Conceived and designed the experiments: AN JC CC DT. Performed the experiments: AN. Analyzed the data: AN JC DT. Contributed reagents/materials/analysis tools: AN CC DT. Wrote the paper: AN JC CC DT.
The authors have declared that no competing interests exist.
Alternative cell differentiation pathways are believed to arise from the concerted action of signalling pathways and transcriptional regulatory networks. However, the prediction of mammalian cell differentiation from the knowledge of the presence of specific signals and transcriptional factors is still a daunting challenge. In this respect, the vertebrate hematopoietic system, with its many branching differentiation pathways and cell types, is a compelling case study. In this paper, we propose an integrated, comprehensive model of the regulatory network and signalling pathways controlling Th cell differentiation. As most available data are qualitative, we rely on a logical formalism to perform extensive dynamical analyses. To cope with the size and complexity of the resulting network, we use an original model reduction approach together with a stable state identification algorithm. To assess the effects of heterogeneous environments on Th cell differentiation, we have performed a systematic series of simulations considering various prototypic environments. Consequently, we have identified stable states corresponding to canonical Th1, Th2, Th17 and Treg subtypes, but these were found to coexist with other transient hybrid cell types that co-express combinations of Th1, Th2, Treg and Th17 markers in an environment-dependent fashion. In the process, our logical analysis highlights the nature of these cell types and their relationships with canonical Th subtypes. Finally, our logical model can be used to explore novel differentiation pathways
T lymphocytes play a key role in the regulation of the immune response in mammals. Various T-helper subtypes (Th1, Th2, Th17, Treg,…) have been identified over the years, characterised by the expression of specific transcription factors and cytokines, which have a critical influence on the selection of different immune responses, driving pro-inflammatory or allergic responses, promoting alternative antibody classes, or preventing (auto)immunity by inhibiting the activation and proliferation of other cells. To gain insight into the heterogeneity and the plasticity of late T-helper lineages, we have built an integrated model of the regulatory network and signalling pathways controlling Th cell differentiation. Relying on a logical modelling framework, we have performed a systematic series of simulations to assess the effects of heterogeneous environments on Th cell differentiation. We have identified stable states corresponding to canonical Th1, Th2, Th17 and Treg subtypes, but also to hybrid cell types co-expressing combinations of Th1, Th2, Treg and Th17 markers in an environment-dependent fashion. Our analysis highlights the nature of these cell types and their relationships with canonical Th subtypes.
Alternative cell differentiation pathways are believed to arise from the concerted action of signalling pathways and transcriptional regulatory networks. However, the prediction of mammalian cell differentiation from the knowledge of the presence of specific signals and transcriptional factors is currently a daunting challenge. In this respect, the vertebrate hematopoietic system, with its many branching differentiation pathways and cell types, is a compelling case study. In particular, numerous publications describe molecular and genetic interactions involved in the control of the late stages of (TCR
Beyond the expression of diverse and cell-specific antigen receptor genes, the appreciation of the heterogeneity of late Th cell lineages emerged from the characterisation of Th1 and Th2 cell types
Additional T-helper subtypes have been recently identified. Regulatory T cells that depend on the transcription factor Foxp3 are capable of preventing (auto)immunity by inhibiting the activation and proliferation of other cells
Effective immunity to many fungi and bacteria requires that the T cell response is dominated by pro-inflammatory effector Th1 or Th17 cells. Allergic reactions, whether beneficial or deleterious, are strictly dependent on Th2 cells. Avoiding spontaneous autoimmunity or controlling the collateral damage of effective immune responses to infection involves a fine balance between regulatory T cells and other Th cells. Genetic defects or accidental failures affecting this delicate balance can lead to irreversible immunopathology.
The issue is not only how heterogeneous are these cell types, but, perhaps more importantly, how is the heterogeneity of Th cells sustained and controlled? How does antigen-specific memory correlate with the predominance of an appropriate Th cell branch? How plastic or resilient are the differentiated Th cells?
