Optogenetic techniques have transformed the biomedical sciences by controlling biological events with high spatial and temporal resolution through triggering signal transduction pathways via light-sensing proteins (Fenno et al., 2011; Möglich and Moffat, 2010; Motta-Mena et al., 2014; Cambridge et al., 2009); however, there is a need for photoactive molecules that can quickly and reversibly regulate cellular events at the RNA level. Photolabile ligands have previously been used to regulate RNA activity, but their photo-uncaging reactions are irreversible and relatively slow, requiring long UV light exposures, and resulting in limited temporal control (Dhamodharan et al., 2018; Wulffen et al., 2012; Walsh et al., 2014; Young et al., 2009; Chaulk and MacMillan, 1998; Lucas et al., 2017). Similarly, aptamers that bind a blue-light photosensitive protein have recently been developed for light-dependent regulation of translation (Weber et al., 2019). However, this system requires the expression of a relatively large regulatory protein that inhibits translation of a downstream mRNA when exposed to light and turns translation on only when slowly dark-adapted and dissociated from the RNA, preventing fast activation of protein expression. In contrast, photoreversible molecules, such as azobenzenes, have been employed to create several examples of photo-sensitive RNAs. For example, a light-responsive ribozyme has shown reversible activity, (Lee et al., 2007) and aptamers that bind photo-isomerizable ligands have been identified (Hayashi et al., 2009; Lotz et al., 2019; Young and Deiters, 2008), but these RNAs have only been characterized in vitro.

Riboswitches, cellular ligand-dependent regulatory RNAs that modulate transcription and translation in bacteria, and splicing, translation, and RNA stability in eukaryotes (Breaker, 2011) represent an attractive platform for the development of photosensitive regulatory RNAs. Riboswitches typically consist of an aptamer domain, which bind cellular targets, such as cofactors and metabolites, and an expression platform, which regulates gene expression by forming structures that affect downstream events, such as ribosome binding (Fuchs et al., 2007; Breaker, 2012). Many riboswitches sense their ligands during transcription; therefore, ligand-dependent conformational changes occur on the timescales of synthesis of these RNAs (Greenleaf et al., 2008; Uhm et al., 2018). Furthermore, in many instances the aptamer and the expression platform are distinct domains of a riboswitch, allowing molecular evolution or engineering to convert genomic riboswitches into synthetic ones through replacement of the metabolite-sensing domains with in-vitro–selected aptamers (Etzel and Mörl, 2017).

We sought to develop a synthetic riboswitch system that regulates gene expression by binding a photoreversible ligand. We chose stiff-stilbene as the molecular scaffold for the ligand design, because stilbenes are well-characterized and exhibit fast photoswitching between the trans and cis states (Quick et al., 2014; Waldeck, 1991; Szymański et al., 2013a). We used in vitro selection (Tuerk and Gold, 1990; Ellington and Szostak, 1990) to isolate a novel RNA that selectively binds only the trans isoform of amino stiff-stilbene (amino-tSS). Chemical probing identified amino-tSS–induced RNA structural changes in both the aptamer domain and a downstream expression platform derived from a bacterial riboswitch. In vitro and in vivo functional analysis showed that the riboswitch, termed Were-1, can induce or inhibit translation of a downstream open reading frame upon exposure to a sub-millisecond pulse of light, through reversible photoisomerization of the ligand. Our results demonstrate how a genetically encoded light-responsive RNA can reversibly regulate gene expression using light, providing a new optogenetic tool to broaden the analysis of complex RNA processes in living cells (You and Jaffrey, 2015; Liang et al., 2011; Jäschke, 2012; Szymański et al., 2013b).

