How to Reliably Define Human CD8⁺ T-Cell Subsets: Markers Playing Tricks

Abstract

In recent years, our understanding about the functional complexity of CD8⁺ T-cell populations has increased tremendously. The immunology field is now facing challenges to translate these insights into phenotypic definitions that correlate reliably with distinct functional traits. This is key to adequately monitor and understand compound immune responses including vaccination and immunotherapy regimens. Here we will summarize our understanding of the current state in the human CD8⁺ T-cell subset characterization field. We will address how reliably the currently used cell surface markers are connected to differentiation status and function of particular subsets. By restricting ourselves to CD8⁺ αβ T cells, we will focus mostly on major histocompatibility complex (MHC) class I–restricted virus- and tumor-specific T cells. This comes with a major advantage as fluorescently labeled peptide-loaded MHC class I multimers have been widely used to identify and characterize these cells.

Over the past two decades, advances in multiparameter flow cytometry drove the appreciation of an astounding heterogeneity within the T-cell compartment in humans. In 1997, by characterizing the membrane expression patterns of the costimulatory receptor CD27 and the tyrosine phosphatase CD45RA, human naive CD8⁺ T cells, expressing both molecules, were phenotypically distinguished from memory and resting effector-type cells, which express just CD27 or CD45RA, respectively (Hamann et al. 1997). Shortly thereafter, in 1999, it was shown that the CD45RA-negative memory pool could be divided into two functional subsets on the basis of the expression of the lymph node homing marker CCR7 (Sallusto et al. 1999). Here, the authors show how CCR7-positive central memory T (Tcm) cells exert limited effector functions while being capable of differentiating into CCR7-negative cells upon stimulation. Instead, stimulated CCR7-negative effector-memory T (Tem) cells display potent effector functions. With the introduction of expression patterns of the costimulatory receptor CD28 in 2002, an extra layer of phenotypic heterogeneity was added to defining antigen-specific CD8⁺ T-cell subsets (Appay et al. 2002). Indeed, the forthcoming definitions that were used to characterize CD8⁺ T-cell subsets practically remain unchanged up to this moment.

Despite the fact that none of these molecules have an immediately apparent causal relation with specific functions of antigen-experienced cells, they do correspond quite strongly to specific functional traits like proliferation capacity, production of cytokines and chemokines, and the ability to kill antigen-presenting cells. However, while these rather rigid definitions of T cells appeared ubiquitously on the CD8⁺ T-cell differentiation charts, newer insights showed how these definitions can be deceptive. In fact, antigen-experienced CD8⁺ T cells cover a range of properties that frequently pass these phenotypic borders. For example, antigen-experienced CD8⁺ T cells have been found expressing a naive phenotype, and nonantigen primed CD8⁺ T cells were found to reside within the effector and memory fractions. Also, the relation between the expression of these cell surface molecules and a functional profile is lost when examining T cells residing for prolonged periods of time in specialized tissues (van Aalderen et al. 2014). Our growing understanding of functional heterogeneity that exists within the memory CD8⁺ T-cell population feeds the requirement for a reevaluation of phenotypic definitions that best reflect distinct roles of CD8⁺ T cells.

OF MICE AND MAN

CD8⁺ T cells are crucial constituents of the part of the adaptive immune response that is tasked with the elimination of cells in which intracellular pathogens multiply and of genetically altered malignant cells. The conventional model for defining functionally distinct subsets is based on murine models where primary infections induce naive CD8⁺ T (Tn) cells to proliferate and differentiate into a population of effector cells. These effector cells provide immediate protection through the acquirement of specific functional traits such as the secretion of inflammatory cytokines, cytolytic activity, and the capacity to migrate to infected tissues. After clearance of infection, the effector cells are destined to die off by apoptosis. However, a smaller fraction of cells, defined as memory precursor cells, persist for a prolonged period of time during which these cells provide a superior immune response upon reinfection with the same pathogen. Although this model still describes the common context for defining human CD8⁺ T-cell subsets, it has also proven insufficient in capturing the actual complexity wherein various exceptions to the commonly accepted rules have been identified, especially in man. For example, while “memory” is classically defined as the capacity of CD8⁺ T cells to induce enhanced immune responses upon reencountering cognate antigens, this definition does not cover the heterogeneity of the substantial proportions of functionally distinct antiviral CD8⁺ T-cell subsets that are commonly found in the circulation of immunocompetent humans that have been infected by herpes viruses but in whom viral loads are undetectable. With new subsets being described at a steady pace through the use of newer techniques based on mass spectrometry, single-cell transcriptomics, and epigenetic analysis, we aim to summarize the current views in the field of human CD8⁺ T-cell subsets (Fig. 1). Better definitions of CD8⁺ T-cell subsets and consensus on minimal phenotypic traits may be highly relevant to characterizing immune responses, including outcomes of vaccination and immune therapy practices, with the appropriate resolution required to fully understand protective immunity in man.

