Background

Acute HIV infection depletes mucosal CD4 T cells, mainly effector memory (TEM) cells, rapidly and profoundly [1–3]. The ensuing chronic phase is largely asymptomatic, even though mucosal tissues are not replenished with TEM cells [4]. Simian immunodeficiency virus (SIV) infection of rhesus macaques (an animal model of human HIV disease) shows that opportunistic control infection in the chronic phase is mediated by remnant mucosal TEM cells supplied by the differentiation of central memory (TCM) cells in lymph nodes [5, 6]. Additionally, human TCM cells are also characterized by their capacity of proliferation and differentiation into TEM cells [7, 8]. Thus, homeostasis of TCM cells is considered a key factor sustaining the asymptomatic stage of HIV infection, while progression to acquired immunodeficiency syndrome is attributed to homeostatic failure of TCM cells [5, 6, 9–12].

It is unclear how this homeostatic equilibrium is lost during chronic infection. CD4 T cell maturation subpopulations (TN, TCM, and TEM) [7] are differentially affected by HIV infection [13, 14]; with TEM cells being HIV’s main target [15]. TCM cells can be infected in a lower proportion by HIV, which has led to propose that direct virion-mediated cytopathicity could gradually eliminate them, leading to poor homeostatic activity [6]. Nevertheless, direct cytopathicity by HIV [16] cannot completely explain CD4 T cell depletion during chronic infection [17–20], which suggests the participation of indirect pathogenic mechanisms, particularly chronic activation [12, 21]. Additionally, CD4 T cells from patients with HIV could be intrinsically altered, as suggested by the limited proportion of HIV-infected patients recovering their pre-infection CD4 T cells counts under virus-controlling antiretroviral therapy [22]. In this regard, we have found intrinsic dysfunctions in activated TCM cells from HIV-infected patients, as a lowered IL-2 response and CD40L induction after T cell receptor (TCR)-mediated stimulation [23, 24], which could decrease their proliferative, differentiation, and survival capacities.

In order to determine if circulating TCM cells from HIV-infected patients have a transcriptome consistent with activation, but simultaneously with altered capacities to divide and survive, we compared the ex-vivo messenger Ribonucleic acid (mRNA) whole-genome expression patterns of CD4 T naive (TN) and TCM cells from HIV+ patients with TN, TCM, and TEM cells from healthy controls. We found a TCM cell signature in HIV-1 infection suggesting that the loss of this subpopulation may be driven by increased cell cycle entry followed by mitotic arrest possibly leading to cell death in a non-senescent or effector-like state.

Methods

Participants

This study was approved by the boards of Instituto Nacional de Enfermedades Respiratorias Ismael Cosío Villegas (reference number B29-11), and Instituto Nacional de Ciencias Médicas y Nutrición Salvador Zubirán (reference number 1403). All patients signed written informed consent according with the Helsinki Protocol. Blood samples were obtained from 9 HIV¯ controls, and 6 HIV+ patients. Patients had median 480 CD4 T cells/μL blood (range 330–757), and median 121 563 HIV-ribonucleic acid (RNA) copies/mL-blood (23 883–41 2584). Among them, patients providing TCM cells had viral loads of 23 883, 81 834 and 107 732 HIV RNA copies/mL-blood, and CD4 T cell counts of 439, 473 and 491 CD4 T cells/μL blood, respectively. Relative telomere length was determined in samples from ten additional HIV¯ controls, and ten additional HIV+patients with median 628 CD4 T cells/ μL-blood (194–1 128) and median 485 882 HIV-RNA copies/mL-blood (3 870–3 500 000). Patients were antiretroviral therapy-naive, free of opportunistic infections and malignancies, and were not taking any immunomodulatory drugs.

Isolation of CD4 T cell subpopulations

Peripheral blood mononuclear cells (PBMCs) were purified from 50 to 60 mL of peripheral blood by sedimentation on Lymphoprep (Fresenius Kabi Norge, Oslo, Norway). CD4 TN (CD45RA+ CCR7+), TCM (CD45RA¯ CCR7+) and TEM (CD45RA¯ CCR7¯) cells were purified from PBMCs using immunomagnetic beads (Miltenyi Biotec, Bergisch Gladbach, Germany).

