Host T Cell Dedifferentiation Effects Drive HIV-1 Latency Stability

ABSTRACT

The development of therapies to eliminate the latent HIV-1 reservoir is hampered by our incomplete understanding of the biomolecular mechanism governing HIV-1 latency. To further complicate matters, recent single-cell RNA sequencing (scRNA-seq) studies reported extensive heterogeneity between latently HIV-1-infected primary T cells, implying that latent HIV-1 infection can persist in greatly differing host cell environments. We show here that transcriptomic heterogeneity is also found between latently infected T cell lines, which allowed us to study the underlying mechanisms of intercell heterogeneity at high signal resolution. Latently infected T cells exhibited a dedifferentiated phenotype, characterized by the loss of T cell-specific markers and gene regulation profiles reminiscent of hematopoietic stem cells (HSC). These changes had functional consequences. As reported for stem cells, latently HIV-1-infected T cells efficiently forced lentiviral superinfections into a latent state and favored glycolysis. As a result, metabolic reprogramming or cell redifferentiation destabilized latent infection. Guided by these findings, data mining of single-cell RNA-seq data of latently HIV-1-infected primary T cells from patients revealed the presence of similar dedifferentiation motifs. More than 20% of the highly detectable genes that were differentially regulated in latently infected cells were associated with hematopoietic lineage development (e.g., HUWE1, IRF4, PRDM1, BATF3, TOX, ID2, IKZF3, and CDK6) or were hematopoietic markers (SRGN; hematopoietic proteoglycan core protein). The data add to evidence that the biomolecular phenotype of latently HIV-1-infected cells differs from that of normal T cells and strategies to address their differential phenotype need to be considered in the design of therapeutic cure interventions.

IMPORTANCE HIV-1 persists in a latent reservoir in memory CD4 T cells for the lifetime of a patient. Understanding the biomolecular mechanisms used by the host cells to suppress viral expression will provide essential insights required to develop curative therapeutic interventions. Unfortunately, our current understanding of these control mechanisms is still limited. By studying gene expression profiles, we demonstrated that latently HIV-1-infected T cells have a dedifferentiated T cell phenotype. Software-based data integration allowed the identification of drug targets that would redifferentiate viral host cells and, by extension, destabilize latent HIV-1 infection events. The importance of the presented data lies within the clear demonstration that HIV-1 latency is a host cell phenomenon. As such, therapeutic strategies must first restore proper host cell functionality to accomplish efficient HIV-1 reactivation.

INTRODUCTION

Despite successful antiretroviral therapy that represses circulating virus below detection levels, HIV-1 can persist for decades, and natural viral eradication is not expected during the lifetime of people living with HIV-1 (PLWH) (1). While an increasing amount of data suggests that viral reservoirs may also exist in non-T cell compartments (2–6), the most comprehensive evidence for a mechanism of viral persistence so far has been presented for latent HIV-1 infection residing in long-lived T cells with a resting CD4 memory phenotype (7–9). The extraordinary stability of the latent HIV-1 reservoir in this cellular reservoir has been explained by homeostatic or antigen-driven proliferation of latently HIV-1-infected cells in the absence of reactivation (10–13) or possibly by viral integration in gene loci that increase the proliferative capacity of the latently infected host cells (14). These findings imply that in vivo, as seen in latently HIV-1-infected T cell lines, proliferation does not trigger HIV-1 reactivation and that latently HIV-1-infected T cells are not necessarily resting at all times (15, 16).

Initially, HIV-1 latency was explained by the presence of a restrictive histone composition or chromatin environment at the viral promoter (long terminal repeat [LTR]) (17–22). Over time, it has been demonstrated that latent HIV-1 does not require chromatin level gene silencing (23) but that transcriptional reprogramming of CD4 T cells during effector-to-memory transition is required to render them permissive to host latent HIV-1 infection events, likely by restricting access to relevant transcription factors (24). Furthermore, the majority of latent HIV-1 infection events are integrated into actively expressed genes, and therefore, regions of open chromatin, histone-level restrictions are unlikely, but paused RNA polymerase II is found at latent HIV-1 LTRs (23, 25–30). This means that at the level of the viral LTR, latent HIV-1 infection events do not differ from any nonexpressed but inducible cellular gene in resting T cells, such as genes for interleukin 2 (IL-2) or IL-8, which have promoters with transcription factor binding site profiles that are highly similar to the HIV-1 LTR (31–33). This raises the question of why it is so extraordinarily difficult to trigger efficient, population-wide reactivation of latent HIV-1. Even in ex vivo experiments, large parts of latent HIV-1 infection events in primary memory T cells from PLWH on antiretroviral therapy (ART) remain unresponsive to strong stimuli such as phytohemagglutinin L (PHA-L) or phorbol myristate acetate (PMA)–ionomycin (16, 23). One explanation would be that cells harboring latent HIV-1 provirus have been permanently altered during the process of infection and display an activation-inert phenotype. To this end, we previously demonstrated that latently infected T cells are phenotypically altered in ways that are consistent with the idea of an unresponsive host cell phenotype. Latently HIV-1-infected T cells exhibit an altered NF-κB response, and manipulation of hallmark proteins associated with T cell anergy (e.g., ICER [inducible cyclic AMP early repressor]) affected the stability of latent HIV-1 infection (34). Kinome analysis experiments describing these protein-level changes identified targets for pharmacological intervention strategies that altered the level of HIV-1 latency establishment and affected the stability of latent HIV-1 infection events. In addition, a large number of cellular restriction or dependency factors have been described as being involved in HIV-1 latency control (34–45).

Adding to the recognized multifactorial nature of latent HIV-1 infection (46), recent single-cell RNA sequencing (scRNA-seq) analysis studies now suggest a high degree of heterogeneity between latently HIV-1-infected T cells derived from PLWH on ART (47) and from a model of latent HIV-1 infection using primary T cells (48). This suggests that latent HIV-1 infection can persist in a wide variety of cellular environments and begets the question of whether the observed heterogeneity is linked to the inability of currently studied stimuli to efficiently trigger reactivation (47–49). To improve current approaches to trigger HIV-1 reactivation as part of a cure for HIV-1 infection, it will be important to identify the driving forces behind this intercell heterogeneity and their potential role in establishing and maintaining latent HIV-1 infection or to identify shared gene expression patterns that could explain host cell control of latent infection but that are obscured by transcriptomic heterogeneity.

Such studies are impossible to conduct in primary T cells from patients, as there is no marker that would identify latently HIV-1-infected T cells in their resting state. However, the Nussenzweig group has been able to generate scRNA-seq data from primary T cells obtained from PLWH in which HIV-1 expression was reactivated by T cell activation and then enriching previously latent, now activated and actively infected cells using antibodies against HIV-1 env/gp120 (47). The authors of that study concluded that these previously latently HIV-1-infected T cells, despite overall transcriptomic heterogeneity, shared altered activated gene expression profiles, mostly associated with immune function.

Starting with the finding that intercell heterogeneity was also observed between latently HIV-1-infected T cell lines, we used high-resolution bulk RNA-seq data to search for shared pathway motifs that were concealed by the transcriptional heterogeneity. This effort identified dedifferentiation toward a hematopoietic stem cell (HSC) phenotype as a shared biomolecular phenotype for latently HIV-1-infected T cell lines. Infection-induced T cell dedifferentiation would conclusively explain the observed transcriptomic heterogeneity between latently infected T cell clones and the observed activation inertness spectrum. Guided by these findings, we were able to identify a similar transcriptomic signature when mining the scRNA-seq analysis data from primary latently HIV-1-infected T cells, suggesting that HIV-1 infection-induced T cell dedifferentiation is a phenomenon shared between latently HIV-1-infected T cell lines and primary T cells.

RESULTS

Intercell heterogeneity, loss of T cell-specific markers, and differential regulation of hematopoietic lineage markers characterize latently HIV-1-infected T cells.

