Defining the Molecular Hallmarks of T-Cell Memory

Abstract

The pool of memory CD8 T cells is comprised of highly specialized subpopulations of cells with both shared and distinct functions. The ongoing study of T-cell memory is focused on how these different subpopulations arise, how the cells are maintained over the life of the host, and how the cells protect a host against reinfection. As a field we have used the convenience of a narrow range of surface markers to define and study these memory T-cell subsets. However, as we learn more about these cells, it is becoming clear that these broad definitions are insufficient to capture the complexity of the CD8 memory T-cell pool, and an updated definition of these cellular states are needed. Here, we discuss data that have recently arisen that highlight the difficulty in using surface markers to functionally characterize CD8 T-cell populations, and the possibility of using the epigenetic state of cells to more clearly define the functional capacity of CD8 memory T-cell subsets.

OUR DEFINITION OF CD8 T-CELL MEMORY

The first defining characteristic of memory T cells is that upon reencounter with antigen, the cells recall effector function faster than the first time the antigen was seen. Memory cells produce interleukin (IL)-2 and interferon (IFN)-γ within 3 h of restimulation, while naive cells take >12 h (Bachmann et al. 2006). In addition, memory cells kill target cells in vitro after 5 h while naive cells have no activity at this time (Zimmermann et al. 1999). Using in vivo killing assays, lymphocytic choriomeningitis virus (LCMV) antigen-specific memory cells kill between 49% and 89% of target cells within 4 h, and completely eliminate targets by 24 h, while naive cells have no killing capacity (Barchet et al. 2000; Barber et al. 2003). Comparable kinetics are found with CD8 memory cells specific for mouse polyoma epitopes (Byers et al. 2003). Similar findings are reported in human memory CD8 T cells. Live virus vaccines such as the yellow fever and smallpox vaccines have been useful to track memory CD8 T-cell generation in humans from the beginning of exposure to infection. These studies have shown a rapid and robust effector CD8 T-cell response with effector cells expressing copious amounts of cytolytic molecules and antiviral cytokines. A robust effector response with similar properties is also seen following other primary viral infections such as HIV, CMV, EBV, Ebola, and Lassa, although the magnitude and duration are varied (Hertoghs et al. 2010; Odumade et al. 2012; McElroy et al. 2015, 2017; Ndhlovu et al. 2015). Most effectors are short lived but a small subset remains to seed the memory pool. These memory CD8 T cells exhibit rapid production of IFN-γ and IL-2 when stimulated in vitro unlike naive CD8 T cells (Akondy et al. 2009). Additionally, polyfunctionality—the ability to produce multiple cytokines—is thought to be an important measure of the quality of memory CD8 T cells and is often used to assess vaccine-elicited responses. Whereas much of the recall potential of human cells is inferred from either in vitro assays of cellular function or transcriptomic and, more recently, proteomic analysis (van Aalderen et al. 2017), these studies collectively point to a defining feature of memory cells as having the ability to rapidly recall an effector state after rechallenge with antigen or T-cell receptor (TCR) stimulation.

The second defining feature of memory CD8 T cells is that they persist without antigen. LCMV-specific memory cells can be transferred into naive hosts and persist indefinitely (Lau et al. 1994). Similarly, Sendai virus antigen-specific memory cells persist for >250 d after transfer into a naive host without any antigen detectable in the host (Hou et al. 1994). Eliminating the role of TCR signaling for survival entirely, MHC-I or β2 microglobulin-deficient hosts support extended survival of memory cells (Hou et al. 1994; Murali-Krishna et al. 1999). In mice, the longevity of CD8 memory has been attributed to stochastic proliferation within the memory pool with a small fraction of cells completing division with an intermitotic interval of 50 d (Choo et al. 2010). In humans, persistence of memory CD8 T cells can be remarkably long lived as has been shown in case of the smallpox and yellow fever vaccines (Crotty et al. 2003; Hammarlund et al. 2003; Fuertes Marraco et al. 2015; Akondy et al. 2017). Both are acute viral infections and make a case that memory CD8 T cells can persist without antigen. Whether this memory (that can be seen as long as 50 yr) is due to longevity of individual human CD8 T cells or due to homeostatic proliferation of these cells has been an interesting question. To address this, we used in vivo deuterium labeling to mark CD8 T cells that proliferated in response to yellow fever virus (YFV)-17D and assessed cellular turnover and longevity by quantifying deuterium dilution kinetics in YFV-specific CD8 T cells using mass spectrometry. We found that the memory pool originates from CD8 T cells that divided extensively during the first month (effector stage) after vaccination. Whereas most of these cells died resulting in decreasing cell counts, the population that remained was maintained by quiescent cells that divide less than once every year (doubling time of >450 d) (Akondy et al. 2017). Together, these studies highlight that the rapid recall of killing and proliferation and the persistence without antigen due to slow homeostatic proliferation are the defining features of memory CD8 T cells.

