Trans-amplifying RNA expressing functional miRNA mediates target gene suppression and simultaneous transgene expression

Abstract

The co-delivery of microRNAs (miRNAs) and protein-coding RNA presents an opportunity for a combined approach to gene expression and gene regulation for therapeutic applications. Protein delivery is established using long mRNA, self-, and trans-amplifying RNA (taRNA), whereas miRNA delivery typically uses short synthetic oligonucleotides rather than incorporating it as a precursor into long RNA. Although miRNA delivery into the cell cytoplasm using long genomes of RNA viruses has been described, concerns have remained regarding low processing efficiency. However, miRNA precursors can be released from long cytoplasmic alphaviral RNA by a cytoplasmic fraction of Drosha. taRNA, a promising vector platform for infectious disease vaccination, uses a nonreplicating mRNA expressing an alphaviral replicase to amplify a protein-coding short transreplicon-RNA (STR) in trans. To investigate the possibility of simultaneously delivering protein expression and gene silencing, we tested whether a taRNA system can carry and release functional miRNA to target cells. Here, we show that mature miRNA is released from STRs and silences specific targets in a replication-dependent manner for several days without compromising the expression of STR-encoded proteins. Our findings suggest that incorporating miRNAs into the taRNA vector platform has the potential for gene regulation alongside the expression of therapeutic genes.

Graphical abstract

Yıldız and colleagues exploit trans-amplifying RNA, a recently established RNA vaccine platform, to deliver functional miRNAs excised from the 3' UTR of protein-coding transreplicons. Target gene knockdown lasted for 96 h and protein expression remained unalterably high, suggesting co-delivery of therapeutic proteins and miRNA with taRNA whenever deemed beneficial.

Introduction

RNA interference (RNAi) is a posttranscriptional regulatory mechanism that uses small noncoding RNAs to direct sequence-specific mRNA degradation or translational repression.¹ In mammalian cells, common exogenous RNAi mediators comprise synthetic small interfering RNAs (siRNAs), short hairpin RNAs (shRNAs), and artificial microRNAs (amiRNAs).¹ To be effective, these small RNAs are channeled into the endogenous miRNA biogenesis pathway. Canonical miRNA maturation begins in the nucleus with the cleavage of primary miRNA (pri-miRNA) transcripts into precursor miRNAs (pre-miRNAs) by the ribonuclease III Drosha. Upon translocation to the cytoplasm, the pre-miRNAs are further processed by another ribonuclease III, Dicer, resulting in ∼22-nt-long miRNA duplexes. Typically, one strand of the duplex, the mature miRNA, associates with the RNA-induced silencing complex to bind mRNA sequences with imperfect complementarity, thereby mediating posttranscriptional silencing.² Artificial RNAi mediators enter the canonical pathway at different stages. Expression cassettes for amiRNAs structurally mimic natural pri-miRNAs and thus follow the entire pathway, requiring processing by Drosha and Dicer to reach maturation.³ However, unlike mature natural miRNAs, mature amiRNAs are perfectly complementary to their target mRNA, similar to shRNAs and siRNA. shRNAs mimic pre-miRNAs and, therefore, enter the pathway downstream of Drosha. Once processed by Dicer, they become target-specific siRNAs.^4,5 Finally, synthetic siRNAs or synthetic mature miRNA duplexes do not require processing and directly associate with RNA-induced silencing complex, leading to RNA cleavage through perfect target complementarity.² Although shRNAs and siRNAs are highly efficient in silencing genes, they can cause cytotoxicity by oversaturating the endogenous miRNA processing machinery.^6,7,8,9 In contrast, amiRNAs are associated with lower toxicity, making them safer and more suitable for in vivo applications.^8,10,11

Efficient delivery of RNAi mediators is crucial for manipulating target gene expression. Although mature miRNA mimics can quickly and efficiently regulate their targets through direct transfection to cells,¹² a more sustained miRNA expression can be achieved by expressing amiRNAs through stable chromosomal insertion and nuclear transcription.¹³ Typically, amiRNAs are inserted downstream of a suitable promoter into lentiviral vectors, which are also the primary choice for stable transcription of protein-coding mRNA.¹⁴ Dual expression of protein and amiRNA can be achieved on single-expression cassettes, with an open reading frame followed by an miRNA gene inserted into the 3' UTR or contained within an artificial intron.^15,16 However, lentiviral vectors carry the risk of insertional mutagenesis.¹³ Alternatively, vectors based on viruses having a single-stranded RNA genome of negative (e.g., Sendai virus [SeV]) or positive polarity (e.g., alphavirus) enable the joint delivery of functional amiRNAs and proteins while avoiding chromosomal insertion.^17,18,19,20 These vectors replicate in the cell cytoplasm with the help of viral RNA-dependent RNA polymerases, also known as replicases.²¹ Incorporated pri-miRNAs undergo processing within the cytoplasm following a noncanonical biogenesis pathway by nuclear Drosha, which relocates to the cytoplasm during viral infection.²² Thus, the viral RNA delivery enables the cotransfer of protein and miRNAs encoded on a single transcript, which has not yet been described for in vitro transcribed mRNA. Until now, for the combined delivery of proteins and miRNAs using in vitro transcribed RNA, the mixing and cotransfection of separately synthesized mRNA and miRNA have been the methods of choice, for example, for the improved derivation of RNA-based induced pluripotent stem cells (iPSCs).²³

