Protein-driven RNA nanostructured devices that function in vitro and control mammalian cell fate

Nucleic acid nanotechnology has great potential for future therapeutic applications. However, the construction of nanostructured devices that control cell fate by detecting and amplifying protein signals has remained a challenge. Here we design and build protein-driven RNA-nanostructured devices that actuate in vitro by RNA-binding-protein-inducible conformational change and regulate mammalian cell fate by RNA–protein interaction-mediated protein assembly. The conformation and function of the RNA nanostructures are dynamically controlled by RNA-binding protein signals. The protein-responsive RNA nanodevices are constructed inside cells using RNA-only delivery, which may provide a safe tool for building functional RNA–protein nanostructures. Moreover, the designed RNA scaffolds that control the assembly and oligomerization of apoptosis-regulatory proteins on a nanometre scale selectively kill target cells via specific RNA–protein interactions. These findings suggest that synthetic RNA nanodevices could function as molecular robots that detect signals and localize target proteins, induce RNA conformational changes, and programme mammalian cellular behaviour.

I n the nucleic acid nanotechnology field, a variety of nanostructures have been designed and constructed to utilize the programmable features of nucleic acids and the defined size and periodicity of the double-helical structure 1,2 . From this field, the concept of 'nanomachine 3 ' or 'molecular robots 4 ' has been investigated, because nucleic acids have the potential to change their conformations and functions based on the principle of simple Watson-Crick base pairing. For example, dynamic DNA nanostructures, such as the DNA walker 5 , the DNA motor 6 and the DNA nanomachine [7][8][9] , have been constructed using DNA-DNA interactions. For biological applications, it is important to develop functional nanodevices that detect various environmental signals (e.g., RNA or protein signals), induce structural changes and produce desired functions (e.g., control mammalian cell fate). Several DNA nanostructures have been generated for potential biomedical and biotechnology applications, such as target cell-surface detection 10,11 , imaging 12,13 , drug delivery 14,15 and chemical reaction control 16 . For example, a DNA-based nanorobot has been designed to detect cancer cellsurface receptors and release a drug in target cells 10 . Stimuliresponsive DNA nanohydrogels with size-controllable properties 17 and pH-or chloride-sensing DNA nanodevices have been constructed inside cells 18,19 . In addition to DNA, RNA has attracted the attention of bioengineers because of the structural diversity of RNA molecules (i.e., structured RNA uses both canonical Watson-Crick base pairing and non-canonical RNA structural motifs to form various two-dimensional and threedimensional (3D) structures) 20,21 . Several RNA nanostructures, such as triangles, squares, nanorings, three-way junctions and prisms, have been constructed in vitro [22][23][24][25][26][27][28][29][30][31][32][33][34][35] and some have been used for cellular applications through the attachment of a functional molecule, such as RNA (e.g., siRNA or aptamer) 25,27,28,32 or protein (e.g., cell-surface binder) 26,27,[31][32][33][34] , on the designed RNA structures. Synthetic RNA scaffolds that control the assembly of enzymes for hydrogen production in bacteria have also been reported 26 . However, the construction of nanostructured devices that control mammalian cellular behaviour by detecting or accumulating intracellular protein signals has not yet been demonstrated.
Inside a cell, many RNA molecules cannot function alone. RNA molecules together with RNA-binding proteins construct nanostructured RNA-protein (RNP) complexes. For example, the ribosome, which is composed of ribosomal RNAs and proteins, is a nature-made, sophisticated RNP nanomachine that catalyses protein synthesis based on the information coded in genes. Clustered regularly interspaced short palindromic repeat-CRISPR-associated proteins (CRISPR-Cas9) are another example of RNP complex-mediated nanodevices that enable the editing of a target region of genomes in a customized manner 36 . Several long noncoding RNAs have been shown to function as natural scaffolds that can control the localization and function of chromatin regulatory proteins 37 . The naturally occurring RNP interactions often control a variety of biological functions through dynamic regulation of the structures and activities of intracellular RNA or protein. Thus, we considered building synthetic RNP nanostructured devices by mimicking natural RNP complexes that have the following properties: (1) RNA-nanostructured devices detect and localize target RNA-binding proteins both in vitro and inside cells; (2) the conformation of the RNA devices is dynamically changed through specific RNP interactions; and (3) the actuation of the RNA devices produces functional outputs dependent on the extracellular and intracellular environment.
Here we report protein-driven RNA nanostructured devices that function in vitro and within live mammalian cells. Specific RNP interactions induce both structural and functional changes in the RNA nanodevices. The actuated RNA devices produce various outputs, such as the activation and repression of RNA aptamers (Fig. 1a, b) and the detection of RNA-binding protein in cells (Fig. 1c). In addition, synthetic RNA scaffolds formed in mammalian cells can selectively control cell-death pathways by detecting endogenous RNA-binding protein or microRNA (miRNA) signals and regulating the assembly and oligomerization of apoptosis-regulatory proteins on a nanometre scale (Fig. 1d).

