Epigenetic Regulation of HIV-1 Latency by Cytosine Methylation

Human immunodeficiency virus type 1 (HIV-1) persists in a latent state within resting CD4+ T cells of infected persons treated with highly active antiretroviral therapy (HAART). This reservoir must be eliminated for the clearance of infection. Using a cDNA library screen, we have identified methyl-CpG binding domain protein 2 (MBD2) as a regulator of HIV-1 latency. Two CpG islands flank the HIV-1 transcription start site and are methylated in latently infected Jurkat cells and primary CD4+ T cells. MBD2 and histone deacetylase 2 (HDAC2) are found at one of these CpG islands during latency. Inhibition of cytosine methylation with 5-aza-2′deoxycytidine (aza-CdR) abrogates recruitment of MBD2 and HDAC2. Furthermore, aza-CdR potently synergizes with the NF-κB activators prostratin or TNF-α to reactivate latent HIV-1. These observations confirm that cytosine methylation and MBD2 are epigenetic regulators of HIV-1 latency. Clearance of HIV-1 from infected persons may be enhanced by inclusion of DNA methylation inhibitors, such as aza-CdR, and NF-κB activators into current antiviral therapies.

In resting CD4 + T cells, HIV-1 is maintained in a latent state by multiple factors that inhibit virus gene expression after integration into cellular DNA. In particular, several studies have highlighted the critical role of chromatin structure at the site of provirus integration in repressing provirus transcription. Sequence-specific transcription factors can recruit histone deacetylases (HDACs) and other chromatin-modifying enzymes to the provirus promoter, resulting in transcriptional repression and virus latency [15][16][17][18][19]. Interestingly, the mechanism by which virus escapes silencing by these sequence-specific factors in a productive infection is unknown. Additionally, resting CD4 + T cells are deficient in transcription factors essential for HIV-1 transcription [20], and latent virus can be reactivated by stimulation of T cell pathways that activate these factors [5][6][7][8]. The provirus integration site can also be a determinant of latency, either by making the provirus susceptible to transcriptional interference from cellular genes [21][22][23][24] or by suppressing virus transcription through the formation of heterochromatin [25]. Post-transcriptional mechanisms affecting the export [26] or translation [27] of HIV-1 mRNAs constitute other blocks to HIV-1 gene expression during latency.
The resting state of CD4 + T cells and the activity of HDACs are two of the best-understood characteristics of latency, but stimulation of resting CD4 + T cells or inhibition of HDACs in HIV-infected patients do not appreciably decrease the latent reservoir when combined with HAART [28][29][30][31][32].
The study of latently infected cells is hampered by their rarity in HIV-infected individuals and the lack of a marker for latent infection. For these reasons, we developed the J-Lat cell lines as an in vitro model of HIV-1 latency [33]. Similar to latently infected CD4 + T cells, the J-Lat cells harbor a full-length HIV-1 genome that is transcriptionally competent, is integrated within actively transcribed cellular genes, and is inhibited at the transcriptional level. Additionally, the latent provirus integrated in the J-Lat cell lines encodes the GFP gene, providing a fluorescent marker of HIV-1 transcriptional activity.
To identify novel mechanisms of HIV-1 latency, we have conducted a cDNA screen in J-Lat cells for genes that reactivate latent HIV-1. This screen identified a portion of methyl-CpG binding domain protein 2 (MBD2), a transcriptional repressor that binds methylated DNA. We found that the HIV-1 promoter is hypermethylated in J-Lat cell lines and in primary CD4 T cells at two CpG islands surrounding the HIV-1 transcriptional start site.
Most importantly, we found that a small molecule inhibitor of DNA methylation, 5-aza-29deoxycytidine (aza-CdR), synergizes with NF-kB activators to promote a dramatic increase in virus gene expression. Aza-CdR is approved for use in humans to treat myelodysplastic syndrome [34] and may promote the reactivation of latent HIV-1 and the clearance of latently-infected cells in combination with HAART in HIV-infected patients.

