The Naming of Memory T-Cell Subsets

Abstract

With ever-improving methods of cell characterization, the field of immunology has enjoyed unprecedented opportunities to resolve distinctions between lymphocyte populations. However, this has led to a proliferation of “subset” designations that threatens to complicate and confuse clear identification of populations that share critical functional traits. This article discusses some of the challenges associated with a uniform approach to assigning subset designations to memory T-cell populations.

Subset: (noun) A part of a larger group of related things.

In biology, names are invaluable as a way to identify and characterize organisms, cells, and molecules. Whereas the formal chemical names for molecules are descriptive and follow established criteria (albeit with some notable exceptions), biological names are often a consequence of the whims of biologists and the vagaries of the accidents that led to their discovery. In modern times, geneticists have been the leaders in this, resulting in evocative gene names such as “bride of sevenless” and “mothers against decapentaplegic” (Thurmond et al. 2019), but even these names have a legitimate, if convoluted, basis that denotes their functional relationship to other genes. Perhaps, especially in immunology, the names that “stick” frequently make very little sense—anyone trying to teach mouse major histocompatibility complex (MHC) nomenclature to a class of students is probably familiar with the looks of dismay at the capricious way this “madman's alphabet” (Klein et al. 1983) is used to label molecules that occupy a central role in cellular immunology.

Luckily, names for major populations of immune cells are somewhat more intuitive—“dendritic cells” really are dendritic, most “T cells” are made in the thymus, and “B cells” are made in the chicken bursa or the mammalian bone marrow. Whereas such names do not hint at these cells’ function, they can help in the process of relating a population to its functional relevance (e.g., we now recognize that the major immune consequences of lacking a thymus or a chicken bursa correlate with the consequences of losing T or B cells, respectively) (Miller 1961; Cooper et al. 1965; Murphy and Weaver 2016).

But the associations implicit in applying a name to a thing pose a risk of being too restrictive, perhaps especially when considering cell function. If a cell has numerous phenotypic and transcriptional characteristics of a “Th1” CD4⁺ T cell, one might be highly confident it will not produce “Th2” cell cytokines upon stimulation, but that does not equate to confidence that this cell will actually produce Th1 cytokines after activation as it might be unable to mount a functional response at all or that the cell will not have important additional functional properties that distinguish it from other Th1 cells. The opposite can also apply; two populations that share key functional properties but differ in expression of what is considered a canonical phenotypic marker may be artificially resolved into distinct subsets with questionable significance.

To be most useful, the name of a cellular subset should carry both positive and negative implications of their characteristics and function; for example, one would expect a neuron to play a role in the nervous system and not produce enzymes specialized to hepatocytes. But there is substantial danger with both inclusion (lumping together distinct cell populations with some shared characteristic) and exclusion (artificially distinguishing cells that share a key trait). For the perspective of this article, I will argue that the rapid pace of cellular and genomic research has led to a proliferation of memory T-cell “subset” names and that immunologists in this field need to decide whether to permit the emergence of a new “madman's alphabet”—with the confusion that this will inevitably foster—or try to reach some consensus about the criteria for memory cell subset naming. This article is not intended to present a crafted plan for how to reach such consensus, however—it will be long on questions, short on conclusions—as the goal is to spark reevaluation of the current status of our naming system and prompt discussion.

This article will also not dwell on how memory T cells differentiate and their relationship to effector cells—these are important, topical issues (and the subject of several recent reviews [Jameson and Masopust 2018; Blank et al. 2019; Omilusik and Goldrath 2019])—rather, the intended focus of this discussion is how the function, phenotype, and homeostasis of memory T-cell populations can be accurately and economically reflected in their subset designations, potentially providing a way to incorporate future research findings in a logical framework.