Results
Construction of protein-driven RNA nanodevices in vitro. To design protein-responsive RNA nanodevices that induce dynamic conformational change, we employed the box C/D kink-turn (K-turn) RNA and a small RNA-binding protein from archaea, called L7Ae, as an appropriate RNP interaction motif 38 . The K-turn adopts a flexible conformation in a protein-free state, whereas L7Ae strongly stabilizes the K-turn with an approximately 60°bending angle. We have previously designed trianglelike RNP nanostructures in vitro using three K-turn-L7Ae motifs inserted into three vertexes of nanostructures composed of double-stranded RNAs [31][32][33] . However, it has remained a challenge (1) to induce both dynamic conformational and functional changes in RNA-nanostructured devices in response to a protein signal and (2) to construct protein-driven, functional RNA nanostructures inside mammalian cells.
To induce an apparent conformational change via the RNP interaction, we designed two RNA strands (long (L) and short (S) strands) in which Watson-Crick base pairs are formed within the strands but where the two ends of the double-stranded RNA do not interact with each other (Fig. 2a; protein-free double-stranded RNAs should form open, nearly linear structures). We designed two different types of RNP nanostructures that formed 'triangle instrument'-like (Tri-shape; L7Ae bends two K-turn motifs to place the two ends of the double-stranded RNA close to each other) or 'Z-shape'-like nanostructures (Z-shape; L7Ae bends two K-turn motifs in opposite directions; Figs 1a and 2a). To design the Tri-shape and Z-shape nanostructures, we incorporated two K-turn motifs into the 33 and 28 base pairs of 3 RNA duplexes corresponding to 3 and 2.5 (two and a half) helical turns, respectively (Supplementary Figs 1 and 2; 2Kt-33-Tri and 2Kt-28-Z). Through the L7Ae-induced conformational change, 2Kt-33-Tri and 2Kt-28-Z should form Tri-shape and Z-shape nanostructures. These structures enable nanometre-scale control of the distance between the RNA termini ( Fig. 2a, red arrows).
We first examined the interaction of L7Ae with 2Kt-33-Tri and with 2Kt-28-Z using an electrophoretic mobility shift assay (EMSA; Fig. 2b, c). Both 2Kt-33-Tri and 2Kt-28-Z interacted with L7Ae in a concentration-dependent manner to give shifted bands, whereas defective K-turn (dKt)-containing RNA variants that were designed to be less capable of binding to L7Ae (2dKt-33-Tri and 2dKt-28-Z; Supplementary Figs 1 and 2) 38 provided no retarded bands ( Supplementary Fig. 3), suggesting that two L7Ae proteins bind to two K-turn motifs in 2Kt-33-Tri and 2Kt-28-Z. We next analysed the structural changes of 2Kt-33-Tri and 2Kt-28-Z in the absence and presence of L7Ae by atomic force microscopy (AFM; Fig. 2d, e). In the absence of L7Ae, 2Kt-33-Tri and 2Kt-28-Z adopted heterogeneous, nearly linear RNA structures due to the flexibility of RNA and the lack of interaction between the two edges of the two RNA strands. In contrast, in the presence of L7Ae, we observed a conformational change in 2Kt-33-Tri and 2Kt-28-Z: we found that the opened, nearly linear RNA structures were dynamically changed to Tri-shape-and Z-shape-like forms due to the two L7Ae-K-turn interactions on 2Kt-33-Tri and 2Kt-28-Z, respectively, which were observed at a resolution of a single RNP interaction by AFM (Fig. 2d, e and Supplementary Fig. 4). These results indicated that two L7Ae proteins bound to two K-turn motifs of the RNA nanostructures and induced the RNA conformational change, as expected.
To further analyse the structural change caused by the RNP interaction, we measured the distance between the two ends of 2Kt-33-Tri and 2Kt-28-Z (shown in red arrows in Fig. 2a) in the absence and presence of L7Ae (Fig. 2f, g). In the absence of L7Ae, the edge distance of 2Kt-33-Tri was widely distributed from 0 to 44 nm. In contrast, the addition of L7Ae narrowed the distribution of the distance (the major population was within 16 nm), because the two ends of 2Kt-33-Tri came close to each other due to the two L7Ae-K-turn interactions that bent the RNA at the two K-turn regions to give a Tri-like shape as the major population (Fig. 2f). Similarly, we observed a distribution change in the distance between the two ends of 2Kt-28-Z in the absence and presence of L7Ae (the average distance was~21 nm; Fig. 2g), which was consistent with the expected distance from the 3D modelling of 2Kt-28-Z (~18 nm in the presence of L7Ae).
Conformational changes that regulate the distance between the two ends of structured RNAs support our concept, i.e., that we can control the distance and arrangement of RNA-conjugated molecules on the two edges through protein sensing as an input signal. To confirm our idea, we attached fluorescent Cy3 and Cy5 molecules to the edge of 2Kt-33-Tri and analysed the L7Ae-induced RNA conformational change using Förster resonance energy transfer (FRET; Supplementary Figs 5 and 6). We observed an increased fluorescence signal (F 665 /F 565 ) of 2Kt-33-Tri in the presence of L7Ae, but we did not observe this signal for the defective RNA, 2dKt-33-Tri, suggesting that FRET occurred likely due to the structural transition of 2Kt-33-Tri induced by the L7Ae-K-turn interaction.