A genetic screen to identify novel regulators of HIV-1 latency
The J-Lat cells are clonal cell lines isolated after infection of Jurkat cells with a HIV-1 virus encoding GFP. Latently infected cells were selected that were GFP-negative at the basal state but became GFP-positive after treatment with TNF-a. Treatment of each cell line with TNF-a reactivated latent HIV-1 to a different extent, depending on the cell line ( Figure 1A). To identify cellular genes that control HIV-1 latency in this system, a complementary DNA (cDNA) library was made from the Jurkat T cell line and cloned into a plasmid encoding the pBMN-CSI-T retrovirus vector, which expresses tomato fluorescent protein as a marker ( Figure 1B).
To confirm that this vector mediates expression of cloned cDNAs at a level sufficient for reactivation of latent HIV-1, a positive control virus was produced that encodes NF-kB RelA, which reactivates latent HIV-1 in J-Lat cells [18]. Infection of J-Lat cell line 6.3 with the RelA-encoding virus caused a 3.5-fold increase in HIV-1 gene expression compared to a control virus that lacks an insert ( Figure 1C).
The cDNA library was packaged into retroviral particles and introduced into the J-Lat 6.3 cell line via infection (Table 1). GFPpositive cells, indicative of reactivated latent HIV-1, were isolated by fluorescence activated cell sorting (FACS). cDNA library inserts were amplified from genomic DNA obtained from these cells by

Author Summary
Current drug therapies inhibit replication of the human immunodeficiency virus (HIV). In patients undergoing these therapies, the amount of HIV is reduced to an undetectable level and HIV-related disease subsides. However, stopping antiviral drug therapy results in the quick return of HIV and of disease. One reason for this is latently infected cells, in which virus replication is temporarily halted. When drug therapy is stopped, virus from these latently infected cells can resume infection and spread to other cells in the patient, resulting in the return of disease. Here, we demonstrate that one mechanism of latency is DNA methylation, in which chemical groups called methyl groups are added to HIV DNA. We also identify a host protein called methyl-CpG binding domain protein 2 (MBD2) that binds methylated HIV DNA and is an important mediator of latency. Furthermore, we demonstrate that a drug that inhibits DNA methylation potently reactivates latent HIV. Novel strategies to eliminate or reduce the latent reservoir are necessary. Our findings may prove useful in the development of novel therapies to efficiently reactivate latent HIV-1, thus making it susceptible to current drug therapies. PCR with virus-specific primers ( Figure S1A) and recloned into pBMN-CSI-T.

MBD2 regulates HIV-1 latency and repression of methylated DNA
One clone identified in this screen, MBD2 1345-1947 , corresponded to nucleotides 1345-1947 of the mRNA encoding the MBD2 transcriptional repressor ( Figure S2). Importantly, the first ATG within this clone is in frame with the authentic MBD2 initiation codon, indicating a truncated protein corresponding to amino acids 388 to 411 of full-length MBD2 could be translated.
MBD2 is a member of the methyl-CpG binding domain family of proteins, which possess methyl-CpG binding domains (MBDs). Similar to other members of this family, MBD2 specifically binds methylated DNA and mediates transcriptional repression by recruitment of the nucleosome remodeling and histone deacetylation (NuRD) complex that includes chromatin remodeling and HDAC activities [35][36][37].
To confirm that MBD2 1345-1947 reactivates latent HIV-1, J-Lat cells were transfected with an expression vector for this polypeptide. Transfection of J-Lat 6.3 with MBD2 1345-1947 induced a 5-fold greater reactivation of latent HIV-1 in comparison to an empty vector control ( Figure 2A). Since MBD2 inhibits transcription of methylated DNA [35], the identification of a C-terminal fragment of MBD2 in our screen indicated that this fragment inhibits endogenous MBD2 function in a dominant-negative manner. Furthermore, identification of this fragment implicated full-length, endogenous MBD2 in the repression of HIV-1 transcription during latency. To establish the role of endogenous MBD2 in HIV-1 latency, J-Lat 6.3 was transfected with a pool of siRNAs corresponding to this factor. This resulted in an 80 percent reduction in the level of MBD2 mRNA compared to cells transfected with a non-targeting control siRNA pool ( Figure 2B, left panel). Depletion of MBD2 resulted in a 300 percent increase in HIV-1 mRNA compared to those transfected with the control siRNA pool ( Figure 2B, right panel). These data demonstrate that MBD2 participates in the repression of HIV-1 transcription during latency.
Since MBD2 inhibits transcription of methylated DNA [35], we believed the C-terminal MBD2 fragment identified in our screen might reactivate latent virus by inhibiting endogenous MBD2 function. To test MBD2  for this activity, we examined its effect on transcription of methylated DNA in a heterologous system. 293T cells were cotransfected with an expression vector for MBD2  and with another plasmid encoding GFP under the control of the CMV promoter (pEGFP-N1). This latter plasmid was either methylated in vitro (meGFP) or left unmethylated (GFP). Plasmid methylation was confirmed by resistance to Hpa II cleavage ( Figure S1B) and reduced GFP expression in transfected 293T cells ( Figure 2C). Importantly, cotransfection of the MBD2 1345-1947 plasmid with methylated pEGFP-N1 increased the proportion of GFP-positive cells from 58 to 72 percent ( Figure 2D, left panel). Furthermore, derepression by MBD2 1345-

