Tunable gene control via RNA splicing with a clinically approved small molecule

Abstract

Precise transgene regulation is crucial for safe and effective gene and cell therapies. Current inducible systems often rely on immunogenic exogenous proteins or non-clinically approved inducers, hindering clinical translation. Here we present RisdiON, a compact inducible system controlled by risdiplam, a clinically approved oral drug that acts via splicing modulation. RisdiON utilizes risdiplam-responsive sequences for precise transgene control via endogenous splicing machinery, bypassing exogenous protein regulators, while its split-ATG architecture ensures the expression of native, tag-free proteins. This approach provides robust, dose-dependent induction with minimal leakiness. We demonstrate that RisdiON controls various transgenes in immortalized cell lines and hiPSCs, enables inducible CAR expression in primary T cells, and regulates Cas9 for precise gene editing. Additionally, we achieve reversible transgene expression in vivo using adeno-associated virus (AAV) delivery. The platform is modular and functional across diverse promoters, offering a safe, titratable, and reversible tool when coupled with an orally bioavailable drug to advance next-generation therapies.

Precise transgene regulation is crucial for the safety and efficacy of next-generation therapies. Here, authors develop RisdiON, a compact gene switch controlled by the approved oral drug risdiplam, showing this system allows for tunable, reversible regulation across diverse applications in vitro and in vivo.

Introduction

Gene and cell therapies hold a promise to cure debilitating diseases, moving beyond mere symptomatic treatments. However, widespread clinical translation remains constrained by safety concerns, regulatory complexities, and the inherent intricacies of these modalities¹. Central to these limitations is the reliance of the vast majority of approved therapies on constitutive promoters for transgene expression². Such an approach is only suitable for genes where constitutive activity is both non-toxic and therapeutically effective³. Additionally, continuous high expression can lead to cellular adaptation, resistance, or significant safety concerns⁴. Moreover, the pathophysiology of many diseases is complex and dynamic; the optimal therapeutic dose is dependent on the patient’s condition, disease or physiological state⁵. Unregulated systems simply lack the capacity for dose titration, a fundamental capability of conventional pharmacology².

To overcome these limitations, various inducible systems have been developed^2,3. While widely used preclinically, these synthetic constructs often face significant hurdles for clinical translation. Common issues include immunogenicity risk associated with exogenous regulatory proteins^6,7 (e.g., TetR, VP16), the introduction of potentially immunogenic peptide tags⁸, and the inherent challenges associated with the use of self-cleaving peptide sequences⁹ (e.g., P2A, T2A). Although 2A elements are intended to produce multiple proteins from a single transcript, cleavage is often incomplete¹⁰, producing residual peptide fragments that may compromise protein function or localization and can create additional immunogenic epitopes⁹. Multi-component designs also increase vector size and engineering complexity, which can push constructs towards viral packaging limits (particularly for adeno-associated viruses (AAV)) and complicate in vivo delivery¹¹. In addition, background expression is a recurrent limitation, as even low levels of leakiness can be detrimental for toxic gene products or lead to immune sensitization³.

These challenges, particularly the immunogenicity and off-target effects associated with transgenic protein-based regulators, have fostered a shift towards the exploration of RNA-level control mechanisms leveraging endogenous cell machinery without immunogenic foreign proteins^9,12–16. Among these, small molecule-mediated alternative splicing, which involves drug-dependent exon inclusion into mRNA, offers a promising strategy. Previous systems, such as the Xon, miniXon/miniXon2G, and the indSMN2^17,18, demonstrated proof-of-concept for splicing-based switches (derived from human SF3B3 and SMN2 genes) using the investigational compound branaplam^19,20, which ultimately ceased clinical development²¹. More recently, DreAM (drug-elicitable alternative splicing module) was reported as a risdiplam-responsive splicing cassette derived from a risdiplam-elicitable exon in murine Zfr²². In the published DreAM implementations, the splicing module is coupled to the downstream cargo using a P2A sequence, which introduces peptide sequence constraints that can be problematic for cargos requiring an unmodified N-terminus or seamless protein products^9,10,22. These considerations indicate a need for a splicing switch that combines seamless protein processing with control by a clinically approved, specific, and safe orally bioavailable small molecule.

Risdiplam^23,24 (Evrysdi™) represents such a molecule. It is an FDA-approved oral small molecule^23,25 for the treatment of Spinal Muscular Atrophy (SMA)²⁶. In SMA patients, mutations in the SMN1 gene lead to a loss of functional protein, while the SMN2 gene, nearly identical to SMN1, predominantly excludes exon 7 due to a single C-to-T base mutation in exon 7 (C6U in RNA)^27,28, resulting in insufficient levels of functional SMN protein^26,29. Risdiplam acts by binding the SMN2 pre-mRNA near the 5′ splice site to stabilize U1 snRNA recruitment, thereby promoting exon 7 inclusion and increasing functional SMN protein^26,30. Risdiplam’s clinical validation³¹, high specificity, and established record of safe long-term administration in pediatric SMA patients and in healthy volunteers^23,32 make it a suitable candidate inducer.

The demand for such precise, small-molecule-mediated gene control extends across diverse therapeutic modalities. In regenerative medicine, for example, human induced pluripotent stem cells (hiPSCs) are a technology that has transformed regenerative medicine by enabling the generation of patient-specific cellular models and autologous therapies³³. The directed differentiation of iPSCs into specific somatic lineages is a sophisticated process that relies on dynamic gene networks and is highly sensitive to gene dosage and timing^34–36. While tetracycline-responsive systems are currently widely used in iPSCs, they require exogenous regulatory proteins and are prone to epigenetic silencing³⁷. Furthermore, because many differentiation workflows already rely on doxycycline-inducible lineage drivers (e.g., NGN2), there is frequently no independent channel to regulate a second transgene without unintentionally coupling it to fate induction.

Similar timing and dosage constraints also plague other therapeutic modalities. In gene editing, constitutive Cas9 expression increases the likelihood of off-target DNA cleavage, a persistent safety concern for CRISPR-based therapies³⁸. Limiting nuclease activity to a defined temporal window can reduce genotoxicity while preserving on-target efficiency. A similar challenge arises in cellular immunotherapy: while chimeric antigen receptor (CAR) T cells  have achieved clinical success in hematological malignancies, their therapeutic window is often narrowed by severe adverse effects, including cytokine release syndrome (CRS) and T cell exhaustion, that stem from unregulated, sustained T cell activation³⁹. Consequently, there is a need for synthetic safety switches that can modulate transgene expression in a titratable and reversible manner⁴⁰. Controlling CAR expression or modulating T cell function can enhance efficacy, longevity, and safety by mitigating toxicities like CRS or preventing T cell exhaustion⁴.

In this work, we introduce RisdiON, a risdiplam-responsive splice switch platform derived from human SMN2 gene sequences. By combining rational design with systematic screening, we engineer SMN2-derived switches that demonstrate strong inducibility and low basal expression. We evaluate RisdiON in hiPSCs, demonstrating its utility as an orthogonal, compact module compatible with various promoters for independent, sustained gene expression control during differentiation. To illustrate the versatility of the RisdiON platform across diverse therapeutic modalities, we implement the switch to regulate Cas9 nuclease expression for gene editing and demonstrate its function in primary T cells as a strategy to modulate CAR-T cell therapies. Finally, we confirm the system’s functionality and full risdiplam-dependent reversibility in vivo following AAV-mediated gene delivery in mice, achieving seamless ON and OFF state transitions.

Results

Generation of risdiplam-inducible gene expression platform RisdiON

We have developed RisdiON, an inducible gene expression system controlled by risdiplam²³, an orally available clinical small molecule splicing modifier approved for treating SMA (Supplementary Fig. 1a). Risdiplam binds to the interface between U1 snRNA and the 5′ splice site (5′SS) of exon 7 of the SMN2 pre-mRNA⁴¹, promoting exon 7 inclusion and restoring the production of the full-length SMN protein (Supplementary Fig. 1b). The core concept of RisdiON relies on risdiplam-mediated inclusion of SMN2 exon 7. In the absence of risdiplam, exon 7 is predominantly skipped, resulting in out-of-frame translation through the introduction of a premature termination codon (PTC) upstream of the gene of interest (GOI), thereby preventing its translation (Fig. 1a). Upon risdiplam administration, exon 7 inclusion restores the correct reading frame, enabling the expression of the GOI. To build the inducible system, we engineered RNA splicing-based gene expression switches derived from the human SMN2 gene that consist of exons 6, 7, a 28 bp fragment of exon 8, a shortened intron 6 (5769 bp > 414 bp), and the full-length intron 7 (Supplementary Fig. 1c-d). Next, we inserted a Kozak sequence and an ATG start codon upstream of exon 6. Finally, guided by existing literature⁴², we engineered point mutations in intron 6 (A > C) and exon 7 (G > A) to reduce basal expression (leakiness) and inserted an A at position 49 of exon 7 to introduce an exon 7-mediated in-frame shift⁴³ (Supplementary Fig. 1d).

Fig. 1. Development of risdiplam-inducible splice switch.

Schematic illustration of the RisdiON splice switch. Shown are the initial design with an SMN2-derived tag (a) and the optimized tag-free, split-ATG design (b), with GA > TG mutation in exon 8 to enable ATG start codon formation after risdiplam treatment. c SMN2 exon 7 deletion screen workflow. a–c Created in BioRender. Mendel, M. (2026) https://BioRender.com/8v4jz75. d SMN2 exon 7 deletion screen strategy. Colors indicate deletion performance relative to Gen3 (corresponding to e). e Heatmap of SMN2 exon 7 deletion effects, displaying mean leakiness (median GFP in DMSO-treated cells, top) and induction ratio (bottom), normalized to Gen3. Annotated exon 7 regions include C6T, TSL1, ESE2, TSL2, and 3′-Cluster. The heatmap displays the mean of five individual biological replicates (n = 5). f RisdiON regulates GFP expression. Schematic (top) and quantification (bottom) of GFP induction in HEK293T cells treated with risdiplam for 48 h. Data points represent four biological replicates (n = 4), horizontal bar indicates the median. ****P < 0.0001. g Benchmarking strategy comparing RisdiON (Max, Tight) against previously published systems (indSMN2, Xon-series, and DreAM_Zf1) using a standardized expression vector. Switches drive GFP expression linked via a P2A peptide and contain a constitutive IRES-mCherry cassette for normalization. Quantification of GFP induction ratio in HEK293T cells transfected with the indicated constructs and treated with risdiplam (h) or branaplam (i) at 0–500 nM. Data points represent four individual biological replicates (n = 4). j Comparison of RisdiON switches with the Tet-ON inducible system. Shown are schematics (top) and induction ratios (bottom) in HEK293T cells treated with either risdiplam or doxycycline for 48 h. Data points represent two biological replicates (n = 2). k RisdiON regulation of NanoLuc reporters. Schematic (top) and induction (bottom) of secreted (SecNLuc), destabilized (NLuc-PEST), and standard NanoLuc reporters in HEK293T cells treated with risdiplam for 48 h. The induction ratio is calculated as relative light units normalized to DMSO control. Data points represent three individual biological replicates (n = 3). ***P = 0.0002, ****P < 0.0001. l RisdiON regulates Cas9-mediated CD81 reduction. Cells were co-transfected with three CD81-targeting sgRNAs and treated with risdiplam for 72 h. Quantification shows the percentage of CD81-positive cells analyzed by flow cytometry. Data points represent four individual biological replicates (n = 4). ****P < 0.0001. Statistics f, k, l: Two-way ANOVA with Tukey’s multiple comparison test. P values are indicated as follows: *P < 0.05 **P < 0.01, ***P < 0.001, ****P < 0.0001. Source data are provided as a Source Data file.

Due to translation initiation in exon 6, the first generation of the RisdiON switch (RisdiON Gen1) expresses the GOI with an N-terminal tag originating from the included SMN2 exon sequences (Fig. 1a). Such exogenous tags can interfere with protein function or cause immunogenicity^8,9. We therefore developed a split-ATG design to produce tag-free proteins. This design exploits the non-canonical 5′ splice site of SMN2 exon 7, terminating with an A nucleotide instead of a canonical G⁴² (Fig. 1b). We hypothesized that replacement of the first two nucleotides of exon 8 with TG (GA > TG) would form a de novo ATG translation initiation site (TIS) upon risdiplam-mediated exon 7 inclusion and ligation to the engineered exon 8. In the absence of risdiplam, exon 7 is skipped and the ATG start codon is not formed, preventing translation. To restrict translation initiation exclusively to the new TIS, the native ATGs in SMN2 exon 6 and any cryptic ATGs within exons 6 and 7 were mutated. Therefore, the split-ATG design directly couples risdiplam-inducible splicing with the conditional formation of a TIS immediately upstream of the GOI, preventing N-terminal extensions and enhancing the utility and translatability of the RisdiON platform.