We propose that an elegant and cost-efficient way to expand the function of synthetic mRNA with minimal effort would be to integrate an amiRNA expression cassette into a protein transcript such that they are collinear. Given that cytoplasmic miRNA processing by Drosha was discovered using alphaviral vectors,^19,22 we hypothesized that alphaviral trans-amplifying RNA (taRNA) would be suitable for this purpose.^24,25 Our recently developed second-generation taRNA vector system is bipartite, combining a nonreplicating mRNA (nrRNA) encoding an alphaviral replicase and a short transreplicon (STR) encoding the transgene.²⁶ Upon transfection and translation, the replicase recognizes the STR in trans and replicates it transiently within the cytoplasm. In the present study, we incorporated amiRNA into the STR 3' UTR and observed a replication-dependent, yet effective target knockdown that correlated with the enzymatic activity of the replicase. Moreover, a polycistronic stem cell-derived miRNA cluster was efficiently processed, leading to the release of multiple mature miRNAs that suppressed target genes for several days. Our findings suggest that the taRNA platform could be exploited for simultaneous expression and repression of protein-coding genes.

Results

Incorporation of pre-miRNA into the 3' UTR of protein-coding transreplicons preserves high protein expression and enables target gene regulation

Previous studies have demonstrated that the release of functional miRNA from the genomic RNA of alphaviruses that replicate in the cytoplasm follows a noncanonical miRNA biogenesis pathway.^19,27 Building on this knowledge, we investigated whether functional miRNAs could be processed and released from the 3' UTR of STRs contained within taRNA, which also replicates in the cytoplasm, and whether they would reach a level sufficient for effective target gene silencing. To avoid the cytotoxic replicase of Semliki forest virus^28,29 used in our previous studies,^24,26 we generated taRNA based on the genome of the Trinidad donkey strain (TRD) of the Venezuelan equine encephalitis virus (VEEV), which has a less toxic replicase.³⁰ As described before for a number of other alphaviruses,^24,26,31 we inserted the VEEV replicase into a nonreplicating mRNA (nrRNA-REPL); for the transreplicon, we chose our second-generation design and constructed VEEV-based STRs (Figure 1A).²⁶

Figure 1.

Incorporation of pre-miRNA into the 3' UTR of protein-coding transreplicons preserves high protein expression and enables target gene regulation

(A) Scheme of a taRNA-miR vector. taRNA comprises two capped and polyadenylated RNA molecules, one nonreplicative mRNA coding for the VEEV replicase (nrRNA-REPL) and a short transreplicon (STR-miR) that is replicated by the VEEV replicase, and coding for a transgene (TG) alongside an miRNA upstream of its 3' UTR and CSE. STR-miR is recognized as primary miRNA by ribonuclease Drosha and processed into precursor miRNA (pre-miRNA) and a 5′- and a 3′-truncated STR fragment. Pre-miRNA is further processed into an miRNA duplex by Dicer, another ribonuclease. (B–E) BHK-lacZ cells were electroporated with the respective taRNA-miR constructs (2.2 pM per STR-miR and 1.6 pM nrRNA-REPL of TRD-VEEV) or without RNA (mock). (B) taRNA-miR emGFP expression. At 24 h after transfection, the rate of emGFP⁺ cells and emGFP-mean fluorescence intensity (MFI) were determined by flow cytometry (n = 3). Total emGFP expression was estimated by multiplying the rate of emGFP⁺ cells with the MFI of emGFP⁺ cells (mean [SD]). (C) Target transcript level. Total cellular RNA was harvested 72 h after transfection to quantify relative transcript levels of lacZ normalized to that of β-actin by qRT-PCR. Mock-electroporated cells served to determine mean fold changes (mean [SD] of n = 3). Statistical significances were tested by 1-way ANOVA; ∗∗∗, p < 0.001; ns, not significant corresponding to mock. (D) Target protein expression. β-Gal expression was measured at the indicated time points and normalized to that of mock-electroporated cells (mean [SD] of n = 3). Statistical significances were tested by 2-way ANOVA; ∗∗, p < 0.01, and ns, not significant corresponding to mock. (E) Cell viability. The viability of BHK-lacZ cells was determined at indicated time points and normalized to that of mock-electroporated cells (mean [SD] of n = 3). Statistical significances were tested by 2-way ANOVA; ns, not significant corresponding to mock.miR-neco, nontargeting miRNA control; miR-scrmbld, no miRNA processing control; nsP, nonstructural protein; vUTR, viral UTR.

