Figure 3. A kinetic model suggests a delayed negative feedback loop through USP18.

(A) A diagram for the simple kinetic model of the IFN-driven gene regulatory network. (B) Fitting errors between simulations and the data with different assigned values of the delay time in USP18 upregulation. (C) Amounts of P_IRF9-YFP induction by the second IFN-α stimulation under different pretreatment conditions, from experimental data (solid bars) and model simulations with the best-fit parameters (open bars). Data were from Figure 1F and were normalized to the non-pretreatment condition (control). (D) Schematic of experimental design with repetitive IFN pulses versus a sustained IFN input. (E) Model prediction of the responses to pulsatile versus sustained IFN inputs in the presence and absence of USP18. Results were normalized to the amount of induction to the pulsatile IFN input in WT. (F) Experimental data of the responses to pulsatile versus sustained IFN inputs in WT and USP18-KD cells. The error bars represent standard deviations of single-cell data.

Figure 3—figure supplement 1. Model fitting results and the parameter analysis.

(A) A diagram for the simple kinetic model of the IFN-driven gene regulatory network with reaction parameters labeled. (B) Fitting results with different delay times. The model fitting results (solid lines) were shown for delay times from 1 to 20 hrs while keeping all the other parameters free (Figure 3B) and were compared with experimental data (solid circles) under different pretreatment conditions. The delay time of 8 hrs gave smallest error between simulation and data (dashed line). (C) Model simulation of the time trace of IRF9 induction in response to 100 ng/ml IFN-α. The model was fit to the end point data from pretreatment experiments with different durations, as well as the time trace data of IRF induction to IFN stimulation. The simulated time trace from the best-fit parameters were shown as the dashed line. The data were from Figure 1C and were represented as the mean (yellow solid lines) and + / - standard deviation (SD) of single cells (shaded areas). (D) Sensitivity analysis of the parameter values. For each parameter, its value was changed across a wide range, while all the other parameters remained at the best-fit values. The simulated results (solid lines) were shown and compared with experimental data (solid circles) under different pretreatment conditions. The dashed line represents the best-fit value. (E) Roles of the positive and feedback loops on model performance. (i) Original model with both positive and negative feedback loops, (ii) model without positive feedback loop, and (iii) model without negative feedback loop were compared and Akaike information criterion (AIC) was computed using the best-fit parameter set for each model and was shown in the figure. Experimental data (solid bars) and model simulations with the best-fit parameters (open bars) for P_IRF9-YFP induction by the second IFN-α stimulation under different pretreatment conditions were shown for each model. Data were from Figure 1F and were normalized to the non-pretreatment condition (control).

Figure 3—figure supplement 2. Pulsatile IFN-α treatment induces higher ISG expression in single cells.

(A) The violin plots showing single-cell distributions of P_IRF9-YFP induction upon 5 × 8-hr pulsatile (blue) or 40-hr sustained (red) IFN-α treatments in WT (top) and USP18-KD (bottom) cells. The normalized mean levels of P_IRF9-YFP induction were 1 (WT, pulsatile), 0.6204 (WT, sustained), 1 (USP18-KD, pulsatile), and 1.0796 (USP18-KD, sustained), respectively. The coefficients of variance (CVs) for IRF9 induction were 0.1846, 0.1953, 0.2419, and 0.2273, respectively. (B) Scatterplots showing STAT1-mCherry versus P_IRF9-YFP induction in single cells in response to 5 × 8 hr pulsatile or 40-hr sustained IFN-α treatments in WT (top) and USP18-KD (bottom) cells. Fluorescent signals without IFN-α treatment were used as control.