During their lifetime, naive and memory Th cells face a changing environment in time and space. The lymphoid tissues are heteregeneous and provide variable local cytokine contexts to circulating Th cells. How robust are the Th subtypes with respect to these heterogeneous environments? Will a Th cell that differentiated into a Th1 phenotype in a lymph node always remain in this state or can it switch to another cell type if it faces a different environment? Evidences for substantial plasticity have been recently reported. For example,
How many instances of such conversions should be expected? Are some Th cells irreversibly committed and others more plastic? It is difficult to address these questions directly due to the current impossibility to follow a single cell as it circulates in the body, during the rather long time scale of cell differentiation. Instead, studies are usually made on cell populations and therefore measure the predominance of one or another cell type. However, using mathematical modelling, these questions can be addressed in terms of stability and robustness of differentiation states of the molecular network underpinning the cellular phenotypes.
Mathematical modelling has been used recurrently to formalise hypothetic regulatory schemes in immunology. Early phenomenological models represented the development of Th1 vs Th2 responses from naive Th0 cells using ordinary differential equations
More recently, models of the cellular networks driving Th cell differentiation and polarisation have been formulated using logical
In this paper, we propose an integrated, comprehensive model of the regulatory network and signalling pathways accounting for the core control of Th cell differentiation. As most available data are qualitative, we rely on a qualitative, logical formalism to perform extensive dynamical analyses. To cope with the size and complexity of the resulting network, we use an original model reduction approach described in detail elsewhere
To assess the effects of heterogeneous environments on Th cell differentiation, we have performed systematic simulations, considering various prototypic environments. As we shall see, stable states corresponding to canonical Th1, Th2, Th17 and Treg subtypes are readily identified, but they are found to coexist with other hybrid cell types that co-express combinations of Th1, Th2, Treg and Th17 markers in an environment-dependent fashion. In the process, our logical analysis highlights the nature of these cell types and their relationships to canonical Th subtypes.
The precise roles of the different molecular species involved in the regulation of T cell differentiation are sparsely known. Even in the cases where direct regulatory interactions have been documented, little or no quantitative information is available on the relative strengths or rates of these processes.
The extended logical formalism
Component(s) | Qualification | Behaviour | Reference(s) |
IFNG_e, TGFB_e, IL{2,4,6,10,12,15}_e, IL{17,21,23,27}_e | External cytokines. | Input of the model representing the external environment. We do not consider the arrest of the activation. | |
APC | Denotes the presence of an Antigen-Presenting Cell. | Input of the TCR module. We do not consider the arrest of the activation. | |
CGC, IFNGR{1,2}, IL{4,6,10,15,27}RA, GP130, IL{2,10}RB | Subchains of the cytokine receptors. | Assumed to be constitutively expressed at functional levels. | |
IL12RB2 | Subchain of IL-12R. | Inhibited by STAT6 (present otherwise). | |
IL12RB1 | Subchain of the IL-12 and IL-23 receptors. | Always present with a higher level (required for IL-12 signalling) in presence of IRF1. | |
IL2RA | High affinity subchain of the IL-2 receptor. | Activated by NFAT, NFKB, STAT5, SMAD3 and FOXP3. | |
IL4RA | Subchain of the IL-4 receptor. | Constitutively expressed, it is upregulated by a high level of STAT5. | |
IFNGR, TGFBR, IL{4,6,10,15,17}R, IL{21,23,27}R | Cytokine receptors, composed of subchains as described in |
Active when their subchains and the cytokine (external or from the same cell) are present. | |
IL12R | IL-12 receptor. | As other receptors but requires a higher level of IL12RB1. | |
IL23R | IL-23 receptor. | As other receptors but also requires ROR |
|