ON/OFF switching of RNA function by changing the RNA shapes. Next, we aimed to control the function of the RNA aptamer (i.e., ON and OFF) using the L7Ae-induced RNA structural change. We chose the split RNA aptamer (binary malachite green aptamer, biMGA) 39 and conjugated each split sequence to the 5′-and 3′-termini of the S-strand in 2Kt-33-Tri or 2Kt-28-Z ( Supplementary Figs 7 and 8). In the presence of L7Ae, activated (ON) and repressed (OFF) states of the aptamer should be formed by the assembly and disassembly of two split RNA fragments through the formation of Tri-shape and Z-shape RNPs, respectively (Fig. 2h).
To modulate the function of biMGA, we optimized the length of the linker region between the RNA devices and biMGA, and the stem sequence of biMGA (Supplementary Figs 9-15 and Supplementary Note 1). We analysed the function of biMGA on the designed RNA nanostructures with malachite green (MG) in the presence and absence of L7Ae by measuring fluorescence spectra. The interaction between L7Ae and the two optimized RNA devices (Tri-MGA-ON and Z-MGA-OFF) controlled the formation of functional RNA aptamers in the ON state and OFF state, respectively (Fig. 2i). AFM analyses of Tri-MGA-ON and Z-MGA-OFF confirmed that the modulation of biMGA activity was indeed due to conformational changes in the designed RNA devices ( Fig. 2j and Supplementary Fig. 16). Thus, we are able to control both the functions (the activation and repression of the aptamer) and structures (Tri-shape and Z-shape) of the RNA nanodevices through detection of the target protein signal.
Actuation of protein-driven RNA nanodevices in living cells.
In cell-signalling pathways, proteins are central to processing information and controlling cell behaviour. We next aimed to  construct and actuate protein-responsive RNA nanodevices within live mammalian cells: the RNA nanostructures should induce a conformational change when detecting a protein expressed in target cells. We thought that our in vitro designed Z-shape nanostructures could be used for protein-detector devices that output fluorescence signals through a specific RNP interaction-mediated conformational change in cells (Fig. 3a). An RNA device with a Cy3-and Cy5-labelled short RNA duplex at the termini should produce an increase in the Cy3 signal at 565 nm (F 565 ) and a decrease in the Cy5 signal at 665 nm (F 665 ) by preventing FRET due to the formation of Z-shape RNP nanostructures. Thus, we first optimized the sequence of the short RNA duplex at the ends of the RNA devices in vitro by measuring the change in the F 565 /F 665 value before and after the RNP formation. We engineered 2Kt-17-Z by introducing Watson-Crick base pairs (from 4 to 8 bp) at the 5′-and 3′-ends of the RNA device (2Kt-17-Z-4, 5, 6, 7 and 8; Fig. 3a and Supplementary Fig. 17) and measured the fluorescence spectra of the RNA variants labelled with Cy3 and Cy5 in the presence and absence of L7Ae (Fig. 3b). The fluorescence spectra of 2Kt-17-Z-4 and 2Kt-17-Z-5 showed few changes in the presence of L7Ae. In contrast, for 2Kt-17-Z-6, 2Kt-17-Z-7 and 2Kt-17-Z-8, F 565 increased and F 665 decreased simultaneously due to the RNP formation. We found the highest fold change in the F 565 /F 665 values before and after the RNP formation when employing 2Kt-17-Z-7 (Fig. 3c). To check size effects of the RNA nanostructures, we also constructed larger Z-shape nanostructures, 2Kt-28-Z, with short RNA duplexes (from 5 to 7 bp) and confirmed that changes in the fluorescence signal were similar to 2Kt-17-Z ( Supplementary Figs 18 and 19). Thus, we chose 2Kt-17-Z-7 for the following cellular experiments.
To deliver and construct protein-responsive RNA nanostructured devices inside mammalian cells, we chose the synthetic RNA delivery approach. This approach may provide a safe tool to and chemically modified mRNA (modRNA; the use of modRNA was expected to reduce the interferon response) 40 coding for L7Ae were co-transfected into HeLa cells. The expression of L7Ae was confirmed by capillary-based immunodetection 1Kt-33-Tri 3Kt-33-Tri 6Kt-33-Tri 9Kt-33-Tri 14Kt-33-Tri 1Kt-33-Tri 3Kt-33-Tri 6Kt-33-Tri 9Kt-33-Tri 14Kt-33-Tri 1rcKt-33-Tri  Fig. 20). To detect the structural change in RNA nanodevices inside cells, we performed a flow cytometry-based FRET assay 41 using Cy3-and Cy5-labelled RNA nanodevices ( Fig. 3a and Supplementary Fig. 21). After 12 and 24 h of RNA transfection, we found that the Cy3/Cy5 signal of 2Kt-17-Z-7 increased in the presence of L7Ae compared with the signal in the presence of noncognate MS2CP protein or a defective RNA device (2dKt-17-Z-7) ( Fig. 3d and Supplementary Fig. 21d-g). Although the enhancement of the signal in the cell was less than that observed in vitro, these results indicated that the increased Cy3/Cy5 signal of 2Kt-17-Z-7 was likely due to the conformational change of the intracellular RNA nanostructure caused by specific L7Ae-K-turn interactions. Thus, the protein signal could actuate the RNA nanostructured device in cell.
