Construction of miRNA-responsive IRES translation activators (MITAs) based on hepatitis C virus (HCV) IRES
We first chose the hepatitis C virus (HCV) IRES to demonstrate our strategy because the structure of HCV IRES has been solved15. HCV IRES is composed of four domains. Previous literature has pointed out that a pseudoknot in domain IV played a crucial role in HCV IRES-mediated translation initiation, and the pseudoknot is sterically adjacent to the 5’ end of HCV IRES16. Thus, we hypothesized that the native structure of HCV IRES would be distorted by inserting an upstream sequence (termed IRES structure distorting sequence, DS) complementary to the pseudoknot region, leading to inhibition of IRES-mediated translation. We generated a bicistronic fluorescence reporter construct to evaluate the translation efficiency of engineered IRES in human embryonic kidney (HEK293) cells using flow cytometry (Fig. 1a). The translation of EYFP is cap-dependent and serves as an internal control. The translation of tagBFP is IRES-initiated, and the ratio of mean fluorescence intensity (MFI) of tagBFP to EYFP denotes the IRES-mediated translation efficiency. In comparison with unmodified native IRES (nIRES), the MFI of tagBFP decreased when the length of DS exceeded 12 nucleotides and was reduced by 10-fold when the length of DS reached 20 nucleotides (termed this structure dIRES, Fig. 1b, Supplementary Fig. 1a–b).
Next, we designed and inserted a rescuing sequence (RS) complementary to the 20-nt DS, hypothesizing that base-pairing between RS and DS would restore the native structure and translation initiation function of HCV IRES (Fig. 1c). By introducing a RS upstream of DS with a 12-nt spacer, MFI of tagBFP increased when the length of RS exceeded 8 nucleotides and recovered to over 70% of nIRES when the length of RS reached 13 nucleotides (termed this structure rIRES, Fig. 1d–f, Supplementary Fig. 1c–e). Additionally, we also observed similar reduced tagBFP translation of dIRES and recovery of rIRES in human hepatocellular carcinoma (Huh7) cells as in HEK293 cells (Supplementary Fig. 1f, g).
To engineer programmable miRNA-responsive IRES translation activators (MITAs), we sought to design secondary structures with miRNA binding sites (miRBS) to sequester RS in a relatively stable structure. To ensure the miRNA-Argonaute (Ago) complex only binds to miRBS without cleavage, we designed miRBS as miRNA sponge17 with 2-nucleotide mismatches with miRNA at positions 10 and 11. Enlighten by toehold switch18 and other riboregulator design19, we constructed three types of MITAs (Fig. 2a–c), namely toehold-like (TL, Fig. 2a), stem-loop (SL, Fig. 2b), and 3-arm-junction (AJ, Fig. 2c). We hypothesized that, in the absence of complementary miRNA, the 13-nt RS was sequestered in the designed structure, and IRES-mediated tagBFP translation was inhibited by the 20-nt DS (OFF state). While in the presence of complementary miRNA, the miRNA-Ago complex binds to the designed miRBS and interrupts the local secondary structure, releasing the 13-nt RS to recover IRES structure and translation activity (ON state). We first designed miRBS complementary to an artificial miR-FF4 and tested all three types of MITA in HEK293 with miRNA mimics co-transfection. Since more base-pairing in the RS-sequestering structure would stabilize the OFF-state configuration while more mismatches would disfavor the OFF-state configuration, we variated the number of base-pairing and mismatches to optimize the MITA structure. We designed a variety of engineered IRES structures, including toehold-like structures with varying base-pairing (TL14 to TL18, Fig. 2d, Supplementary Fig. 2a, b), stem-loop structures with different mismatches in stems (SL0, SL2 to SL6, Fig. 2e, Supplementary Fig. 2c, d), and 3-arm-junctions with different mismatches in junctions (AJ0, AJ2 to AJ5, Fig. 2f, Supplementary Fig. 2e, f). We transfected the miRNA mimics at low (10 nM), medium (30 nM), and high (50 nM) concentrations. All three types of MITA exhibited reduced IRES-mediated tagBFP translation activity with control mimics