Figure 1.

CD8⁺ T-cell subset lineage associations. (Tn) naive, (Trt) recent thymic emigrants, (Tscm) stem cell memory, (Tcm) central memory, (Tpm) peripheral memory, (Tem) effector memory, (Tex) exhausted phenotype, (Tfh) follicular helper, (Trm) resident memory, (Te) effector, (Tvm) virtual memory, (Tim) innate memory, (Tllec) long-lived effector cells.

HOW TO DEFINE NAIVE CD8⁺ T CELLS: MARKERS PLAYING TRICKS

Naive T cells are classically defined as T cells that have not yet encountered their cognate antigen. For a long time, Tn cells have been classified on the basis of the differential expression of CD45RA, CD27, CCR7, CD62L, and CD28. This provided a workable system where Tn cells could be readily distinguished from Tcm, Tem, and quiescent effector cells that (re)express CD45RA (also coined Temra cells). The use of CD45R isoforms to define T-cell subsets warrants specific discussion. In 1986, it was shown that activation of naive T cells in vitro leads to the conversion of the CD45RA isoform into the CD45RO isoform. This was also assumed to be an irreversible event (Smith et al. 1986). Therefore R0-expressing cells were considered as primed cells, whereas RA-expressing cells were presumed to be naive. Whereas the first assumption still holds, the second one was contested by reports demonstrating that quiescent antigen-specific effector-type cells can also express CD45RA (Hamann et al. 1997). As such, it then appeared that the switch from RA to R0 is, in fact, a reversible process. This process remains poorly understood to date but might be related to the fact that when cells become truly quiescent, CD45RA may be reexpressed. Repetitive stimulation either via (cross-reactive) ligands, low amounts of antigenic peptides available in the system, and/or the impact of stimulatory cytokines, notably interleukin (IL)-15, might maintain CD45R0 expression within the pool of primed T cells. Caution should still be used when using CD45 isoforms to define Tn cells as recently primed and activated CD8⁺ T cells that are in the G₂ phase of their cell cycle can express CD45RA and CD45R0 simultaneously (Hamann et al. 1997). As such, previously Tn cells that have just been activated and are actively cycling may briefly express CD45RA while not being truly naive. The kinetics by which antigen-primed Tn cells lose their CD45RA expression and acquire CD45R0 are still not completely understood. Consequently, Tn cells that are in transition to CD45R0 memory cells can still be found within the naive phenotype gates (CD45RA⁺CD27⁺CCR7⁺CD62L⁺ CD28⁺).

The cardinal property of the Tn cell population is the expression of a highly diverse preimmune T-cell receptor (TCR) repertoire that is essential for an effective immune response against most, if not all, microbial challenges. Whereas generated in the thymus, Tn cells were initially aligned to the phenotype of T cells detected in cord blood, where the reasoning was that cord blood T cells are virgin and would never have encountered their antigen or been exposed to inflammatory conditions. This classic definition of naive CD8⁺ T cells began to show its first cracks with the publication of papers that demonstrated various degrees of phenotypic heterogeneity within this population. First, Pircher and colleagues demonstrated that a substantial proportion (∼20%) of αβ CD8⁺ T cells in human cord blood expressed killer cell lectin-like receptor subfamily G member 1 (KLRG1), a coinhibitory receptor that appeared to be expressed by T cells that had undergone more divisions after having excited the thymus according to the amount of TCR excision circles (Marcolino et al. 2004). Cord blood CD8⁺ T cells predominantly express CD45RA, with little detectable expression of CD45R0 (Hamann et al. 1997; Haluszczak et al. 2009). In addition, virtually all cord blood cells coexpress CD27 and CD28, both costimulatory receptors required for proper priming, as well as the lymph node–homing molecules CCR7 and CD62L (or L-selectin) (Picker et al. 1993; Tedder et al. 1995; Sallusto et al. 1999; Unsoeld and Pircher 2005). In the absence of priming, Tn cells maintain a chromatin configuration at effector gene loci that precludes them from producing proinflammatory cytokines such as interferon γ (IFN-γ) and tumor necrosis factor α (TNF-α). However, Tn cells can produce IL-2, which is required for proliferation of T cells that occurs upon TCR-mediated activation (Hamann et al. 1997).

