Integrating resident memory into T cell differentiation models

Abstract

Advances in the field of T cell memory, including the discovery of tissue residency, continue to add to the list of defined T cell subsets. Here, we briefly review the role of resident memory T cells (T_RM) in protective immunity, and propose that they exhibit developmental and migrational plasticity. We discuss T cell classification, the concept of cell type versus ‘subset’, and the difficulty of inferring developmental relationships between cells occupying malleable differentiation states. We propose that popular subset strategies do not perfectly define boundaries of developmental potential. We integrate T_RM into a ‘terrace’ model that classifies memory T cells along a continuous axis of decreasing developmental potential. This model also segregates cells on the basis of migration properties, although different migration properties are viewed as parallel differentiation states that may be permissive to change.

Introduction

“We must be clear that when it comes to atoms, language can be used only as in poetry. The poet, too, is not nearly so concerned with describing facts as with creating images and establishing mental connections”

-Niels Bohr

Primary T cell responses are “inside out”

The adaptive immune system must balance massive clonal diversity in the naïve lymphocyte repertoire with surveillance efficiency. T cells are tactile, and probe for antigen on the surface of cells. To increase the expediency by which very rare (naïve) T clones detect cognate antigen, they limit their surveillance to secondary lymphoid organs (SLOs) rather than explore the entire organism. Recognition of antigen within SLOs (typically internal sites that drain body surfaces) results in priming and activation of small populations of naïve T cells. Primed naïve T cells undergo massive clonal expansion. These now-activated T cells radiate out to solid organs and barrier sites where they survey the tissue, recognize infected cells, and execute effector functions that contribute to the restoration of homeostasis. This could be considered an inside (originating in draining LNs) out (then migrating to peripheral tissues) immune response (Figure 1A).

Figure 1. The anatomic topology of primary and recall T cell responses.

A) During a primary infection, antigen drains from peripheral tissues to draining lymph nodes (LN) which activates naïve T cells specific for that pathogen. Cells expand in lymphoid tissue and migrate out to peripheral tissues to control the infection (which could be referred to as an inside-out immune response). B) At steady-state, memory T cells patrol for reinfection and are heterogeneous, consisting of resident cells that are absent from circulation (T_RM), cells that circulate through blood and lymph, and cells that recirculate though tissue using blood and lymph as conduits. C) Following secondary exposure to antigen, T_RM can proliferate, give rise to an expanded local population of long-lived progeny, redistribute to draining LNs where they may remain resident, and possibly rejoin the circulation. T_RM redistribution from the periphery to LNs and the circulating pool could be referred to as an outside-in immune response. Not pictured: Reactivation of LN circulating memory T cells (i.e., T_CM) recapitulates an inside-out immune response. D) Upon sensing cognate antigen within tissue, T_RM reactivate and alert the local immune system of a reinfection through chemokine and cytokine production. This leads to upregulation of interferon stimulated genes (ISGs), maturation of DCs, activation of T cells and NK cells, adhesion molecule upregulation and recruitment of CD8+ T cells and B cells. T_RM also proliferate and upregulate cytotoxic molecules, likely contributing to killing of infected cells.

Evidence that resident memory T cells mediate “outside in” responses

Following resolution of infection, the host is left with a stable pool of memory T cells that patrol the body for reinfection. Such memory T cells are orders of magnitude more numerous than the naïve cells they differentiated from and can respond to antigen rapidly. Memory T cells represent a heterogeneous population of cells that circulate through blood and lymph [1,2], cells that extend this recirculation through non-lymphoid tissues (NLTs) [3–5], and cells that reside within NLT and, at steady-state, infrequently re-enter the circulation (Figure 1B). The latter are collectively referred to as resident memory T cells (T_RM) [6,7]. T_RM are remarkably abundant and broadly distributed throughout most of the body including internal organs, skin, and the mucosal barriers that form the most common sites of pathogen entry [8–11]. The positioning of T_RM in frontline tissues allows them to be first responders in the event of reinfection.

