Systems immunology of regulatory T cells: Can one circuit explain it all?

Abstract

Regulatory T (Treg) cells play vital roles in immune homeostasis and response, including discrimination between self- and non-self-antigens, containment of immunopathology, and inflammation resolution. These diverse functions are orchestrated by cellular circuits involving Tregs and other cell types across space and time. Despite dramatic progress in our understanding of Treg biology, a quantitative framework capturing how Treg-containing circuits give rise to these diverse functions is lacking. Here, we propose that different facets of Treg function can be interpreted as distinct operating regimes of the same underlying circuit. We discuss how a systems immunology approach, involving quantitative experiments, computational modeling, and machine learning, can advance our understanding of Treg function, and help identify general operating and design principles underlying immune regulation.

Negative regulation in mammalian immune homeostasis and response

Mammalian regulatory T (Treg) cells are classically defined as a subset of CD4⁺ T cells that express the forkhead box P3 (FOXP3) transcription factor [1,2], high CD25 (IL2RA, the high affinity subunit of the IL-2 receptor) [3-5], and are characterized by additional molecular signatures such as CD127⁻[6] and CTLA4⁺ [7]. In contrast to conventional T (Tconv) cell subsets, which include CD4⁺ FOXP3⁻ and CD8⁺ T cells, Tregs are predominantly immunosuppressive and disruption of Treg function has been implicated in various autoimmune pathologies [8,9]. The T cell receptor (TCR) repertoires of Treg and Tconv cells are different, with Treg cells showing bias towards self-antigen binding [10-15] (Box 1). Decades of research has revealed multiple key molecular and cellular mechanisms that underlie the immunosuppressive activity of Tregs. While an in-depth review of these mechanisms is not our goal (see [8], [10], and [16], among others, for such reviews), we note that none of these mechanisms appear to be unique to Tregs and can be found in other T and immune cell subsets. Yet, Treg function is required and appears to be irreplaceable by other cell types in diverse functional contexts [16]. Here, we posit that the dynamic interactions among Tregs, Tconvs, and antigen-presenting cells (APCs) (Figure 1 and Table 1; throughout, please consult Table 1 for additional details whenever Figure 1 is referenced), together with the difference in the TCR repertoire between Treg and Tconv cells, can account for a diverse array of Treg functions. Given the complex interactions involving feedback and feedforward elements among Treg, Tconv, and other immune cells and molecules, a “bottom-up” systems biology-based approach— involving both quantitative experimental analyses and computational modeling, with a focus on the network topology of the interactions among Tregs and other cell types— is well suited for advancing a predictive and quantitative understanding of Treg biology.

Box 1. Discovery of Tregs and their development in the thymus.

It was more than fifty years ago that studies using thymectomized mice first suggested that a thymus-derived lymphocyte subset could play a functional role in preventing inflammatory and autoimmune pathology [119,120]; this together with the subsequent observation that this pathological phenotype could be rescued by the adoptive transfer of CD4⁺ T cells hinted that the relevant T cell subpopulation must be CD4⁺ [121]. This subpopulation was later shown to be CD25⁺ [3]. Subsequently, CD4⁺ CD25⁺ Treg cells were identified in human blood and thymus [122,123], followed by studies demonstrating key functional roles for this cell subset in diverse immune response contexts, including infection [109,124], autoimmunity [8], cancer [125], and organ transplantation [126].

During thymic development, thymocytes that strongly bind self-peptides presented by medullary thymic epithelial cells (mTECs) undergo apoptosis, a process called negative selection [127]. However, some CD4⁺ CD8⁻ thymocytes that receive relatively strong T cell receptor (TCR) stimulation from self-peptides upregulate the expression of both CD25 and FOXP3 and subsequently differentiate into Tregs, acquiring an immunosuppressive phenotype [10-14]. A recent study suggested that the timing and duration of agonist signaling in the mouse thymus could play a crucial role in determining the fate of T cells during thymic development. The study showed that the development of Tregs occurs late in the thymic development process in response to strong but disrupted stimulation by self-antigens presented on mTECs [128]. Thus, overall, the TCR repertoire of Tregs tends to be enriched for self-specificity relative to that of Tconv cells. After thymic development, CD4⁺ Tconv cells in the periphery may also upregulate FOXP3 expression and acquire a Treg-like phenotype upon antigen encounter under special conditions, for example, in the presence of TGF-ß [129] or retinoic acid [130,131], or in specific tissues such as the colon [132], and the placenta [133]. Such cells may also be biased towards self-specificity [15].

Key Figure, Figure 1. Model for the Treg-Tconv-APCs cellular circuit.

Each edge represents the overall effect of multiple potential molecular and cellular mechanisms. The mechanisms and processes corresponding to each edge are summarized in Table 1. In the text, each edge is referred to by the number indicated in the figure.

Table 1.

Interactions between Treg, Tconv, and antigen-presenting cells, as shown in Figure 1.

