During acute antigenic exposure, such as acute viral infections or vaccination, naïve CD8⁺ T cells are activated by professional antigen presenting cells carrying cognate antigenic peptide/MHC-I complex in secondary lymphoid organs. Activated T cells undergo massive proliferation and further differentiate into effector CD8⁺ T cells with profound alterations in effector molecule production, migration pattern, transcriptional network, metabolic program, and epigenetic landscape. Shortly after antigen clearance, effector T cells undergo contraction and further differentiation into long-lived memory T cells with superior recall capacity (1). However, when antigen presence is prolonged (such as chronic infection and tumor), effector T cells rapidly turn to a different path towards exhaustion with greatly reduced effector function and population size. In recent decades, reviving exhausted T cells have been established as one of the common goals in tumor immunotherapies (2). Thus, it is essential to advance our understanding of exhausted T cells. Here, we will summarize recent findings related to the migration and tissue residency of a subset of exhausted T cells expressing transcription factor Tcf-1 (T cell factor-1).

Shortly after the discovery of T cell exhaustion, it has been realized that exhausted T cells are not homogenous. Instead, a broad spectrum of T cell subsets together constitute exhausted T cell population. Initially, different exhausted T cell subsets were distinguished by various levels or composition of inhibitory receptors (3, 4). Later, different exhausted T cell subsets with various expression of T-box transcription factors T-bet and Eomes levels were discovered (5). More recently, Tcf-1⁺ subset of exhausted T cells has been defined as the progenitor or stem-like subset to sustain the whole exhausted T cell population (6–10) ( Box 1 ). Tcf-1⁺ cells further differentiate into transitional subsets (e.g., CX3CR1⁺ cells) as well as terminally exhausted cells (e.g., CD101⁺ cells) (11–13). Most importantly, Tcf-1⁺ subset is the one responding to PD-1 or PD-L1 blockade in both chronic viral infection and tumor settings (7, 9, 14). Further, Tcf-1⁺ exhausted T cells are the main target of therapeutic tumor vaccines (15). Together, Tcf-1⁺PD-1⁺ stem-like or progenitor exhausted T cells are the key CD8⁺ subset which can be self-sustained and further differentiate into other exhausted T cell populations.

Interestingly, a recent paper has further defined a subset of Tcf-1⁺PD-1⁺ T cells carrying T_CM (central memory T cells) marker CD62L and CCR7, which is highly enriched for stem-cell or progenitor activity and largely responsible for PD-1 blockade induced T cell expansion (16). These Tcf-1⁺PD-1⁺CD62L⁺ stem-like CD8⁺ T cells is critically dependent on transcription factor Myb, which reminds us about a similar Myb-dependent CD8⁺ T_CM subset generated after acute viral infection (17). Accumulating evidence has documented the similarity between T_CM cells generated after acute infection (18) and stem-like exhausted T cells during chronic antigen exposure, especially regarding the Tcf-1-dependent genetic signature. At molecular level, it has been recently demonstrated that during memory T cell recall responses, there are a large collection of immediate responsive genes, including glycolytic enzymes, cell cycle controllers and transcriptional regulators. Tcf-1 is essential to keep these genes ready for future recall response via maintaining their 3D genomic interaction with distal enhancers (19). Consistent with this role of Tcf-1 in memory T cells, Tcf-1 and closely related transcription factor Lef-1 controls the 3D structure and crosstalk between distal genomic elements partially via interacting with CTCF (CCCTC-binding factor) in naïve T cells (20, 21). Thus, it is safe to conclude that Tcf-1-control transcription and epigenetic programs represents one of the central themes for most T cells (e.g., naïve, T_CM and stem-like) with robust expansion and differentiation capacity.