To engineer the split-ATG system, we took a stepwise approach. First, we created the RisdiON Gen2 switch by simply replacing the initial two nucleotides of the SMN2 exon 8 fragment with TG and retaining the remaining 26 nucleotides (Supplementary Fig. 2a). The reduction of the SMN2-tag from sixty-four to seven amino acids led to a substantial increase in GFP signal and maintained a splicing efficiency comparable to RisdiON Gen1 (Supplementary Fig. 2b–g). Following this finding, we further refined the RisdiON Gen2 to a Gen3 switch by shortening exon 6 and introns 6 and 7 and by eliminating the remaining SMN2 exon 8 sequence. Although RisdiON Gen3 is a compact (618 bp), tag-free, and highly inducible system, it showed significantly increased basal GFP expression (leakiness, Supplementary Fig. 2a-i). Previous studies reported that short introns can promote the inclusion of exons with weak splice sites⁴⁴, such as SMN2 exon 7. Therefore, we evaluated multiple iterations of intron 6 and 7 lengths and found that a 25-nucleotide increase in intron 7 length significantly reduced basal expression (construct pMM472, Supplementary Fig. 3a–c), leading to the RisdiON Gen4 design that allowed for tight control and high inducibility (Supplementary Fig. 2a–i). A subsequent 5-nucleotide non-overlapping deletion screen of introns 6 and 7 confirmed that this improvement was primarily due to optimized intron length. Apart from splice site-proximal regions, no other internal regulatory elements were found to significantly alter switch performance (Supplementary Fig. 3d–g). Overall, these intron optimization efforts reinforced the importance of the splice site-proximal regions and intron length for splicing regulation.

While the aforementioned backbone optimizations were crucial to establishing a tight and highly inducible system, the exon 7 sequence itself is essential to splicing efficiency. SMN2 exon 7 harbors multiple cis-regulatory elements that govern exon inclusion and splicing efficiency^25,45, as well as sequence determinants that enable risdiplam- and branaplam- dependent modulation of splicing⁴¹. Thus, targeted modifications within exon 7 had the potential to further enhance RisdiON performance. To explore this, we conducted comprehensive deletion and mutation screens of exon 7, where we leveraged the RisdiON Gen3 backbone, which showed higher basal expression and moderate inducibility, to obtain an advantageous window for detecting enhancements (Fig. 1c–f, Supplementary Fig. 3h, i). An overlapping 1- to 5-nucleotide deletion screen across exon 7 (Fig. 1d) identified regulatory hotspots, not only confirming established sites (C6T^28,46, TSL1⁴⁷, ESE2⁴¹, and the 3′-Cluster⁴⁷) but also revealing previously uncharacterized ones that strongly affect both leakiness and induction (Fig. 1e, consistent with leaky/tight and low/high induction table). For improved RisdiON performance, further validation and optimization efforts focused on regions showing both reduced leakiness and increased induction (Fig. 1e, regions highlighted in blue). A favorable 3-nucleotide deletion (TAA) identified in the 3′-Cluster (Fig. 1e) was engineered into the intron-optimized Gen4 backbone, creating RisdiON Max, which displayed lower basal expression and improved induction ratio (Supplementary Fig. 2a–i).

To further explore these regulatory hotspots, a homology-directed repair (HDR)-based screen was implemented to introduce randomized mutations into the exon 7 region (Supplementary Fig. 3h–k). Cells were sorted by FACS into several sub-populations. Enriched variants emerged from a direct comparison of GFP high population (characterized by low background and high, risdiplam-dependent GFP expression) with GFP constitutive population. The latter contains mutations showing constitutively active GFP expression, independent of risdiplam (Supplementary Fig. 3h–i). Several variants were selected for further testing and Var-392, characterized by a 4-nucleotide deletion (CCTT) in the adjacent intron 6, demonstrated significantly lower leakiness compared to RisdiON Max (Supplementary Fig. 3j, k). Subsequent incorporation of Var-392 into RisdiON Max significantly lowered leakiness, resulting in RisdiON Tight with improved OFF state control (Supplementary Fig. 2a–i).

The performance of the iteratively optimized RisdiON constructs, in particular the Max and Tight versions, was evaluated and benchmarked against the previously published indSMN2 switch¹⁷ in HEK293T cells (Fig. 1f). Both Max and Tight constructs demonstrated robust, dose-dependent GFP reporter expression in response to risdiplam, achieving approximately 200-fold induction at highest risdiplam concentrations, exceeding the 10-15-fold induction observed with indSMN2 and Gen1 (Fig. 1f). We further confirmed dose-dependent increase in SMN2 exon 7 inclusion for all the constructs by semi-quantitative RT-PCR (Supplementary Fig. 2c–e). Notably, for RisdiON Max, the rate of exon 7 inclusion approximately doubled with each two-fold increase in risdiplam concentration, a trend that mirrors the doubling intensity observed in the functional GFP readouts. RisdiON Tight exhibited a similar dose-dependent trend at higher concentrations, though exact quantification at lower doses was limited by the detection threshold of the RT-PCR assay (Supplementary Fig. 2d–g). Quantitative GFP expression analysis further indicated that the Max and Tight vectors achieved higher induction ratios and median GFP fluorescence intensities than earlier generations, while maintaining low leakiness (Supplementary Fig. 2f–i).

To evaluate how the immediate downstream exonic sequence influences both splice-site recognition and translation initiation, we screened 61 codon variants adjacent to the splice-generated ATG in K562 cells using the RisdiON Max architecture (pMM955 library) and characterized the induction rate and basal leakiness (Supplementary Fig. 4a, b). Start (ATG) and stop codons were explicitly excluded from the randomized NNN position to ensure translation initiation remained strictly dependent on splice-mediated ATG formation.

Flow cytometric analysis revealed that RisdiON Max was broadly compatible with diverse downstream sequences: most variants retained high inducibility and clustered within a narrow performance window (log2 fold change within ±1 of the library mean) (Supplementary Fig. 4c–g). Quantitative analysis identified the first nucleotide of the codon (N1, corresponding to the +4 position from the ATG) as the dominant modulator of switch performance, whereas nucleotides at positions 2 and 3 (N2, N3) had negligible impact (Supplementary Fig. 4d–e). We observed a clear base class dichotomy at N1, where pyrimidines (C, T) consistently yielded higher induction rates than purines (A, G) (Supplementary Fig. 4d, f). In parallel, N1 strongly influenced basal leakiness, which inversely correlated with induction dynamic range: purines increased basal signal in the DMSO condition, compressing fold induction, whereas pyrimidines suppressed background signal (Supplementary Fig. 4e, g). Cytosine at N1 produced the strongest suppression of basal signal, and codons such as CGA, CGC, and CTT exhibited the lowest background in the full codon heatmap (Supplementary Fig. 4g). These effects likely reflect a combined contribution of local exon boundary context at the engineered acceptor and translation level tuning driven by the +4 nucleotide⁴⁸, rather than a strict downstream sequence requirement for splicing. Collectively, the data indicate that RisdiON Max is broadly tolerant to diverse N-terminal sequences, but for applications requiring maximal OFF state stringency, we recommend engineering the target protein N terminus to begin with cytosine.

To further define the limits of the RisdiON architecture, we investigated whether the residual background expression observed in the OFF state resulted from non-canonical translation initiation rather than basal splicing. While non-canonical translation initiation codons, such as GTG, can support weak translation initiation, they could potentially facilitate protein expression in the absence of the drug^49–51. We therefore engineered variants pMM1051 and pMM1052, based on the RisdiON Max (pMM953) backbone, designed to disrupt potential non-canonical start sites found at the exon 6 terminus and exon 8 junction of the parental pMM953 (RisdiON Max) and pMM954 (RisdiON Tight) vectors (Supplementary Fig. 5a).

Quantitative analysis across a risdiplam dose range (0–500 nM) demonstrated that eliminating these sites neither reduced basal leakiness nor significantly improved the induction ratio (Supplementary Fig. 5b, c). The induction ratios for pMM1051 and pMM1052 remained comparable to their parental controls, with representative flow cytometry overlays confirming nearly identical GFP expression profiles in both DMSO and induced conditions (Supplementary Fig. 5d). Critically, the frequency of basal expression (% GFP+ cells) and the absolute median fluorescence intensity in the OFF state were similar to, or slightly higher than, the parental Max control (Supplementary Fig. 5e–f). The failure of these mutations to attenuate background expression confirms that cryptic start sites are not the drivers of basal expression, indicating that the residual signal is instead attributable to leakiness at the level of basal exon inclusion.

To benchmark the performance of RisdiON Max and Tight, we evaluated their performance against state-of-the-art splicing switches: the risdiplam-regulated DreAM²² (DreAM_Zf1) and the branaplam-regulated Xon-series (Xon, miniXon, and miniXon2G)^17,18. For a rigorous head-to-head comparison, all systems were embedded in a standardized vector, where the switch drives GFP expression via an upstream P2A self-cleaving peptide, utilizing a constitutive IRES-mCherry normalization cassette (Fig. 1g). Because the DreAM and Xon architectures inherently rely on 2A linkages, this P2A configuration was utilized strictly to establish a uniform testing baseline. Following stable transfection into HEK293T cells, constructs were analyzed by flow cytometry, with detailed gating strategies and representative flow cytometry plots provided in Supplementary Fig. 6a, b.

Upon treatment with risdiplam, both RisdiON Max and Tight demonstrated greater dynamic ranges compared to DreAM_Zf1, whereas the SF3B3-derived Xon switches showed poor responsiveness, confirming the specificity of risdiplam for the SMN2-based systems (Fig. 1h, Supplementary Fig. 6c). Conversely, under branaplam treatment, all the tested systems exhibited robust, dose-dependent activation (Fig. 1i). Notably, RisdiON Max (non-P2A version) achieved induction ratios comparable to or surpassing those of the optimized Xon switches and DreAM_Zf1 (Fig. 1i, Supplementary Fig. 6d, e), highlighting its versatility and high performance across multiple small-molecule inducers.

Following the validation of their performance against other splicing-based switches, we next compared the optimized RisdiON Max and Tight systems with a conventional Tet-ON inducible system for GFP expression in HEK293T cells (Fig. 1j). At maximal induction (500 nM risdiplam), both RisdiON Max and Tight reached expression levels similar to Tet-ON. However, RisdiON displayed a linear, dose-dependent signal increase allowing for fine-tuned expression, in contrast to the much steeper, nearly binary (OFF-ON) response of the Tet-ON system (Fig. 1j). Furthermore, RisdiON Tight displayed comparably low basal GFP expression in the uninduced state compared to Tet-ON. Specifically, RisdiON Tight exhibited a lower percentage of GFP-positive cells (Supplementary Fig. 6h), though the median GFP intensity remained comparable between the two systems (Supplementary Fig. 6i).

To further differentiate RisdiON from alternative inducible systems, we performed a comparative toxicity assessment across six human cell models (HEK293A, HEK293T, Jurkat, K562, U2OS, and hiPSCs) using CellTiter-Glo cell viability profiling over 72 h. We systematically evaluated the safety profiles of the splicing modifiers risdiplam (Supplementary Fig. 7g–i) and branaplam (Supplementary Fig. 7j–l) alongside the conventional Tet-ON inducer doxycycline (Supplementary Fig. 7d–f) and the proteasome inhibitor bortezomib (serving as a toxicity control; Supplementary Fig. 7a-c). Under matched treatment conditions, risdiplam was consistently better tolerated than branaplam, as reflected by higher IC50 values across the panel (Supplementary Fig. 7m–o). Branaplam showed pronounced cytotoxicity at 72 h, with IC50 values as low as 176 nM in hiPSCs and 184 nM in Jurkat cells, whereas risdiplam displayed a broader tolerability window, including IC50 values > 2000 nM in four of the six models. In the most sensitive context (hiPSCs), risdiplam still showed improved tolerance relative to branaplam (72 h-IC50: 412 nM vs 176 nM) (Supplementary Fig. 7m–o).

Importantly, the maximum risdiplam concentration used for functional RisdiON induction (500 nM) remained below the IC50 thresholds for most cell models and within the tolerability range for hiPSCs under these conditions. Doxycycline was well-tolerated across the tested range, with minimal impact on viability (Supplementary Fig. 7d–f). Together, these data confirm that risdiplam provides a more robust therapeutic window for gene regulation compared to alternative splicing-based small molecules.

The performance of the Max and Tight switches was further validated using alternative reporters, such as NanoLuc luciferase (Supplementary Fig. 8a). To characterize both the dynamic response and maximal expression potential of the system, we utilized destabilized NanoLuc luciferase (NLuc)-PEST and stable NanoLuc variants (NanoLuc and secreted NanoLuc), respectively. The stable NanoLuc reporter accumulated to significantly higher levels, achieving a ~ 1000-fold induction ratio compared to the approximately 64-fold induction observed with the rapid-turnover NLuc-PEST variant (Fig. 1k and Supplementary Fig. 8b). These results confirm that the overall induction magnitude is highly dependent on the intrinsic stability of the protein of interest rather than constrained by sequence-specific interference within the switch architecture. Notably, this high dynamic range was consistent across both CMV and EF1α promoters (Fig. 1k, Supplementary Fig. 8b), demonstrating the system’s modularity and robust function across different transcriptional contexts. Furthermore, the RisdiON system exhibited rapid induction kinetics within hours of risdiplam treatment and a swift decrease in signal upon drug withdrawal, highlighting its reversibility (Supplementary Fig. 8c, d).