Interaction	Corresponding molecular / cellular mechanisms (non-exhaustive list)	References
Edge 1	Uptake of self- and foreign antigens, via endocytosis, micropinocytosis, or phagocytosis, by dendritic cells, macrophages, and other antigen-presenting cells	[48]
Edge 2	Presentation of self / foreign peptides bound to MHC Class II proteins to Treg cells, together with co-stimulatory signals; the peptides, if recognized by the TCR on the Treg cell, can trigger Treg clonal expansion	[17,18]
Edge 3	Presentation of self / foreign peptides bound to MHC Class II proteins to Tconv cells, together with co-stimulatory signals; the peptides, if recognized by the T cell receptor on the Tconv cell, can trigger Tconv clonal expansion and differentiation into effector Tconv cells	[49,50]
Edge 4	Expansion of Treg cells in response to the IL-2 secreted by activated CD4+ Tconv cells.	[19,38]
Edge 5	Suppression of Tconv cell expansion and function by Tregs, mediated by Treg-driven, CD25-dependent depletion of IL-2, competition for the stimulatory signals provided by APCs (mainly in SLOs), generation of immunosuppressive metabolites, direct killing of Tconv cells, and secretion of immunosuppressive cytokines (IL-10, TGF-ß, etc.) (mainly in peripheral tissues)	[20-23,28-31]
Edge 6	Proliferation of Tconv cells in response to IL-2 produced by activated CD4+ Tconv cells; includes both autocrine and paracrine signaling	[19,38,51]

The Treg-Tconv circuit

APCs including dendritic cells (DCs) can present self or foreign peptides bound to major histocompatibility complexes (MHC; MHC Class I for presentation to CD8⁺ T cells and MHC Class II for presentation to CD4⁺ T cells such as Tregs) to both Treg and Tconv cells (edges 1-3, Figure 1). If the TCR on a Tconv cell binds to the peptide-MHC complex with sufficient avidity, the TCR stimulation, together with the co-stimulatory signals provided by the APC, can trigger Tconv cell proliferation and differentiation into effector and memory cells (edge 3, Figure 1). Similarly, stimulation of the TCR on Treg cells can contribute to their proliferation (edge 2, Figure 1) [17,18]. Activated CD4⁺ Tconv cells secrete IL-2 which further promotes their proliferation (edge 6, Figure 1) and that of nearby Treg cells in the microenvironment (edge 4, Figure 1). Antigen presentation by DCs followed by the proliferation and differentiation of T cells predominantly takes place in the secondary lymphoid organs (SLOs) such as draining lymph nodes. Note that Treg proliferation can be triggered independent of signaling from TCR binding to a self-peptide-MHC complex (edge 2, Figure 1) [17-19]: Tregs can proliferate in response to Tconv-produced IL-2 alone (edge 4, Figure 1). Tregs, in turn, suppress the activation and expansion of Tconv cells via multiple mechanisms (edge 5, Figure 1), such as the titration of IL-2 produced by Tconv cells via high CD25 expression on Tregs [20,21], competition for co-stimulatory signals from DCs (e.g., via CTLA4-mediated depletion of surface CD80 / CD86 on DCs [22]), and the potential removal of the peptide-MHC II complex from DCs via trogocytosis [23].

These Treg-Tconv interactions are not limited to SLOs. CD4⁺ effector Tconv cells secrete IL-2 at infection sites and in tumors, driving the expansion of both effector Tconv and Treg cells (edges 4 and 6, Figure 1). Tregs exhibit immunosuppressive activity in tissues outside of SLOs including the skin [24], synovia [25], mucosal barrier sites [26], and solid tumors [27], acting via mechanisms similar to the ones discussed above, although some of the molecular players can be different. These include ectoenzyme CD39− and CD73-mediated generation of the immunosuppressive metabolites adenosine monophosphate and adenosine, direct killing of effector Tconv cells by Tregs via perforin / granzyme or a Fas ligand-mediated pathway [28], and the production of anti-inflammatory cytokines such as IL-10 [29,30] and TGF-ß [31]; these are capable of suppressing both innate and adaptive responses (edge 5, Figure 1). Treg homing to non-lymphoid organs has been shown to be essential for suppressing tissue-specific autoimmune responses [32], and the same processes may be co-opted by tumors to suppress anti-tumor immunity [33].

The interplay between Treg and Tconv cells involves incoherent feedforward and negative feedback circuits (Figure 1). While the dynamical behavior of such circuits can be somewhat non-intuitive, they are prevalent and well-studied in the context of signaling and transcriptional networks performing diverse biological functions [34,35]. The dynamical behavior of the circuit in Figure 1 depends on the relative “strengths” of the interactions depicted by edges 2, 3, 4, and 5; their relative strengths, in turn, depend on whether self- or non-self-antigens are presented, given that the TCR repertoire of Tregs is biased towards recognition of and higher binding affinity for self-antigens [10-14]. The strengths of interactions depicted by edges 2 and 3 also depend on the co-stimulatory and co-inhibitory signals provided by the APCs to Tconv cells and Tregs. These, in turn, can depend on the presence of inflammatory cytokines and other innate immune response signals. Thus, while the circuit in Figure 1 has a simple network topology, it can operate in multiple dynamical regimes depending on the context (e.g., the kinetic parameters governing the operation of the network): an inflammatory encounter such as infection (wherein foreign antigen presentation is accompanied by pro-inflammatory signaling) versus homeostatic conditions (under which self-antigen presentation is dominant, accompanied by little to no inflammatory signaling). Depending on the regime the system is operating under, the feedback arms (edges 4 and 5, Figure 1) can exert varying levels of suppressive effects on the activated Tconv cells. The same circuit could also operate to avoid excessive tissue damage as a byproduct of inflammatory responses to a pathogen. Moreover, the operation of the Figure 1 circuit is spatially distributed. As mentioned above, Tregs can operate in peripheral tissues to suppress effector Tconv responses with or without inflammation [36], and under inflammatory conditions, Treg-DC interactions in peripheral tissues can subsequently influence antigen presentation by DCs in SLOs [37]. There is also a more local, spatial dimension to the operation of this circuit: for example, fluorescent imaging of Tconv activation markers in mouse lymph nodes under homeostatic conditions (i.e., in the absence of infection) has shown that the local density of Tregs surrounding autoreactive Tconv cells can impact the extent of Tconv activation and proliferation [19,38].

