Memory T and memory B cells share a transcriptional program of self-renewal with long-term hematopoietic stem cells

Results

To broadly compare the gene expression profiles of HSC and T cell populations, we used Affymetrix GeneChip technology. For the initial analysis, we first used the Genechip data generated by Ivanova et al. (14) in which phenotypically well characterized and functionally defined mature HSC populations were compared with both St-HSC and LCP. As a source of monoclonal populations of naïve, effector, and memory T cells, we used the OTI T cell receptor transgenic mice because this strain greatly facilitates the functional definition and purification of the CD8⁺ T cells at various stages of differentiation. We purified naïve, effector, and memory CD8⁺ T cells as described in ref. 15. Each population expressed the expected markers (e.g., cell-surface) and exhibited diagnostic functional activities. In particular, we demonstrated that the memory T cells generated in our system self-renew by following them in a cohort of mice over the course of several months. OTI memory T cells maintained constant cell numbers, cell-surface phenotypes, and cytokine secretion profiles (C.J.L., A.W.G., C.B., and D.M., unpublished results), consistent with the published reports demonstrating that CD8⁺ memory T cells homeostatically self-renew over the lifetime of an individual (16, 17).

Having established functionally defined cell populations, we grouped the raw data (.cel files) and collectively preprocessed them with the “rma” method in the affylmgui statistical analysis package (detailed in Methods). Diagnostic transcripts behaved as expected, confirming that the data accurately depict the gene-expression profiles of naïve, effector, and memory cells. For example, the microarray-measured levels of CD44, IL-7 receptor, IL-15 receptor, IL-2 receptor, Bcl-2, Bcl-X, granzyme A, granzyme B, and IFN-γ transcripts all showed the same order of expression in naïve, effector, and memory cells as has been reported for their mRNA and/or protein in previous studies (ref. 15; see also Fig. 5A, which is published as supporting information on the PNAS web site). Furthermore, the profiles were remarkably similar to those independently generated by using a different T cell receptor-transgenic mouse/viral infection system, providing an important external validation of the experimental approach (ref. 18; Fig. 5B). Finally, the GeneChip data correctly predicted differential expression as measured by quantitative PCR (Q-PCR) ≈90% of the time, although often dramatically underestimating Q-PCR-estimated fold changes (Fig. 5C).

To identify transcripts augmented in memory cells, we first compared the gene expression profiles of memory and naïve cells by displaying them as a function of P value versus fold change (Fig. 1A). The x axis represents the log₂ (fold change), so that transcripts that are equally expressed in the two cell populations are zero and fall on the gray midline. A transcript whose expression is relatively enriched in a given population shows up away from the midline, toward the side where its expression is highest. The y axis represents the false discovery rate-corrected P value, with those transcripts with the lowest P values having the highest likelihood of being truly different in the two cell populations being compared. Thus, those transcripts whose expression is the lowest and farthest from midline are the most likely enriched in the population toward which they are skewed. The transcripts whose levels were most increased in memory cells were selected, and their expression values in memory versus effector cells were secondarily plotted (Fig. 1B). Thereby, a set of 98 transcripts enriched in memory CD8⁺ T cells relative to both naïve and effector CD8⁺ T cells was delineated. Analysis of these enriched transcripts within the various stem cell populations revealed a preferential representation in Lt-HSC-enriched genes compared with either LCP or St-HSC, both in numbers (73% and 78% respectively) and P values (Fig. 1C).

Inspection of the plots depicting expression of memory-enriched genes revealed that the degree of skewing by both P value and fold change appeared to be greater in the Lt-HSC versus LCP than the Lt-HSC versus St-HSC comparison. Indeed, those transcripts whose expression values correlated best with the biggest difference in self-renewal capacity, i.e., those most enriched in Lt-HSC, vis-à-vis LCP, were also enriched in Lt-HSC relative to St-HSC and in St-HSC relative to LCP (Fig. 1D in red). Furthermore, the memory-enriched transcripts appeared to be progressively lost with differentiation. Their loss correlated with progressive loss of self-renewal function in all three progenitor populations. In addition, because St-HSC retain full differentiation potential, the enriched transcripts are unlikely to represent genes solely involved in lymphocyte biology or fate commitment.