Furthermore, the naive T-cell repertoire is supported by thymic output, especially during the early years of life, and it was shown that Tn cells that had emigrated from the thymus recently can be identified by their expression of CD31 and CD103 (McFarland et al. 2000; Kimmig et al. 2002). Because the thymic output progressively declines during aging, the abundance of such recent thymic emigrants declines as well. Functionally it was demonstrated that this specific subset of Tn cells was capable of producing the inflammatory IL-8 (or CXCL8), which has the potential to activate innate cells including neutrophils and γδ T cells, contesting the dogma that T cells in very early life are intrinsically anti-inflammatory (Gibbons et al. 2014; van den Broek et al. 2016). Also, it was discovered that heterogeneity exists with regard to the levels of CD5 expression. CD5 is a surface marker of which the expression levels directly correlate to the TCR affinity for self-ligands (Azzam et al. 1998). Functionally, it was shown that Tn cells with the highest CD5 expression were also the most proliferative within lymphopenic hosts and would give rise to a specific subset of memory CD8⁺ T cells referred to as virtual memory (Tvm) T cells. It is therefore important to recognize that defining antigen-inexperienced T cells as naive brings the caveat of classifying CD45RA⁺CD27^–CD5⁺ Tvm cells but also the subset of CD5^low innate memory (Tim) T cells, distinguishable in mice by the lack of NKG2D expression, which emerge from the thymus as naive due to their independence of antigen-specific TCR priming (White et al. 2017). In the last decade, single-cell RNA sequencing, epigenetic analysis, and proteomic approaches have revealed a more substantial heterogeneity in the composition of phenotypically naive T-cell populations (Mold et al. 2010; Wong et al. 2016; Smith et al. 2018). It remains to be fully elucidated how this diversity relates to the differentiation of phenotypic and functionally different effector and memory CD8⁺ T-cell populations.

Because it was demonstrated that years after infection with Epstein–Barr virus (EBV), polyomavirus BK, or vaccination with smallpox or yellow fever viruses a substantial fraction of antigen-specific memory T cells displayed a naive CD45RA⁺CD27⁺CD28⁺CCR7⁺CD62L⁺ phenotype (Miller et al. 2008; Akondy et al. 2009; Long et al. 2013; van Aalderen et al. 2016), it was accentuated that not all cells within the apparently homogeneous Tn cell population are truly naive. This resulted in the identification of a long-lived small subset of memory cells that could maintain their own pool size over time through self-renewal referred to as stem-cell memory (Tscm) cells (Gattinoni et al. 2011; Lugli et al. 2013; Di Benedetto et al. 2015). These clonally expanded Tscm cells expressed a naive-like phenotype in addition to memory markers such as CD95, IL-2Rβ, CXCR3, CD58, and CD11a. Functionally, these cells, like other memory CD8⁺ T cells, proliferate rapidly and produce IFN-γ and TNF-α upon antigen reexposure. Tscm cells can be found in the blood, spleen, bone marrow, and lymph nodes but are absent from the peripheral tissues (Gattinoni et al. 2017). The ontogeny of these Tscm cells remains a subject of debate. Studying hematopoietic stem cell transplantation, it was postulated that Tscm cells differentiate directly from naive precursor cells and emerge early after priming as it occurs in vivo (Cieri et al. 2015; Roberto et al. 2015). In contrast, using yellow fever virus vaccination as a model, followed by an assessment of longevity and epigenetic landscape, Tscm cells were placed at the end of the hierarchical tree of CD8⁺ T-cell differentiation (Akondy et al. 2009; Youngblood et al. 2017).

In summary, the CD45RA⁺CD27⁺CD28⁺ CCR7⁺CD8⁺ T-cell population actually holds antigen-experienced memory cells. In other words, whereas labeling naive cells with the above-mentioned phenotype might be informative when looking at the overall population, it also comes with substantial caveats when specific subsets of antigen-reactive cells are being analyzed.