We note that the brief introduction to Treg biology provided here is coarse-grained and focuses only on one cellular circuit, even though the underlying molecular biology of Tregs and their interactions with other cells is highly complex. Additional details, including Treg subsets and their plasticity, and molecular mechanisms of Treg-mediated immunosuppression have been reviewed elsewhere [8,16,28,39,40]. Treg functions in different biological contexts including viral infections [41,42], the tumor microenvironment [43], autoimmune diseases [44], and immunodeficiencies [9,45], have also been previously reviewed. Our central thesis here is that the network introduced above (Figure 1) can serve as a common framework for analyzing and understanding Treg behavior and function across a range of homeostatic and immune response scenarios. Moreover, such a circuit can be extended to incorporate more specific aspects of Treg function. For example, tissue-specific mechanisms of Treg-Tconv interactions can be considered, e.g., by the inclusion of certain gut microbes and gut-specific Treg subsets in the circuit to study Treg function in gut homeostasis and pathology [46]. Similarly, the presence of fetal antigens may be incorporated to probe the function of the Figure 1 circuit in the establishment and maintenance of fetal tolerance by Tregs [47].

Tregs in homeostasis

Tregs play a crucial role in regulating autoimmunity. While negative selection in the thymus prunes self-reactive Tconv cells, recent experiments have revealed an unexpected prevalence of self-reactive Tconv cells exhibiting an activated phenotype under homeostatic conditions [19,38,52-54]. Multiplex imaging of Tconv activation markers in C57BL / 6 mouse lymph nodes has shown that even in the absence of foreign antigens from infection, commensal microbes, or diet, about 2% of CD4⁺ Tconv cells in mouse lymph nodes are IL-2-secreting and express programmed cell death protein 1 (PD-1) [19,38]. Consistent with this observation, the Il2 promoter has been shown to be transcriptionally active in Tconv cells under homeostatic conditions (by the use of reporter mice in which the green fluorescent protein transcription is under the control of the Il2 promoter) [54]. These Tconv cells, posited to be activated by self-antigens (“self-activated”), proliferate vigorously in Treg-deficient / Treg-depleted mice [55], or when Treg function is impaired [38]; this indicates that their activation is constrained by Treg cells under homeostatic conditions. Are such self-reactive Tconv cells indeed “safe”? Do they serve a biological function [52]? The Figure 1 circuit provides a counterintuitive perspective on how such self-activated cells can, in fact, serve to maintain Treg population size at homeostasis, a function enabled by IL-2-mediated interactions between self-activated Tconv cells and Tregs.

Do self-reactive Tconv cells help maintain Treg population size at homeostasis?

In the thymus, negative selection involves stochastic encounters between thymocytes and self-antigen-presenting medullary thymic epithelial cells (mTECs). Consequently, some self-reactive thymocytes, by random chance, may not encounter mTECs presenting the corresponding self-antigens. These thymocytes could leak into the periphery as self-reactive Tconv cells [56-62] and become activated upon recognition of cognate self-antigens [19,38] (edges 1 and 3, Figure 1). In mice, together with Tconv cells that are capable of self-reactivity but have not yet been activated, self-reactive Tconv cells are estimated to comprise 4-10% of the total CD4⁺ Tconv pool in the periphery [53,54,63].

Given that negative selection and other processes regulating T cell development are unlikely to remove all self-reactive Tconv cells [64-66], an intriguing question is whether these cells play any useful role under homeostatic conditions. IL-2 is known to be crucial for Treg maintenance [67]: Treg cells in the periphery exhibit fast turnover [68] and the balance between proliferation and apoptosis in Tregs is IL-2 dependent [69]. However, the cellular source(s) of IL-2 in Treg maintenance are not well understood. Since CD4⁺ Tconv cells are known to be the primary producers of IL-2 (they can do so in response to TCR stimulation and co-stimulatory signals provided by DC and other APCs), self-reactive CD4⁺ Tconv cells could be a potential source of IL-2 that is responsible for Treg maintenance [70-72]. Furthermore, Tregs are continuously maintained in the blood and SLOs even in the absence of foreign antigens [39], suggesting that the IL-2 needed for overall Treg maintenance could at least be derived partly from CD4⁺ Tconv cells activated under homeostatic conditions. We thus propose that the self-antigen-driven activation of self-reactive CD4⁺ Tconv cells (edge 3, Figure 1) and the resultant secretion of IL-2, which can drive Treg proliferation (edge 4, Figure 1), is an important mechanism for peripheral Treg maintenance. Note that in the case of effector or memory Tregs, which can be pathogen-specific [73,74] and are believed to reside mainly in non-lymphoid tissues [75,76], IL-2 can be dispensable for homeostatic maintenance: in one study, treatment with IL-2-neutralizing antibodies did not deplete ovalbumin-specific, skin-resident memory Tregs in a transgenic mouse model with inducible ovalbumin expression in the skin [75]. These Treg subsets seem to rely on IL-7 [75,77] (secreted by stromal cells [78]) or ICOSL (presented by DCs) [71] for continued maintenance. Consequently, their population sizes are unlikely to be perturbed by changes in the population size of CD4⁺ Tconv cells, which do not secrete IL-7 or present ICOSL. The dynamical behavior of such Treg subsets is outside the scope of the framework discussed here.