A complete understanding of how host cells of latent HIV-1 infection events are altered to allow latent HIV-1 infection events to persist is a crucial step toward the development of curative strategies for HIV-1 infection. As part of such an effort, it will be important to determine whether the observed transcriptional intercell heterogeneity actually obscures a shared core motif that would be responsible for HIV-1 latency control (47–49). To accomplish this goal, we first addressed the question of whether (i) the intercell heterogeneity that was reported for latently HIV-1-infected primary T cells would be the result of differences between primary T cells that already existed prior to infection or (ii) heterogeneity would be induced by the actual infection event. To answer this question, we generated transcriptomic profiles for latently infected T cell lines that were all derived under similar experimental conditions. Jurkat T cells, although not a completely uniform population at the transcriptomic level (50), can be considered relatively uniform in their biomolecular phenotype compared to primary T cells, which already have highly diverse transcriptional phenotypes prior to infection (e.g., naive versus memory T cells) (51). The presence of intercell heterogeneity between Jurkat cell-derived latently HIV-1-infected Jurkat T cell clones would thus suggest that transcriptional heterogeneity is not solely a function of preexisting differences but can be induced by the actual infection event.

We initially selected three latently HIV-1-infected T cell lines that were generated using full-length, replication competent green fluorescent protein (GFP) reporter viruses. These cell lines represent the three major HIV-1 integration site phenotypes, and the HIV-1 infection events reproduced typical features of latent infections. In CA5 T cells, the virus is integrated in an intron in the same-sense orientation relative to the transcriptional direction of the host gene. In EF7 T cells, the virus is integrated into an intron in the antisense orientation, and CG3 T cells hold a rare intergenic integration event (Fig. 1A). Data from assays for transposase-accessible chromatin with high-throughput sequencing (ATAC-seq) suggest that the intergenic integration event in CG3 cells occurred in an area of open chromatin in the parental cells (Fig. 1B).

FIG 1.

Genomic and transcriptional characteristics of latent HIV-1 infection events in CA5, EF7, and CG3 T cells. (A) The majority of latent HIV-1 infection events in vivo are found in introns or exons of actively expressed host genes. The latent HIV-1 infection events in CA5 and EF7 T cells are integrated in the same-sense orientation or the antisense orientation relative to the transcriptional direction of the host genes, whereby the latent HIV-1 infection event in CA5 T cells is integrated into a pair of overlapping genes (RBM12 and CPNE1; map location, 20q11.21). The latent infection event in CG3 T cells is integrated in an intergenic region between the TIGD5 and the PYCRL genes. (B) ATAC-seq analysis suggests that that the provirus in CG3 T cells is integrated into a region of open chromatin in Jurkat cells. (C) To determine basic transcriptional characteristics of the latent HIV-1 LTR in CA5, EF7, and CG3 T cells, chromatin immunoprecipitation (ChIP) experiments were performed using anti-RNAP II and anti-H3Ac antibodies. Primers that spanned positions −14 to +113 of the HIV-1 LTR were used for real-time PCR to detect factor association with the HIV LTR. These data represent triplicate ChIPs and are representative of two experiments. (D) Comparison of generated mRNA transcript levels between control cells (Jurkat) and the latently infected T cells for the respective integration site genes (CA5 and EF7) or the genes upstream or downstream of the intergenic integration event (CG3). (E) Reactivation profiles of CA5, EF7, and CG3 T cells in response to increasing concentrations of PMA. (F) Reactivation profiles of CA5, EF7, and CG3 T cells in response to increasing concentrations of bryostatin. (G) Determination of maximum reactivation level following PMA stimulation of 50 latently infected T cell clones generated in parallel to CA5 (dark red), CG3 (bright red), and EF7 (blue).

Consistent with RNA polymerase II (RNAP II) pausing being an established, critical checkpoint for HIV-1 transcription in latently HIV-1-infected T cells, chromatin immunoprecipitation assays (ChIPs) identified high levels of RNAP II and acetylated histone 3 (AcH3), an indicator of open chromatin (23, 25–30), at the viral LTRs in all three T cell lines (Fig. 1C). It is important to state that the integration events did not affect the level of generated mRNA transcripts, of either the respective integration site genes or the genes in proximity to the intergenic integration event (Fig. 1D). The absence of any gene regulation effects directly caused by viral integration makes it unlikely that the integration site plays any role for possible global host cell gene regulation effects that would have to occur to generate transcriptomic heterogeneity (47–49). Unsurprisingly, the three clonal cell lines exhibit different reactivation profiles in responses to protein kinase C (PKC) agonists such as PMA or bryostatin, with latent infection in CG3 cells being the most difficult to reactivate (Fig. 1E and F). Inferior reactivation levels in CG3 T cells could be caused by its unique integration site in an intergenic region; however, CG3 reactivation levels are well within the range of commonly observed reactivation levels seen for latent infection events integrated into actively expressed genes when generating latently infected T cell lines in this system (Fig. 1G). While we could select for the most reactivation-responsive T cell clones for latent integration events into actively expressed genes, this was not possible for latent infection events in intergenic regions. Out of hundreds of tested latently infected T cell clones, CG3 T cells were the only ones hosting an intergenic integration event. Perceived inferior reactivation levels in CG3 are thus most likely the result of selection bias.

Genome-wide RNA expression profiling of the latently HIV-1-infected T cells by RNA-seq identified a total of 3,093 genes that were differentially expressed in the three latently infected T cell lines compared to Jurkat T cells (adjusted P value [P_adj] < 0.05; 2-fold difference). Heat map representation of these gene sets visualized the extensive transcriptomic differences (heterogeneity), not only between latently infected and uninfected control T cells but also between the individual latently infected T cell clones (Fig. 2A). A set of 888 differentially regulated genes (29%) was found to be shared between the three latently HIV-1-infected T cell clones, while the expression of 1,296 genes (42%) was uniquely altered in just one T cell clone, a measure of the high level of heterogeneity between these clones (Fig. 2B). Principal-component analysis (PCA) found that based on their transcriptomic signature, all latently infected T cells clearly differed from the uninfected Jurkat T cells. CA5 and CG3 T cells were more closely related to each other than either was related to EF7 T cells (Fig. 2C). This analysis demonstrates that intercell heterogeneity reported for latently HIV-1-infected primary T cells is also present in T cell clones.

FIG 2.

Extensive RNA-level heterogeneity between latently HIV-1-infected T cell lines. (A) Expression and hierarchical clustering of genes that were found differentially expressed throughout the three latently infected T cell lines (CA5, EF7, and CG3) compared to Jurkat T cells (likelihood ratio test; adjusted P value < 0.01). (B) Venn diagram of genes found to be differentially expressed between each latently infected clone and Jurkat cells (≥2-fold change and adjusted P value of <0.01). (C) Principal-component analysis assessing the similarity between Jurkat control cells and the three latently HIV-1-infected T cell lines.

Pathway motif analysis using the Molecular Signatures Database (MSigDB) revealed that despite this heterogeneity, the individual clones shared certain signatures in which they differed from uninfected control cells. As seen in Table 1, the hematopoietic cell lineage motif (KEGG) was the highest-ranked shared signature, suggesting that latently HIV-1-infected T cell clones could have dedifferentiated toward a more hematopoietic stem cell-like phenotype.

TABLE 1.