To isolate a new aptamer fused to a functional expression platform, we constructed an RNA pool derived from a bacterial SAM-I riboswitch (Winkler et al., 2003) by replacing its ligand-binding domain with a 45-nucleotide random sequence, partially randomizing its anti-terminator and terminator hairpins, and retaining its translation initiation sequences (Figure 1—figure supplement 1). We synthesized a photoactive ligand – a trans stiff-stilbene with an amino-terminated linker designed to maintain good cell permeability (Figure 1a). The ligand was also designed to have a narrow window for photoregulation of both isomerizations in order to keep the rest of the visible spectrum available for the potential readouts through luminescent or fluorescent reporters. Amino-tSS was characterized using UV-Vis and NMR spectroscopy to confirm photoisomerization to the cis conformation at 342 nm and back to the trans conformation at 372 nm, and to ensure that both isoforms are stable on timescales relevant to pulsed gene expression (Figure 1—figure supplement 2, Figure 1—figure supplement 3, Figure 1—figure supplement 4). The RNA pool- was selected in vitro to bind amino-tSS coupled to carboxylate agarose beads and eluted under denaturing conditions (Ellington and Szostak, 1990). We hypothesized that a pool of amino-tSS–binding aptamers would include motifs that do not bind the cis photoisoform of the ligand and that the tSS–binding conformation stabilizes the expression platform in a single state that affects either transcription or translation of a downstream open reading frame (ORF). After six rounds of in vitro selection, we cloned the pool into bacterial plasmids and tested individual sequences for amino-tSS binding by monitoring RNA–dependent changes in the fluorescence of the amino-tSS. One sequence (Figure 1c) showed markedly increased fluorescence of amino-tSS (Figure 1—figure supplement 3). This sequence, termed Werewolf-1 (Were-1) for its potential light-dependent conformational changes, was chosen for further analysis.

To assess the ligand-dependent structural modulation of Were-1, we performed multiple RNA structure-probing experiments, including digestions with T1 and S1 nucleases (Peng et al., 2012; Reynolds and Gottesfeld, 1985), terbium (III) footprinting (Harris et al., 2014), in-line probing (Regulski and Breaker, 2008), and selective 2' hydroxyl acylation by primer extension (SHAPE) (Wilkinson et al., 2006) using a range of amino-tSS concentrations (Figure 1b and Figure 1—figure supplement 4). The changes in the pattern of RNA probing suggested that Were-1 undergoes conformational modulation upon introduction of amino-tSS in both the sequence derived from the randomized region (the aptamer domain) and the expression platform (Figure 1b and Figure 1—figure supplement 7, Figure 1—figure supplement 8, Figure 1—figure supplement 9). Control experiments with amino-tSS analogs, such as trans-stilbene (tS), 4,4-trans-dihydroxystilbene (tDHS), and S-adenosyl methionine (SAM), showed no change in the probing patterns (Figure 1—figure supplement 7, Figure 1—figure supplement 8, Figure 1—figure supplement 9), suggesting that Were-1 is specific for amino-tSS. Analysis of the T1 probing experiments revealed a K_D of 1.1 µM, based on the change in band intensity with increasing amino-tSS for nucleotide G42, normalized to a control band G72 (Figure 1b). A 108 µM K_D was calculated based on the same band intensity change with increasing amino-cSS (Figure 1—figure supplement 9), revealing a 100-fold specificity for the target ligand, amino-tSS. The average amino-tSS K_D derived form the T1 nuclease probing (at positions 42, 46, 77, and 80; normalized to the band intensity at position 72) was of 1.5 µM, whereas the nuclease S1 probing revealed a K_D of 0.4 µM (based on band intensity change for positions A44-G46), and the terbium (III) footprinting yielded a somewhat higher apparent K_D of 4.8 µM (calculated based on the change in intensity at positions A113 and U107, normalized to G134 control band; Figure 1—figure supplement 7).

To establish the location of the amino-tSS aptamer domain, we modeled the secondary structure of Were-1 based on the probing data and created mutants hypothesized to affect ligand binding affinity or RNA structural stability (Figure 1c), and tested them in vitro and in vivo. The secondary structure modeling did not support a conformation containing a Rho-independent transcriptional terminator, in part because the selected sequence contained two mutations (C90A and U92A), which are predicted to disrupt the stability of a full-length transcription-terminating helix (Figure 1c). The structure-probing experiments (Figure 1b and Figure 1—figure supplement 6, Figure 1—figure supplement 7, Figure 1—figure supplement 8, Figure 1—figure supplement 9) supported a secondary structure consisting of four helical segments. Ligand-dependent changes mapped primarily to the loops of the molecule, suggesting that it does not undergo rearrangement in secondary structure upon amino-tSS binding. Instead, the structure probing experiments suggest that the loops either contact the ligand directly or make tertiary contacts to form its binding pocket.

