Autocatalytic base editing for RNA-responsive translational control

Abstract

Genetic circuits that control transgene expression in response to pre-defined transcriptional cues would enable the development of smart therapeutics. To this end, here we engineer programmable single-transcript RNA sensors in which adenosine deaminases acting on RNA (ADARs) autocatalytically convert target hybridization into a translational output. Dubbed DART VADAR (Detection and Amplification of RNA Triggers via ADAR), our system amplifies the signal from editing by endogenous ADAR through a positive feedback loop. Amplification is mediated by the expression of a hyperactive, minimal ADAR variant and its recruitment to the edit site via an orthogonal RNA targeting mechanism. This topology confers high dynamic range, low background, minimal off-target effects, and a small genetic footprint. We leverage DART VADAR to detect single nucleotide polymorphisms and modulate translation in response to endogenous transcript levels in mammalian cells.

Fig. 1. Autocatalytic DART VADAR sensors are a practical implementation of ADAR-mediated RNA-responsive translational control.

a Schematic presenting an overview of a basic ADAR-mediated RNA-responsive translational switch. These sensors are activated by the specific hybridization of target transcripts, followed by the enzymatic deamination of the mismatched A in the central stop codon. b Illustrated are three different ways to design such circuits; these, as well as pros (in green) and cons (in orange) inherent in these designs, are summarized in panels i, ii, and iii. (i) These sensors can be designed such that there is no exogenous supplementation of ADAR. In such systems, the low levels of endogenous ADAR carry out editing of a subset of sensor molecules. (ii) Other implementations rely on constitutive overexpression of exogenous ADAR^,, which efficiently mediates editing of sensor molecules while sacrificing ease-of-delivery and increasing the unnecessary consumption of cellular resources. (iii) A potential solution that builds on these approaches is based on conditional expression of exogenous ADAR. Here, endogenous ADAR mediates editing in a subset of sensor molecules, prompting the translation of the circuit payload, including ADAR itself. After this initial step, exogenously produced ADAR increases the frequency of editing events and consequently yields higher dynamic range. m7G: mRNA cap; 2A: self-cleaving 2A peptide; AAA: poly(A) tail. c Exogenous supplementation of ADAR improves sensor performance. Numbers following the CCAs indicate the nucleotide position of the central target triplet, using the start codon as position +1. The value of each bar corresponds to the output fold-change (FC), which is the ratio of the geometric mean of mNeonGreen expression in the presence and absence of trigger. Error bars represent 95% confidence intervals for the fold-change values, determined from at least 2000 cells. d Fluorescence microscopy of mNeonGreen illustrates CCA60 sensor performance against iRFP720 in HEK293FT cells, 48 hr after transfection (Scale bar: 300 µm). e Sequencing data confirms increased A-to-I editing of the CCA60 sensor in the presence of trigger and exogenous ADAR p150. Error bars correspond to the standard deviation for n = 3 biological replicates. The sequence logo demonstrates ADAR-mediated editing is specific to the central A in the UAG stop codon. f DART VADAR relies on an initial editing step by endogenous ADAR, which is then amplified by exogenous ADAR.

Fig. 2. Optimization of DART VADAR inputs and outputs.

a The performance of sensors (black) targeting a sequence in the coding sequence of a transcript is hypothesized to be diminished due to dehybridization by ribosomes translating the trigger sequence (orange). Given this, the performance of sensors designed against secreted proteins or 3’UTR sequences is expected to be enhanced as stalled and dissociated ribosomes are less likely to disrupt sensor-trigger hybridization. SP: signal peptide; UTR: untranslated region. b ADAR-based sensors yield higher dynamic range when designed to target the 3’UTRs of transcripts, or coding sequences of secreted proteins. “allstop” indicates that all the sites in the sensor aligning with CCA sites in the target were made into editable UAG codons (as opposed to only the central codon). The value of each bar corresponds to the output fold-change (FC), which is the ratio of the geometric mean of mNeonGreen expression in the presence and absence of trigger. Error bars represent 95% confidence intervals for the fold-change values, determined from at least 2000 cells. c MCP-ADAR2dd is a compact RNA base editor, and the MCP-MS2 binding system facilitates the specific recruitment of the enzyme to the sensor edit site. NES: nuclear export sequence; ZBDs: Z-DNA binding domains; dsRBDs: dsRNA-binding domains; NLS: nuclear localization signal; MCP: MS2 major coat protein; m7G: mRNA cap; 2A: self-cleaving 2A peptide; AAA: poly(A) tail. d We tested the functionality of sensors containing MS2 hairpins without ADAR, with ADAR p150, or with MCP-ADAR2dd. The OFF and ON states correspond to mNeonGreen expression in the absence and presence of iRFP720 trigger mRNA, respectively. e MCP-ADAR specifically activates the translation of payloads encoded in sensor transcripts containing MS2 RNA hairpins.

Fig. 3. Autocatalytic DART VADAR sensors are responsive, specific, and sensitive.

a Schematic illustrating the DART VADAR circuit design. b We modified ADAR-based sensors by adding flanking MS2 hairpins in three configurations. Numeric annotations correspond to the number of base pairs between the edit site and MS2 hairpin. L, left; R, right; C, center. c As it forms an autocatalytic loop, DART VADAR is a closed-loop (CL) system. We benchmarked DART VADAR’s performance against an open-loop (OL) control, in which ADAR is constitutively expressed in trans. d The fold-change (FC) of the geometric mean of mNeonGreen expression levels is plotted for OL and CL variants. Points above the dashed line represent sensors that perform better with autocatalysis. e Closed-loop sensors with negative x-axis (basal activity ratio, CL to OL) values demonstrated a reduction of background sensor activation, and closed-loop sensors with positive y-axis (fold-change ratio, CL to OL) values showed an increase in dynamic range. f Open- and closed-loop sensors have different transfer curves for a given sensor expression level. The iRFP720 bin #1 corresponds to a “no-trigger” condition. For each variant, expression is normalized to the maximal sensor activation. g DART VADAR sensors can be designed to specifically activate in response to a point mutation (n = 3 biological replicates). WT: wild-type p53; Y220H: mutant p53. OL: open-loop, without constitutive ADAR supplementation; OL + MCP-ADAR: open-loop, with constitutive MCP-ADAR supplementation. Error bars correspond to the standard deviation for n = 3 biological replicates. h C2C12 differentiation can be steered towards the myoblastic or the osteoblastic lineage. Top right: Hoechst and CFSE staining demonstrates the presence of multinucleated syncytia (arrows) two days post-induction of differentiation. Bottom right: alkaline phosphatase activity was detected in C2C12 cells treated with BMP-2 for 8 days. The images are representative of results from n = 3 biological samples. (Scale bars: 150 µm) i RT-qPCR analysis highlights lineage-specific markers in undifferentiated and differentiated C2C12 cells. Bars represent mean and standard deviation measured on n = 3 technical replicates. j Sensors targeting endogenous Myh7 and Myog mRNAs are specifically activated two days post-induction of differentiation. Backbone: stop-less sensor; N1, N2: sensors for osteoblastic differentiation. k Sensors targeting endogenous Alp mRNA are activated on day 8 post-treatment with BMP-2. Backbone: stop-less sensor; N1, N2: sensors for myogenic differentiation.