Characterization and mitigation of gene expression burden in mammalian cells

Abstract

Despite recent advances in circuit engineering, the design of genetic networks in mammalian cells is still painstakingly slow and fraught with inexplicable failures. Here, we demonstrate that transiently expressed genes in mammalian cells compete for limited transcriptional and translational resources. This competition results in the coupling of otherwise independent exogenous and endogenous genes, creating a divergence between intended and actual function. Guided by a resource-aware mathematical model, we identify and engineer natural and synthetic miRNA-based incoherent feedforward loop (iFFL) circuits that mitigate gene expression burden. The implementation of these circuits features the use of endogenous miRNAs as elementary components of the engineered iFFL device, a versatile hybrid design that allows burden mitigation to be achieved across different cell-lines with minimal resource requirements. This study establishes the foundations for context-aware prediction and improvement of in vivo synthetic circuit performance, paving the way towards more rational synthetic construct design in mammalian cells.

Fig. 1. Resource sharing and the origin of gene expression burden.

a Characterization of gene expression burden. Expression of independent exogenous genes impacts on host cellular resources. Thus, perturbations in one gene’s expression (hereby named X-tra) affect the expression of a second gene (hereby named capacity monitor). b Modeling of gene expression in a resource-limited environment. Modeling of gene expression is generally performed under the assumption of unlimited resources. A simple framework enables the straightforward transformation of such a model to a system that incorporates resources explicitly. The transformation involves a simple function that scales the original reaction rate. c Mitigation of gene expression burden. A simple microRNA-based circuit motif is capable of mitigating the burden-induced coupling of X-tra and the capacity monitor. It should be noted that the dynamic range of X-tra also slightly decreases as a consequence of mitigation. However, as it will be discussed in the results section and shown in Supplementary Fig. 19, the absolute expression of X-tra is higher with mitigation.

Fig. 2. Burden imposed by genetic circuits in mammalian cells.

a Left: As the total plasmid amount increases, the total expression plateaus. Right: Titration of two plasmids expressing the fluorescent proteins mCitrine and mRuby3 from EF1α promoters in ratios from 1:4 to 4:1 (total of 50 ng, top right; or 500 ng of DNA, bottom right). N = 3 biological replicates. Source data are provided as a Source Data file. b Two plasmids were co-transfected, one constitutively expressing capacity monitor and tTA from a strong constitutive promoter and the other expressing X-tra from a tTA responsive promoter. Capacity monitor levels counterbalance the increase in X-tra expression. Flow cytometry data are normalized to the expression at maximal Dox. N = 3 biological replicates. Source data are provided as a Source Data file. c mRNA quantification of X-tra and a capacity monitor expressed at different molar ratios. As the X-tra increases, the mRNA levels of the capacity monitor decreases. N = 4 biological replicates. qPCR analysis was performed 48 h post-transfection and data show fold change ± SE. Source data are provided as a Source Data file. d Cells transfected with a plasmid expressing two fluorescent proteins from a bidirectional promoter were sorted according to high, intermediate, or no fluorescence (Supplementary Fig. 4) for mRNA extraction. mRNA levels expressed from endogenous genes decrease in cells with intermediate and high fluorescence. N = 3 biological replicates. Data show fold change ± SE. Individual values are plotted in Supplementary Fig. 28. Source data are provided as a Source Data file. e Capacity monitor levels are higher with an HDV ribozyme rapidly degrading the capacity monitor mRNA than with an inactive mutant, suggesting a sequestration of transcriptional resources. N = 3 biological replicates (N = 2 for HDV−, 1.6 ng/μL DOX). Source data are provided as a Source Data file. f The synthetic intron shows higher X-tra levels compared to a control and leads to reduced capacity monitor levels. N = 4 biological replicates. Source data are provided as a Source Data file. g Repressed X-tra expression leads to increased capacity monitor levels. N = 2 biological replicates for L7Ae and N = 4 for Ms2-cNOT7. Source data are provided as a Source Data file. h When X-tra is downregulated by miR-221 endogenously expressed in HEK293T cells, the capacity monitor levels increase. All flow cytometry data were acquired 48 h post-transfection and are plotted as mean ± SE. SE standard error, r.u. relative units. N = 2 biological replicates. Source data are provided as a Source Data file. Unpaired two-sided T-test. P value: ****<0.0001, ***<0.0005, **<0.005, *<0.05.