IL4R | IL-4 receptor. | As other receptors, high level of receptor requires a high level of IL4RA. | |
IL2R | IL-2 receptor, composed of three subchains (CGC, IL-2R |
CGC and IL2RB are mandatory, while IL2RA is only needed for higher levels of IL2R. | |
TCR, CD28 | T Cell Receptor and its co-receptor. | Activated by APC. | |
IKB | Denotes I |
Inhibited by the TCR pathway. | |
NFKB | Denotes NF |
Inhibited by IKB and FOXP3. | |
IRF1 | Transcription factor. | Activated by STAT1. | |
STAT1 | Transcription factor. | Activated by IFNBR, IFNGR and IL27R. | |
STAT3 | Transcription factor. | Activated by IL6R, IL10R, IL21R, IL23R, and IL27R. | |
STAT4 | Transcription factor. | Activated by IL12R and inhibited by GATA3. | |
STAT5 | Transcription factor. | Activated by IL2R, IL4R, and IL15R. High levels of IL2R or IL4R are required for high levels of STAT5. | |
STAT6 | Transcription factor. | Activated by IL4R. | |
proliferation | Denotes cell proliferation. | Triggered by high levels of STAT5, its arrest is not considered here. We assume that cell proliferation is required for the production of all cytokines but IL2. | |
NFAT | Transcription factor. | Activated by TCR and CD28. We assume it is required for the production of all cytokines. | |
TBET | Denote T-bet, the master switch for the Th1 subtype. | Activated by itself and STAT1 and inhibited by GATA3. | |
RUNX3 | Transcription factor | Activated by TBET. | |
GATA3 | Denotes GATA-3, the master switch for the Th2 sub type. | Activated by itself and STAT6 and inhibited by TBET. | |
FOXP3 | Transcription factor specific to Treg cells. | Activated by NFAT, TGFB (through SMAD3), and IL2 (through STAT5) and inhibited by IL6 (through STAT3). Based on promoter binding data, we further assume inhibition by STAT1 and RORGT. | |
RORGT | Denotes ROR |
Self-maintained and activated by STAT3 and TGFBR. Potential intermediate in STAT3 activation by TGFBR. | |
TGFB | Denotes TGF- |
Produced by the Treg, assumed to be activated by FOXP3. | |
SMAD3 | Signal transduction component. | Activated by TGFB. | |
IFNG | Denotes IFN- |
Activated by NFAT, proliferation, TBET/RUNX3 and STAT4/IRAK. Activation by NFAT inhibited by FOXP3. Inhibited by STAT3. | |
IL2 | Denotes interleukin-2. | Activated by NFAT anf NFKB. STAT5 and STAT6 cooperate to inhibit IL2 production. FOXP3 cooperates with NFAT to inhibit IL2. TBET cooperates with RelA (NF |
|
IL4 | Denotes interleukin-4. | Activated by GATA3, NFAT and proliferation. TBET and RUNX3 inhibit IL4 cooperatively. FOXP3 blocks its activation by NFAT. STAT1 inhibits IL4 through IRF1. | |
IL10 | Denotes interleukin-10. | Activated by NFAT, proliferation, GATA3 IL6 and TGFBR (probably through STAT3). | |
IL17 | Denotes interleukin-17. | Activated (cooperatively) by STAT3 and RORGT and inhibited by IL2 (through STAT5) and FOXP3. We further assume inhibitions by STAT1 and STAT6. | |
IL21 | Denotes interleukin-21. | Activated by STAT3. | |
IL23 | Denotes interleukin-23. | Activated by STAT3. |
This table lists the regulatory components considered in the Th cell differentiation network model, along with their qualifications, behaviours and related references.
Next, logical rules are defined for each regulatory component to specify its
Given a regulatory graph and starting from a (set of) initial state(s), successor states can be recursively computed. This results in a new graph, called
Considering a state, if the activity level of a component differs from the target level defined by its logical rule, there is an updating call on the corresponding variable. We generally assume that a state has as many successors as updated components (fully asynchronous dynamics), potentially leading to alternative behaviours. The computation of the state transition graph can be restricted by considering a set of priority classes. Each regulatory component is then associated to a priority class and is updated only in the absence of concurrent updating call with a higher priority
Considering a state transition graph, it is particularly interesting to identify the
The asynchronous dynamical analysis of increasingly large regulatory graphs can be very challenging due to the exponential growth of the state transition graphs. A solution consists in reducing the model by removing (intermediate) components.