The versatility of the RisdiON Max switch is apparent with its ability to successfully function with various promoters in both HEK293T and K562 (Supplementary Fig. 9a, b). Across all tested conditions, the RisdiON system effectively induced gene expression at low nanomolar risdiplam doses (Supplementary Fig. 9c, d). While basal expression generally correlated with promoter strength, as evidenced by reduced leakiness of CMV in K562 (weak activity) versus HEK293T (strong activity) cells, deviations observed with promoters like CAG suggest that sequence-specific contexts also influence splicing stringency (Supplementary Fig. 9e, f).

To demonstrate the utility of the RisdiON system for functional gene regulation, we placed the Streptococcus pyogenes Cas9 nuclease under the control of the RisdiON Tight switch in HEK293T cells (Fig. 1l). When co-transfected with sgRNA guides targeting the endogenous cell surface marker CD81, these cells showed risdiplam dose-dependent reduction of CD81 expression, indicative of successful gene knockout (Fig. 1l). While the RisdiON system exhibited limited leakiness measured by low CD81 reduction in the absence of risdiplam, the efficiency of inducible Cas9 was comparable to that of a constitutively expressed Cas9 (Fig. 1l).

In summary, the RisdiON platform is compact (~640 bp), fully modular, and compatible with various promoters, GOIs and regulatory elements, enabling a plug-and-play integration into existing expression systems. These findings establish a clinically compatible, small-molecule inducible gene expression platform based on FDA-approved risdiplam, positioning it as a valuable tool for advanced gene expression control in research and therapeutic applications.

RisdiON enables multiplexed, dose-dependent, and reversible control of gene expression in human iPSCs and iPSC-derived neurons

Following the systematic engineering of the RisdiON platform (Fig. 1, Supplementary Figs. 1–4), we evaluated the efficacy of RisdiON Max and Tight and their abilities to regulate GFP expression in neurogenin-2 (NGN2)-inducible human induced pluripotent stem cells (hiPSCs) (Fig. 2a). NGN2 is a transcription factor that, when inducibly expressed in hiPSCs, rapidly and reproducibly drives their differentiation into a relatively homogeneous population of excitatory neurons⁵². Because CMV promoters have been shown to be susceptible to silencing in hiPSCs^53,54, the gene cassette was instead placed downstream of an EF1α promoter (Fig. 2a). A clear risdiplam dose-dependent increase in GFP reporter expression was observed, with RisdiON Max inducing GFP expression at lower concentrations compared to RisdiON Tight (Fig. 2a). Quantification by flow cytometry showed overall reduced basal GFP expression for RisdiON Tight, as evidenced by lower median GFP expression (Fig. 2b) and significantly smaller proportion of GFP-positive cells in DMSO (Supplementary Fig. 10b). Despite the lower expression, RisdiON Tight achieves a higher induction ratio compared to RisdiON Max due to its decreased leakiness (Supplementary Fig. 10b) and is thus suitable for applications where minimal basal expression is preferred over high expression levels (Supplementary Fig. 10a). Overall, RisdiON’s consistent responsiveness across various risdiplam levels facilitates tunable gene control at clinically relevant risdiplam concentrations.

Fig. 2. Functional validation of risdiplam-inducible splice switch in hiPSCs.

a RisdiON regulates GFP expression in hiPSCs. Schematic of the pEF1α-driven RisdiON GFP expression vector (top) and representative fluorescence images (bottom) of NGN2-inducible hiPSCs with stable integration of RisdiON Max (pMM582) or RisdiON Tight (pMM810). Cells were treated for 24 h at the indicated risdiplam concentrations. Representative fluorescence images from n = 3 experiments (GFP: green; DAPI: blue. Scale bar: 100 µm (20x objective)). b Flow cytometry analysis of median GFP fluorescence intensity in NGN2-hiPSCs, transfected as in a and treated for 48 h at indicated risdiplam concentrations. Data points represent two biological replicates for pMM810 (n = 2) and four biological replicates for pMM582 (n = 4), the horizontal bar indicates median. c Risdiplam-inducible Cas9-mediated gene editing. Schematic of the gene cassette driving CD81 sgRNA and RisdiON Tight-Cas9 (top) and flow cytometry analysis of CD81 reduction (bottom). Stable lines were treated with risdiplam for 48 or 72 h; the graph displays the percentage of CD81-positive cells. Data points represent technical replicates of two individual experiments (n = 2). Two-way ANOVA with Tukey’s multiple comparison test was performed. ****P < 0.0001. d Schematic of the NGN2-mediated hiPSC-to-neuron differentiation protocol. Timeline showing the 14-day protocol with key reagents addition. Arrows indicate risdiplam treatment for experiments detailed in e, f where cells were treated for 24 h before imaging. Created in BioRender. Mendel, M. (2026) https://BioRender.com/uox1g9r. e Risdiplam-inducible GFP expression in NGN2-hiPSCs. Left: Schematic of RisdiON Tight (pMM810) and Tet-ON NGN2 constructs. NGN2-hiPSCs were differentiated as in d and treated with 250 nM risdiplam or DMSO at the indicated times (arrows). Right: Representative fluorescence images from n = 3 experiments (GFP: green; DAPI: blue). Scale bar, 100 µm (20x objective). f Multiplexed control of gene expression in NGN2-hiPSCs. Left: Strategy for the orthogonal control of three inducible systems: Tet-ON-NGN2, RisdiON-GFP, and VA-OFF-MitoTimer. Right: Representative images of DAPI (blue), GFP (green), and MitoTimer-RFP (red) in neurons on day 8 of differentiation. Cells were treated with 250 nM risdiplam and/or 250 µM vanillic acid (VA) 24 h prior to imaging. Labels indicate the specific activity states of RisdiON and MitoTimer. Scale bar, 100 µm (20x objective). Representative images from n = 3 experiments. Source data are provided as a Source Data file.

Next, we applied RisdiON Tight to control Cas9 expression in hiPSCs, a critical step for minimizing off-target edits and preserving genomic integrity during gene editing. The gene cassette incorporated a dual-promoter system with an EF1α promoter driving RisdiON Tight-Cas9 expression and a U6 promoter for CD81 sgRNA expression (Fig. 2c). Puromycin-selected hiPSCs showed a risdiplam dose- and time-dependent decrease in CD81 surface expression, indicating successful inducible gene editing (Fig. 2c).

Beyond direct gene editing, precise temporal control of gene expression is critical during cellular differentiation. We therefore evaluated the RisdiON system during the directed differentiation of hiPSCs into neurons using a doxycycline (dox)-inducible NGN2-based protocol⁵⁵ (Fig. 2d). NGN2 hiPSCs with stable expression of the RisdiON Tight-GFP reporter (pMM810) were subjected to neuronal differentiation and monitored for GFP expression following treatment with either 250 nM risdiplam or DMSO. Fluorescence microscopy revealed robust risdiplam-dependent GFP induction throughout the 14-day neuronal differentiation, demonstrating the RisdiON system’s sustained functionality and responsiveness (Fig. 2e). This highlights its utility for stage-specific gene expression studies in developing neuronal lineages.

Finally, to demonstrate its multiplexing potential, the RisdiON platform was integrated with two additional inducible systems: a doxycycline-inducible NGN2 cassette and a vanillic acid (VA)-repressible VA-OFF module⁵⁶. The VA-OFF system actively silences target gene expression upon VA administration, and in this context, was used to control the expression of a MitoTimer⁵⁷. MitoTimer is a mitochondrial-targeted fluorescent protein that undergoes an oxidative shift in emission from green to red as it matures, distinguishing newly synthesized (green) from older (red) mitochondria, thereby enabling mitochondrial maturation and turnover assessment. After eight days of doxycycline-mediated neuronal differentiation, the three modules could be operated in parallel within the same cultures: doxycycline supported NGN2-driven differentiation, VA modulated MitoTimer output, and risdiplam induced RisdiON Tight-GFP expression (Fig. 2f). The RisdiON Tight-GFP reporter showed robust induction at 250 nM risdiplam from an undetectable basal level under these imaging conditions (Fig. 2f). In contrast, although VA treatment reduced MitoTimer expression, residual signal indicated appreciable background from the VA-OFF system under these conditions. This integrated inducible setup was also active in undifferentiated hiPSCs (Supplementary Fig. 10c). Combined, the data highlight the tight regulation and the robust inducibility of the RisdiON platform within a complex, multi-regulated cellular environment, offering broad utility for complex biological studies.

Risdiplam-mediated control of CAR expression in primary human T cells

Despite significant successes in haematological malignancies, CAR-T cell therapies face challenges including CRS and limited persistence due to T cell exhaustion⁵⁸, often exacerbated by constitutive and high-level CAR expression. To address these challenges, we developed a regulatable CAR expression system that uses the RisdiON platform to enable precise control over CAR expression on T cells (Fig. 3a).

Fig. 3. Risdiplam-inducible CAR-T cells reach full killing competency at lower levels of CAR expression and milder activation status.

a Schematic representation of the RisdiON CAR-T system. Administration of risdiplam induces expression of chimeric antigen receptors (CARs) on engineered T cells, enabling precise modulation of CAR-T activity. Created in BioRender. Mendel, M. (2026) https://BioRender.com/5xyzygo. Representative flow cytometry of CAR surface expression. Shown are comparisons of RisdiON BCMA-CAR-T against untransduced wild type (WT) T cells (b) and conventional BCMA-CAR-T cells (c), alongside a risdiplam dose-response in RisdiON cells (d). e Dose-response of risdiplam-induced CAR expression for BCMA- and CD19-targeted CARs after 16 h of incubation. Two technical replicates from three donors (n = 3), mean ± SD. f Flow cytometry of activation markers (PD1, CD25, CD69) and phenotype (CD62L/CD45RO) in the absence of antigen for BCMA-CAR-T cells after 14 days of culture. Two technical replicates from one individual donor (n = 1), data presented as individual values or mean (frequency of CAR-T cells). g Dose-response relationship between risdiplam and killing of RPMI-8226 target cells using RisdiON BCMA-targeted CAR-T cells, measured using CytoTox-Glo™ Assay after 48 h of co-culture. Two technical replicates from three donors (n = 3), mean ± SD. h Comparative phenotypic analysis of RisdiON and conventional BCMA-targeted CAR-T cells. Classification into TTE, TEM, TCM, and TSCM phenotypes was assessed by flow cytometry gating on CD62L and CD45RO after 48 h of co-culture. Two technical replicates from two donors (n = 2). i Dose-response relationship between risdiplam and killing of CD19+ Raji target cells using RisdiON CD19-targeted CAR-T cells, measured using CytoTox-Glo™ Assay after 48 h of co-culture. Technical triplicates from two donors (n = 2), mean ± SD. j Comparative phenotypic analysis of RisdiON and conventional CD19-targeted CAR-T cells. Classification into TTE, TEM, TCM, and TSCM phenotypes was assessed by flow cytometry gating on CD62L and CD45RO after 48 h of co-culture. Two technical replicates from three donors (n = 3). Two-way ANOVA with Tukey multiple comparisons correction was used for intra-phenotype comparisons. ***P = 0.0004, ****P < 0.0001. k Flow cytometry of inhibitory T cell receptors (PD1, Lag3) on RisdiON versus conventional CD19-CAR-T cells after 48 h of co-culture. Two technical replicates from three donors (n = 3), mean ± SD. Unpaired, two-tailed Student’s t-test was performed. ***P = 0.0002, ****P < 0.0001. Data presented as mean (TSCM: T memory stem-like cells; TCM: central memory T cells; TEM: effector memory T cells; TTE: terminal effector T cells). Source data are provided as a Source Data file.

We first validated RisdiON inducibility in Jurkat T cells by stably expressing either a B cell maturation antigen (BCMA)-targeted CAR, a CD19-targeted CAR, or GFP as a control, all regulated by RisdiON Max. Both BCMA- and CD19-targeted CAR constructs demonstrated clear risdiplam-dependent surface expression by flow cytometry (Supplementary Fig. 11a, b). Furthermore, an NFAT-luciferase reporter assay in these engineered Jurkat cells showed risdiplam dose-dependent activation upon co-culture with cognate antigen-expressing target cells, indicating risdiplam-controllable functional CAR signaling (Supplementary Fig. 11c–f).