Subpopulation purity was determined according to the expression of CD4, CD45RA and CCR7, using anti-CD4-APC-Cy7, anti-CD45RA-APC (BD Biosciences, San José, CA, USA), and anti-CCR7-PE (Miltenyi Biotec) fluorochrome-conjugated antibodies (See Additional file 1). Cells were analyzed in a FACSCanto II flow cytometer (BD Biosciences). Cells with purity >90% were used. Membrane CD38 was detected with an anti-CD38-biotin (Miltenyi Biotec) plus streptavidin PerCp-Cy5.5 (Biolegend, San Diego, CA, USA).

RNA extraction and microarray analysis

Total RNA was obtained from three TN, three TCM, and three TEM CD4 T cell samples from healthy controls, and three TN and three TCM CD4 T cell samples from HIV+ patients, using RNeasy Mini Kit (Qiagen, Venlo, Netherlands). Each RNA sample proceeded from a different subject. Scarcity of patients’ TEM cells precluded obtaining sufficient RNA.

Microarray gene expression analysis used equimolar concentrations of total RNA from T cell subpopulations. Complementary deoxyribonucleic acid (cDNA) synthesis, amplification, and gene expression profiling were performed according to the manufacturer’s instructions (Affymetrix WT Sense Target labeling assay manual, California, USA). Labeled DNA was added to hybridization cocktail and injected into the array (GeneChip Human Gene 1.0 ST Array, Affymetrix). Washing and staining steps were performed in the GeneChip Fluidics Station 450 (Affymetrix). Probe arrays were scanned using a GeneChip Scanner 3000 7G (Affymetrix). Data were deposited in GEO, series record GSE73968.

Background correction and normalization were performed with Robust Multiarray Average Method (RMA) [25] using Bioconductor package [26] of R [27]. A Principal component analysis (PCA) of normalized signals from all genes in each microarray was performed using R [27].

Modeling gene expression was performed using linear models of Limma package [28]. The B-statistic was used as significant measure to define differentially expressed genes. This statistic is computed as the posterior odds of differential expression. It is reformulated in terms of a moderated t-statistic in which posterior residual standard deviations are used in place of ordinary standard deviations. Essentially, the B-statistic compromises between individual gene variance estimates and a single variance estimate for all genes. The probabilities are transformed to a scale that goes from –Inf to Inf using log odds. The B-statistic is analogous to the adjusted p-value, which addresses statistical significance for multiple comparisons. Here, genes with Log 2 Fold Change (FC) ≥ |0.5| and Log odds > 0 were considered as differentially expressed. Limma statistics such as adjusted p-value and the B statistic can be seen in Additional file 2. FDR Benjamini Hochberg multiple testing correction [29] was applied to control the number of false positives. Both B statistic and adjusted p-value showed consistency across differentially expressed genes. Unsupervised 2-way hierarchical clustering analysis of gene expression data was performed using Euclidian distance and average linkage with gplots [30] of R [27]. Venn diagrams were made with Venny 2.0.2 [31].

Functional enrichment analyses were performed with Data Base for Annotation, Visualization and Integrated Discovery (DAVID) [32, 33], Gen Set Enrichment Analysis (GSEA) [34] and Ingenuity Pathway Analysis (IPA, QIAGEN Redwood City, CA, USA). DAVID uses a Fisher Exact test in order to determine gene-enrichment in annotation terms. A gene set is enriched when the proportion of genes in a list that falls into an annotation term differs from the background model. The EASE score is a modified Fisher exact p-value. Basically, if n is the number of genes in the list that falls into a given annotation term, n-1 is used to compute the p-value [32, 33]. Gene set enrichment methods also implement strategies for addressing the issue of multiple testing hypotheses. GSEA uses a ranking procedure to produce a gene list from the full expression matrix. This is done by computing an Enrichment Score (ES(S)). It controls the ratio of false positives to the total number of gene sets attaining a fixed level of significance using FDR [34]. IPA assesses enrichment (i. e. biological functions that could be increased or decreased given the observed gene expression patterns) using a Fisher exact p-value. Additionally, it computes a Z score that allows inferring upstream transcriptional regulators and expectable enriched functions, based on statistical significance by comparing the match between observed and predicted up/down regulation patterns. The null model is referred as activation Z-score [35]. Predicted regulation patterns are based on previously reported causal relationships between relevant genes and functions [35].

