IDMME and IDMDE: inducible CRISPR-dCasRx platforms for spatiotemporal RNA m5C editing Authors: Jinming Xu, Jiaju Xu, Congcong Cao, Yichang Shu, Letian Shen, Xuhong He, Qi Huangfu, Chengfang Sun, Weikai Wang, Jingchao Wei, Ming Cai, Bohan Wang, Aolin Li, Yuchen Liu, Jiaming Wen Published: 2026 ABSTRACT AbstractRNA 5-methylcytosine (m5C) is a dynamic epigenetic mark implicated in tumorigenesis, however, existing editors lack spatiotemporal regulation and reversibility. Here, we develop abscisic acid (ABA)-inducible CRISPR-dCasRx systems for programmable m5C methylation (IDMME) and demethylation (IDMDE). These editors enable site-specific, low off-target m5C modification through ligand-dependent assembly of split effector domains. We further integrate photocaged ABA to achieve light-controlled activation. Application in renal carcinoma models shows that targeted m5C editing modulates transcript function and suppresses tumor growth in vitro and in vivo. This platform provides a versatile, spatiotemporally controllable approach for dissecting RNA epigenetic mechanisms and advancing RNA-based therapeutic strategies.Supplementary InformationThe online version contains supplementary material available at 10.1186/s13059-026-04150-7.Keywords: RNA m5C methylation, CRISPR dCasRx, Epigenetic therapy, Chemical inducible proximity, Renal cell carcinoma Background With the continuous development of high-throughput sequencing and detection techniques, post-transcriptional RNA modifications have gained recognition as critical elements in the regulation of epigenetic landscapes [1]. More than 170 types of RNA methylations have been identified to date, among which the most prominent include N6-methyladenosine (m6A), 5-methylcytosine (m5C), N1-methyladenosine (m1A), and N7-methylguanosine (m7G). Among these modifications, m5C is particularly notable due to its dynamic involvement in various aspects of RNA metabolism, including mRNA nuclear export [2], stability [2–4], and translation efficiency [5]. Catalyzed by the NOL1/NOP2/SUN domain (NSUN) methyltransferase family and reversed by ten-eleven translocation (TET) proteins, m5C modification influences cancer progression and recurrence by altering RNA–protein interactions in a dysregulated manner [4, 6]. Among NSUN family members, NSUN2 and NSUN6 are key m5C methyltransferases, with NSUN2 being more extensively utilized in genetic engineering. NSUN2 is evolutionarily conserved, mediating 89–97% of functional m5C sites across most species and being widely distributed in coding and non-coding regions. In contrast, NSUN6 is largely human-specific, accounting for only 23.7% of m5C sites, and its activity is constrained by species-specific factors [1, 7, 8]. RNA modification-based targeted therapy holds great promise for cancer treatment. A deeper understanding of the mechanisms by which RNA modifications contribute to tumor progression can aid in the development of novel therapeutic strategies aimed at inhibiting tumor advancement and improving patient prognosis [9]. Notably, compared to DNA-targeted gene editing, an RNA modification-based strategy exhibits lower genotoxicity, while offering advantages over protein-targeted inhibitors or antibodies by reducing off-target effects and associated toxicities. However, conventional m5C modulation strategies that rely on the broad-spectrum manipulation of methyltransferases NSUN2/6 or the demethylase TET2 fail to achieve locus-specific precision. This limitation hampers the functional analysis of individual m5C sites and leads to indiscriminate transcriptional dysregulation of both oncogenic and tumor-suppressive pathways [10–12]. To address these limitations, we aimed to develop tools for site-specific m5C editing, enabling precise writing and erasure of specific m5C sites in diseases like renal cell carcinoma (RCC). CRISPR-Cas systems, fused with RNA-modifying enzymes, have revolutionized site-specific nucleic acid editing [13]. The catalytically inactive forms of Cas proteins were previously utilized for in vivo translation enhancement and RNA editing [14]. We previously developed a CRISPR-Cas13d-eIF4G-based mRNA translation enhancement system to treat calcium oxalate kidney stones and also engineered a CRISPR/dCasRx-SINEB2 system for the targeted activation of transcripts encoding anti-tumor proteins [15, 16]. An efficient RNA knockdown system