Building on these findings, we applied the RisdiON system to primary human T cells. We generated second-generation self-inactivating lentiviral vectors encoding either a BCMA- or CD19-targeted CAR under RisdiON Max control. Primary human T cells were transduced, expanded, and compared to constitutively expressed CAR constructs (see workflow in Supplementary Fig. 11g). Flow cytometry analysis revealed minimal CAR expression in inducible BCMA-CAR-T cells (RisdiON BCMA-28z) without risdiplam (Fig. 3b). Upon addition of increasing doses of risdiplam, CAR expression increased, reaching levels comparable to constitutively expressed (conventional) BCMA-CAR-T cells (Fig. 3c, d). A clear risdiplam dose-dependent relationship with CAR expression in addition to reproducible dynamic range were observed for both RisdiON BCMA-28z and RisdiON CD19-28z cells across three independent donors (Fig. 3e). Importantly, in the absence of antigen stimulation and risdiplam, inducible RisdiON CAR-T cells exhibited baseline expression levels of the inhibitory receptor PD1, activation markers CD69 and CD25, and T cell differentiation markers comparable to those of non-inducible CAR-T cells and untransduced wild type T cells (Fig. 3f). Specifically, we characterized four phenotypic subsets representing distinct maturation states based on the expression of CD45RO and CD62L: stem cell-like memory T cells (TSCM; CD45RO- CD62L + ), central memory T cells (TCM; CD45RO + CD62L + ), effector memory T cells (TEM; CD45RO + CD62L-), and terminal effector T cells (TTE; CD45RO- CD62L-). This finding suggests that the RisdiON system does not cause tonic signaling or premature T cell activation.

Next, we assessed the cytotoxic potential of the inducible CAR-T cells. RisdiON BCMA-CAR-T cells demonstrated risdiplam dose-dependent killing of BCMA + RPMI-8226 target cells, achieving maximal lysis comparable to that of constitutively expressed CARs (Fig. 3g). Despite achieving comparable maximal killing effects, phenotypic analysis revealed a critical difference: RisdiON BCMA-CAR-T cells stimulated with the highest risdiplam concentration maintained a higher proportion of less differentiated T memory stem-like (TSCM) cells, a subset frequently associated with improved persistence and clinical efficacy^59–61. In contrast, unregulated conventional BCMA-CAR-T cells showed a tendency towards more differentiated effector memory (TEM) and terminal effector (TTE) phenotypes (Fig. 3h). Similar risdiplam dose-dependent cytotoxicity was observed for RisdiON CD19-CAR-T cells incubated with CD19+ Raji target cells (Fig. 3i). Phenotypic analysis of RisdiON CD19-CAR-T cells showed a statistically significant trend towards a less differentiated phenotype (higher TSCM and TCM) compared to conventional CD19-CAR-T cells (Fig. 3j). Furthermore, RisdiON CD19-CAR-T cells exhibited significantly lower expression of the inhibitory receptors PD1 and Lag3 compared to conventional CD19-CAR-T cells during killing assays at maximal risdiplam concentrations (Fig. 3k), suggesting a less exhausted T cell state.

Together, these results demonstrate that the RisdiON platform is compatible with lentiviral vectors, enabling tunable, small molecule-regulated CAR expression in primary human T cells. This system achieves minimal background CAR expression, promotes the maintenance of less differentiated memory phenotypes, and allows for controllable effector function with reduced expression of exhaustion markers.

Evaluation of the RisdiON switch in modulating AAV9-mediated gene expression in BALB/c mice

Having established the RisdiON platform’s efficacy and favorable characteristics for ex vivo cell therapy applications using lentiviral delivery, we next sought to evaluate its performance and regulatory capacity in vivo using AAV vectors. We first determined the optimal AAV dose for inducible transgene expression in mice. BALB/c mice were intravenously injected with varying doses of AAV9 vector encoding a firefly luciferase reporter (Fluc-PEST) under the control of a CAG promoter and the RisdiON Max switch component (Fig. 4a, b). Phosphate-buffered saline (PBS) was used as a negative control. Three weeks post-injection, risdiplam was administered at 15 mg/kg via oral gavage followed by intraperitoneal injection of D-luciferin to enable bioluminescence signal detection (Fig. 4b).

Fig. 4. Risdiplam-controlled reversible gene expression in vivo via repeated dosing.

a Schematic of the engineered AAV vector genome (AAV9 serotype, AAV2 genotype) encoding a red-shifted firefly luciferase (Fluc) fused to a PEST degradation domain for rapid turnover. Expression is driven by a minimal CAG promoter and regulated by the pMM569 RisdiON Max splice switch. The cassette also includes a Woodchuck Hepatitis Virus Posttranscriptional Regulatory Element (WPRE) and a bovine growth hormone (bGH) polyadenylation signal. b Overview of the in vivo study design. Female BALB/c mice were intravenously (I.V.) injected with AAV9-RisdiON-Fluc-PEST at 1 × 10¹³ vector genome (vg) per kg body weight (vg/kg) or with PBS as a negative control (Neg. Ctrl). Three weeks post-AAV administration, mice received risdiplam (15 mg/kg) via oral gavage. Bioluminescence imaging (BLI, using the Vilber Newton system) was conducted five hours after risdiplam administration, following intraperitoneal (I.P.) injection of D-luciferin. Created in BioRender. Mendel, M. (2026) https://BioRender.com/8hsub1k. c Schedule of the repeated dosing of risdiplam and D-luciferin to evaluate the reversible dynamic of the switch. d Representative in vivo bioluminescence images displaying Fluc-PEST activity, with ten animals in the AAV-injected group and eight animals in the PBS-treated negative control group. e Quantification of bioluminescent signal, measuring total photon flux per second (ph/s). ns = not significant, ****P < 0.0001. f Quantification of transduction efficiency by droplet digital PCR (ddPCR). ddPCR was performed on genomic DNA extracted from liver tissue samples to determine the absolute copy number of the WPRE sequence contained within the AAV9-RisdiON-Fluc-PEST cassette (shown in a). Data is presented as viral genome copies per cell. Individual data points are presented (n = 8 for the Neg. Ctrl group; n = 10 for the AAV9-transduced group), the horizontal line indicates the mean value. ****P < 0.0001. (e, f) Statistical analysis was performed using one-way ANOVA with post-hoc Tukey HSD test. Source data are provided as a Source Data file.

Bioluminescent imaging conducted four hours after risdiplam administration revealed an AAV titer-dependent expression of Fluc-PEST (Supplementary Fig. 12a) with peak signal intensity observed at the highest AAV dose (1 × 10¹³ vg/kg). After 72 hours, the signal was absent, indicating the system’s transition to an OFF state due to both the short half-life of Fluc-PEST and the absence of risdiplam (Supplementary Fig. 12a, b). Based on these results, we selected the highest AAV dose (1 × 10¹³ vg/kg) to further assess the dynamic and reversible control of the RisdiON switch in vivo via repeated subchronic risdiplam administrations (Fig. 4c). Fluc-PEST activity images (Fig. 4d) and quantitative analysis (Fig. 4e) clearly illustrate the system’s dynamic and fully reversible control, showcasing cycles of induction and reversion to an OFF state in the absence of risdiplam. Notably, OFF state signals were indistinguishable from PBS controls, indicating undetectable leakiness, and were consistent with droplet digital PCR results showing viral genome copies only in AAV-injected animals (Fig. 4f). The RisdiON system provides AAV dose-dependent expression that is dynamically and reversibly regulated by risdiplam’s presence or absence, exhibiting minimal background activity and validating its robust potency in vivo.

Discussion

Here we present RisdiON, a versatile RNA splicing-based platform for gene expression regulation. Through its inherent design, the RisdiON system overcomes key limitations of existing technologies. Primarily, RisdiON operates via endogenous splicing machinery. This contrasts sharply with traditional systems like Tet-ON that rely on foreign bacterial proteins (e.g., rtTA), which are prone to eliciting immune responses and gene silencing^6,7,37. In addition to avoiding exogenous elements, RisdiON offers more precise, dose-dependent control, while matching the performance of Tet-ON in both its OFF and ON states, showing comparable minimal and maximal expression (Fig. 1j, Supplementary Fig. 6h, i). Furthermore, its split-ATG architecture circumvents a key drawback of existing splicing-based systems, such as Xon¹⁷ and its derivatives (e.g., miniXon2G¹⁸) or DreAM²², which append peptide tags or P2A sequences that can trigger immunogenicity or result in aberrant protein maturation. Finally, through iterative sequence optimization, a compact switch (~640 bp) was generated that maintains low basal activity even under strong promoters while enabling high induction ratios (up to ~ 200-fold for the Max and Tight variants).

A comprehensive NNN-codon screen validates the system’s robustness by revealing the compatibility of the split-ATG system, with diverse N-terminal sequences at the start of the gene of interest (Supplementary Fig. 4). Mechanistically, this flexibility is consistent with the AG-scanning model of 3′ splice site selection⁶². The upstream polypyrimidine tract (PPT) of SMN2 intron 7 efficiently recruits the U2AF complex, establishing a dominant acceptor site definition that renders recognition largely independent of the immediate downstream exonic sequence^63,64. This strong upstream anchoring allows the spliceosome to tolerate deviations from the canonical +1 guanine, such as the thymine in our TG design, thereby facilitating reliable splicing across varying genetic contexts.

Beyond downstream sequence compatibility, we investigated the structural and transcriptional drivers of basal leakiness. Our mutagenesis studies demonstrate that the residual background stems from low-level basal inclusion of exon 7 rather than translation initiation at non-canonical sites within the switch architecture (Supplementary Fig. 5). Consistent with a splicing-based mechanism, basal output was promoter- and cell-type-dependent (Supplementary Fig. 9). For example, CMV drove substantially higher expression in HEK293T than in K562 cells, accompanied by higher basal signal, whereas fold induction did not scale proportionally across these contexts. Notably, the CAG promoter exhibited unexpectedly high basal leakiness relative to its overall output, but the underlying cause remains unclear (Supplementary Fig. 9). One plausible explanation is that promoter and 5′ UTR architecture modulate co-transcriptional spliceosome engagement and kinetics, thereby altering basal exon inclusion independently of bulk transcript abundance^65,66. Together, these observations suggest that promoter context can shape RisdiON performance beyond overall transcript output. While RisdiON Tight was engineered for minimal background, flow cytometry revealed residual basal activity that scaled with transgene copy number, as estimated by the mCherry normalization cassette (Supplementary Fig. 5, 6). Importantly, the impact of such leakiness is cargo-dependent: while highly potent effectors like Cas9 can induce permanent edits even at trace levels⁶⁷, lower-potency therapeutic proteins typically remain below the phenotypic threshold. Thus, the acceptable background level must be defined by the specific mechanism of the therapeutic cargo.

Another key differentiator of the RisdiON platform is its regulation by risdiplam, a clinically approved, orally bioavailable drug with a well-defined safety profile. This contrasts favorably with prior splicing-based switches that rely on branaplam^17,18, an investigational compound with a narrower tolerability window^21,68. Our comparative profiling demonstrated that risdiplam²³ maintains a significantly wider safety margin across diverse human cell lines (Supplementary Fig. 7). While highly proliferative populations, such as hiPSCs, exhibited dose- and time-dependent sensitivity, cell viability remains largely uncompromised at the concentrations required for RisdiON induction. This favorable safety profile is corroborated by published transcriptomic analyses evaluating the cellular response to both risdiplam and branaplam^21,68. While high concentrations of both drugs drive widespread splicing deregulation, treatment with low concentrations leads to very modest transcriptome changes in both cases^21,68. Crucially, previous branaplam-based switches rely on concentrations (up to 100 nM) that can trigger these widespread transcriptomic changes^17,18. In contrast, RisdiON operates efficiently within a window that limits off-target events. Prior off-target analyses of risdiplam identify a small set of secondary splice targets, with FOXM1 implicated primarily at high concentrations, offering a plausible mechanism for the dose-dependent proliferation sensitivity seen in fast-cycling cells such as iPSCs²³. Because this off-target engagement occurs primarily at extreme concentrations, operating within RisdiON’s optimal dynamic range effectively uncouples robust transgene induction from global transcriptomic disruption and cytotoxicity, offering a direct path for clinical translation.

These specific attributes of the RisdiON platform present compelling solutions for diverse gene and cell therapy applications. For instance, Cas-based genome editing requires transient activity and minimal OFF-state expression to prevent off-targets³⁸, DNA damage, and immunogenicity⁶⁹. Our demonstration of precise, tunable Cas9 expression in hiPSCs, achieving efficient gene editing with very low leakiness (Fig. 2c), directly addresses these requirements. Beyond isolated gene editing, RisdiON’s modularity significantly expands its utility for complex cellular engineering in iPSCs. Because many directed differentiation protocols already rely on doxycycline-inducible lineage drivers, deploying a second Tet-regulated transgene couples the secondary gene to the differentiation trigger and constrains independent tuning. As evidenced by our multiplexing experiments in differentiated neurons, RisdiON can be operated in parallel with doxycycline-driven differentiation and a second small-molecule system (vanillic acid) in the same neuronal cultures (Fig. 2f). Overall, RisdiON expands the practical control space in hiPSC differentiation by adding a compact, orthogonal small molecule channel that enables sustained and independently tunable regulation of secondary genes, reporters, or safety modules alongside doxycycline-based differentiation programs.

