Engineering eToeholds from IRES modules
To engineer an RNA-sensing riboswitch that functions in eukaryotic cells, we chose to modify viral IRES modules, which possess structure-dependent translational activity10,11,12,13,14 (Fig. 1b). Although IRESs have been adapted to sense small molecules15, IRES-based systems have not been designed to respond to trRNAs. We first selected IRES modules from viral databases and tested them in a human embryonic kidney 293 (HEK293) cell-based transfection assay (Extended Data Fig. 1a). T7 RNA polymerase (RNAP) was co-transfected into cells and used to produce IRES transcripts, as T7 RNAP does not recruit 5′ capping. The presence of IRES modules resulted in a ~9-fold enhancement in expression of in cis mKate (Fig. 1c and Extended Data Fig. 1b–i). We decided to pursue cricket paralysis virus (CrPV), kashmir bee virus (KBV) and acute bee paralysis virus (ABPV) IRES modules as the basis for further development of eToeholds, with a focus on the CrPV IRES owing to its well-characterized structure and reported functionality in a wide range of eukaryotic systems10,11,12.
Having verified IRES activity, we next hypothesized that inserting short complementary RNA segments into the IRES sequence would disrupt its secondary structure by forming new loops, thereby reducing translational initiation ability, and that breaking of these introduced loops by sense–antisense action of a trRNA would rescue IRES functionality. We tested this hypothesis by inserting complementary sense and antisense segments (two pieces) of nucleic acids into sites within the IRES template, theorizing that base pairing between these sequences in the resulting transcript would distort the functional configuration of the IRES. To enable recovery of IRES function in the presence of trRNA, we designed these complementary introduced segments to bind to a chosen trRNA sequence. Our goal was for base pairing between the trRNA and the new insertion to be sufficient to break the IRES-disrupting loop, so we designed the complementary insertions to be of unequal lengths. The longer piece (40–50 base pairs) was chosen to be the reverse complement of a portion of the trRNA, whereas the shorter piece (6–15 base pairs) was chosen to be the reverse complement of a portion of the longer piece. Based on previous literature11,16, we identified eight sites that were predicted to tolerate insertions and would not on their own erase CrPV IRES activity (Fig. 1d).
We chose green fluorescent protein (GFP) mRNA to act as the trRNA and screened different CrPV IRES sequences with complementary sequences inserted at the eight possible sites. To monitor IRES activity, we used an in cis mKate gene downstream of the modified IRES sequences and co-transfected cells with these constructs as well as a plasmid constitutively expressing GFP (Fig. 1e). Each site combination is named with the format of long piece site number–short piece site number (for example, 1–2). We found that several site combinations (1–2, 1–8, 2–7, 6–7 and 8–6) behaved as expected, producing higher mKate signal when co-expressing GFP. Based on these results, we decided to focus on site combinations 6–7 and 8–6, which reproducibly showed 1.7-fold increases in the percentage of mKate positive cells in the presence of GPF trRNA, which we confirmed via GFP fluorescence.
We observed that our modified IRES appeared to retain substantial translational ability despite the newly introduced insertions. Previous studies suggest that IRES pseudoknots are critical for ribosome recruitment10,17. We reasoned that simultaneously distorting the IRES module, as in our current strategy, and breaking the IRES pseudoknot using the same insertion sequences could further reduce the basal expression of the module. To test this hypothesis, we designed new eToeholds by choosing the shorter insert to be adjacent to sequences present in critical native IRES pseudoknots, with complementary long inserts that included partial sequences of the native IRES pseudoknot. We theorized that, in the absence of trRNA, the annealing between our inserts would impair correct pseudoknot folding. We termed the insert site where this would occur as the base pair breaking site (BB site; Fig. 1f). Designing eToeholds with a BB site led to a reduction of the off (no trigger) state to background levels and increased the on-to-off fold change from 1.7 to 2.5 for site combination 8–6 (Fig. 1g and Extended Data Fig. 4). To further characterize thermodynamic requirements for eToehold switching, we altered the length of the short insert at the 8–6 site, thus changing annealing temperature. Although some eToeholds displayed a correlation between annealing temperature and output, others did not (Extended Data Fig. 2). As the longer eToehold sequence might be of sufficient length to induce an RNA interference response, we assessed RNA levels of IRES constructs designed to bind 40–50-nucleotide-long complementary ‘trigger’ RNA sequences, compared to a non-binding orthogonal sequence (Supplementary Sequences). We observed no significant differences in IRES construct RNA levels in two of the three cell types that we assessed, but we did find ~2-fold reduction in the presence of a ‘trigger’ RNA in primary human fibroblasts (Supplementary Table 2). We also found no significant differences in the production of inflammation-associated cytokines, including CCL2, CCL5 and IL-6, between cells transduced with ‘trigger’ RNA compared to non-binding orthogonal RNA, although transfection alone significantly increased cytokine production (Extended Data Fig. 3)18.
