Memory T cells are heterogeneous in terms of their phenotype and functional properties. We investigated the molecular profiles of human CD8 naive central memory (TCM), effector memory (TEM), and effector memory RA (TEMRA) T cells using gene expression microarrays and phospho-protein-specific intracellular flow cytometry. We demonstrate that TCM have a gene expression and cytokine signaling signature that lies between that of naive and TEM or TEMRA cells, whereas TEM and TEMRA are closely related. Our data define the molecular basis for the different functional properties of central and effector memory subsets. We show that TEM and TEMRA cells strongly express genes with known importance in CD8 T cell effector function. In contrast, TCM are characterized by high basal and cytokine-induced STAT5 phosphorylation, reflecting their capacity for self-renewal. Altogether, our results distinguish TCM and TEM/TEMRA at the molecular level and are consistent with the concept that TCM represent memory stem cells.
Immunological memory is a fundamental feature of the adaptive immune system. It enables the immune system to respond more rapidly and vigorously to infectious pathogens that have been encountered previously. In particular, memory CD8 T cells play a major role in host defense by rapid recognition and lysis of virus-infected cells. A memory response differs both quantitatively and qualitatively from a primary response (1, 2, 3). Thus, compared with a naive population, the precursor frequency of Ag-specific memory cells is increased and, furthermore, these cells have an enhanced capacity to respond to Ag. Despite recent progress, a clear understanding of the molecular and cellular basis of T cell memory is still lacking.
Within human CD8 T cells, van Lier and colleagues (4) first demonstrated the presence of phenotypically and functionally distinct subsets of primed T cells by analyzing expression of CD27 and CD45RA. Although naive CD8 T cells express both of these cell surface glycoproteins, cells expressing CD27 but not CD45RA were reported to have functional properties suggestive of “memory” cells and those that expressed CD45RA but not CD27 had functional properties suggestive of “effector” cells (4).
Subsequently, Lanzavecchia and coworkers (5, 6) used expression of CCR7 and CD45RA to define subsets of CD8 T cells. According to this scheme, naive T cells (TN)3 (3) express both CCR7 and CD45RA whereas primed CD8 T cells can be considered as belonging to one of three different subsets. Two of these lack expression of CD45RA and thus lie broadly within the van Lier memory subset. Of these, central memory cells (TCM) express CCR7 while effector memory (TEM) cells lack expression of CCR7. In humans, but not in mice, there is a third T cell memory subset, TEMRA, that includes cells that express CD45RA but lack expression of CCR7.
TCM and TEM can be distinguished by their different homing and effector capacities (6). Like naive cells, TCM express CD62 ligand (CD62L) and CCR7 and home to secondary lymphoid organs. In contrast, expression of a different set of chemokine receptors (e.g., CXCR3) allows TEM and TEMRA to gain access to inflamed peripheral tissues. Human TEM, and TEMRA in particular, are more differentiated in terms of effector function than TCM (4, 7, 8, 9, 10, 11). They display potent ex vivo cytotoxicity and produce Th1 cytokines upon stimulation, whereas TCM mainly produce IL-2 and Th2 cytokines. Further studies have shown that TCM have a higher proliferative potential and greater resistance to apoptosis, whereas TEM/TEMRA have a skewed TCR repertoire and are characterized by a “senescent” replicative history (9, 10, 12, 13, 14).
The importance of both TCM and TEM subsets for the control of infectious diseases and the effectiveness of vaccines has been shown in several murine studies (15, 16, 17). In mice, Ahmed and colleagues (18) have demonstrated a linear differentiation pathway TN→ effector → TEM → TCM following acute lymphocytic choriomeningitis virus (LCMV) infection. Two other models of CD8 memory T cell differentiation have been proposed: the signal strength/progressive differentiation model by Lanzavecchia and coworkers (6) and the “independent” differentiation model by Pannetier and colleagues (19). However, in humans the developmental relationship among TCM, TEM, and TEMRA is still controversial.
