Petri Net computational modelling of Langerhans cell Interferon Regulatory Factor Network predicts their role in T cell activation

Langerhans cells (LCs) are able to orchestrate adaptive immune responses in the skin by interpreting the microenvironmental context in which they encounter foreign substances, but the regulatory basis for this has not been established. Utilising systems immunology approaches combining in silico modelling of a reconstructed gene regulatory network (GRN) with in vitro validation of the predictions, we sought to determine the mechanisms of regulation of immune responses in human primary LCs. The key role of Interferon regulatory factors (IRFs) as controllers of the human Langerhans cell response to epidermal cytokines was revealed by whole transcriptome analysis. Applying Boolean logic we assembled a Petri net-based model of the IRF-GRN which provides molecular pathway predictions for the induction of different transcriptional programmes in LCs. In silico simulations performed after model parameterisation with transcription factor expression values predicted that human LC activation of antigen-specific CD8 T cells would be differentially regulated by epidermal cytokine induction of specific IRF-controlled pathways. This was confirmed by in vitro measurement of IFN-γ production by activated T cells. As a proof of concept, this approach shows that stochastic modelling of a specific immune networks renders transcriptome data valuable for the prediction of functional outcomes of immune responses.


Connecting the network components with edges to represent the identified interactions.
a. IRF, IRF TP, and corresponding DNA binding sequence were connected with stimulatory edges (black arrows) including all possibilities detailed in the Table S3 (e.g. IRF8 can bind to EICE with ETS, to AICE with AP-1 or ISRE with IRF1), preserving the and/or logic (i.e. IRF8 cannot bind to ISRA without IRF1: transition "and", IRF 1 can connect to ISRA either on its own or hetero-dimerised with IRF8: transition "or"). The Boolean logic gates "and" and "or" have been recreated using two nodes to transition ("and") and two transition to node ("or") (Signal flow through "and" ad "or" gate is presented in Figure S2) f. Genes identified by ChIP-seq analysis of IRF1,4, and 8 were associated with the DNA binding sequence, and with the output biological process. Assumption: controlling IRF homo/heterodimer determines DNA binding sequence. If a gene can be controlled by two transcription factors/two DNA sequence (e.g. IL18 via IRF1 or IRF8/ETS complex) both possibilities were included in the diagram.

Adding entry transitions for input nodes: (transitions: black bars)
a. An entry transition was added before each entry node to allow setting up initial marking of the network and input of the numerical data.

Converting diagram edges into appropriate interactions (stimulatory: black arrows, inhibitory: red open diamonds)
a. Each edge drawn is initially a black stimulatory edge. To convert the interaction to an inhibitory, the arrow was replaces with an open diamond shape end. For clearer visualization the inhibitory edges are colored red.  presented in octagons on the right side of the diagram. The diagram is drawn in a Petri net notation, where the interacting elements of GRN (nodes, gene transcripts) are interspaced with transitions (vertical black lines, and black diamonds). The diagram captures the combinatory nature of immune activation, depending on the levels of expression, timing and interactions between the regulatory elements. The flow of the signal through the diagram can be modelled mathematically using experimental or simulated data and activity flow visualised in BioLayout Express3D.

b-e) Effect of signal transmission through "and" and "or" Boolean gates
Petri Net network motifs demonstrating the principles of signal flow through "and" (b,d) and "or" (c,e) gates with input from single (b,c) and multiple (d,e) transitions. Initial network marking = 100, token accumulation after gate are shown in the right column, 100 time blocks, 500 runs, simulation under the conditions of standard distribution. In silico profiles of gene expression in programmes "A" and "B" , measured at the output node when the input nodes are marked as per the gene expression values during LCs stimulation with TNF-α and TSLP, Signalling Petri Nets: BioLayout Express3D, 100 time blocks, 500 runs. b) Expression profiles of individual genes in "Programme A" as measured in the microarray experiment. c) Expression profiles of individual genes in "Programme B" as measured in the microarray experiment.

Figure S4: Ability of LC to cross-present antigens is modified by TNFα and TSLP.
Activation of antigen-specific CD8+ T cells by medium (white), TNFα (grey), TSLP (black) and a combination of TNFα and TSLP (black checkerboard grey) matured LCs, pulsed with a long peptide antigen requiring cross-presentation, IFN-γ production measured in co-culture ELISpot assay, n=2 in triplicate, mean +/-SE. Human epidermal biopsies (a) were exposed to PI3Kγ inhibitor or control media for 48h.      Table S5: Genes regulated by expression programme "A" and "B" in the IRF-GRN        c d e f Table S1. Search strategy to identify components of the IRF GRN network search term number of publications "Interferon regulatory factor" or IRF and antigen presentation 71 "Interferon regulatory factor" or IRF and dendritic cell and T cell stimulation 22 "Interferon regulatory factor" or IRF1 or IRF4 or IRF8 and *transcripton partner* as per the transcription partner list 510 Interferon regulatory factor or IRF1 or IRF4 or IRF8 and ChIP-seq 15