Close this search box.

Engineering a synthetic gene circuit for high-performance inducible expression in mammalian systems – Nature Communications

In silico design and analysis of synthetic circuits for high-performance inducible gene expression

Among the potential gene network motifs, we focussed on those that may yield reduced leakiness levels14. We thus mathematically modelled and compared three alternative circuit topologies for inducible gene expression as shown in Fig. 1b, against the naïve configuration (NC): (i) the coherent feedforward loop type 4 (CFFL-4)15; (ii) the mutual inhibition (MI) topology14; and (iii) a combination of these two topologies that we named Coherent Inhibitory Loop (CIL). All these circuits make use of an additional species Y to inhibit the reporter gene Z in the absence of the inducer molecule, thereby suppressing leaky expression. We used ordinary differential equations and dynamical systems theory to analyse the performance of these three networks, assuming realistic biological parts (Supplementary Note 1).

Analytical results and numerical simulations of the circuits, when using the very same parameters for the common biological parts, confirmed that all three exhibit improved performances over the naïve configuration, in terms of lower leakiness, high maximum expression, and increased fold induction, as reported in Fig. 1c-e and Supplementary Note 1, albeit with notable differences. In the CFFL-4, the leakiness is smaller than the one of the NC thanks to the inhibitory action of Y over Z, in the absence of the inducer molecule (Fig. 1c); however, as X does not fully repress Y upon inducer molecule treatment, the maximal expression of Z is also smaller (Fig. 1d), thus leading to only a modest increase in Fold Induction (Fig. 1e). The MI improves on the CFFL-4 in terms of maximum expression (Fig. 1d), as Y is now repressed also by Z in addition to X. The CIL combines the advantages of both circuits, and it exhibits the best performance as compared to the NC configuration in terms of all the three features, as shown in Fig. 1c-e. To further explore the robustness of these findings, we conducted additional numerical simulations by varying the model’s parameters, whose results are shown in Fig. 1f and Supplementary Note 1. For all the parameter values tested, the CIL circuit exhibited the best performance whereas the CFFL-4 was the worst. Based on these analyses, we decided not to biologically implement a CFFL-4 system and instead focused on the biological implementation of the MI and CIL circuits.

Experimental implementation of the mutual inhibition (MI) circuit by means of pre-gRNA processing of the CasRx endoribonuclease

To experimentally implement mutual inhibition (Fig. 2a), we looked for a biological implementation that was compact and could be applied to any gene of interest Z. We thus turned to CRISPR-Cas endoribonucleases which have been recently repurposed to act as post-transcriptional regulators by exploiting their pre-gRNA processing mechanisms16. Indeed, CRISPR-endoribonucleases can cleave specific short sequences known as direct repeats (DRs) on their cognate pre-gRNAs, generating shorter guide RNA (gRNA) sequences; hence, these DRs have been repurposed as cleavage motifs to stabilize or degrade user-defined mRNA transcripts by placing them in the mRNA untranslated regions (UTRs)16. Specifically, in our implementation shown in Fig. 2a, we employed the CasRx endoribonuclease to implement species “Y”, while species “Z” is the Gaussia Luciferase (gLuc) reporter gene bearing the DR sequence in its 3’UTR. Because of the CasRx’s distinctive feature of irreversibly bind its processed gRNA17, we reasoned that this configuration could implement a mutual inhibition between species Y and Z. Here, Y is able to negatively regulate Z, as the CasRx cleaves the DR in the 3’UTR of the gLuc mRNA thus leading to the loss of its polyA tail and subsequent degradation; at the same time, we assumed that Z could be able to inhibit Y by “sponging out” the CasRx, which irreversibly binds to the DR and it is thus unable to cleave additional Z mRNAs.