In this study, we have conducted gene expression and kinase phospho-protein profiling of CD8 memory subsets to investigate their molecular programs and their cytokine responsiveness and to gain insight into their relationship at the molecular level. We elected to define memory subsets according to the Lanzavecchia model in view of the capacity of the model to distinguish between two apparently functionally distinct subpopulations of CD45RA−-primed T cells (TCM and TEM) and the extensive debate in the literature concerning the lineage relationship between TCM and TEM or TEMRA cells. Our results define a molecular basis for the different functional properties of human CD8 T cell memory subsets and place TCM cells between TN and TEM/TEMRA cells in terms of their molecular signatures.
Materials and Methods
Isolation of CD8 T cell subsets
PBMC were obtained from buffy coat preparations from four healthy donors by density gradient centrifugation using LymphoPrep (Nycomed) in accordance with institutional ethics approval. We isolated CD8+ T cells by positive immunomagnetic selection using dynabeads (Dynal) with detachment of the anti-CD8 mAb. Purity of the selected CD3+CD8+ cells was >98% as assessed by flow cytometry. The CD8+N (CCR7+CD45RA+), TCM (CCR7+CD45RA−), TEM (CCR7−CD45RA−), and TEMRA (CCR7−CD45RA+) populations on a MoFlow Cytometer (DakoCytomation). Purity of isolated subpopulations was 93–98%. Cell purification procedures were conducted at 4°C to minimize in vitro-induced changes in gene expression.
Preparation of cRNA and array processing
Total RNA was extracted from purified CD8 T cell populations using TRI reagent (Sigma-Aldrich) followed by RNA cleanup with the RNeasy kit (Qiagen). We confirmed integrity of the total RNA by Lab-on-a-Chip 2100 Bioanalyzer (Agilent) quality control. Double-stranded cDNA was synthesized using a modification of the SMART-PCR protocol described by Petalidis et al. (20). This protocol has been validated regarding the fidelity of amplification and compares favorably with the direct labeling approach in terms of sensitivity, speed and cost-effectiveness. Briefly, 300–600 ng of total RNA was reverse transcribed using a modified SMART CDS Primer IIA containing a 5′ T7 promoter sequence. After double-stranded DNA synthesis, the cDNA was subjected to 15 rounds of PCR according to the manufacturer’s recommendations. Biotin-labeled cRNA was generated from 2 μg of double-stranded cDNA by one round of in vitro transcription with the BioArray High Yield RNA Transcript Labeling kit (Enzo). cRNA yields were >50 μg with A260:A280 GAPDH ratios ranged from 0.8 to 1.2. We performed replicate microarray experiments with RNA from four independent donors. R2 values derived from scatter plots of signal intensity values were 0.98 or greater for individual replicates of CD8 T cell populations. All data have been deposited in the European Bioinformatics Institute ArrayExpress public database. Accession number: E-TABM-40.
Microarray data analysis
We used the comprehensive software package BRB-ArrayTools (21, 22) for data analysis. First, the robust multiarray average algorithm was applied for probe-level normalization, background correction, and to calculate log expression measures (23). Then, to minimize the negative effects of random noise, we performed the following filtering steps: First, transcripts showing minimal variation across the set of arrays (log intensity variation, p > 0.01 with the test hypothesis that gene i has the same variance as the median variance) were removed. Second, transcripts with mostly unreliable expression (GCOS present call ≤4 in 16 samples) were also excluded. This resulted in 10,854 probe IDs that were used for further unsupervised analysis to examine the relationship among samples. First, we used standard average-linkage hierarchical clustering to cluster the samples using the Pearson correlation as the distance metric. Genes were median centered before clustering. The so-called R (reproducibility) measure described in Ref.24 was used to evaluate the robustness of the clusters. The R measure is based on perturbing the expression data with Gaussian noise, reclustering, and measuring the similarity of the new clusters to the original clusters. R measure = 1 indicates perfect cluster reproducibility. Second, multidimensional scaling (MDS) analysis was conducted using the Pearson correlation to measure between sample distances. MDS is related to principal components analysis and allows the representation of high-dimensional data in three-dimensional space. The Pearson correlation subtracted from unity was used to measure the distance between two CD8 T cell subsets by calculating the average linkage distance (distance range, 0–2) (25). The average linkage distance represents the mean of all pairwise distances (linkages) between samples from the two CD8 subsets concerned.