MEMORY T-CELL SUBSETS ARE MADE UP OF FUNCTIONALLY DISTINCT SUBSETS THAT CAN BE DEFINED BY SURFACE MARKERS

One of the seminal discoveries related to T-cell heterogeneity has been the observation that naive and memory T cells can be broadly delineated by the differing expression of CD45 isoforms. Building upon this observation, a major conceptual advance in the ability to partition memory T cells into distinct subsets occurred when Sallusto et al. demonstrated that the homing marker CCR7 could be used to further divide memory cells into two subsets, each associated with specialized circulating properties. Central and effector memory subsets were initially defined in humans as cells that were CCR7⁺CD45RA^– and CCR7^–CD45RA^–, respectively (Sallusto et al. 1999). Central memory cells had superior proliferative capacity in response to antigen, while the CCR7^– cells had better cytokine-producing ability and markers of tissue homing. Similar subsets with analogous functions were described in mice, with central memory cells expressing both CD44 and CD62L, while effector memory cells only expressed CD44 (Wherry et al. 2003). Following the phenotypic demarcation of these memory subsets, it was observed that within tissue there was a subset of memory cells that resembled effector memory cells (Masopust et al. 2001). Subsequent work using parabiosis experiments found that these cells within the tissue did not reenter circulation of the host but remained resident in the tissue, thus gaining the “resident memory” label (Klonowski et al. 2004). CD69 and CD103 have been identified as markers that distinguished these resident cells from circulating effector memory cells and are now standard markers to define these cells (Klonowski et al. 2004; Mackay et al. 2013).

Historically, the four markers, CD45RA, CCR7, CD28, and CD27, have gained widespread use to distinguish naive and antigen-experienced human T-cell subsets (Hamann et al. 1997). More recently, a stem-like CD8 T-cell population has been defined that looks like a naive cell, CD45RA⁺CCR7⁺, but coexpresses other markers like CD122 and CD95 (Gattinoni et al. 2011). These cells have the capacity to differentiate into central and effector memory cells, and a sizeable proportion of tetramer-positive yellow fever cells had this phenotype after >5 yr after vaccination (Akondy et al. 2017). Technical advances in profiling approaches that can analyze a broader range of molecules within a single cell have revealed even more heterogeneity when surface markers and cytokine production are combined (Newell et al. 2012). It has become apparent that T cells are remarkably diverse and using only a few parameters to parse them is not sufficient. Whereas all these cell types have unique functional characteristics, they all share the traits of long-term persistence without antigen and rapid recall of effector functions. Historically, deconstructing heterogeneity among total CD8 T cells has been of great value.

THE LIMITATIONS OF SURFACE MARKERS TO DEFINE FUNCTIONALITY

When we have information about the antigen specificity of a population of T cells, we know when these cells encountered antigen, if the antigen is still present, and can therefore be sure that they fit the strict definition of a memory cell. When the antigen specificity is not known, we infer function based on similar markers that have been defined in antigen-specific settings discussed above. The limitation of this approach is that these markers can be up-regulated on a T cell in a context other than memory, and maybe dynamically expressed at different time points related to antigen stimulation or cytokine environment. For example, Ova-specific OT-1 CD8 T cells transferred into irradiated B6 hosts up-regulate CD44 as they undergo homeostatic proliferation, giving the impression that these cells are memory cells (Goldrath et al. 2000, 2004). These cells have not seen antigen and have different rates of proliferation and different chemokine and cytokine production profiles compared to memory CD8 cells (Goldrath et al. 2000; Cheung et al. 2009). Importantly, this transition to the “memory” state is only transient. After 120 d, cells revert to the naive CD44^– state and have naive-like levels of IFN-γ production after 8 h of stimulation (Goldrath et al. 2000). Markers that define central and effector memory cells are also dynamic. When LCMV antigen-specific CD44⁺CD62L^– effector memory CD8 T cells are transferred into naive recipient mice, they reacquire CD62L expression (Wherry et al. 2003). A similar situation exists for CD69 as a marker of resident memory, as it is immediately expressed by CD8 T cells upon TCR engagement or type 1 IFNs (Yokoyama et al. 1988; Testi et al. 1989; Shiow et al. 2006).