Furthermore, RisdiON promises value in cell therapies like CAR-T, a clinically approved treatment of several cancers⁷⁰. Current challenges of CAR-T therapy include CRS, immune effector cell-associated neurotoxicity syndrome (ICANS) and T cell exhaustion. These hurdles are tightly linked to the inability to precisely regulate CAR expression levels post-infusion due to the use of constitutive promoters for the CAR-T engineering^4,71. While innovative strategies like integrating CARs into endogenous loci (e.g., the TRAC locus) to harness physiological T cell activation dynamics have shown markedly better efficacy and reduced T cell exhaustion⁷², their regulation ties to T cell activation, lacking exogenous, titratable control. RisdiON offers a complementary path: its regulation of CAR expression in primary human T cells enables dose-dependent target cell killing while preserving a less differentiated T cell phenotype with lower inhibitory receptor expression compared to constitutive CARs (Fig. 3 and Supplementary Fig. 11). Our system supports dose-titration and reversibility, two properties highly desirable in managing the above-mentioned adverse events. Moreover, the oral bioavailability of risdiplam may enable non-invasive outpatient control, supporting paradigms like low-dose ramp-up, pause-on-toxicity, or on-demand recall in the event of tumor relapse⁷³. Thus, the RisdiON platform enhances key safety and efficacy determinants that are difficult to achieve with current genetic tools^4,71. While the functional validation of the RisdiON system is established in this study, its clinical translation will require rigorous in vivo assessment, in specialized models to fully demonstrate its clinical safety advantages over conventional CAR-T cells.

Finally, the compact, modular RisdiON switch offers a clinically deployable solution for post-delivery modulation of transgene expression in AAV-mediated gene therapy. We demonstrated robust in vivo AAV dose-dependent activation and full reversibility following oral risdiplam administration in mice (Fig. 4, Supplementary Fig. 12), predominantly in the liver, consistent with AAV9 tropism^74,75. Notably, the use of a destabilized reporter (Fluc-PEST) allowed for rapid decay of expression after drug withdrawal, reflecting tight temporal control (Fig. 4c, d). The measured five-hour fold induction reflects a short risdiplam pulse, and we hypothesize that later time points would show even higher induction levels. While this proof-of-concept study utilized a high dose to ensure robust signal detection during these narrow temporal windows, the exact protein level and duration of expression is influenced by the transgene product and its intrinsic protein half-life. Consequently, future disease-specific application studies will prioritize the evaluation of standard clinical-range dosing.

Importantly, the system demonstrated full, risdiplam-dependent reversibility, seamlessly transitioning between ON and OFF states (Fig. 4). This dynamic capability, regulated by a safe, orally deliverable drug, would be highly valuable for fine-tuning gene expression, halting toxicity, or on-demand therapeutic needs.

The post-transcriptional nature of RisdiON regulation renders it agnostic to promoter, capsid, and genome configuration, enabling integration with tissue-specific or self-complementary AAVs. This opens opportunities to apply inducible expression in previously inaccessible contexts such as retina, brain, and cardiac tissue, where control over transgene dosage and timing can substantially improve both safety and efficacy. In summary, this work introduces a clinically grounded, modular platform for small molecule–regulated gene expression. Crucially, the dynamic range of the RisdiON switch aligns with the plasma concentration range in the clinical studies³², i.e. up to maximal peak values of 485 nM (~ 194 ng/mL; mean Cmax). Our in vitro characterization demonstrates high sensitivity with effective induction at low nanomolar levels comparable to the clinical plasma concentration range, while in vivo validation confirms robust, reversible switching within the clinically safe maximal exposure range. By combining risdiplam’s pharmacological maturity as an approved drug with RisdiON’s tight regulation, compactness, and precise RNA-level splice control, this system paves the way for safer, more programmable gene switches as well as gene and cell-based therapies.

Methods

Ethical statement

The animal experiments were conducted at Nuvisan ICB (Berlin, Germany). All procedures were performed in accordance with the German Animal Welfare Act and were approved by the local ethics committee, Landesamt für Gesundheit und Soziales (LaGeSo) Berlin, under license number E0132/23. This study did not involve human participants. Research involving human-derived materials followed relevant institutional and national guidelines. The established hiPSC line (BIONi010-C-13; Cat# 66540561, ECACC) was originally derived with informed consent and ethical approval under Bioneer A/S protocols, verified by ECACC and EBiSC. Primary human T cells were isolated from healthy donor volunteer's blood samples obtained with informed consent from a blood donation center (Blutspende SRK, Zurich, Switzerland) with approval of the Cantonal Ethics Committee (Zurich). All personal identifiers were removed by the center, prior to experimentation to ensure donor confidentiality.

Cloning and plasmid design

All expression constructs were designed in Geneious Prime (Dotmatics, v2025.1.1) and synthesized by GenScript. For mammalian cell expression, the expression cassette is driven by the indicated promoters and transcription is terminated by a bovine growth hormone (bGH) polyadenylation (polyA) signal. The vector backbone also includes a separate expression cassette for a puromycin resistance gene (PuroR) under the control of a PGK promoter, enabling stable cell line selection. To minimize interference between the two expression cassettes, they are separated by an A2 insulator sequence⁷⁶. Detailed sequences of all constructs used in this study are provided in Supplementary Note 1.

Recombinant DNA/RNA for CAR production

Desired gene segments were synthesized by GeneArt (Thermo Fisher Scientific, Regensburg, Germany) from synthetic oligonucleotides and PCR products using automated gene synthesis. Inserts for T cell transduction were cloned by restriction/ligation into a second generation lentiviral expression system^77,78. The CAR transgene expression was controlled by a eukaryotic translation elongation factor 1 alpha (EF1α)-derived promoter and terminated by Woodchuck Hepatitis Virus Posttranscriptional Regulatory Element (WPRE).

Oligonucleotide production

All oligonucleotides (primers, ssDNA donors, and gRNAs) were synthesized and purified by Integrated DNA Technologies (IDT). Lyophilized oligonucleotides were resuspended in UltraPure™ DNase/RNase-Free Distilled Water (Cat# 10977015, Invitrogen) to a stock concentration of 100 μM. Detailed sequences for all oligonucleotides used in this study are provided in the corresponding Methods sections, as well as in Supplementary Table 1 of the Supplementary Information file.

Cell culture and cell lines

Human embryonic kidney 293 T (HEK293T) (Cat# CRL-3216, ATCC) and HEK293A (Cat# CLPZ02046, Roche) cells were maintained in DMEM High-Glucose (Cat# 11965092, Gibco) supplemented with 10% Fetal Bovine Serum (FBS; Cat# A3160402, Gibco) and 1% penicillin-streptomycin (10,000 U/mL) (Cat# 15140122, Gibco). Cells were passaged every 2-3 days to maintain exponential growth, using 0.05% Trypsin-EDTA (Cat# 25300054, Gibco) for detachment. Puromycin 1 µg/mL (Cat# A1113803, Gibco) or 15 µg/mL blasticidin (Cat# A11139-03, Gibco) was included in the medium to select for stably transfected or nucleofected cells exhibiting antibiotic resistance. U2OS (Cat#HTB-96, ATCC) were maintained in McCoy’s 5 A (Modified) medium (Cat# 36600021, Gibco) supplemented with 10% FBS (Cat# A3160402, Gibco) and 1% penicillin-streptomycin (10,000 U/mL) (Cat# 15140122, Gibco). Jurkat cells (Cat# ACC-282, DSMZ) were cultured in RPMI 1640 medium with GlutaMAX™ (Cat# 61870-036, Gibco) supplemented with 10% FBS (Cat# A3160402, Gibco) and 1% penicillin-streptomycin (Cat# 15140122, Gibco). Cells were maintained at 0.5–1.5 × 10⁶ cells/mL. For antibiotic resistance selection, 1 µg/mL of puromycin (Cat# A1113803, Gibco) was added to the media. K562 cells (Cat# CCL-243, ATCC) were maintained in RPMI 1640 medium with GlutaMAX™ (Cat# 61870-036, Gibco) supplemented with 10% FBS (Cat# A3160402, Gibco) and 1% penicillin-streptomycin (Cat# 15140122, Gibco). Cells were passaged every three days to maintain exponential growth. Selection was performed using 1 µg/mL puromycin (Cat# A1113803, Gibco). SUDHL4 (Cat# ACC 495, DSMZ), NALM-6 (Cat# ACC 128, DSMZ), Raji (Cat# CCL-86, ATCC) and RPMI-8226 (Cat# CCL-155, ATCC) were maintained in complete RPMI 1640 medium (Cat# 11875093, Gibco), supplemented with 1x GlutaMAX (Cat# 35050061, Gibco) and 10% FBS (Cat# F4135-500 ML, Sigma-Aldrich). All cells were grown in a 5% CO₂, 95% air-humidified incubator at 37 °C. All commercial cell lines were authenticated by their respective vendors. All used cell lines were routinely tested for mycoplasma contamination and found to be negative. Inducible expression was controlled using risdiplam (RO7034067, Roche), doxycycline (Cat# D5207, Sigma), vanillic acid (Cat# 94770, Sigma) or branaplam hydrochloride (Cat# HY-19620A, MedChemExpress). Compounds were added using a D300e Digital Dispenser (Tecan).

Human induced pluripotent stem cell (hiPSC) culture and maintenance

Human pluripotent stem cells (hiPSCs) with doxycycline (dox)-inducible NGN2 expression (BIONi010-C-13; Cat# 66540561, ECACC) were maintained under feeder-free conditions. Culture vessels were coated with Biolaminin 521 LN (Cat# LN521-05, Biolamina), diluted to 12.5 µg/mL in PBS containing Ca²⁺ and Mg²⁺ (PBS + /+; Cat# 14040133, Gibco) for at least 4 h at 37 °C or overnight at 4 °C. Before seeding, the coating solution was aspirated. Cells were cultured in mTeSR™ Plus medium (Cat# 100-0276, STEMCELL Technologies). For routine passaging, hiPSCs at approximately 80% confluency were dissociated into single cells using Accutase (Cat# AT-104, Innovative Cell Technologies) for approximately 3–5 min at 37 °C. Dissociation was stopped by adding mTeSR™ Plus medium. Cells were typically seeded at a density of 20,000 cells/cm² onto Laminin 521-coated vessels in mTeSR™ Plus medium, supplemented with fresh 10 µM ROCK inhibitor (Cat# Y-27632, STEMCELL Technologies). After 24 hours, the medium was exchanged to mTeSR™ Plus medium without ROCK inhibitor. For thawing, cryopreserved cells were rapidly warmed, resuspended in mTeSR™ Plus medium with 10 µM ROCK inhibitor, centrifuged (200 x g, 5 min), and seeded at 25,000–30,000 cells/cm². For imaging and FACS experiments, cells were split and seeded at 20,000 cells/cm² in Biolaminin 521 LN pre-coated 96-well plates (#3904, Corning). After 24 h, cells were treated with risdiplam and analyzed 48 h later. Cell counts were performed using 0.4% Trypan Blue (Cat# 15250061, Gibco).

Neuronal differentiation of hiPSCs

For initiation of neuronal differentiation, hiPSCs maintained as above were dissociated using Accutase (~ 10 min, 37 °C), counted, and seeded at a density of 20,000 cells/cm² onto Biolaminin 521-coated 96-well plates in mTeSR™ Plus medium, supplemented with 10 µM ROCK inhibitor (Day −1). On the following day (Day 0), the medium was completely replaced with Neuronal Induction medium (N2 medium), consisting of DMEM/F12 medium (Cat# 31331028, Thermo Fisher Scientific) with N-2 supplement (Cat# 17502048, Thermo Fisher Scientific), supplemented freshly with 2 µg/mL doxycycline (Cat# D5207, Sigma-Aldrich). On Day 2, the media was replaced with Neuronal medium (NBB27), consisting of Neurobasal medium (Cat# 21103049, Gibco), supplemented with 1x B-27™ supplement (Cat# 17504044, Gibco), 1x GlutaMAX™ (Cat# 35050061, Gibco), 2 µg/mL doxycycline, 10 ng/mL Brain-Derived Neurotrophic Factor (BDNF; Cat# AF-450-02, PeproTech), and 1:100 Laminin (Cat# 11243217001, Roche) (final concentrations after plating). Full media changes were performed using NBB27 medium supplemented with 10 ng/mL BDNF and varying additives: 2 µg/mL doxycycline and 10 µM DAPT (Cat# 100-1046, STEMCELL Technologies) on day 3 (200 µL/well); 2 µg/mL doxycycline, 10 µM DAPT, and 1:500 Roche Laminin on day 6 (100 µL/well); 2 µg/mL doxycycline and 1:500 Roche Laminin on day 7 (100 µL/well); and 1:500 Roche Laminin on day 8 (200 µL/well). From Day 10 onwards, half-media changes (100 µL/well) were performed every 2-3 days using NBB27 medium, supplemented with 10 ng/mL BDNF and 1:500 Roche Laminin.

Stable cell line generation using PiggyBac

Stable cell lines were generated using the PiggyBac transposon system⁷⁹ to achieve stable genomic integration of the expression cassettes. The desired RisdiON expression plasmid was transfected with a helper plasmid encoding the PiggyBac (PBO) transposase at a 3:1 ratio using Lipofectamine 2000 (Cat# 11668019, Invitrogen). The RisdiON expression vectors were designed to include a puromycin resistance gene (PuroR) driven by a PGK promoter, enabling the generation of stable pools.