co-transfection, indicating the MITAs were in the OFF state in the absence of the desired miRNA. At medium and high transfection concentrations of miR-FF4 mimics, the tagBFP translation activity mostly recovered, indicating the presence of miR-FF4 triggered the MITAs to the ON state. Notably, the results also unveiled a dose-dependent activity of the designed MITAs, validating the miRNA sensor functionality of our MITA design. To confirm that miRNAs did not affect the basal IRES expression, we included control mimics, miR-FF4 mimics, and miR-21 mimics to test the influence of miRNA on IRES itself (Supplementary Fig. 2g). These control experiments demonstrated that miRNA mimics did not affect the basal expression of IRES (nIRES, dIRES, and rIRES). To further verify the specificity of our MITA design, we co-transfected our miR-FF4 responsive MITAs with either control mimics or non-target miR-199a mimics. These control experiments showed that the miR-FF4 responsive MITA was not responsive to non-target miRNA mimics, which confirmed the specificity of our MITA design (Supplementary Fig. 2h). Taken together, these results indicated that the translation activity of HCV IRES could be modulated by changing the conformation of IRES structure, and the conformational change could be designed to be miRNA-responsive.
Construction of logic gates and cell-type classifiers based on the PROMITAR platform
Having demonstrated the strategy of programming MITAs in response to specific artificial miRNA mimics, we then sought to engineer MITAs to sense dual miRNA mimics to implement logic gates. We tandemly ligated two TL16 motifs for miR-FF4 and miR-199a (a miRNA that showed little expression in HEK293, Supplementary Fig. 3a) respectively, and tested in HEK293 with miRNA mimics co-transfection (Fig. 3a, b, Supplementary Fig. 3c, d). With either miR-FF4 or miR-199a mimics co-transfected, IRES-mediated tagBFP translation recovered, indicating that the tandem TL16 design could implement the OR logic function. We then designed the SL3 motif with both miR-FF4 and miR-199a binding sites (Fig. 3c, d, Supplementary Fig. 3e, f). With either miR-FF4 or miR-199a mimics co-transfected, tagBFP translation was inhibited as the negative control. With both miR-FF4 and miR-199a mimics co-transfected, tagBFP translation mostly recovered, indicating that the two-input SL3 design could serve as an AND logic gate. Furthermore, we also investigated the potential influence of miRBS arrangement on our system by altering the order of miRBS within OR-gates and AND-gates (Supplementary Fig. 3g, h). The results were consistent with our prior findings, indicating that our logic-gate design remained effective irrespective of the miRBS arrangement. To implement the NOT logic function, we generated a programmable miRNA-responsive IRES translation repressor (MITR) by directly inserting fully complementary binding sites of miR-FF4 both upstream and downstream of the nIRES (Fig. 3e, f, Supplementary Fig. 3i, j). The MITR exhibited a slightly lower translation level compared with nIRES in the absence of miR-FF4. With miR-FF4 co-transfected, the tagBFP translation was significantly inhibited, indicating that MITR could implement the NOT logic function.
After characterizing our MITA and MITR design with exogenous miRNA mimics, we then sought to employ MITA to sense endogenous miRNA. We designed miRBS complementary to miR-21, a miRNA that is highly expressed in most cancer cells, such as Huh7 but not in HEK293 (Supplementary Fig. 3a, b). Since IRES usually differs in the basal expression among different cell lines4, and we indeed found that HCV IRES showed 1.2–1.5 times higher expression in Huh7 than in HEK293 when these cells were transfected with our bicistronic reporter plasmids (Supplementary Fig. 1f, g). Hence, we normalized the reporter tagBFP expression by nIRES or rIRES expression in different cell lines to eliminate the inherent difference of IRES expression between these cell lines, which allowed us to exclusively illustrate the individual impact of our engineering strategies on IRES translation.