We also looked at those transcripts selectively down-regulated in memory CD8⁺ T cells relative to both naïve and effector cells (Fig. 1 E and F). One hundred two transcripts were identified with expression levels selectively down-regulated in memory CD8⁺ cells relative to both naïve and effector cells. Again, we observed a strong correlation between expression trends in memory CD8⁺ T cells and Lt-HSC. The transcripts that were absent or down-regulated in memory CD8⁺ cells were also relatively depleted in Lt-HSC (Fig. 1G). The transcripts most down-regulated in Lt-HSC versus LCP also appeared to be progressively up-regulated as cells differentiated into St-HSC and LCP (Fig. 1H in red). Together with the data presented in Fig. 1 A–D, this result demonstrated that the majority of T cell memory enriched transcripts are also enriched in Lt-HSC, whereas the majority of T cell memory depleted transcripts are depleted in Lt-HSC.

Next we addressed whether a similar correlation could be observed between Lt-HSC and the other self-renewing mature lymphocyte population, memory B cells. Thus, we turned to an independently generated collection of Affymetrix GeneChip data on memory, naïve, germinal center, and plasma B cells. To generate antigen-specific B cells in vivo, we immunized mice with the T-dependent immunogen NP-CGG, and antigen-specific cells were harvested at various time-points (detailed in Supporting Methods, which are published as supporting information on the PNAS web site). These cells were functionally verified by transfer into RAG^−/− hosts and measuring T-dependent antibody production (Fig. 6, which is published as supporting information on the PNAS web site).

In a fashion parallel to the above analysis, we compared the gene-expression profiles of memory and naïve B cells (Fig. 2A). The transcripts whose levels were most increased in memory cells were selected and subsequently plotted versus germinal center cells (Fig. 2B) and then plasma cells (Fig. 2C). Thereby, a set of 272 transcripts enriched in memory B cells relative to both naïve, germinal center, and plasma B cells were delineated. Analysis of these enriched transcripts within the various stem-cell populations revealed a preferential representation in Lt-HSC-enriched genes compared with either LCP or St-HSC, both in numbers (71% and 79%, respectively) and P values (Fig. 2D). Just as was observed in the T cell analysis, the degree of skewing of the memory B cell-enriched transcripts correlated inversely with the progressive loss of self-renewal capacity (Fig. 2E in red). Clearly then, a large fraction of those transcripts augmented in memory B cells relative to naïve, germinal center, and plasma cells were also selectively enriched in Lt-HSC.

A set of 481 transcripts selectively down-regulated in memory B cells relative to the other B cell populations was also delineated (Fig. 2 F–H). Again, a strong correlation between expression in memory B cells and Lt-HSC was observed, as transcripts depleted in memory B cells were diminished in Lt-HSC relative to both LCP and St-HSC (Fig. 2I). Those transcripts most down-regulated in Lt-HSC versus LCP also appeared to be progressively up-regulated as cells differentiated into St-HSC and LCP (Fig. 2J in red). These observations demonstrate that the majority of B cell-memory-enriched genes were augmented in Lt-HSC, whereas the majority of B cell-memory-depleted transcripts were depleted in Lt-HSC. This separate B cell data set provides an important independent confirmation of the T cell data comparisons.

Those transcripts whose expression was up-regulated in both memory T and memory B cells relative to their non-self-renewing counterparts were tabulated (Fig. 3A). Virtually all (92% and 85%) of these shared transcripts were enriched in Lt-HSC (Fig. 3B). Similar results were obtained when those genes down-regulated in both memory populations were compared, with 88% and 84% also down-regulated in Lt-HSC (Fig. 3C). We propose that these transcripts underlie the most restrictive transcriptional definition of immune memory, and virtually all of them were coordinately regulated in Lt-HSC. The transcripts expressed in concert among memory T, memory B, and Lt-HSC likely represent a transcriptional profile of self-renewal in these diverse hematolymphoid cells.