HIV-1 latency is associated with cytosine methylation in provirus CpG islands
Cytosine methylation is an epigenetic modification that inhibits transcription when CpG islands, clusters of CpG dinucleotides proximal to a transcription start site, are hypermethylated [38]. To determine whether the HIV-1 genome encodes CpG islands, the methprimer program [39] was used search the HIV-1 provirus nucleotide sequence. Two CpG islands were identified flanking the transcription start site at positions -194 to -94 and 180 to 368 ( Figure 3A). These islands overlap with two regions that were previously shown to be nucleosome-free [40] and rich in transcription factor binding sites [41], two features usually associated with bona fide CpG islands [38]. To determine whether HIV-1 CpG islands are methylated during latency, their methylation state was analyzed by bisulfite-mediated methylcytosine mapping. Figure S3 shows nucleotide sequence of the HIV-1 promoter, positions of CpG islands, and the particular CpGs subjected to methylation analysis. We found that both CpG islands were hypermethylated in four different J-Lat cell lines, with the majority of CpGs methylated more than 70 percent of the time ( Figure 3B and Figure S4A-D). In sodium bisulfite-treated DNA, cytosine was converted to thymine in greater than 99 percent of all CpN dinucleotides (N = A, T, or C), confirming efficient bisulfite conversion of non-methylated cytosines ( Figure S5A).
Cytosine methylation recruits transcriptional repressors to the HIV-1 promoter MBD2 mediates transcriptional repression by acting as a bridge between hypermethylated CpG islands and chromatin modifying enzymes, including HDACs [42]. To test whether MBD2 is recruited to the HIV-1 provirus in vivo, we performed chromatin immunoprecipitation (ChIP) assays. Chromatin from J-lat cells was incubated with MBD2 antisera and the immunoprecipitated material analyzed by quantitative PCR for presence of HIV-1 provirus. We observed recruitment of MBD2 to CpG island 2 of the HIV-1 genome, but observed no recruitment to CpG island 1 in comparison to a negative control ( Figure 4B, first panel). Treatment of J-Lat 6.3 with aza-CdR, an inhibitor of DNA methylation, caused up to a 50 percent decrease in methylation, depending on the CpG analyzed ( Figure 4A and Figure S4E), demonstrating that HIV-1 DNA methylation is reversible. It should be noted that the data for PBS-treated J-Lat 6.3 in Figures 3B and 4A are from the same experiment. Importantly, MBD2 recruitment to CpG island 2 was eliminated when cytosine methylation was inhibited by treatment of the cells with aza-CdR ( Figure 4B, second panel). Next, we tested for the presence of HDAC2, an MBD2 cofactor, at CpG island 2 during latency. Comparable to MBD2, HDAC2 was recruited to CpG island 2 during latency and was lost after treatment with aza-CdR ( Figure 4B, third panel). In contrast, inhibition of methylation by aza-CdR was associated with increased Sp1 recruitment to CpG island 2 ( Figure 4B, fourth panel). These data demonstrate that during latency, multiple components of the NuRD complex recognize the methylated HIV-1 CpG island 2 and that this recruitment may be pharmacologically reversed.