Fig. 2: The CASwitch inducible gene expression system.
figure 2

a Experimental implementation of the mutual inhibition. CasRx acts as species Y. The Gaussia Luciferase (gLuc) with a Direct Repeat (DR) in the 3’Untranslated Region (UTR) acts as species Z. CasRx binds to the DR and cleaves the polyA tail (AAA) of the gLuc mRNA leading to its degradation, thus achieving Y-mediated repression of Z. Following cleavage, the CasRx irreversibly binds to the DR forming the gRNA-Cas binary complex which cannot cleave additional mRNAs, thus possibly implementing the Z-mediated repression of Y. b Experimental validation of CasRx-mediated mRNA degradation. Cells were transfected with CasRx and gLuc plasmids at the indicated relative concentrations. The bar-plot reports the mean Relative Luciferase in arbitrary units (A.U.) obtained by dividing the average Luciferase A.U. value at each molar ratio by the average Luciferase A.U. value in the absence of CasRx. Error bars correspond to the standard deviation. n = 4 biological replicates (white dots). c CASwitch v.1.: rtTA3G and CasRx are constitutively expressed from the pCMV promoter, while gLuc with the DR is placed downstream of the pTRE3G promoter. d, e Experimental validation of CASwitch v.1 (red) and comparison with the Tet-On3G expression system (black) at the indicated concentrations of doxycycline. n = 4 biological replicates. Relative Luciferase A.U. is computed as the Luciferase A.U. value of each data point divided by the average value of the Tet-On3G system at 1000 ng/mL, in both log-scale and linear-scale, and in (e) as fold-induction computed as the Luciferase A.U of each data point divided by the average value in the absence of doxycycline. f The CASwitch v.2: rtTA3G is constitutively expressed from a pCMV promoter, CasRx is driven by the pCMV/TO that can be repressed by the rtTA3G, while the gLuc with the DR is placed downstream of the pTRE3G promoter. g, h Experimental validation of CASwitch v.2 (green) and comparison with the state-of-the-art Tet-On3G gene expression system (black) at the indicated concentrations of doxycycline. n = 4 biological replicates. MI: Mutual Inhibition circuit topology; CIL: Coherent Inhibitory Loop circuit topology. Source data are provided as a Source Data file.

To experimentally test this hypothesis, we co-transfected HEK293T cells with CasRx along with one of three different gLuc transcript variants, as reported in Fig. 2b. These variants bear different numbers of DR motifs at their 3’ UTR: either no DR motif, one DR motif, or four DR motifs (4xDR). Our rationale was that by introducing more than one DR motif, we could “sponge” CasRx more effectively and thus alleviate repression of the target gLuc mRNA. Indeed in this scenario, one gLuc-4xDR mRNA should able to bind four CasRx, rather than only one, as in the case of the gLuc-DR. Results are shown in Fig. 2b: in the absence of CasRx, all the three gLuc transcripts yield the same luciferase expression level, independently of the number of DRs in their 3’UTR, thus excluding perturbations of mRNA stability caused by the DR itself. In the case of the gLuc-DR transcript (with one DR), the relative increase in the amount of co-transfected CasRx resulted in an exponential decrease in luciferase expression, with up to 100-fold reduction in luminescence. On the contrary, for the gLuc-4xDR, the CasRx repression efficiency was strongly reduced, thus supporting the hypothesis of a DR-mediated “sponging” of CasRx, although we cannot exclude alternative mechanisms. Encouraged by these results, we sought to implement the MI circuit using the CasRx endoribonuclease, developing the CASwitch v.1 system, as shown in Fig. 2c.

We chose as species X the tetracycline transactivator (rtTA3G) transcriptional factor, which is a fusion protein that combines a tetracycline-responsive DNA-binding domain with a strong transcriptional activation domain13. In the presence of the doxycycline, rtTA3G binds to multiple copies of the tetracycline operon (TO) sequence present in its cognate pTRE3G synthetic promoter, thereby inducing the expression of the downstream gene of interest18. In the CASwitch v.1 system, both the CasRx and the rtTA3G are constitutively expressed from the CMV promoter, while the gLuc harbours one DR in its 3’UTR and it is placed downstream of the pTRE3G promoter, as schematically shown in Fig. 2c.

We experimentally compared the performances of the CASwitch v.1 and the Tet-On3G by transiently transfecting HEK293T cells with three plasmids: (1) the pCMV-rtTA3G, (2) the pTRE3G-gLuc-DR for the CASwitch v.1, or the pTRE3G-gLuc for the Tet-On3G, and (3) the pCMV-CasRx at a relative molar ratio of 1:5:1. Observe that for the Tet-On3G system, the gLuc has no DR in its 3’UTR, but we co-transfected the CasRx anyway to exclude potential biases caused by cellular burden. We then quantified gLuc expression by luminescence measurements at varying concentrations of doxycycline. Results are reported in Fig. 2d-e and demonstrate that the CASwitch v.1, in the absence of doxycycline, strongly reduces leaky gene expression by >1-log when compared to the Tet-On3G system (Fig. 2d); at the same time, the maximal expression upon doxycycline treatment was only slightly reduced (Fig. 2d). Notably, the reduced leakiness and the retention of high maximal expression resulted in a very significant gain in terms of fold-induction by more than 1-log (Fig. 2e).