Risdiplam solution preparation

Risdiplam powder (RO7034067, Roche) was stored at −20 °C. For in vitro experimental use, a 200 µM stock solution was prepared by dissolving the powder in 100% DMSO. Complete dissolution was facilitated by incubation at 37 °C for approximately 1 h with regular vortexing. Stock solutions were aliquoted and stored at −80 °C for long-term use and at −20 °C for short-term storage use.

For in vivo studies, risdiplam was prepared as a suspension in a vehicle containing 10 mM ascorbic acid and 0.01 mg/mL sodium thiosulfate pentahydrate in reverse osmosis water, adjusted to a pH of 3.0 ± 0.1. This formulation was administered to animals via oral gavage to deliver a final dose of 15 mg/kg.

Imaging

Risdiplam-induced GFP expression was assessed using two imaging systems. HEK293T cells were imaged using an Incucyte S3 Live-Cell Analysis System (Sartorius). Bright field and GFP images (300 ms exposure) were acquired using a 10x or 20x objective lens every six hours for a total duration of 48 hours. Imaging for hiPSCs and differentiated neurons was performed using an Operetta High-Content Imaging System (PerkinElmer). For this, cells were cultured in 96-well PhenoPlates (Cat# 6057302, Revvity), and images were captured using a 20x air objective lens.

Nucleofection

Nucleofection of HEK293T, hiPSCs, and Jurkat cells was performed using 4D-Nucleofector® X Unit (Cat# AAF-1003X, Lonza) according to the manufacturer’s protocol. For HEK293T cells nucleofection with CD81-targeting sgRNAs, 1 × 10⁶ HEK293T cells were collected, washed with DPBS and centrifuged. The cell pellet was resuspended in 100 µL of SF Cell Line Nucleofector^TM Solution (Cat# V4XC-2032, Lonza) containing three CD81 sgRNAs (CD81_sgRNA1: UUGGCUUCCUGGGCUGCUA, CD81_sgRNA2: GCAGCCCUCCACUCCCAUGG, and CD81_sgRNA3: GGCGCUGUCAUGAUGUUCGU), at a concentration of 150 nM each. Resuspended cells were transferred to Nucleocuvette^TM Vessels and nucleoporated in 4D-Nucleofector® X Unit using CM-130 program. Subsequently, 400 µL of fresh media was added to the vessels and cells were incubated for 5 min at room temperature before being seeded into 96-well plates and treated with risdiplam. For hiPSCs, 1 × 10⁶ cells per reaction were collected, washed once with Ca²⁺/Mg²⁺-free DPBS (Cat# 14190144, Gibco), centrifuged and resuspended in 100 µL of P3 Primary Cell 4D-Nucleofector Kit L (Cat# V4XP-3012, Lonza). Resuspended cells were mixed with 2 µg of donor plasmid and 0.65 µg of PiggyBac-encoding plasmid and transferred to a Nucleocuvette^TM Vessel (Lonza). Cells were nucleoporated using the CB-150 program. Subsequently, 400 µL of culture media (mTeSR™ Plus) was added to the vessels and cells were incubated for 5 min at room temperature. Next, the entire volume was transferred to 12-well plates pre-coated with Biolaminin 521 and containing 0.5 mL of fresh mTeSR™ Plus medium. For Jurkat cells nucleofection, 1 × 10⁶ cells per construct were collected, washed once with Ca²⁺/Mg²⁺-free DPBS, centrifuged and resuspended in 100 µL of SE Cell Line Nucleofector^TM Solution (Cat# V4XC-1024, Lonza). Resuspended cells were mixed with 2 µg of splice-switch plasmid and 0.65 µg of PiggyBac-encoding plasmid and transferred to a Nucleocuvette^TM Vessel (Lonza). Cells were nucleoporated using the CL-120 program. Subsequently, 400 µL of culture media (RPMI 1640) was added to the vessels and cells were incubated for 5 min at room temperature. Next, the entire volume was transferred to 12-well plates containing 0.5 mL of fresh media. Cells were incubated for 48 h, followed by puromycin selection.

Arrayed screen for RisdiON optimization

To systematically optimize the RisdiON switch, two separate arrayed screens were performed targeting the exonic and intronic sequences of the SMN2-derived cassette. The first screen targeted SMN2 exon 7, utilizing a library of RisdiON Gen3 constructs with 1- to 5-nucleotide overlapping deletions that targeted nucleotides 3 through 50 of the SMN2 exon 7 sequence. A second screen probed the regulatory roles of intronic sequences using a library built on the RisdiON Max backbone, containing systematic, 5-nucleotide non-overlapping deletions across both intron 6 and intron 7. Both screens were conducted in K562 cells. Cells were transfected with individual constructs from the deletion library and RisdiON Gen3/RisdiON Max as a positive control. Transfection was performed in 96-well plates by seeding 20,000 cells per well. Stable integration was achieved using a PiggyBac (PBO) transposon system with Lipofectamine 2000 (Cat# 11668019, Invitrogen) as the transfection reagent. Following transfection, cells underwent a 14-day selection with puromycin at 1 µg/mL to establish stable pools, which were then expanded for analysis. The resulting pools were treated for 48 h with 250 nM risdiplam or a DMSO vehicle control, with compounds administered by a D300e Digital Dispenser (Tecan). GFP expression was monitored using an Incucyte S3 Live-Cell Analysis System and subsequently quantified by flow cytometry to assess performance based on basal expression (leakiness) and induction ratio using CytoFLEX LX Flow Cytometer (BeckmanCoulter). FACS data were analyzed using FlowJo (BD, v.10.10.0). Results were normalized to the RisdiON Gen3/RisdiON Max constructs.

HDR screen for RisdiON variant optimization

Reporter Cell Line Generation and Characterization. A pooled CRISPR-Cas9 Homology Directed Repair (HDR) screen was conducted using HEK293T cells stably expressing the RisdiON pMM490 GFP reporter. RisdiON pMM490, a derivative of RisdiON Gen2, features a shortened exon 6, deleted exon 8, and a PGK-puromycin cassette replaced with an IRES-puromycin element downstream of the GFP (Supplementary Note 1). The pMM490 vector was introduced into HEK293T cells via PiggyBac transposition. To minimize variability from reporter copy number, cells with low GFP expression were subsequently isolated by fluorescence-activated cell sorting (FACS), establishing the HEK293-pMM490-Low reporter line. This line was estimated to contain 1-2 copies of the reporter, which helps prevent the masking of mutational effects on splicing.

HDR Library Design and Transfection. For the HDR-based pooled screen, mutations were introduced into the SMN2 reporter sequence using oligonucleotide donors (ODNs) synthesized with degenerate sequences. These ODN templates (100 µM stock) were designed to target five regions within the SMN2 reporter: the 3′ splice site (3′ss): (ssDNA-3′SS, AAACAAAATGCTTTTTAACATCCATATAAAGCTATCTATATATAGCTATCTATATCTATATAGCTATTTTTTTTAACTTCCTTTATTTTCCTTCCAGGANNNNNNNNAAAATCAAAAAGAAGGAAGGTGCTCACATTCCTTAAATATAAGGAGTAAGTCTGCCAGCATTATGAAAGTGAATCTTACTTTTGTAAAACTTT); exon 7: (ssDNA-Exon7, TATATATAGCTATCTATATCTATATAGCTATTTTTTTTAACTTCCTTTATTTTCCTTCCAGGATTTTAGACAAAATCAAAAAGAAGGAAGGTGCTCNNNNNNNNNNAATATAAGGAGTAAGTCTGCCAGCATTATGAAAGTGAATCTTACTTTTGTAAAACTTTATGGTTTGTGGAAAACAAATGTTTTTGAACATTTAA), an exonic splicing enhancer (ESE), (ssDNA-ESE2, ATCCATATAAAGCTATCTATATATAGCTATCTATATCTATATAGCTATTTTTTTTAACTTCCTTTATTTTCCTTCCAGGATTTTAGACAAAATCAAAAAGAANNNNNNNNNNNACATTCCTTAAATATAAGGAGTAAGTCTGCCAGCATTATGAAAGTGAATCTTACTTTTGTAAAACTTTATGGTTTGTGGAAAACAAA), and two distinct 5′ splice site (5′ss) regions (designated −1 and −2), (ssDNA-5′SS-1,AGCTATCTATATCTATATAGCTATTTTTTTTAACTTCCTTTATTTTCCTTCCAGGATTTTAGACAAAATCAAAAAGAAGGAAGGTGCTCACATTCCTTANNNNNNNNGAGTAAGTCTGCCAGCATTATGAAAGTGAATCTTACTTTTGTAAAACTTTATGGTTTGTGGAAAACAAATGTTTTTGAACATTTAAAAAGTTC), (ssDNA-5′SS-2, AGCTATCTATATCTATATAGCTATTTTTTTTAACTTCCTTTATTTTCCTTCCAGGATTTTAGACAAAATCAAAAAGAAGGAAGGTGCTCACATTCCTTAAATATAAGGAGTNNNNNNNNNAGCATTATGAAAGTGAATCTTACTTTTGTAAAACTTTATGGTTTGTGGAAAACAAATGTTTTTGAACATTTAAAAAGTTC).

Four single guide RNAs (sgRNAs) targeting the RisdiON reporter were initially designed and functionally validated in the HEK293-pMM490-Low cells. HDR_sgRNA1 (CGUAAUACUUUCACUUAGAA) and HDR_sgRNA2 (AUUUAUAUUCCUCAUUCAGA) were selected for the pooled library generation due to their specific targeting of the reporter sequence without cleaving the endogenous SMN2 gene. sgRNA1 was paired with the ODN targeting the 3′ss. sgRNA2 was used with ODNs targeting exon 7, the ESE, and both 5′ss regions, with pairings based on the proximity of the sgRNA cut site to the intended ODN integration site. For delivery of the HDR templates, 1.2 × 10⁶ HEK293-pMM490-Low cells were nucleofected per reaction. Each nucleofection contained 360 pmol of the appropriate sgRNA, 4 µL of Cas9 protein, and 2 µL of the 100 µM ODN donor. Nucleofections were performed using a 4D-Nucleofector™ system (Lonza) with SF Cell Line Nucleofector™ Solution (Cat# V4XC-2032, Lonza) in Nucleocuvette^TM Vessel (Lonza), employing program CM-130.

Cell Sorting by FACS and DNA extraction. Three days post-nucleofection, the HEK293-pMM490-Low cell pools were subjected to FACS. Cells were first sorted into two populations: GFPconstitutive (cells exhibiting constitutive GFP expression, irrespective of risdiplam treatment) and GFP low population (cells with low basal GFP expression). The GFP low population (1.2 × 10⁶ cells collected) was cultured further and subsequently treated with 150 nM risdiplam for 48 h to assess splicing induction. Following risdiplam treatment, these cells were sorted by FACS into three fractions based on GFP intensity: no response (1.5 × 10⁶ cells), medium response (GFP+ medium, lower than the pMM490 positive control, 0.5 × 10⁶ cells), and strong response (GFP+ high, comparable with the pMM490 positive control, 0.8 × 10⁶ cells). Parental HEK293-pMM490-Low cells treated with risdiplam under identical conditions served as a positive control for induction. All sorted cell populations were expanded to over 2 × 10⁶ cells before genomic DNA (gDNA) extraction. Genomic DNA was extracted from the expanded sorted cell populations using the DNeasy Blood & Tissue Kit (Cat# 69504, Qiagen). Following gDNA extraction, 50 µL of DNA from the HEK293T-sorted populations (at concentrations of 200–1000 ng/µL) was sent to Microsynth for sequencing. The concentration of double-stranded DNA was quantified using Lumiprobe Pico488, and DNA integrity was assessed using a Fragment Analyzer or QIAxcel Fragment Analyzer.

Library Preparation and NGS sequencing. For sequencing library preparation, 3.2 µg of gDNA per sample was used as the template for a customized first-step PCR. To maintain library complexity, 32 individual PCR reactions were performed per sample (20 µL volume containing 5 µL of template). The mutated region of the SMN2-GFP reporter was specifically targeted using forward primer MM61 (5′-AGCTTTCATGCTATCTTAACTGC-3′) and reverse primer MM50 (5′-ACCTTTCAACTTTCTAACATCTG-3′). This primer pair incorporated partial Illumina adapter sequences (tails) and was designed to yield a ~ 400 bp amplicon while minimizing amplification of the larger corresponding region from the endogenous SMN2 locus. The PCR products from these 32 reactions were then pooled for each sample. The pooled product was purified and subsequently used as a template for a single, second-step PCR per sample to complete the library construction by adding full Illumina adapters and indices. Final libraries were sequenced on an Illumina NovaSeq platform using an SP flow cell with 2×250 bp paired-end reads.