Synergistic reactivation of latent HIV-1 by aza-CdR and NF-kB activators
The finding that methylation of CpG islands flanking the HIV-1 transcription start site can be reversed with aza-CdR suggests that aza-CdR could reactivate latent HIV-1. Aza-CdR alone, however, showed little effect in terms of mean fluorescence intensity or the proportion of GFP-positive cells ( Figures 4C and 4D). As previously observed, treatment with TNF-a reactivated latent HIV-1 in only a fraction of the cell population ranging from 16 to 41 percent depending on the J-Lat cell line studied ( Figure 1A). In contrast, dual treatment of latently infected J-Lat clonal cell lines (lines 6.  Table S1). Each of these increases was nearly 20-fold greater than the additive effect of the two reagents (Table S1). Synergistic activation of transcription was specific for the HIV-1 promoter. Analysis of J-Lat 6.3 by RT-PCR found that aza-CdR and TNF-a synergistically activated HIV-1 transcription ( Figure S6A  synergistic reactivation was observed when aza-CdR was combined with the HDAC inhibitor suberoylanilide hydroxamic acid (SAHA), with an effect about two-fold greater than the additive effect of the drugs ( Figure S7).
We show in four different J-Lat cell lines that near-complete reactivation of latent HIV-1 required treatment with both an NF-kB activator and an inhibitor of DNA methylation ( Figure 4D, lower panel). J-Lat A2 is another clone that harbors a latent HIVderived vector encoding only the viral promoter and Tat. In contrast to the other cell lines analyzed here, latent virus in J-Lat A2 did not require aza-CdR for full reactivation. TNF-a alone reactivated the majority of latent virus in J-Lat A2 (Figures 4D, lower panel, and 4F) [33]. These data show that treatment with a methylation inhibitor is necessary for full reactivation of some, but not all, J-Lat cell lines.

Cytosine methylation contributes to HIV-1 latency in a polyclonal cell population
To confirm that cytosine methylation is regularly associated with HIV-1 latency, a polyclonal population of latently infected Jurkat T cells was generated by infection with virus produced from the R7/E 2 /GFP clone. All HIV-1 proteins are expressed from this full length HIV-1 molecular clone, except Nef, which is replaced with GFP, and Env, which is suppressed by a frameshift mutation. FACS was used to separate latently infected/uninfected GFP-negative cells from productively infected GFP-positive cells ( Figure 5A). To compare the infection rate of this population to that of the J-Lat cells, quantitative PCR for HIV R7/E2/GFP sequence was performed on genomic DNA from the polyclonal population 14 and 72 days post infection. The quantity of HIV-1 DNA was normalized to cellular DNA using PCR primers that anneal upstream of the b-actin gene. The level of HIV DNA in these cells ranged from 9-to 14-fold less than that detected in J-Lat cells, indicating a lower rate of infection ( Figure 5B). Bisulfitemediated methylcytosine mapping of HIV-1 DNA from the productive population found hypomethylation, with no detectable methylation at most CpGs. In direct contrast, methylcytosine mapping of the latent population found hypermethylation, with the majority of CpGs methylated more than 68 percent of the time ( Figure 5D and Figure S4F). In sodium bisulfite-treated DNA, cytosine was converted to thymine in greater than 99 percent of all CpN dinucleotides (N = A, T, or C), confirming efficient bisulfite conversion of non-methylated cytosines ( Figure S5B).
Reactivation of latent HIV-1 was also examined in this population. After approximately two months, the proportion of cells with active HIV-1 remained stable at 0.65% ( Figure 5C). Cells were then treated with TNF-a, aza-CdR, or TNF-a plus aza-CdR. TNF-a reactivated latent HIV-1, with a 1.5-fold greater proportion of cells with active virus ( Figure 5E). Importantly, latent HIV-1 was also reactivated by aza-CdR alone, with a twofold greater proportion of cells with active virus ( Figure 5D). These observations indicate that, after infection of Jurkat cells in vitro, a subset of latently infected cells exists that can be reactivated solely by inhibition of DNA methylation.

HIV-1 latency is associated with cytosine methylation in primary cells
The similarities of J-Lat cells to latently infected CD4 + T cells have established the utility of this experimental system for identifying and characterizing mechanisms of HIV-1 latency. However, because J-Lat cells divide autonomously and possess other aberrations associated with cellular transformation, cytosine methylation was analyzed in a recently developed primary cell model of latency [43]. In this system, naïve CD4 + T cells are purified from uninfected donors and activated under conditions that drive them to become memory cells with either a Th1, Th2, or non-polarized (NP) phenotype [44]. These differentiated cells are then infected with HIV-1 and viral expression is monitored. The phenotype of NP cells generated ex vivo ( Figure S8) closely resembles that of central memory CD4 + T cells found in vivo, which persist for years in secondary lymphoid organs and can differentiate into effector memory CD4 + T cells [45]. A high rate of HIV-1 latency is observed in NP memory CD4 + T cells [43].
To determine if HIV-1 latency is associated with cytosine methylation in primary CD4 + T cells, bisulfite-mediated methylcytosine mapping was performed on CD4 + T cells activated under NP, Th1, and Th2 polarizing conditions and infected with HIV-1. Cells were infected with virus produced from the DHIV virus clone [46], in which CpG island 2 is conserved. Five days postinfection, p24 gag was detected in all three subsets ( Figure 6A). At this early time point, the HIV-1 CpG island in the NP and Th1 populations was hypomethylated, with most CpGs methylated only 0 or 10 percent of the time, respectively ( Figure 6D and Figure S4G). Significant methylation was detected in Th2 cells, with most CpGs methylated 33 percent of the time ( Figure 6D and Figure S4G). Two weeks post-infection, NP cells had returned to a quiescent state and HIV-1 gene expression, as measured by intracellular p24 gag expression, was low ( Figure 6C, left panel). However, stimulation with antibodies against CD3 and CD28 dramatically increased HIV-1 gene expression, indicating a large population of latently infected cells ( Figure 6C, right panel). Importantly, CpG island methylation in latently infected NP cells was greater than in productively infected NP cells, with the majority of CpGs methylated 67 percent of the time ( Figure 6E, S5D, and S4G). In sodium bisulfite-treated DNA, cytosine was converted to thymine in greater than 98 percent of all CpN dinucleotides (N = A, T, or C), confirming efficient bisulfite conversion of non-methylated cytosines ( Figure S5C). These data confirm that T cell quiescence is associated with methylation of HIV-1 CpG islands and latency in memory CD4 + T cells.