was subsequently engineered in our group using CRISPR-CasRx and hammerhead ribozyme [17]. Additionally, within the CRISPR/dCas9-SAM gene activation system, we developed a tumor-targeted hollow mesoporous silica nanoparticle to load gene-editing components for enhancing gene activation efficiency [18]. As for programmable RNA methylation, studies have demonstrated that dCas13 fusions with the m6A demethylase ALKBH5 or the m1A-modifying enzyme ALKBH3 enable targeted RNA methylation editing [19, 20]. However, the substantial size of Cas proteins and the prolonged expression of CRISPR/RNA editing enzymes could disrupt the natural functions of target RNAs and cells. Furthermore, current m5C editors are often hindered by suboptimal efficiency, insufficient optimization, and a significant deficiency in spatiotemporal control [21]. These limitations have impeded both mechanistic investigations into m5C dynamics and the therapeutic applications that require precise temporal or spatial regulation, particularly in diseases such as RCC. Precise spatiotemporal control over biological processes is critical in precision therapy, particularly in tumors where proteins, even those encoded by oncogenes, can exhibit beneficial roles for physiological homeostasis. Chemical induction systems address this need by enabling tunable, reversible, and dose-responsive manipulation of cellular activities. A key example is the chemically induced proximity (CIP) system, where small molecules such as abscisic acid (ABA) trigger interactions between adaptor proteins (e.g., PYL and ABI) fused to effector domains, allowing programmable control over downstream events [22, 23]. Moreover, other chemically induced dimerization systems, such as FKBP-FRB and BCL-xL-ABT-737 similarly enable small-molecule-guided regulation of genetic or epigenetic states [11, 24]. By integrating these approaches with RNA-targeting CRISPR systems, we could achieve dynamic and quantitative control of m5C editing, thereby overcoming the limitations associated with static or irreversible editors. Herein, we developed a chemically induced CRISPR-dCasRx RNA m5C methylation editing system for RCC therapy, designed for compatibility with both in vitro and in vivo applications. This system enables precise temporal control over m5C modifications at disease-relevant targets in RCC cells, serving as a versatile tool to dissect the functional roles of m5C dynamics in tumor progression and to establish precision therapeutic strategies via the induction of apoptosis and ferroptosis. By integrating target-specific m5C modulation with spatiotemporal regulation, our approach advances RNA epigenetic research while demonstrating potential for clinical applications through its dual applicability in in vitro experimental studies and in vivo therapeutic interventions. Results Engineered dCasRx systems for bidirectional RNA m5C editing An ideal site-specific m5C editing platform requires precise targeting capability, bidirectional modulation (methylation and demethylation), and reversible control. To achieve these objectives, we selected the enzymatically inactivated Cas13d ortholog derived from Ruminococcus flavefaciens XPD3002 (RfxCas13d, designated dCasRx) due to its well-established RNA-guided RNA-targeting activity. We engineered dCasRx by fusing it with either the m5C writer NSUN2 (which was selected over NSUN6 for its conserved predominance in mediating most functional m5C sites across species [7, 8]) or the eraser TET2, to enable m5C installation or removal at specific RNA sites through guide RNA (gRNA) targeting (Fig. 1A). Dual nuclear localization signals (NLS) flanking the dCasRx-effector fusion proteins were incorporated to ensure nuclear targeting, thereby enabling engagement with nascent transcripts. The resulting m5C editing complexes were designated NLS-dCasRx-NLS-NSUN2 (dCasRx-NSUN2) and NLS-dCasRx-NLS-TET2 (dCasRx-TET2), respectively (Fig. 1B). The design and validation of CRISPR-dCasRx-mediated m5C methylation editors (DMME). A Schematic of the DMME dCasRx-NSUN2 RNA methylation system. The editor was engineered through fusion of the NSUN2 methyltransferase with catalytically inactive dCasRx, enabling gRNA-guided m5C deposition at defined RNA loci via nuclear-localized enzymatic activity. B Domain architecture of the dCasRx-NSUN2 m5C methylation module with dual NLS for nuclear