To test whether Were-1 can directly couple light-induced states of the ligand to the activity of the expression platform in vitro, the aptamer was tested for amino-tSS–dependent conformational changes using a strand-displacement assay that mimics mRNA binding by the bacterial ribosome (Martini et al., 2016; Zhang and Winfree, 2009; Martini et al., 2015). We designed a DNA duplex in which the longer strand has a toehold sequence corresponding to the reverse complement of the Shine-Dalgarno sequence of Were-1 (Supplementary file 1 - Supplementary Table 1) and a fluorophore to assess whether the shorter DNA strand, containing a quencher chromophore, is displaced through RNA:DNA hybridization with Were-1 (Figure 2—figure supplement 1a). Polyacrylamide gel electrophoresis (PAGE)-purified Were-1 bound the toehold readily, but the strand displacement was diminished in the presence of amino-tSS (Figure 2—figure supplement 1b). Testing the toehold binding at various concentrations of amino-tSS revealed a dose dependence with a half–maximal inhibition of toehold binding at ~6 µM (Figure 2—figure supplement 1c).

We next asked whether the ribosome mimic binds this RNA during in vitro transcription (Figure 2a and Figure 2—figure supplement 2). In the absence of the ligand, the RNA bound the toehold efficiently, showing a robust increase of fluorescence immediately after transcription initiation. In contrast, the addition of high concentration (14.8 µM) of amino-tSS strongly abrogated new binding of the toehold, as revealed by almost a full reduction in the slope of the fluorescence expansion curve (Figure 2b). Intermediate concentrations of amino-tSS were tested to assess RNA binding and specificity, yielding a ligand-dependent response with a half-maximum of ~4 µM amino-tSS (Figure 2c).

Probing-derived secondary structure modeling of Were-1 suggested formation of a ligand-insensitive hairpin (positions 69–84). To confirm the presence of this structural element, we created a variant containing a single mutation (G69C), that was predicted to disrupt the stem, and a presumed compensatory mutant (G69C/C84G) (Figure 1c). The G69C variant showed diminished response to amino-tSS, whereas the G69C/C84G double-mutant exhibited partially restored activity, suggesting that these two positions are indeed part of a helix. Other variants, C89G and C91G, designed based on parts of the sequence that showed amino-tSS–dependent changes in the structure-probing experiments, both showed decreased sensitivity to the ligand, suggesting that they are essential for ligand binding. Furthermore, when testing toehold binding using the purified cis isoform of the stiff stilbene (amino-cSS), as well as other stilbenes, such as tS, tDHS, and trans stiff stilbene (tSS; Were-1 ligand lacking the aminolated linker), no significant changes in fluorescence were observed (Figure 2c and Figure 2—figure supplement 3). These results demonstrate that amino-tSS stabilizes the RNA in an ‘OFF’ (ribosome-inaccessible) conformation in a dose-dependent manner and with high ligand specificity.

In order to test whether the Were-1 RNA interaction is selective for amino-tSS, and potentially acts as an amino-tSS riboswitch, we created a construct consisting of the putative riboswitch, including its minor start codon from Bacillus subtilis that was present in the starting pool, followed by a firefly luciferase (Fluc) ORF lacking its endogenous start codon. We chose luciferase for the readout because, unlike fluorescent proteins, it does not require excitation light and quantitatively yields photons immediately after the enzyme is synthesized. Based on the toehold assays, we hypothesized that in absence of amino-tSS, Were-1 would be transcribed in a conformation promoting the translation of the luciferase enzyme, whereas the presence of amino-tSS would stabilize a conformation preventing efficient translation initiation, downregulating the luciferase expression (Figure 2d). Using a purified bacterial in vitro transcription/translation system, luciferase production was measured in the presence and absence of amino-tSS. In absence of the ligand, the construct exhibited robust luciferase production, demonstrating that the unbound aptamer promotes protein production from a downstream ORF (Figure 2e). In contrast, when amino-tSS was added, protein production decreased in a dose-dependent manner (Figure 2f). To confirm that the ligand itself did not impact the in vitro translation system, luminescence was tested with a control plasmid lacking the riboswitch, and no amino-tSS sensitivity was observed (Figure 2—figure supplement 4).