Semi-quantitative real-time PCR

We used B2M, GAPDH, POLR2A, and TBP as reference genes to normalize expression. RNA proceeded from the samples used for microarray analysis. cDNA was synthesized from ~100 ng total RNA with Transcriptor First Strand cDNA Synthesis Kit (Roche Applied Science, Mannheim, Germany), using random hexamers and performing one cycle of 10 min 25 °C; 30 min 55 °C, and 5 min 85 °C. cDNA was stored at −20 °C until use. PCR amplifications were performed by high-throughput gene expression analysis using DNA binding dye Evagreen (SsoFast MasterMix, Biorad, California, USA) for product detection, and specific primers for each gene (DELTAgene Assays, Fluidigm Corporation, California, USA). Specific target pre-amplification of each cDNA and a cleanup step were performed as described elsewhere [36].

We performed semiquantitative RT-PCR using Fast Gene Expression Analysis with EvaGreen (Biorad), following the Biomark System Protocol (Fluidigm Corporation, California, USA). Assay mixes (100 μM of each pair of primers, 2X Assay Loading Reagent, and 1X TE buffer), sample mixes (pre-amplified cDNA, 2X SsoFast MasterMix (BioRad), and 20X DNA Binding Dye Sample Loading Reagent (Fluidigm), were loaded into a 96.96 Dynamic Array (Fluidigm), using the IFC Controller HX (Fluidigm), and were then transferred to a BioMark HD device (Fluidigm) for the PCR cycles (40 min 70 °C, 30s 60 °C; 60s 95 °C, then 30 cycles of 5 s 96 °C, 60s 60 °C). Melting curves were determined at the 60 to 95 °C rise, with a temperature change rate of 1 °C/3 s. Ct values were obtained with Fluidigm Real-Time PCR Analysis Version 4.1.3 software (Fluidigm).

Only Ct values <30 and amplicons with only 1 melting curve were used. Geometric means of four reference genes were used to normalize expression data [37]. Relative expression was calculated as ΔΔCt. Expression of each gene was determined with six technical replicates per sample. Normality was verified using Kolmogorov-Smirnov test, which quantifies the distance between the empirical distribution of the sample and the cumulative distribution of the reference distribution, which in this case is assumed to be normal. Group differences were analyzed with Student’s t test. Data management and statistics were done with Reshape [38] and fBasics [39] packages of R [27].

Relative telomere length

Telomere PNA kit/FITC (Dako, São Paulo, Brazil) was used following the manufacturer’s instructions, including thymocytes from 6-week old mice as reference for normalization. Briefly, samples were prepared by mixing 106 mouse thymocytes and 106 TCM cells. The mixture was distributed into four tubes. 150 μl of FITC-labeled peptide nucleic acid (PNA) probe solution was added into two tubes while 150 μl of unlabeled PNA probe solution was added into the other two. Samples were hybridized in a pre-warmed heating block (TB2 Thermoblock, Biometra, Göttingen, Germany) set at 82 °C, 10 min, and left overnight at room temperature. Samples were washed twice. Between washing steps, samples were heated to 40 °C in a pre-warmed TB2 Thermoblock (Biometra) for 10 min. Samples were resuspended in 250 μL of DNA staining solution (1X), and stored overnight at 4 °C, away of light. Then, samples were analyzed by flow cytometry in a FACSCanto II (BD Biosciences).

Results

TCM cells from HIV+ patients are not more differentiated and are not more senescent

Unsupervised principal component analysis of normalized whole genome expression data segregated samples of each maturation subpopulation, and further separated samples originating from persons with HIV and samples from controls (Fig. 1a). Thus, phenotype-based classification of differentiation subpopulations [7, 8] reliably reflected distinct gene expression programs, as previously reported [40–42], which were altered by HIV infection.