localization. C Expression of dCasRx-NSUN2 fusion protein was examined by western blot assay. D Subcellular localization of the fusion protein in HEK-293T cell after transfection with the NSUN2 or dCasRx-NSUN2 construct were checked by confocal imaging using antibody against NSUN2. E Schematic representation of positions of the m5C site within HDGF mRNA and regions targeted by three gRNAs. RT-qPCR assay was employed for detecting the mRNA expression of HDGF in HEK-293T cell co-transfected with wild-type (WT) CasRx and NT-gRNA (control) or gRNA1/2/3. F The m5C level of HDGF in HEK-293T co-transfected dCasRx-NSUN2 and NT-gRNA (control) or gRNA1/2/3 were measured by m5CRIP-qPCR. G Western blot results of HA-tag demonstrated expression of DMME in HEK-293T cells when transfected with dCasRx-NSUN2-FL (1-767aa), dCasRx-NSUN2-1 (66–430 aa), dCasRx-NSUN2-2 (170-341aa). H The m5C level changes under different treatment conditions using the DMME. I, J The expression levels of HDGF in HEK-293T cells under the same treatment as (H) were detected by RT-qPCR and western blot assay. Three independent experiments were presented as mean ± SD (n = 3). Statistical analysis was performed using one-way ANOVA followed by Tukey’s multiple comparisons test, defined as *P < 0.05, **P < 0.01, and ***P < 0.001 Validation of CRISPR-dCasRx-mediated m5C methylation editors (DMME) To confirm the expression and nuclear localization of the DMME system, HEK-293T cells were transfected with the dCasRx-NSUN2 fusion construct. Immunoblotting analysis demonstrated robust expression of the full-length dCasRx-NSUN2 protein in the transfected cells (Fig. 1C). Confocal microscopy revealed that the integration of dual nuclear localization signals (NLS) reprogrammed the subcellular trafficking of endogenous NSUN2 in mammalian cells, driving the nuclear localization‌ of the dCasRx-NSUN2 fusion protein (Fig. 1D). As a proof-of-concept, we targeted two established m5C-modified loci: cytosine C1052 in the 3’UTR of HDGF mRNA and cytosine C313 in the CDS of SLC7A11 [25, 26]. Three gRNAs were designed to flank each specific site (± 50 nucleotides). Co-transfection of wild-type CasRx with HDGF- or SLC7A11-specific gRNAs induced significant reductions in the respective mRNA levels (Fig. 1E; Additional file 1: Fig. S1A, B), confirming their ability to direct CasRx to the target loci. We next evaluated the m5C editing capacity of the DMME system by co-transfecting dCasRx-NSUN2 with each gRNA. m5CRIP-qPCR analysis demonstrated that the dCasRx-NSUN2 induced site-specific m5C hypermethylation at the HDGF C1052 site. Notably, gRNA3 (designed to target a 30-nucleotide region upstream of the m5C site) exhibited the maximal catalytic activity, yielding a 3.4-fold methylation enrichment relative to baseline levels (Fig. 1F). Parallel analysis of SLC7A11 C313 demonstrated analogous editing dynamics, with gRNA3 achieving a 3.3-fold methylation enhancement (Additional file 1: Fig. S1C), confirming the versatility of the system across different transcript regions. To address challenges associated with the large size of full-length NSUN2 (767 amino acids), which often hampers recombinant expression and editing efficiency, we rationally designed and evaluated truncated variants. Through sequence alignment with orthologs, we identified a conserved core region spanning residues 32–440, which contains the catalytically essential residues Cys271 and Cys321 (Additional file 1: Fig. S2A) [25, 27]. Consistent with earlier reports, the functional core of NSUN2 comprises a Rossmann-fold catalytic domain (residues 171–429) and a PUA domain (residues 54–147) responsible for nucleic acid binding [28, 29]. Cross-referencing with domain annotations from Pfam [30, 31] and PROSITE [32] guided the selection of two candidate fragments: NSUN2-1 (residues 66–430) retained the complete SAM-dependent methyltransferase domain, including the PUA-containing N-terminal segment (66–169) and the C-terminal extension (342–430); and NSUN2-2 (residues 170–341), which represented a minimal construct encompassing the core catalytic pocket. Both truncations substantially reduced protein size while preserving the essential catalytic machinery. Structural predictions using AlphaFold2 confirmed that each variant maintained an intact catalytic cleft with Cys271 and Cys321 in close spatial proximity, consistent with