Based on migration pattern, acute antigen exposure induced memory T cells can be categorized into central memory (T_CM), effector memory (T_EM) and tissue resident memory (T_RM) T cells (22) (Box1). Because of the broad TCR repertoire, the frequency of T cells bearing TCR with a given specificity is often extremely low. To efficiently protect the whole body against potential antigenic evasion, continuous migration and patrolling for cognate antigen appearance is a build-in feature of T cell biology. Thus, the very existence of T_RM cells, which are largely separated from the circulation at steady states and confined to a specific tissue represents an intriguing “outlier”. Numerous efforts have been devoted to investigating the differentiation, molecular regulation and function of T_RM cells (23). T_RM cells are direct decedent of effector T cells. They often strategically located at previous pathogen entering sites or peripheral tissues experienced local inflammation and damage. In adult human and immunized animals, T_RM cells can be detected in most non-lymphoid tissues, including both mucosal and non-mucosal sites as well as the tissues which have been traditionally considered as immune-privileged sites (24–26). Number wise, T_RM represents the most abundant T cell population in most antigen-experience individuals.

Several local signals are actively involved in T_RM differentiation. For example, TNF (tumor necrosis factor), IL-33, extracellular ATP and local ICOS signals can promotes T_RM formation (27–30). Here, we will limit our discussion to two of the most well-studied signals for T_RM differentiation. First, we will focus on TGF-β (transforming growth factor-β), which is cytokine essential for CD103 (encoded by Itgae) induction on activated CD8⁺ T cells. CD103 is a commonly used marker for mucosal T_RMs and critically involved in mucosal T_RM retention via interaction with its ligand E-cadherin (31–33). It is well established that TGF-β signal delivered to CD8⁺ T cells is broadly required for T_RM differentiation, including most mucosal T_RM with CD103 expression (31–34) and some non-mucosal T_RM lacking CD103 (35). Further, continuous TGF-β signal is required for long-term maintenance of T_RM cells in both skin and intestine (36, 37). Interestingly, dendritic cells deliver basal TGF-β to naïve T cells inside secondary lymphoid organs. This basal TGF-β signaling before T cell activation will keep naïve T cells semi-ready for later priming towards T_RM differentiation (38). Thus, during T_RM differentiation and maintenance, TGF-β signal is required at different locations and different stages. Similar to most dogmas in biology, the requirement for TGF-β in T_RM is not universal. Prominent exceptions do exist, i.e., T_RMs isolated from upper respiratory tract and liver are formed independent of TGF-β signal following acute infection (39, 40). It is interesting to note that some TGF-β-dependent T_RM population carry higher levels of inhibitor receptor PD-1 expression (39, 41).

The second signal we would like to discuss here is local antigen. As T_RM is often formed at the site of local infection, which is likely associated with enhanced local antigen presentation. T_RM induction in the brain, the sensory ganglia, the lung and the cornea requires local antigen recognition (24, 25, 42, 43). However, local antigen is not essential for T_RM formation in the skin, the gut and the female reproductive tract (43–45). For example, chemical-induced local sterile inflammation can effectively attract in vitro activated CD8⁺ T cells to form skin T_RM, which is a commonly used and convenient technique in T_RM field (43). However, even for skin T_RMs, local antigen significantly boosts their formation (46, 47). After T_RM formation, it is generally believed that long-term maintenance of T_RM is TCR-independent, which is first demonstrated in skin-resident γδT cells (48), later confirmed in both CD8⁺ and CD4⁺ T_RMs (49–51). Together, although not universally required, TGF-β and local antigen often promote initial T_RM formation. For long-term T_RM maintenance, TGF-β is likely involved while antigen is not required.