We identified genes that were differentially expressed among the four CD8 subsets using a random variance F test. The random variance F test permits sharing information among genes about within-class variation without assuming that all genes have the same variance (26). It is therefore well suited to the analysis of microarray data with a relatively small number of replicates. Genes were considered statistically significant if their p < 0.001. We used a multivariate permutation test (21) to provide 95% confidence that the false discovery rate (FDR) was <1% (1000 random permutations). The multivariate permutation test is nonparametric and does not require the assumption of Gaussian distributions. Using the significance of microarray analysis algorithm (27) produced similar results. We also performed a global test of whether the expression profiles differed between the different CD8 populations by permuting the labels of which arrays corresponded to which population. For each permutation, the p values were recomputed and the number of genes significant at the 0.001 level was noted. The proportion of the permutations that gave at least as many significant genes as with the actual data was the significance level of the global test.
Gene ontology (GO) overrepresentation analysis
Onto-Express was used to translate gene lists into functional profiles and to identify overrepresented GO annotation categories (28). Multiples (probes corresponding to the same gene) were removed before analysis. Enrichment of GO categories and associated p values were calculated based on hypergeometric distribution statistics. We obtained similar results when applying a χ2 test. Multiple testing correction was conducted with the Bonferroni step-down (Holm) procedure (α = 0.05). The Holm procedure is based on the family-wise error rate and a conservative global measure of type I error.
Analysis of cytokine signaling by phospho-specific intracellular FACS
29). Briefly, PBMC were fixed in 1.5% formaldehyde for 10 min at room temperature and permeabilized in methanol for 15 min on ice. This was followed by a rehydration step before staining the cells for 1 h at room temperature. We used CD62L-FITC/-PE, CD45RA-allophycocyanin, and CD8-PerCP Abs for surface staining and phospho-specific Abs STAT1 (Y701)-PE, STAT3 (Y705)-PE, STAT4 (Y693)-Alexa Fluor 488, STAT5 (Y694)-PE, STAT6 (Y641)-Alexa Fluor 488 (all BD Biosciences), NF-κB-p65 (S536)-Alexa Fluor 488 (Cell Signaling Technology) for intracellular staining. Isotype-matched irrelevant Abs were used as controls. Specificity of intracellular phospho-staining was verified by using blocking phospho-peptides (1 μg/sample; Santa Cruz Biotechnology). Geometric mean fluorescence intensities (GMFI) of phospho-STAT/-NF-κB proteins were calculated for both unstimulated and stimulated CD8 T cell subsets. We determined differences in phosphorylation by obtaining the log2 ratios of GMFI of stimulated vs unstimulated cell populations. Differences in basal phosphorylation were compared by calculating each sample’s GMFI log2 ratio divided by the minimum among all samples. We then performed phosphorylated kinase clustering analysis using Cluster and Treeview programs (Michael Eisen, University of California, Berkeley, CA). One-way ANOVA was used to determine the significance of differences in phosphorylation between CD8 T cell subsets. Post hoc testing was conducted using Tukey’s significant difference test (α = 0.05).2
Relationships between CD8 memory T cell expression signatures
We isolated highly purified CD8+ TN (CCR7+CD45RA+), TCM (CCR7+CD45RA−), TEM (CCR7−CD45RA−), and TEMRA (CCR7−CD45RA+⇓A). In contrast, TEM and TEMRA samples from the same donors formed individual clusters. Thus, two main clusters could be distinguished: TN/TCM and TEM/TEMRA. We confirmed the robustness of these two main clusters by obtaining a high overall cluster reproducibility measure (R index = 0.949, dendrogram cut at level of four clusters). Second, we performed MDS analysis to assess the distances between the expression signatures of the CD8 T cell subsets. MDS could separate TEM and TEMRA, but these two populations were still in close proximity (Fig. 1⇓B). In contrast, their transcriptional profiles placed TCM replicates between TN and TEM/TEMRA. We also calculated average linkage distances between TN and the different memory subsets as a measure for their relatedness. This showed that the gene expression profile of TCM was closer to TN than were the profiles of TEM or TEMRA (Fig. 1⇓C). Furthermore, we determined the number of differentially expressed genes between the CD8 T cell subsets by individual pairwise comparisons: It was highest for