Discussion
Here, we describe a novel, phenotype-based screen to identify cellular proteins that control HIV-1 latency. This screen identified the transcriptional repressor MBD2 and led to the discovery that the latent HIV-1 provirus is hypermethylated in an in vitro model for HIV-1 latency and in primary lymphocytes latently infected with HIV-1. Based on these observations, we designed and tested a novel strategy for reactivation of latent HIV-1 using the synergistic activities of an inhibitor of cytosine methylation and activators of NF-kB signaling.
HIV-1 latency is likely to be a multifactorial process and a number of different mechanisms have been proposed to account for the establishment and the maintenance of the latent phenotype [13,14,20]. NF-kB signaling reactivates latent HIV [47][48][49][50], but data reported here and elsewhere [16,51,52] indicate that a significant proportion of latent HIV-1 remains silent when NF-kB is activated in the J-Lat clones or other cells. We show here that inhibiting provirus methylation leads to an almost complete reactivation of latent HIV-1 in the J-Lat cell lines when combined with activators of NF-kB. These data are consistent with the model that sequence-specific transcription factors and cytosine methylation cooperate to maintain HIV-1 latency. In the latent state, HDAC1 is recruited to the HIV-1 promoter by several sequencespecific factors including NF-kB p50 [18], CBF-1 [19], and Yin-Yang 1 [15]. Additionally, in microglial cells CTIP-2 has been shown to recruit HDAC1 to the HIV-1 promoter [53]. Our new observations demonstrate that MBD2 is also recruited to the latent HIV-1 promoter via the second CpG island ( Figure 7A). We propose that MBD2 silences transcription by recruitment of the   NuRD complex or other factors. This is supported by our finding that another component of NuRD, HDAC2, is also recruited to hypermethylated CpG island 2 during latency. NF-kB activation relieves one component of the transcriptional block, causing decreased CBF-1 [19] and NF-kB p50 homodimer recruitment to the HIV-1 promoter, as well as increased binding of the NF-kB RelA activator [18] (Figure 7B). Inhibition of cytosine methylation relieves another component of the transcriptional block, causing decreased MBD2 and HDAC2 recruitment to HIV-1 CpG island 2 ( Figure 7C). The combination of NF-kB activation and methylation inhibitors eliminates both transcriptional blocks, causing a synergistic increase in HIV-1 transcription and reactivating virus in the majority of cells ( Figure 7D).
In polyclonal Jurkat cells, the magnitude of HIV-1 reactivation appeared to be smaller than for the J-Lat clones. This was not, however, because TNF-a or aza-CdR were ineffective, but because of the small proportion of latently infected cells in this population compared to the J-Lat cells, each of which harbor a provirus. To ensure no more than one provirus per cell, they were infected at a low multiplicity that left approximately 90 percent of the cells uninfected. Quantitative PCR for HIV DNA demonstrated the small proportion of infected cells in this population compared to the J-Lat clones. Virus reactivation by TNF-a and aza-CdR is highly significant, but is somewhat obscured by the large background of GFP-negative uninfected cells.
The role of epigenetic mechanisms in suppression of HIV-1 transcription during latency has not been fully addressed to date. Sequence-specific transcription factors contribute to latency by recruiting HDACs and other repressors to the virus promoter. These findings present a paradox, however, because latent virus can be of wild-type nucleotide sequence [33], and yet transcription is suppressed. Here, we present evidence that HIV-1 latency is also maintained at the epigenetic level by the methylation of provirus DNA and recruitment of MBD2. This protein brings transcriptional repressors to methylated DNA, and the MBD2 1345-1947 fragment isolated from the screen may reactivate latent HIV-1 by disrupting the interaction of MBD2 with an interaction partner. Importantly, after each round of DNA replication, cytosine methylation is faithfully reproduced in a process that is directed by previously methylated DNA [54]. Thus, identical DNA sequences can be either active or silenced depending on their methylation status. Our and previous findings suggest that both HDACs and cytosine methylation contribute to HIV-1 latency, in agreement with a growing body of evidence demonstrating cooperation between these two gene silencing mechanisms [55,56].
The rarity of latently infected cells and the lack of a marker for latent HIV-1 infection necessitate the use of in vitro model systems for detailed studies of this process. Transformed cells such as the Jurkat line may show aberrant DNA methylation patterns at specific loci [57], possibly complicating analyses of cytosine methylation and HIV-1 latency. However, when Jurkat cells are infected with HIV-1 the proportion of cells that become latently infected vs. productively infected is small, suggesting that transformation does not result in the indiscriminate methylation and repression of HIV-1. Furthermore, high-resolution analysis of cytosine methylation in primary and transformed cells has found less aberrant methylation of CpG island promoters in transformed cells than had been previously hypothesized based on candidate gene studies [58]. Importantly, we confirmed the association between HIV-1 latency and cytosine methylation in a primary cell model of HIV-1 latency.
The findings reported here, based upon a near full-length HIV-1 with wild type LTR and Tat sequences, add to previous studies that have used mutated forms of the HIV-1 promoter to describe a role for methylation of HIV-1 DNA in latency [59][60][61]. One report, however, has described latent HIV-1-derived vectors, or ''minigenomes,'' that lack all virus genes except for Tat. J-Lat clone A2 harbors exactly such a minigenome. Importantly, these latent minigenomes are not methylated [62] and are almost fully reactivated by TNF-a treatment, unlike the full-length genome [33]. Apparently, screens to isolate latently infected clones produced very different results when mini-instead of full-length genomes were used. For the minigenomes, removal of virus genes and repositioning of Tat out of its normal genomic context are likely to have altered transcriptional control. This alteration may have influenced the type of latently infected cell recovered from the screen. The screens that produced the J-Lat cells also selected for mechanisms that silence HIV-1 within several days after infection. Other screens for cells that silence HIV-1 at later time points have identified additional silencing mechanisms [19].
Our results indicate that cytosine methylation can be an important component of HIV-1 latency. In the case of the fulllength J-Lat clones, a high degree of cytosine methylation is detected during latency. In the case of minigenomes such as J-Lat A2 or that characterized by Pion et al, the persistent lack of methylation may permit efficient reactivation by TNF-a alone. Pion et al also describe a lack of cytosine methylation in latently infected PBMCs, but the large proportion of productively infected cells in the analyzed population complicates this assay.
Novel approaches are required to reactivate latent HIV-1 in infected persons. Therapies that interfere with cytosine methylation are attractive candidates to reactivate suppressed virus and purge the latent HIV-1 reservoir. In uninfected human subjects, aza-CdR causes decreased CpG island methylation and reactivation of a silenced gene [63,64]. In HIV-infected individuals, a similar decrease in methylation should be attainable and could reactivate latent virus. Furthermore, mechanisms by which aza-CdR induces hypomethylation are well understood [65,66] and this pharmaceutical is approved for use in humans. Aza-CdR acts directly upon the HIV-1 provirus, because it reactivates HIV-1 transcription in the presence of cycloheximide. HIV-1 was reactivated to a lesser extent in this experiment, and this could result from cellular toxicity or inhibition of an indirect component to reactivation. Any indirect component to HIV-1 reactivation would not, however, make aza-CdR any less effective a drug for reactivation of latent HIV-1 in humans. Aza-CdR synergizes with prostratin, a phorbol ester that triggers reactivation of latent HIV-1 in the absence of T cell activation and inhibits de novo virus infection [67]. Thus, the combination of aza-CdR and prostratin may reactivate latent HIV-1 while minimizing additional HIV-1 infection and side effects associated with T cell activation [29]. Therefore, the inclusion of cytosine methylation inhibitors in antiretroviral therapy could represent a significant step toward elimination of the latent HIV-1 reservoir and clearance of virus from infected patients.