Although T7 promoter (PT7) is generally thought to be highly specific for T7 RNAP, mammalian RNA polymerase II has been shown to bind PT7 and initiate substantial levels of transcription19. We reasoned that part of the unexpected translation from eToeholds in the absence of trRNA might be due to recruitment of endogenous RNA polymerase II and subsequent generation of transcripts with 5′ caps, which independently induces translation and overrides the need for a functional IRES. Accordingly, we screened exogenous promoter sequences for transcription systems orthogonal to PT7 to test whether off-state translation of IRES-controlled reporter would be reduced. We found that the promoter for SP6 (PSP6) resulted in significantly lower basal expression than PT7 and its analogues (Extended Data Fig. 5a). We next tested upstream recruitment of RNA polymerase I, which does not result in 5′ capping and has been shown to decrease RNA polymerase II binding20,21,22, reasoning that this could further decrease basal expression. We identified an upstream activation factor binding DNA sequence from Saccharomyces cerevisiae that successfully reduced basal expression (UAF2; Extended Data Fig. 5b). By combining these components, we minimized basal expression and reached an on-to-off trigger mRNA-based induction of 15.9-fold (Fig. 2a and Extended Data Fig. 6). Similar fold changes upon induction while retaining the use of the T7 RNAP and promoter could be achieved by using a bicistronic design and adding stop codons and stem loops23 before the IRES modules (Extended Data Fig. 7a).
a, Effect of switching promoter-polymerase systems and adding an RNA polymerase I upstream activation sequence on expression of eToehold-gated transgene (mKate). b, eToehold activity, as assessed by mKate expression, in the presence of designed trRNA and unmatched RNA. c, Schematic of bicistronic RNA polymerase II-driven eToehold-gated RNA, including stop codons and stem loops. d, eToehold or IRES activity, assessed by mKate expression, of constructs with and without specific features shown in c. See Supplementary Table 1 for construct details. e, eToeholds were engineered using alternative IRES modules with higher translational activity. These new eToeholds were assessed through creation of stable HEK293T cell lines and subsequent trigger plasmid transfection; nanoluciferase expression and detection outputs are shown (Supplementary Sequences and Supplementary Table 1). Data are presented as mean values with error bars representing s.d. of three technical replicates. All experiments were repeated at least three times. a.u., arbitrary units; EV71, enterovirus 71; HCV, hepatitis C virus; PV, poliovirus.
Having developed an approach to achieve robust fold changes in transgene expression upon trRNA induction, we next tested the specificity of our system to desired trRNA sequences. We designed eToeholds to detect GFP, Azurite and yeast SUMO mRNA and found that our designs specifically sensed their trRNA sequence (Fig. 2b). We further tested the sensitivity of eToeholds to their cognate trRNA by introducing mismatches in the two insertion sequences. We found that eToeholds were sensitive to mismatches in the annealing region (Extended Data Fig. 7b). Additionally, we tested the generalizability of eToehold design to other IRESs. We synthesized eToeholds using both KBV and ABPV IRES modules and observed similar fold changes in trigger-induced translation compared to CrPV IRES module-based eToeholds (Extended Data Fig. 7c). Moreover, we found that RNA sequences that could bind to the short eToehold insertion but not to the longer insertion were not sufficient to activate the eToehold (Extended Data Fig. 7d). Taken together, these findings indicate that eToeholds can be readily designed for different mRNAs with high specificity and that sense–antisense activation is broadly generalizable for IRES-mediated translation.
We next sought to adapt our eToehold system to expression by endogenous eukaryotic polymerases. This would obviate dependency on exogenous polymerases for producing uncapped transcripts, thereby reducing construct size and increasing compatibility of our eToehold system with current cell engineering and gene therapy methods. Although RNA polymerase I24,25 has been shown to produce uncapped transcripts, we found that reliance on RNA polymerase I alone to generate the eToehold transcripts leads to significantly decreased on-to-off ratios and increased basal expression (Extended Data Fig. 5c). Therefore, we decided to explore methods of reducing translational activity in the presence of canonical 5′ capping. By adding stop codons and stem loops23 after a gene controlled by a constitutive promoter, we were able to reduce the basal expression of downstream mRNA despite the reliance on RNA polymerase II (Fig. 2c,d). Furthermore, by inserting an IRES or eToehold module between the stem loops and the coding sequence of a desired gene, thereby creating a bicistronic construct, we retained trRNA-mediated control over translation of the second gene.