Fig. 1
figure 1

Gene expression does not support greater differentiation or senescence of TCM cells from HIV+ patients. a Principal component analysis of the entire microarray data set of each subpopulations from HIV+ patients (red triangles and red circles) and controls (blue triangles, blue circles and blue squares). The first three principal components are shown, accounting for 55% of variance in a three dimensional plot. b Heat map resulting from hierarchical clustering of genes related with normal differentiation (pairwise comparisons between not infected subpopulations). Each row represents a differentially expressed gene. Each column represents each independent sample. The unsupervised two-way hierarchical clustering is shown as a dendrogram for genes (left), and a dendrogram for samples (top). In the upper dendrogram (samples) the independent resulting nodes, each one corresponding to a maturation subpopulation, is encircled in green. c Sequential downregulation of selected naïve-associated genes, and sequential upregulation of selected effector-associated genes when samples are arranged according to the linear differentiation model. Data are represented as means ± 1 SEM of three donors (blue) and three patients (red). TBX21 expression difference between TCM and TCM HIV was analyzed with Student’s t test. d Relative telomere length of central memory CD4 T cells (TCM) from HIV+ patients (red triangles) and controls (blue circles), Student’s t-test was used to compare groups. We were unable to obtain sufficient RNA from TEM cells from patients due to their small number

Full size image

We asked if differential gene expression by patients’ and controls’ TCM cells reflected greater differentiation of patients’ cells (towards effector stages) [23, 24]. Using the criteria defined in methods (Log2FC ≥ |0.5| and Log (odds) > 0) we looked in the whole transcriptome for all differentially expressed genes in the following pair-wise comparisons of CD4 T cell subpopulations from controls: TCM vs. TN, TEM vs. TCM, and TEM vs. TN (arrows a, b and c in Fig. 2a). The resulting 1858 differentially expressed genes are subsequently referred to as differentiation-related genes (corresponding to subpopulations in distinct stages of differentiation). We performed an unsupervised 2-way hierarchical clustering analysis of these 1858 differentiation-related genes (Fig. 1b, and Additional file 3). TN and TCM cells from patients grouped with their counterparts from controls (Fig. 1b). Samples of a same subpopulation were assigned to a same node (green circles 1, 2, and 3 on Fig. 1b), regardless of their HIV status. The expression of the differentiation-related genes progressively decreased or increased in the order of linear differentiation (TN → TCM → TEM ), agreeing with previous reports [40–43] (Fig. 1c). For instance, LEF1, ACTN1, FOXP1, IL6ST and CERS6 reportedly undergoing down-regulation in naive T cells after antigen recognition and differentiation [44–48], along with TAF4B, appeared progressively down regulated when samples were ordered according to the linear model of peripheral differentiation (Fig. 1c, left panel). These changes agree with previous reports [42]. Conversely, differentiation and effector function-associated transcripts, like EOMES, TBX21 (t-bet), PRDM1 (Blimp-1) [49, 50], GZMA and PRF1 [51, 52], were gradually increased in the same order (Fig. 1c right panel). A same pattern was followed by the expression of KLRG1,an indicator of replicative senescence [53, 54] (Fig. 1c). TBX21 (t-bet) was the only gene with increased expression in TCM cells from patients, compared with controls (p = 0.003), which, along with the increased expression of IL12R e IL18R, suggests a Th1-skewed response driven by HIV infection. A Th1-skewed response was also predicted by Ingenuity Canonical Pathway analysis (See Additional file 4). Thus, TCM cells from patients did not seem to be more differentiated than their counterparts from controls, but appeared polarized to Th1 functions.

Fig. 2
figure 2

Unique TCM cell signature in HIV infection. Differential expression was defined as Log2 of fold change (Log FC) ≥ |0.5|, and Log (odds) > 0. a Pairwise comparisons of samples of CD4 T cells subpopulations from HIV+ and HIV¯ groups indicated by arrows a and d. Number of genes differentially expressed in each comparison are shown. Blue circles, controls’ samples; red squares, HIV+ patients’ samples. b Venn diagram of sets of differentially expressed genes. Each pairwise comparison is depicted by a colored oval. The number of differentially expressed genes found in more than one comparison appear in the intersections. c Heat map displaying a two-way unsupervised hierarchical clustering of 210 differentially expressed gens distinguishing HIV+ patients’ TCM cells (red bar) and controls’ TCM cells (blue bar), grouped in dendrograms. Each column represents an independent sample (biological replica) of each subpopulation, numbered 1 to 3. Each row corresponds to a differentially expressed gene

Full size image

We then asked if patients’ TCM cells had a longer replicative history, which would entail a shortening of telomeres. We did not find any difference in relative telomere length between TCM cells from patients and controls (p = 0.737, Fig. 1d), agreeing with KLRG1 expression [53, 54], and suggesting that they are not in a more senescent state.

