.04nM-625nM panobinostat and 0.15nM-640nM romidepsin for 4h at 37°C in complete culture medium. Cells were washed and stained (45min, 4°C) with the Aqua LIVE/DEAD staining kit together with antibodies to cell surface proteins (CD3, CD4, CD8, CD45RA, CD27, CCR7). After a 15min-fixation step with formaldehyde 4% (Fischer Scientific, F79–500), cells were permeabilized (45min, 4°C) with the Perm Wash buffer (BD Biosciences, 554723), and stained (45min, RT) with anti-acetylated H3 and H4 antibodies. Levels of acetylated H3 and H4 in gated subsets of CD4 T cells were determined by flow cytometry on a BD LSR II.

Measure of NF-κB activation

PBMCs were rested overnight in complete medium (RPMI 1640, Life Technologies) supplemented with 10% Fetal Bovine Serum (Wisent) and 1% Penicillin/Streptomycin (Life Technologies) and exposed to 0.27nM-350nM PMA, 0.69nM-900nM bryostatin, 0.69nM-900nM ingenol or 0.069μM-90μM prostratin for 15min at 37°C in PBS/2% Human Serum (Atlanta Biologicals, 540110). Antibody to CCR7 and LIVE/DEAD reagent were added to the culture medium during the stimulation. Cells were then fixed (10min, 37°C) with the Cytofix Fixation Buffer (BD Biosciences, 554655) and stained (1h, room temperature) with antibodies to cell surface markers (CD3, CD4, CD8 and CD27). Cells were permeabilized (20min, on ice) with the ice-cold Permeabilization Buffer III (BD Biosciences, 558050), and rehydrated (15min, on ice) in PBS/Human Serum 10%. Finally, PBMCs were stained (30min, RT) with anti-pS129 NF-κB and anti-CD45RA antibodies. Levels of pNF-κB in gated subsets of CD4 T cells were determined by flow cytometry on a BD LSR II.

Measure of P-TEFb activation

PBMCs were rested overnight in complete medium (RPMI 1640, Life Technologies) supplemented with 10% Fetal Bovine Serum (Wisent) and 1% Penicillin/Streptomycin (Life Technologies). After a 20min-incubation at 4°C, cells were exposed to 10μM JQ-1, 0.08nM-250nM bryostatin or 0.08nM-250nM ingenol for 30min at 37°C in PBS/2% Human Serum. Cells were then stained (30min, 4°C) with the Aqua LIVE/DEAD staining kit together with antibodies for cell surface markers (CD3, CD4, CD8, CD45RA, CD27, CCR7). After a 20min-fixation with methanol free formaldehyde 4% (Polysciences, 04018–1), cells were permeabilized (30min, 4°C) with the Perm Wash buffer (BD Biosciences, 554723) and stained (30min, RT) with the anti-pS175 CDK9 antibody. Levels of pCDK9 in gated subsets of CD4 T cells were determined by flow cytometry on a BD LSR II.

Measures of miRNA levels

200×10⁶ PBMCs from 5 participants were used to isolate CD4 T cells by negative magnetic selection with the EasySep™ Human CD4 T Cell Enrichment Kit (StemCell technology, 19052). Purified CD4 T cells were stained with Aqua LIVE/DEAD, CD3, CD4, CD45RA, CD27 and CCR7 in order to sort the T_N, T_CM, T_TM and T_EM subsets on a BD Aria III. After sorting, cells were subjected to RNA extraction (mirVana miRNA Isolation Kit, ThermoFisher, AM1560). RNAs from each subset were polyadenylated and retrotranscribed according to the manufacturer’s instructions (miRNA 1^st strand cDNA Synthesis kit, Agilent Technologies, 600036). Quantitative real time PCR (PowerUp SYBRGreen Master Mix, Thermo Fisher, A2574) was used to quantify each miRNA and GAPDH was used as reference gene. Sequences of oligonucleotides used to amplify miRNA were designed as previously described by Balcells et al. (Balcells et al., 2011) and are indicated in Supplementary Table 2. Amplification of miRNA and GAPDH was carried out in a 10-μl reaction mixture comprising 2μL cDNA, 5μL SYBRGreen Master Mix, and 1 μL of the forward and reverse primers (500nM). The PCR cycle conditions were as follows: Uracil-DNA glycosylase (UDG) activation at 50°C for 2min, Taq polymerase activation at 95°C for 2min, and 50 cycles of amplification (95°C for 1s, 60°C for 30s), followed by a melt curve analysis (95°C for 15s, 60°C for 1min, 95°C for 15s). Amplification was performed on the QuantStudio5 instrument (Applied Biosystems). Relative expression of miRNA in each subset was determined following normalization on the levels of GAPDH transcripts.