We have recently proposed an algorithm to automatically compute the logical rules for a user-defined reduced model
This reduction method ensures the preservation of a number of dynamical properties of the original model. In particular, stable states and more complex attractors are conserved. However, additional cyclical attractors may arise from the isolation of transient cycles of the original system. An attractor which is reachable in the reduced model is also reachable in the full model, but the reverse is not always true, as the reduction may lead to the loss of some trajectories (see
One asset of the logical framework is the possibility to analyse the dynamical roles of the
While the presence of a positive circuit is necessary for the existence of multiple stable states, this condition is not sufficient. Indeed, when regulatory circuits are embedded in large networks, external regulators may prevent some circuits to generate the expected dynamical behaviour. Circuit functionality contexts can then be defined in terms of constraints on the values of the regulators of circuit members (see
Developed in Java, the software
We used GINsim to build a comprehensive logical model by integrating a large set of documented molecular actors and interactions. The resulting model is available in the model repository linked to the GINsim website. As GINsim allows the user to document a model by associating annotations (textual comments, links and references) with each component, extensive documentation has been integrated in the model file (see
In the case of the model presented here, we have to deal with many input nodes that collectively handle local environmental conditions. Varying the values of these inputs, all possible fates of a given cell can be computed in terms of attractors reachable in these diverse local environments.
Our simulations aim at determining how naive cells can differentiate into specialised cells, depending on local environments, and how these differentiated cells respond to environmental changes. In this respect, we have defined a set of prototypic local environments (
As we wish to take into account the roles of various external cytokines on the activation and differentiation of Th cells, we expect to find many possible stable states, some differing from each other only by minor component activities, and thus corresponding to the same Th subtype. Similar stable states can be grouped and denoted by the value of a vector encompassing the major transcription factors (GATA-3, T-bet, ROR
The logical formalism enables a modular approach for the construction of large regulatory network models, where parts of the network are defined and studied separately, before merging them to generate a comprehensive model. The Th cell network has been constructed by combining several “modules” illustrated in
The first building block is the TCR signalling module involved in the control of all aspects of Th cell life cycle, including activation and differentiation. A comprehensive Boolean model of this signalling cascade has been recently published
Cytokines play an important role in the control of T cell differentiation. Cytokine signalling proceeds via the JAK-STAT pathway. Cytokine binding to its receptor chains leads to the phosphorylation of specific JAK and STAT factors, the latter being translocated into the nucleus where they activate the transcription of target genes.
A generic logical model for cytokine signalling is shown in
The generic cytokine module has been replicated and adapted for different cytokines, taking into account the relevant receptor chains, and the specific JAK and STAT components downstream (see
Cytokines | Chains | Targets |
IFNG | IFNGR1, IFNGR2 | STAT1 |
IL4 | IL4RA, CGC | STAT5, STAT6 |
IL6 | IL6RA, GP130 | STAT3 |
IL10 | IL10RA,IL10RB | STAT3 |
IL12 | IL12RB1, IL12RB2 | STAT4 |
IL15 | IL2RB, CGC, IL15RA | STAT5 |
IL21 | CGC, GP130 | STAT3 |
IL23 | IL12RB1, GP130 | STAT3 |
IL27 | GP130, IL27RA | STAT1, STAT3 |
List of the cytokines considered in our model, each corresponding to an instance of the generic module shown in
To enable the analysis of the model in terms of stable states, we provisionally ignore SOCS-dependent negative feedbacks.
As IL2 plays a particular role in the control of Th cell differentiation and proliferation, it deserves a more detailed description. The IL2/IL2R module shown in
The 13 input components are colored in black. Ellipses denote Boolean components while rectangles denote ternary components. Green arrows denote activations, whereas red blunt ones denote inhibitions. A peculiar blue arrow denotes the unique dual interaction. The greyed-out components have been reduced to generate the regulatory graph displayed in
IL2 signalling is also involved in the activation of induced cell death (for a review, see
All the modules must then be integrated to build a comprehensive model. The TCR signalling module functions as an input, since it is not regulated by any other component. Several cytokine receptors share subchains and targets. For example, the common gamma chain (CGC) is shared by IL2, IL4, IL7, IL9 and IL15 receptors that lead to the activation of STAT5 (cf.