a functional active site (Additional file 1: Fig. S2B, C). Western blot analysis validated the successful synthesis of both truncation variants (Fig. 1G). At the HDGF C1052 site, NSUN2-1 induced a 6.0-fold m5C elevation, surpassing NSUN2-FL (2.1-fold) and NSUN2-2 (3.7-fold, Fig. 1H), with concordant increases in mRNA and protein levels (Fig. 1I, J). This advantage extended to SLC7A11 C313, where NSUN2-1 achieved 4.7-fold methylation versus the 2.7-fold increase by NSUN2-FL (Additional file 1: Fig. S1D). Corresponding increases in SLC7A11 mRNA and protein levels were also observed (Additional file 1: Fig. S1E, F). The superior performance of NSUN2-1 likely stems from its balanced preservation of catalytic residues and reduced molecular weight, which may enhance nuclear import and target site occupancy compared to the bulkier full-length enzyme. Based on these results, we selected NSUN2-1 for subsequent studies to maximize methylation efficiency. This truncation strategy effectively isolated the catalytic core of NSUN2, enabling precise m5C editing without compromising enzymatic activity. Development of inducible CRISPR-dCasRx-mediated m5C methylation editors (IDMME) While existing dCasRx-NSUN2 fusion systems enable site-specific m5C editing, their inability to provide temporal control over methylation limits dynamic studies of epigenetic regulation. To address this, we integrated the DMME dCasRx-NSUN2-1 system with a chemically inducible proximity tool to develop an ABA-inducible, temporally specific methylation editor, designated IDMME (Fig. 2A). The IDMME comprises two modular components: (1) dCasRx-PYL, consisting of dCasRx fused to the PYL domain, and (2) ABI-NSUN2-1, comprising the ABA-insensitive protein (ABI) linked to NSUN2-1. Both constructs were tagged with HA or Myc epitopes and incorporated nuclear localization signals (NLS) to ensure efficient nuclear enrichment and editing efficiency (Fig. 2B). As a negative control, we generated an enzymatically inactive ABI-NSUN2-1Mut variant by introducing C271A/C321A mutations into the NSUN2-1 catalytic domain, thereby ensuring that observed m5C changes were specifically attributable to NSUN2-1 catalytic activity. Immunofluorescence confirmed the nuclear localization of dCasRx-PYL and ABI-NSUN2-1 (Fig. 2C), while western blot demonstrated the expression of dCasRx-PYL, ABI-NSUN2-1, and ABI-NSUN2-1Mut in HEK-293T cells (Fig. 2D). The dCasRx-PYL/gRNA complex selectively bound to target RNA transcripts, whereas ABI-NSUN2-1 lacked inherent targeting capacity. However, ABA-induced heterodimerization between PYL and ABI recruited ABI-NSUN2-1 to dCasRx-PYL, enabling precise m5C modification (Fig. 2A). Notably, this interaction was reversible upon ABA removal. Co-immunoprecipitation (Co-IP) assays validated the ABA-dependent association between dCasRx-PYL and ABI-NSUN2-1. In HEK-293T cells co-transfected with both components, ABA triggered stable complex formation. Conversely, ABA withdrawal dissociated the dCasRx-PYL/ABI-NSUN2-1 complex, restoring baseline interaction levels (Fig. 2E). These results confirmed the temporal controllability of IDMME, establishing it as a tool for the time-resolved analysis of m5C-mediated epigenetic dynamics. Design and validation of inducible CRISPR-dCasRx-mediated m5C methylation editors (IDMME) via ABA dimerization. A ABA-triggered reversible methylation: dCasRx-PYL targeted cytosine loci via gRNA. ABA induced PYL/ABI dimerization to recruit NSUN2-ABI for site-specific m5C deposition; removal of ABA reversed methylation by complex dissociation. B Molecular architecture of IDMME components. Constructs express: dCasRx-PYL-HA: NLS-flanked targeting module; Myc-ABI-NSUN2-1: Catalytically active m5C methyltransferase (residues 66–430); Myc-ABI-NSUN2-1Mut: Catalytically dead mutant, NSUN2 C271A C321A. C Subcellular localization analysis by confocal microscopy. HA-tagged dCasRx-PYL and Myc-tagged NSUN2-ABI colocalize with nuclear DAPI staining (blue). Scale bar: 17.5 μm. D Western blot validation of recombinant protein expression in HEK-293T cells. E Co-immunoprecipitation (Co-IP) analysis demonstrated ABA-dependent interaction between HA-tagged dCasRx-PYL and Myc-tagged ABI-NSUN2 in HEK-293T cells. The complex dissociated within 24 h post-ABA withdrawal, confirming chemically reversible recruitment of the m5C methyltransferase Inducible and