THE NATURE OF THE PROBLEM: FROM ONE, MANY

Despite it being a quintessential feature of the adaptive immune system, the cellular basis for immune memory has only been well defined within the last few decades. A pivotal moment came with the realization that memory T cells were not a homogenous group, but rather showed characteristics indicating specialized functions. The landmark report from Sallusto, Lanzavecchia, and colleagues, published 20 years ago (Sallusto et al. 1999), revealed two phenotypically and functionally distinct populations of memory CD4⁺ T cells in human blood, which differed in expression of both trafficking molecules and effector function. One population, termed central memory (Tcm) cells, expressed the chemokine receptor CCR7 and high levels of CD62L (L-selectin), both of which were known to be important for trafficking into lymphoid tissues and produced large amounts of IL-2 when restimulated. Another memory subset, termed effector memory (Tem) cells, lacked CCR7 expression and were negative or low for CD62L expression—and hence would be unable to enter typical lymph nodes from the blood—and were initially described to be more poised for production of effector cytokines such as IFN-γ. Modifications, such as a focus on CD62L rather than CCR7 expression in studies on mouse Tcm arose in part because of purely pragmatic issues, in this case the difficulty of staining for mouse CCR7.

These terms have had a profound impact on the field, being widely adopted and standing the test of time remarkably well. But it is important to ask whether they are sufficient to capture the diversity of memory subsets. Indeed, a notable limitation of these original studies and hundreds of subsequent reports is that they focused primarily on memory T cells in the blood and lymphoid tissues; extension to analysis of cells in nonlymphoid sites, such as the skin, gut, lung, etc., revealed cells that shared some characteristics with blood Tem. These could be incorporated into the model by picturing them as Tem caught in the act of actively surveilling nonlymphoid tissues before returning to the circulation (although even early studies found that these cells had distinct phenotypic and functional properties, straining the idea that this population was in rapid equilibrium with the circulation). Instead, it has become clear over the intervening years that the vast majority of memory T cells in the parenchyma of nonlymphoid sites show minimal, if any, ability to recirculate but rather are “resident” memory (Trm) cells that persist long term in those tissues during normal homeostasis. Resident memory cells cannot, by definition, be sampled in the blood, yet they are recognized as playing a critical and unique role in immune protection at barrier tissues, as well as potentially contributing to autoimmune and immunopathological diseases (Schenkel and Masopust 2014; Park and Kupper 2015; Masopust and Soerens 2019). This “division of labor” in separating a mobile but readily recruitable population of circulating memory cells and a sessile pool of local first responders offers a layered strategy to deploy successive waves of immune reactivity following antigen reencounter.

Even within the circulating Tem-like pool, further subsets have been established. Painstaking studies indicate that not all blood-borne Tem are capable of entering nonlymphoid sites and returning to the circulation, and that this property is limited to a subset of what have been called “peripheral” memory (Tpm) cells, initially characterized by intermediate expression of the chemokine receptor CX₃CR1 (Gerlach et al. 2016). These cells access nonlymphoid sites and are thought to reenter the circulation after afferent lymphatic drainage that may use CCR7 and bypass the lymph node parenchyma. Another Tem-like subset, which expresses high levels of CX₃CR1 and shares characteristics with a subset also described as “long-lived effector cells” (hence referred to as Tllec in this article) (Olson et al. 2013; Böttcher et al. 2015; Gerlach et al. 2016; Herndler-Brandstetter et al. 2018; Omilusik et al. 2018), appears to be confined to the blood with minimal access to lymph nodes at all (Olson et al. 2013; Gerlach et al. 2016).

So—are we done? Would Tcm, Trm, and two subsets of circulating Tem (Tpm and Tllec) cover all the known memory subsets, at least with respect to division by trafficking? Unfortunately, the answer is unclear, in part because each of these subsets are open to further phenotypic and functional resolution, the significance of which is not always clear. For example, use of a few additional phenotypic markers resolves additional subsets of Tem and Tcm (Hikono et al. 2007). Are these merely subtle variations on a theme with minimal biological importance or do they represent critical functional diversity within the pool, with potentially great significance in immune surveillance or response? Similarly, memory T cells with stem-like properties (and hence termed Tscm), which share properties of both Tcm and naive T cells, have been identified (Zhang et al. 2005; Gattinoni et al. 2011; Restifo and Gattinoni 2013). Should we consider these a branch of an existing memory pool or an entirely separate subset?

The appeal of a “subset” designation is that it will allow one to anticipate consistent properties and distinctions between two or more groups within a larger group. If that larger group is “immune system cells,” distinctions are stark and numerous: T cells and dendritic cells differ in thousands of well-defined ways related to their development, gene expression, and function. T cells and B cells, or CD4⁺ and CD8⁺ T cells, are progressively more similar to each other, but still distinguishable by numerous characteristics. However, iterative dissection of cell populations will inevitably result in fewer properties being used to distinguish populations and less confidence that these differences are accurate or meaningful. Use of tools such as single-cell RNA-seq and phenotypic analysis based on dozens of markers can generate ever more fine-tuned differences between individual cells (see, for example, Newell et al. 2012): but does that help define key traits or just highlight expected—and potentially transient—biological variability among cells that might cross a range of differentiation states but essentially be similar?