RNA scaffold-based cell-fate control by assembling proteins. We next aimed to control cell fate (i.e., selective cell killing) by detecting and assembling an apoptosis regulatory protein, caspase-8 (Casp-8), through specific RNP interactions on designed RNA scaffolds formed in target cells. In naturally occurring systems, Casp-8 contains the death effector domain (DED), which is responsible for the assembly and activation of Casp-8 molecules through proximity-induced oligomerization 42 . We replaced DED with L7Ae, expecting that the fused L7Ae should bring multiple Casp-8 catalytic domains together on the RNA scaffold to induce caspase activation without requirement of the DED-mediated Casp-8 chain assembly. We therefore fused L7Ae with DED-defective Casp-8 (dCasp-8) and examined whether the RNP interaction-mediated accumulation of L7Ae-dCasp-8 on the RNA scaffold induces cell death (Fig. 1d).
To control the assembly of L7Ae-dCasp-8 on the RNA scaffold, we newly designed a set of RNA scaffolds that contained 1, 3, 6, 9 or 14 K-turn motifs (1Kt-33-Tri, 3Kt-33-Tri, 6Kt-33-Tri, 9Kt-33-Tri or 14Kt-33-Tri, respectively), which would collect a different number of L7Ae-dCasp-8 molecules on the RNA scaffold at nanometre scale ( Supplementary Fig. 22). The specific RNP interactions between 1-(3-, 6-, 9-or 14-)Kt-33-Tri and L7Ae were confirmed in vitro by EMSA and AFM ( Fig. 4a and Supplementary Figs 23 and 24). In contrast, interactions between defective RNA scaffolds with reversecomplement K-turn motifs (rcKt; they do not interact with L7Ae; Supplementary Figs 23 and 25) and L7Ae were not. We found that the structures of the RNA scaffolds were clearly changed by the L7Ae addition ( Supplementary Fig. 24). Furthermore, we directly observed that the number of L7Ae proteins assembled on the scaffolds increased as the number of K-turn motifs on the scaffolds increased (Fig. 4a). We confirmed L7Ae-dCasp-8 expression in cells by capillary-based immunodetection ( Supplementary Fig. 26a). We also performed a translational repression assay 40 and confirmed that the fusion of dCasp-8 to L7Ae maintained L7Ae-K-turn interactions in cells ( Supplementary Fig. 26b, c). To investigate whether the designed RNA scaffolds could induce cell death through the accumulation of L7Ae-dCasp-8, each RNA device and L7Ae-dCasp-8-coding modRNA were co-transfected into HeLa cells and the cells were analysed by phase microscopy or flow cytometry after 24 h of transfection. Notably, we observed that 14Kt-33-Tri with L7Ae-dCasp-8 effectively induced cell death (Fig. 4b, top right picture and Fig. 4c, left), whereas the defective RNA scaffold (14rcKt-33-Tri) with L7Ae-dCasp-8 did not (Fig. 4b, bottom right picture and Fig. 4c, right). This result indicated that L7Ae-dCasp-8 alone does not induce cell death effectively, but the accumulation of L7Ae-dCasp-8 on the RNA devices facilitates cell-death induction. In addition, we observed that the dead cell populations increased as the number of Kt on the RNA scaffolds increased (9Kt and 14Kt), whereas the number of rcKt on the scaffolds did not affect cell death (Fig. 4c, d). These results suggest that we can tune cell-death signals by changing the number of dCasp-8 assembled on the designed RNA scaffold.
To further investigate whether cell death was induced by Casp-8-mediated apoptosis through specific RNP interactions, we constructed inactive variants of Casp-8 by replacing the catalytically active cysteine residue of Casp-8 with serine (Casp-8-CS) 43 and transfected modRNAs encoding various protein variants (L7Ae, dCasp-8, L7Ae-dCasp-8, L7Ae-dCasp-8-CS, Casp-8, L7Ae-Casp-8 and L7Ae-Casp-8-CS) together with RNA scaffolds (14Kt-33-Tri or 14rcKt-33-Tri) ( Fig. 4e and Supplementary Fig. 27). We confirmed that specific interactions between L7Ae-(d)Casp-8 and 14Kt-33-Tri are necessary for the induction of effective cell death, because L7Ae or (d)Casp-8 alone did not induce cell death even in the presence of 14Kt-33-Tri ( Fig. 4e and Supplementary Fig. 27). Notably, cell death was not induced by full-length Casp-8 (or L7Ae-Casp-8) alone, whereas effective cell death was observed in 14Kt-33-Tri with L7Ae-Casp-8, indicating that the expression level of Casp-8 from modRNA is not enough to activate Casp-8 under our experimental condition, but 14Kt-33-Tri could increase the local concentration of L7Ae-Casp-8 to initiate cell death. This result suggests that RNA scaffolds may function as a signal amplifier capable of inducing cell death. In addition, we observed that 14Kt-33-Tri with L7Ae-(d)Casp-8-CS abolished the cell-death induction, indicating that Casp-8 activity is essential for triggering apoptosis mediated by the RNP interaction. We also measured caspase activity using a fluorogenic substrate (IETD-AMC). The highest caspase activity was observed in cells transfected with 14Kt-33-Tri and L7Ae-dCasp-8-coding modRNA, and the addition of a caspase inhibitor (IETD-FMK) abolished the observed activity (Fig. 4f). Together, these results confirm that we can selectively control cell-death signalling by regulating the oligomerization of pro-apoptosis regulatory proteins on synthetic RNA scaffolds formed in cells.