PROPERTIES OF HUMAN CD8⁺ T CELLS DURING PRIMARY INFECTION

Naive CD8⁺ T cells are constantly patrolling the secondary lymphoid tissue in search of antigenic peptides presented by major histocompatibility complex (MHC) class I molecules. Upon encountering their cognate antigen presented by dendritic cells in the lymph nodes, primed Tn cells become activated. This is followed by a cascade of signaling sequences that drive the proliferation and differentiation of these cells into effector and memory precursor cells that have the ability to recirculate and migrate to the inflamed sites. Importantly, although memory precursor cells are already present during the peak response of the primary infection, these cells do not yet express functional traits of mature memory cells as this apparently requires a progressive differentiation path (Kaech et al. 2002). Several models have been postulated to describe the mechanisms governing the differentiation of primed naive T cells into either short-lived effector cells or long-lived memory cells. Whereas some models postulate that the affinity and duration of the TCR stimulation during priming of Tn cells is orchestrating the fate decision of cells (Chang et al. 2007; Zehn et al. 2009; Henrickson et al. 2013), paradoxical outcomes demonstrate, based on progeny tracing of single-cell transferred or genetically barcoded naive T cells, that under a variety of conditions the dominant pathway of CD8⁺ T-cell differentiation is one naive cell with multiple fates (Stemberger et al. 2007; Gerlach et al. 2010).

The assumption that all naive T cells have an equal potential to differentiate into memory and effector subsets does not explain how cells adopt different fates, producing different proportions of cell populations (Plumlee et al. 2013). Recently, a link between fate choice decisions and developmental stages was demonstrated (Smith et al. 2018). In mice, it was recently shown that naive CD8 T cells generated early in life were biased to develop into Tvm cells that exhibit innate-like functions while naive CD8 T cells made during adulthood will selectively respond to their cognate antigens (Smith et al. 2018). Presumably, the early effector cells provide a measure of innate immune protection until a fully mature adaptive immune system is established. Future studies should validate whether a similar mechanism applies to human T cells and how this is related to hallmarks of an aging immune system in normal and thymectomized adults.

Important regulators of the transcriptome expressed in cells are major epigenetic processes including DNA methylation, histone modification, and organization of the chromatin architecture. These processes all direct how genes are transcribed and ultimately translated into proteins that determine the cell's functional capabilities. Various studies on human and murine CD8⁺ αβ T cells showed how the epigenetic landscape changes considerably as naive T cells differentiate into effector and memory T cells, influencing the transcription of genes coding for fate-determining transcription factors like Forkhead box protein O1 (FOXO1), Krüppel-like factor 2 (KLF2), lymphoid enhancer-binding factor 1 (LEF1), and transcription factor 7 (TCF7), all involved in inducing a memory phenotype in T cells, and eomesodermin (EOMES), T-bet (TBX21), and PR domain zinc finger protein 1 (also known as Blimp-1 or PRDM1), involved in inducing traits fitting effector cells. Epigenetic regulation also concerns targets of these transcription factors, such as granzyme A (GZMA), granzyme B (GZMB), perforin (PRF1), IFN-γ, and KLRG1, typical markers of effector cells, and IL-2α (IL2RA), CD27, TNF-α, CCR7, and CD62L (L-selectin or SELL) (Araki et al. 2008, 2009; Denton et al. 2011; Scharer et al. 2013; Shin et al. 2013; Russ et al. 2014; He et al. 2016; Abdelsamed et al. 2017). These alterations nicely show how external stimuli, in this case leading to T-cell activation, result in rapid transcriptional changes. Furthermore, epigenetic alterations are heritable, but can also be reversible or remain intact upon withdrawal of the antigenic stimulus during homeostatic proliferation by memory cells (Abdelsamed et al. 2017).

In mice, short-lived effector cells can be defined during the acute phase by an IL-7Rα (CD127)^low/killer cell lectin-like receptor subfamily G member 1 (KLRG1)^high phenotype (Kaech et al. 2003; Joshi et al. 2007; Sarkar et al. 2008). Because most effector cells die quickly when antigen loads decline, these cells were fittingly named short-lived effector cells or SLECs. In contrast, memory-precursor cells, referred to as MPECs, are long lived and express an IL-7Rα^high/KLRG1^low phenotype (Kaech et al. 2003; Joshi et al. 2007; Sarkar et al. 2008). Importantly, as with the other subset-defining markers, KLRG1 and CD127 are functionally not required for the generation of SLEC or MPEC (Goldrath et al. 2002; Gründemann et al. 2010). Vaccination with live attenuated yellow fever vaccine and transplantation of virus-infected organs into seronegative donors have allowed for similar observations in humans. Two weeks after vaccination, comparable numbers of abundantly cycling IL-7Rα^low and granzyme B–expressing cells were detected in the circulation (Miller et al. 2008). Likewise, the early T-cell responses that target human cytomegalovirus (hCMV) in primary-infected kidney transplant recipients were accompanied by a strong expansion of similar cells (Gamadia et al. 2003). Transcriptional profiling of these hCMV-specific cells confirmed that these cells have all the features of the acutely expanded effectors that have been defined in the experimental mouse system (Kaech et al. 2002; Hertoghs et al. 2010).