TN vs TEMRA and TCM vs TEMRA and lowest for TEM vs TEMRA (data not shown), again confirming the dichotomy TN/TCM vs TEM/TEMRA. Finally, we conducted a global permutation test to assess whether gene expression profiles between the CD8 subsets differed. The significance values (TEM vs TEMRA p = 0.05714; all other pairwise comparisons p = 0.02857) again suggest that TEM and TEMRA have the most similar gene expression profiles among all CD8 T cell subsets. In summary, analysis of our results, using several different approaches, indicate that the TEM and TEMRA subsets are closely related, whereas TCM have an expression signature that is distinct from that of the other primed T cells and is more closely related to the TN population.N and TCM all four replicates clustered together (Fig. 1
Genes differentially expressed between CD8 T cell memory subsets
Next, we identified genes that showed significant differential expression among the four CD8 T cell subsets. We used stringent statistical criteria, including a permutation test to minimize the FDR (see Materials and Methods). A total of 2092 probe IDs corresponding to 940 named genes met our criteria for differential expression: p < 0.001, FDR <1% with 95% confidence.4 Hierarchical clustering of the differentially expressed genes revealed six main clusters (Fig. 2⇓). Number of probe IDs in each cluster was as follows: 103 (cluster 1), 1369 (cluster 2), 92 (cluster 3), 16 (cluster 4), 129 (cluster 5), and 383 (cluster 6). Two major patterns could be identified among the clusters: First, genes with low expression in TN with increasing expression from TCM → TEM → TEMRA (clusters 2 and 3), termed “effector memory signature.”4 Second, genes with high expression in naive cells with decreasing expression from TCM → TEM → TEMRA (clusters 5 and 6), termed “naive signature.”4 Cluster 1 consisted of genes with higher expression in TCM and TEM.4 Cluster 4 comprised genes with a “TCM-specific” expression pattern. Surprisingly, very few genes fell within this category and probe IDs for CD28 accounted for 3 of the 16 included in this cluster (Fig. 2⇓). Overall, >70% of all differentially expressed genes had expression levels in TCM that were intermediate between their expression levels in the TN population and the TEM and TEMRA subsets. These analyses again suggest that, at the level of gene expression, TCM represent a differentiation state that is intermediate between TN cells and the TEM and TEMRA subsets.
The effector memory signature of human CD8 T cells
Genes in clusters 2 and 3 (effector memory signature) appear to represent genes that underpin the known high effector capacity of TEM and TEMRA. We identified GO categories that displayed statistical overrepresentation among cluster 2 and 3 genes (hypergeometric statistics, Holm multiple testing correction, α = 0.05): GO Biological Processes such as “immune response,” “cellular defense response,” “cytolysis,” and “protein transport” were overrepresented among the effector memory signature genes (Table I⇓). Thus, TEM and TEMRA highly expressed genes encoding lytic granule proteins like granzyme A (GZMA), granzyme B (GZMB), granzyme H (GZMH), and perforin (PRF1) as well as TNFSF10 (TRAIL) and TNFSF6 (FASL) that mediate perforin-independent apoptosis of target cells.4 Also TEM and TEMRA showed strong expression of genes involved in protein sorting to granules and granule transport/exocytosis such as HPS3, MYO5A, RAB27A, and RABGGTA. Humans with genetic defects of any of these genes have impaired T cell cytotoxicity (30). Finally, genes encoding inflammatory cytokines (e.g., IFNG (IFNγ)) and chemokines (e.g., CCL5 (RANTES)) were also present in clusters 2 and 3. Consistent with this, in TEM and TEMRA subsets, we found higher expression of transcription factors that control effector function in CD8 T cells (31), i.e., EOMES, TBX21 (T-BET), REL, NFATC2, and NFATC3.4
In the case of some proteins, particularly those expressed at the cell surface, Abs capable of detecting expression are available. Using these we have confirmed higher expression of CCR7, CD62L, CD27, and CD28 in TN and TCM subsets at the protein level (data not shown). Apart from effector molecules, differential expression of other genes from clusters 2 and 34 between different CD8 T cell subsets has previously been described at the protein level: ITGAL (CD11a), ITGB2 (CD18), ITGAM (CD11b), ITGA4 (CD49d), KLRB1 (CD161), KLRD1 (CD94), killer Ig-like receptor family members, and TNFRSF6 (CD95). Therefore, the results of the protein expression studies are consistent with the results of our gene expression study. Overall, this further strengthens the validity of our gene expression data.