T_RM that encounter cognate antigen rapidly alert neighboring cells of a reinfection event, which has been termed a “sensing and alarm” function [12,13]. Within hours of antigen sensing, T_RM alarm the immune system through the release of proinflammatory cytokines, establishing an antiviral state locally within the tissue, activating NK cells and T cells, promoting dendritic cell maturation and recruiting circulating innate and adaptive immune cells into the tissue (Figure 1D). These rapid T_RM-mediated responses can accelerate protection against reinfection and in many cases are sufficient for protection without the contribution of circulating memory T cells [14–16]. T_RM may also contribute to protection by killing infected cells directly, and are important for tumor immunosurveillance [17–19] as well as certain autoimmune [20–22], allergic, and inflammatory pathologies [23–25].

T_RM arise from recently activated T cells that migrate to NLTs [26], and are thought to result from an inductive differentiation program that depends on tissue-derived developmental cues [27–30]. T_RM differentiation is associated with downregulation of CD62L, CCR7, and KLF2 (a transcription factor that regulates recirculation) [31], expression of Hobit, Blimp1 [32], and/or Runx3 [33], coupled with cell surface expression of CD69 and sometimes CD103 (which may contribute to retention within tissues), although these are imperfect markers [8,26,34,35]. Indeed, a recent report referred to CD4+ ‘T_RM’ that recirculated between blood and nonlymphoid tissues in the steady state (violating the definition of residence), yet expressed phenotypic signatures in common with T_RM [36].

T_RM have been reported to express markers, transcripts, and functions that are shared with effector and exhausted T cells, supporting one view that T_RM are terminally differentiated cells; unable to mount recall responses that give rise to an amplified resident memory population, or differentiate into other subsets [37–39]. In contrast to this view, recent and ongoing work reveals that T_RM are able to drive autonomous expansion of their population in response to antigen [38,40,41], can relocate to SLOs [42], and even stably re-enter circulation where they may transmogrify into memory T cells with different properties (R. Fonseca et al., submitted) [43]. This can be likened to an ‘outside-in’ topology; recall responses can be initiated at barrier sites (on the outside of the body) where T_RM will proliferate, differentiate, migrate, and, after resolution of infection, may stably populate other tissue compartments including blood and lymphoid tissue (inside the body) (Figure 1C). In the event of systemic reactivation, we propose that those ‘Ex-T_RM’ that have re-joined the circulating pool may remain poised for preferential homing back to previous sites of residence and reacquisition of a stable T_RM program.

With advancing technologies, it has become apparent that there is considerable heterogeneity among memory T cells, including amongst those that share the property of residence [39,44,45]. This heterogeneity has encumbered the historical practice of delineating the diversity of memory into a few discrete and monolithic subsets. In this review, we will address what it means to be a subset, discuss common terminology through the lens of our emerging understanding of memory CD8+ T cell heterogeneity and plasticity, and propose where to integrate T_RM into the broader lexicon and conceptualization of the T cell lineage.

Memory T cell subsetting: Developmental relationships

Our language concerning memory T cells often applies terms (and associated concepts) from developmental biology. Cellular differentiation is a progression of specification and determination events that result in discrete cell types that have reduced developmental potential. A cell lineage traces this differentiation history of a cell or tissue and ‘lineage commitment’ is the developmental point where a cell is irreversibly committed to becoming a certain cell type [46]. The T cell lineage, as with other cell types, ultimately begins with a single zygote, followed by iterative cell divisions and differentiation. Eventually, hematopoiesis gives rise to many lineage-committed cells of the immune system, including T cells. Thus, naïve T cells are the result of many cell divisions and differentiation steps (Figure 2). T cells themselves branch developmentally into clear lineages that do not typically transdifferentiate. Branches may include commitment to αβTCR vs γδTCR expression and/or single positive CD4 vs. CD8 expression. When mature naïve T cells recognize antigen, that likely also initiates an irreversible differentiation process coupled with division, that results in a large population of activated cells; many of which will die and some of which will further develop into memory cells. Important questions arise when we consider the possible branch points, or lack thereof, in this differentiation path following activation of naïve T cells. The way we conceptualize this process is intertwined with how we define subsets of cells and infer their developmental relationship.

Figure 2. Memory T cell lineage.

Memory T cells are a product of a series of lineage specification events that define cell type. These commitments become fixed under physiologic conditions and cells are unable to transdifferentiate or dedifferentiate (e.g. B cells do not become T cells). We argue that memory T cell subsets should not be viewed as a few discrete and distinct cell types that occupy clear steps in a linear differentiation path, but rather comprise a heterogeneous constellation of cells. Hematopoietic stem cell (HSC); Common myeloid progenitor (CMP); Common lymphoid progenitor (CL); Lymph node (LN).