reversible m5C editing via the IDMME system To validate the ABA-controlled m5C editing capability of the IDMME system, we systematically assessed its target-specific RNA recruitment and catalytic precision. RNA immunoprecipitation (RIP) assays targeting HDGF mRNA demonstrated that ABA treatment triggered the interaction between the split IDMME components (dCasRx-PYL and ABI-NSUN2-1) and the target RNA, whereas no binding was observed in the absence of ABA (Fig. 3A). To investigate the inductive effect of ABA on m5C methylation by IDMME, we demonstrated that m5C levels at the HDGF C1052 site increased in a time‑ and dose‑dependent manner, reaching a 6‑ to sevenfold enrichment and approaching saturation under 100 μM ABA treatment for 24 h (Fig. 3B-D). Notably, the m5C editing efficiencies achieved by IDMME were comparable to those of constitutive NSUN2-1 overexpression or non-inducible dCasRx-NSUN2-1 systems. These results indicated that the truncation itself (NSUN2-1) provides a highly active catalytic core, which accounted for the high baseline efficiency across these constructs. Catalytically-dead mutations in NSUN2-1 completely abolished these effects, confirming that the modification was enzymatically driven (Fig. 3E). Inducible and Reversible m5C methylation of C1052 on HDGF via the IDMME System. A The enrichment level of Myc-ABI-NSUN2-1 fusion protein on HDGF in HEK-293T cells co-transfected with IDMME System with or without ABA treatment were examined by RIP-qPCR. B The m5C level of HDGF C1052 in HEK-293T co-transfected with IDMME system with or without ABA treatment were examined by m5CRIP-qPCR analysis. C The methylation effect of 24 h IDMME treatment under different concentration of ABA treatment was evaluated by m5CRIP-qPCR. D The methylation effect of IDMME system under 100 μM ABA addition for different time was evaluated by m5CRIP-qPCR. E The m5C level changed under different treatment conditions using the ABA-inducible (IDMME system) or non-inducible m5C methylation system. F The expression levels of HDGF in HEK-293T cells under the same treatment as (E) were detected by RT-qPCR. G The relative expression of HDGF mRNA in HEK-293T cells at different time points after actinomycin D treatment under the same treatment as (E). H The m5C level and (I) the mRNA level change before or after ABA removal (for 24 h). J The methylation effect at different time points after washing out ABA. ABA was added to induce m5C writing for 24 h before the time-course experiments of ABA removal. Three independent experiments were presented as mean ± SD (n = 3). Statistical analysis was performed using two-way analysis of variance (A, G), two-tailed Student’s t-test (B, H, I), one-way ANOVA followed by Tukey’s multiple comparisons test (C-F, J), defined as *P < 0.05, **P < 0.01, and ***P < 0.001 The generalizability and specificity of the IDMME platform were further validated across multiple targets and cell lines, including HEK-293T and 769-P. Site-specific gRNAs triggered significant m5C deposition, as detected via m5CRIP-qPCR at HDGF C1052 (6.2-fold), SLC7A11 C313 (4.9-fold), and KAT7 C717 (18.3-fold) without cross-modification at non-targeted regions (Additional file 1: Fig. S3A). To evaluate the editing performance of IDMME across diverse epigenetic backgrounds, we classified target sites into three categories based on their basal methylation levels as previously reported: low (5–20%), medium (20–40%), and high (> 40%) [7, 8]. Single-base resolution targeted bisulfite sequencing (TBS-seq) revealed that upon ABA-induced activation, the IDMME system consistently elevated m5C methylation levels at all tested sites in both cell lines, with higher editing efficiency observed at targets with low and medium basal methylation levels (Additional file 1: Fig. S4). Further functional assessment revealed that m5C deposition at the HDGF C1052, SLC7A11 C313 and KAT7 C717 sites elevated both mRNA and protein levels, with an efficacy comparable to NSUN2-1 overexpression and non-inducible dCasRx-NSUN2-1 systems (Fig. 3F, Additional file 1: Fig. S3B, C). RNA decay assays using actinomycin D confirmed that this regulation was due to enhanced transcript stability, as evidenced by the significantly prolonged half-life of target mRNAs (Fig. 3G). Crucially, the system provided temporal reversibility; upon ABA withdrawal, m5C levels at the HDGF