To test the applicability of our system in IRES modules that evolved to use mammalian, and specifically human, translational systems, we designed eToehold modules adapted from IRES sequences of hepatitis C virus26, poliovirus27 and enterovirus 71 (refs. 28,29). These new eToeholds also demonstrated an ability to sense specific trRNAs (Fig. 2e). Compared to the CrPV constructs, these new eToeholds, which we call human-optimized Toeholds (hToeholds), produced greater than one order of magnitude more output protein. As a comparison, monocistronic 5′ capped mRNA produced even higher protein output than these hToeholds, consistent with previous findings that monocistronic constructs cause higher levels of mRNA and protein output than bicistronic constructs30.
We next sought to test the functionality of eToehold switches in different eukaryotic systems. To test the eToeholds in a single-celled eukaryote, we created strains of yeast (S. cerevisiae) expressing GFP upon galactose induction, as well as an eToehold designed to produce iRFP670 in the presence of GFP mRNA. We found that, although IRES-mediated expression was low, eToeholds were inducible by galactose-controlled GFP expression, whereas our control constructs (unmodified KBV and CrPV; Extended Data Fig. 8a,b) were not. We next tested eToeholds in plant and mammalian cellular extracts (wheat germ and rabbit reticulocyte lysate) to assess their functionality in cell-free systems. We transcribed different target RNAs and their matching eToeholds and confirmed that eToehold-mediated translation in these extracts was dependent on the presence of trRNA (Extended Data Fig. 8c,d). Furthermore, this induction was dose dependent, suggesting that eToeholds can potentially provide readouts of relative intracellular RNA levels.
eToeholds for detection of endogenous and viral mRNA
As eToeholds have demonstrated an ability to detect exogenously introduced transcripts in mammalian cells, we hypothesized that they could serve as live cell biosensors for viral infection. We generated lentiviral constructs containing eToeholds specific for Zika and SARS-CoV-2 sequences, respectively, that produced either nanoluciferase or an Azurite fluorescent protein as a readout. We transduced these constructs into the Vero E6 cell line, which is commonly used for viral studies and is derived from African green monkey kidney. We found that infection produced up to 9.2-fold increased luminescent signal in cells engineered to express Zika-specific eToeholds (Fig. 3a,b) and, furthermore, demonstrated dose-dependent responsiveness to Zika virus infection at greater sensitivities than existing live cell biosensor approaches31 (Extended Data Fig. 9a). To test the specificity of these eToeholds, we also infected transduced cells with a related Flavivirus virus—dengue virus32—and found that the eToehold response was specific to Zika infection (Fig. 3b). We observed similar results using eToeholds encoding an Azurite fluorescent protein (Extended Data Figs. 9b,c and 10). To test the ability of eToeholds to function as a live cell sensor for SARS-CoV-2, we engineered stable cell lines with eToeholds designed to sense SARS-CoV-2 transcripts. Upon transfection with constructs expressing fragments of SARS-CoV-2, we found that these eToehold-engineered cells could distinguish between the SARS-CoV-2 trRNAs and non-target RNAs (Extended Data Fig. 9d,e).
a, Stable cell lines were created with eToehold modules designed to sense infection with Zika virus. b, Luminescent signal from cells engineered to express nanoluciferase upon Zika infection after mock, Zika or dengue infection. Cells engineered with CrPV-gated nanoluciferase were used as a positive control. c, Stable cell line created with eToehold modules designed to sense exposure to heat by detecting heat shock protein mRNA. d, HeLa cells were transfected with constructs that contained a GFP reporter and eToehold-gated Azurite. e, Constructs designed to translate Azurite protein in the presence of mouse Tyr were transfected into B16, D1 or HEK293T (not shown) cells. f, Expression of eToehold-gated Azurite in B16, D1 or HEK293T cells after transfection. Data are presented as mean values with error bars representing s.d. of three technical replicates (four in Fig. 3f) exposed to the same conditions. All experiments were repeated at least three times. hsp, heat shock protein.
Finally, we explored the potential of the eToehold system for sensing endogenous transcripts, which would enable applications in identifying and targeting specific cell states and cell types. To assess the ability of eToeholds to determine cell state, we designed eToeholds to produce an Azurite protein reporter in response to transcripts of heat shock proteins hsp70 and hsp40, which are upregulated upon exposure to higher temperatures, in HeLa cells. We found that the eToehold constructs increased Azurite production up to 4.8-fold after growth at 42 °C for 24 h, as compared to routine 37 °C culture (Fig. 3c,d). Next, to assess the ability of eToeholds to differentiate between different cell types, we designed eToeholds to sense mouse tyrosinase (Tyr) mRNA, which is abundant in melanin-forming cells. With Azurite as a reporter, we observed a 3.6-fold increase in signal using Tyr-sensing eToeholds transfected into B16-F10 murine melanoma cells, as compared to two control cell lines (HEK293T and D1 marrow stromal cell; Fig. 3e,f). These results indicate that eToeholds can regulate transgene expression based on levels of intracellular, endogenous transcripts, thereby demonstrating their potential for targeting therapies to specific cell types.