To further test the riboswitch, we incorporated the construct into a bacterial plasmid and induced its expression in Escherichia coli cells (Figure 2g). Bioluminescence, due to luciferase expression and activity, was robust in the absence of ligand, and when the cells were incubated in the presence of amino-tSS, bioluminescence was again diminished in a dose-dependent manner (Figure 2h and i). To confirm the specificity of Were-1 for amino-tSS, we incubated cells in the presence of amino-cSS, tS and tDHS, and observed no significant change in bioluminescence (Figure 2i). To determine the effect of an alternate start codon on Fluc expression, we changed the UUG start codon in the Were-1-Fluc plasmid to AUG. Bioluminescence was higher in the AUG samples compared to the wild-type UUG construct and showed a similar amino-tSS–dependent response (Figure 2—figure supplement 5). Testing the above-described mutants confirmed the G69/C84 interaction, and the importance of the C89 and C91 positions for ligand binding. Taken together, our results demonstrate that Were-1 controls amino-tSS–dependent protein expression in vitro and in vivo, acting as a translational riboswitch.

We next asked whether Were-1 could regulate gene expression in a light-dependent manner, acting as a photoriboswitch. For this to occur, the trans isoform of the stiff stilbene must photoisomerize to its cis state, preventing binding of the Were-1 aptamer domain and promoting expression of a downstream ORF.

Riboswitches are sensitive to co-transcriptional events because they are capable of adopting different RNA conformations as they are transcribed by RNA polymerase (Frieda and Block, 2012). To first monitor changes in RNA folding over time, we used the toehold-fluorophore system to determine whether the DNA duplex was in a bound state (no strand displacement) versus an unbound state (strand displacement releasing the quencher DNA, yielding fluorescence) during transcription. Our results show that when the amino-tSS-bound Were-1 structure was irradiated at 342 nm of light, the toehold fluorescence increased, suggesting that the RNA increased the binding to the ribosome mimic present on the DNA toehold. Furthermore, when exciting amino-cSS at 372 nm of light to switch the ligand to its trans isoform, we observed a decrease in toehold fluorescence growth, indicating that the photo-generated amino-tSS was able to re-bind the Were-1 RNA. The irradiation was repeated until all toehold was bound, with each switch showing consistent results (Figure 3—figure supplement 1a). This experiment shows reversible, wavelength-dependent binding of the ribosome mimic, emulating light-dependent protein expression from a downstream open reading frame. Furthermore, analysis of the toehold binding as a function of exposure length revealed that even millisecond pulses of light at 340 ± 10 nm (Φ_q = 1.4*10⁻² W/cm²) result in a detectable change in toehold fluorescence when compared to unexposed transcription reaction, suggesting that the system has the capacity to act on timescales that are faster than a single nucleotide addition during RNA synthesis (Figure 3—figure supplement 1b), which for E. coli is in tens of milliseconds range (at 37°C) (Manor et al., 1969; Bremer and Yuan, 1968).

To study the system further, we used the Were-1-luciferase construct and the purified bacterial in vitro transcription/translation system and tested whether luciferase expression could be regulated by our putative photoriboswitch. We found that when the reaction was irradiated at 342 nm of light, luminescence increased, suggesting that Were-1’s conformation changed to expose the RBS, enabling luciferase expression. When the cis stilbene was photoisomerized to the trans state (amino-tSS) at 372 nm of light, luciferase protein production slightly decreased (Figure 3—figure supplement 2). These data are consistent with an in vitro activity of a photoriboswitch.