As a first step, we inserted a validated and commercially available lentiviral amiRNA expression/reporter cassette into the STR (STR-miR). The STR-miR thereby comprised the emerald GFP (emGFP) followed downstream by an optimized pre-miR-155 backbone containing an amiRNA against the bacterial lacZ mRNA (miR-lacZ) encoding β-galactosidase (β-gal).⁵ We retained the original alphaviral 5' and 3' UTRs and conserved sequence elements (CSEs) upstream and downstream to this cassette to ensure replication. As expected, an excision of the pre-miR hairpin from an STR-miR RNA molecule by Drosha would leave behind replication-incompetent side products (Figure 1A). To investigate whether miRNA processing would measurably reduce STR replication and expression, we generated a scrambled pre-miR-lacZ control to disrupt the miRNA secondary structure (Figure S1) and prevent recognition and processing by Drosha. We also inserted the available negative control into the STR comprising an amiRNA that, according to the manufacturer, does not target any known vertebrate gene (miR-neco). As a test system, we generated BHK-21 cells stably transduced with the lacZ gene to express β-gal (BHK-lacZ).

To estimate the presumably maximal achievable knockdown efficiency in this artificial system, we first transduced BHK-lacZ cells with both purchased lentiviral miR vectors. Thanks to very high transduction rates (∼90% emGFP⁺ cells), we achieved >90% knockdown efficiency of β-gal and lacZ expression within 96 h, whereas the negative control did not alter expression (Figure S2). Next, we transfected BHK-lacZ cells with the nrRNA-REPL of VEEV-TRD and STR-miR (together designated taRNA-miR), resulting in comparable emGFP expression for all three STR-miR constructs (Figure 1B). Within 72 h of transfection, taRNA-miR-lacZ significantly reduced lacZ transcript levels by 70% and β-gal protein levels by 50%, whereas expression remained unaltered in controls (Figures 1C and 1D). Moreover, taRNA-miR transfections did not hamper cell viability (Figure 1E).

Although lentiviral vectors were more effective in silencing lacZ, these results are the first evidence that taRNA-miR can suppress target expression in a sequence-specific manner while maintaining reporter gene expression and cell viability.

Replication of STR-miR is required for target knockdown and replicase activity determines the extent of knockdown

We extended our analysis by confirming the effectual miRNA delivery using STRs. We replaced the miR-lacZ with two amiRNAs targeting firefly luciferase (STR-miR-luc1, STR-miR-luc2) and generated BHK-21 cells that stably expressed luciferase (BHK-luc). To investigate whether STR replication is required for target gene knockdown, we cotransfected BHK-luc with both STR-miR-luc constructs, along with an nrRNA encoding either the replicase of VEEV-TRD (TRD-REPL), a replication-deficient mutant (inactive-REPL), or a hyperactive replicase (hyper-REPL) for increased replication rates.³² The day after transfection, we observed that the emGFP expression level of the cells reflected the type of replicase used (Figure S3). Specifically, with the inactive replicase, the capped STR-miR translated to a basal GFP expression, which was amplified >50 times by the TRD-REPL, and further increased about 10-fold by the hyper-REPL (Figure S3). Regarding luciferase expression, we found no significant silencing in cells cotransfected with STR-miR-luc and inactive-REPL (Figure 2A, left). Cotransfecting the TRD-REPL reduced luciferase expression by 50%–60% (Figure 2A, center), and cotransfecting hyper-REPL culminated in 80% silencing (Figure 2A, right). Importantly, cell viability (as measured by relative luciferase light units) remained unaffected regardless of the replicase variant used (Figure 2B). These data show that taRNA-miR suppresses a target gene in a replication-dependent manner without being cytotoxic. For clinical translation, electroporation is unlikely to be a viable transfer method. Therefore, we explored the efficiency of taRNA-miR-luc constructs in BHK-luc cells by complexing the RNA with commercial liposomes. The formulated taRNA-miR exhibited comparable transgene expression levels and target knockdown efficiency in transfected cells, as previously achieved through electroporation (Figures S4A and S4B). Once again, cell viability remained unaffected (Figure S4C), emphasizing that the efficiency of the platform is independent of the delivery method.

Figure 2.

Replication of STR-miR is required for target knockdown, and replicase activity determines the extent of knockdown

BHK-21 cells stably expressing firefly luciferase (BHK-luc) were electroporated with 1.1 pM of indicated STR-miR and co-delivered with 0.4 pM of either inactive replicase (inactive-REPL), replicase of VEEV-TRD (TRD-REPL), or hyperactive replicase (hyper-REPL). Control cells were electroporated without RNA (mock). (A) Target gene expression. Luciferase expression was measured at indicated time points and normalized to that of mock electroporated BHK-luc cells (mean [SD] of n = 3). Statistical significances were tested by 2-way ANOVA; ∗, p < 0.1; ∗∗, p < 0.01; ∗∗∗, p < 0.001; ∗∗∗∗, p < 0.0001, and ns, not significant corresponding to mock. (B) Cell viability. Viability of BHK-luc cells was inferred by relative luciferase intensity at indicated time points after transfection (mean [SD] of n = 3). RLU, relative light units.

taRNA-miR mediates sustained suppression of an endogenous target gene in primary cells, and mature miRNAs accumulate during replication