Cytokine signaling signatures of CD8 T cell memory subsets
We observed differential expression of cytokine receptor mRNA in our microarray analysis (Table II⇓). For example, naive CD8 T cells were characterized by higher expression of receptors for IL-6 (IL-6ST) and IL-7 (IL-7R). In contrast, TEM and TEMRA showed preferential expression of receptors for IL-2 family cytokines (IL-2RB, IL-2RG) and for inflammatory Th1-type cytokines (TNFR2, IL-12RB1, IL-18R1, IL-18RAP, IFNGR1). This prompted us to investigate cytokine signaling in CD8 T cell subsets at the single-cell level using multiparameter flow cytometry (32, 33). In doing so, we aimed to generate a functional data set that could be subject to clustering and scaling analysis and give further insight into the relationship between the TCM and TEM/TEMRA subsets.
Activated STAT and NF-κB proteins that transduce cytokine signals were detected with phospho-specific Abs by intracellular FACS staining. Methanol permeabilization can compromise detection of some surface Ags (29), and we observed loss of discrimination for CCR7 surface staining (data not shown). Therefore, we substituted CCR7 with CD62L as a surface marker to identify CD8 T cell subsets in conjunction with CD45RA. CCR7 and CD62L expression largely overlap in CD8 T cells (Ref.10 , and own data not shown). In all, for each of the four T cell subsets, we analyzed basal expression levels of intracellular phosphorylated STAT1, STAT3, STAT4, STAT5, STAT6, and NF-κB. We also analyzed expression levels of phosphorylated STAT1 in cells stimulated with IFN-β or IFN-γ, phosphorylated STAT3 in cells stimulated with IL-6 or IL-10, phosphorylated STAT4 in cells stimulated with IL-12 or IFN-β, phosphorylated STAT5 in cells stimulated with IL-2, IL-7, or IL-15, phosphorylated STAT6 in cells stimulated with IL-4, and phosphorylated NF-κB in cells stimulated with IL-18 or TNF-α. Examples of phospho-specific intracellular staining are shown in Fig. 3⇓. Interestingly, ex vivo CD8 T cells had elevated levels of P-STAT1 and P-STAT5, which could be increased further by cytokine stimulation (Fig. 3⇓A). We confirmed specificity of intracellular phospho-staining by using blocking phospho-peptides (data shown). In addition, we found that, as known from the literature, some cytokines did not induce phosphorylation of certain STAT proteins, e.g., STAT1 and STAT3 in response to IL-4 or STAT5 and STAT6 in response to IL-10 (data not shown). Importantly, in TN and TCM, we observed high STAT3 phosphorylation in response to IL-6 and in TN, TCM, and TEM, we observed high STAT5 phosphorylation in response to IL-7 (Fig. 3⇓B). In contrast, TNF-α induced higher NF-κB phosphorylation in TEM/TEMRA. These results are in line with the differential mRNA expression of receptors for IL-6, IL-7, and TNF-α derived from the microarray data.
In all we performed experiments on samples of blood taken from 10 different donors and found differences in levels of basal STAT/NF-κB phosphorylation between the four different T cells subsets. Interestingly, TCM were characterized by significantly higher cumulative basal STAT/NF-κB phosphorylation compared with TEM and TEMRA subsets (TCM vs TEM, p = 0.006, TCM vs TEMRA, p = 0.001, one-way ANOVA/Tukey’s post hoc test). We also found higher basal P-STAT1 (p = 0.037) and P-STAT5 levels (p = 0.013) in TCM compared with TEMRA.