Highest-ranked shared pathway motifs between latently HIV-1-infected T cell lines

Description	P value			Gene set name (no. of genes [K])
	CA5	EF7	CG3
Hematopoietic cell lineage	4.57e−12	1.23e−10	1.98e−12	KEGG_HEMATOPOIETIC_CELL_LINEAGE (87)
Hemostasis	5.86e−13	1.85e−10	2.27e−11	REACTOME_HEMOSTASIS (678)
Cytokine signaling in immune system		2.27e−11	5.18e−12	REACTOME_CYTOKINE_SIGNALING_IN_IMMUNE_SYSTEM (858)
Adaptive immune system	2.46e−8		5.18e−10	REACTOME_ADAPTIVE_IMMUNE_SYSTEM (818)
Genes defining early response to estrogen	9.73e−10		1.87e−17	HALLMARK_ESTROGEN_RESPONSE_EARLY (200)
Genes involved in cholesterol homeostasis	8.43e−10		5.25e−9	HALLMARK_CHOLESTEROL_HOMEOSTASIS (74)
RNA polymerase II transcription	6.2e−8		3.39e−7	REACTOME_RNA_POLYMERASE_II_TRANSCRIPTIPTION (1,375)

Heterogeneity between latently HIV-1-infected T cells was not limited to this experimental model of latent HIV-1 infection. In a different model system, we had previously performed transcriptomic analysis to compare latently HIV-1-infected T cells in which CD3/CD28 stimulation would trigger HIV-1 reactivation with latently HIV-1-infected T cells that had become T cell receptor (TCR)/CD3 activation inert (52). These latently infected cells were generated using a GFP reporter T cell population that was infected with a primary HIV-1 patient isolate (HIV-1 WEAU) (53). Analysis of this data set using the same analytical pipeline used for the analysis shown in Fig. 1 confirmed the presence of high levels of heterogeneity between latently infected T cells (Fig. 3). As for the other model system of latent HIV-1 infection, motif analysis again identified hematopoietic cell lineage as the highest ranked altered gene signature (P_adj = 2.53E−06) and cytokine/cytokine receptor interaction as the second highest (P_adj = 2.40E−05).

FIG 3.

Extensive RNA-level heterogeneity between reporter T cell lines latently infected with an HIV-1 primary patient isolate. (A) Expression and hierarchical clustering of genes that were found differentially expressed in the latently infected reporter T cell lines (JWEAU-A10 and JWEAU-C6) compared to control T cells (likelihood ratio test; adjusted P value < 0.01). (B) Venn diagram of genes found to be differentially expressed between each latently infected clone and Jurkat cells (≥2-fold change and adjusted P value of <0.01). (C) Principal-component analysis assessing the similarity between Jurkat control T cells and JWEAU-A10 or JWEAU-C6 T cells.

Immediate implications of these results are that (i) HIV-1 latency is indeed maintained in otherwise extensively different cellular environments (47–49), and (ii) that the actual infection event must contribute to the generation of intercell heterogeneity observed between latently HIV-1-infected T cells. Also, dedifferentiation effects, if functionally confirmed, could explain the intercell heterogeneity and activation inertness of latently HIV-1-infected T cells that were described previously (52).

Protein network interactions and signaling hierarchies confirm dedifferentiated phenotype of latently HIV-1-infected T cells.

We noted that the genes in the hematopoietic cell lineage KEGG gene set focused on descriptor proteins expressed on the cell surface, but transcription factors that are known to drive lineage differentiation were underrepresented. If latently infected T cells were to have a dedifferentiated phenotype, these cells should display systematic differential regulation of transcription factors associated with immature T cell development stages or even hematopoietic stem cells. As a consequence of gene regulation effects at the transcription factor level, latently HIV-1-infected T cells should also exhibit a consistent loss of markers present on mature T cells. We thus curated a list of transcription factors with known central roles in hematopoietic lineage development and genes characteristic for mature T cells.

Transcription factors such as POU5F1 (Oct3/4), Lin28, MEIS1, LEF1, KLF5, MYCN, LBP9/TFCP2L1, YAP1, STAT5, and GATA-2 have been reported to play fundamental roles regarding survival, self-renewal, and metabolism control of hematopoietic stem cells (54–81). Expression of these genes was overall differentially regulated in latently HIV-1-infected T cells compared to control T cells (Fig. 4A), with each T cell clone exhibiting an individual gene expression signature within the HSC transcription factor motif. These differences were also reflected in the differential expression patterns of HSC markers, such as CD9, CD34, CD84, or CD109 (82–91). While the transcription factor composition within the HSC motif differed between individual T cell clones, the downstream effects regarding the regulation of T cell markers that are present in the control T cells were uniform. All latently infected T cells exhibited downregulation of mRNA expression of the T cell lineage marker CD4, the pan-T cell marker CD2, CD5, CD7, the costimulatory CD28 molecule, and CD40L or the antigen-presenting CD1 family members (Fig. 4B). The expression of other lymphoid genes such as RAG1 and RAG2, DNTT, BLK, or IL7R which are essential for the maturation of T cells was also found reduced or no longer present in latently HIV-1-infected T cells (92–95). This principal phenotype was reproduced in the latently HIV-1-infected J-WEAU T cells, which were characterized by similar but not completely identical dysregulation effects for HSC genes and subsequent downregulation of T cell markers (Fig. 4C and D). Flow-cytometric analysis for some of the key markers confirmed that gene expression changes correlated with changes at the protein expression level (Fig. 4E).

FIG 4.

Latently HIV-1-infected T cells have a dedifferentiated phenotype. (A) Heat map representation of genes that are central to hematopoietic stem cell biology comparing gene expression levels in the parental Jurkat T cells and the latently HIV-1-infected CA5, EF7, and CG3 T cells. (B) Heat map representation of genes typically expressed by mature T cell comparing gene expression levels in the parental Jurkat T cells and the latently HIV-1-infected CA5, EF7, and CG3 T cells. (C) Heat map representation of genes that are central to hematopoietic stem cell biology comparing gene expression levels in the parental Jurkat T cells and the latently HIV-1-infected JWEAU-A10 and JWEAU-C6 T cells. (D) Heat map representation of genes typically expressed by mature T cells, comparing gene expression levels in the parental Jurkat T cells and the latently HIV-1-infected JWEAU-A10 and JWEAU-C6 T cells. (E) Flow-cytometric validation of protein level expression of differentially regulated T cell markers in latently infected T cell clones.

We next used network analysis to challenge the idea that latently HIV-1-infected T cells have a phenotype reminiscent of HSC. To determine how the differentially regulated genes or rather their protein products would align in a protein-protein interaction network (PIN), we utilized MetaCore network analysis software. MetaCore is based on the manually curated results of approximately one million peer-reviewed publications; this rich data set is thus able to resolve subtle gene expression changes into detailed cellular networks not based on signal amplitude but instead by linkage numbers. Linkage-based analysis accounts for the possibility that seemingly small changes to gene expression, or in extension protein activity, can have significant downstream effects, should a gene product control a large number of targets. We would posit that if cell dedifferentiation effects are a central component of latency control, then genes/proteins that are involved in hematopoietic cell lineage development would be central to a PIN generated from the data comparing the parental cells with the latently HIV-1-infected T cell lines.

The generated PIN (Fig. 5A) had an unusually steep hierarchy in which only six transcription factors (POU5F1 [Oct3/4], MEIS1, LBP9, MYCN, YAP1, and ERG), all genes with described roles in hematopoietic stem cell biology, had 20 or more edges/links with other network components/nodes (Fig. 5B). Of the altered genes, 28.6% had only a single link to the network and 12.5% had only two links. Within the network, components assigned to the Gene Ontology (GO) cell differentiation motif (traced in Fig. 5C) constituted 42% of the network, and in their vast majority had several links to other network components. A detailed heat map representation of the relative expression levels of the central network nodes emphasized that gene expression changes within the motif differed between the T cell clones (Fig. 5D), in ways similar to the overall heterogeneity of gene expression patterns between T cell clones (Fig. 2 and 3).

FIG 5.

Network analysis identifies key transcription factors involved in host cell dedifferentiation. (A) Protein-protein interaction network of the genes in the shared latency control signature generated using the direct interaction algorithm of MetaCore software. (B) Top 25 network nodes ranked by the number of their interactions. (Inset) Number of nodes by number of interactions/edges. (C) Nodes and interactions involved in cell differentiation processes in the direct protein-protein interaction network are traced. (D) Heat map of differentially expressed transcription factors that control differentiation and hematopoietic stem cell development.