Having repeatedly achieved targeted gene knockdown by taRNA-miR in reporter cell lines, we aimed to downregulate an endogenous transcript under more physiological conditions in primary human dermal fibroblasts (HDF). To this aim, we designed three STR-miR constructs expressing a dual GFP-secreted nano-luciferase (SecNLuc) reporter gene and inserted amiRNAs targeting human p53 (miR-p53-1, miR-p53-2, miR-p53-3). As controls, we used an STR encoding GFP-SecNLuc without the miRNA cassette and the previously used miR-neco. HDF possess an innate immune response that inhibits self-amplifying RNA (saRNA) and taRNA replication,^26,33 which is activated by the transfection of in vitro transcribed, unmodified RNA, as well as double-stranded RNA intermediates generated during RNA replication.^34,35 To counteract the innate immune response, we cotransfected cells with nrRNAs encoding the vaccinia virus immune evasion proteins E3 and B18R (EB), which have been shown to significantly enhance saRNA expression³³ and to boost taRNA expression and reduce cytotoxicity resulting from the cellular innate immune response (Figure S5). Hence, all subsequent taRNA-miR-p53 experiments in HDF were conducted in the presence of cotransfected EB nrRNA. We found that all three taRNA-miR-p53 constructs downregulated p53 expression at both the transcript and protein levels compared to the miR-neco control, with miR-p53-2 being most efficient, achieving an 80% knockdown efficiency (Figures 3A and 3B). Neither viability of the transfected HDF (Figure S6A) nor the taRNA-encoded transgene expression was impaired compared to controls (Figure S6B).

Figure 3.

taRNA-miR mediates sustained suppression of an endogenous target in primary cells, and mature miRNAs accumulate during replication

(A–E) HDF were electroporated with the indicated taRNA-miR constructs (0.8 pM/RNA) and 0.2 pM E3 mRNA and 0.2 pM B18R nrRNA or without RNA (mock). As controls, 30 nM synthetic scrambled siRNA or synthetic siRNA against TP53 were transferred by lipofection. (A) Target transcript level. Total cellular RNA was harvested 72 h after transfection to quantify relative transcript levels of TP53 normalized to that of HPRT by qRT-PCR. taRNA-miR-neco transfected cells served to determine mean fold changes. (B) Target protein expression. p53 protein levels were assessed 72 h after transfection by western blotting. The depicted blot is representative of 3 independent experiments (left). Expected molecular weights in kilodaltons (kDa) of p53 and GAPDH are indicated. p53 expression levels were quantified by densitometry, normalized to GAPDH expression per lane (right). The p53 protein level in taRNA-miR-neco transfected cells served to determine mean fold changes. (A and B) Data shown as mean (SD) of 3 independent experiments; statistical significance tested by 1-way ANOVA; ∗, p < 0.05; ∗∗, p < 0.01; ∗∗∗, p < 0.001, and ns, not significant corresponding to taRNA-miR-neco. (C) TP53 transcript levels over time. Total cellular RNA was harvested at indicated time points after transfection to quantify relative transcript levels of TP53 normalized to that of HPRT by qRT-PCR. Mock-electroporated cells served to determine mean fold changes. Statistical analysis was a 2-way ANOVA; ∗∗, p < 0.01; ∗∗∗, p < 0.001; ∗∗∗∗, p < 0.0001, and ns, not significant corresponding to mock. (D) STR-miR RNA levels over time. Total cellular RNA was harvested at indicated time points after transfection to quantify relative transcript levels of STR-miR normalized to that of HPRT by qRT-PCR. To detect STR-miR transcripts, primers specific for SecNluc (transgene) were used. (E) Mature miR-p53-2 levels. Total cellular RNA was harvested at indicated time points after transfection to quantify absolute levels of mature miR-p53-2 processed from STR-miR-p53-2 in cells cotransfected with inactive (inactive-REPL) or active replicase (TRD-REPL) determined by qRT-PCR. All of the data are shown as mean ± SD, 3 independent experiments.

To investigate the kinetics of TP53 gene knockdown and its correlation with STR replication and release of mature miRNA, we performed time course experiments using the most efficient miR-p53-2. Our results showed that 24 h after transfection, TP53 transcript levels were maximally reduced by 80% and remained suppressed for at least another 72 h (Figure 3C). Concurrently, the STR-miR copy number increased within the first 12 h in response to nrRNA-REPL, and then decreased (Figure 3D), negatively correlating with the knockdown kinetics (Figure 3C). We also observed a good correlation between TP53 knockdown and the level of mature miRNAs released from replicating STR-miR (Figure 3E). Both reached sustained maximal levels from ∼24 to 96 h after transfection (Figures 3C and 3E). Interestingly, mature miRNA was also measurably released from STR-miR-p53-2 transfected cells with inactive REPL, albeit at a much lower level (Figure 3E), indicating that transfected RNAs are generally accessible for cytoplasmic pre-miR processing. In summary, we demonstrated that taRNA-miR suppressed an endogenous target gene in primary cells for several days, which is correlated with the presence of processed mature miRNAs.