Naïve T cells differentiate into ‘effector’ cells. Effector T cells have a somewhat ambiguous definition; sometimes being defined based on a functional property such as cytolysis or antiviral cytokine production, sometimes being defined as a cell that will soon die, and sometimes being defined as a cell that was ‘recently’ activated. The definition of memory is not particularly precise either, but usually invokes the idea that a significant period of time has elapsed since the naïve T cell was primed (although even this vague definition could be considered controversial). Superimposed on these vagaries, memory T cells are dissected into multiple discrete subsets, on the basis of phenotype, function, migration, perceived cell fate, or combinations of properties. There is obvious importance in being able to categorize cells into groups with shared properties, and subsetting imputes a division of labor among T cell varieties best suited to different desirable biological outcomes. Accordingly, vaccines that favor ‘central’ vs. ‘effector’ memory may have different efficacies that are contextual, rejuvenation of the ‘exhausted’ T cell subset may aid in clearing chronic infections or cancer [47,48], and CAR T cells might perform optimally if they comprised stem cell memory [49–52]. This has inspired a need to not only classify memory, but also to understand the developmental processes that contribute to the ontogeny of one subset or another.

But, how should we think about memory T cell subsets? To what extent does subsetting comprise an extension of irreversible developmental decisions or lineage commitments? Do some subsets give rise to others, but not the other way around? Or, could some subsets represent alternative differentiation states that are interconvertible? Are the qualitative axes by which subsets have been delineated robustly capturing differences in developmental relationships? Or rather, are they just highlighting functional differences of interest, without being founded on a more traditional developmental strategy for classification? Given recent advances in the field, has an opportunity been presented to restructure memory T cell classification, or to at least be more explicit about the definitions of each class and acknowledge when development principles are being eschewed in favor of (sometimes loosely defined) properties?

Criteria for subsetting and ontogeny models

There are a number of properties by which memory T cells have been Balkanized into subsets, and this process has carried implications, both semantic and conceptual. Firstly, subsets imply largely discrete homogenous entities (Figure 3A). Moreover, subset ontogeny has been diffracted through the prism of developmental biology, which imputes unidirectional lineage relationships and progressive loss of plasticity, and there has been a temptation to insert newly defined subsets into an existing lineage hierarchy. Confounding understanding, many subsets have been proposed, though not always clearly defined, with some gaining more traction than others. Often, a designation is based on a conveniently measurable cell-surface marker or transcription factor. These markers may imperfectly correlate with diverse and difficult-to-measure properties that relate more closely to function (e.g. what cytokines are made, trafficking patterns, proliferative capacity, developmental plasticity, etc.). Soon, the marker, and an associated subset name, becomes a synonym or substitute for the biology. In addition, the semantics (e.g. central vs. effector memory) can belie the actual properties (central memory T cells make effector cytokines while ‘effete’ might be a better label for some effector memory T cells) [53]. These problems are not unfamiliar to anyone working in the field of T cell memory and can be especially vexing to those in other fields [54,55].

Figure 3. Models of T cell differentiation.

A) Defined subsets of memory T cells imply unidirectional differentiation between discrete cell types. B) Initial modeling of electron dynamics implied discrete orbits of movement (left), whereas the modern view highlights a probabilistic distribution, represented as an electron cloud (right). C) The ‘terrace’ model depicts naïve cell activation (1) and differentiation into memory T cells (2) along a continuum of developmental potential (depicted as terraces), though most activated T cells (depicted in the river) die (3). Migration status is used to further bifurcate memory T cells into a resident and circulating pool. This migration status is largely fixed, though may not be permanent for all cells (4). Subset nomenclature of circulating memory T cells is estimated, but does not reflect discrete cell types with stepwise loss of differentiation potential (5). The ‘terrace’ model is based on developmental potential and migration properties. Stem cell memory T cells (T_SCM); Central memory T cells (T_CM); Effector memory T cells (T_EM); Exhausted T cell (T_EX).