Overall, these findings do not imply that any of the central factors of the network directly control latent HIV-1 infection, but the data support the presence of a functional HSC-like phenotype. Although previous work has suggested that latent HIV-1 infection events reside in hematopoietic progenitor T cells (96, 97), our data do not support the idea that latently infected cells are HSCs but instead suggest that latently infected T cells are dedifferentiated toward an HSC-like phenotype. To claim biological relevance for this hypothesis, it would be important to show that the observed altered transcriptomic signature has measurable consequences and thus demonstrate the presence of functional features that are typical for HSCs. Among others, stem cells (i) can efficiently silence retroviruses and (ii) have an altered metabolic profile that prefers glycolysis, two features that would have in vivo relevance and that can be experimentally explored (64, 98, 99).

Latently HIV-1-infected T cells suppress de novo HIV-1 infection.

An established characteristic of stem cells is their unparalleled ability to silence incoming retroviruses (100–103). It seems therefore plausible that the HSC-like phenotype of latently HIV-1-infected T cells contributes to the suppression of the integrated infection events into a latent state. If correct, we would expect that latently HIV-1-infected T cells also have an increased capacity to establish latent HIV-1 superinfection events. To test this hypothesis, we infected either Jurkat T cells (control) or the latently HIV-1-infected CA5 and EF7 T cells with increasing amounts of an HIV-mCherry reporter virus, to distinguish between the original infection (GFP) and new infection events, for a total of 30 individual infection cultures for each cell type. The HIV-mCherry reporter virus was vesicular stomatitis virus G protein (VSV-G) pseudotyped to render infections CD4 independent. Reverse transcriptase (RT) inhibitors were added to all cultures on day 2 postinfection to prevent de novo infections and the establishment of a nonintegrated latent reservoir. On day 3 postinfection, we determined the levels of active HIV-mCherry infection in each cell culture as a reference point to calculate the size of the latent HIV-1 reservoir as a function of the initially active infection levels (Fig. 6).

FIG 6.

Latently HIV-1-infected T cells force HIV-1 superinfection events into a latent state. Jurkat T cells (×) or latently HIV-1-infected CA5 (gray squares) and EF7 (gray circles) T cells were infected with increasing amounts of an HIV-mCherry reporter virus, each for a total of ∼30 individual infection cultures. RT inhibitors were added on day 2 postinfection to prevent de novo infections. On day 3 postinfection, active HIV-mCherry infection in each cell culture was determined by flow-cytometric analysis as a reference to calculate the size of the latent HIV-1 reservoir as a function of the active infection levels. On day 10 postinfection, a sample of each cell culture was stimulated with PMA to activate latent HIV-1 infection events. This size of the latent reservoir in each culture was determined by subtracting the percentage of mCherry-positive cells in the absence of stimulation from the percentage of mCherry-positive cells in the paired PMA-stimulated cultures. To present reservoir size as a function of the initial infection level, the percentage of latently HIV-1-infected T cells in each culture was then plotted over the active infection level of the culture as determined on day 3 postinfection.

As previously described, on day 10 postinfection, we stimulated a sample of each cell culture with PMA to activate otherwise latent HIV-1 infection events to determine the size of the latent-infection reservoir (104). This was calculated by subtracting the percentage of mCherry-positive cells in the absence of stimulation in the parental control cultures from the percentage of mCherry-positive cells in the paired PMA-stimulated cultures. The size of the latent HIV-mCherry reservoir was indeed much higher for all infections of the latently HIV-1-infected CA5 or EF7 T cells than for uninfected control T cells (Fig. 6), despite the superinfection events triggering reactivation of the original infection event and the subsequent provision of HIV-1 Tat, which is known to counteract latency establishment. This would suggest that the observed biomolecular phenotype of latently HIV-1-infected T cells indeed has functional consequences consistent with previous reports on the ability of stem cells to silence retroviral infection events (100–103).

Latently HIV-1-infected T cells have been metabolically reprogrammed.

HSCs have a metabolic profile that favors glycolysis, an adaptation to their localization in often hypoxic environments, such as the bone marrow. This metabolic profile is maintained even if oxygen is available, as glycolysis, in contrast to oxidative phosphorylation, does not favor the production of reactive oxygen species that would be a threat to the functional and genomic integrity of stem cells during their extended lifetime. The proglycolytic phenotype of HSCs has been reported to be programmed by the HSC transcription factor MEIS1, which controls the downstream regulation of many glycolytic enzymes (64). MEIS1 was found central to the PIN controlling latently HIV-1-infected T cells (Fig. 5A and B).

For the actual glycolysis pathway, latently HIV-1-infected T cells expressed higher levels of RNA for several of the genes, including hexokinase 1 (HK1), glucose-6-phosphate isomerase (GPI), phosphofructokinase (PFKL), aldolase A (ALDOA), and lactate dehydrogenase A (LDHA), the enzyme that converts pyruvate, the final product of the glycolysis pathway, into lactic acid/lactate. Throughout all genes of the glycolysis pathway, the mean increase was 1.5-fold and the median increase 1.3-fold (Table 2). To determine whether these changes in RNA levels translate into a functional phenotype, we tested whether the latently HIV-1-infected T cell clones would produce higher levels of lactate under nonhypoxic conditions. Uninfected T cells (Jurkat) and latently HIV-1-infected T cells (CA5, EF7, and CG3) were seeded at identical cell densities into 6-well plates, and lactate production was measured after 24 h. To adjust lactate production to differences in proliferative capacity, cell numbers were determined at this time point using a capillary-based flow cytometer. In these experiments, all three latent clones produced significantly more lactate (Fig. 7A), confirming the metabolic preference for glycolysis that was suggested by the RNA-seq data and consistent with a stem cell-like phenotype.

TABLE 2.

Expression level changes of genes in the glycolysis pathway

Gene	Fold change of read counts in clonea
	CA5	EF7	CG3
HK1	2.2	2.5	2
GPI	1.3	1.2	1.2
PFKL	1.5	1.1	1.5
ALDOA	1.3	1.4	1.4
ALDOC	5	3.4	3.3
GAPDH	1	1.3	1
PGK1	1.2	1.1	1.2
PGM1	0.9	1.2	1.1
PGM2	1.1	0.9	1.1
PKM	1.3	1.1	1.2
LDHA	1.3	1.5	1.5

^aData are for the averages of two independent experiments, showing fold induction relative to read counts in Jurkat T cells.

FIG 7.

Metabolic changes of latently HIV-1-infected T cells can be addressed to destabilize latent HIV-1 infection events. (A) To determine whether latently HIV-1-infected T cells would be metabolically reprogrammed to preferentially use glycolysis, we measured lactate production by Jurkat, CA5, EF7, and CG3 T cells. The cells were seeded into fresh culture medium, and supernatants were harvested after 24 h to measure lactate concentration using a colorimetric lactate assay kit (Sigma). The experiment was performed in three independent replicates, and data were normalized for cell numbers at 24 h and are presented as means and standard deviations. The latently HIV-1-infected CA5 (B), EF7 (C), and CG3 T cells (D) were treated overnight with the LDHA inhibitor (R)-GNE-140 (5 or 10 µM). The cells were then stimulated with increasing concentration of either bryostatin or prostratin to trigger HIV-1 reactivation. Reactivation was measured as the percentage of GFP-positive T cells, as determined by flow-cytometric analysis.

As pharmacological tools for metabolic reprogramming are available, we explored whether the observed glycolytic phenotype is part of latency control. We treated CA5, EF7, and CG3 T cells with the LDHA inhibitor (R)-GNE-140 and then stimulated with increasing concentrations of either bryostatin or prostratin to trigger HIV-1 reactivation, as these PKC agonists are more therapeutically tractable than PMA.