A polycistronic miRNA cluster is processed from taRNA-miR, and targeted genes are suppressed for several days

Although synthetic miRNAs typically target only the specific gene of interest, natural miRNAs are known to regulate a network of target genes to different extents. Many miRNA genes are arranged in clusters,³⁶ which further expands their pleiotropic effects. Hence, we aimed to investigate whether the taRNA-miR platform could functionally deliver a natural miRNA gene cluster and lead to the simultaneous release of multiple mature miRNAs. To address both questions, we chose the polycistronic human miR-302/367 cluster, which contains five miRNA hairpins, namely miR-302b, -c, -a, -d, and -367 and is highly expressed in human embryonic stem cells and iPSCs. This cluster regulates many genes involved in cell signaling, the cell cycle, and the epigenetic regulation of pluripotency.³⁷ We replaced the synthetic miR-155 backbone with the miR-302/367 cluster in the STR 3' UTR (Figure 4A) and cotransfected HDF with either taRNA-miR-302/367 or taRNA-miR-neco, along with EB nrRNA. As a positive control for target gene regulation, we also transfected cells with an equimolar mix of five synthetic mature miRNAs corresponding to the miR-302/367 cluster. Transgene expression in taRNA-miR-302/367 transfected cells was not compromised by the inclusion of this polycistronic miRNA cluster (Figure S7), and we individually detected all five mature miRNAs 3 and 6 days after transfection, at levels that were ∼100–1,000 times higher compared to that in miR-neco transfected cells (Figure 4B). Although the mature miRNAs released from taRNA-miR-302/367 did not reach the levels naturally expressed in iPSCs (Figure 4B), two known target genes of the cluster, DAZAP2 and TGFβR2,³⁷ were significantly suppressed in taRNA-miR-302/367 transfected HDF compared to miR-neco transfected cells, indicating that physiologically relevant levels were reached (Figure 4C). In fact, although the transfection of synthetic miRNA achieved >10 times higher levels than of taRNA-miR-302/367, target gene suppression was only ∼30% more effective (Figure 4C). In established iPSCs generated using nrRNA-based reprogramming, TGFβR2 and DAZAP2 expression was barely detectable (Figure 4C).

Figure 4.

A polycistronic miRNA cluster is processed from taRNA-miR, and targeted genes are suppressed for several days

(A) Illustration of STR-miR-302/367 vector. STR-miR-302/367 incorporates the natural human miR-302/367 cluster composed of the 5 miRNAs miR-302b, -c, -a, -d, and -367. (B and C) HDF were electroporated with the indicated taRNA-miR constructs (0.8 pM/RNA) and 0.2 pM E3 and 0.2 pM B18R mRNA or without RNA (mock). Lipofection of HDF with synthetic mature miRNA miR-302a–d and 367 (0.4 μM each) served as positive control. (B) Relative miRNA levels. Total cellular RNA was harvested 3 and 6 days after transfection to quantify relative mature miRNA levels of miR-302s (miR-302a–d) and miR-367 normalized to that of SNORD48 by qRT-PCR. taRNA-miR-neco transfected cells served to determine mean fold changes (mean [SD] of n = 3). (C) Target transcript levels. Total cellular RNA was harvested 3 and 6 days after transfection to quantify relative transcript levels of TGFβR2 and DAZAP2 normalized to that of HPRT by qRT-PCR. Mock-electroporated cells served to determine mean fold changes (mean [SD] of n = 3). Statistical analysis was a 1-way ANOVA; ∗, p < 0.05; ∗∗, p < 0.01; ∗∗∗, p < 0.001; ∗∗∗∗, p < 0.0001, and ns, not significant corresponding to mock.

These findings suggest that the taRNA-miR platform can effectively deliver a natural polycistronic miRNA gene, resulting in the release of all of the enclosed miRNAs at levels sufficient for target gene regulation.

Discussion

In this study, we demonstrate that the taRNA vector platform can effectively deliver functional miRNAs and protein coding sequences simultaneously. We observed that amplification of a transgene that harbors miRNA sequences is necessary for achieving target gene regulation, and that targets remain suppressed for several days. The incorporation of miRNA into taRNA thereby extends the functionality of this highly immunogenic vaccine vector platform.