TCM gene expression signature in HIV infection

Having ruled out a greater differentiation of patients’ TCM cells, we investigated if the gene expression signature of these cells revealed a functional state that could explain loss of homeostatic capacity. Using the criteria defined in methods (Log2FC ≥ |0.5| and Log (odds) > 0), we looked for genes that were differentially expressed by TCM cells from HIV+ patient and TCM from controls. We found a total of 210 differentially expressed genes. We refer to this 210- gene list as the gene expression signature of TCM cells in HIV infection (See Fig. 2a arrow d, b red oval and Additional file 2). This gene expression signature was obtained from the transcriptome independently of the list of 1858 differentiation-related genes. Among these 210 differentially expressed genes, 137 were absent in all other pairwise comparison (Fig. 2a, b). Hierarchical clustering analysis showed clear and consistent differences in the relative expression of these 210 genes between patients and controls (Fig. 2c). Of note, biological replicates were very homogeneous.

We analyzed the HIV TCM signature with the enrichment analysis tools IPA, GSEA, and DAVID. These different analyses consistently yielded four general functional categories that were modified in TCM cells from patients: cell cycle, DNA damage and repair, apoptosis, and immune responses (Table 1). Notably, the 137 genes uniquely distinguishing TCM cells from patients and controls (Fig. 2b red oval) sufficed to yield the same four functional categories when analyzed with DAVID and IPA. This suggests that the enriched functions largely depend on the TCM cell signature. GSEA rendered a larger set of altered immune functions, likely because it uses data from the entire microarray, and because it detects more modest changes when the members of a function or pathway show a strong correlation [34] (Table 1). GSEA identified Toll-like receptors (TLR), type I interferons, IL1, and NLRs signaling, plus NFκB activation, all of them related to an inflammatory milieu. Independently, Ingenuity Upstream Regulator Analysis [35] assigned the greatest z-scores and the most significant p-values to the activity of IL-1B, TNF, NFκB complex, and CCL5, as possible upstream molecules eliciting the expression changes constituting the TCM signature (See Additional file 5, upstream analysis). Cell cycle, DNA damage and repair, and apoptosis (greatly related functions [55, 56]) appeared consistently in the output of all enrichment analysis tools (Table 1). There were several functional categories closely related with G0/G1/S transition and G2/M checkpoints in the output of GSEA analysis. IPA, which weighs its predictions, displayed increased proliferation and cell survival, decreased apoptosis and decreased cell death. In contrast, IPA’s output simultaneously indicated an increase in cytostasis, movement to interphase, and a decrease in mitosis (Table 1).

Table 1 Enriched categories of functions according to TCM gene expression signature in HIV infection
Full size table

We re-analyzed mRNA expression by RT-PCR of 91 genes of the HIV TCM signature that were associated with enriched functions, and B2M, GAPDH, POLR2A, and TBP as reference genes. This analysis validated 75 genes (82%) of the signature (See Additional file 6). Five genes failed amplification, and 11 were not differentially expressed when assessed by RT-PCR. The expression of reference genes did not differ between samples (See Additional file 7). An analysis with IPA using only the 75 validated genes yielded the same enriched functions and pathways as microarray data (Fig. 3a, b, c, d, e, and Additional file 8). While some genes were related with more than one function, many were related exclusively with one function (Fig. 3f), supporting an unambiguous prediction. Finally, analyzing the TCM cells used for microarray and RT-PCR analyses, we found that surface expression of the CD38 protein was more frequent among TCM cells from patients than among those from controls, consistent with mRNA results (Fig. 3g).

Fig. 3
figure 3

Cell cycle impairments are predicted using RT-PCR-validated genes from the signature. Each graph (a to e) represents the group of validated genes by RT-PCR associated with an increased enriched function (red title and positive z-score) or decreased enriched function (blue title and negative z-core). f Venn diagram depicting the differentially expressed genes within sets corresponding to three IPA predictions. Intersections correspond to genes appearing in more than one prediction. g Frequency of surface expression of CD38 on TCM cells from two groups, corresponding to gene expression results

Full size image

A model of TCM cell death in HIV infection

Since the predictions of increased proliferation and increased cytostasis were incompatible, and the prediction of reduced apoptosis did not agree with previous evidence [13, 57–60], we took into account that enrichment tools base their predictions on a broad set of previous findings, ranging from very particular to very general ones. Accordingly, we investigated if the predictions were based on more demarcated processes, and if these processes were compatible. With this purpose, we reviewed the references supporting IPA predictions, allocating the indicated genes to the cell cycle phase that they regulated. We found that genes in all the predictions could be assigned to particular phases in the cell cycle, and implied no conflicts (Fig. 4), with the exception of some proliferation-predicting genes, which did not relate with cell cycle in the supporting evidence.