Given that LDHA inhibition can selectively kill tumor cells that preferentially use glycolysis, one possibility was that LDHA inhibition would mostly affect the viability of latently HIV-1-infected T cells (105), but at the utilized concentrations, (R)-GNE-140 had no effect on cell viability. However, as predicted, overnight (R)-GNE-140 pretreatment (5 or 10 µM) lowered the concentration of bryostatin and prostratin required to trigger HIV-1 reactivation. As expected, given the heterogeneity between the T cell clones, the effect varied between cell lines and the utilized secondary stimulus. For CA5 T cells, (R)-GNE-140 treatment mostly boosted bryostatin-triggered HIV-1 reactivation and allowed reactivation in otherwise inert cells (Fig. 7B). In EF7 cells, the priming effect of (R)-GNE-140 in EF7 cells was relatively small and resulted in only a minimum increase in the achievable reactivation levels at the highest bryostatin or prostratin concentrations (Fig. 7C). However, (R)-GNE-140 pretreatment of CG3 T cells increased the ability of bryostatin to trigger reactivation at the population level from 20% to 60% of the cells and for prostratin from 40% to over 80% (Fig. 7D). These findings extend previous reports showing that altered cellular metabolism is a key element of cellular HIV-1 susceptibility (106, 107). In summary, these findings demonstrate that the HSC-like phenotype seen at the transcriptomic level translates into functional traits of HSCs and provide definite evidence that metabolic changes are a part of the control of latent HIV-1 infection.

Retinoic acid-driven host cell redifferentiation triggers spontaneous HIV-1 reactivation.

To identify drug targets that could reverse a dedifferentiated host cell phenotype, we screened the RNA-seq data set for additional altered genes directly involved in cell differentiation that could be therapeutically addressed. This analysis revealed that expression of the gene for ALDH1A2, an enzyme in the vitamin A pathway responsible for converting retinol/retinaldehyde to retinoic acid (RA), was downregulated by 50 to 95% in all latently HIV-1-infected T cells, while the expression of other ALDHs remained largely unchanged. In the absence of ALDH1A2, latently HIV-1-infected T cells should not be able to convert vitamin A, which is abundantly present in fetal bovine serum (FBS), into retinoic acid, as ALDH1A1 and ALDH1A3 are not expressed. Given that retinoic acid has cell differentiating abilities and has been reported to enhance the T cell response to TCR/CD3 activation (108), failed conversion of vitamin A into retinoic acid would permit or accelerate the observed dedifferentiation of latently HIV-1-infected T cells.

As an extension of these observations, we hypothesized that supplementation of retinoic acid would bypass the lack of ALDH1A2 activity and redifferentiate latently infected T cells, thereby promoting HIV-1 reactivation either directly or through secondary activating stimuli. As seen in Fig. 8A, supplementation of the medium with RA had no immediate effect on the stability of latent HIV-1 infection. However, on day 5 after RA treatment and without a need for additional agonists, HIV-1 reactivation became apparent. This effect was not achieved by a bolus application of RA but required repeated addition of RA. Following repeated addition of RA over a 10-day period, latent HIV-1 infection transitioned to an active expression state in 80% of CA5 and EF7 T cells and 30% of CG3 T cells.

FIG 8.

Host cell redifferentiation by RAR or RXR antagonists destabilizes latent HIV-1 infection. To determine the effect of redifferentiation strategies on latency stability, the three latently HIV-1-infected T cell clones (CA5, EF7, and CG3) were treated with increasing amounts of retinoic acid or bexarotene, either added every second day or as a one-time treatment (black diamonds). (A) HIV-1 reactivation levels determined as percent GFP-positive cells over time following treatment of CA5, EF7, or CG3 T cells with different concentrations of retinoic acid. (B) HIV-1 reactivation levels triggered by different concentrations of the RXR agonist bexarotene.

As retinoic acid receptors (RAR-α, RAR-β, and RAR-γ) signal by binding to DNA in a heteromeric complex with RXR to retinoic acid response elements (RARE) in promoters, we also tested the effect of the FDA-approved RXR agonist bexarotene on the stability of latent HIV-1 infection events (Fig. 8B). Similar to what was observed for RA, no immediate effect was apparent. Tangible reactivation was attained following a single treatment starting on day 7 posttreatment. Repeated bexarotene addition only modestly improved reactivation levels. On day 10 posttreatment, HIV-1 reactivation was observed in ∼30% of CA5 and EF7 T cells and 20% of CG3 T cells, confirming that suppression of RAR/RXR signaling is part of HIV-1 latency control. The delayed onset of these effects would suggest that RA or bexarotene acted by altering the composition of cellular factors controlling latent HIV-1 infection, rather than directly activating the virus. The addition of RA or bexarotene may partially restore the original host cell phenotype, as the parental Jurkat T cells generally promote active HIV-1 infection. These data emphasize that heterogeneity affects the spectrum of differentiation effects which is exemplified by the relative recalcitrance of CG3 cells to redifferentiation, likely because of the presence of other unidentified restrictions. This is supported by the overall reactivation profiles of CG3 cells, which are also relatively resistant to reactivation triggered by PMA, prostratin, or bryostatin (Fig. 1E and F).

scRNA-seq analysis suggests the presence of an altered T cell differentiation and activation response phenotype in latently HIV-1-infected primary T cells.

Finally, we sought to explore whether changes to the T cell differentiation state would be found in latently HIV-1-infected primary T cells from PLWH. In the absence of a marker that identifies latently HIV-1-infected primary T cells, it is impossible to directly determine the biomolecular baseline phenotype of patient-derived latently infected T cells. The only possibility to identify primary latently HIV-1-infected T cells is to obtain CD4⁺ T cell populations from PLWH on fully suppressive ART, to activate these purified T cells, and to subsequently identify and analyze T cells that express HIV-1 gp120 following activation. As we have already shown that latently HIV-1-infected T cells have a differential activation response phenotype (52), we hypothesize that for these previously latently, now activated and actively HIV-1-infected T cells, a biomolecular phenotypic description would primarily reveal altered T cell activation motifs, but preexisting divergent differentiation patterns may still be discoverable. For this analysis, we capitalized on the availability of single-cell RNA-seq analysis data describing the transcriptomic differences between T cells from HIV-1 patients that were selected for activation-induced gp120 expression and activated control T cells (47).

The Nussenzweig group reported that among the 10 most significantly enriched biological processes (GO analysis), eight were related to immune system function and further identified gene regulation effects associated with the interferon and antiviral response (47). In our analysis, k-means clustering and subsequent GO motif analysis of the primary T cell data principally reproduced the findings made in latently HIV-1-infected T cell lines and, as expected, identified changes to cell activation patterns: cell metabolism (P_adj: 6.1e−8; cluster 2), cell activation (P_adj: 6.0e−16; cluster 5) immune processes (P_adj: 2.0e−16; cluster 5), and, importantly, cell dedifferentiation (hematopoietic stem cell differentiation; P_adj: 2,8e−19; cluster 6) (Fig. 9A). To ensure that our analysis pipeline would reliably reproduce findings by the Nussenzweig group (47), we performed principal-component analysis. As in the original analysis, the majority of latently HIV-1-infected T cells had a transcriptomic phenotype that was distinct from that of control T cells, but a small group of latent HIV-1 infection events were also hosted in T cells that would not be phenotypically distinguishable from normal T cells (Fig. 9B), confirming that our analytical pipeline produced comparable results.