A dynamic equilibrium: Treg population size control by the Treg-Tconv circuit

While self-activated CD4⁺ Tconv cells can supply IL-2 to maintain Tregs, the size of this self-activated Tconv population must be kept under control to avoid systemic autoimmunity. Note that IL-2 can also mediate a positive self-feedback loop (edge 6, Figure 1): a small number of activated Tconv cells can promote their own proliferation as well as that of neighboring Tconv cells via IL-2-mediated autocrine / paracrine signaling. Small fluctuations in IL-2 production could thus trigger a full-blown proliferative response in self-reactive Tconv cells when the local IL-2 concentration stochastically exceeds a certain threshold. Therefore, this self-driven system must be constantly kept in check to avoid unwarranted immune activation. Without invoking additional mechanisms, Tregs alone can carry out this crucial function via the negative interaction depicted by edge 5 in Figure 1. At the individual cell level, each self-activated Tconv cell in mice SLOs has been shown to undergo a brief period of IL-2 secretion and proliferation but the clonal population is ultimately pruned by the suppressive effects exerted by nearby Tregs [38]. At the systems level, several studies have demonstrated an inverse relationship between the size of the peripheral Treg pool and the frequency of IL-2-secreting CD4⁺ Tconv cells [53,54,63]. Consistently, ablation of Tregs has been shown to drive the proliferation of self-reactive Tconv clones in different mouse organs [79]. Local Treg density (or the Treg population size in a local spatial domain) can function as a checkpoint on Tconv activation and IL-2 secretion, serving as a determinant of whether a self-reactive Tconv cell will be activated in response to cognate antigen presentation by DCs. When a Tconv cell is activated, local Treg population size can impact the duration of IL-2 secretion; thus, the population size of Tconv cells may depend on Treg density. The “stringency” of this checkpoint, in turn, would depend on the relative strengths of the interactions depicted by edges 3, 5, and 6 in Figure 1 [23,80].

Overall, the Tconv response is governed by the behavior of a classic incoherent feedforward circuit [34]: the transient activation of a downstream circuit component (in this case, Tconv cells) in response to an upstream stimulatory signal (self-antigens) is one characteristic phenotype of this circuit motif. This transient response is regulated by the IL-2-mediated feedback between Tconv and Treg cells which can tune the local population size of Tregs as a function of the extent of Tconv activation and IL-2 production (edge 4, Figure 1). Together, the IL-2-mediated Treg-Tconv interaction can shape the size of both the Treg and the self-activated Tconv pools.

The circuit in Figure 1 suggests that Tregs and self-activated Tconv cells can co-exist in a dynamic steady state characterized by homeostatic population sizes: in mice, self-activated CD4⁺ Tconv cells and Tregs are reported to comprise 1-2% and 10-15% of all CD4⁺ T cells, respectively [19,54,71,72]. This steady state is dynamic: there is ongoing activation of self-reactive Tconv cells. We suggest that the IL-2 produced by these cells is a key contributor to maintaining the Treg pool. The Treg pool, in turn, can prevent the unrestricted proliferation of self-activated Tconv cells, pruning them out after a brief period of IL-2 secretion and proliferation. Such a steady state is expected to be stable in response to perturbations. If the Treg population size increases due to an external perturbation, the limited availability of IL-2 can bring it back down; conversely, a decrease in the Treg pool size can increase the number of IL-2 secreting Tconv cells, driving Treg expansion and returning the self-activated Tconv population size to the steady state level [81]. Similarly, a decrease in the IL-2 secreting Tconv pool can shrink the Treg pool while an increase in the IL-2 secreting Tconv pool size can expand the Treg pool, restoring the dynamical balance in both cases. Larger perturbations, however, may disrupt this dynamical balance. For example, impaired Treg expansion after hemopoietic stem cell transplantation is associated with graft-versus-host disease in humans [82,83], which may be attenuated via the administration of IL-2 to restore Treg expansion [84].

Decision boundary between homeostasis and progression towards autoimmunity: Do Tconv cells collectively make decisions via quorum sensing?

When can the self-antigen binding, IL-2-secreting Tconv cells escape Treg suppression and expand towards full-blown autoimmunity? Given that the physical Tconv-DC interaction (edge 3, Figure 1) driving the proliferation and differentiation of Tconv cells [85,86] tends not to last more than 48 hours (at lease in mice) [87], the IL-2 produced by fellow Tconv cells is likely necessary to drive and sustain a proliferative Tconv response [86]. Due to IL-2-mediated positive feedback (edge 6, Figure 1), an activated Tconv pool larger than a certain size may drive its own expansion. This population size threshold defines the “decision boundary” separating the homeostatic regime discussed above from the path towards full-blown autoimmunity (Figure 2, left panel). The underlying mechanism may also be viewed as IL-2-mediated quorum sensing: Tconv cells may collectively determine, based on their population size as “measured” by the local IL-2 concentration, whether to mount a full-blown proliferative response or not. Theoretical and experimental studies have provided indirect support for such a collective decision-making model [88,89]. Two related plausible mechanisms could underlie such behavior: (i) repeated encounters between a recently activated Tconv cell and high-density IL-2 niches formed by other activated Tconv cells, and (ii) formation of Tconv cell clusters that share a dense local pool of IL-2 [86,90]. The aforementioned decision boundary is likely to depend on the size of the IL-2 secreting Tconv pool, IL-2 diffusion rate, and the extent of co-stimulatory signaling from APCs and other innate immune cells; all of these parameters may also be modulated by Tregs [38,91,92]. Overall, spatiotemporal quantitative modeling of the underlying signaling and cell-cell interaction network, together with experiments, is needed to unravel the quantitative biology of this decision-making process.

Figure 2. Schematic of the “decision boundary” in the Treg-Tconv-APCs cellular circuit.