An extreme extension of the iterative process would be to conclude that every memory T cell is actually a distinct “subset.” This progression may obey rules of logic, and might even be true at some level, but is operationally unhelpful if we want to apply this information to anticipate the behavior of these cells, such as how a memory T-cell population will respond to challenge for example, or how long the population is expected to persist in the animal. To be valuable, we need to be able to identify subsets based on major and predictable shared properties that have functional relevance. The trick is going to be reaching consensus about what properties are functionally meaningful and implement policy through which these can be practically and consistently indicated through subset designations.

SUBSETS BASED ON FUNCTIONAL DIFFERENCES BETWEEN MEMORY T CELLS

T-cell function can be characterized in many ways. We have already considered T-cell trafficking properties, which is one component. Another is effector function, which are the responses induced by antigen reencounter or other triggers of memory T-cell activation.

In comparison with the diversity and prominence of CD4⁺ T-cell “Th” subsets, each with distinct constellations of effector cytokines, memory CD8⁺ T cells are usually considered to be more homogeneous, most being limited to production of type I cytokines like IFN-γ and TNF, and potent cytolytic activity. That does not reflect the full range of memory CD8⁺ T cells; for example, Tc17 (CD8⁺ T cells making Th17-like cytokines) are observed in skin inflammatory diseases (Hijnen et al. 2013; Cheuk et al. 2014) and some CD8⁺ T cells produce IL-10 during ongoing antiviral immune responses (Sun et al. 2009; Trandem et al. 2011), although in both of these cases it is unclear whether these properties persist into the stable memory pool. At another extreme, exhausted CD8⁺ T cells, those facing persistent antigen exposure in the form of chronic viral infections or tumors, display only a subset of the normal functions of effector or memory CD8⁺ T cells, for example, showing impaired TNF production and cytolysis while secretion of IFN-γ may be minimally compromised (McLane et al. 2019). Cytokine production by memory T cells is not solely induced by T-cell receptor (TCR) stimulation; most memory CD8⁺ T cells act similarly to natural killer (NK) cells in their ability to produce IFN-γ in response to various combinations of inflammatory cytokines in the complete absence of known TCR ligands (Berg and Forman 2006; Freeman et al. 2012), essentially operating as innate immune cells.

How many subsets would cover these different functions? We could certainly say that Tc1 and Tc17 are distinct, but what about a Tc1 cell that transiently expresses IL-10, while one of its counterparts does not? Should those be called Tc1/10 cells? And how would we even know which cells had transiently expressed a cytokine if we wanted to extrapolate to the clinic (where fate mapping reporters are not an option)? We can use phenotypic, transcriptional, and epigenetic criteria to distinguish exhausted CD8⁺ T cells from “normal” effector or memory cells, but we can do the same thing with exhausted cells themselves and define multiple sub-subsets. PD-1 blockade can reanimate the effector function of exhausted CD8⁺ T cells (at least transiently) but this does not affect all exhausted cells equally. Cells that express CXCR5 and TCF1, together with low levels of Tim-3, are preferentially “rescued,” and during normal homeostasis these cells are thought to help sustain the chronic antigen-specific CD8⁺ T-cell population (Im et al. 2016; McLane et al. 2019). Do these cells need their own subset designation? And if so, would CXCR5 expression alone be sufficient to identify these cells (even though it is unknown whether expression of this chemokine receptor by that population relates to their amenability for checkpoint blockade)?

Related to the topic of exhausted T cells, there has been much discussion in the field about whether and how to discuss memory in the face of chronic antigen exposure. One could legitimately say that there is no such thing as memory while immune cells are still encountering antigen, which persists during chronic infections or uncured cancers. Still, studies in mice have shown that exhausted CD8⁺ T cells can mediate effective control of acute infections as well as “true” memory cells, and that these cells go on to form a memory population that nevertheless retains characteristics of the exhausted cells (Utzschneider et al. 2016; Blank et al. 2019). Similarly, checkpoint blockade can lead to elimination of the source of persistent antigen, but the exhausted CD8⁺ T-cell population is only transiently changed in gene expression and then reverts to a “memory” population expressing markers of exhaustion in the absence of antigen (Pauken et al. 2016; Blank et al. 2019). Hence, exhausted cells at least have the potential to become durable memory cells, and therefore will continue to be considered in this article.