FIG 9.

scRNA-seq identifies cell dedifferentiation patterns in latently HIV-1-infected primary T cells from patients. Available scRNA-seq data from primary control T cells and latently HIV-1-infected T cells from PLWH on ART following activation-induced reactivation were mined for the presence of signals suggesting differences in the differentiation status (47). (A) Results of k-means clustering followed by GO motif analysis. The listed GO motifs represent the T cell-relevant top motifs for each cluster. (B) Principal-component analysis of scRNA-seq data from patients 603 and 605 shows the diversity of phenotypes of latently HIV-1-infected T cells (blue) and control T cells (red) (47). (C) A list of genes that were found differentially expressed between control and latently HIV-1-infected T cells was used to generate a protein-protein interaction network analysis. Blimp-1 (PRDM1) and IRF4 act as the central network hubs. (D) Central network nodes and their number of connections/edges to other network components. (E) Box-and-whisker plots for differentially expressed genes associated with hematopoietic lineage development found to be expressed in >20% of the control cells or the latently HIV-1-infected T cells. Gene names are in gray.

Network analysis linked only 80 of the 379 differentially expressed genes (P_adj < 0.01; 2-fold) into a direct interaction network (Fig. 9C). This low linkage efficacy was similar to the observations in T cell lines and consistent with the presence of dedifferentiation-induced stochastic gene expression patterns that likely drive the reported intercell heterogeneity (47). Similar to our observations in T cell lines, the network had a very steep hierarchy, with interferon regulatory factor 4 (IRF4) and PR/SET domain 1 (PRDM1), also termed B lymphocyte-induced maturation protein-1 (Blimp-1), forming the central hubs of the network (Fig. 9D). These proteins are primarily known for their roles in interferon signaling and immune response but also play an essential role in T cell development and stem cell biology. IRF4 regulates the choice between T lymphoid-primed progenitor and myeloid lineage fates during embryogenesis (109). The transcription factors Blimp-1 and IRF4 jointly control the differentiation and function of effector regulatory T cells (110), and the transcriptional repressor PRDM1 by itself has a reported role in hematopoietic lineage development and is known to silence stem cell-related genes (111, 112).

We finally explored whether additional information on specific genes involved in lineage development could be extracted from the scRNA-seq data set. scRNA-seq was developed to discover heterogeneity in cell populations, and the commonly used analysis pipelines were designed accordingly. Many genes that are identified as differentially expressed based on commonly used statistical analysis algorithms often produce detectable signals in only a very small fraction of the analyzed cell events (<10% of events). However, this effort sought to identify a shared mechanism of latency control that should be generally present in all cells. Our data generated in T cell lines suggest that while there are differential gene regulation effects in the HSC motif of the different latently HIV-1-infected T cell clones, many of the motif-associated genes were similarly regulated. Guided by these findings, we would expect that for gene expression changes to be biologically relevant and not just statistically significant, they should occur in the majority of the analyzed cell events. Considering that the ability of scRNA-seq to detect signals is affected by factors such as the presence of transcriptional bursts, that gene expression is often not continuous, and that the overall signal resolution of the method is limited, we chose to perform a focused analysis of genes that were expressed in at least 20% of the cell events in the control or in the latently infected sample group. Of the ∼28,000 RNA products that can be theoretically assessed, 12,708 produced a signal in at least a single analyzed cell, but only 168 genes were expressed in at least 20% of the events and were differentially regulated by statistical analysis. Of these, according to published literature, 35 (21%) were associated with hematopoietic stem cell development, clearly supporting the idea that dedifferentiation effects could be a contributing factor to HIV-1 latency stability in primary T cells (Table 3).

TABLE 3.

Differentially expressed genes associated with HSC differentiation in latently HIV-1-infected primary T cells

Symbol	P adj	Log2 fold change	Name	Reference(s)
TOX	5.99E−12	9.37	Thymocyte selection-associated high-mobility group box	113, 114
BCL2L11	1.90E−09	8.69	BCL2-like 11 or BIM	132
BATF3	1.67E−07	8.36	Basic leucine zipper ATF-like transcription factor 3	115 – 117
IRF4	5.89E−06	5.10	Interferon regulatory factor 4	169 – 171
PRDM1	8.06E−06	3.65	PR/SET domain 1 or BLIMP-1	104, 172–176
HUWE1	1.21E−05	5.35	HECT, UBA, and WWE domain-containing E3 ubiquitin protein ligase 1	56, 177, 178
ANXA1	1.94E−05	-7.23	Annexin A1	179
SYNE1	2.33E−05	6.04	Spectrin repeat-containing nuclear envelope protein 1	178
NOL9	1.11E−04	6.23	Nucleolar protein 9	180
GPR183	3.45E−04	4.58	G protein-coupled receptor 183	181
TLE3	4.03E−04	4.06	TLE family member 3, transcriptional corepressor	182, 183
NAMPT	4.62E−04	4.10	Nicotinamide phosphoribosyltransferase	184, 185
RDH10	9.12E−04	5.62	Retinol dehydrogenase 10	186
RAPGEF2	4.90E−03	5.03	Rap guanine nucleotide exchange factor 2	187
GPR171	5.37E−03	3.16	G protein-coupled receptor 171	188, 189
TIGIT	7.99E−03	5.28	T cell immunoreceptor with Ig and ITIM domains	190
PIM2	1.49E−02	−2.23	Pim-2 proto-oncogene, serine/threonine kinase	191
FOXP1	1.94E−02	3.03	Forkhead box P1	192 – 194
CDK6	2.10E−02	4.62	Cyclin dependent kinase 6	118 – 121
PDE4B	2.14E−02	4.12	Phosphodiesterase 4B	195
KMT2A	2.90E−02	4.58	Lysine methyltransferase 2A	196, 197
ADAM 10	3.06E−02	3.50	ADAM metallopeptidase domain 10	198, 199
SRGN	3.41E−02	1.88	Serglycin	200 – 202
SFPQ	4.08E−02	3.69	Splicing factor proline and glutamine rich	203
LGALS1	4.08E−02	−3.39	Galectin 1	204, 205
BRD2	4.89E−02	3.79	Bromodomain containing 2	206, 207
MAF	4.93E−02	1.83	MAF BZIP transcription factor, or c-maf	208, 209
IKZF3	5.01E−02	2.43	Ikaros family zinc finger 3, or Aiolos	126, 210–215
PIK3CD	5.51E−02	−3.66	Phosphatidylinositol-4,5-bisphosphate 3-kinase catalytic subunit D	216
EIF4EBP2	6.92E−02	3.77	Eukaryotic translation initiation factor 4E binding protein 2	217, 218
ZEB2	7.02E−02	4.33	Zinc finger E-box binding homeobox 2	219 – 222
ID2	7.22E−02	2.78	Inhibitor of DNA binding 2	223 – 227
STAT3	7.55E−02	2.64	Signal transducer and activator of transcription 3	228 – 230
SRRT	7.86E−02	3.88	Serrate, RNA effector molecule	231, 232
PTPN1	9.55E−02	3.63	Protein tyrosine phosphatase non-receptor type 1	233, 234

Gene regulation effects for a selection of these genes are shown in Fig. 9E. Serglycin (SRGN; hematopoietic proteoglycan core protein), which is considered an HSC marker, was expressed in 94% of the latently infected cell events, and expression levels were 2-fold increased over that in the control cells that expressed this gene. Among other known factors involved in hemopoietic lineage or lymphocyte development contained in this data set, differentiation-relevant genes that were upregulated include the already discussed PRDM1/Blimp-1 and IRF4, TOX (thymocyte selection-associated high-mobility group box) (113, 114), BATF3 (basic leucine zipper ATF-like transcription factor 3) (115–117), CDK6 (cyclin-dependent kinase 6) (118–121), IKZF3 (Ikaros family zinc finger 3; Aiolos) (122–127), ID2 (inhibitor of differentiation 2) (128, 129), E1A-like inhibitor of differentiation (EID1) (130, 131), RDH10 (retinol dehydrogenase 10) (132), and HUWE1 (HECT UBA and WWE domain-containing E3 ubiquitin protein ligase 1), reported to regulate hematopoietic stem cell maintenance and lymphoid commitment (56).