The figure depicts the decision boundary separating the regime of maintenance of a homeostatic activated Tconv pool size from the regime of extensive Tconv clonal expansion and a full-blown immune response. In the absence of infection, the activated Tconv pool size is maintained firmly on the left side of the decision boundary, i.e., in the homeostatic regime (left panel), and cannot undergo further expansion. In response to infection by a foreign pathogen (right panel), the decision boundary shifts left towards a smaller activated Tconv pool size; together with an increased local pool size of foreign antigen-activated Tconv cells, this ensures that the resultant pool of foreign antigen-specific Tconv cells is sufficient for mounting a full-blown response characterized by rapid clonal expansion. The shift in the decision boundary can be driven by multiple factors including increased Tconv stimulation by foreign antigens, as well as increased antigen presentation and co-stimulatory signaling by dendritic cells and other antigen presenting cells due to innate immune activation, among others. The increased pool size of the foreign antigen-activated Tconv cells can be induced by Tconv proliferation in response to strong TCR stimulation from foreign antigens and by increased chemotaxis of Tconv cells to DC-containing niches.

Tregs in infection and inflammatory responses

Regime change: symmetry breaking between self- and non-self-antigens in response to infections

In response to infection by a foreign pathogen, the circuit shown in Figure 1 must allow for the expansion and differentiation of pathogen-specific Tconv cells needed for pathogen clearance. At the same time, self-reactive Tconv cells must be kept in check to prevent autoimmune pathology. How, then, does the circuit break the ‘symmetry’ between foreign-antigen-activated and self-antigen-activated Tconv cells? What signaling changes are needed for such symmetry breaking to happen? Since both foreign- and self-antigens are presented in the SLOs draining infected tissues, and since the sizes of foreign- and self-antigen-specific naïve Tconv clones can be comparable [52,61], symmetry breaking must involve expanding the pathogen-specific Tconv cells to a greater extent than the self-activated Tconv cells. In addition, the decision boundary discussed above would need to be shifted so that the presence of a relatively small number of Tconv cells recognizing pathogen-specific antigens is sufficient for them to escape Treg-mediated suppression, proliferate, and differentiate in order to effectively respond to the pathogen in question [93] (Figure 2, right panel).

The circuit in Figure 1 can orchestrate such an outcome. First and foremost, unlike homeostatic conditions, infection by a foreign pathogen is accompanied by an inflammatory response that upregulates MHC expression, co-stimulatory signaling, and phagocytic potential in APCs; APC trafficking to SLOs is also upregulated [94]. These activated APCs have enhanced capacity to present antigens and mitigate CTLA4-mediated Treg suppression. In parallel, activated pathogen-specific Tconv cells upregulate CD25 expression to better compete with Tregs for IL-2 [95]. Tconv cells can also upregulate chemokine receptors (e.g., CCR5 and EBI2) that receive signals from activated DCs (e.g., via CCR5 ligands CCL3 and CCL4) [96-98] to facilitate the formation of intrafollicular Tconv aggregates. This could increase the co-localization and, thus, the local density of foreign antigen-specific, IL-2 secreting Tconv cells [97]. The overall effect of these processes would be expected to strengthen the interactions depicted by edges 3 and 6 and attenuate those depicted by edges 2 and 5 (Figure 1). As a result, even a small pool of pathogen-specific Tconv cells could escape Treg suppression and expand to form IL-2-sharing microenvironments, thus moving towards a full-blown anti-pathogen response (Figure 2, right panel).

At the same time, can self-reactive Tconv cells take advantage of the inflammatory environment in SLOs during an infection to trigger autoimmunity? At least two mechanisms may act to mitigate this. First, the skewing of Tregs’ TCR repertoire towards self-antigens could allow Tregs to be better at intercepting the presentation of self-antigens to self-reactive Tconv cells, thereby resulting in stronger suppression (edge 5, Figure 1) of self-reactive Tconv cells as compared to pathogen-specific Tconv cells [23]. Treg competition for self-antigens might be enhanced by increased Treg-DC interactions during inflammation, facilitated by increased CCL22-CCR4 (expressed on DCs and Tregs, respectively) interaction due to upregulation of CCL22 in response to infection-triggered toll-like receptor 9 (TLR9) signaling in DCs [99]. Consequently, self-specific Tconv cells would be, at best, “weakly” activated and thus, less responsive to the aforementioned chemotactic guidance from activated DCs to form or join the IL-2 sharing microenvironments established by foreign antigen-specific Tconv cells. At the same time, strong TCR stimulation by foreign antigens and co-stimulation from activated DCs could upregulate CD25 expression in pathogen-specific Tconv cells [49,100,101]; these cells could thus further outcompete self-reactive Tconv cells exhibiting weaker CD25 expression for IL-2 stimulation. In fact, foreign antigen-reactive Tconv cells could also outcompete Tregs for IL-2 to form IL-2-sharing microenvironments [95]. Thus, we argue that the same circuit (Figure 1) operating in a distinct dynamical (and spatial) regime during infection and / or inflammation (i.e., under a different spatial organization of cells as well as different strengths and kinetics of interactions among them, compared to homeostasis) could allow for an appropriate response against infections without triggering autoimmunity.