We then examined the normalized tagBFP translation of all three types of MITA in Huh7 and HEK293 cells (Fig. 3g, Supplementary Fig. 4a, b). As a result, all MITAs showed higher MFI of tagBFP in Huh7 than HEK293, and TL16, SL3 and AJ2 exhibited the highest fold changes (up to 3-fold). Interestingly, SL5, SL6, AJ4, and AJ5 showed higher MFI of tagBFP compared with rIRES, probably because the Ago proteins enhanced the translation initiation activity of HCV IRES as previously reported20.
To rule out the possibility that the designed MITAs’ structural changes could be induced by intracellular context instead of target miRNAs, we transfected each version of exogenous miR-FF4 responsive MITA plasmid constructs into HEK293 and Huh7 cells without the co-transfection of miR-FF4 mimics (Supplementary Fig. 4c). The differential expressions of MITAs comprising non-target miRBS (i.e., miR-FF4 binding sites) between the HEK293 and Huh7 were similar to the inherent IRES expression difference between HEK293 and Huh7, contrasting with the differential expression pattern of miR-21 responsive MITAs. These results indicated that our MITA design was not modulated by the cellular context. To verify the miRNA specificity of our miR-21 responsive MITAs design, we tested miR-21 responsive MITAs by introducing miR-21 inhibitors in Huh7 and miR-21 mimics in HEK293 (Supplementary Fig. 4d, e). The results showed a decrease in tagBFP reporter expression with miR-21 inhibitor co-transfection in Huh7, and an increase in tagBFP reporter expression with miR-21 mimics co-transfection in HEK293. These results confirmed the specificity of our miR-21 responsive MITA design. In addition to Huh7 cells, we also tested the performance of miR-21-responsive MITAs in HeLa cells, another cancer cell line exhibiting high miR-21 activity21 (Supplementary Fig. 4f). Consistent with our observations in Huh7 cells, miR-21 responsive MITAs also exhibited higher tagBFP reporter expression in HeLa cells than in HEK293 cells. To further explore our MITA design’s capacity to respond to a diverse range of miRNA signals, we employed another endogenous miRNA, miR-18a, which showed high activity in HEK293 but low activity in Huh7 cells (Supplementary Fig. 3a, b). The results showed that the tagBFP reporter expression was higher in HEK293 cells than that in Huh7 cells (Fig. 3h, Supplementary Fig. 4g, h). Given the opposite miRNA activity levels of miR-18a and miR-21 in HEK293 and Huh7 cells, the reversed tagBFP differential expression patterns of corresponding designs of miR-18a-responsive and miR-21-responsive MITAs provided further evidence for the applicability of our MITA design for detecting diverse miRNA signals.
Combining the MITA and MITR designs, we then sought to employ our PROMITAR platform to construct cell-type classifiers. We tested several miRNAs reported previously that showed high repression activity in HEK293 (Supplementary Fig. 3a, b)22. We chose miR-21 as Huh7-high activity miRNA to construct MITA modules, and miR-106a-5p and miR-18a as HEK293-high activity miRNAs to implement NOT logic gates. We constructed vectors containing miR-21 responsive TL16, SL3, or AJ2 MITA with 4 tandem repeats of HEK293-high miRBS upstream of the MITA and downstream of the tagBFP reporter (Fig. 3i). We transfected each of these plasmid constructs into Huh7 and HEK293 cells with mKate fluorescence protein as an internal control. As a result, we observed up to a 7-fold increase of tagBFP fluorescence intensity in Huh7 compared with HEK293 (Fig. 3j). Notably, we only observed a ~2-fold increase of fluorescence signal when we employed MITA (Fig. 3g) or MITR (Fig. 3j). Collectively, these results indicated a cooperating effect between MITA and MITR in regard to cell-type classification. Since the basal expression of HCV IRES in Huh7 cells was ~1.2–1.5 times higher than that in HEK293 cells, combined with the higher normalized expression of MITA and classifiers in Huh7 relative to HEK293, we observed that MITA and classifiers showed much higher absolute expression in Huh7 cells (up to 9-fold, Supplementary Fig. 4i, j).