This analysis is certainly limited by the lower detection sensitivity of scRNA-seq and largely excludes genes that are expressed at lower levels, but the data clearly support the presence of an altered differentiation phenotype in primary latently HIV-1-infected T cells, suggesting that dedifferentiation effects are shared between latently HIV-1-infected clonal T cell lines and latently HIV-1-infected primary T cells.

DISCUSSION

Current efforts to eradicate the latent HIV-1 reservoir generally assume that latent HIV-1 infection events reside in functional host T cells. However, the description of a large, latently HIV-1-infected T cell population that despite potent ex vivo activation would not allow HIV-1 reactivation is not aligned with this idea (23). Functional memory T cells would be expected to respond to stimulation with the generation of an intracellular environment that is highly conducive to HIV-1 expression, but this was not the case for T cells constituting more than half of the latent HIV-1 reservoir (23). In vitro, we recently confirmed that host T cells of latent HIV-1 infection events show a spectrum of activation responsiveness, with a significant portion of the latently infected T cell population being completely inert to TCR/CD3 activation (52). In addition, recent single-cell RNA-seq studies of latently infected T cells identified a high degree of heterogeneity between individual latently HIV-1-infected T cells (47–49), suggesting that HIV-1 latency can exist in host cell environments that differ greatly. Host cell heterogeneity may at least in part explain the differential responsiveness of latently infected T cells to activation. Detailing the underlying biomolecular mechanisms of intercell heterogeneity and activation inertness, and determining whether these phenomena are actually linked, would be important to identify means to improve current cure strategies.

Our finding that extensive transcriptional heterogeneity was not limited to latently infected primary T cells (47–49) allowed us to dissect the underlying molecular biology of this phenomenon by utilizing bulk RNA-seq analysis in clonal T cell lines. These T cell line-derived high-resolution data were then used to guide motif discovery in scRNA-seq data generated using latently HIV-1-infected T cells from PLWH, a method that is not optimized for the discovery of shared motifs, as it generally produces inferior data resolution and is largely limited to the identification of medium- and high-level-expressed genes. For both types of latently HIV-1-infected-T-cell systems, we provide evidence for the presence of gene regulation motifs that indicate host cell dedifferentiation effects, possibly toward an HSC phenotype, that were concealed by a large degree of intercell heterogeneity and likely stochastic gene regulation effects. For the scRNA-seq data from primary T cells, it was remarkable that this phenotype could be detected (Fig. 9; Table 3), as by the design of the experiment, the analyzed cells represent the activation-responsive T cell subpopulation and did not even include the activation-inert latently infected T cell population, which should have a more pronounced phenotype (52).

As we introduce the concept of dedifferentiation as a bimolecular mechanism of latency control in T cells, it is important to emphasize that dedifferentiation of immune cells is not an unusual phenomenon but rather a necessity for normal immune function. Dedifferentiation of mature immune cells has been described in the context of T cell lineages and B cell reversion to progenitor cell types (133, 134), although the relevant dedifferentiation triggers are unknown. It thus seems conceivable that HIV-1 infection, in the few T cells that survive the cytopathic viral effect, could serve as a trigger for some form of uncoordinated dedifferentiation effects and result in the described biomolecular host cell phenotype that supports and stabilizes latent HIV-1 infection but also drives intercell heterogeneity.

The finding that latently HIV-1-infected T cells derived from T cell lines and from patients seemed to be dedifferentiated toward an HSC-like phenotype would provide conclusive explanations for several reported characteristics of latently HIV-1-infected T cells. Early developmental stages of T cells respond with differential activation patterns compared to mature T cells, explaining the observed activation inertness spectrum (135–138). As stem cells have self-renewal capacity, an HSC-like phenotype would also provide a biological rational for the capacity of latently infected T cells to proliferate in the absence of external stimulation (14, 139).

The idea that dedifferentiation is part of a cellular program that stabilizes latent infection events is indirectly supported by reports that cell-differentiating agents destabilize latent HIV-1 infection. It has been demonstrated earlier that the cell-differentiating compound hexamethylene bisacetamide (HMBA) triggers HIV-1 reactivation in some experimental systems (140, 141). We demonstrated previously that drugs that exhibit secondary cell-differentiating abilities (e.g., dactinomycin, aclacinomycin, and cytarabine) actually prime latent HIV-1 infection events in primary T cells for reactivation (142). It is important to note that all histone deacetylase (HDAC) inhibitors also have cell-differentiating abilities. One of the most prominent HDAC inhibitors used in previous HIV-1 reactivation studies, suberoylanilide hydroxamic acid (SAHA; vorinostat), was initially conceived as a cell-differentiating agent to treat cancer (143–147). Similarly, BRD4 inhibitors, such as JQ1, another class of compounds considered latency reversal agents, also affect T cell differentiation. For example, JQ1 has been reported to affect the differentiation of naive CD4 T lymphocytes into Th17 cells (148).

There is also reported precedent for a possible role of the identified individual proteins for host cell differentiation and HIV-1 control. For example, two of us (K.K.M. and A.J.H.) reported that PRDM1/Blimp-1 represses basal and Tat-mediated HIV-1 transcription (149). Blimp-1 was found to bind an interferon (IFN)-stimulated response element within the HIV-1 provirus and was displaced following T cell activation. Blimp-1 was highly expressed in memory CD4⁺ T cells, a population enriched for latent infection events, making Blimp-1 an intrinsic factor that possibly predisposes CD4⁺ memory T cells to latent HIV-1 infection. Blimp-1, together with ZNF683, has further been reported to instruct a transcriptional program of tissue residency in lymphocytes (104), which may explain why tissue-resident T cells are a prominent reservoir of latent HIV-1 infection (150). Similarly, we have previously shown that targeting STAT3 destabilizes latency and promotes HIV-1 reactivation (52).

Another example is the role of ALDH1A2 in latency control. We have demonstrated loss of ALDH1A2 in latently infected T cells, the enzyme that converts retinol into retinoic acid, and the ability of continuous retinoic acid addition to drive HIV-1 reactivation, possibly by host cell redifferentiation. These observations could provide a biomolecular explanation for the finding by Zhang et al. that describe how retinoic acid supplementation of medium used in HIV-1 outgrowth cultures significantly improved the efficacy of virus recovery from latently HIV-1-infected primary T cells derived from patients on antiretroviral therapy and allowed viral outgrowth in samples that were negative under standard culture conditions (151). Loss of ALDH1A2 could also be of possible interest for maintaining the pool of latently HIV-1-infected T cells, as inhibition of aldehyde dehydrogenase activity and retinoid signaling has been shown to induce the expansion of human hematopoietic stem cells (152).

Taken together, the data support the idea that latently HIV-1-infected T cells are in an atypical differentiation state that provides an HSC-like phenotype, which could drive stochastic gene expression effects that would explain intercell heterogeneity and support activation inertness. This would explain why stimuli that normally activate T cells cannot uniformly stimulate latently HIV-1-infected T cells and trigger HIV-1 reactivation. Our findings emphasize that future therapeutic reactivation protocols should be based on a well-timed combination of differentiating and activating agents to overcome the inefficiency of current treatment approaches.

MATERIALS AND METHODS

Cell culture and reagents.

All T cell lines and populations were maintained in RPMI 1640 supplemented with 2 mM l-glutamine, 100 U/mL penicillin, 100 µg/mL streptomycin, and 10% heat-inactivated fetal bovine serum (FBS). FBS was obtained from HyClone (Logan, UT) and was tested on a panel of latently infected cells to ensure that the utilized FBS batch did not spontaneously trigger HIV-1 reactivation (153, 154). Jurkat T cells were initially obtained from the ATCC (clone E6-1; ATCC TIB-152). The latently infected CA5 and EF7 T cells were described earlier (153, 155). In CA5 T cells, the virus is integrated into an intron of the RMB12/CPNE1 gene in the same-sense orientation relative to the transcriptional direction of the host gene. In EF7 cells, the virus is integrated into an intron of the WHSC1 gene in the antisense orientation relative to the transcriptional direction of the host gene. In CG3 T cells, the virus is integrated in an intergenic, noncoding region between the TIGD5 and the PYCRL gene. The latently HIV-1-infected JWEAU-A10 and JWEAU-C6 cells have been described in detail (52).