It bears consideration whether immunology should borrow a page from physics. Niels Bohr elegantly visualized electrons as occupying discrete orbits, which advanced the field in its time and enabled further discoveries. However, the Bohr model had limitations, as many models do, and it was superseded by Erwin Schrodinger’s proposal that electrons fall within a probabilistic distribution that only approximate Bohr’s orbits, which was conceptualized as the ‘electron cloud’ (Figure 3B). T cell subsets may be fuzzy at the margins, or even represent a continuum of differentiation states upon which borders (and labels) have been roughly assigned.

We propose a model of T cell differentiation that is likened to terraced farming and based on developmental potential and migration (residence or circulation), while acknowledging that the current subset nomenclature may not be synonymous with these criteria (Figure 3C). Indeed, this model embraces the fact that simple subset designations will not capture discrete cell types. In the terrace model, naïve T cells can differentiate into diverse memory T cells that comprise a spectrum of developmental potentials. Unlike naïve T cells, memory T cells have passed through a stage of activation (symbolized as being in a river). Most activated T cells rapidly lose all developmental potential, and die (falling off the waterfall) without acquiring a stable memory differentiation state (occupying a flat terrace). It should be noted that activated T cells that will become memory cells may meet definitions of transiently being an ‘effector’ because they express anti-pathogen molecules and/or acquire epigenetic imprints of being competent to do so [56–58]. In this vernacular, however, being an ‘effector’ does not provide information about future differentiation potential, thus the word effector has been removed from the terrace model. As another point of emphasis, individual memory T cells exhibit a spectrum of developmental potential and are not clustered into a few discrete subsets. Therefore, existing T cell nomenclature only loosely captures the developmental relationship among circulating memory T cells (a few popular subsets have been depicted in Figure 3C while acknowledging some overlap in developmental potential). Moreover, the progression towards terminal T cell differentiation tends to correlate with impaired proliferative capacity and vulnerability to death.

This model is compatible with developmental biologist Conrad Waddington’s widely accepted creode, but rejects a common immunological interpretation that memory T cell subsets signify a few discrete cell types that are linearly organized by developmental potential and ontogeny, and dispenses with the notion that T cells with effector functions arise from terminally differentiated memory T cells [59]. This model also highlights migration because of its intrinsic relevance to understanding T cell biology given the primacy of location in relation to T cell antigen detection, metabolism, phenotype, and function. Thus, resident and circulating memory T cells exist on parallel descents of developmental potential. The property of residence or recirculation is not necessarily a fixed property, and TCR stimulus or inflammation may be a catalyst for change. Thus, developmental potential (irreversible), and migration status (amenable to change) form two continuous axes along which memory T cells may be placed.

Conclusions:

T cells are unusual somatic cells. They can remain quiescent for years, then undergo explosive proliferation and give rise to many progeny with diverse metabolisms, longevities, trafficking patterns, and functions. The growing application of single cell profiling accentuates memory T cell transcriptional, epigenetic, and posttranslational diversity. The establishment of resident memory T cells as a broadly distributed population united by a migration criterion has introduced an additional layer of complexity (and, yet another subset). T_RM exhibit tissue-specific idiosyncrasies and even neighboring cells in the same organ may be dissimilar. The extent of memory T cell heterogeneity appears to be vast and may exceed the number of discrete subsets that would be practical to define in daily discourse, and this has made classification difficult. The subset labels most commonly applied (e.g., ‘effector’, ‘central’, ‘exhausted’, ‘resident’) intend to connote some useful demarcating qualities to organize the complexity.

The terrace model captures the essence of cellular differentiation and is based on a continuum of developmental potential. It deals with migrational heterogeneity, plasticity, and the issue that current subset naming conventions can be ambiguous or impermanent. One major shortcoming of the terrace model is that developmental potential and migration properties can be difficult to measure. However, it may provide a modern framework that embraces the complexity of T cell memory. Ultimately, a common language would help consolidate our current understanding of memory T cells, clarify gaps in knowledge, broaden the audience for our research, and accelerate therapeutic developments.

Highlights.

• T_RM reside in tissues without recirculating.

• Memory T cells are more heterogenous than implied by a few discrete subset labels.

• Popular subset labels do not perfectly define boundaries of developmental potential.

• T_RM are not terminally differentiated and exhibit plasticity.

• A ‘terrace’ T cell differentiation model, based on developmental potential and migration, is proposed.