C1052 site declined rapidly within 6 h, and both methylation and mRNA expression fully returned to baseline within 24 h (Fig. 3H-J). Collectively, these findings establish IDMME as a precise and reversible tool for RNA epigenetic engineering across diverse methylation contexts. To evaluate the sequence specificity of the IDMME system, we identified the top three putative off-target sequences for HDGF-gRNA#3 using BLASTN (https://blast.ncbi.nlm.nih.gov/) and designed primers for m5CRIP-qPCR analysis. As demonstrated above, the IDMME system exhibited superior targeting precision when editing the HDGF C1052 site. Three non-target transcripts (CDK5RAP3, PTCH1 and PRRT4) exhibiting around 80% sequence homology to the HDGF gRNA#3 binding region (nt1004-1026) were selected as potential off-target candidates (Additional file 1: Fig. S3D). Post-editing m5C quantification via m5CRIP-qPCR revealed no significant methylation alterations at these sites (Additional file 1: Fig. S3E). Parallel validation using the SLC7A11 C313 model confirmed this targeting specificity, with non-target loci maintaining baseline methylation levels (Additional file 1: Fig. S3F, G). These findings demonstrated IDMME’s capacity for precise RNA methylation without detectable off-target activity. IDMME-mediated methylation at SLC7A11 C313 regulated ferroptosis in RCC The IDMME platform’s capacity to precisely modulate site-specific m5C levels prompted an investigation into whether RNA methylation-dependent mRNA stability could elicit therapeutic responses in RCC. A previous study revealed that NSUN2 recruits the m5C reader YBX1 to recognize and bind m5C-modified sites on SLC7A11 mRNA, enhancing transcript stability and expression levels to drive ferroptosis suppression [26]. We engineered RCC 769-P cells with a non-inducible DMME (dCasRx-NSUN2-1) and an ABA-inducible IDMME system (dCasRx-PYL + ABI-NSUN2-1 + ABA), alongside control groups including non-induced IDMME (-ABA) and a catalytically inactive IDMME (ABI-NSUN2-1Mut), to target the SLC7A11 C313 site. Site-specific m5C editing of SLC7A11 mRNA was exclusively observed in the DMME and ABA-induced IDMME groups, which demonstrated a 3.1-fold increase in m5C methylation, accompanied by elevated mRNA and protein expression (Fig. 4A-C). In contrast, non-induced IDMME and the enzymatically dead mutant showed negligible activity, confirming the system’s dual dependency on ABA induction and catalytic integrity. IDMME-directed m5C methylation at SLC7A11 C313 suppressed ferroptosis and drove tumor progression. A The m5C level changed under different treatment conditions using the ABA-inducible (IDMME system) or non-inducible m5C methylation system in 769-P cells. B The mRNA and (C) protein expression levels of SLC7A11 in 769-P cells under the same treatment as (A). D CCK-8 assay and (E) colony formation assay revealed the proliferative activity of 769-P cells under the same treatment as (A). F Flow cytometer measured the cell death rates (PI-positive) of 769-P cells under the same treatment as (A). G Intracellular lipid peroxides MDA levels were measured with an MDA assay kit. H Representative data and statistical graph of lipid ROS using BODIPY 581/591 C11 probe in 769-P cells under the same treatment as (A) were shown. Three independent experiments were presented as mean ± SD (n = 3). Statistical analysis was performed using one-way ANOVA followed by Tukey’s multiple comparisons test (A, B, E–H), and two-way analysis of variance (D), defined as ns. not significant, **P < 0.01, and ***P < 0.001 Functional validation centered on SLC7A11-mediated ferroptosis, a regulated cell death pathway mechanistically linked to cystine uptake and lipid peroxidation. CCK-8 and colony formation assays revealed significantly changed proliferative capacity in DMME and ABA-induced IDMME groups (Fig. 4D, E). Flow cytometry demonstrated markedly lower PI positivity in these groups compared to controls, indicative of reduced cell death (Fig. 4F). Given that lipid peroxidation is a hallmark of ferroptosis, we quantified malondialdehyde (MDA) and lipid peroxidation levels. The DMME and ABA-induced IDMME groups exhibited attenuated lipid ROS accumulation, as evidenced by decreased MDA levels and reduced BODIPY 581/591 C11 fluorescence intensity (Fig. 4G, H). Critically, these effects were dependent on ABA induction and catalytically