Furthermore, the four different T cell subsets responded differently, in terms of changes in STAT/NF-κB phosphorylation, to stimulation with cytokines (Fig. 4⇓). Despite their already elevated P-STAT5 levels, overall TCM had higher induction of STAT5 phosphorylation in response to IL-2 family cytokines (IL-2: TCM vs TN, p < 0.001, TCM vs TEM, p = 0.044, TCM vs TEMRA, p < 0.001; IL-7: TCM vs TEMRA, p < 0.001, IL-15: TCM vs TN, p = 0.002, TCM vs TEMRA, p = 0.011). In TN we detected strong phosphorylation of STAT1 in response to IFN, STAT3 in response to IL-6, STAT5 in response to IL-7, and STAT6 in response to IL-4. Overall, TN and TCM displayed the highest cumulative cytokine-stimulated phospho-response (TN vs TEMRA, p = 0.01, TCM vs TEMRA, p = 0.02). Finally, we found preferential phosphorylation of NF-κB in response to IL-18 and TNF-α in the TEM and TEMRA subsets (IL-18: TEM vs TN, p < 0.001, TEM vs TCM, p < 0.001, TEM vs TEMRA, p = 0.008; TNF-α: TEM vs TN, TEM vs TCM, TEMRA vs TN, and TEMRA vs TCM, all p < 0.001). Thus, human CD8 T cell subsets are characterized by the differential activation of cytokine signaling pathways.
We clustered CD8 T cell subsets based on their cytokine signaling profiles. The 18-parameter phospho-signature, based on the 6 basal and 12 stimulated levels of phosphorylated STAT and NF-κB molecules, again placed TCM between TN and the TEM and TEMRA subsets (Fig. 5⇓A). Interestingly, the responses to the proinflammatory cytokines IL-12, IL-18, and TNF-α clustered together and formed a bigger cluster with IL-10 and IL-15. In contrast, IFNs were found in a cluster with IL-4, IL-6, and IL-7. We also performed MDS analysis based on our FACS phospho-proteome data. Similar to the gene expression data, TCM took up an intermediate place between TN and TEM/TEMRA (Fig. 5⇓B). Consistent with this, the average linkage distance between TCM and TN was smaller than that between TEM or TEMRA and TN (Fig. 5⇓C). Thus, the cytokine signaling signatures support the observations from transcriptional profiling in terms of the molecular relationship of CD8 memory cells.
In this study, we identified both the gene expression and cytokine signaling profiles of CD8 TN, TCM, TEM, and TEMRA cells. Our results provide a molecular basis for the different functional properties of different memory subsets, especially in terms of their self-renewal and effector capacity. They also shed light on the relationship between the primed T cell subsets.
In contrast with a recent study (34) that used the van Lier classification of CD8 T cells, we used the Lanzavecchia model as a basis for our experiments. This allowed us to compare the properties of the TCM subset with those of other T cell populations. Interestingly, we found a clear dichotomy between TCM and the two effector memory (TEM and TEMRA) CD8 T cell subsets at the molecular level. Analysis of both gene expression and cytokine signaling showed that TCM cells were significantly more closely related to TN cells than TEM or TEMRA cells were. Furthermore, the results of the gene expression analysis imply that TCM cells form a population that is broadly intermediate between TN and TEM or TEMRA cells. The observation that there are very few truly “TCM-specific” genes is most consistent with the idea that these cells form part of a continuum of differentiation or functional states (35, 36, 37). However, it is possible that some of the defining features of TCM are epigenetically imprinted, but not yet actively transcribed, and are therefore not identified by transcriptional profiling.
In many respects, the genetic profile of CD8 TCM cells suggests they are equipped with features of both naive and effector memory cells. Thus, TCM possess some stem cell-like qualities, such as a high proliferative capacity, that are also found in the TN cell population. In this regard, TCM displayed high basal and IL-2 family-induced STAT5 phosphorylation. Cytokines of the IL-2 family, which mainly signal through STAT5, are known to play an important role in the differentiation and homeostasis of CD8 T cells (38, 39). Thus, IL-15 can induce the homeostatic proliferation of CD8 memory T cells, whereas IL-7 has recently been implicated in the survival of murine CD8 memory precursors, which can be identified by selective expression of the IL-7Rα (40). Furthermore, a recent study found that STAT5 regulates the self-renewal and differentiation of human memory B cells (41). Therefore our results might explain why TCM have a greater expansion potential and self-renewal capacity than effector memory cells, TEMRA in particular. In contrast, we found that both effector memory subsets strongly express genes involved in CD8 T cell effector function. However, our data also show that the gene expression profile of TCM is biased to some extent toward effector differentiation, characteristic of effector memory cells. Thus, mRNAs encoding effector molecules like perforin, granzyme A, IFN-γ, and RANTES were up-regulated in TCM compared with TN.