Whereas the SLEC and MPEC differentiation pathways described in mice provide an accurate model of CD8⁺ αβ T-cell development targeting a primary infection with lymphocytic choriomeningitis virus or the intracellular bacterium Listeria monocytogenes (LM), it could also suggest that if this definition has any bearing on the human situation, the circulation of healthy adults would predominantly hold naive and IL-7Rα^high/KLRG1^low memory cells. Remarkably, however, large populations of CD8⁺ T cells targeting viruses like hCMV or EBV, with an IL-7Rα^low/KLRG1^high SLEC phenotype, can readily be detected in the circulation of immunocompetent asymptomatic individuals in whom viral loads cannot be detected in the circulation (Gamadia et al. 2001a; Appay et al. 2002; Remmerswaal et al. 2019). Moreover, <1% of such cells express Ki-67, a marker of actively cycling cells also in humans, and it was shown that for hCMV-specific cells targeting the same epitopes, clonotypes stably persist for periods spanning more than 5 years (Klarenbeek et al. 2012; Griffiths et al. 2013; Remmerswaal et al. 2019). Therefore, these populations are unlikely to be short lived. While studying primary hCMV infections, various phenotypically distinct effector cell populations can be identified. Human CMV-encoded protein pp65-specific effector cells predominantly have an IL-7Rα^low/KLRG1^low phenotype (previously named early effector cells [EECs] in mice) and an IL-7Rα^low/KLRG1^high SLEC phenotype during the first days after infection. Nevertheless, smaller populations of IL-7Rα^low/KLRG1^high MPECs and IL-7Rα^high/KLRG1^high (double-positive effector cells [DPECs]) could also already be detected at the same time points (Remmerswaal et al. 2019). Interestingly, all four acute phase subsets expressed T-bet, a T-box transcription factor responsible for inducing effector traits in T cells, and granzyme B, a serine protease that induces apoptosis in cells after injection into their cytoplasm by the T cells, denoting them as all having apparent immediate cytotoxic potential (Remmerswaal et al. 2019).

Surprisingly, the acute phase antiviral CD8⁺ αβ T-cell response progresses differently during primary EBV infection (van Aalderen et al. 2015; Remmerswaal et al. 2019). Here, acute phase EBV BamHI Z replication activator (BZLF-1)-specific CD8⁺ αβ T cells predominantly expressed an IL-7Rα^low/KLRG1^high MPEC and IL-7Rα^low/KLRG1^low EEC phenotype. Nevertheless, also here, all possible combinations of IL-7Rα/KLRG1 expression were seen with smaller populations of IL-7Rα^low/KLRG1^high SLEC and IL-7Rα^high/KLRG1^high DPEC cells were also detected during this phase. Furthermore, none of these EBV BZLF-1-specific acute phase cells expressed granzyme B when they emerged, despite a considerable expression of T-bet (Remmerswaal et al. 2019). Instead, virtually all were expressing eomesodermin (Eomes), another T-box transcription factor with significant homology to T-bet, and granzyme K, another serine protease of which the role in combating viral infections is poorly understood (Bovenschen and Kummer 2010).