active NSUN2, as neither the non-induced nor the enzymatically dead systems mitigated ferroptosis. These data demonstrated that IDMME-mediated m5C deposition at defined loci enables precise regulation of specific transcript abundance and consequent phenotypic alterations in RCC. Development of an inducible CRISPR-dCasRx-mediated m5C demethylation editors (IDMDE) Building upon the modular design of IDMME, we engineered an inducible demethylation module by fusing dCasRx-PYL-HA with Myc-ABI-TET2CD, which incorporates the catalytic domain of TET2 (1129–1936, Δ1481-1843) for m5C removal (Fig. 5A) [33]. A catalytically inactive mutant (Myc-ABI-TET2CD Mut) was generated as a negative control (Fig. 5B). Transfection of HEK-293T cells with the IDMDE components confirmed expression of both fusion proteins via western blot and their colocalization in the nucleus by immunofluorescence (Fig. 5C, D). Design and validation of inducible CRISPR-dCasRx-mediated m5C demethylation editors (IDMDE). A Schematic of ABA-inducible m5C demethylation at defined RNA loci via dCasRx-PYL/ABI-TET2CD targeting. B Molecular architecture of the inducible m5C demethylation system: dCasRx-PYL, ABI-TET2CD, and ABI-TET2CD Mut (TET2 ΔHxD mutant: H1382Y/D1384V). C Western blot verification of Myc-ABI-TET2CD and Myc-ABI-TET2CD Mut fusion protein expression in HEK-293T cells. D Confocal microscopy analysis of subcellular localization revealed nuclear enrichment of Myc-ABI-TET2CD (red) in transfected HEK293T cells, colocalized with DAPI-stained nuclei (blue). Scale bar: 17.5 μm. E Schematic representation of positions of the m5C site within PBX1 mRNA and regions targeted by three gRNAs. m5CRIP-qPCR assay was employed for detecting the m5C level of PBX1 in HEK-293T cells co-transfected with dCasRx-TET2CD and NT-gRNA (control) or gRNA1/2/3. F The m5C level changed under different treatment conditions using the ABA-inducible (IDMDE system) or non-inducible m5C demethylation system. G The expression levels of PBX1 in HEK-293T cells under the same treatment as F were detected by RT-qPCR. H The relative expression of PBX1 mRNA in HEK-293T cells at different time points after actinomycin D treatment under the same treatment as (F). I m5C methylation dynamics in HEK-293T cells under IDMDE modulation. Volcano plots displaying m5C methylation differences between IDMDE-treated (+ ABA) and untreated (-ABA) groups, guided by PBX1 C6935-targeting gRNA. Venn diagram analysis of overlapping m5C sites across experimental conditions. Data derived from two biologically independent Nanopore DRS-seq replicates. J Volcano plots displaying mRNA level differences between IDMDE-treated (+ ABA) and untreated (-ABA) groups, guided by PBX1 C6935-targeting gRNA. Venn diagram analysis of overlapping transcripts across experimental conditions. Data derived from two biologically independent Nanopore DRS-seq replicates. K Integrative-genomics-viewer tracks displaying the differential m5C levels of PBX1 between IDMDE-treated (+ ABA) and untreated (-ABA) groups. Three independent experiments were presented as mean ± SD (n = 3). Statistical analysis was performed using one-way ANOVA followed by Tukey’s multiple comparisons test (E–G), and two-way analysis of variance (H), defined as *P < 0.05, **P < 0.01, and ***P < 0.001 To validate the site-specific demethylation capacity of the IDMDE system, we targeted two established m5C-modified loci: C6935 in the 3’UTR of PBX1 and C1798 in the CDS of TRAF7. Three gRNAs targeting the PBX1 C6935 were engineered and delivered alongside dCasRx-TET2CD into HEK-293T cells, with gRNA#1 demonstrating the highest demethylation efficacy, yielding a 75% methylation reduction (Fig. 5E). A comparative analysis of IDMDE-mediated demethylation versus conventional TET2CD overexpression and the non-inducible dCasRx-TET2CD system was subsequently conducted. m5CRIP-qPCR analysis revealed that ABA-activated IDMDE induced significant m5C erasure at PBX1 C6935 with approximately 73% methylation reduction, albeit slightly attenuated compared to the TET2CD-overexpressing (84%) and dCasRx-TET2CD systems (78%, Fig. 5F). PBX1 mRNA downregulation was consistently observed across all functional demethylation platforms, confirming system-wide coordination between epigenetic editing and transcript regulation (Fig. 5G). This pattern was recapitulated at the TRAF7 C1798, where IDMDE-mediated