Cell culture and drug treatment
Jurkat and J-Lat cells were cultured in RPMI (Invitrogen) with 5% FBS (Gemini Bio-Products) and 5% Fetalplex (Gemini Bio-Products). For analysis of virus reactivation by flow cytometry, aza-CdR (Sigma) and TNF-a (Biosource) treatments were for 24 h, after which medium was replaced. Reactivation was assayed after an additional 48 h. For ChIP and bisulfite-mediated methylcytosine mapping, cells were treated for 30 h with aza-CdR. For cycloheximide experiments, cells were treated for 24 hours, either with or without 40 ng/ml cycloheximide.
Plasmid and cDNA library generation 20 mg of pEGFP-N1 (Clontech) was methylated at CpGs with M. Sss I (New England Biolabs) according to the manufacturer's protocol. DNA was purified and subjected to a second round of methylation. To generate pBMN-CSI-T, the multiple cloning site (MCS) and GFP gene from pBMN-I-GFP (Addgene plasmid 1736) were replaced with the MCS from pDNR-LIB (Clontech) and the tomato fluorescent protein. Also, the human cytomegalovirus (hCMV) immediate early promoter was inserted upstream of the MCS. For production of RelA-expressing retrovirus, RelA was cloned from pCMV4(hind), kindly provided by W. Greene, into a version of pBMN-CSI-T lacking the hCMV promoter. MBD2  was cloned into pBMN-CSI-T as part of cDNA library generation. The cDNA library was generated using the Creator SMART cDNA Library Construction Kit (Clontech) with oligodT-purified (Quickprep mRNA Purification Kit, Amersham) RNA isolated (TRIzol, Invitrogen) from Jurkat T cells. Amplified cDNAs were cloned into pBMN-CSI-T and electroporated into E. coli strain DH5a. The library was amplified 240,000-fold by plating of bacteria on solid medium, and DNA was extracted from aliquots (Plasmid Maxi Kit, Qiagen).