These results suggest that L7Ae-dCasp-8 protein assembly on 14Kt-33-Tri is essential to initiate cell death. In addition, we observed an increased level of dead cells by the addition of miR-302 or miR-21 inhibitors with the corresponding switches in hiPSCs and HeLa cells, respectively (Fig. 6b, c), indicating that endogenous miR-302 and miR-21 can control cell-death signals activated by the formation of RNP nanostructures. We also observed that the addition of a miR-302 or miR-21 mimic with the corresponding switches repressed the death of HeLa cells and hiPSCs, respectively (Fig. 6b, c). Together, these results suggest that we can control RNA scaffold-mediated cell death by detecting endogenous miRNAs and modulating protein signals in target cells.

Discussion
In this study, we demonstrated the construction of proteintriggered, functional RNA nanostructured devices in vitro and within mammalian cells. Both the structures and the functions of the devices can be controlled via specific RNP interactions, which produce various output signals (i.e., ON/OFF switching of aptamer formation, intracellular protein-responsive actuation and cell-fate control by detecting intracellular protein or miRNA signals). Protein-responsive RNA-nanostructured devices may regulate various biological processes, because the spatiotemporal regulation of naturally occurring RNP complexes has a crucial role in the control of gene expressions, signal transductions and cell-fate conversions.
Our system provides a tool for controlling the localization and accumulation of signal proteins on RNA scaffolds that were designed on a nanometre scale. In fact, we found that the oligomerization of L7Ae-dCasp-8 proteins (i.e., more than nine dCasp8 proteins) on a designed RNA scaffold is important for the induction of effective cell death. In naturally occurring systems, the assembly and activation of Casp-8 are induced by the formation of death-inducing signalling complex (DISC), which is a key step in the death receptor-mediated activation of apoptotic pathways. Dickens et al. 46 reported that DISC contains ninefold more Casp-8 than Fas-associated protein with death domain (FADD) and they proposed a DED chain assembly model for DISC formation. They demonstrated that the recruitment of many Casp-8 molecules to a single FADD molecule was required for DISC formation and subsequent apoptotic cell death. The DISC model seems to agree well with our observation that the assembly of more than nine L7Ae-dCasp-8 proteins on a single RNA scaffold facilitated effective Casp-8 activation and cell-death induction (Fig. 4). To further examine the effect of RNA structural configuration on cell-killing activity, we designed an alternative scaffold that formed 'zigzag-shaped' nanostructures (14Kt-27,28-Zig) by RNP interaction (Supplementary Fig. 32). The cell-killing activity of 14Kt-27,28-Zig was comparable to that of 14Kt-33-Tri ( Supplementary Fig. 33), suggesting that the assembly and oligomerization of dCasp-8 proteins on the RNA scaffolds are important for cell-death induction.
The molecular design of RNA nanodevices can be applied to various RNA motifs that bind to target proteins, including naturally occurring motifs and in vitro selected aptamers. Indeed, we designed and constructed alternative RNA nanodevices that control their structures by responding to U1A protein by employing U1A-binding motifs 47 (Supplementary Fig. 34). We observed that U1A protein interacted with the RNA device to induce RNA conformational change ( Supplementary Fig. 35). To actuate RNA nanodevices in response to a target protein within mammalian cells, RNA-binding proteins of interest (e.g., Lin28A) could be fused to functional proteins (e.g., dCasp-8) and the protein-binding RNA motifs can be incorporated into the RNA nanodevices. We designed a Lin28A-resposive, cell-fate control system using the 9pre-let7d-33 nanodevice (Fig. 5). We found that Lin28A expressed in HeLa-Lin28A cells or hiPSCs inhibited Lin28A-dCasp-8-mediated cell killing activity, probably due to the competitive binding between Lin28A and Lin28A-dCasp-8 to 9pre-let7d-33. These data suggest that the RNA scaffold-mediated cell death system could be modular and controllable by detecting various protein signals.