For all HIV-1 reactivation experiments, GFP was used as a surrogate marker of HIV-1 expression (142, 156). GFP expression was routinely measured 24 or 48 h postinduction using flow-cytometric analysis, and data are presented as percent GFP-positive cells. In all experiments where compounds or drugs were added to latently HIV-1-infected T cells, these compounds/drugs were not removed prior to the addition of a secondary stimulus. Flow-cytometric analysis was performed on a Guava EasyCyte (Guava Technologies, Inc.) or an LSRII (Becton Dickinson) instrument. Data were analyzed using FlowJo software (TreeStar, Ashland, OR). All antibodies were purchased from Becton Dickinson (Franklin Lakes, NJ). The phorbol esters phorbol 12-myristate 13-acetate (PMA) and prostratin were purchased from Sigma. The PKC activator bryostatin was purchased from EMD Millipore. The LDHA inhibitor (R)-GNE-140 was purchased from Selleckchem.

Expression plasmids.

To obtain an mCherry-HIV reporter virus, the enhanced GFP (eGFP) codon in the pBR43IeG-NA7nef plasmid (157), a proviral vector designed to coexpress Nef and eGFP from a single bicistronic RNA, was replaced by the mCherry codon. pBR43IeG-NA7nef was digested with NcoI/SmaI to remove the eGFP codon and replaced by a mCherry amplicon where the ATG was deleted. The ATG for eGFP/mCherry is contained in the NcoI site (10134 to 10139 bp). The following primers were used: NA7(mCherry)-NcoI fwd, GATGATAATACCATGGTGAGCAAGGGCGAGGagg, and NA7(mCherry)-SmaI rev, AAAAAGTGGCTACCCGGGCTACTTGTACAGCTCGTCCATGCC.

Lactate quantitation.

Cells were seeded in 6-well plates at 150,000 cells/mL in a total volume of 10 mL per well to ensure no effects from liquid removal over the course of the kinetic. At time zero and every 12 h after, 400 µL was removed from each well and split in half, with one part counted on a capillary flow cytometer to determine cell numbers (proliferation capacity) (Guava EasyCyte; Guava Technologies, Inc.). The remaining portion was spun at 300 × g for 5 min to pellet cells. Per the manufacturer’s recommendation, the supernatant was spun through a deproteinizing column (Spin-XR UF 500, 10,000 molecular weight cutoff [MWCO]; Corning no. 431478) to remove extracellular lactate dehydrogenase. All deproteinized supernatants were immediately frozen at −80°C until assayed. To measure lactate levels, supernatants were assayed using a colorimetric reaction (lactate assay kit, MAK064-1KT; Sigma-Aldrich). A standard curve was constructed using the kit-provided 100 mM lactate sample, diluted 1:500, and then in an 11-point curve of 1:2 dilutions, as well as a blank. Samples were thawed and diluted 1:25 in assay buffer and then again for final dilutions of 1:250 and 1:625. Samples, standards, and blanks were mixed 1:1 with 25 µL of master mix (buffer, enzyme, probe) in an opaque walled, clear bottom 384-well plate. Absorbance of the colorimetric reaction was read on a Cytation3 (BioTek) spectrophotometer at 570 nm, and the standard curve was calculated using GraphPad Prism 7.05 (GraphPad). Statistics were calculated using Prism’s two-way analysis of variance (ANOVA) with Dunnett’s correction for multiple comparisons. Results for three biological replicates assayed at the same time are shown, and data are representative of three independent experiments.

HIV-1 latency establishment assays.

To determine latency establishment levels under defined experimental conditions, 20 to 30 individual cell cultures of either Jurkat control cells or latently HIV-1-infected CA5 or EF7 T cells were infected over a wide range of multiplicities of infection (MOIs) using a VSV-G-pseudotyped HIV-mCherry reporter virus. The choice of mCherry as a fluorescent marker allowed us to distinguish between superinfection events (mCherry) and the original latent infection events with an HIV-GFP reporter virus (157). Viruses were pseudotyped with VSV-G to render infection efficacy CD4 independent. On day 3 or 4 postinfection, active HIV-mCherry infection levels and reactivated HIV-GFP infection levels were determined by measuring the percentage of mCherry- and GFP-positive cells using flow-cytometric analysis. To determine the size of the latent HIV-mCherry superinfection reservoir formed between days 14 and 21 postinfection, a sample of each infection culture was taken and stimulated with PMA overnight, which triggered the reactivation of latent HIV-1 superinfection events (mCherry-positive cells). By subtracting the percentage of residual active infection in untreated cell cultures from the level of active infection in a sample of each culture in which HIV-1 reactivation had been triggered by PMA, using mCherry expression as a surrogate marker of HIV-1 superinfection, we could calculate the size of the latent reservoir that had been established.

ChIP-qPCR.

Chromatin immunoprecipitations were performed as previously described (26). Antibodies used were as follows: anti-RNAP II (Santa Cruz Biotechnology), anti-AcH3 (Upstate Biotechnology) and rabbit IgG (Upstate Biotechnology). Quantitative real-time PCR analysis was carried out using SYBR green reagents and the primers 5′-TGCTTTTTGCCTGTACTGGGTCTC-3′ and 5′-GCACACACTACTTGAAGCACTCAAG-3′, which amplify the −14 to +113 region of the HIV-1 LTR.

ATAC-seq.

ATAC-seq data for parental Jurkat T cells were generated. For each sample, 50,000 to 100,000 cells were collected. ATAC-seq was conducted as described elsewhere (158), and 2 × 50 rapid-run paired-end sequencing was performed on a HiSeq 2500 system (Illumina). An average of 35 million sequencing reads (50-bp length) per sample were obtained from the Illumina analysis pipeline. Sequence reads were mapped to the human genome hg19 using bowtie2 (159). Uniquely matched reads were retained, and PCR duplicate reads were removed. Peak calling was performed by MACS2 with a P value threshold of 1 × 10⁻⁵ (160). ATAC-seq results were visualized using the Integrated Genome Browser (IGB) software (161).

RNA-seq.

For bulk RNA-seq analysis, total RNA was isolated by using miRNeasy minikit (Qiagen). Illumina library preparation and sequencing were performed by Genewiz (New Jersey, USA). Sequencing adapters were trimmed from reads using Trim Galore! (Babraham Bioinformatics). Trimmed reads were aligned to human Hg38 using STAR (24), and reads mapping to individual genes were counted using HTSeq-Count. Raw read counts were normalized, and differential expression was analyzed using DESeq2 (25). Genes that differed by at least 2-fold from uninfected controls and had an adjusted P value of <0.01 were used for network and pathway analysis. Pathway analysis was performed using MetaCore software. Single-cell RNA-seq data from the study reported in reference 47 were obtained from the public-domain NCBI GEO server (GSM2801437). Data sets from patients 603 and 605 were used for the analysis. Additional patient information can be found in reference 162.

Data analysis.

Several software solutions were used to analyze or interpret the data, including MetaCore software (Clarivate) for network analysis, iDEP.91 (RNA-seq) (163), Network analyst (164, 165), MetaScape (166), and the Molecular Signatures Database (MSigDB) (167, 168). Pathway analysis functions of these public-domain software programs were used to analyze the data in the context of described biological processes, pathways or networks and to detect possible shared signal signatures in the gene expression. To generate protein-protein interaction networks using the differentially regulated genes identified by RNA-seq as input, we used the MetaCore direct interaction algorithm, which exclusively uses the identified genes as seed nodes. MetaScape and MSigDB were used to compute overlap between our identified data describing changes to the transcriptome or proteome and curated reference data sets.