Q-PCR provided confirmation of the expression of several of those transcripts coordinately enriched in memory T cells and Lt-HSC. Lt-HSC were purified as Lin^−/lo, Sca⁺, c-kit⁺, CD34⁻, and Flt3⁻ cells (19–21). Likewise, St-HSC were purified as Lin^−/lo, Sca⁺, c-kit⁺, CD34⁺, and Flt3⁻ cells and LCP as Lin^−/lo, Sca⁺, c-kit⁺, CD34⁺, and Flt3⁺ cells. Although this LCP population differs from the one used in the GeneChip analysis in its Sca expression, both populations lack self-renewing capacity and show a limited differentiation capacity (14, 19–22). In nearly every individual comparison, the Q-PCR data confirmed what was observed in the microarray analysis (Fig. 7, which is published as supporting information on the PNAS web site). Indeed, the differences observed by Q-PCR were often much greater than those estimated from the chip data. Two noticeable exceptions were the cytokine receptors IL-18R and IL-7R. IL-18R did not appear to be enriched in either HSC population, whereas IL-7R transcript levels were high in Lt-HSC, low in St-HSC, and highest in LCP.

To observe how these shared transcripts partitioned in other stem cell comparisons, we considered our T cell data in conjunction with other array comparisons of stem cell populations. Although a number of additional hematopoietic stem cell global gene expression studies have been performed (23–25), previous work has shown that cross-platform comparisons do not correlate well (26). Therefore, we focused our analyses on studies performed with Affymetrix-based experiments and used the complete data sets published by Ivanova et al. (14), Ramalho-Santos et al. (27), and Akashi et al. (28). We are aware of only one other recently published Affymetrix data set comparing HSC populations (29), but we were unable to obtain the original files in time for inclusion in this publication. Collectively, the three independent data sets analyzed included adult and fetal hematopoietic stem cells as well as embryonic and neural stem cells. The vast majority of those genes whose expression we identified as coenriched in memory T cells and memory B cells (vis-à-vis their designated counterparts) were also augmented in the adult hematopoietic stem cell populations of Ramalho-Santos et al. (27) and Akashi et al. (ref. 28; Fig. 4). Furthermore, most of the shared transcripts were also increased in fetal-liver hematopoietic cell precursors of Ivanova et al. (14). These findings provide independent confirmation of our results, suggesting that this common “self-renewal” molecular signature might be a general feature of hematopoietic stem cell populations. However, there was not a consistent enrichment of the shared transcripts in either neural stem cell or ES cell populations, arguing that this particular molecular signature may be restricted to the self-renewing cells of the hematopoietic system.

Discussion

We sought to provide biological evidence for or against the hypothesis that memory T and memory B cells have reacquired the expression of molecules characteristic of long-term stem cells, coincident with their ability to self-renew. The data presented demonstrate that for both memory T and B cells, a significant subset of their transcripts was also found in Lt-HSC. Indeed, virtually all of those selected transcripts whose expression was most closely coordinated in B and T memory cells were similarly regulated in Lt-HSC. These observations provide evidence supporting the hypothesis that the self-renewal pathways used in memory T and B cells are related to those of hematopoietic stem cells.

Although nearly all of the transcripts shared between memory B and T cells were also found in Lt-HSC, there were many more transcripts shared between only one memory population and Lt-HSC (Tables 1–3, which are published as supporting information on the PNAS web site). Because hematopoietic stem cells are absolutely critical for the survival of the organism, it is likely that they rely on redundant pathways involved in self-renewal, only some of which are used in a given memory lymphocyte population. One explanation for the limited overlap observed between memory T and memory B cells is that memory T cells have reactivated different self-renewal pathways than memory B cells. Support for this explanation can be found by looking at those transcripts shared between Lt-HSC and only one memory population. Bmi-1 is conserved between memory B cells and Lt-HSC but is not enriched in memory T cells. Bmi-1 is a polycomb family member involved in self-renewal of hematopoietic stem cells (30), leukemia cells (31), and central and peripheral nervous system cells (32). Conversely, memory T cells and Lt-HSC, but not memory B cells, have up-regulated Iex-1 and Spi-2A. These transcripts are known regulators of apoptosis that function in memory CD8⁺ T cell survival (33–35). These examples support the hypothesis that a given memory cell lineage may have reactivated only a subset of the redundant pathways expressed in HSC.