To detect endogenous miRNAs and control protein signals on the RNA scaffold, we combined our protein-responsive RNA nanostructured devices with miRNA-responsive mRNAs (miRswitches) 45 . The combination of 14Kt-33-Tri and miR-L7Ae-dCasp-8 switch selectively induced cell death in HeLa cells or hiPSCs by responding to the endogenous miRNA signatures (miR-21 or miR-302) and by regulating Casp-8-mediated apoptosis signals on the RNA scaffold. Transfection of full-length Casp-8-or L7Ae-Casp-8-encoding mRNA did not induce cell death (Fig. 4e), indicating that simple use of miRNA switches that express Casp-8 cannot control the cell phenotype efficiently. In contrast, RNA scaffold 14Kt-33-Tri amplified cell-death signals produced from miR-L7Ae-dCasp-8 switch and increased the number of dead cells in the absence of the target miRNAs. We propose that the combination of miRNA switches and synthetic RNA scaffolds can facilitate the control of miRNA-responsive cell-signalling pathways for the following reasons: (1) the simple use of miRNA switches that encode cell-death inducers may cause leaky expression of the proteins even in the presence of miRNA, which could blur the threshold between different cell types, because the expression balance between miR-switches and miRNA is critical for controlling cell death; (2) synthetic RNA scaffolds that assemble and oligomerize target proteins at nanometer scale can amplify the effect of protein signals by increasing the local concentration of the proteins on the scaffold: signal amplification processes using RNA scaffolds facilitate to control cell fate; and (3) RNA scaffolds can tune the signals by changing the assembly number of the apoptosis regulatory proteins on the scaffold (Fig. 4a-d).
To develop functional nanodevices in mammalian cells for future therapeutic applications, the introduction and formation of nanostructures inside target cells should be safe and efficient. Thus, we thought that the formation of RNP nanostructures in cells by RNA-only delivery is ideal for the following reasons: (1) the gene delivery of synthetic RNA rather than DNA is more likely to avoid potential genomic damage caused by random genomic integration of devices 40 ; (2) modRNA can produce RNA-binding proteins without an immune response 40 ; (3) the function of RNA devices can be transiently actuated inside cells, reducing the risk of potential side effects; and (4) the self-folding of RNA strands in vitro before transfection may facilitate the efficient detection of intracellular proteins via RNP interactions in the cytoplasm and the subsequent control of cell functions, because neither transcription nor RNA export from the nucleus are required in the case of RNA delivery; (5) RNA delivery approaches may have diverse medical applications, such as RNA vaccines 48 , cell reprogramming 49 , cell purification 45 and cell-fate control 40 .
The construction of protein-responsive nucleic acid nanodevices in vitro has been reported. Examples of such nanodevices include nanomechanical DNA origami that detects proteins at the single-molecule level 8 , DNA nanorobots that recognize the cell surface and control cell signalling by utilizing the interaction between DNA aptamers and protein 10 , and protein-triggered DNA nanomachines on gold nanoparticles that control the assembly and cleavage of DNA components 50 . However, the detection of proteins by DNA nanodevices has been limited to in vitro or cell surfaces. Protein-assembled RNA nanostructures have also been constructed in vitro [31][32][33][34] , although these RNA nanostructures had rather static scaffolds and their conformations and functions were not controlled by the protein signal. Our RNA nanostructured devices can be actuated via their conformational changes both in vitro and inside mammalian cells in response to extracellular and intracellular protein signals, respectively. In addition, our RNA devices are signal transducers that are capable of converting the input protein signal into the work of RNA (e.g., the formation of active aptamers) or protein (e.g., apoptosis induction) by employing a dynamic RNA scaffold. We believe that RNA and RNP nanodevices can be employed as a material to construct molecular robots that regulate their structures and produce desired functions by sensing various types of extracellular and intracellular information.
DNA and RNA preparation for constructing RNA nanostructures. DNA templates and primers were purchased from Greiner Japan. DNA templates for in vitro transcription were prepared by PCR using KOD-Plus-Neo DNA Polymerase (TOYOBO). The DNA templates were purified using a MinElute PCR Purification Kit (QIAGEN), according to the manufacturer's protocol. RNAs were transcribed by a MEGAshortscript Kit (Ambion), according to the manufacturer's protocols. The transcribed RNAs were purified by denaturing polyacrylamide gel electrophoresis (PAGE). After the recovery of the RNAs, the RNA concentration was measured using NanoDrop (Thermo Scientific). The DNA templates of 1Kt-33-Tri and 1rcKt-33-Tri were generated by PCR using synthetic oligo-DNAs. The DNA templates of 3, 6, 9 and 14Kt-33-Tri, and the corresponding mutants (3, 6, 9 and 14rcKt-33-Tri) were amplified from pcDNA4/TO-14Kt-33-Tri by PCR using the corresponding primer sets. The DNA templates were transcribed using a MEGAshortscript Kit (Ambion) and a modified protocol. Psuedouridine-5'triphosphate and 5-methylcytidine-5'-triphosphate (TriLink BioTechnology) were used instead of uridine-5'-triphosphate and cytidine-5'-triphosphate. The transcription of 9pre-let7d-33 was conducted with only cytosine modification to maintain the RNP interaction between pre-let-7d and Lin28A. Reaction mixtures were incubated at 37°C for 4-16 h. TURBO DNase (Ambion) was added to the reaction mixtures to remove template DNAs. After incubation at 37°C for 30 min, the RNAs were purified using a FavorPrep Blood/Cultured Cells total RNA extraction column (Favorgen Biotech) and then treated with Antarctic Phosphatase (New England Biolabs). After incubation at 37°C for 30 min, the resulting RNAs were purified by an RNeasy MinElute Cleanup Kit (QIAGEN) and denaturing PAGE.