The functionality of the combined STR miRNA design was not apparent. Trans-amplification takes place in the cytoplasm, and the miRNA sequences we incorporated into the STRs resemble pri-miRNAs, usually processed in the nucleus by Drosha, following the canonical miRNA biogenesis pathway.³⁸ However, studies have shown that miRNAs incorporated into the genome of RNA viruses localizing exclusively in the cytoplasm can also be processed by Drosha that has relocalized to the cytoplasm.^22,39 Moreover, alternative splicing can generate cytoplasmic isoforms of Drosha that cleave pri-miRNAs outside the nucleus,^40,41 suggesting altogether that STR-miRs are processed by Drosha. We demonstrated that STR-miRs, harboring target-specific miRNAs, effectively mediate target mRNA suppression, indicating successful processing. Importantly, we found that regulation of the target did not compromise the ability of STR-miRs to be amplified or to express co-encoded proteins. This was a consideration, because mature amiRNAs released during processing could theoretically counteract replication or translation by self-targeting complementary sequences within their own related hairpin structure present in the STR-miR, an issue also discussed in the context of lentiviral miRNA delivery.⁴² In our experiments, neither of these potential issues seemed to materialize. Although self-targeting with siRNA serves as a natural antiviral defense mechanism in most invertebrates and plants,⁴³ RNAi-mediated cleavage may encounter challenges by complex secondary structures that mask the target sequence, protecting it from the RNA-induced silencing complex.¹³ Accordingly, the hairpin structure in the STR-miR, with its complementary sequence to the mature miRNA, could be shielded from processing. In addition, RNA replication occurs within membranous compartments sealed by replicase, referred to as spherules.⁴⁴ These spherules create an isolated and protective environment for the replicated RNA, potentially shielding it from self-targeting by released miRNAs. Given that we observed no detectable decrease in STR-encoded protein expression, the processing of STR-miRNAs to generate mature miRNAs did not quantitatively affect transgene levels. This suggests that the high replication rates can afford the sacrifice of a small proportion of the transfected or de novo synthesized STR copies for miRNA production, and potential self-targeting would not significantly influence replication. This hypothesis finds support in observations with engineered flaviviruses containing miRNA transcripts, where at least 100-fold fewer miRNAs were produced in infected cells than genomic RNA copies.¹⁸

In our study, the kinetics of taRNA-miR mediated target knockdown, and the simultaneous accumulation of mature amiRNA correlated directly with the kinetics of STR-miR replication. Although a low number of mature miRNAs were released from STRs in the absence of replicase, target suppression was only enabled after amplification to high STR-miR copy numbers. Conversely, producing more STR-miR transcripts by enhancing trans-replication with a hyperactive replicase led to a stronger target knockdown. The knockdown efficiency mediated by taRNA-miR persisted for several days in our study, but it did not reach the efficiency of stable lentiviral miRNA transfer. These results suggest that the processing of cytoplasmic RNA is rather inefficient and requires RNA amplification to achieve biological effects, which is not observable when using nonreplicating RNA. Thus, miRNA and protein cotransfer can be considered a unique selling point for replicating RNA compared to nonreplicating mRNA.

To improve the efficiency of miRNA production or to target multiple genes simultaneously, several strategies have been used, including engineering pri-miR backbones to deliver multiple siRNA sequences chained in a row.^5,45,46 amiRNA chaining can also be achieved by taRNA-miR because it produces several (natural) miRNAs from a polycistronic miRNA cluster, thereby regulating several targets simultaneously. However, it is worth noting that the efficiency of miRNA hairpin processing can vary considerably depending on the miRNA backbone used. In our study, we observed that the miR-367 hairpin was processed most efficiently among the miRNAs in the miR-302/367 cluster. This finding is consistent with the observations made with SeV-based vectors developed for the long-term cytoplasmic production of miRNAs.¹⁷ That study showed that the SeV vector-derived miR-367 hairpin is an exceptionally effective backbone for amiRNA production. In fact, the knockdown efficacy with miR-367 hairpin-based amiRNAs significantly outcompeted commonly used hairpins of miR-30 and miR-124, as well as the miR-155 used in this study. Because taRNA replicates in the cytoplasm like SeV, we anticipate identifying more effective miRNA backbones for taRNA-miR in future studies. Furthermore, miRNA processing efficiency could also be enhanced by improving pre-miR hairpin cleavage sites, thereby increasing pre-miR processing by Drosha,^47,48 or by overexpressing Drosha or Argonaute-2, which may be limiting factors.⁴⁹ Argonaute-2, in particular, has been identified as a rate-limiting factor, with Argonaute-2 overexpression increasing siRNA levels and Argonaute-2-knockdown leading to a significant reduction of most miRNA levels.⁴⁹ Furthermore, future optimization and validation of the taRNA-miR as a platform should include the analysis of potential off-target effects and the emergence of unexpected miRNA from the vector backbone, which we have not investigated in this study.

In primary cells, the innate immune response inhibits taRNA expression. We addressed this issue with the same approach that we used to improve gene expression by self-amplifying RNA - cotransfecting viral immune evasion proteins E3 and B18.³³ As effective as this method was, it may be possible to avoid the use of recombinant proteins or additional coding mRNAs by taRNA-encoded amiRNAs that downregulate RNA sensors or interferon-stimulating genes. Thus, immunomodulating amiRNAs could be a more viable option for clinical translation of the platform.

Overall, taRNA has the potential to be a powerful tool for codelivering therapeutic proteins and miRNAs. Incorporating them into untranslated regions of STRs is a simple and effective way to upgrade the functionality of the taRNA platform, requiring only a minor elongation of the STR sequence that has no noticeable impact on the manufacturing of the in vitro transcribed RNAs.

