A Minimal Regulatory Network of Extrinsic and Intrinsic Factors Recovers Observed Patterns of CD4+ T Cell Differentiation and Plasticity

CD4+ T cells orchestrate the adaptive immune response in vertebrates. While both experimental and modeling work has been conducted to understand the molecular genetic mechanisms involved in CD4+ T cell responses and fate attainment, the dynamic role of intrinsic (produced by CD4+ T lymphocytes) versus extrinsic (produced by other cells) components remains unclear, and the mechanistic and dynamic understanding of the plastic responses of these cells remains incomplete. In this work, we studied a regulatory network for the core transcription factors involved in CD4+ T cell-fate attainment. We first show that this core is not sufficient to recover common CD4+ T phenotypes. We thus postulate a minimal Boolean regulatory network model derived from a larger and more comprehensive network that is based on experimental data. The minimal network integrates transcriptional regulation, signaling pathways and the micro-environment. This network model recovers reported configurations of most of the characterized cell types (Th0, Th1, Th2, Th17, Tfh, Th9, iTreg, and Foxp3-independent T regulatory cells). This transcriptional-signaling regulatory network is robust and recovers mutant configurations that have been reported experimentally. Additionally, this model recovers many of the plasticity patterns documented for different T CD4+ cell types, as summarized in a cell-fate map. We tested the effects of various micro-environments and transient perturbations on such transitions among CD4+ T cell types. Interestingly, most cell-fate transitions were induced by transient activations, with the opposite behavior associated with transient inhibitions. Finally, we used a novel methodology was used to establish that T-bet, TGF-β and suppressors of cytokine signaling proteins are keys to recovering observed CD4+ T cell plastic responses. In conclusion, the observed CD4+ T cell-types and transition patterns emerge from the feedback between the intrinsic or intracellular regulatory core and the micro-environment. We discuss the broader use of this approach for other plastic systems and possible therapeutic interventions.


Construction of the logical functions
We considered that a node is active if there is enough amount of protein or gene expression to be functional and affect the differentiation of CD4+ T lymphocytes. A transcription factor is active if it is present in enough quantity and in a conformation that can alter the expression of its target genes. A transcription factor or cytokine is active if it is present in enough quantity and in a conformation that can form a functional complex with its receptor. A receptor is active if it forms a complex that can activate its downstream signaling. A STAT proteins is active if it is phosphorylated and forms a dimer capable of translocating to the nucleus and affecting the expression of its target genes.

Basal levels
A protein or gene may be expressed at a basal level, but does not necessarily affect the differentiation of the cell at that level of expression. For example, GATA3 is necessary for T cell maduration and for CD4+ T-cell survival and maintenance. The deleterious mutation of GATA3 is letal, and Lck-Cre conditional deletion models lack CD4+ T cells or have impaired survival and maintainance. GATA3 high also drives the differentation into Th2 (Ho, Tai and Pai 2009). In this case we considered that the basal level of GATA3 low corresponded to zero, while GATA3 high was one.

Weak interactions
Weak interactions were ignored in our model. Interactions between genes and proteins are weak when they increase or decrease the expression of a gene or protein but are not necessary or sufficient to cause changes in differentiation. For example, The IL-2 receptor (IL2-R) is necessary for the activation of CD4+ T cells and plays a central tole in the differentiation towards Th2 and iTreg. IL-2R is composed of three subunits IL-2Rα, IL-2Rβ, γ c . The three subunits together form a high affinity receptor, while IL-2Rβ and γ c form a medium affinity receptor, both complexes are functional. IL2 increases the expression of IL-2Rα and IL-2Rβ and Foxp3 increases the expression of IL-2Rα. The result is that the IL-2R can form a functional complex (IL2R = 1) in the presence of IL-2 with or without Foxp3, even if the transcription factor affects its expression levels and affinity (Liao 2011).

Boolean Logic Reduction Method
To simplify the network we employed a Boolean reduction method proposed in Villarreal et al, 2012. For simplicity, we illustrate only the simplification scheme of the interactions between IL-2 and Foxp3. Interleukin 2 (IL-2) can be produced by the T CD4+ lymphocytes or by other cells of the immune system (IL2e). IL-2 binds the IL-2 receptor (IL-2R), which causes the phosphorylation and dimerization of STAT5. The phosphorylation of STAT5 can be inhibited by SOCS1, which binds the IL-2R. STAT5 activates the transcription of IL-2, Foxp3 and increases the transcription of IL-2R. Foxp3 can induce its own transcription and inhibit the transcription of IL-2. These interactions can be characterized by a set of logical propositions which satisfy the following mapping: IL2 t+1 = IL2e t or (ST AT 5 t and not F OXP 3 t ) IL2R t+1 = IL2 t and not SOCS1 t ST AT 5 t+1 = IL2R t F OXP 3 t+1 = ST AT 5 t and F OXP 3 t (1) Considering that the expression level of node N at a time t is represented by N t the attractors (steady states) that represent different phenotypes are determined by the condition N t+1 = N t . In that case, the mapping becomes a set of coupled Boolean algebraic equations. This results in the identity: (3) We employ this identity to determine the system's attractors: ST AT 5 = IL2 and not SOCS1 (4) ST AT 5 = (IL2e or (ST AT 5 and not F OXP 3)) and not SOCS1 Thus, the regulatory network attractors are summarized by the expression values of the nodes pertaining to a concise set of Boolean expressions: ST AT 5 = (IL2e or (ST AT 5 and not F OXP 3)) and not SOCS1 F OXP 3 = ST AT 5 and F OXP 3 (6)

Reduction of logical regulatory graphs
To verify the Boolean approach we compared our results with those obtained with the software GINsim (Naldi et al, 2009). GINsim uses decision diagrams to iteratively remove regulatory components and actualizes the components to maintain the indirect effects. The method preserves the dynamical properties of the original model. The simplification with GINsim returns a similar network as the one obtained with the boolean logic reduction method.