Fig. 4
figure 4

Allocation of HIV-related TCM signature genes in the cell cycle phases they regulate. These genes suggest a promoted progress from G0 to S, followed by arrest in G2/M cell cycle phases, and decreased apoptosis

Full size image

The HIV TCM signature is compatible with an enhanced progression from G0 to S phase (CCNE1 [61, 62], MKI67 [63, 64], IL12RB2 [65, 66], ADAM9 [67], E2F8 [68], MXD1 [69], MCM10 [70], BCL6 [71–73]), but not with progression to later stages. Simultaneously, the expression patterns of other genes suggest arrest in G2/M (CHK1 [74], GADD45B [75], PER1 [76], NFKBIA [77], RBBP8 [78], KIF11 [79]). For instance, accumulation of CCNE1 is necessary for G1 → S transition, but its overexpression is associated with chromosome destabilization and DNA damage [61, 62]. In turn, CHK1 expression is required for cycle arrest in G2 following DNA damage [74]. These observations suggest that the cell processes expectable from the HIV TCM signature can be integrated within the cell cycle, and they indicate increased cycling up to S phase, followed by arrest in G2/M. Remarkably, since cycle arrest leads to cell death [55, 56], it was unexpected to find that the expression patterns of CSTA [80], RNASEL [81] NR4A2 [82] and NFKBIA [77] predicted an inhibition of caspase-3 mediated apoptosis (Fig. 4).

Discussion

We identified and validated a gene expression signature of TCM cells in the context of chronic HIV infection that led us to a model of TCM cell death in HIV infection, also supported by additional observations and previous reports.

It could be argued that a model based on mRNA presence, but not demonstrating the presence of encoded proteins [83], is unfounded. In this regard, recent studies and reassessments found that mRNA changes explained 87 to 92% of corresponding protein changes [84, 85], as found in mouse dendritic cells responding to lipopolysaccharide [86]. Further, 81% of protein levels were explained by mRNA levels using a large scale proteome and mRNA analysis in NIH3T3 cells [87]. Therefore, we consider that we have evidence to propose a model of TCM cell death in HIV infection.

Our interest in intrinsic TCM cell alterations in patients with HIV originated from studies showing their importance in CD4 T cell homeostasis under HIV infection [10, 12, 57, 88–90]. Our previous studies on activated (CD38+) TCM cells, particularly those from HIV-infected patients, showed IFN-γ-skewed cytokine responses that were un-connected to CD40L induction, along with a lowered IL-2 production [23, 24]. Given this functionality, TCM cells seemed differentiated towards an effector fate [40, 42, 43, 91, 92]. However, in the present study we found that TCM cells’ gene expression profile was incompatible with the TEM maturation category. Additionally, we found no decrease in relative telomere length (RTL) of patients’ TCM cells, which is expectable of cells with a longer replicative history [7, 8], such as TEM- cells. Moreover, KLRG1 expression, which is proportional to replicative history [53], was similar in TCM cells from patients and controls, but greatly increased in more differentiated TEM cells. These findings support that HIV infection was not associated with an enhanced TCM cell differentiation. Nevertheless, TCM patients’ cells showed an increased expression of the Th1-associated transcription factor T-bet [49, 93], suggesting that our previous findings were attributable to Th1 polarization rather than differentiation.

The homogeneity and consistency of the TCM signature in HIV infection (Fig. 2c) contrasts with the great differences between patients’ viral loads (23883, 81834 and 107732 HIV RNA copies/mL-blood). This may seem more important considering that even transient changes in viral load can greatly influence gene expression in total CD4 T cells [94]. However, patients’ CD4 T cell counts, the strongest predictor of subsequent disease progression [95] are close (439, 473 and 491 CD4 T cells/mm3 blood, respectively), and lie above the threshold for the occurrence of most opportunistic infections (200 cells/mm3 blood [95]). This could suggest that the TCM gene expression signature is not dependent of the magnitude of viral replication during the chronic phase of infection, but it could rather be related with irreversible events from the initial phase of infection and/or with the magnitude of circulating CD4 T cell loss in the chronic phase.