Although initial CD8⁺ T_RM research was largely focused on non-lymphoid tissues, it was quickly realized that T_RM could form inside secondary lymphoid organs [i.e., spleen and lymph nodes (LN)] after systemic viral infection although the population size was small (52). In systemic LCMV (lymphocytic choriomeningitis virus) infection model, lymphoid organ CD8⁺ T_RM does not express CD103. They carry typical T_RM markers CD69⁺Ly6C^-CD62L^- and core T_RM gene signature. Importantly, these secondary lymphoid organ T_RMs are not migratory as demonstrated in parabiosis experiments (53). They are direct derivative of upstream non-lymphoid tissue T_RM. In other words, non-lymphoid tissue T_RM re-activation leads to robust T_RM accumulation inside draining LNs (53). Consistently, pet store mice or “dirty” mice with a complicated exposure history to a broad collection of environmental pathogens carried significantly increased T_RM population in secondary lymphoid organs (53). In local influenza virus infection model, a significant population of CD69⁺CD103⁺CD8⁺ T_RM subset can be identified in lung draining LNs (54–56). Repetitive infection promotes LN T_RMs (54) and CD8⁺ T cells carrying different TCR specificity exhibit distinct LN T_RM potential (56), suggesting a possible role of antigen in LN T_RM formation. However, antigen is not required for LN T_RM maintenance (55). Similar to systemic LCMV infection, LN T_RM is generated via retrograde migration from upstream lung T_RMs during influenza viral infection (55). Functionally, these draining LN T_RM may represent an expanded local defense to reinforce the first line of T_RM-dependent immunity at the upstream non-lymphoid tissues.

Interestingly, a large number of memory CD8⁺ T cells in human LNs and spleen carry typical T_RM markers CD69 and CD103 (57). In addition, a CD69⁺CD103⁺ CD8⁺ T cell subset has been identified in human tonsil and specific for Epstein Barr Virus (EBV) (58). The identity, migration and function of these human T cells remains a mystery. Based on the observation in mice (especially the results from dirty mice), it is conceivable that these CD69⁺CD103⁺ CD8⁺ T cells in human secondary lymphoid organs may contain a significant T_RM subset. Thus, CD8⁺ T_RM can form inside secondary lymphoid organs in both mouse and human. In mouse acute infection models, these LN T_RMs are derived from upstream non-lymphoid tissue T_RMs. In other words, they may have a travel history to periphery tissues before settling down in the draining LNs.

In the original papers that discovered Tcf-1⁺PD-1⁺ subset during chronic LCMV infection, a few interesting features of Tcf-1⁺ stem-like T cells emerged. First, they are largely located inside secondary lymphoid organs (LNs or splenic lymphoid follicles). Second, they are almost absent in the peripheral blood (7). Demonstrated via parabiosis experiments, most Tcf-1⁺PD-1⁺ T cells are tissue-resident and largely separated from the circulation after the establishment of chronic LCMV infection (59). Incorporating T_RM marker CD69, both Tcf-1⁺ stem-like and Tcf-1^- effector subsets can be further divided into CD69⁺ and CD69^- populations. Importantly, both Tcf-1⁺CD69⁺ and Tcf-1^-CD69⁺ subsets are excluded from the circulation and negatively enriched for circulating T cell gene signature (60). Tcf-1⁺CD69⁺ cells are largely located inside lymphoid follicles while Tcf-1^-CD69⁺ ones are splenic red pulp-resident (60). These results demonstrate that during systemic chronic viral infection, a significant portion of exhausted CD8⁺ T cells acquire certain features of T_RM inside lymphoid organs. Based on these findings, it will be interesting to address the questions why Tcf-1⁺ stem-like CD8⁺ T cells prefers a lymphoid environment and whether the lymphoid-residency is functionally important for stem-like T cell differentiation or maintenance.