Hence, it becomes clear that functionally distinct populations can be assigned to the KLRG1-positive SLEC populations. In contradiction to the SLEC/MPEC model nomenclature, CX3CR1⁺CXCR3^–KLRG1⁺ can be long-lived effector cells (Tllec) in both mice and human. These cells are predominantly found in blood and share phenotypic features with (resting) effector cells but also demonstrate functional traits of memory cells (Böttcher et al. 2015). The expression of intermediate levels of CX3CR1 can also be used to define yet another population of memory cells that is referred to as peripheral memory (Tpm) cells in line with their ability to recirculate through peripheral tissues (Gerlach et al. 2016). Interestingly, in the circulation of healthy persons, a small population of IL-7Rα^low/KLRG1^low CD8⁺ T cells is found. These cells are enriched for Ki-67-expressing cells and show at the transcriptional level all features of primary effector cells (Remmerswaal et al. 2019). These findings again urge caution, which is needed when translating an established phenotype into a specific functional trait from one specific experimental situation to another. Apart from the subtle differences that may exist between particular models, more general and difficult-to-tackle issues such as species differences and natural infection into outbred subjects also have to be taken into account.

PROPERTIES OF HUMAN CD8⁺ T CELLS DURING MEMORY RESPONSES

The hallmark of immunological memory is the ability to respond faster and more effectively to an antigen rechallenge. Because T cells mediate local protection, they are equipped with the migratory traits to provide surveillance in the tissues that are at risk for pathogen exposure. As different pathogens apply different infection routes or strategies to breach barrier tissues and evade the immune system, infections by different pathogens lead to the generation of phenotypically and functionally distinct CD8⁺ memory T-cell subsets. Different migratory patterns are classically used to characterize CD8⁺ T-cell populations. It is important to recognize that these definitions do not encompass other functional traits of memory T cells such as effector mechanisms and dysfunctional states. Based on migration traits, we can distinguish CD45RA^– CCR7⁺CD27⁺ central memory T cells that migrate through lymphoid tissues and CD45RA^– CCR7^–CD27^– effector memory T cells that migrate through nonlymphoid tissues and blood. A key mechanism that regulates T-cell migration in blood and lymph are gradients of the egress factor S1P, which can bind to the S1PR1 expressed by virtually all T cells (Schwab and Cyster 2007). Since 2013, it was appreciated that a significant fraction of Tem phenotype cells in nonlymphoid tissues and secondary lymphoid sites are noncirculating and, as such, are largely in disequilibrium with the blood (Sathaliyawala et al. 2013; Thome et al. 2014). These cells were named tissue-resident memory T (Trm) cells and were found to stably express CD69 and/or CD103 and to harbor unique gene expression and metabolic and epigenetic characteristics (Sheridan et al. 2014; Mackay et al. 2016; Milner et al. 2017). Expression of CD69 appeared to be crucial for Trm cells to stay on site as it interferes with the S1PR1 receptor and thereby stalls Trm cells in the tissue (Skon et al. 2013). Expression of CD103, an αE integrin that pairs with β7 integrin to bind epithelial cell–expressed E-cadherin, is found mainly on intraepithelial Trm cells. Although several questions concerning the ontogeny of Trm cells remain unanswered, Trm precursor cells were found to arise simultaneously as the generation of recirculating memory cells (Milner et al. 2017). Hence, infections leave behind a number of Trm cells at initial sites of infection. Evolutionarily, this strategy may have been selected to counterbalance residual or latent pathogens. Additionally, as recirculating memory T cells are largely absent from noninflamed tissues, remaining Trm cells may provide immediate protection, thereby saving precious time to control pathogen invasion and replication. Indeed, in several animal models, Trm cells have been shown to mediate rapid elimination of pathogenic viruses, superior to that of circulating T cells and even in absence of antibodies. Therefore, the generation, through vaccination of epithelial surfaces rather than intramuscularly, is of utmost interest for boosting the efficacy of vaccination regimens (Park and Kupper 2015).