SUBSETS BASED ON DERIVATION AND HOMEOSTASIS OF MEMORY T CELLS

Aside from trafficking potential and function, memory subsets have been named based on the way in which they were induced and/or maintained.

The homeostasis of memory T cells can be regulated by several factors. The cytokines IL-7 and IL-15 are important for both CD4⁺ and CD8⁺ T-cell memory, while the TCR is considered to be dispensable for memory T-cell maintenance (Sprent and Surh 2011; Raeber et al. 2018). But while generally valid, these “rules” are frequently broken. TCR engagement (or even expression) appears unnecessary for persistence of memory CD8⁺ T cells following acute antigen exposure, yet antigen reencounter is thought to be important for long-term durability of some CD4⁺ T-cell memory cells and maintenance of the exhausted CD8⁺ T-cell pool (Surh and Sprent 2008; McLane et al. 2019). Memory T cells in different locations or circumstances show distinct cytokine requirements. So while IL-15 is required for the basal proliferation and the long-term persistence of the circulating memory CD8⁺ T-cell pool and the Trm in some nonlymphoid tissues, Trm maintenance in other sites appears completely independent of IL-15 and some Trm cells appear to undergo minimal basal proliferation, if any at all (Mackay et al. 2013; Schenkel et al. 2016). Whether these TCR or cytokine receptor engagements affect T-cell function beyond maintenance of the memory population is unclear. The role of cytokines does not end there; for example, recent work indicates that circulating memory CD8⁺ T cells arising after skin infection need to encounter TGF-β produced by keratinocytes for long-term maintenance, while memory T cells of the same specificity primed systemically are sustained without this requirement (Hirai et al. 2019). These findings suggest pathways driving memory T-cell homeostasis may differ depending on the initial site of antigen encounter. Evidence is building that there are divergent metabolic requirements for distinct memory CD8⁺ T-cell subsets as well (O'Sullivan et al. 2014; Phan et al. 2016; Pan et al. 2017), although whether this relates to their trafficking patterns and/or specific cytokine requirements is unclear.

Cytokines can also drive the generation of memory-like T cells in the absence of foreign antigen recognition. This pathway was discovered from studies on the homeostasis of naive T cells in which naive mouse T cells were transferred into lymphopenic hosts (Sprent and Surh 2011; Raeber et al. 2018). The unexpected outcome was that naive T cells not only persisted but proliferated and acquired the phenotype of memory T cells (Sprent and Surh 2011). In some cases, that response could be mapped to recognition of commensal microbes or overt autoreactivity, but some generation of memory-like T cells occurs in germ-free and even in “antigen-free” mice (Haluszczak et al. 2009; Kim et al. 2016). Nevertheless, generation of these memory-like cells in lymphopenic settings required encounter with self-MHC molecules (class I for CD8⁺ T cells, class II for CD4⁺ T cells) indicating TCR engagement was involved, yet these responses did not involve overt autoreactivity (Sprent and Surh 2011). IL-7 and IL-15 were also important for this response, potentially acting sequentially in initiating and sustaining the memory-like populations (Jameson 2002; Sprent and Surh 2011). In addition, another cytokine, IL-4, was found to drive the induction of memory-like CD8⁺ T cells within the thymus (Jameson et al. 2015). Intriguingly, despite this response being driven by a canonical Th2 cytokine, the resulting memory-like CD8⁺ T cells behave as Tc1, producing IFN-γ upon stimulation (Jameson et al. 2015). These memory-phenotype cells (termed Tmp in this article), including populations that have been variously labeled innate, virtual, or homeostatic memory cells (Tim, Tvm, and Thm, respectively) as well as other names depending on context and investigator whim, appear to be functional, being able to participate in “recall” responses similarly (but not identically) to “true” memory cells elicited by deliberate foreign antigen immunization (Lee et al. 2013; Jameson et al. 2015; Renkema et al. 2016; White et al. 2017; Smith et al. 2018). Furthermore, memory-phenotype populations include cells specific for unencountered foreign antigens (Haluszczak et al. 2009; Smith et al. 2018), meaning that they have the opportunity to participate in primary immune responses alongside with their naive counterparts.