Which cell types and differentiation pathways can be predicted from the logical model just described? The different cell types correspond to attractors in the state transition graph,
As our regulatory network encompasses too many components to enable a direct analysis of the full state transition graph, we have applied the reduction method described in Section “Model reduction”. Regarding the selection of the components for reduction, we face a compromise between computational performance and biological readability. Selecting TCR, CD28, cytokines receptors and their subchains, along with the intermediate components RUNX3, IRF1, SMAD3, IKB and NFKB for reduction, we obtain a regulatory graph containing 34 components (see
This graph has been obtained by applying the reduction method described in Section “Model reduction” to the full model shown in
A further reduction leads to a graph conserving the 13 inputs, but only 12 internal nodes: TBET, GATA3, RORGT, FOXP3, STAT3, STAT5, IL2, IL4, IFNG, IL17, TGFB and proliferation. Using this compact model, we could compute the full state transition graphs for relevant input combinations. This led us to identify 28 context-dependent stable states (corresponding to the greyed-out cells in
Each of the four master genes considered (TBET, GATA3, RORGT and FOXP3) is positively auto-regulated. The first five rows correspond to the canonical Th cell subtypes expressing no (Th0) or a single master regulator (Th1, Th17, Th2, Treg). The remaining rows correspond to hybrid Th cell subtypes that express more than one of the master regulators,
Regulatory circuits are known to play a role in the emergence of essential dynamical properties (cf. Section “Feedback circuit analysis”). The regulatory graph presented in
Considering that each functional positive circuit can lead to two attractors, the four auto-activations can possibly generate
To analyse how cell fate depends on initial and environmental conditions, we have iterated several round of simulations, using the reduced model shown in
Each row corresponds to one prototypic environment, defined in terms of combinations of APC and of seven different cytokine inputs. Presence/absence of the different inputs is denoted by grey/white cells. The coloured tile code defined in the first column is used in
A grey cell denotes the activation of the corresponding component (column entries) for the corresponding stable state (row entries). Black cells denote higher activity levels (in the case of multi-level components). Note that the values of the input nodes are omitted here. A state stable for a given input combination may become unstable for other input values. Relationships between these stable states and selected environmental conditions (described in
To check the stability of the identified Th cell subtypes in changing environments, we have performed a series of simulations using each of the stable states listed in
This figure summarises several simulation rounds, displaying the context-dependent stable states (column entries) reached depending on eliciting initial states (row entries) and environmental conditions (coloured tiles). The coloured tile code for environmental conditions is defined in
The Th cell subtypes observed
In the absence of any input from the environment, resting Th0, Th1, and Th2 are the only three attractors. These three cell types can thus be considered as reference states, to which the other stable states can be associated depending on the expression of characteristic markers. However, these groups or
By and large, Th1 cells tend to remain within the Th1 state constellation, which includes activated, resting, and anergic variants. For example, stimulation by APC will activate resting Th1 into activated Th1 cells, inclusively in a pro-Treg environment. In either a pro-Th2 or pro-Treg environment, Th1 cells will become anergic, while they will tend to express ROR
Th2 differentiated cells expressing GATA-3 are even more robust than Th1 cells and will always remain within the Th2 constellation (denoted in blue in
A subset of Th17 cells expressing ROR
Treg cells can be maintained only in the presence of TCR-stimulus delivered by APC and of IL-2 produced by other T cells. However, in strong polarising environments, Tregs may differentiate into mixed cell types associated with the Th1 or Th2 cell constellations. For example, ROR
We have reconstructed the molecular network controlling the activation and differentiation of Th cells and asked how many stable states to expect, considering that these cells face a changing local environment during their life span.
Components and cross-regulatory links were extracted from the literature and, in some cases, from previous logical models
This cautionary remark notwithstanding, our current model recapitulates the differentiation of naive cells into Th1, Th2, Th17 and Treg subtypes. Strikingly, our model also gives rise to hybrid states expressing markers characteristic of two or more canonical cell types. Cyclic attractors (potentially corresponding to oscillatory behaviour) have been found only in restricted environmental situations, whose biological relevance remains to be assessed. Furthermore, our model analysis emphasises an unexpected plasticity of the canonical cell types. Indeed, according to the model, both Foxp3+ regulatory T cells and Th17 cells are highly plastic and labile, whereas Th1 and Th2 subtypes are more readily maintained across different environmental conditions.