To further evaluate the robustness of the CASwitch v.1 system, we repeated the same experiments at higher relative concentrations of CasRx, as reported in Supplementary Fig. 1. This resulted in a further suppression of leakiness, but also in a reduction of the maximal achievable expression, suggesting that controlling CasRx expression is an important design parameter to achieve the desired inducible system properties.

Overall, our results demonstrate that the constitutively expressed CasRx, combined with its cognate direct repeat (DR) in the 3’UTR of a target mRNA, can serve as a plug-and-play strategy to significantly enhance the performance of transcriptional inducible gene expression systems.

Experimental implementation of the Coherent Inhibitory Loop (CIL) by transcriptional inhibition of CasRx boosts CASwitch performance

We set out to further enhance the performances of the CASwitch v.1 by specifically focusing on the increase in the maximal achievable expression upon doxycycline treatment. To this end, guided by the modelling results in Fig. 1b, we sought to biologically implement the CIL circuit by replacing the constitutive pCMV promoter driving the CasRx with a modified version, named pCMV/TO, as shown in Fig. 2f. The pCMV/TO promoter has two TO sequences downstream of the TATA binding box of the pCMV19, hence, upon doxycycline administration, rtTA3G binds to these elements and causes a steric hindrance to the PolII resulting in a partial repression of CasRx transcription. We first confirmed the effective doxycycline-dependent inhibition of the pCMV/TO promoter (Supplementary Fig. 2). Subsequently, we verified that switching the pCMV promoter with the pCMV/TO promoter did not affect CasRx expression and its effect on its downstream target (Supplementary Fig. 3). Finally, we proved that the pCMV/TO enables doxycycline-mediated repression of CasRx expression and relief of CasRx-mediated degradation of the target mRNA (Supplementary Fig. 4).

We thus leveraged the pCMV/TO-mediated transcriptional control of the CasRx to implement the CASwitch v.2, as shown in Fig. 2f, and we experimentally compared its performances to that of Tet-On3G system. Results are reported in Fig. 2g,h in terms of luciferase expression at varying concentrations of doxycycline. The CASwitch v.2 exhibited more than 1-log reduction in leakiness compared to the art Tet-On3G system, yielding results similar to those obtained with the CASwitch v.1 (Fig. 2g); this time, however, in agreement with the in silico analysis, it was able to fully recover the maximal achievable expression to the level of the original Tet-On3G (Fig. 2g), thus leading to a very large amplification of fold induction levels of up to 3000-fold (Fig. 2h).

To assess the robustness of CASwitch v.2, we tested its performance against that of state of the art Tet-On3G system by: (i) changing the plasmid molar ratio among the circuit components; (ii) testing it in a different mammalian cell line; and (iii) changing the promoter that drives the rtTA3G.

Results on the performance against changes in plasmid molar ratios are presented in Supplementary Fig. 5. Different amounts of plasmids can affect basal and induced levels of gene expression; hence one may presume that the Tet-On3G system performance could be improved by simply changing the plasmid ratios. Interestingly, the CASwitch v.2 (red and blue lines in Supplementary Fig. 5b,c) maintains its enhanced performance over the Tet-On3G system (yellow and green lines) independently of the plasmid ratio used.

Results on the performance of the CASwitch v.2 in HeLa cells are shown in Supplementary Fig. 6a-b, where it is evident that it retains its improved performance over the Tet-On3G system, consistently with the results observed in HEK293T cells.

Results on the impact of replacing the pCMV promoter driving rtTA3G in the CASwitch v.2 system with two alternative promoters with lower expression strengths (pEF1a and pPGK) are shown in Supplementary Fig. 7. In the cases tested, the CASwitch v.2 exhibited a better performance versus the Tet-On3G system by exhibiting a lower leakiness while maintaining the maximal expression (Supplementary Fig. 7b) thus leading to a higher fold induction (Supplementary Fig. 7c), with a slight decrease in performance for the weakest pPGK promoter. Notably, the use of the pEF1a yielded the highest fold induction, hence we chose to express the rtTA3G from this promoter in following experimental applications of the CASwitch v.2.