Determination of HDACi uptake in CD4 T cell subsets

400×10⁶ PBMCs from 2 participants were thawed and T_N, T_CM, T_TM and T_EM subsets were sorted as described above on a BD Aria III. Each subset was exposed to 500nM vorinostat, 5nM panobinostat, and 10nM romidepsin (corresponding to EC50 measured in total CD4 T cells) for 4h at 37°C in complete medium. Intracellular concentration of the active forms of the HDACi was determined by liquid chromatography–mass spectrometry (LC-MS/MS) analysis in each subset. Briefly, protein precipitation with methanol was used to extract LRAs from CD4 T cells. 10ng/mL glibenclamide was used as internal standard solution for all LRAs. For vorinostat and panobinostat, 50μL of internal standard solution were added to 200μL of cell lysate. The mobile phase consisted of 0.1% formic acid in acetonitrile (A) and 10mM ammonium formate pH 3 (B); ratio A:B varied from 8:92 to 90:10 (v/v), at a flow rate of 300μL/min. 10μL of the extract were injected into LC-MSMS system. For romidepsin, 250μL of internal standard solution were added to 500μL of cell lysate. The solution was evaporated to dryness at 50°C under a gentle stream of nitrogen and reconstituted with 100μL of methanol. The mobile phase consisted of acetonitrile (A) and 0.01% formic acid (B); ratio A:B varied from 8:92 to 90:10 (v/v), at a flow rate of 300μL/min. 5μL of the extract were injected into LC-MSMS system. Mass spectrometer (Thermo Scientific TSQ Quantiva Triple Quadrupole) was interfaced with ultra-performance liquid chromatography (UPLC) system (Thermo Scientific UltiMate 3000 XRS) using a pneumatic assisted heated electrospray ion source. MS detection was performed in positive ion mode, using selected reaction monitoring.

Quantification of p24+ cells

Quantification and characterization of p24+ cells were performed by using the HIV-Flow procedure as previously described (Pardons et al., 2019). Briefly, CD4 T cells were isolated by negative magnetic selection as described above and resuspended at 2×10⁶ cells/mL in RPMI + 10% Fetal Bovine Serum with antiretroviral drugs (200nM raltegravir, 200nM lamivudine). Cells were pre-incubated for 1h with 5μg/mL Brefeldin A (Sigma, B2651) and BFA was maintained in the culture until the end of the stimulation. As a positive control, cells were stimulated with 162nM PMA and 1μg/mL ionomycin for 24h. To test the ability of different LRAs at inducing p24 expression, cells were exposed to 100nM panobinostat, 100nM romidepsin, 10μM JQ-1, 500nM ingenol and their combinations. After 24h of stimulation, cells were collected, resuspended in PBS and stained with the Aqua L/D staining kit for 30min at 4°C. Cells were then stained with antibodies for CD3, CD4, CD8, CD45RA, CCR7, CD27 in PBS + 4% human serum for 30min at 4°C. Cells were fixed and permeabilized for 45min at 4°C using the FoxP3 Transcription Factor Staining Buffer Set (eBioscience, 00–5523-00), and stained with anti-p24 (clone KC57) and anti-p24 (clone 28B7) antibodies for an additional 45min at room temperature. The frequency of p24+ cells (KC57+, 28B7+) was determined by flow cytometry (BD LSRII) in gated CD8 negative T cells (CD3+CD8-). In all experiments, purified CD4 T cells from an HIV-negative control were included to set the threshold of positivity.

QUANTIFICATION AND STATISTICAL ANALYSIS

All data were analyzed using Graphpad Prism v6.0h. For each concentration of LRA, we calculated the fold increase in mean fluorescence intensity (MFI) over the non-stimulated condition in the population of interest (total CD4 T cells or gated subsets). To generate dose response curves, a nonlinear regression [log(agonist) vs. response (3 parameters)] was applied to the data. From the dose-response curves, we determined the fold induction at the top plateau, which represents the maximal efficacy of a given LRA. To determine the EC50, data were normalized with the lowest and highest values defined as 0% and 100% respectively, and a nonlinear regression [log(agonist) vs. normalized response (variable slope)] was applied to the data. To compare levels of cellular factors between CD4 T cell subsets, Friedman tests with Dunn’s correction for multiple comparisons were used. To compare levels of HIV reactivation between memory CD4 subsets, non-parametric Wilcoxon tests were used. The numbers of samples analyzed are indicated in the Figure legends.

DATA AND CODE AVAILABILITY

This study did not generate any unique datasets or code.