Regime reversal: returning to homeostasis after an infection

The concentration of the foreign antigen is an important determinant of inflammation resolution. As the infection clears and the foreign antigen concentration wanes, DCs (and other APCs) lose high MHC expression and co-stimulatory signaling. Pathogen-specific Tconv cells, in turn, receive less TCR stimulation and co-stimulation, and both CD25 expression and IL-2 secretion decline [102]. Together, these changes can restore Treg capacity to suppress the expansion of foreign-antigen specific Tconv cells. With the strengthening of interactions depicted by edges 2, 4, and 5, and the weakening of edges 3 and 6 interactions (Figure 1), the size of the foreign antigen-specific Tconv pool would decline and the decision boundary would shift back to a higher threshold corresponding to a larger size of the activated Tconv cell population. The strong coupling between pathogen clearance (and inflammation resolution) and strengthening of Treg dominance could further ensure that self-reactive Tconv cells not take advantage of the pathogen-induced inflammatory environment to trigger an autoimmune response. Presumably, as inflammation resolves and the foreign antigen clears completely, the overall system would revert back to homeostasis with the decision boundary (see above) specified by the population size of the self-activated Tconv cells. Failure of Treg-mediated regulation at this stage could contribute to immunopathology even after the foreign antigen has cleared.

The Treg-Tconv circuit as a rheostat balancing pathogen clearance and immunopathology

The circuit in Figure 1 might serve another function during infection: limiting self-damage from the immune response (Figure 3). The effector immune response for clearing an infection can potentially cause damage to the body’s own cells and tissues, presumably increasing the amounts of self-antigens that APCs could present to T cells in SLOs. This could increase TCR stimulation in Tregs and promote their expansion and retention in SLOs. As a negative feedforward response, Treg activation and expansion might attenuate excessive Tconv activation and differentiation in SLOs, thereby restraining immune pathology. Of note, activated Tregs can also migrate to the site of the inflammatory response [103] to suppress effector responses via multiple pathways, for example, via the production of immunosuppressive metabolites (e.g. adenosine monophosphate and adenosine), secretion of immunosuppressive cytokines (e.g., IL-10 and TGF-ß), and direct killing of effector T cells [28,40]. We propose that the ratio between the quantities of foreign- and self-antigens taken up and presented by APCs at any given time during an infection response is an important quantity for determining the balance between pathogen clearance and immunopathology. When there is excessive tissue damage, self-antigen presentation and Treg activity increase (since the TCR repertoire of Tregs is biased towards self-antigen binding), which could then attenuate the Tconv effector response and the associated excessive immunopathology. In contrast, if the effector Tconv response is excessively suppressed by Treg-mediated negative feedback, the amount of foreign antigen presentation would exceed self-antigen presentation, tilting the balance towards pathogen clearance. Thus, the circuit shown in Figure 1 may have evolved to achieve a fine balance between pathogen clearance and immunopathology.

Figure 3. Treg cells are key to mounting an effective anti-pathogen response while restricting excessive immunopathology.

With differing binding affinities of the T cell receptor (TCR) repertoires of Treg and Tconv cells to self- and non-self-antigens, the Treg-Tconv cellular circuit (Figure 1) can mount an effective anti-pathogen immune response without causing excessive immunopathology. Infection by a foreign pathogen, accompanied by innate immune activation and presentation of foreign antigens, promotes an effector Tconv response which is needed for pathogen clearance. However, if pathogen clearance by effector Tconv cells results in excessive self-tissue damage, the self-antigens generated and presented to Tregs can strengthen Treg-mediated immunosuppression, helping to reign in the effector Tconv response and tissue damage. The circuit shown in Figure 1 is capable of striking a delicate balance between Tconv and Treg responses, allowing for effective pathogen clearance while restraining immunopathology.

Dysregulation of the Treg-mediated balance between pathogen clearance and immunopathology can have severe consequences. For example, during Toxoplasma gondii infection in C57BL/6 mice, CD4⁺ Tconv cells produce large amounts of IFN-γ at the expense of IL-2, resulting in a collapse of the Treg population [104]; the Treg pool collapse can be further promoted by PD-1-PD-L1 signaling-mediated Treg suppression [105]. Although these mice were able to restrict pathogen expansion, they exhibited systemic immunopathology including intestinal inflammation and necrosis, liver dysfunction, and eventual death. These issues could be ameliorated with IL-2-anti-IL-2 complexes, which enhanced Treg survival to restore the balance between pathogen clearance and immunopathology. In the case of septic shock, Tregs can undergo excessive proliferation after the acute phase, as shown in human patients [106], possibly due to the high concentration of self-antigens from tissue damage. The resulting Treg-mediated immunosuppression might increase a septic shock survivor’s susceptibility to new infections and cancer [107,108].

Thus far, our discussion has assumed that it is possible to clear the pathogen without excessive immunopathology. In the case of chronic infections (i.e., without complete pathogen clearance), the pathogen itself may need to be added as an additional component to the Figure 1 circuit. If the “just right” immune response, i.e., one that avoids excessive self-damage, fails to clear the pathogen, a different kind of dynamical steady state could emerge, one involving continual activation of Tconv cells and Tregs by foreign (from the pathogen) and self-antigens (from tissue damage), respectively. Such a regime would be characterized by a “tug of war”-like situation with the pathogen not clearing completely. It has been hypothesized that such a scenario might arise particularly when the pathogen has evolved strategies to weaken effector Tconv cell responses while strengthening Treg responses [109]. While the framework discussed thus far does not explicitly incorporate such a dynamical steady state, it may be extended to do so by explicitly modeling how different pathogens may impact the nature and strengths of the various interactions (i.e., edges) in the network shown in Figure 1.