Overall, these results confirm that the CASwitch v.2 represents a general strategy to endow transcriptional inducible gene expression system with very low leakiness but with unaltered maximal expression, hence resulting in very large gain in fold induction.

CASwitch v.2 is an easy-to-implement solution to improve the performance of mammalian whole-cell biosensors

As the CASwitch v.2 greatly enhances the fold induction levels of the Tet-On3G inducible gene expression system, we decided to deploy it to increase the performance of established transcription-based biosensors20. As a case in point, we deployed the CASwitch v.2 to improve the performance of a previously published copper biosensor20 in mammalian cells, as shown in Fig. 3a. In this biosensor, a luciferase reporter gene is placed downstream of a synthetic metal-responsive promoter (pMRE). This promoter is bound by the endogenous metal response element binding transcription factor 1 (MTF-1)21 in the presence of zinc (Zn), copper (Cu), or cadmium (Cd) driving expression of the downstream reporter gene. As most biosensors, this configuration has several limitations, including low expression of the reporter gene and a narrow dynamic range, defined as the ratio between the maximum achievable biosensor response and its leakiness (Fig. 3b, c—blue line). To address these limitations, we modified the CASwitch v.2 system by replacing the pCMV promoter driving the expression of rtTA3G, with the metal-responsive promoter pMRE, as shown in Fig. 3a, with the goal of simultaneously enhancing the copper biosensor’s absolute expression and amplifying its dynamic range.

Fig. 3: CASwitch v.2 as a whole-cell biosensor.
figure 3

a Schematics of three alternative experimental implementations of a copper biosensor. Upon copper administration, the endogenous MTF-1 transcription factor binds its cognate synthetic promoter pMRE that either directly drives expression of Firefly Luciferase (fLuc) expression (pMRE Biosensor), or drives expression of the rtTA3G transactivator, which in turn induces the expression of the fLuc through the pTRE3G in the presence of doxycycline (Tet-On3G Biosensor). In the CASwitch v.2 Biosensor, the pMRE promoter drives expression of the rtTA3G, which in turn induces expression of the fLuc harbouring a DR and inhibits expression of the CasRx through the pCMV/TO promoter. b,c Experimental validation of the three biosensors at the indicated concentrations of copper chloride in HEK293T cells. Firefly luciferase (fLuc) expression was evaluated by luminescence measurements and normalised to Renilla firefly (rLuc) luminescence. Fold-induction in (c) is obtained by dividing each data point by the average luciferase expression in the absence of copper. n = 4 biological replicates, albeit for CuCl2 equal to 25uM which shows 3 replicates. MTF-1: metal-responsive transcription factor 1; pMRE: synthetic metal responsive promoter; DR: direct repeat sequence; rtTA3G: reverse tetracycline TransActivator 3G; pTRE3G: Tetracycline Responsive Element promoter 3G; pCMV/TO: modified CMV promoter with two Tetracycline Operon (TO) sequences. Source data are provided as a Source Data file.

To evaluate the effectiveness of the CASwitch v.2 plug-in strategy, we compared it to an additional biosensor configuration as shown in Fig. 3a, where the pMRE promoter drives the expression of the rtTA3G transcription factor, which in turn drives expression of fLuc from the pTRE3G promoter. This configuration, in the presence of doxycycline, effectively implements a transcriptional amplification of the reporter gene expression, which however should not improve the dynamic range as both leaky and maximal gene expression should increase.