At a fundamental level, it is important the field recognize that not all memory-phenotype T cells were generated when naive T cells responded to a foreign antigen, and that a typical primary immune response likely involves participation by both naive and memory-like cells (White et al. 2017; Smith et al. 2018). Some phenotypic differences in the memory-like cells and “true” memory cells have been defined in mice, although it is unclear whether these will prove consistent in all cases and whether they translate to humans (White et al. 2017). This raises the key pragmatic issue: in real-world situations, rather than experimental animal models, can we confidently assert whether a memory-phenotype population arose from encounter with a foreign antigen, an overt autoreactive response, or through a homeostatic mechanism? Knowing the antigen specificity of the cells (using peptide/MHC tetramers or antigen stimulation, for example) can help address this, but only to a certain extent. We and others observed that memory-phenotype cells could account for a sizable fraction of foreign antigen-specific CD8⁺ T cells in unimmunized mice (even in germ-free mice) (Haluszczak et al. 2009; Smith et al. 2018), and Davis and colleagues found a high percentage of viral antigen-specific human CD4⁺ T cells were memory phenotyped in seronegative individuals (Su et al. 2013). Whether these cells arise due to ill-defined cross-reactive immune responses (heterologous memory) or are truly Tmp arising by homeostatic mechanisms is difficult to resolve.

And whereas the mechanisms by which memory(-like) cells are maintained by TCR signals, IL-7, IL-15, or IL-4, during homeostasis and bystander responses is important to understand, that does not mean it will always be useful (if even possible) to use that information as a basis to distinguish subsets of memory cells.

SUBSETS AND THE RISK OF DISTINCTIONS WITHOUT A DIFFERENCE

Cellular subsets in biology are usually taken to mean well-defined groups that split a larger group; neurons differ from hepatocytes in countless ways, but both are cells. At the other extreme, however, two cell populations might differ by observable criteria but actually share all their “important” functional characteristics. For example, proliferation in an activated lymphocyte population is not tightly synchronized, but it might be hard for an immunologist to convince her skeptical colleagues that there is a critical difference between a T cell that has divided seven times with one that has divided eight times over a certain time period, even though that distinction can be made experimentally.

These examples raise two issues: First, experimental tools can be used to distinguish cell populations, but these may not represent “subsets” but rather transiently distinct activation or differentiation states. Second, the closer two “subsets” are, the more the risk of identifying a distinction without a difference.

Mouse cellular immunologists rely on distinctions without a difference all the time: T cells from mice differing in congenic markers (e.g., bearing CD45.1 vs. CD45.2 alleles) are presumed to be identical in their development and functional potential, yet those populations are easily distinguished. One could label them as CD45.1 and CD45.2 “subsets,” but that is presumed to have no biological significance (assuming no other genes are altered). So how do we know when the difference between two memory T-cell “subsets” is significant or “meaningful”? The sad truth is that many times we do not have evidence for this, especially when phenotypic markers of unclear function are used to distinguish subsets.

The designation of a new subset is usually taken (whether intended or not) to indicate that these cells differ in substantial ways from other known subsets. The name does not give an indication of this; a neophyte would not know whether Tcm were more similar to Tscm than Trm based on the name, and the emphasis is on how a “new” subset is distinguished from an old one.