Preparation of DNA template for modRNA synthesis. The 5′-and 3′untranslated regions (UTRs) were generated by using PCR with the proper sets of synthetic oligo-DNAs. The protein-coding regions of the desired proteins were prepared by PCR with the corresponding protein-containing plasmids using proper primer sets. The template plasmids were digested with the DpnI restriction enzyme (TOYOBO) for 30 min at 37°C. The PCR products were purified using a MinElute PCR Purification Kit (QIAGEN), according to the manufacturer's protocol. DNA templates for modRNA synthesis were generated by PCR using the primer set T7FwdG3C (0.3 µM) and Rev120A (0.3 µM) and using the PCR products of the 5′-UTR PCR (10 nM), 3′-UTR PCR (10 nM), and a coding region (0.2 ng µl −1 ). DNA templates for enhanced green fluorescent protein (EGFP)-myc-coding modRNA synthesis were generated by PCR using the primer set CGCT7G2 (0.3 µM) and Rev120A (0.3 µM), the PCR products of 5′-UTR PCR (10 nM) and an EGFP coding region. DNA templates for miRNA-responsive modRNAs (miRNA switches) were prepared by using the PCR products of miRNA target sitecontaining 5′-UTR (5′-UTR-T21-5p PCR or 5′-UTR-T302a-5p PCR) instead of the above 5′-UTR PCR. The PCR products were purified with the above-mentioned protocol.
Preparation of modRNAs. mRNAs were generated from the corresponding DNA templates using a MEGAscript T7 Kit (Ambion) and a modified protocol. Psuedouridine-5'-triphosphate and 5-methylcytidine-5'-triphosphate (TriLink Bio-Technology) were used instead of uridine-5'-triphosphate and cytidine-5'triphosphate. A mixture of Anti-Reverse Cap Analog (TriLink) and guanosine-5'triphosphate (4:1) was used instead of guanosine-5'-triphosphate. Reaction mixtures were incubated at 37°C for 4 h and then treated with TURBO DNase (Ambion) at 37°C for 30 min. The RNAs were purified using a FavorPrep Blood/ Cultured Cells total RNA extraction column (Favorgen Biotech) and then treated with Antarctic Phosphatase (New England Biolabs) at 37°C for 30 min. The resulting mRNAs were purified using an RNeasy MinElute Cleanup Kit (QIAGEN).
Electrophoretic mobility shift assay. Mixtures of L-RNA and S-RNA (each 20-50 nM) in 20 mM phosphate buffer, pH 7.0, 1.5 mM MgCl 2 and 150 mM KCl were heated at 80°C for 3 min and then cooled at room temperature for 10 min. After the addition of L7Ae (50-1000 nM), the mixtures were incubated at room temperature for 10-30 min. The mixtures were mixed with 10 × loading buffer (0.25% bromophenol blue, 30% glycerol) and were subjected to electrophoresis through a native polyacrylamide gel with 0.5 × Tris/Borate/EDTA buffer at room temperature. After electrophoresis, the gels were stained for 10 min with SYBR Green I and II (Lonza). Images of the gels were collected using a Gel Doc EZ Imager (BIO-RAD) or a Typhoon FLA 7000 laser scanner (GE Healthcare).
Atomic force microscopy. The RNA and RNP nanostructures were observed in solution. Their samples were prepared as described for the EMSA. The AFM images were collected using Nano Live Vision (Research Institute of Biomolecules Metrology Co.) or BIXAM (Olympus). Small cantilevers (9 µm long, 2 µm wide and 130 nm thick; BL-AC10EGS, Olympus) having a spring constant of~0.1 N m −1 and a resonant frequency of~300-600 kHz in water were used to scan the sample surface. The 320 × 240-pixel images were collected at a scan rate of 0.2 frames per second. A fresh mica surface was coated with 0.1% 3-aminopropyltriethoxy silane. The samples (20-50 nM RNA nanostructure with or without 600-1,000 nM protein) were prepared in RNP binding buffer and diluted five-(for RNP nanostructures) or tenfold (for RNA nanostructures) with AFM observation buffer (20 mM Tris-HCl (pH 7.6) and 10 mM MgCl 2 ). Then, they were applied to the mica for 5 min at room temperature and were washed with buffer solution. The image sequences were analysed using ImageJ (http://imagej.nih.gov/ij/) software.  Cell cultures. HeLa cells (originally obtained from ATCC) were cultured at 37°C in 5% CO 2 in Dulbecco's modified Eagle's medium-High Glucose (Nacalai Tesque) containing 10% fetal bovine serum (Japan Bio Serum), non-essential amino acids (Invitrogen), sodium pyruvate (Sigma) and antibiotic antimycotic solution (Sigma). HeLa cells stably expressing Lin28A (a kind gift from Dr. Kozo Tomita, Tokyo University) were established by transfecting pIRES-puro vector containing Flagtagged Lin28A sequence into HeLa cells. Positive clones were obtained by selection with puromycin (0.5 μg ml −1 ) for 2 weeks. Human iPS cells (201B7, a kind gift from Dr Masato Nakagawa, Kyoto University) were maintained in feeder-free condition 52 with StemFit AK03 (Ajinomoto) on laminin-511 E8 (iMatrix-511, Nippi). Flow cytometric-based FRET assay. A total of 5 × 10 4 cells were seeded into 12-well plates. After a 24 h culture, RNA nanodevices (8 pmol) and modRNA (500 ng) were co-transfected into the cells using 2 µl of Stemfect RNA Transfection Reagent (STEMGENT). The medium was changed after 2 h. After 12 or 24 h of transfection, the cells were washed with phosphate-buffered saline (PBS) and incubated in 200 µl of 0.25% Trypsin-EDTA (Nacalai Tesque) at 37°C for 3 min. After the addition of 100 µl of Dulbecco's PBS containing 2% fetal bovine serum, the cells were analysed with a FACSAria cell sorter (BD Bioscience). A 532 nm laser and a 575/25 nm optical filter were used to detect Cy3 fluorescence. A 532 nm laser and a 670/30 nm optical filter were used to detect the Cy5 fluorescence attributed to FRET. In the flow cytometric analysis, dead cells were removed using a dot plot of forward scatter (FSC) versus side scatter (SSC). The evaluation of the FRET signal change by the actuation of an RNA nanodevice was performed by calculating the mean of the ratio between the Cy3 and Cy5 fluorescence intensities. Enhancement of the Cy3/Cy5 mean was calculated from the Cy3/Cy5 mean normalized to cells transfected with 2dKt-17-Z-7 and either MS2CP-or L7Ae-coding modRNA. The obtained means were normalized to the mean of negative control cells transfected with MS2CP-conding modRNA and 2dKt-17-Z-7 for calculating relative Cy3/Cy5 mean intensities.