Although the mechanisms that regulate Tn cells differentiation into memory populations with different migratory traits are not fully understood, it became clear that after the initial commitment, following activation of Tn cells, memory T cells can alter their migratory behavior upon antigen reencounter. For example, upon infection, activated recirculating CD8⁺ T cells localize to sites of inflammation via a CXCR3-mediated mechanism (Hu et al. 2011; Sung et al. 2012; Kastenmüller et al. 2013). To reach the infected sites, circulating memory T cells may need to migrate across the endothelium (Verma and Kelleher 2017). Although relatively unexplored, it was demonstrated that this migration induces an LFA-1-mediated signaling cascade in T cells that induces genetic signatures of TGF-β and Notch pathways. Whereas both pathways have been previously linked to the formation and maintenance of Trm cells, it is tempting to speculate transendothelial migration might help imprint the traits of tissue residency (El-Asady et al. 2005; Casey et al. 2012; Hombrink et al. 2016). These tissue-recruitment mechanisms show how infections can mediate the composition of the circulating non-naive T-cell repertoire. It has been long recognized how persisting herpes virus infections with hCMV and EBV affect the circulating CD8⁺ T-cell repertoire (Gamadia et al. 2001a,b, 2004; van Aalderen et al. 2015). The circulation of hCMV-seropositive individuals (concerning ∼60% of the Western adult population) generally contains substantial amounts of Ki-67-negative CD45RA⁺CCR7^–CD28^–CD27^– cells that highly express T-bet and granzyme B, whereas this specific population is virtually absent in hCMV-seronegative individuals (Gamadia et al. 2004; van Aalderen et al. 2015). Instead, the circulation of EBV-seropositive individuals, when compared to EBV- and hCMV-seronegative individuals, generally contains Ki-67-negative CD8⁺ αβ memory T-cell populations expressing considerable amounts of Eomes and granzyme K. Interestingly, hCMV- and EBV-specific memory cells display the same distinctions with regard to these transcription factors and serine proteases as was seen during the acute phase of infection (van Aalderen et al. 2015). Therefore, it appears that these phenotypic and functional distinctions between memory cells in humans are imprinted very early during the acute phase of disease, and are stably maintained during the latency phase when patients are asymptomatic and viral loads are undetectable. Importantly, the size of the CD8⁺ αβ memory T-cell population with immediate cytotoxic capacity (i.e., CD45RA^+/–CCR7^– CD28^–CD27^+/– cells expressing T-bet and granzyme B) correlates positively with the size of the hCMV-specific CD8⁺ αβ memory T-cell pool. This suggests that this memory expansion is related to a progressive loss of markers like CD45RA, CCR7, CD28, and CD27 from the surface of the CD8⁺ αβ T cells.

Phenotypic differences are even seen for CD8⁺ T cells targeting different epitopes on different proteins of the same virus. For instance, CD8⁺ T cells targeting epitopes from the BZLF-1 protein differ significantly from CD8⁺ T cells targeting epitopes from the EBV nuclear antigen (EBNA) proteins EBNA-1 and EBNA-3a (van Aalderen et al. 2015, 2016; Remmerswaal et al. 2019). In this case, BZLF-1-specific CD8⁺ T cells, enriched in the CD45RA^–CD27^– Tem cell subset, express a substantial amount of the T-box transcription factor T-bet and granzyme B (the expression of which is induced directly by T-bet), while EBNA-specific cells express significantly lower amounts of these molecules and are more often expressing IL-7Rα. It is believed that divergent expression kinetics by which viral proteins are expressed during viral infection are driving T-cell heterogeneity. Specifically, EBV-BZLF-1 is expressed early during the lytic phase of infection and EBNA proteins during different latency programs of EBV, and it is likely that these phases are defined not only by differences in the degree of antigenic exposure of the virus-specific T cells, but also by differences in anatomical niches, viral replication sites, and inflammatory milieus. Similar phenotypic divergence is observed when comparing CD8⁺ T-cell responses to hCMV-derived immediate early 1 (IE-1), believed to have a decisive role in acute infection in fibroblasts, and structural phosphoprotein pp65, maximally expressed after the initiation of viral DNA replication (Khan et al. 2002). The association between infection route and T-cell phenotypes and function is largely explored in mice models. To illustrate, both intranasal and intraperitoneal infection with influenza in mice give rise to influenza-specific CD8⁺ T cells with similar antigen specificities. However, there is good evidence that CD8⁺ Trm cells, specifically induced after intranasal infection, can provide superior protection against subsequent infection with influenza, although both routes of infection efficiently produced influenza-specific splenic Tem cells (Wu et al. 2014). Similar results were obtained when comparing oral versus intravenously models of Listeria infection in mice (Sheridan et al. 2014).