The lineage relationship between CD8 T cell subsets remains controversial. Our study is cross-sectional in design and therefore does not directly address questions relating to T cell lineage. It is however interesting to consider the results of our study in the light of recently proposed models. Lanzavecchia and colleagues (6) have suggested a signal strength model of T cell memory differentiation. According to this model, TCR signal strength determines the memory cell fate a naive T cells acquires: A low Ag dose leads to the differentiation TN → noneffector → TCM, whereas a high Ag dose to TN → effector → TEM differentiation. Their work also suggests that human TCM further differentiate into TEM following Ag stimulation and into TEMRA in response to homeostatic cytokines (9). The results of our microarray analysis showing that the gene expression profile of TCM cells is largely intermediate between TN and TEM or TEMRA subsets would be consistent with this type of model. In contrast, another model based on the idea that TCM and TEM are independent populations has been proposed by Pannetier and coworkers (19), who found that the TCR repertoires of influenza-specific TCM and TEM are largely distinct. We did not find many truly TCM- and TEM-specific genes and this argues against an independent TCM/TEM differentiation pathway. Finally, Ahmed and coworkers (18) have defined a linear differentiation pathway TN → effector →TEM → TCM in a murine model of acute LCMV infection. Interestingly, this TEM → TCM conversion does not occur in chronic LCMV infection (42). A second study by Pannetier and colleagues (43) using murine H-Y-specific CD8 T cells confirmed this linear differentiation pathway although there was also evidence for an independent TCM/TEM differentiation pathway. The observation that TCM cells are intermediate between TN and TEM cells does not obviously support this type of linear differentiation model. However, the present study is cross-sectional in nature and therefore we do not exclude the possibility that TEM convert to TCM. Interesting questions about the plasticity of the genetic program of the CD8 memory T cell remain and further work will be needed to address these issues. Finally, it is noteworthy that the studies by Ahmed and Pannetier have not examined the TEMRA subset in relation to TCM and TEM as we did in the present work. Effector memory cells re-expressing CD45RA form a sizeable population in human peripheral blood and are likely to have an important role in human CD8 T cell memory.
In conclusion, our genomic and phospho-proteomic study demonstrates a dichotomy between CD8 TCM and TEM/TEMRA cells at the molecular level and suggests that human TCM cells represent an intermediate state between the TN and the TEM/TEMRA populations in terms of CD8 memory differentiation and function. By defining molecular signatures for CD8 memory subsets, we provide a framework for the further molecular characterization of human CD8 T cell memory.
We thank Ann Atzberger for expert cell sorting, Hayley Woffendin of the LRF Microarray Facility for assistance in performing the GeneChip analysis, Sumit Bhattacharyya for submitting the microarray data to the EBI ArrayExpress database, and Charles Bangham for advice and reading of this manuscript.
The authors have no financial conflict of interest.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
↵1 This work was supported by grants from the Medical Research Council (U.K.) and the Leukaemia Research Fund. T.W. is a Medical Research Council Clinical Research Fellow and M.F.C. is a Medical Research Council Senior Clinical Fellow.
↵2 Address correspondence and reprint requests to Dr. Tim Willinger, MRC Human Immunology Unit, Weatherall Institute of Molecular Medicine, The John Radcliffe, Oxford OX3 9DS, U.K. E-mail address:
↵3 Abbreviations used in this paper: TN, naive T cell; FDR, false discovery rate; GMFI, geometric mean fluorescence intensity; GO, gene ontology; MDS, multidimensional scaling; TCM, central memory T cell; TEM, effector memory T cell; TEMRA, effector memory RA T cell; CD62L, CD62 ligand; LCMV, lymphocytic choriomeningitis virus; ID, identifier.
↵4 The online version of this article contains supplemental material.
- Received June 21, 2005.
- Accepted August 24, 2005.
- Copyright © 2005 by The American Association of Immunologists