Post-transcriptional regulatory feedback encodes JAK-STAT signal memory of interferon stimulation

Abstract

Immune cells fine tune their responses to infection and inflammatory cues. Here, using live-cell confocal microscopy and mathematical modelling, we investigate interferon induced JAK-STAT signalling in innate immune macrophages. We demonstrate that transient exposure to IFN-γ stimulation induces a long-term desensitisation of STAT1 signalling and gene expression responses, revealing a dose- and time-dependent regulatory feedback that controls JAK-STAT responses upon re-exposure to stimulus. We show that IFN-α/β1 elicit different level of desensitisation from IFN-γ, where cells refractory to IFN-α/β1 are sensitive to IFN-γ, but not vice versa. We experimentally demonstrate that the underlying feedback mechanism involves regulation of STAT1 phosphorylation but is independent of new mRNA synthesis and cognate receptor expression. A new feedback model of the protein tyrosine phosphatase activity recapitulates experimental data and demonstrates JAK-STAT network’s ability to decode relative changes of dose, timing, and type of temporal interferon stimulation. These findings reveal that STAT desensitisation renders cells with signalling memory of type I and II interferon stimulation, which in the future may improve administration of interferon therapy.

Variability of the innate immune response is globally constrained by transcriptional bursting

“…These included the highly variable and abundant genes including chemokine family Ccl5, Ccl4, Ccl3, Ccl2 as well as IL1b and TNFa. While the scRNA-seq can be in principle treated as time-series data (e.g., across the replicates from individual mice) [34], our current understanding of TLR signalling suggest that due to endotoxin resistance and desensitisation [56][57][58], the regulatory network, and thus model structures and parameters, are time-varying rather than stationary [59]. We therefore treated each data time-point (and replicate) separately, which also allowed more efficient implementation to fit 1,507 mouse, and 1,079 orthologue conditions.…”