It is interesting to note that T cells have often been misunderstood based on their phenotype in many early studies, CD45RA expression was sufficient to qualify as a naive T cell. Subsequently, CD45RA expression by cells lacking CCR7 got them labeled as “terminally differentiated” and an association with poor quality memory. The availability of tetramers and multiparameter analysis has clarified some of these issues. By tracking tetramer-positive cells over the course of a vaccine response, it was shown that during the peak of the effector response, antigen-specific CD8 T cells were CCR7^–CD45RA^–, but after only a month tetramer-specific cells had reexpressed CD45RA and cells observed several years later were CD45RA⁺ (Fig. 1; Precopio et al. 2007; Miller et al. 2008; Akondy et al. 2009). Whereas it can be argued that these phenotypes are restricted to blood, a rare analysis of total T cells in several human tissues has provided a major insight. Thome et al. showed that the TCM subset is most prominently seen in human CD4 T cells but represents only a small fraction in CD8 T cells among every tissue examined (Thome et al. 2014). We have observed that in case of YFV-specific tetramer⁺ memory CD8 T cells, the CD45RA^–CCR7⁺ TCM phenotype is rare (if at all present) among blood memory cells and is only seen transiently in the initial stage when activated YFV-specific CD8s are expanding (Precopio et al. 2007). It remains to be seen whether this observation can be extended to other antigenic specificities and/or other tissues.

Figure 1.

Dynamic changes in T-cell markers. (Left) The schematic shows the typical phenotypic expression of CD45RA and CCR7 in yellow fever virus (YFV)-specific CD8 T cells during differentiation and how they fit into the widely used memory differentiation subsets. Note that the naive and long-lived memory CD8 T cells are both CD45RA⁺CCR7⁺. Epigenetic marks such as the methylation patterns seen near the perforin (PRF) and granzyme B (GZMB) loci distinguish them (Akondy et al. 2017). The cells in the blue-shaded area represent early memory cells and show much more heterogeneity when other markers are included. For example, CD28, CD62L, CD127, CX3CR1, CXCR3, CD38, and CD57 are expressed at varying intensities and “gating” based on these could result in potentially 127 more subsets. (Right) The schematic shows how epigenetic changes can occur independently of transcriptional changes and may represent a more stable marker of a cellular state. Importantly, while memory cells may on the surface resemble naive or other cell populations, memory cells retain many epigenetic marks of their former effector function like open access to the GZMB locus. (CM) Central memory, (EM) effector memory, (EMRA) effector memory CD45RA⁺, (N) Naive, (E) Effector, (M) Memory.

Given the above-described challenges in using surface markers to define T-cell subsets, the optimal situation to define a memory cell is knowing the antigen specificity of the cell, knowing that the cell encountered antigen, and how long the antigen is present in the host. Obviously in many situations this level of detail is simply not feasible. Cancer, autoimmunity, and transplantation studies are exceedingly difficult to determine antigen specificity of the T cells responding, many viruses are never truly cleared by the host (EBV, CMV), or the pathogen is regularly reencountered (influenza). While these difficulties are hard to overcome, we need to at least consider the history of antigen encountered before broadly defining cells as memory subsets based on a few surface markers. For example, in mice, the CD44⁺ population contains memory cells, cells undergoing homeostatic proliferation, and virtual memory cells. Each cell looks very similar using a narrow set of markers, but interesting and unique biological mechanisms control each cell. If we give cells the same name, we will assume they have the same function, and may discourage more in-depth analysis of the biology at play. This is not an easily solved problem. Below we suggest that in the case where the history of antigen encounter is unknown, the epigenetic state of a T cell may be an alternative marker to help differentiate between cells that are actual memory and those that might just look like memory based on two or three markers.