Cell-death assay. A total of 5 × 10 4 cells were seeded into 24-well plates. After a 24 h culture, the corresponding RNA nanostructures (0.1 pmol) and modRNA (10 ng) were co-transfected into the cells using 1 µl of Stemfect RNA Transfection Reagent (STEMGENT). After 2 h, the medium was changed. Cell images were taken after 24 h of transfection. The supernatants and cells were recovered and stained with Pacific Blue-labelled Annexin V (Life Technologies). Dead cells were stained with SYTOX AADvanced Dead Cell stain (Life Technologies), if needed. The stained cells were analysed using a FACSAria cell sorter and LSR Fortessa (BD Bioscience). A 405 laser and a 488 nm laser were used for the excitation of Pacific Blue and SYTOX AADvanced Dead Cell Stain, respectively; 450/50 and 660/20 nm optical filters were used to detect the fluorescence of Pacific Blue and SYTOX AADvanced Dead Cell Stain, respectively. Cells with high Pacific Blue fluorescence intensity were counted as apoptotic and dead cells.
Translational repression assay. A total of 5 × 10 4 HeLa cells were seeded into 24-well plates and incubated for 24 h. K-turn (Kt) motif-containing EGFP (Kt-EGFP)-coding modRNA (100 ng) and a mixture of L7Ae-or L7Ae-dCasp-8coding modRNA with MS2CP-coding modRNA (total 250 ng) were co-transfected into the cells using 1 µl Stemfect RNA Transfection Reagent (STEMGENT). The medium was changed after 2 h. After 24 h transfection, the cells were recovered after Trypsin-EDTA treatment and analysed with LSR Fortessa (BD Bioscience). A 488 nm laser and a 530/30 nm optical filter were used to detect EGFP fluorescence. In the flow cytometric analysis, dead cells were removed using a dot plot of FSC versus SSC. The evaluation of translational repression was performed by calculating the mean of the EGFP fluorescence intensities. The obtained means were normalized to the mean of cells transfected with Kt-EGFP-coding modRNA (100 ng) and MS2CP-coding modRNA (250 ng).
Immunodetection of proteins. A total of 2 × 10 5 cells were seeded into six-well plates. Cells transfected with various amounts of modRNAs encoding L7Ae-myc-His6, L7Ae-dCasp-8-myc-His6 and Lin28A were prepared in a similar manner as described in Cell-death assay using 4 μl Stemfect RNA Transfection Reagent (STEMGENT). After 24 h transfection, the supernatant and adhesive cells were collected and washed by PBS. The collected cells were lysed in M-PER Mammalian Protein Extraction Reagent (Thermo) for 5 min, then the debris of the cells were removed by centrifugation. The total amount of protein in the supernatant was determined by BCA assay kit (Thermo). GAPDH, Myc-tag and Lin28A were detected by an automated capillary-based immunodetection system (Wes SimpleWestern, ProteinSimple) using 100-fold diluted Anti-GAPDH (Santa Cruz, FL-355, sc-25778), 20-fold diluted Anti-Myc (Santa Cruz, sc-789) and 100-fold diluted Anti-Lin28A (R&D Systems, AF3757) antibodies as primary antibodies, respectively. HRP-conjugated Anti-rabbit (ProteinSimple, 042-206) and 100-fold diluted Anti-goat (Thermo, 81-1620) antibodies were used as secondary antibodies. The expression levels of the target proteins were quantified by Compass software (ProteinSimple).
Statistical analysis. The statistical significance was analysed by Welch's t-test (KaleidaGraph 4.5.1, HULINKS). The levels of significance are denoted as *P < 0.05, **P < 0.01, ***P < 0.001, NS not significant. All data are presented as the mean ± SD.
Data availability. The data that support the findings of this study are available from the corresponding author upon request.