Bioinformatic analysis. DNA library construction, sequencing, and primary data analysis were performed by Microsynth AG (Balgach, Switzerland). The produced paired-end reads, which passed Illumina’s chastity filter were subject to de-multiplexing and trimming of Illumina adaptor residuals and primer sequences using Illumina’s bcl2fastq software version, v2.20.0.422. The quality of the reads was assessed using the software FastQC version 0.11.8 and sequencing reads below an average Q-score of 24 or with any uncalled bases (N) were excluded from further analysis. Surviving paired-end reads were merged to reconstruct full-length amplicon sequences in silico considering a minimum overlap of 15 bases using the software USEARCH, version 11.0.667. Merged reads that contained ambiguous bases or were shorter than 100 bp were also discarded. Subsequently, the resulting merged reads for each sample were divided into three equally sized, non-overlapping, sequential portions to generate technical triplicates (designated a, b, and c). This strategy was employed to create replicate datasets, facilitating statistical evaluation of variant distributions. A comparative dereplication analysis was performed on these technical triplicates; the merged reads in each of the newly formed samples were dereplicated (clustering with a 100% match in sequence length and identity) to detect unique sequence patterns and their frequency using the software USEARCH, version 11.0.667. For the subsequent differential abundance analysis of the detected sequence variants, a minimum valid cluster size of 1500 reads was applied. Normalization of the raw counts and differential abundance analysis were carried out using the R software package DESeq2, version 1.44.0. This comprehensive analysis pipeline enabled the precise identification and quantification of genetic variations within the SMN2 reporter across the different sorted cell populations, allowing for the assessment of their impact on splicing regulation. Variants enriched in the GFP high inducible population relative to the GFPconstitutive population, such as Var-392, were selected for subsequent validation through individual cloning and functional characterization against parent constructs.

Arrayed screen for first amino acid codon influence on RisdiON performance

To systematically investigate the influence of the first amino acid codon on the performance of the splice switch, an arrayed screen was performed utilizing a library of pMM955 constructs. The library was built on the RisdiON Max backbone, where the codon immediately downstream of the splice-generated TG site was randomized (NNN). To ensure strict reliance on the splice switch for translation initiation, the canonical start codon (ATG) and stop codons (TAA, TAG, TGA) were excluded from the randomized position.

The screen was conducted in K562 cells. Transfection was performed in 96-well plates. Stable integration was achieved using a PiggyBac (PBO) transposon system with Lipofectamine 2000 (Cat# 11668019, Invitrogen) as the transfection reagent. Following transfection, cells underwent a 14-day selection with puromycin to establish stable pools. The resulting pools were treated for 48 hours with 200 nM risdiplam or a DMSO vehicle control, with compounds administered by a D300e Digital Dispenser (Tecan). Reporter expression was quantified by flow cytometry using a CytoFLEX LX Flow Cytometer (BeckmanCoulter). Flow cytometry data was analyzed using FlowJo (BD). To assess performance, GFP expression was normalized to the constitutive mCherry marker, and results were analyzed as log2 fold-change relative to the library mean.

CD81 staining

HEK293T cells, cultured in 96-well plates were first detached with 40 µL of Trypsin-EDTA (0.05%), phenol red (Cat# 25300054, Gibco) and then resuspended in an additional volume of 100 µL DPBS to a total volume of 140 µL. Cells were transferred to fresh 96-well microplates, PP, U-bottom, 0.3 mL (Cat# 701330, Brand), centrifuged and washed once with DPBS. Next, cells were resuspended in 50 µL of 1:10 mix of APC-conjugated CD81 (1:10, APC, clone JS-81, Cat# 561958, BD) and DPBS and incubated for 15 min at 4 °C. After a centrifugation step (180 x g, 3 min), the staining solution was removed and cells were washed three times with DPBS. Finally, cells were resuspended in DPBS and analyzed by FACS using CytoFLEX LX Flow Cytometer (BeckmanCoulter).

Flow cytometry analysis

HEK293T and Jurkat cells were analyzed by flow cytometry. HEK293T were detached by adding 40 µL of 0.05% Trypsin-EDTA and further resuspended in 100 µL DPBS. A volume of 120 µL was then transferred to a new 96-well U-bottom plate (Cat# 701330, Brand), centrifuged and resuspended in 120 µL autoMACS® Running Buffer (Cat# 130-091-221, Miltenyi Biotec). Finally, CytoFLEX LX Flow Cytometer (Beckman Coulter) was used for the analysis. For Jurkat cells, the FACS analysis was performed as described above, excluding the initial detachment with trypsin. To calculate the percentage (%) of GFP-positive (GFP+ ) cells, % of GFP+ cells in DMSO was normalized to the % of GFP+ cells at the maximal risdiplam concentration (max induction). This normalization was performed to exclude cells that do not express the RisdiON system and to prevent a potential underestimation of induction levels.

Luciferase reporter assays

NanoLuc luciferase (NLuc) activity was quantified using the Nano-Glo® Luciferase Assay System (Cat# N1120, Promega) following the manufacturer’s instructions. Briefly, the Nano-Glo® substrate was prepared by combining one volume of Nano-Glo® Luciferase Assay Substrate with 50 volumes of Nano-Glo® Luciferase Assay Buffer at room temperature. Firefly luciferase (Fluc) activity was measured using the ONE-Glo™ Luciferase Assay System (Cat# E8120, Promega), following a similar procedure. The ONE-Glo™ reagent was prepared by dissolving the ONE-Glo™ Substrate in ONE-Glo™ Buffer at room temperature. For both assays, a volume of reagent equal to the culture medium (100 µL per well in a 96-well plate) was added to the cells. After a 3-minute incubation at room temperature, luminescence was measured using a PHERAstar FSX Microplate Reader (BMG Labtech), with gain settings adjusted based on signal strength.

Cell viability assay

Cell viability was assessed using the CellTiter-Glo® Luminescent Cell Viability Assay (Cat# G7570, Promega), which quantifies ATP as an indicator of metabolically active cells. HEK293A, HEK293T, Jurkat, K562, U2OS, and hiPSCs were seeded in 96-well Clear Bottom White Polystyrene TC-treated Microplates (Cat# 3610, Corning) at appropriate densities to maintain log-phase growth during the treatment window. After 24 h, cells were treated with increasing concentrations of risdiplam (RO7034067, Roche), branaplam hydrochloride (Cat# HY-19620A, MedChemExpress), doxycycline (Cat# D5207, Sigma), or bortezomib (Cat# HY-10227, MedChemExpress) using a D300e Digital Dispenser (Tecan). The maximum concentrations tested were 10,000 nM for bortezomib, 20,000 nM for doxycycline, and 2000 nM for risdiplam and branaplam. DMSO-treated cells served as the 100% viability control, while media-only wells served as the background (0% viability) control. Plates were incubated for 24, 48, or 72 h at 37 °C in a 5% CO₂ incubator. Following incubation, a volume of CellTiter-Glo® reagent equal to the volume of cell culture medium present in each well was added. The plates were placed on an orbital shaker for 2 minutes to induce cell lysis, followed by a 10-minute incubation at room temperature to stabilize the luminescent signal. Luminescence was recorded using a PHERAstar FSX Microplate Reader (BMG Labtech). Raw luminescence values were normalized to the DMSO control as 100% viability and media control as 0% viability to calculate the percentage of viable cells. IC50 values were determined using non-linear regression analysis.

NFAT-luciferase reporter assay

Engineered Jurkat NFAT-luciferase reporter T cells expressing either RisdiON Max-CD19 CAR, RisdiON Max-BCMA CAR, or RisdiON Max-GFP were prepared for co-culture. Prior to co-culture, effector cells were pre-incubated for 48 h with the specified concentrations of risdiplam (0-400 nM) or DMSO vehicle control to induce CAR or GFP expression. Target cells (SUDHL4 or NALM6) were harvested, centrifuged, and resuspended in fresh RPMI-1640 complete medium at a final concentration of 1.5 × 10⁶ cells/mL. For the assay, 20 µL of the target cell suspension (30,000 cells) were seeded per well of a 96-well Clear Flat Bottom Polystyrene TC-treated Microplate (Cat# 3595, Corning). To maintain induction, pre-treated effector Jurkat T cells were resuspended in RPMI complete medium supplemented with the corresponding concentration of risdiplam or DMSO. These effector cells were then added to the target cells (10,000 effectors in 80 µL) to achieve an Effector:Target (E:T) ratio of 1:3. Control wells (CTRL) received effector cells (pre-treated with the highest concentration of risdiplam or DMSO as appropriate) and media without target cells. The final co-culture volume per well was 100 µL. The co-culture plates were incubated for 4 to 6 hours at 37 °C in a 5% CO₂ incubator. Following incubation, 100 µL of ONE-Glo™ Luciferase Assay Reagent (Cat# E8120, Promega) was added to each well according to the manufacturer’s instructions. The content of each well was mixed gently, and the entire volume was transferred to a white, opaque-walled 96-well assay plate (Cat# 3917, Corning). Luminescence was measured immediately using the PHERAstar FSX Microplate Reader, with equal gain for all replicates. Data were recorded as relative light units (RLU).

Kinetics measurement

The inducibility (ON/OFF switch) of the constructs was evaluated in HEK293T cells by monitoring NanoLuc luciferase (NLuc) expression. For the ON-switch assay, cells were seeded in a 96-well plate and treated with 200 nM risdiplam one hour post-seeding. The treatment was repeated at sequential time points (e.g., 2, 4, 6, and 8 h). For the OFF-switch assay, the previous procedure was reversed. Cells were treated with 200 nM risdiplam for 16 h, after which the culture medium was replaced with risdiplam-free medium at time points indicated in the figure. For both assays, luminescence was assessed using the Nano-Glo® Luciferase Assay System (Cat# N1120, Promega).

RNA isolation

HEK293T cells with stably incorporated splice switch constructs were seeded on 6-well plates (0.5 × 10⁶ cells/well) and treated with DMSO or risdiplam for 24 h. After 24 hours, cells were washed with DPBS (-/-), trypsinised, resuspended in DMEM complete media and transferred to 1.5 mL Eppendorf tubes. Next, cells were centrifuged at 300 x g for 5 min at room temperature, washed once with PBS (-/-) and cell pellets were frozen at −80 °C. RNA was isolated using RNeasy Mini Kit (Cat# 74104, Qiagen) according to the manufacturer’s instructions. After the isolation RNA concentration was measured and all samples were diluted with UltraPure™ DNase/RNase-Free distilled water (Cat# 10977015, Invitrogen) to the same concentration. The isolated RNA was stored at −80 °C and subsequently used for RT-PCR.

cDNA synthesis

Two µg of the isolated total RNA were used for the cDNA synthesis using SuperScript IV VILO (Cat# 18090050, Invitrogen) according to the manufacturer’s protocol. In short, RNA was either mixed with SuperScript VILO reverse transcriptase (RT) (+RT) or without RT (–RT), 4x master mix and water in 0.2 mL 8-Strip Non-Flex PCR tubes (Cat# I1402-3700, Starlab) to the final volume of 20 µL. Tubes were incubated at 25 °C for 10 min, followed by 50 °C for 30 min and 80 °C for 5 min. Samples were diluted to 100 µL with UltraPure™ DNase/RNase-Free distilled water (Cat# 10977015, Invitrogen) to the final cDNA concentration of 20 ng/µL and stored at −20 °C.

PCR

Synthesized cDNA was used for PCR reaction to analyze splicing efficiency of the tested constructs. PCR reaction was set up using 10 µL of Q5 Hot Start HF 2x Master Mix (Cat# M0494L, NEB), 1 µl forward primer MM16 (5′-TTGACCTCCATAGAAGACACC-3′) and 1 µL reverse primer MM40 (5′-GCAGATGAACTTCAGGGTCAG-3′) to the final concentration of 0.5 µM each, 6 µl H₂O and 2 µL cDNA to the total volume of 20 µL in 0.2 mL 8-Strip Non-Flex PCR tubes. PCR reaction conditions were as follows: 98 °C for 1 min, 15 cycles of 98 °C for 15 s, 64 °C for 15 s, and 72 °C for 30 s. The protocol concluded with 72 °C for 2 min and a hold at 4 °C. A limited cycle number was used to ensure amplification remained in the linear range for semi-quantitative analysis. Generated PCR products were resolved for 20 min on 2% E-Gel 48-well Agarose Gels (Cat# G820802, Invitrogen) with E-Gel 50 bp DNA Ladder (Cat# 10488099, Invitrogen) and imaged using E-Gel™ Power Snap Electrophoresis System (G8300). Densitometric analysis was performed using Fiji (ImageJ) software, version v1.54. The integrated density of the bands corresponding to exon inclusion (upper band) and exclusion (lower band) was quantified. Background signal was subtracted from each measurement using a region of interest (ROI) of identical size in a designated background area. Splicing efficiency (percent spliced-in, PSI) was calculated as the ratio of the inclusion band intensity to the total intensity of both bands: [Intensity inclusion / (Intensity inclusion + Intensity exclusion)] × 100.

Animal study

Study design. Mice were group-housed (5 animals per cage, single-sex groups) in a 14/10 hour light/dark cycle (lights on at 06:00) in a temperature and humidity-controlled room (22 °C ± 2 °C; 55% ± 10% humidity) with free access to water and food in the AAALAC-accredited animal facility Nuvisan ICB GmbH under barrier conditions. All animals were sourced at 7 weeks of age from the certified vendor Janvier Labs (22 Rte des Chênes Secs, 53940 Le Genest-Saint-Isle, France) and were of the inbred strain BALB/cAnNRj. The animal experiment was approved by the Berlin State Authority (Landesamt fuer Gesundheit und Soziales, license number E0132/23) and was internally reviewed and approved by the Nuvisan ICB Animal Welfare Committee in compliance with AAALAC accreditation guidelines. The animals were assigned to the groups randomly.

