Fig. 3.

Interactions between miRNA target site sets in the 3′ UTR sensors appear antagonistic. a Schematic for single-input sensors bearing one set of target sites (top) and 3-input sensors bearing three different sets of target sites (bottom) where all miRNA target sites are located in the 3′ UTR. b miRNA repression data and model fits for different single-input sensors. Plots are ordered by increasing number of high activity miRNA target site sets. miRNA activities are denoted as low (L), high (H), or very high (H*). c Comparison of predicted miRNA repression to data obtained for 3-input sensors. Predictions for 3-input sensors are computed based on M and K_m parameters measured from single-input sensors. Using the assumption that miRNA target sites act antagonistically yields predictions (red) which are equivalent to taking the maximum activity (minimum mKate2 expression) of the three single-input sensors for each EBFP2 expression level. Additive predictions (gray dashed) are made using the Chou–Talalay method. Synergistic predictions are made by multiplying fold repression from each of the single-input sensors within each EBFP2 transfection bin. Predictions using the antagonistic or additive models are markedly better than those from a synergistic model, indicating that miRNA repression for multi-input sensors is not a simple multiplicative effect. While antagonistic and additive models were close in these examples, predictions can diverge drastically when several miRNA inputs are combined and when input activities are very similar. d Analysis of prediction errors for 3-input sensors. Errors are measured by computing the maximum fold difference between predictions and data across all bins of EBFP2 expression (max fold error) or by computing the mean squared error (MSE). For both metrics and for all tested combinations of low/high/very high (*) activity miRNAs, antagonistic interaction explains the data better than an additive or synergistic interaction