A thorough pathway enrichment analysis of the HIV TCM signature suggested enhanced cell cycle entry and proliferation. Under closer inspection, however, we observed that proliferation-predicting genes corresponded to functions upstream of S phase. Conversely, this analysis also predicted cell cycle arrest due to functions occurring in G2 or M. If cell cycle promotion and arrest occur in a same TCM cell, cycling would not imply cell division [58], since G2/M arrest would lead to death [96]. Additionally, overexpression of DNA damage and repair-related genes by patients’ TCM cells are consistent with a failed division after S phase.

Our model could integrate partial observations from previous studies on cell cycle in HIV infection, providing a wider view of the fate of TCM cells. In a previous study, CD4 T cells from patients with HIV that were ex vivo in S phase (mostly TCM) would die after in vitro stimulation more frequently than CD8 T cells [58]. However, patients’ cells were not compared with CD4 T cells from HIV¯ controls. Contrastingly, our comparison with cells from controls, and our characterization of CD4 T cell subpopulations, suggested that cycle-related death was due to HIV infection, and involved predominantly TCM cells.

Another group reported arrest in G1 based on the accumulation of cycling cells in G1 among total circulating CD4 T cells [97]. However, since TCM cells comprise only about 25% of circulating CD4 T cells [42], they were possibly not well represented in that study. Nevertheless, an increased proportion of Ki67+ cells in the G1 phase provided evidence that CD4 T cells more frequently entered cell cycle in HIV infection, as previously demonstrated [98, 99]. Since dead cells are readily removed from blood [100], it is possible that the reported ex vivo increase of CD4 T cells in G1 was the result of rapid removal from blood of cells that died in a further phase. Our findings were consistent with a promoted entry into cell cycle (Fig. 4), and notably, they suggested that arrest occurred in the later phases G2 and/or M.

Our results could imply that incorporation of nucleotide analogues by TCM cells from patients with HIV could reflect entrance to a fatal cycle, rather than proliferation [101, 102]. Evidence of division and viability will be required to dismiss this possibility. In our model, TCM cell turnover reflects to some extent, death, and not actual proliferation. Accordingly, we found no difference in telomere length between TCM cells from controls and patients.

An analogous implication pertains to apoptosis. Previous studies have reported increased apoptosis in total [58, 103, 104] and in TCM [58, 105] CD4 T cells from patients with HIV, which is an explanation of CD4 T cell loss in chronic infection [106]. These and other studies [13, 57–60] inferred apoptosis by demonstrating Annexin V binding to viable cells; however, it has been demonstrated that Annexin V binding can be increased in other cell death pathways [107–110]. In this regard, we found that HIV infection altered the expression of a considerable number of genes that indicated that apoptosis would not be favored. A possible explanation of this discordance with previous studies could be that different cell death pathways may coexist with caspase 3-mediated TCM cell death in HIV infection, as suggested by other studies [107, 111]. Also, TCM subpopulation is heterogeneous [10, 112], and cells under apoptosis might not be reflected in microarray data. Therefore, we propose that an additional programmed cell death pathway may be in involved in TCM cell death after cell cycle arrest. A likely pathway is pyroptosis, an inflammatory programmed cell death pathway driven by pro-inflammatory signals, such as bacterial lipopolysaccharide and IL-1β [113]. This pathway has been suggested by previous reports [111, 114], and agrees with the presence in blood of triggers of pyroptosis as bacterial lipopolysaccharide (LPS) [115] and IL-1β [116] in HIV infection. Additionally, increased concentrations of IL-18, a cytokine liberated during pyroptosis, has been found in the blood of patients with HIV [117].

Conclusions

In summary, we propose a model of CD4 TCM cell death in chronic HIV infection based on a gene expression signature unique to this subpopulation. According to it, CD4 TCM cell loss in HIV infection may be driven in vivo by increased cell cycle entry followed by G2/M arrest, possibly leading to a non-apoptotic cell death, arguably pyroptosis. Further experimental work is required to validate the processes involved in this model.