AAV dose selection study. BALB/c male mice, aged 8–12 weeks, were enrolled in the study. The mice were randomly divided into the experimental groups, with each group consisting of five mice for the preliminary AAV dose finding study. The study aimed to assess the bioluminescent signal following AAV9 injections and subsequent risdiplam small molecule treatment to determine the required AAV dose. Mice were injected intravenously (i.v.) with either phosphate-buffered saline (PBS, negative control group) or AAV9 vectors. Three weeks post-AAV injection, all groups were subjected to bioluminescence imaging. Prior to imaging, mice were injected intraperitoneally (i.p.) with D-luciferin (150 mg/kg body weight). Bioluminescence imaging was performed using the NightOWL imaging system (Berthold Technologies). Risdiplam was administered to the respective groups via oral gavage (i.g.) four hours before D-luciferin administration and imaging.

ON-OFF dynamics study. BALB/c female mice, aged 8–12 weeks, were enrolled in the study. The mice were randomly divided into the experimental groups, with each group consisting of ten or eight mice. The study aimed to assess the bioluminescent signal following AAV9 injections and subchronic administration of risdiplam small molecule to determine the reversibility of the RisdiON system. Mice were injected intravenously (i.v.) with either phosphate-buffered saline (PBS, negative control group) or AAV9 vectors. Three weeks post-AAV injection, the experimental timeline of the subchronic injections started with an immediate imaging of the bioluminescent signal to determine background (day 0 = OFF 1 state). Prior to imaging, mice were injected intraperitoneally (i.p.) with D-luciferin (150 mg/kg body weight). Bioluminescence imaging was performed using the Vilber Newton system (Berthold Technologies). Risdiplam was administered at the respective time lines (day 1 = ON 1, day 5 = ON 2) via oral gavage (i.g.) five hours before D-luciferin administration and imaging.

AAV production protocol

HEK293T cells were cultured in DMEM supplemented with 10% FBS + 2 mM L-Glutamine until they reached 50–70% confluency. The cells were then transfected with pHelper, pAAV ITR-expression, and pAAV Rep-Cap vectors (triple-transfection method) using PEI reagent in Nunc™ Cell Factory™ system (Thermo Fisher Scientific). 72 hours post-transfection, AAV particles were harvested from the cells. The crude lysate was prepared by lysing the cell pellet with 0.1% Triton and digesting with Benzonase. Primary capture of AAV particles was performed using Poros CaptureSelect AAV-X (Thermo Fisher Scientific). A secondary purification step was conducted using iodixanol gradient centrifugation to enrich for full capsids, followed by formulation in PBS supplemented with 0.001% Pluronic.

For quality control, vector genome titer was determined by performing qPCR with ITR-specific primers (ITR_qPCR_F: 5′-CGGCCTCAGTGAGCGA-3′, and ITR_qPCR_R: 5′-ggaacccctagtgatggagtt-3′). The purity of the final product was assessed via 10% SDS-PAGE using Stain-Free Imaging Technology (Bio-Rad). A purity level of more than 95% was confirmed if only VP1, VP2, and VP3 proteins were detectable. Endotoxin levels were measured using the Endosafe-nexgen-PTS Spectrometer from Charles River Laboratories, with all constructs showing endotoxin levels below the detection limit.

Quantification of viral genome copies by droplet digital PCR

Genomic DNA extraction from tissue. Frozen tissue samples were pulverized in liquid nitrogen using a Covaris dry cryo-pulverization system. Genomic DNA was extracted from approximately 25 mg of pulverized tissue (resuspended in 50 µL of PBS) using the Macherey-Nagel NucleoSpin Tissue Kit (#740952) according to the manufacturer’s instructions. Lysis was performed at 56 °C (2 h at 300 rpm, then 30 min at 1000 rpm). Following purification, DNA was eluted with 100 µL of Buffer BE. DNA concentration and purity were quantified via UV absorbance (Tecan Infinite 200 microplate reader, i-control, version v2.0.10.0).

Droplet digital PCR (ddPCR) for WPRE detection. WPRE (Woodchuck Hepatitis Virus Post-transcriptional Regulatory Element) quantification was performed on a Bio-Rad QX200 ddPCR system. The assay used a proprietary 20x FAM-labeled primer-probe mix (Cat# dEXD39054642, Bio-Rad) targeting a 76 bp region of WPRE (GenBank FJ797421.1). Normalization was performed using a murine genomic reference assay (Cat# dMmuCNS671723376, Bio-Rad). Each 20 µL reaction contained 10 µL of 2x ddPCR Supermix for Probes (No dUTP), 1 µL of the 20x WPRE assay, 3 U of MspI (for intracartridge digestion), and 50 ng of template DNA. Droplets were generated with the Bio-Rad QX200 Droplet Generator, and PCR amplification was performed on a Bio-Rad C1000 Touch Thermal Cycler (105 °C heated lid, 2 °C/sec ramp). Cycling conditions: 95 °C for 10 min (enzyme activation); followed by 40 cycles of 94 °C for 30 s (denaturation) and 55 °C for 1 min (annealing/extension). The protocol concluded with enzyme deactivation at 98 °C for 10 min and a hold at 4 °C.

Measurement and data analysis. Post-amplification, droplets were analyzed using the QX200 Droplet Reader. Absolute copy number quantification was performed using QuantaSoft software, version v1.7 and Poisson statistics. Thresholds were manually set, guided by positive (synthetic target sequence specific to the target) and no-template controls. Data visualization included HPRT1 normalization protocols⁸⁰.

Production of second-generation self-inactivating lentiviral vectors

Lipofectamine LTX™-based (Cat# 15338500, Invitrogen) transfection of ~ 60% confluent HEK Lenti-X™ 293 T cells (Cat# 632180, Takara) was performed with transgene-encoding lentiviral vectors as well as packaging vectors pCAG-VSVG and psPAX2 at a 2:1:2 molar ratio. After 48 h, the supernatant was collected and centrifuged for 10 min at 500 x g to remove remaining cells and concentrated 10-fold using lenti-X concentrator (Cat# 631231, Takara) by centrifugation and resuspension according to the manufacturer’s protocol.

Isolation, transduction and culture of healthy donor T cells

Lymphocytes were isolated from fresh buffy coats from anonymized healthy donors (Blutspende SRK Zürich) by density gradient centrifugation using Histopaque®−1077 (Cat# 10771-500 ML, Sigma-Aldrich). The enriched cell population was further purified separately by MACS® separation using human CD4 and CD8 T cell isolation kits (Cat# 130-096-533 and Cat# 130-096-495, respectively, Miltenyi Biotec). The purified T cell populations were frozen in liquid nitrogen as 5 × 10⁶ cells/mL aliquots in advanced RPMI 1640 (Cat# 12633012, Thermo Fisher Scientific), supplemented with 20% FBS (Cat# F4135-500ML, Sigma-Aldrich), 1x GlutaMAX™ (Cat# 35050061, Thermo Fisher Scientific), and 20% (volume/volume) DMSO (Cat# D2650-100ML, Sigma-Aldrich). Frozen CD4 and CD8 T cell aliquots were thawed in a 37 °C water bath and washed and cultured in TheraPEAK® T-VIVO® cell culture medium, containing 25 ng/mL interleukin 7 (IL7; Cat# 130-095-364, Miltenyi Biotec) and 50 ng/mL IL15 (Cat# 130-095-766, Miltenyi Biotec). T cells were activated using Immunocult CD3/CD28/CD2 T cell Activator cocktail (Cat# 10990, STEMCELL Technologies) for 24 h according to the manufacturer’s protocol. After addition of 8 µg/mL Polybrene (Cat# TR-1003-50UL, Sigma-Aldrich) and Lentiboost P (1:100) (Cat# SB-P-LV-101-12, Sirion Biotech), previously purified second-generation self-inactivating lentiviral vectors were added to the activated T cells at a multiplicity of infection (MOI) of 5. Engineered or wild type T cells were cultured in G-Rex® 24-well cell culture plates (Cat# 80192 M, Wilson Wolf) or G-Rex® 6 well cell culture plates (Cat# MSPP-80660M, Wilson Wolf) at densities of 0.2 ×  10⁶ cells/mL to 4 × 10⁶ cells/mL. T cells were expanded for a minimum of 7 days prior to all analysis and experiments, unless otherwise stated.

Flow cytometric analysis in vitro

Cell surface expression of chimeric antigen receptors was analyzed by staining with fluorescently-labeled anti-G4S antibody (AF647, clone E7O2V, 1:50, Cat# 69782, Cell Signaling Technologies). The phenotype of T cells in vitro was assessed by flow cytometric analysis of cell surface receptors using the following antibodies: CD25 (BV605, clone BC96, 1:200, Cat# 302632, BioLegend), Lag3 (PE, clone 3DS223H, 1:100, Cat# 12-2239-42, eBioscience), CD69 (PE, clone FN50, 1:100, Cat# 310906, BioLegend), PD1 (BV786, clone EH12.2H7, 1:100, Cat# 329930, BioLegend), CD8 (BV711, clone RPA-T8, 1:100, Cat# 301044, BioLegend), CD62L (BV650, clone DREG-56, 1:100, Cat# 304832, BioLegend), CD45RO (BUV395, clone UCHL1, 1:50, Cat# 564291, BD). After staining, cells were acquired using a BD LSRFortessa™ Cell Analyzer and analyzed using FlowJo™ software, version v10.8 (BD Life Sciences).

Killing assay

For the killing assays, RPMI-8226 and Raji cancer cells were seeded in the maintenance media (RPMI 1640 complete media) in a 96-well plate with edge reservoir (Cat# 167425, Thermo Fisher Scientific). Engineered T cells were washed twice in DPBS and resuspended in the same media. CAR-T cell populations were normalized to equal percentage of CAR-positive T cells by dilution with the respective donor wild type T cells before being added to the target cells. Engineered T cells and cancer target cell lines were used at a constant effector to target cell ratio (1:1, 10,000 and 10,000 cells). Stimulation of CAR expression was performed using 2-fold or 10-fold serial dilutions of risdiplam starting at 500 nM. Relative amount of cell death was determined using CytoTox-Glo™ Assay (Cat# G9291, Promega) according to the manufacturer’s protocol. Data were normalized to cancer cells (RPMI-8226 or Raji) only control and maximum killing effect was achieved by using conventional BCMA-/CD19-CD28-CD3z CAR-T or a digitonin (Cat# G9291, Promega) control (0–100%). Data were further analyzed using GraphPad Prism software, version v10.6.1.

Statistics and reproducibility

Statistical analyses and graph generation were performed using GraphPad Prism software, version v10.6.1, FlowJo software, versions v10.10.0 and v10.8, and the R package DESeq2, version v1.44.0. Volcano plot was generated with ggVolcanoR⁸¹. Exact sample sizes (n) and the number of independent replicates to ensure reproducibility are specified in the individual figure legends. No statistical method was used to predetermine sample size; rather, sample sizes were based on standard practices for cell-based and in vivo assays. No data were excluded from the analyses.

Data are presented as either individual data points, mean ± standard deviation (SD), mean ± standard error of mean (SEM) or median (as indicated in the figure legends). Statistical significance was defined as P < 0.05 and determined using unpaired Student’s t-tests, or one-way, two-way, and Welch’s ANOVAs with appropriate post-hoc tests (e.g., Tukey HSD), as specified in the figure legends.

For animal studies, mice were randomly divided into the experimental groups. Otherwise, the experiments were not randomized. The investigators were not blinded to allocation during experiments and outcome assessment.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

M.M., T.S., D.S., S.T., A.F., D.M., S.L. performed experiments. M.M., T.S., D.S., S.T., A.F., D.M., S.L., M.E., L.S. performed data analyses. M.M., D.S., A.F., F.R., L.S. wrote the manuscript. R.J., F.R., L.S. conceived the study. D.V., H.R., F.R., L.S. guided and directed the research. All authors have read and approved the manuscript.

Peer review

Peer review information

Nature Communications thanks Yuxuan Guo, and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

Funding for this study was provided by F. Hoffmann-La Roche Ltd and Genentech Inc.

Data availability

The FASTQ files generated from the HDR screen experiment have been deposited in the GEO database under accession code GSE320370. The data that support the findings of this study are available within the paper and its Supplementary Information. Source data are provided with this paper.

Competing interests

This study is supported by F. Hoffmann-La Roche Ltd and Genentech Inc. M.M., D.S., T.S., S.T., A.F., D.M., S.L., D.V., H.R., R.J., M.E., F.R., and L.S. were employees, master or PhD students of F. Hoffmann-La Roche at the time of performing the research. H.R. is currently employed at Skyhawk Therapeutics (Switzerland). D.S. and S.L. were supported by the F. Hoffmann-La Roche doctoral fellowship.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-73673-1.