Plasmid transfection, cell infection, and screening
J-Lat cells were transfected by electroporation using Kit R and program O-28 (Amaxa Biosystems). HIV-1 reactivation was assayed by flow cytometry four days post-transfection. HIV-1 R7/E 2 /GFP pseudotyped with the vesicular stomatitus virus G (VSV-G) protein was produced by cotransfecting 293T cells with pEV1335 and a plasmid encoding VSV-G by the calcium phosphate method. Supernatant was harvested 48 h post-transfection and frozen at 280uC in aliquots. Aliquots were thawed, diluted 1:160, and used to infect Jurkat T cells overnight at a multiplicity of 0.1 infectious units per cell with 2 ml supernatant per 1 million cells. Three days postinfection, GFP-negative and -positive cell populations were isolated by FACS. Retrovirus pseudotyped with VSV-G was produced as described previously [68] by cotransfection of Phoenix-ampho cells with pBMN-CSI-T or plasmids derived thereof and a plasmid encoding VSV-G. Supernatant was harvested 48 h post-transfection and J-Lat cells were infected overnight at a ratio of 250,000 cells to 2 ml supernatant with centrifugation at 2500 rpm for the first 1.5 h. 293T cells were transfected by the calcium phosphate method. Cells were cotransfected with a plasmid encoding the tomato fluorescent protein and either methylated or unmethylated pEGFP-N1, and the tomato-positive population was analyzed for GFP expression. For measurement of J-Lat activation, cells were infected with undiluted virus and analyzed by flow cytometry 2 days post-infection. For cDNA screening, cells were infected at a multiplicity of 0.15 infectious units per cell. GFP-positive cells were purified by FACS two days post-infection, cultured for two days, and genomic DNA was isolated (DNeasy Tissue Kit, Qiagen). The cDNA inserts were amplified from genomic DNA by PCR using oSK57 (59-AAATGGGCGGTAGGCGTGTACGGTG-39) and oSK58 (59-GCGGCTTCGGCCAGTAACGTTAGGG-39) as primers, cloned into pBMN-CSI-T, and identified by determination of nucleotide sequence (Molecular Cloning Laboratories).