We evaluated the expression of fLuc from the three configurations at increasing concentrations of copper and at a fixed concentration of doxycycline. Results are reported in Fig. 3b,c: the standard copper biosensor exhibited considerable leakiness and low levels of reporter gene expression even at high copper concentrations, thus resulting in a low signal-to-noise ratio with a maximum induction of only 10-fold. The second configuration with the rtTA3G resulted in a significant increase in luciferase expression levels at all copper concentrations, however, as expected, it did not lead to dynamic range amplification, as it also increased the leaky reporter expression in the absence of copper. Conversely, the CASwitch v.2 configuration effectively reduced leakiness in the absence of copper, while achieving higher luciferase expression than that of the standard copper biosensor (Fig. 3b). This resulted in a large increase in the biosensor’s signal-to-noise ratio with a maximum induction of up to 100-fold, hence 1-log more than the other two configurations (Fig. 3c). Of note, the CASwitch v.2 yielded higher fold-induction levels at four times lower copper concentration, thus also enhancing its sensitivity. Taken together, these findings support the application of the CASwitch v.2 system to improve the efficacy of existing transcriptional-based biosensors that experience limitations in terms of a narrow dynamic range. The expansion of the biosensors’ dynamic range through the integration of CASwitch v.2 will yield a more sensitive and reliable biosensor, capable of detecting lower concentrations of the analyte with increased confidence.

CASwitch v.2 is a tight gene expression system for reliable inducible control of toxic gene expression

We investigated the application of the CASwitch v.2 system in tightly controlling the expression of toxic genes, this feature is very useful for some industrial applications such as recombinant protein production, where the unintended accumulation of the protein of interest due to leakiness impairs host cell viability and lowers production yields (e.g, viral proteins). As a proof-of-principle, we used the CASwitch v.2 system to express the Herpes Simplex Virus Thymidine Kinase-1 (HSV-TK), which exerts cytotoxic effects in the presence of nucleotide analogues such as ganciclovir (GCV)22. To this end, as shown in Fig. 4a, we added a Direct Repeat in the 3’UTR of the HSV-TK gene and placed it downstream of the pTRE3G promoter in the CASwitch v.2 circuit. We then evaluated cell viability in the presence of ganciclovir, either with or without doxycycline and compared it to the one obtained by using the state-of-the-art Tet-On3G gene expression system. To account for cytotoxic effects associated with transfection, we co-transfected cells with a non-coding plasmid in the “Mock” condition, against which all other cell viability measurements were normalized to. Furthermore, constitutive expression of HSV-TK provided a reference for the maximum achievable toxicity. Results are reported in Fig. 4b, c and show no cytotoxic effects for the CASwitch v.2 system in the absence of doxycycline. In contrast, the Tet-On3G system exhibited high cell toxicity, resulting in ~50% cell death in the absence of doxycycline. These findings confirm that the CASwitch v.2 system has very low leakiness, highlighting its efficacy in controlling toxic genes expression.

Fig. 4: CASwitch v.2 enables inducible expression of toxic genes and inducible production of Adeno Associated Viruses.
figure 4

a Three alternative constructs to express the cytotoxic HSV-TK gene. pCMV-HSV-TK: positive control, with constitutive expression of HSV-TK. Tet-On3G: the constitutively expressed rtTA3G induces the cytotoxic HSV-TK gene harbouring a DR in its 3’UTR, binding to pTRE3G in the presence of doxycycline. CASwitch v.2: the same as the Tet-On3G but for the presence of the CasRx downstream of the pCMV/TO. b Viability of HEK293T cells transfected with the indicated constructs and grown in the presence of ganciclovir. Mock transfected cells represent the negative control. Cell viability is reported as a percentage of the viability of mock transfected cells in the absence of doxycycline. The error bars represent the mean and standard deviation of biological replicates across two independent experiments (n = 9). Statistical analysis with ANOVA (one-tailed) after determining equal or unequal variances by D’Agostino & Pearson test (****P-value < 0.0001) c Crystal violet staining of transfected HEK293T cells to highlight viable cells. d Plasmids required for AAV production. Two alternative experimental implementations for inducible expression of the Helper genes using either the Tet-On3G system or the CASwitch v.2 are also shown. e Assay for testing AAV vector inducible production yield by means of viral transduction. Created with Biorender. f, g Flow cytometry of cells transduced with cell lysates of HEK293T cells transfected with the indicated configurations. At least 10,000 cells were analysed for each point. The bar-plot in (g) reports, for each experimental condition, the mean value of the percentage of transduced cells across of biological replicates for two independent experiments (n = 6) with error bars corresponding to the standard deviation. Statistical analysis by ANOVA (one-tailed), after determining equal or unequal variances by D’Agostino & Pearson test (****P-value < 0.0001). HSV-TK: Herpes Simplex Virus Thymidine Kinase; AAV: Adeno-Associated Virus; E2A(DBP): Early 2A DNA Binding Protein gene; E4(Orf6): Early 4 Open reading frame 6 gene; VaRNA-I: Viral associated RNA-I; Rep: AAV-2 Replication genes; Cap: AAV-2 Capsid genes. Source data are provided as a Source Data file.