Perhaps part of the issue is that, like much of biology, the designation of immune cell subsets in recent years has been driven from the bottom up, identifying subpopulations by various means and then trying to piece together their functional properties and importance. Whereas this has proven extremely valuable in determining the cellular basis of known, functionally important elements of the immune response (e.g., which lymphocytes make antibodies, and which do not) the task becomes more difficult when we do not know what functional properties we are trying to explain. Returning to memory T-cell trafficking is illustrative of this point. The worldview that held Tcm and Tem as the major memory subpopulations was potentially sufficient to explain much of memory T-cell biology. Whereas Tcm could be seen as the reservoir of quiescent memory, recirculating through lymphoid sites and recruitable into an anamnestic response, Tem cells were their more active and adventurous counterparts, using the circulation to access diverse nonlymphoid sites to survey for reinfection. There was no “need” for Trm in this self-contained model, yet Trm were found and their discovery led (eventually) to a reevaluation of the diversity of memory T-cell trafficking patterns and an appreciation of the functional value that having a sessile population of tissue-resident memory cells offers. Importantly, the subset was identified first, and the discovered properties of Trm demanded reevaluation of how memory T-cell subset trafficking patterns is harmonized for optimal recall responses. Other examples in immunology could be used; we did not know there was a “need” to explain the existence of differentiated Th1 and Th2 T-cell clones until they were identified (Mosmann et al. 1986), and then their properties demanded incorporation into the worldview of immune function. But that certainly does not mean that every subset distinction we find automatically implies that those cells have a meaningful functional difference. The important issue facing immunology now (and not just in the field of memory T-cell subsets) is deciding when distinctions between two cell populations provide a valuable and functionally important separation.

WHAT ARE MEMORY T CELLS AND WHAT ARE PRACTICAL AND USEFUL WAYS TO SUBSET THEM?

The reader may have expected a section on “What is a memory cell” to have come up earlier, but it was important to outline the various ways that T cells with memory characteristics can arise and function, and a discussion of the pros and cons of subsetting, to properly grapple with this question. Incorporating those elements, here is a potential definition of T-cell memory:

A memory T cell is derived from a proliferating naive T cell, which loses numerous naive characteristics, persists beyond the immediate effector phase (if present), and exhibits changed functional properties compared to naive T cells, such as non-naive trafficking patterns or the potential for rapid effector responses (such as cytokine production or cytolysis).

This definition would include both “true” memory cells that have responded to a foreign antigen (or, potentially, to a self-antigen in an overt autoimmune response) but also memory-like cells that have arisen by homeostatic or unknown differentiation pathways (see Table 1).

Table 1.

T-cell “supergroups”

Designation	Criteria for inclusion	Defined subsets (or sub-subsets)	Notes
Memory T cell	Derived from a proliferating naive T cell, which loses numerous naive characteristics, persists beyond the immediate effector phase (if present), and exhibits changed functional properties such as non-naive trafficking patterns or the potential for rapid effector responses (such as cytokine production or cytolysis)	All Tcm and Tem supergroup populations
Memory T-cell classes
“True” memory T cell	Memory cells with known specificity for a previously encountered foreign antigen or an autoantigen	Tcm and Tem supergroup cells	“Non-naive phenotype” includes cells that do not bear a “classic” memory phenotype
Memory-phenotype T cell (Tmp)	Memory T cells with unknown specificity, or with known specificity for an unencountered antigen	Thm, Tim, Tvm	This group would also include heterologous memory cells if the cross-reactive epitopes are not clearly defined
		All Tcm or Tem supergroup cells of undefined specificity
Major subsets
Central memory supergroup	Non-naive phenotype T cells that express CCR7 and CD62L and display some functional characteristics of memory(-phenotype) cells	Tcm, Tscm, Tmnp	“Non-naive phenotype” includes cells that do not have a “classic” memory phenotype, such as CD45RA-RO+ in humans; hence, Tscm and Tmnp are included here; some Tpm have a Tcm phenotype but may be in transition from Tem-like Tpm
		Most Tmp
		Some Tpm (?)
Effector memory supergroup	Non-naive phenotype T cells that lack expression of CCR7 and/or CD62L and display some functional characteristics of memory(-phenotype) cells	Tem, Trm, Temra, Tllec	Tex show several functional defects, but still exhibit memory-like responses, such as cytolysis; the existence of long-lived Tfh memory is controversial
		Most Tpm
		Some Tmp
		Tex
		Tfh

Within those groups, we can incorporate the distinct subsets of cells with memory-phenotype and memory-function that have already been discussed. Interestingly, these can be quite well incorporated into an expanded definition of Tcm and Tem, which we will call Tcm and Tem “supergroups” (Table 1). This supergroup format builds on the almost ubiquitous usage of Tem and Tcm in the literature and may also allow for a more ordered organization of memory T-cell populations and scrutiny of the criteria used to identify subsets that are prominent in the literature. Specifically, designation of a memory T-cell subset needs to reflect substantial functional and gene expression properties that distinguish cells beyond differences in cherished phenotypic traits.