EPIGENETIC REGULATION OF DEVELOPMENTAL PLASTICITY AS A NEW DEFINITION OF MEMORY T-CELL DIFFERENTIATION

Whereas great strides have been made in defining the cellular properties that contribute to T-cell immunological memory, the above-described advances also bring to light a major challenge the field now faces: phenotypic markers are no longer sufficient for accurately defining the differentiation status of effector and memory T cells. Rather than relying purely on cell-surface markers, the focus has now shifted toward identifying a unifying molecular mechanism that promotes and reinforces cell-subset-associated functions. Building upon studies focused on early development and hematopoiesis, T-cell immunologists have begun to study the role of epigenetic modifications in mediating preservation of cell-fate-associated gene expression programs during the processes of effector and memory T-cell differentiation, memory T-cell homeostatic self-renewal, and T-cell exhaustion.

CD8 T-cell effector and memory differentiation involves extensive epigenetic reprogramming of naive CD8 T cells. Recent studies focused on examining the origin of memory CD8 T cells revealed that a subset of T cells that are destined to develop into long-lived memory T cells, termed memory precursors (MPs), transition through an effector stage of differentiation, imparting them with epigenetic programs that facilitate effector functions (Russ et al. 2014; Gray et al. 2017; Youngblood et al. 2017). Notably, human memory T cells can maintain these effector-associated epigenetic programs for years during antigen-independent homeostatic self-renewal (Fig. 1; Abdelsamed et al. 2017; Akondy et al. 2017). The heritable maintenance of these effector-associated programs likely serves as a mechanism enabling a dividing population of memory T cells to remain poised to rapidly recall effector function upon antigen reexposure. While effector potential appears to be stably imprinted in long-lived memory T cells, memory T cells also retain plasticity in their developmental potential. Such plasticity is clearly shown during the antigen-driven process of an effector recall response; however, the cell's developmental plasticity is further demonstrated by its ability to undergo antigen-independent changes in memory subset specification. A prime example of such antigen-independent reprogramming is the progressive reacquisition of naive-associated gene-expression programs during the effector to memory stage of an immune response following resolution of an acute viral infection. This process has been recently described for the lymphoid homing molecule L-selecting (CD62L). Initially down-regulated during the effector stage of an immune response, the L-selectin promoter acquires transcriptionally repressive DNA methylation programs, which are subsequently erased as effector cells develop into long-lived memory T cells. While providing important insight into the developmental origin of memory T cells, these data also highlight the potential utility of epigenetic programs as a way to define the differentiation status of a T cell. It is now clear that each CD8 T-cell subset has a unique DNA methylation signature across the genome. Identifying differentially methylated regions (DMRs) between the T-cell subsets at key loci has been used to establish a unique “epigenetic fingerprint” that can be used to determine the cell's differentiation status.

The field's recent appreciation for the breadth of epigenetic remodeling events that occur during CD8 T-cell differentiation has enabled the conceptualization of these molecular events as a universal definition of T-cell differentiation. In an effort to formalize this definition, our group has recently used a machine-learning approach to analyze genome-wide DNA methylation profiles among a variety of mouse and human CD8 T cells to design a novel predictive “T-cell multipotency index” (Fig. 2; Abdelsamed et al. 2020; Fonseca et al. 2020; Zebley et al. 2020). The human version of this index was generated by using naive and HIV-specific CD8 T cells as the “developmental bounds” for identifying key CpG sites that could be used to determine the differentiation status of T cells (Abdelsamed et al. 2020; Zebley et al. 2020). The multipotency index was validated using previously characterized CD8 T-cell subsets and now has the capacity to predict the relative differentiation status of CD8 T cells. Similar to the human index, a murine version was established using exhausted CD8 T cells established from the classic chronic viral infection model, LCMV. These bioinformatic tools can be used to predict the differentiation status of CD8 T cells based on the methylation of specific CpG sites. Application of this tool in both mice and humans has now allowed for accurate delineation of terminally differentiated and stem-like CD8 T cells in the context of type 1 diabetes. Furthermore, the development and successful utilization of this tool in different species supports epigenetic (DNA methylation) programming as a universal mechanism for defining memory T-cell differentiation.

Figure 2.