CASwitch v.2 as a platform for Adeno-Associated Virus (AAV) production for gene therapy

Adeno-Associated Virus (AAV) vectors have emerged as highly promising tools for in-vivo gene therapy in clinical applications23. However, current large-scale industrial bioproduction face challenges in terms of efficiency and scalability, as it mainly relies on transient transfection of HEK293 cell lines24,25. Attempts to develop more scalable systems, such as AAV producer cell lines with stable integration of inducible gene systems to control the expression of viral genes, have been hampered by the toxicity associated with leaky expression of viral genes26,27,28,29,30. In this context, the CASwitch v.2 expression system may offer a reliable solution having the ability to significantly reduce leakiness while maintaining high levels of maximal achievable expression.

As shown in Fig. 4d, transient triple transfection manufacturing of AAV vectors requires three plasmids: (i) a Transgene plasmid encoding the desired transcriptional unit to be packaged, (ii) a Packaging plasmid, and (iii) a Helper plasmid. The Packaging plasmid in our implementation carries the wild-type AAV2 Rep and Cap genes, while the Helper plasmid contains the E2A, E4, and VA RNAI genes derived from Human Adenovirus 5 (HAdV-5)31. As the HAdV-5 genes are polycistronic and expressed from distinct promoters, we first determined the minimal set of viral genes necessary for AAV vector production. Previous studies have shown that the E2A(DBP) and E4(Orf6) coding sequences, along with the VARNA-I ncRNA, are essential for AAV vector production32. Therefore, we designed constructs expressing E2A(DBP) and E4(Orf6) as a single transcript by means of two alternative strategies: the EMCV-IRES33 or P2A-skipping ribosome sequence34. By interchanging the positions of E2A(DBP) and E4(Orf6) in the bicistronic transcriptional units, we generated four different Adenovirus Helper plasmids (named µHelper, pAµH1-4) about half the size of the original plasmid, as reported in Supplementary Fig. 8a. We compared these constructs by quantifying AAV production yield through quantitative PCR (qPCR). All µHelper plasmids led to AAV production, albeit to a lesser extent than the full-length Helper plasmid. Among these, the pAµH-3 plasmid (pCMV-E2A[DBP]-IRES-E4[Orf6]) exhibited the highest yields, as shown in Supplementary Fig. 8b. We attributed the lower production yield to the absence of the VaRNA-I ncRNA. Indeed, co-transfection of VaRNA-I along with puH-3 restored production efficacy (Supplementary Fig. 9).

To achieve inducible expression of the Helper genes using the CASwitch v.2 system, we introduced the direct repeat (DR) element into the 3’ untranslated region (UTR) of the E2A(DBP)-IRES-E4(Orf6) cassette and placed it downstream of pTRE3G (p3G-AµH3-DR), as depicted in Fig. 4d. We then qualitatively assessed the capability of the CASwitch v.2 system in controlling expression of helper genes for inducible AAV vector production in the context of transient triple transfection manufacturing and compared it to that of the state-of-the-art Tet-On3G system. Specifically, we employed EGFP as the transgene for generating AAV vectors, with fluorescence quantification in transduced cells serving as a qualitative indirect measure of production yields (Fig. 4e). We assessed production yields both in the presence and absence of doxycycline, providing a qualitative evaluation of the Tet-On3G and the CASwitch v.2 systems’ performance in AAV vector production, as reported in Fig. 4e-g. Infection results confirmed that when controlling Helper genes’ expression with the Tet-On3G system, viral production occurred even in the absence of doxycycline, because of leaky expression of the viral Helper genes. Conversely, when controlling Helper genes with the CASwitch v.2 system, there was a significative reduction in AAV production in the absence of doxycycline, as measured by the percentage of infected cell, while maintaining high production yields in its presence. Despite viral production not being completely shut off, this proof-of-principle experiment shows that with proper fine-tuning, the CASwitch v.2 system could represent an effective solution to prevent unintended toxic viral gene expression, thus paving the way for the development of inducible AAV producer cell lines.