After primary infections are cleared, memory precursors differentiate toward memory cells exalting states of poised effector function regulated by transcriptional and epigenetic circuits, allowing a rapid recall response upon antigen reencounter. These states have proven to be fluid as during chronic infections and cancer, as a consequence of repeated antigen exposure, antigen-experienced CD8⁺ T cells get functionally restrained. Different states of functional impairment can be recognized. While senescent cells expressing CD57 and CD85j, whose abundance increased with age, lost their capacity to proliferate but retained their polyfunctional cytokine production and expression of cytotoxic mediators, T cells classically referred to as exhausted (Tex), are characterized by hierarchical loss of cytokine production and impaired proliferative potential (Wherry and Kurachi 2015; Alpert et al. 2019; Gorgoulis et al. 2019). Although broadly accepted, the reference to exhaustion, meaning inert, appeared to be misleading as it is now appreciated that Tex cells have altered functionality to prevent immunopathology (Blank et al. 2019). For example, part of the tumor Tex cell population has been shown to produce the chemokine CXCL13, required for tertiary lymphoid structure formation, generally associated with favorable clinical outcome in cancer (Sautès-Fridman et al. 2019). Furthermore, rather than a functional state, based on the big difference in epigenetic states, Tex cells have been postulated to be a distinct T-cell subset from effector and memory T-cell populations with a yet elusive ontogeny (Philip et al. 2017; Khan et al. 2019). Interestingly, progenitor Tex cells have been shown to express CXCR5, a marker used to define CD45RA⁺CD27⁺ follicular-like CD8⁺ T cells (Tfl) that exert strong tumor activity (Brummelman et al. 2018). As CXCR5⁺CD8⁺ Tfl cells are generally identified within the B-cell follicle and germinal center, the functional role of Tfl cells may largely depend on the immune context. Phenotypically, Tex cells can be uniformly characterized by the expression of the inhibitory receptor programmed cell death 1 (PD-1); large sections of Tex cells also coexpress inhibitory receptors including TIM3, LAG-3, CTLA-4, and TIGIT. Caution should be used when using PD-1 strictly to define Tex cells, as expression of this inhibitory marker is induced by TCR signaling. As such, recently activated T cells can express PD-1 (Yokosuka et al. 2012; Pauken and Wherry 2015). In addition, PD-1 expression is also a common feature of human Trm cells that retain their poised effector state (Kumar et al. 2017), showing the definition of Tex cells by PD-1 alone to be inadequate. Recently, functional heterogeneity was shown within the Tex cell population. Whereas expression of the transcription factor TCF1 defines Tex cells with self-renewing capacities, those lacking TCF1 are considered to be terminally differentiated exhausted T cells (Im et al. 2016; Siddiqui et al. 2019). In addition, in mice, it was demonstrated that the transcription factor TOX is a critical regulator of the transcriptional and epigenetic programming of Tex cells and that in the absence of TOX, Tex cells do not form and TCF1⁺ CD8 T cells are poorly maintained (Alfei et al. 2019; Khan et al. 2019). It remains to be elucidated whether similar heterogeneity applies to the human Tex cell population.

CAN WE MAKE PROGRESS IN CHARACTERIZING CLINICALLY RELEVANT SUBSETS OF T CELLS?

The key findings summarized in the previous paragraphs allow a couple of conclusions. First, subdivisions made by the currently available marker sets may be appropriate when scrutinizing the global differentiation state of circulating CD8⁺ T cells but may fall short when zooming in on small yet important subsets of cells. This lack of error-free phenotyping may be unfortunately feeding the debate about the functional traits and ontogeny of antigen-experienced T-cell populations. In addition, while several definitions of CD8⁺ T-cell populations have been dependent on immunological context, characterizations based on phenotypes alone fall short. Second, the currently employed experimental infection models in specific pathogen-free (SPF)-housed mice have produced invaluable information on the sequences of immune reaction and importantly on the key transcriptional circuits involved. Still, when they are used to depict the fine details of CD8⁺ T-cell responses in outbred humans that are frequently challenged by specific pathogens they may yield an oversimplified description of the (patho)physiology. Third, viruses differ in the dominant responses they evoke. These differences are related to tropism, evolutionary balanced escape of the immune system (like for herpes viruses), and clearance versus latency. Still, whatever the dominant mechanism, it may be submitted that different antiviral defense programs all share traits needed for a general functionality of T cells, but also each comprise specific adaptations needed to better control individual viruses.

The emergence of techniques that allow transcriptional and epigenetic interrogation of cells at single-cell levels now provides an opportunity to obtain an unbiased view on both naive and antigen-experienced CD8⁺ cells. This type of analysis could lead to a “minimal” fingerprint definition of the naive pool, including other markers and traits than the ones that we—some of them by chance—have learned to work with. Likewise, these techniques would be truly helpful to appreciate the complexity of the antigen-primed pool. Likewise, transcriptional and epigenetic fingerprints may be generated that can improve the resolution for identifying T-cell populations with more explicit functional traits.