A dynamical systems perspective: mapping parameters to phenotypes using machine learning

Throughout this manuscript, we have described how the various functions of Tregs might be qualitatively understood within the framework of the cellular circuit shown in Figure 1. These functions include maintaining a homeostatic immune state in the absence of infections, coordinating signals from self- and non-self-antigens to mount an effective adaptive immune response to pathogens, limiting immunopathology during, and orchestrating inflammation resolution after pathogen clearance. To make these descriptions more precise and quantitative, the theory of dynamical systems can be a useful tool [110]. In fact, we have already used the language of dynamical systems in our discussions, e.g., the notion that different functions of Tregs are characterized by the Treg-Tconv circuit operating in different “parameter regimes”. A useful tool for dynamical systems analysis is a phase portrait, which provides an overview of how the behavior of the system changes as a function of key parameters. In the context of the Treg-Tconv interplay, an infection changes several key parameters of the system as discussed above and pushes the system to cross a “bifurcation point” from a Treg-dominated regime to a Tconv-dominated one. As the infection resolves, the system crosses back over the bifurcation point to a homeostasis-like state. In Box 2, an example of a phase portrait analysis describing the maintenance of homeostasis and initiation of an infection response is provided.

Box 2. A phase portrait of the Treg-Tconv interplay.

Here is an example of phase portraits and how they can be helpful in understanding the role of Treg-Tconv interplay in homeostasis, infection response, and autoimmune dysregulation. Each panel in Figure I shows a two-dimensional phase portrait, indicating behavior as a function of Treg and activated IL-2-secreting Tconv pool sizes. The different curves represent nullclines: trajectories corresponding to steady states of Treg and activated Tconv pools. Intersection of two nullclines defines the dynamic equilibrium at which the system will exist. The background color in each panel shows the probability of Tconv clonal expansion in different regions of the Treg-Tconv space. We consider three scenarios here. The left panel shows the behavior at homeostasis: the dynamic equilibrium point is in a region with near zero probability of clonal expansion. Consequently, there will be no spontaneous clonal expansion at homeostasis. The middle panel shows how clonal expansion can be achieved by increasing the activated Tconv population size at steady state (indicated by the new, dashed nullcline) and shifting the region with possible clonal expansion to the left. This moves the equilibrium point to a region with high probability of clonal expansion. The panel thus corresponds to the behavior for the case of foreign-antigen-activated Tconv cells in the presence of innate immune activation. In the right panel, clonal expansion of Tconv cells emerges from a decrease in the Treg pool size which once again moves the equilibrium point to a region of with high clonal expansion probability. This represents the case of Treg dysfunction which is accompanied by the spontaneous expansion of self-reactive Tconv cells in the absence of infection, as seen in various autoimmune conditions involving loss of Treg activity [8,134,135].

The phase portraits shown here, while merely schematic for illustration purposes, provide an intuitive, qualitative description of the Treg-Tconv interplay. These portraits can be made more quantitative using experimental data and calibrated mathematical models, and applied to make quantitative predictions of system behavior, including how homeostasis is disrupted upon therapeutic or pathologic perturbations in individual patients.

Figure I. Illustration of how phase portraits corresponding to the dynamics of the Treg-Tconv interplay in different biological contexts can be helpful in understanding Treg behavior and function across contexts.

The type of dynamical behavior described above can be quantitatively modeled by differential equations that track the temporal evolution of the system states (e.g., transcriptional state, number, and location of cells). One of the challenges of analyzing such models is that many of the underlying parameter values (e.g., cell counts, ‘on’ and ‘off’ rates of certain molecular interactions) are unknown and can span a large range. One approach that can help overcome this challenge is to vary the parameters of the system over biologically plausible ranges and then map the parameter regimes corresponding to distinct dynamical behaviors to different biological contexts and functions [111,112]. Such an approach differs from the more classical systems biology modeling approach of fitting the kinetic parameters of a dynamical system to a chosen set of experimental data. Our recently developed machine learning-based framework, MAchine learning of Parameter Phenotype Analysis (MAPPA), utilizes simulations of biological dynamical systems over a range of parameter sets followed by machine learning to identify mappings between parameter combinations (e.g., different parameter regimes) and quantitative phenotypes of a biological system [112]. When applied to a multiscale model of the Treg-Tconv interplay, MAPPA identified a key role for Treg spatial micro-domain size in restricting the homeostatic expansion of self-activated Tconv cells. MAPPA also predicted a non-linear threshold for Tconv “escape”: Treg micro-domain size can be reduced by as much as 40% without apparent effects on Tconv activation, but a reduction beyond this threshold would result in significant Tconv activation. This prediction was experimentally confirmed by measuring the number of activated (phosphorylated pSTAT5⁺) CD4⁺ Tconv cells in mouse lymph nodes in response to varying levels of Treg ablation, all under homeostatic conditions [38].

The Treg-Tconv circuit discussed here is but one example where a dynamical systems perspective focused on circuit topology can be a useful framework. Such a perspective, involving the analysis of the range of dynamical behaviors that can be exhibited by molecular and cellular circuits, followed by mapping of these ranges to specific biological contexts, could prove useful for analyzing immune responses across spatial and time scales, e.g., the germinal center reaction [113].

Limitations of the proposed framework

While the simple Treg-Tconv-APCs cellular circuit shown in Figure 1 can explain immune response dynamics across multiple contexts, the framework proposed here has certain limitations. The framework and our discussion focus mainly on naïve Tconv responses and how these responses are modulated by Treg cells; recall responses driven by memory Tconv cells may exhibit different dynamics corresponding to a distinct regime of the Figure 1 circuit or may involve a circuit with a different topology. The Figure 1 circuit also does not include Treg-B cell interactions, and thus cannot be used to understand the Treg-mediated control of humoral immunity. In addition, the circuit does not incorporate the role of IL-2 produced by and secreted from other immune cell types, including DCs and natural killer (NK) cells, which were shown to express IL-2 under homeostatic conditions in a recent study involving IL-2 reporter mice [114]. Finally, the circuit does not incorporate the wide-ranging cross-talk between immune cells and the various non-immune cell types in tissues, including fibroblasts [115] and endothelial cells [116], which can play important roles in fighting off infections and in the maintenance of tissue homeostasis. These limitations arise, partly, from the inherent simplicity of the framework and may be addressed by incorporating additional immune processes of interest into the Figure 1 circuit.