Still other transcripts shared between one memory population and Lt-HSC have functions consistent with their potentially playing a role in self-renewal. Memory T cells and Lt-HSC share expression of TNF receptor II (p74R) and TNF receptor-associated factor 1 (Traf-1). These proteins associate within the cell, resulting in a signal that inhibits apoptosis (36, 37). Also present on the shared memory T cell list were several members of the RAS/mitogen-activated protein kinase pathway, known to be involved in decisions by stem cells to undergo proliferation, apoptosis, and differentiation (38–40). Likewise, the memory B cells and Lt-HSC share expression of several classes of transcripts that likely function in self-renewal. Tcf4 and Tcf12 are potentially downstream of β-catenin signaling, itself known to play a role in self-renewal in several stem cell systems (41). Finally, Mef2a and Mef2d are members of a class of transcription factors known to help translate calcium signals in neurons into long-term survival (42, 43). Taken together, the presence of these particular transcripts supports the general hypothesis that memory cells have selectively reactivated different self-renewal molecular pathways found in Lt-HSC.

Even though our data point to different self-renewal pathways being reactivated in either memory B or T cells, we observed several transcripts whose expression was shared between Lt-HSC and both memory populations. Of these jointly shared transcripts, only IL-7R has been shown to play a role in memory T cell self-renewal (5). Although IL-7R clearly plays a role in B cell progenitor differentiation, its role in memory B cell function is unknown. IL-7R is likely to be functionally required for memory B cell self-renewal. However, it is unlikely to function at the level of stem cells, because the IL-7R protein is not expressed on Lt-HSC cell surfaces. A more straightforward explanation for its Lt-HSC expression it that IL-7R gene transcription lies downstream of a common self-renewal pathway. Alternatively, there are several shared transcripts that are more likely to play a functional role in self-renewal. In particular, the signaling molecules mitogen-activated protein kinase 12 and PKC-ζ and the transcription factor Pou6f1 represent potentially convergent nodes in the network of self-renewal pathways.

Identification of these transcripts lends significant impetus for further testing of their functional relevance to hematopoietic and memory cell self-renewal. In particular, our data suggest that the polycomb complex that includes Bmi-1 is likely to function in memory B cell self-renewal in addition to its already reported role in hematopoietic stem cells. Given the role of polycomb genes in the maintenance of cellular memory of chromatin modification and transcriptional repression, these molecules are particularly intriguing candidates for functioning in immunologic memory. Further, it is worth considering the possibility that the separate pathways identified in our analysis might functionally converge within the cell. For instance, Bmi-1 itself has recently been shown to associate with and be phosphorylated by 3pK (mitogen-activated protein kinase AP kinase 3), which lies downstream of several mitogen-activated protein kinase pathways (44).

There has been a great deal of debate concerning the validity and reproducibility of defining a general molecular signature of stem cells (14, 27, 45, 46). Although “stemness” certainly requires an aspect of self-renewal, there are many additional functions and/or states that might be shared by the broad range of stem cells examined in the previous studies. Furthermore, it is not difficult to imagine arriving at similar phenotypes via divergent pathways, particularly within different lineages. Because memory T and memory B cells are descended from long-term and short-term HSC, we suggest that the focused comparisons presented herein provide unique insights into self-renewal within the hematopoietic system. Indeed, those genes shared between both memory populations were coordinately regulated in all three of the published HSC Affymetrix data sets we analyzed in Fig. 6. However, when the T cell data were considered in conjunction with previously published ES cell and neural stem cell data sets (14, 27), there was not a consistent enrichment of the shared transcripts in either of these two. This finding suggests that the molecular signature we defined may be restricted to the self-renewing cells of the hematopoietic system, a finding consistent with the published work of others showing conservation within, but not across, lineages (45, 46).

Our results have important implications beyond the identification of a self-renewal signature. For example, these shared transcripts are excellent candidates for those reactivated in the self-renewal program of leukemic stem cells (47–51). Indeed, an increase in the expression of the polycomb complex component Bmi-1 has been implicated in leukemogenesis (31, 52). Second, given the recent reports that memory CD8⁺ T cell self-renewal is preferentially localized to the bone marrow (53), it is an intriguing possibility that memory T cells and hematopoietic stem cells may have partially overlapping niches within the marrow that support their self-renewal. Finally, the data provides a glimpse of the shared biochemical mechanisms with which hematopoietic cells undergo self-renewal.

Methods

Supplementary Material