Materials and methods

Plasmids and RNA

Plasmids serving as templates for in vitro transcription of mRNA encoding the VEEV replicase (GenBank: L01442), an inactive replicase variant (described by Spuul et al.⁵⁰), a hyperactive replicase variant (described by Michel et al.³²), and latest generation of TR (shortened TR [STR]) were generated as previously described.^24,26 Two lentiviral vectors containing the emGFP-pre-miRNA expression cassettes were purchased (BLOCK-iT Lentiviral Pol II miR RNAi Expression System with emGFP Kit, catalog no. K4925-00, Invitrogen) and used as PCR templates to clone the miRNA cassettes into STR vectors. The mature miRNA sequence targeting either the bacterial lacZ gene or predicted to be nontargeting is flanked by loop sequences from the murine miR-155 sequence,⁵¹ which directs the excision of the engineered miRNA from a longer RNA polymerase II (RNA Pol II) transcript (pri-miRNA). All of the other artificial miRNA sequences made for the insertion into the miR-155 backbone were designed using the BLOCK-iT RNAi designer, a companion online tool (https://rnaidesigner.thermofisher.com/rnaiexpress/). The mature miRNA sequences were as follows: miR-neco: AAATGTACTGCGCGTGGAGAC; miR-lacZ: AAATCGCTGATTTGTGTAGTC; miR-luc1: AGCCCATATCGTTTCATAGCT; miR-luc2: ATACCTGGCAGATGGAACCTC; miR-p53-1: TCCACACGCAAATTTCCTTCC; miR-p53-2: AGTAGATTACCACTGGAGTCT; and miR-p53-3: CAAACACGCACCTCAAAGCTG. In silico designed pre-miRNA cassettes were ordered by custom gene synthesis (Genewiz) and cloned between the transgene-coding sequence and the alphaviral 3' CSEs of the STR plasmid. Sequences can be found in Table S1. Synthesis, cotranscriptional capping (β-S-ARCA cap), and purification of RNA were previously described.^52,53 Concentration, purity, and integrity of synthetic RNA were assessed by spectrophotometry (NanoDrop 2000c, Thermo Fisher Scientific) and capillary electrophoresis (Fragment Analyzer; Agilent).

Cell culture

Unless indicated otherwise, all growth media and supplements were supplied by Life Technologies/Gibco. Fetal calf serum (FCS) was purchased from Sigma. BHK21 cells (American Type Culture Collection [ATCC]; CCL-10) and derived transductants were grown in Eagle’s minimum essential medium (EMEM) supplemented with 10% FCS. HDF (ATCC; PCS-201-010) were grown in fibroblast medium with 2% FCS and 1% fibroblast growth supplement (Fibroblast Medium Kit; Innoprot). Human foreskin fibroblasts (ATCC; SCRC-1041) were grown in EMEM supplemented with 15% FCS, 1% nonessential amino acids, and 1% sodium pyruvate. All of the cells were cultivated at 37°C in a humidified atmosphere equilibrated to 5% CO₂.

RNA transfection

RNA was electroporated into cells at room temperature using X-VIVO 15 serum-free medium (Lonza) as electroporation buffer and applying defined pulses with a square-wave electroporator (BTX ECM 830, Harvard Apparatus). BHK-21 cells and the derived transductants were electroporated at 750 V/cm with 1 pulse of 16 ms; HDF were electroporated at 625 V/cm, with 3 pulses of 16 ms interrupted by 400 ms intervals; and HFF cells were electroporated at 500 V/cm with 1 pulse of 24 ms. Lipofection of cells with p53 siRNA (Santa Cruz, sc29435) or control siRNA (Santa Cruz, sc-37000) were performed using Lipofectamine RNAiMax (Thermo Fisher Scientific) following the manufacturer’s instructions. Lipofection of cells with taRNA was performed using Lipofectamine MessengerMax (Thermo Fisher Scientific) following the manufacturer’s instructions. Molarities or amounts of RNAs used in the experiments are indicated in the figure legends. After transfection, cells were incubated without refreshing medium until analysis.

Luciferase, β-gal, and viability assay

Firefly luciferase or β-gal expression was assessed using either the Bright-Glo Luciferase Assay System or the Beta-Glo Assay System, respectively, according to the manufacturer’s instructions (Promega). The viability of transfected cells was assessed using a luminescence-based method assaying ATP concentration over time (CellTiter-Glo assay; Promega) according to the instructions of the manufacturer. Relative viability was calculated by normalizing the value of each sample to the value of cells transfected without RNA. Bioluminescence (photons per second) of all of the assays was measured using a microplate luminescence reader Infinite M200 (Tecan Group).

Flow cytometric analysis

To determine fluorescent protein expression, transfected cells were harvested, washed once with PBS, and fixed with PBS containing 4% formaldehyde. Expression of fluorescent proteins was assessed using FACS Canto II flow cytometer and the companion FACSDiva software (BD Biosciences). FlowJo version 10 software was used for further data analyses (BD Biosciences).