Construction and validation of MITAs and cell-type classifiers in circular RNAs (circRNAs) for targeted cancer cell identification and elimination
To apply MITA modules and cell-type classifiers in circRNAs, we constructed plasmid templates containing permuted intron-exon (PIE)4 for in vitro transcription (IVT) of circRNA (Fig. 4a). With firefly luciferase as a reporter, we first transfected circRNAs encoding nIRES, dIRES, and rIRES into HEK293 and Huh7 cells, respectively. We observed the same inhibition and recovery effect as plasmid transfections (Fig. 4b, Supplementary Fig. 5a). Next, we transfected circRNAs encoding TL16, SL3, or AJ2 miR-21 responsive MITA into Huh7 and HEK293 cells, respectively (Fig. 4c, Supplementary Fig. 5b). The TL16 or SL3 design exhibited up to 3-fold luminescence signal increase, and the AJ2 exhibited a 6-fold signal increase in Huh7 compared with HEK293. These results indicated that our MITA modules were also miRNA-activated in circRNA. Finally, we examined the cell-type classifiers in circRNA (Fig. 4d, e, Supplementary Fig. 5c). We observed up to 8-fold changes of luminescence signal in Huh7 compared with HEK293, indicating that our cell-type classifier circRNAs could accurately identify desired cancer cells, and might serve as potential cancer-targeting therapeutics. Notably, we observed that the basal expression of HCV IRES in Huh7 cells was ~3 times higher than that in HEK293 cells with circRNA transfection (Supplementary Fig. 5a), which meant that our design enabled much higher absolute expression in Huh7 (up to 9-fold for MITA, Supplementary Fig. 5b; up to 18-fold for classifier, Supplementary Fig. 5c), facilitating potential applications such as cell classification and selective killing of cancer cells. To further elucidate the potential therapeutic applications of our cell-type classifier circRNAs based on the PROMITAR platform, we replaced the luciferase reporter with the N-terminal section of Gasdermin D (GSDMD), a pore-forming protein that is involved in the inflammatory cell death pathway known as pyroptosis23 (Fig. 4f). We then performed the lactate dehydrogenase (LDH) cytotoxicity assay to evaluate the cancer-cell-specific cytotoxicity of our MITA-containing GSDMD-encoding circRNAs. We observed a substantial difference in cytotoxicity levels, with an 18-fold higher LDH activity in Huh7 cells compared to HEK293 cells (Fig. 4g). Furthermore, we transfected the GSDMD-encoding cell-type classifier circRNA into Huh7 and HEK293 cells and observed an even higher difference in cytotoxicity (up to 89-fold, Fig. 4h, i, Supplementary Fig. 5d). To further evaluate the performance of our targeted cancer killing circRNAs in heterogeneous cell populations, we engineered two fluorescently tagged stable cell lines: Huh7 cells expressing EYFP and HEK293 cells expressing mScarlet (Supplementary Fig. 6). These cell lines were co-cultured, and subsequently transfected with lipid nanoparticles (LNPs) encapsulating either our cytotoxic GSDMD-encoding circRNA classifiers or noncytotoxic circRNAs as controls. Notably, we observed a substantial decrease in the ratio of surviving EYFP+ Huh7 cell percentage to the surviving mScarlet+ HEK293 cell percentage after transfection with the classifier circRNA (Fig. 4j, k). These results indicated that the cell-type classifier functioned correctly and corroborated the functionality of our cell-type classifier in selectively inducing pyroptosis in Huh7 cells within mixed cellular populations. Collectively, these findings highlighted the potential applicability of our cell-type classifiers for targeted cancer therapy applications.