A series of recent findings focused on chemokine receptor CXCR3 have shed light on these critical questions. Using either acute (61) or chronic LCMV infection model (62), it has been demonstrated that CXCR3 is essential for the differentiation from Tcf-1⁺ stem-like to Tcf-1^- effector T cells. In the absence of CXCR3, there is an increased accumulation of Tcf-1⁺ subset inside secondary lymphoid organs. There are two ligands for CXCR3 in C57BL/6 mice, namely CXCL9 and CXCL10. Interestingly, CXCL9 producing cells (e.g., XCR1⁺ cDC1) are concentrated inside T cell zone while CXCL10 producing cells (e.g., conventional Dendritic Cell 2, or cDC2 and inflammatory monocytes) are mainly outside T cell zone. Thus, it is mainly via CXCL10/CXCR3 interaction to attract Tcf-1⁺ T cells to move out of T cell zone (61, 62). These findings have been validated in a different chronic parasite infection model (i.e., Toxoplasma gondii, or T. gondii infection) (63). During T. gondii infection, Tcf-1⁺ CD8⁺ T cells expressing high levels of CXCR3. In responding to CXCL10, these stem-like T cells migrate out of lymphoid follicles and form clusters with cDC2 in the bridging channels of spleen. Importantly, these T. gondii-specific Tcf-1⁺CD8⁺ T cells isolated from the spleen carry a typical T_RM phenotype (i.e., Cd69⁺Itgae⁺Klf2^-S1pr1^-S1pr5 ^-) although this result is from RNA-seq, not confirmed at protein levels (63). Together, these investigations on CXCR3 and CXCL10 provide us an excellent example that the lymphoid location of Tcf-1⁺ stem-like T cells is tightly associated with their maintenance ( Figure 1 ). Leaving lymphoid environment is accompanied by immediate effector differentiation.

Another key signal delivered to stem-like T cells is TGF-β. Although TGF-β is often considered as a cytokine with broad distribution, Tcf-1⁺ stem-like T cells carry TGF-β activating integrin (αvβ8) to keep a TGF-β-rich microenvironment around themselves (64). TGF-β is produced as inactive latent form. Active TGF-β has an extremely low solubility at neutral pH and therefore active TGF-β is likely to have a very short functional distance. Thus, local TGF-β-activating mechanisms (e.g., αvβ8 integrin) are essential for TGF-β function in vivo.

The function of TGF-β on stem-like T cells is multifaceted ( Figure 2 ). First, TGF-β restrains mTOR (Mammalian Target of Rapamycin) activity in stem-like T cells to maintain their long-term responsiveness (64). Second, TGF-β directly suppress the differentiation of CX3CR1⁺ effector T cells and promotes the formation of CD101⁺ terminally exhausted T cells (64, 66, 67). Importantly, the impacts of TGF-β are significantly enhanced during the later stages of chronic infection (66). Finally, we have demonstrated that TGF-β suppresses Tcf-1⁺➔CX3CR1⁺ differentiation partially via enforcing their lymphoid tissue residency. In the absence of TGF-β receptor, stem-like T cells exhibited defective lymphoid tissue retention, which is associated with further effector differentiation. Forcing TGF-βR deficient stem-like T cells to stay inside lymphoid follicles via integrin α4 blocking partially corrects the defects. This result suggests that manipulating the location of Tcf-1⁺ T cells alone is sufficient to control their differentiation (67).

In tumor settings, Tcf-1⁺PD-1⁺ cells are initially identified among tumor infiltrating lymphocytes (TIL), which is out of a secondary lymphoid organ. It is later discovered that a lymphoid-like microenvironment exists inside solid tumors to host Tcf-1⁺ stem-like T cell subset and physically separates them from tumor cells (68, 69).