Flow cytometry and FACS
Cell fluorescence was measured with the FACSCalibur or LSRII (BD Biosciences). Cell sorting was performed with the FACS Vantage DiVa (BD Biosciences). To phenotype CD4 + T cells, they were stained with the following mAbs: phycoerythrinconjugated (PE)-anti-CD4, TC-anti-CD45RA, or PE-anti-CXCR4 (Caltag). Flow cytometry and sorting data were analyzed with FlowJo software (Treestar) or Cellquest (BD Biosciences), in the case of primary cells. Analysis was restricted to the live population, as defined by the forward versus side scatter profile. Flowjo transforms fluorescence plots to a linear scale at the origin, permitting intelligible display of cells with low fluorescence.
To assess intracellular p24 gag expression, cells were fixed and permeabilized with Citofix/Cytoperm (BD Biosciences). Cells were washed with Perm/Wash buffer (BD Biosciences) and were stained with anti-p24 antibody (AG3.0). Cells were washed with Perm/Wash Buffer and incubated with Alexa Fluor 488 goat antimouse IgG (H+L) in 100 ml of Perm/Wash buffer. Cells were washed with Perm/Wash buffer and samples were analyzed by flow cytometry. HIV-1 p24 gag -positive gates were set by comparison with uninfected cells treated in parallel.

Statistical analyses
The effect of MBD2 1345-1947 upon GFP expression (Figure 2A) was evaluated by a two-tailed, two sample Student's t-test with a null hypothesis of no effect. Reactivation of latent HIV-1 ( Figure 5D) was evaluated with a one-tailed, two sample Student's t-test with a null hypothesis of no increase in GFP expression. In bisulfite-mediated methylcytosine mapping experiments, at least nine independent clones of sodium bisulfite-treated HIV-1 DNA were analyzed from each sample. For J-Lat cell lines in the latent state ( Figure 3B) and CD4 + T cells ( Figures 6D and E), a onetailed, single sample Student's t-test was performed for each CpG with a null hypothesis of no methylation. For J-Lat 6.3 treated with either aza-CdR or a PBS control ( Figure 4A), a one-tailed, twosample Student's t-test was performed for each CpG with a null hypothesis of no decrease in methylation after aza-CdR treatment. For sorted populations of GFP-negative and -positive Jurkat T cells ( Figure 5B), a one-tailed, two sample Student's t-test was performed for each CpG with the null hypothesis that the GFPpositive population did not have less methylation.

Chromatin immunoprecipitation and quantitative PCR
ChIP was performed as described previously [70] with modifications. J-Lat 6.3 cells Cells were diluted to 5610 5 per ml, lysed, and sonicated (Model 500 Ultrasonic Dismembranator, Fisher Scientific). Lysates were incubated overnight with 5 mg of antibody against MBD2 (Upstate Cell Signaling Solutions cat. 07-198), HDAC2 (Santa Cruz Biotechnology cat. sc-7899) or Sp1 (Santa Cruz Biotechnology cat. sc-59). Immune complexes were recovered by incubation for 1 h with protein A agarose beads (Invitrogen). Immunoprecipitated DNA was quantified by quantitative PCR using the 7900HT Sequence Detection System (Applied Biosystems) and 26 Hot Sybr real time PCR kit (Molecular Cloning Laboratories). Negative control DNA was assayed using 5USBACT and 3USBACT as primers, CpG island 1 was assayed using oSK92 (59-TCAGTTCAGATAATTT-CAGTTGTCC- 39) and oSK93 (59-CCCAGTACAGG-CAAAAAGCA-39) as primers, and CpG island 2 was assayed using oSK89 (59-AAGCGAAAGGGAAACCAGAG-39) and oSK90 (59-TCTCCCCCGCTTAATACTGA-39) as primers. SDS 2.3 software (Applied Biosystems) was used to analyze precipitated DNA relative to input and to confirm the specificity of each PCR reaction by melting curve analysis.

Differentiation, infection, and activation of CD4 + T cells ex vivo
PBMCs were obtained from leukopaks from unidentified, healthy donors. Naïve CD4 + T cells were isolated by MACS microbead negative sorting using the naïve T cell isolation kit (Miltenyi Biotec). The purity of the population was always higher than 95%. Naïve T cells were primed with beads coated with anti-CD3 and anti-CD28 (Dynal/Invitrogen) as previously described [44].
Seven days after stimulation, cells were infected by spinoculation. Seven days after infection, cells were reactivated with beads coated with anti-CD3 and anti-CD28 for 72 h in the presence of IL-2 at a ratio of 1 bead per cell. The integrase inhibitor 118-D-24 did not have any effect on viral reactivation.