Based on our results, canonical Foxp3+ regulatory T cells would not be truly lineage committed, but would rather correspond to a context-dependent stable state of the underlying regulatory network. This is surprising considering the fundamental role of these cells in avoiding autoimmunity. According to our model analysis, the maintenance of a regulatory T cell phenotype would require sustained TCR/CD28 and IL-2 signals. Indeed, the absence of TCR stimulation can lead to the loss of Foxp3 expression and a conversion into a Th0 phenotype. Depending on inputs, these cells can regain Foxp3 expression but also differentiate into other cell types. Our model further predicts that Treg cells may differentiate into Th1 or Th2 subtypes in proper polarising environments. This plasticity of regulatory Foxp3+ T cells is supported by several recent reports
In our model analysis, Treg lability clearly depends on the assumption that Foxp3 expression requires Stat5 and AP-1/NFAT transcriptional activities. This might turn out to be an oversimplification as we might have overlooked some mechanisms preventing regulatory cells to become conventional T cells and vice-versa. Some data suggest that locus epigenetic regulation is needed for sustainable Foxp3 expression
Now, if our prediction of pervasive T cell plasticity is correct, how can the classes of immune responses in which they predominate be so robust? This point is particularly intriguing in the context of natural tolerance, presumably depending on labile Tregs, or yet regarding persistent memory responses to bacterial infections. Here again, the stability of the responses could stem from regulatory feedbacks at the level of the T cell populations, as intercellular interactions and cell population dynamics might sustain and stabilise specific combinations of Th subtypes. Regarding Treg mediated tolerance, stable and robust coexistence of conventional Th cells and regulatory T cells has been obtained using a cross-regulatory model encompassing positive and negative feedback circuits at the population level in
Among the model stable states, those corresponding to hybrid cell types expressing two or more Th master regulators are particularly striking. Most of these hybrid cell types co-express ROR
Altogether, these observations support the existence of several of our predicted environment-dependent Th subtypes, thereby warranting the investigation of the other predicted hybrids (
The number of reported cytokines grows rapidly. Tentatively, some of these cytokines could be predominantly expressed by specific, yet undiscovered, cell types. One might thus wonder what part of our conclusions could be retained as our knowledge on the Th cell differentiation network grows. In this respect, we have extended earlier models for Th1/Th2 bipolarisation
However, even this well established paradigm has been recently challenged by reports demonstrating simultaneous and sustained GATA-3 and T-bet co-expression in reprogrammed Th2 cells
More generally, the characterisation of the roles of additional cytokines, transcription factors, and regulatory RNAs, or yet the delineation of epigenetic regulatory mechanisms, should enable proper model extensions and refinements, thereby leading to refined predictions, while preserving the functional roles of the most salient regulatory modules.
In conclusion, our results indicate that the pool of CD4 Th cells is highly heterogeneous and that the structure of the gene network promotes this heterogeneity. Diversity of antigen receptors and their crossreactivity are the most salient features of the adaptive immune system of the vertebrates, allowing the host to recognize and potentially react to numerous antigens. The heterogeneity and plasticity of Th cell types emphasised here might further contribute to the capacity of the immune system to deal with wide contingencies. Arguably, such plasticity does not fit well with the classical depiction of T cell differentiation potential in terms of a branching tree. Instead, our computational study points to a reticulate network of alternative, environment-dependent, differentiation and reprogramming events.
Complete annotated Th model. Compressed archive (with extension ZGINML) for the complete, annotated Th differentiation model, including the model file (XML file with extension GINML) and parameter files. This file can be directly opened with the freely available GINsim software.
(0.01 MB ZIP)
Reduced Th differentiation model. Compressed archive (with extension ZGINML) for the reduced Th differentiation model, including the model file (XML file with extension GINML) and parameter files. This file can be directly opened with the freely available GINsim software.
(0.01 MB ZIP)
Th differentiation model documentation. Documentation for the complete Th differentiation model,which include supporting data and links to relevant databases for each model component.
(0.18 MB PDF)
The collaboration between IGC and TAGC teams was triggered by the workshop