Perhaps a good example of this is what is included in the “Tcm supergroup.” Building on the original descriptions of Tcm, one could define these as T cells that display CCR7 and CD62L yet differ from naive cells in phenotypic and functional ways. This would include cells with “classic” memory markers such as high expression of CD44 in mice, or CD45RA^–/CD45RO⁺ phenotype in humans, but it would not end there. For example, the phenotypic and gene expression differences between Tscm and “classic” Tcm are quite modest (Gattinoni et al. 2011; Fuertes Marraco et al. 2015); yet the fact that Tscm cells are CD45RA⁺ and CD45RO-ve has drawn considerable attention and driven the designation of Tscm as a distinct “subset” (Restifo and Gattinoni 2013). One of the roughly 20 genes differentially expressed by human Tcm and Tscm (Gattinoni et al. 2011) is Hnrpll, which regulates the expression of CD45RA versus CD45RO, explaining the phenotypic distinction. But it is far from clear whether these alterations in CD45 isoform drive any functional differences between Tcm and Tscm. Rather, it is chiefly because CD45RA/RO expression has historically been used as a basis to distinguish naive versus memory human T cells that the Tscm phenotype is notable. Likewise, it was primarily the fact that mouse Tscm can be identified by their CD44^low phenotype, in contrast to the CD44^high phenotype of “typical” mouse memory cells, that caused them to stand out (Zhang et al. 2005; Restifo and Gattinoni 2013). In both species, the properties of Tscm appear to make them far closer to Tcm than other populations. Perhaps these should be more clearly recognized as a branch, or potentially a differentiate state, of the larger Tcm pool (see Table 1).

Along the same lines, most CD8⁺ Tmp are Tcm phenotype and only differ subtly from “true” memory cells in their response to foreign antigen encounter (Lee et al. 2013; White et al. 2016, 2017). Despite the fact that Tmp cells have not responded to a foreign antigen, endured an immunogenic response, or navigated through a discernible effector phase, they acquire most of the functional and “recall” characteristics of conventionally primed memory cells. Hence, their inclusion as a branch of the Tcm supergroup is justified and helps predict their immune potential.

The different types of memory T cells in the Tem supergroup are more divergent—there is no overlap in the trafficking pattern of Tllec (which appear to persist chiefly in the blood) versus Trm and the populations differ in numerous phenotypic and gene expression characteristics—yet both are united in the inability to access lymphoid tissues via HEV and being poised for cytolytic and other effector functions (Jameson and Masopust 2018). Important functional differences, in this case trafficking potential, do not limit the value of being able to unite them as a related group. That is not to say, of course, that an assay of blood Tllec could substitute for a direct evaluation of Trm, which highlights the practical issue of how we test whether the memory populations defined in mouse tissues actually align with populations in humans, where blood is often the only tissue available. Pioneering studies from Farber's group, exploring memory subsets in human organ donors, have shown that answering such questions is feasible (Thome et al. 2014, 2016; Gordon et al. 2017).

Finally, what might be a comprehensive subset naming system must be balanced with what is pragmatic. In most cases, we cannot know that a “memory” cell population found in, say, a human blood sample is derived from a proliferating naive cell or that it has persisted through the effector phase (except in the rare situations when the time of antigen exposure is known, such as following vaccination [Miller et al. 2008; Akondy et al. 2017]) and, again, sampling blood will tell us nothing about Trm. In the case of homeostatic memory cells, it is doubtful that an “effector phase” ever existed, and in the case of chronic infection one could argue that “the effector phase” never truly ends. But if modalities of major functional properties can be applied and correlated with consistent phenotypic traits, differences in how memory cells were generated, where they are in the kinetics of a response, and whether they express what is considered a canonical “memory marker” may become less important and the practical value of memory subsets more clear.

CONCLUDING REMARKS

As warned at the opening, this article is not intended to provide an answer to the question of how we should name memory T-cell subsets, but rather to highlight the need to reevaluate the foundations for the current nomenclature—in particular to resist preconceived notions of what constitutes “important” markers (which are often based on historical accident rather than defined functional significance) or the assumption that we can know the derivation of a memory- or naive-phenotype cell based on these markers. At the same time, to be practical and useful, a nomenclature system should not be overly complex. Finding this balance will require extensive and open-minded discussion within the T-cell memory “community.”