Concluding remarks

In this Opinion, we have argued that the seemingly disparate functions of Tregs can be understood as different dynamical regimes of the same underlying cellular circuit (Figure 1), which incorporates the multi-scale interplay between Treg, Tconv, and APCs operating across time and space. We suggest that this cellular circuit, analyzed from a dynamical systems perspective, provides a useful unifying framework for understanding Treg behavior and function in health and disease. Our perspective also suggests that the TCR repertoire of Tregs, with its bias towards binding to self-antigens, is likely key to Tregs’ indispensable and versatile functionality. While the TCR repertoires of Treg and Tconv cells are known to be different (with Treg bias towards self-antigen binding), quantitative characterization of the similarities and differences between their repertoires remains to be undertaken, and might be facilitated by machine learning-based approaches [117,118] (see Outstanding Questions). Going further upstream to T cell development in the thymus, a quantitative understanding of the interactions between thymocytes and thymic epithelial cells could further illuminate how the fates of thymocytes, i.e., whether to survive, die, or become Tregs, are determined to ultimately shape the TCR repertoires of Tconv and Treg cells in the periphery. Given that the overall outcome of the Treg-Tconv interactions is determined by the kinetics of molecular and cellular processes taking place in peripheral tissues and SLOs, analysis of Treg-Tconv interactions at high spatial and temporal resolution, together with dynamical systems and machine learning analysis, can help turn the next chapter in Treg biology.

Outstanding questions.

• A key unknown parameter is the quantitative extent of overlap between the Tconv and Treg TCR repertoires: how much more self-antigen biased is the repertoire of Treg cells in comparison to Tconv cells?

• How different does this circuit operate in different tissues?

• How do the abundance and phenotype of Treg and Tconv cells in the microenvironment impact the decision boundary between homeostasis and infection response?

• What is the interplay between negative selection in the thymus and Treg-mediated control in the periphery (post thymic selection) in regulating autoimmune responses? To what extent does the leakage of self-antigen specific Tconv cells from the thymus optimize such control?

Highlights.

• Regulatory T cells (Tregs) interact with conventional T (Tconv) cells and antigen-presenting cells (APC) in a complex circuit involving both feedforward and feedback regulation. This circuit can operate in multiple dynamic regimes and is key to Treg function in homeostasis, infection responses, and immune resolution.

• The Treg-Tconv-APC circuit, together with the T-cell receptor repertoire of Tregs being biased towards the recognition of self-antigens, restricts the activation of self-reactive Tconv cells under homeostatic conditions while allowing for a robust infection response from pathogen-specific Tconv cells.

• Recent experimental advances characterizing Treg-Tconv-APC interplay in space and time have advanced our understanding of Treg biology in health and disease.

• Combining quantitative experiments with multi-scale mathematical models and machine learning approaches might be key to advancing our understanding of Treg function across diverse biological contexts and for developing putative Treg-targeting therapies.

Significance.

Regulatory T cells perform critical functions in diverse biological contexts in health and disease. Here, a unifying framework is proposed that interprets the different facets of Treg function as distinct dynamical operating regimes of a circuit involving feedforward and feedback interactions among regulatory T cells, conventional T cells, and antigen presenting cells.

Glossary

Trogocytosis

from the ancient Greek trogo, meaning ‘gnaw’; extraction of surface molecules expressed on the surface of a cell by another cell which may then express the extracted molecule on its own surface.

Medullary thymic epithelial cells (mTECs)

population of cells in the thymic medulla that express self-antigens from various peripheral tissues and interact with immature thymocytes, helping to screen out T cells with strongly self-reactive T cell receptors.

Graft-versus-host disease

complication that occurs after allogenic transplantation of hematopoietic stem cells wherein donor immune cells from the transplant (graft) recognize the recipient’s (host’s) tissues as foreign and attack the host’s cells, causing systemic inflammation.

Quorum sensing

form of cellular decision-making, involving aspects of both autocrine and paracrine signaling, wherein cells in a population respond collectively to changes in the local population density. Each cell in the population can produce, detect, and respond to one or more extracellular signaling molecules. As the local population density increases, the signaling molecule accumulates in the environment. The accumulation can be detected by cells in the population to track changes in the local population, and to mount a synchronous response (via change in gene expression, for example).

Toll-like receptors

family of trans-membrane receptors expressed on cell membranes and endosomes that can recognize chemical patterns associated with foreign pathogens (e.g., microbial cell wall components, viral genetic material, etc.), as well as tissue damage, triggering innate and adaptive immune responses.

Phase portrait

geometric representation of the behavior of a dynamical system wherein the axes are state variables and system trajectories are represented by curves (or surfaces) in the state space.

Bifurcation point

point in the parameter space of a dynamical system where a small change in one or more system parameters can dramatically change the dynamical behavior of the system. Changing a system parameter(s) across a bifurcation point can change the stability of a steady state (from stable to unstable, or vice versa), make a steady state disappear, or lead to the emergence of a new steady state, among other possible behaviors.

Nullcline

geometric shape (such as a curve or a surface) connecting points in the state space of a dynamical system for which one of the state variables is in steady state (i.e., the state variable does not change with time). The point(s) where all the nullclines of a dynamical system intersect correspond(s) to stable or unstable equilibrium point(s) of the system.