Recent results have established cDC1-delivered tumor antigen is critical to establish tumor draining LNs as a reservoir of Tcf-1⁺ T cells and to sustain anti-tumor immunity (70, 71). Our recent work has revealed that tumor draining LN (TDLN) harbors a large population of Tcf-1⁺ CD8⁺ T cells with a CD69⁺CD103⁺ T_RM phenotype (72). The differentiation of T_RM-Tcf-1⁺ T cells requires both TGF-β signaling and tumor antigen. Tumor vaccine, especially vaccine adjuvant promotes the differentiation from T_RM to non-T_RM in a type I IFN-controlled way. This result is consistent with the finding in an acute viral infection model, where type I IFN suppresses T_RM formation (35). The loss of T_RM feature is critical for the active migration of stem-like T cells from TDLN to tumor site to control tumor growth. In addition, the loss of T_RM identity may represent the first step of CX3CR1⁺ effector T cell differentiation. Another key finding is that Tcf-1⁺ CD8⁺ T cells gradually differentiate into T_RM inside TDLNs, i.e., the appearance of T_RM-Tcf-1⁺ cells is significantly delayed comparing with that of Tcf-1⁺ cells in TDLNs. Only large tumor TDLN carries a significant population of T_RM-stem CD8⁺ T cells. This finding likely explains the discrepancy between our results and most previous animal research focusing on early-stage tumor (i.e., when tumor is palpable). For example, in contrast to the lack of efficacy in our hands for large tumors, tumor vaccine is often effective when given early (15). Using photoconvertible mice, Tcf-1⁺ T cell migration between tumor and TDLN can be easily identified in early-stage tumor (when tumor size is small) (73). It is possible that similar to retrograde migration in acute infection settings, TDLN T_RM-stem CD8⁺ T cells are derived from tumor infiltrating T cells although this idea has not been tested experimentally. Considering all these results, we believe that tumor-specific Tcf-1⁺CD8⁺ T cells accumulate inside TDLNs and gradually differentiate into T_RM-stem and lose migratory capacity when tumor reaches a certain size ( Figure 3 ). TGF-β and tumor antigen promote, while type I IFN inhibits the establishment of T_RM-Tcf-1⁺ cells in TDLNs. It is conceivable that most cancer patients carry large tumors and likely harbor a significant portion on T_RM-stem in TDLNs. The migration from TDLNs to tumor is essential for CD8⁺ T cells to directly attack solid tumors. Thus, targeting T_RM-stem in TDLN and mobilizing TDLN stem-like CD8⁺ T cells will be one of the keys to boost tumor immunotherapies, including tumor vaccines.

Together, lymphoid residency is an essential component of Tcf-1⁺ exhausted T cells in both chronic viral infection and tumor immunity. The regulation of lymphoid residency for Tcf-1⁺ T cells is critical to control effector differentiation and is an essential speed-limiting step for tumor vaccine response. However, T_RM is not the only fate for lymphoid Tcf-1⁺ exhausted CD8⁺ T cells. A significant portion of lymphoid Tcf-1⁺ CD8⁺ T cells does not differentiate into T_RM. The regulation of T_RM vs non-T_RM Tcf-1⁺ T cells under different tumor immunotherapy settings remains unknown. The lineage relationship between T_RM-Tcf-1⁺ vs non-T_RM-Tcf-1⁺ cells is unclear. Importantly, whether T_RM-Tcf-1⁺ T cells are critically involved in all chronic antigen exposure settings awaits future investigation. For example, in an autoimmune diabetes setting, pancreas draining LN Tcf-1⁺ CD8⁺ T cells do not carry enhanced CD69 and express high levels of Klf2 (74), which is associated with circulating T cells (75). Similarly, in a melanoma and autoimmune vitiligo setting, LN Tcf-1⁺ T cells express high levels of Klf2 and Tcf-1^- LN effector T cells become T_RM (76). Thus, it is possible that a unique mechanism exists to keep autoimmune-induced Tcf-1⁺ CD8⁺ T cells as circulating cells in lymphoid organs. Nevertheless, recent publications have highlighted the importance of the lymphoid location of Tcf-1⁺ T cells. Better understanding the control of residency vs migration of Tcf-1⁺ T cells represents one of the keys to advance our knowledge of Tcf-1⁺PD-1⁺ T cell biology and facilitate the future design of T cell-based immunotherapies.

CM and NZ researched, wrote and edited the manuscript. All authors contributed to the article and approved the submitted version.