Finally, we used E. coli containing the Were-1-Fluc construct to determine whether gene expression can be regulated with a pulse of light. As shown in Figure 2i, bioluminescence was greatly diminished in the presence of amino-tSS. Upon exposing the bacteria to 342 nm light, we saw a modest increase in bioluminescence, relative to the unexposed control samples (Figure 3a and b, Figure 3—figure supplement 3), suggesting that upon photoisomerization of amino-tSS to the cis isoform with a pulse of light, Were-1 changed its conformation to expose its RBS and support luciferase production. Given that photoisomerization of stiff stilbenes from trans to cis is not quantitative (meaning, only about 40% of the ligand is photoisomerized; see Figure 1—figure supplement 2b), the trend in protein expression at individual amino-tSS concentrations can be rationalized in the following way: at low ligand concentrations, translation is weakly inhibited and photoisomerization has little impact on the rate of protein synthesis; at intermediate ligand concentrations (near the IC₅₀ concentration, that is, at the most concentration-sensitive segment of the ligand-response curve) photoisomerization leads to a change in amino-tSS concentration that results in the biggest change in translation initiation; and, at high ligand concentrations, photoisomerization results in little derepression of translation initiation because enough amino-tSS remains to continue inhibiting protein synthesis. Thus the incomplete photoisomerization of the stiff stilbene results in the bell-shaped ligand dependence upon light exposure. Additionally, in another experiment, we excited cells with a different wavelength of light (500 nm) that should not impact the isomerization of the ligand (Figure 3—figure supplement 3). Both controls showed lower bioluminescence compared to the 342 nm–excited samples, suggesting that the increased protein expression was due to derepression of translation initiation upon photoisomerization of amino-tSS to amino-cSS, although the amplitude of the effect is variable, perhaps due to variability of Were-1–regulated protein expression at different cell densities.

To further analyze the system, we tested the temporal dependence of Were-1–regulated Fluc production by exposing an amino-tSS–containing bacterial culture to 342 nm light for various lengths of time. Relative to unexposed controls at the same ligand concentration, the highest luciferase expression was seen at an exposure time of 500 µs (Figure 3b). The origin of the decrease in protein expression induction at longer exposures is currently unknown. E. coli viability begins to decrease above ~0.1 J/cm² (at 325 nm light exposure, increasing to ~1 J/cm² at 390 nm) (Vermeulen et al., 2008) and our photon flux was 0.014 and 0.055 W/cm² for the 342 nm and 390 wavelengths of light, respectively. This means that at exposures above ~10 s, the photon flux in our experiments would begin to affect the cell viability. Our experiments were performed at UV exposures well below lethal doses at these wavelengths, even at the longest exposure times; therefore, we do not expect DNA or RNA damage to cause the exposure dependence shown in Figure 3b. We confirmed the cell viability using cells expressing the Were-1-Fluc construct in absence of amino-tSS, showing insensitivity to the UV even at multi-second exposures; therefore, the decrease of protein expression at longer exposures likely results from a cellular activity of the amino-tSS that is likely unrelated to the Were-1 riboswitch activity. Nonetheless, Were-1–regulated induction of protein expression using submilisecond pulses of light represents a significant advance in the RNA optogenetics field, even if the amplitude of the responses is rather modest.

When testing the same system using amino-cSS, bioluminescence decreased in a dose-dependent manner with increasing amino-cSS concentration after exposure to 390 nm light (Figure 3c). This result strongly suggests that by isomerizing amino-cSS to its trans isoform with a pulse of light, Were-1 was able to sequester its RBS to inhibit luciferase production. Here, higher amino-cSS concentrations resulted in greater inhibition of translation, in line with the proportional conversion of amino-cSS to amino-tSS (Figure 3c). Testing the temporal response of Were-1–regulated Fluc expression in the presence of amino-cSS revealed that Were-1 again regulates protein expression optimally at short exposures, showing the highest inhibition at a millisecond of light exposure (Figure 3d). Based on the isomerization data (Figure 1—figure supplement 2, Figure 1—figure supplement 3, Figure 1—figure supplement 4), the long-exposure effect is likely due to the ligand reaching a semi-photostationary state combined with an unknown effect on the ligand observed for the longer exposures of the trans isoform. No difference in cell density was observed among the experiments, implying that neither the ligand, nor the light pulses affect the bacterial growth, and again suggesting negligible photo-damage to the cells.