An epigenetic-based predictive index that defines T-cell differentiation status. Whole-genome DNA methylation profiles were generated from mouse and human CD8 T-cell populations spanning the differentiation spectrum from naive to functionally exhausted. Using genome-wide DNA methylation programs from these well-defined differentiation states, a machine-learning approach defined key CpG sites in which methylation status is predictive of T-cell differentiation. Application of this epigenetic atlas allowed for generation of a multipotency index with a normalized score (0–1) of T-cell differentiation. (Tscm) Stem memory T cell, (Tcm) central memory T cell, (Tem) effector memory T cell, (Texh) exhausted T cell. (Figure from Abdelsamed et al. 2020; reprinted, with permission, from the authors.)

The above-described T-cell multipotency index is based upon our recent appreciation for the critical role de novo DNA methylation programming plays in limiting the differentiation potential of CD8 T cells. However, CD8 T-cell differentiation involves rearrangement of both DNA methylation and histone modifications. Histone modifications need to be incorporated along with DNA methylation to further develop a more comprehensive epigenetic-based definition of CD8 T-cell subset differentiation. Gray et al. demonstrated that histone modifications indicative of transcriptionally permissive chromatin are acquired at both pro-memory and pro-effector genes in MP cells consistent with their capacitary to further develop into long-lived memory T cells (Gray et al. 2017). Similarly, Pace et al. investigated the epigenetic restriction of memory T-cell differentiation by investigating the histone methyltransferase Suv39H1. Broadly, the authors demonstrated that the silencing of chromatin at stem/memory-associated genes was coupled to CD8⁺ T effector terminal differentiation (Pace et al. 2018). These data also support the developing concept that T-cell multipotency is intimately coupled to changes in epigenetic programs.

An ability to define a T cell's differentiation status based on a universal epigenetic signature not only helps resolve open questions regarding the origin of memory T cells, but also has significant translational importance as we move forward into an era of T-cell-based immunotherapies (Zebley et al. 2020). As described above, DNA methylation serves as a mechanism for maintaining acquired effector-associated gene-expression programs during a T cell's acute exposure to antigen. Building on this concept, our group and others have recently explored the role of epigenetic programs in limiting a T-cell's proliferative capacity and effector potential during chronic antigen exposure. The progressive decline of effector potential that a T cell experiences during chronic stimulation, commonly referred to as T-cell exhaustion, is coupled to the acquisition and reinforcement of specific epigenetic programs (Pauken et al. 2016; Sen et al. 2016; Jadhav et al. 2019). Importantly, blocking acquisition of de novo epigenetic events during the chronic stimulation of a T cell prevents the development of several exhaustion-associated qualities (Ghoneim et al. 2017). Collectively, these studies indicate that epigenetic programs can be used to define the T cell's progression toward the commitment of an exhausted fate. Recent efforts to further identify transcriptional regulators that contribute to the development of exhaustion-associated epigenetic programs have revealed several transcriptional regulators, including Tox and Nr4a, which promote the early establishment of such epigenetic reprogramming events (Alfei et al. 2019; Khan et al. 2019; Scott et al. 2019; Seo et al. 2019). Such insights now make it possible to directly or indirectly (by modifying the specificity determinants) block exhaustion-associated epigenetic programs, directing the cell's fate toward a stem-like, long-lived memory population. These findings, taken together with recent studies demonstrating that a stem-like population of T cells is coupled to the clinical efficacy of ICB and CAR T-cell therapies, suggest that therapeutic strategies that promote stem-associated epigenetic modifications may enhance current immunotherapy approaches.

CONCLUDING REMARKS

In summary, surface markers for T-cell memory have greatly helped us define the functions of different T-cell populations over the years. However, we are now at a point where there are so many subsets of cells with unique and interesting biology associated with them, that the simple panels we have used are no longer sufficient to capture the complexity of the T-cell population. Furthermore, because of the dynamic expression of various markers our understanding of memory differentiation is somewhat limited because only the characteristics in an instance of time are captured with no knowledge of their history of differentiation. The above-described results and novel tools provide a foundation for further refining the epigenetic-based definition for T-cell differentiation and developmental potential. In the future, strategies to purposefully modify the epigenetic profile of human CD8 T cells will allow for rational design of T cells resistant to exhaustion and yield a long-lived pool of T cells that can maintain a prolonged antitumor response.