Fig. 3. Impact of miRNA target sites number and location on burden.

a Schematics of experimental design to infer miRNA-mediated cellular resources redistribution. EGFP (capacity monitor) and mKate (miRNA sensor) are encoded on the same bidirectional CMV promoter plasmid. One or 3 TS for miR-31 (TS) are added either in the 3′ or 5′UTR of mKate. Control: no miR-31 TS. Hypothesis: in the absence of miR-31 regulation, capacity monitor and miRNA sensor are expressed to a certain level (top). In the presence of miR-31, lower miRNA sensor levels correlate with higher capacity monitor expression (middle). This condition is reversed by an miR-31 inhibitor (bottom). b Fold change of miRNA sensor and capacity monitor protein levels compared to control (set to 1). EGFP increases up to fivefold with the strongest downregulation of mKate (3 TS 5′UTR). Flow cytometry data were acquired 48 h post-transfection and are plotted as mean ± SE. SE standard error, r.u. relative units. N = 6 biological replicates. Source data are provided as a Source Data file. Unpaired two-sided T-test. P value: ****<0.0001, **<0.005, *<0.05. c When miR-31 activity was impaired by a miR-31 inhibitor, the rescue of mKate expression corresponds to reduced EGFP levels, whereas both fluorescent proteins do not vary in the control. The heatmaps represent the fold change derived by flow cytometry data, calculated as the ratio between the geometric mean of six biological replicates and the corresponding geometric mean in the control condition. Source data are provided as a Source Data file. Bar plots and statistical analysis are reported in Supplementary Fig. 12.

Fig. 4. A resource-aware mathematical modeling framework.

a General framework for transforming molecular interaction network models. Existing models of molecular interaction networks can be transformed to include shared limiting resources by substituting k_i, the reaction rate of a resource-limited production, with k^eff_i. Shown above an exemplary resource-limited production are the detailed interactions between the substrate and the shared resource. b Limited shared resources reproduce non-monotonous dose response in open-loop and incoherent feedforward circuit topologies. On the left, a graphical representation of a model for both the open-loop (OLP) and incoherent feedforward (IFF) topologies from Lillacci et al.. Transcriptional activation is modeled by a Hill-type function. The solid arrows denote reactions assumed to follow the law of mass action. The model incorporates resources as introduced in panel a. These reactions are depicted as double-headed arrows. The model was fit to data obtained by transiently transfecting HEK293T cells with increasing amounts of plasmid encoding tTA-Cerulean. The data and the fit are shown on the right. c Limited shared resources reproduce non-monotonous dose-response in feedback and hybrid circuit topologies. The model shown on the left is the same as in panel b with an additional negative feedback from miR-FF4 to tTA-mRNA. These topologies correspond to the feedback (FBK) and hybrid (HYB) topologies from Lillacci et al. The activation of gene expression by tTA-Cerulean is modeled by a Hill-type function as shown in the center. Reactions with double-headed arrows denote resource-limited production reactions as introduced in panel a. Solid arrows are assumed to follow the law of mass action. The model was fit to experimental data obtained from transient transfections with increasing amounts of plasmid encoding tTA-Cerulean. A description of the models can be found in Supplementary Note 4 and the parameter values obtained by fitting are summarized in Supplementary Table 7. Data were obtained 48 h after transfection and are plotted as mean ± SE. SE standard error. N = 3 biological replicates. Source data are provided as a Source Data file.

Fig. 5. Mitigating the effects of resource limitation with microRNA-based iFFL.

a The microRNA-based incoherent feedforward loop (iFFL) motif. b Mitigation system based on endogenous microRNA. At high copy number of the X-tra, resources are drawn away from the production of the GOI and miR-31. By sensing the resource availability and repressing the GOI less when there are fewer resources, the miRNA reduces the effect of limited resources. c Two plasmids were co-transfected into H1299 cells which respectively express the X-tra and GOI genes (EGFP and mKate respectively (b)), and the molar ratio of the X-tra:GOI plasmid was progressively increased. The presence of miR-31 TS in mKate 5′UTR mitigates effects due to resource sharing. The parameter values obtained by fitting are summarized in Supplementary Table 8. N = 3 biological replicates. d Mitigation system based on synthetic miRNA. In the presence of many copies of the X-tra gene, resources are drawn away from the production of both the GOIs and the miR-FF4. Due to lower production of miR-FF4 the GOIs are less repressed. This compensates for the reduced availability of resources. e A plasmid encoding both the fluorescent protein mCitrine and an intronic microRNA expressed from the mRuby3 gene (GOI₁, GOI₂ and miR-FF4 (d)) was co-transfected into HEK293T cells with increasing amounts of a plasmid expressing the X-tra gene (miRFP670 (d)). The impact of resource limitation on both GOIs was reduced when they contained three miR-FF4 targets in their 3′UTRs compared to when they contained three mismatched miR-FF5 targets. The parameter values obtained by fitting are summarized in Supplementary Table 9. N = 3 biological replicates. Source data are provided as a Source Data file. A description of the models can be found in Supplementary Note 5. Flow cytometry data were acquired 48 h post-transfection and are plotted as mean ± SE. SE standard error, r.u. relative units.