We next asked whether Were-1 could reversibly regulate gene expression in vivo, providing an optogenetic tool to reversibly regulate cellular events at the RNA level. Using the same E. coli construct, we measured bioluminescence over two hours in samples that were covered during excitations, and therefore unexposed to light, samples that were exposed to a sub-millisecond of 342 nm light, and samples that were first exposed to 342 nm and later excited at 390 nm for a millisecond. Initial values prior to excitation showed no significant difference in bioluminescence twenty-five minutes post induction (Figure 3e). Samples that were then exposed to 342 nm light showed an increase in luciferase expression, compared to the unexposed control. Lastly, bioluminescence of the samples that were subsequently exposed to 390 nm light gradually decreased due to low production of new enzymes post exposure (the existing luciferases presumably continued to yield photons), compared to those that were exposed only to 342 nm, which continued protein production unabated. After ~2.5 hr, samples that were exposed to 342 nm light were 5.5 times more luminescent than unexposed samples and 1.8 times more luminescent than the samples exposed to both 342 nm and 390 nm, suggesting that luciferase expression was regulated reversibly. As an additional control, the G69C mutant was exposed alongside Were-1 and showed no significant difference when G69C excited with 342 nm light, or 342 nm and 390 nm (Figure 3—figure supplement 4). These results strongly suggest that Were-1 is a photoriboswitch that can reversibly regulate protein expression in vivo.

Regulation of cellular events with light offers many advantages over other approaches, particularly through ultrafast and spatially resolved deployment of the incident photons in live animals, making optogenetic tools invaluable for the mechanistic dissection of many biological systems (Fenno et al., 2011). Most optogenetic tools have been developed based on optical receptors, such as rhodopsins and flavin-binding proteins (Möglich and Moffat, 2010), which necessitate their adaptation to the pathways interrogated by the optogenetic tools. Methods based on these receptors are thus limited by the availability of the appropriate protein motifs and their photoactive ligands or prosthetic groups, such as flavin and retinal. Regulatory photoreceptor proteins that bind synthetic ligands have not been, to our knowledge, developed. On the other hand, in vitro selection of RNAs offers a facile approach to identify aptamers that bind a wide variety of molecular targets, including photoactive molecules (Lotz et al., 2019). Furthermore, riboswitches, cellular ligand-binding RNAs, have been identified in many bacteria, where they typically regulate transcription and translation, and eukaryotes, where they regulate translation, and RNA splicing and stability (Breaker, 2011; Zhang et al., 2020). Riboswitches therefore offer an attractive mechanism for the development of synthetic regulatory elements, with expression platforms linked to aptamers binding synthetic ligands, although only a few examples of fully de novo selected synthetic riboswitches exist (Hallberg et al., 2017).

We show that novel riboswitches, regulated by synthetic ligands, can be evolved from random libraries fused to expression platforms using in vitro selection. Werewolf-1 is the first synthetic riboswitch that binds only one photoisoform of its ligand and reversibly regulates bacterial protein expression at the RNA level. This work provides a new optogenetic tool for probing RNA-level biomolecular processes in living cells, and serves as a platform to evolve new RNAs of similar function. We expect that Were-1 and similar photoriboswitches will allow reversible photoregulation of a variety of RNA-centered cellular events with a high spatiotemporal resolution in bacteria and multicellular organisms alike, advancing the field of optogenetics. In particular, the fast action of the Were-1 photoriboswitch offers new opportunities for mechanistic studies of translational cis regulatory elements, such as riboswitches, in vitro and in vivo. Short pulses of UV light can be deployed using light sources commonly found on fluorescence microscopes coupled to mechanical shutters used in film cameras, which allow exposures down to ~100 µs. Such excitation systems thus allow triggering of photoswitches on timescales shorter than individual steps in RNA and protein synthesis (Manor et al., 1969; Bremer and Yuan, 1968; Maaløe and Kjeldgaard, 1966), and many folding transitions in these macromolecules (Vogel and Jensen, 1994; Young and Bremer, 1976; Al-Hashimi and Walter, 2008; Crothers, 2001). In contrast to photocaged ligands, the photoreversibility of the Were-1 photoriboswitch is expected to yield short pulses of translation, allowing finely tuned expression of proteins with the potential of triggering the production of a small number of regulatory proteins in single cells, as well as in individual cells in consortia of microorganisms (biofilms) or complex organisms.