Quantitative real-time reverse transcriptase PCR (qRT-PCR) for mRNA and miRNA

To assess mRNA expression levels in cells, total RNA was extracted from cell lysates (RNEasy kit; Qiagen), quantified by spectroscopy (NanoDrop 2000c, Thermo Fisher Scientific) and reverse transcribed with oligo(dT)18 primer using the Superscript IV Reverse Transcriptase (Invitrogen). The cDNA products were diluted 1:10 with nuclease-free water to serve as templates for qRT-PCR, which were performed using the ABI 7300 Real-Time PCR System, the companion SDS version 1.4 analysis software (Applied Biosystems), and the QuantiTect SYBR Green PCR Kit (Qiagen). Protocols followed the manufacturer’s instruction, with 15 min at 95°C and 40 cycles of 30 s at 95°C, 30 s at 60°C, and 30 s at 72°C. Analyses were performed using the 2^–ΔCT or 2^–ΔΔCT method,⁵⁴ normalized to the reference gene HPRT (HDF and HFF), or β-ACTIN (BHK-21 cells). The following specific primers were used: lacZ, forward: 5′-GTACGTCTTCCCGAGCGAAA-3′, reverse: 5′-CTGTTGACTGTAGCGGCTGA-3′; β-actin, forward: 5′-CCTGTATGCCAACACAGTGC-3′, reverse: 5′-ATACTCCTGCTTGCTGATCC-3′; SecNluc, forward: 5′-CTGGACCAAGTCCTTGAAC-3′, reverse: 5′-CGCTCAGACCTTCATACG-3′; TP53, forward: 5′-ACACTCGCTTCTGAATCATC-3′, reverse: 5′-GAGACCATTCATAAGCAACG-3′; TGFβR2, forward: 5′- TGAGTCCTTCAAGCAGACCGA-3′, reverse: 5′-ACACACCATCTGGATGCCCTG-3′; DAZAP2, forward: 5′- CGAACAGGAAGAGGACGAAA-3′, reverse: 5′-CAGGGTAGGTTGGCTGTGTT-3′; HPRT, forward: 5′-TGACACTGGCAAAACAATGCA-3′, reverse: 5′-GGTCCTTTTCACCAGCAAGCT-3′. To assess miRNA expression levels in cells, small RNA-containing total RNA was extracted from cell lysates (mirVana Kit, Thermo Fisher Scientific), quantified by spectroscopy (NanoDrop 2000c, Thermo Fisher Scientific), and reverse transcribed using the miRCURY LNA RT Kit (Qiagen). The cDNA products were diluted 1:60 with nuclease-free water to serve as templates for qRT-PCR using a Bio-Rad C1000 Touch Thermal Cycler, the companion CFX version 3.1 analysis software (Bio-Rad), and the miRCURY LNA SYBR Green PCR Kit (Qiagen). Protocol followed the manufacturer’s instruction with 2 min at 95°C, and 40 cycles of 10 s at 95°C and 60 s at 56°C. Analyses were performed using standard curves for absolute quantification of the pre-miR-p53-2 and 2^–ΔΔCT method for relative quantification of all five miRNAs of the miR-302/367 cluster,⁵⁴ normalized to the reference small RNA gene SNORD48. Specific LNA-enhanced primers were custom designed to specifically amplify mature miRNA sequences (miRCURY LNA miRNA custom PCR assays; Qiagen), including hsa-SNORD48 (NR_002745), hsa-miR-302a-3p (TAAGTGCTTCCATGTTTTGGTGA), hsa-miR-302b-3p (TAAGTGCTTCCATGTTTTAGTAG), hsa-miR-302c-3p (TAAGTGCTTCCATGTTTCAGTGG), hsa-miR-302d-3p (TAAGTGCTTCCATGTTTGAGTGT), hsa-miR-367-3p (AATTGCACTTTAGCAATGGTGA), and miR-p53-2 (AGTAGATTACCACTGGAGTCT).

Western blot

Total cell extracts were generated by dissolving the cell pellets in RIPA buffer and 100× Halt Protease and Phosphatase Inhibitor Cocktail (Thermo Scientific), resulting in a final buffer concentration of 1×. Samples were incubated for 30 min at 4°C on a rotary wheel, followed by a centrifugation step (16,200 × g, 4°C for 15 min) to remove cell debris. The protein concentration in cell extracts (supernatant) was measured by Pierce BCA protein assay (Thermo Fisher Scientific). Equal amounts of protein were loaded onto SDS-PAGE gels and protein transfer on nitrocellulose membrane (GE Healthcare) was performed by semidry western blot. Nonspecific binding to the membranes was blocked with 5% (w/v) skim milk powder solutions in 1× PBS-Tween. Immunostaining with primary antibodies against p53 (Santa Cruz, sc-126) and glyceraldehyde 3-phosphate dehydrogenase (GAPDH) (GeneTex, GTX627408) was performed overnight at 4°C, followed by secondary antibodies. Protein detection by chemiluminescence was performed using the Lumi-Light Western Blotting substrate (Roche) or Dura reagent (Thermo Fisher Scientific) and the ImageQuant LAS 4000 detection system (GE Healthcare). Quantification of signal intensity was performed using ImageQuant TL software (GE Healthcare). To compensate for unequal sample loading, relative expression values of proteins were normalized to corresponding relative signal intensities of loading controls.

Statistics

The data of independent experiments were summarized and displayed as mean ± SD. All of the statistical analyses were performed with GraphPad Prism 9. Tests applied to the experiments are mentioned in the respective figure captions.

Data and code availability

All of the data supporting the statements and conclusions made by the authors are included in the figures and in the supplemental information.