To evaluate the versatility of our cell-type classifiers, we sought to alter the miRNA binding sites (miRBSs) of the MITA and MITR modules in the cytotoxic GSDMD-encoding circRNA, while maintaining the optimized AJ2 secondary structure of the MITA module (generating classifiers 1 to 4, Supplementary Fig. 7a). We first swapped the miRBSs of the previous classifier (containing miR-21 MITA and miR-18a MITR) to generate a new classifier with miR-18a MITA and miR-21 MITR, expecting this to confer HEK293 cell-targeting cytotoxicity. We then performed LDH assays and co-culture flow cytometry experiments (Supplementary Fig. 7a, c, and e, f, Supplementary Fig. 8a, b, and d–g), and we observed reversed cell killing effects for this new classifier 2 compared with the original classifier 1, consistent with the altered miRBS order.
We further extended our classifier design principle for distinguishing Huh7 cells from another liver-derived line, L-02. We first identified let-7c24 as a highly active L-02-specific miRNA marker using flow cytometry with both linear mRNA and circRNA transfection (Supplementary Fig. 7b). Both linear mRNA and circRNA transfection experiments showed that let-7c exhibited much higher activity in L-02 cells than Huh7 cells, with a more substantial repression activity of let-7c in circRNA transfection compared with linear mRNA transfection. With miR-21 as the Huh7-specific marker and let-7c as the L-02-specific marker, we designed two additional classifiers: classifier 3 with miR-21 MITA and let-7c MITR to distinguish Huh7 from L-02, and classifier 4 with let-7c MITA and miR-21 MITR to distinguish L-02 from Huh7 (Supplementary Fig. 7a). Transfection experiments showed classifier 3 induced stronger Huh7 cell death compared to L-02, while classifier 4 was more cytotoxic to L-02 versus Huh7 (Supplementary Fig. 7d and g, h, Supplementary Fig. 8b, c, and h–k). Collectively, these results demonstrated that our PROMITAR-based cell-type classifier design principle can distinguish cell types based on their unique miRNA expression profiles.
Extending the design principles of the PROMITAR platform to classical swine fever virus (CSFV) IRES
To further demonstrate the versatility of our strategy, we then sought to expand the PROMITAR platform by applying our design principle to the classical swine fever virus (CSFV) IRES, which owns a similar pseudoknot structure as HCV IRES25 (Fig. 5a). Firstly, we also designed a 20-nt DS targeting the pseudoknot of the CSFV IRES (also, termed this structure dIRES) and observed a similar inhibition effect as in HCV IRES (Fig. 5b, Supplementary Fig. 9a), suggesting that our strategy could extend to other IRES. Then we also introduced a series of RS upstream of the 20-nt DS and observed up to 70% recovery rate when the length of RS reached 13 nucleotides (Fig. 5c, d, Supplementary Fig. 9b, c). Following a similar approach as we designed the MITA and MITR with HCV IRES, we developed customized MITA and MITR modules based on the CSFV IRES by incorporating miRBSs responsive to miR-21 and miR-FF4, respectively. With the 20-nt DS and 13-nt RS, we designed RNA secondary structures with miR-21 binding sites sequestering the RS for MITAs as in HCV IRES, and designed MITR by inserting fully complementary binding sites of miR-FF4 both upstream and downstream of the CSFV IRES. We transfected the MITA-containing reporter plasmid into Huh7 and HEK293 cells (Fig. 5e, Supplementary Fig. 9d, e), and co-transfected MITR-containing reporter plasmids with miRNA mimics in HEK293 cells (Fig. 5f, g, Supplementary Fig. 9f, g). Notably, these constructs based on CSFV IRES performed in a manner closely analogous to the previously tested HCV IRES-based designs. Collectively, these results suggested that our design principle offered a versatile framework for developing other PROMITAR platforms capable of engineering diverse IRES sequences while preserving the desired miRNA-responsive functionality.