demethylation approached the efficacy of TET2CD overexpression and the non-inducible dCasRx-TET2CD system (Additional file 1: Fig. S5A). Notably, TRAF7 also exhibited mRNA and protein downregulation following m5C loss (Additional file 1: Fig. S5B). Further actinomycin D chase assays demonstrated that TET2-mediated demethylation destabilized PBX1 mRNA, remarkably reducing its half-life (Fig. 5H). These results above confirmed that the IDMDE system strictly requires ABA induction and an intact catalytic domain for activity, achieving efficient site-specific m5C erasure. The key advantage of IDMDE over global demethylase-based strategies lies in its capacity to preserve genome-wide methylation fidelity, characterized by minimal perturbation of non-target methylation patterns, while achieving precise site-specific editing. To comprehensively assess the on-target specificity and genome-wide methylation fidelity of IDMDE, an integrated analysis using Nanopore DRS-seq for m5C sequencing and RNA-seq revealed that ABA-induced IDMDE treatment elicited highly selective epigenetic and transcriptional alterations. Specifically, only 92 significantly down-regulated m5C sites (2.83% of total m5C loci) and 562 significantly down-regulated transcripts (2.49% of total transcripts) met the thresholds of P < 0.05 and |log2FC|> 1. Notably, the PBX1 C6935 site demonstrated prominent demethylation efficacy (log2FC = −1.18, P < 0.05), ranking within the top 1.07% of edited sites by effect magnitude, accompanied by a corresponding reduction in principal PBX1 transcript levels (Fig. 5I, J). The integrative genomics viewer (IGV) quantitatively confirmed decreased m5C occupancy at PBX1 C6935 from 17.9% to 12.6%, while whole-genome methylation patterns remained largely stable (Fig. 5K). These coordinated epigenetic and transcriptional changes of PBX1 substantiated the precision of IDMDE in achieving targeted demethylation without provoking widespread off-target effects, establishing its functional specificity in coupling site-specific methylation editing with downstream gene regulation. To evaluate the targeted demethylation efficiency of the IDMDE system across varying basal m5C levels at single-base resolution, we performed TBS-seq in HEK-293T and A-498 cells. A panel of targets covering low (PBX1), medium (SOCS3), and high (TRAF7) basal methylation levels was examined. Upon ABA-induced activation, the IDMDE system consistently and significantly reduced m5C methylation at all tested sites in both cell lines, demonstrating effective demethylation activity regardless of basal methylation status (Additional file 1: Fig. S6). These findings confirmed that IDMDE enables efficient and versatile targeted m5C demethylation across diverse cellular contexts and target backgrounds, establishing it as a broadly applicable tool for RNA epigenetic engineering. IDMDE-mediated demethylation at TRAF7 C1798 reduced malignancy in RCC To investigate the m5C demethylation potential of IDMDE in RCC, A-498 cells were transfected with the ABA-inducible IDMDE system or the non-inducible dCasRx-TET2CD, alongside controls including uninduced IDMDE (-ABA) and the catalytically inactive IDMDE (ABI-TET2CD Mut), targeting the oncogenic TRAF7 C1798 site. m5CRIP-qPCR analysis revealed that both the non-inducible dCasRx-TET2CD and the ABA-activated IDMDE significantly reduced the TRAF7 C1798 methylation levels by 86% and 73%, respectively (Fig. 6A), resulting in TRAF7 mRNA destabilization and subsequent protein downregulation (Fig. 6B, C). The uninduced IDMDE and the catalytic mutant (TET2CD Mut) exhibited minimal demethylation, confirming the ABA dependency and enzymatic activity requirements. Functional evaluations of TRAF7 demonstrated tumor-suppressive effects in both the non-inducible dCasRx-TET2CD and the ABA-inducible IDMDE systems. CCK-8 assays showed significant inhibition of cell proliferation in both groups, which was corroborated by flow cytometry revealing elevated apoptosis rates (Fig. 6D, E). Transwell analyses further showed a remarkable reduction in migratory capacity in both the non-inducible and ABA-inducible groups (Fig. 6F). Collectively, these findings established that the ABA-inducible IDMDE system enabled site-specific m5C demethylation at the TRAF7 C1798 site, achieving precise epigenetic reprogramming to destabilize oncogenic transcripts, suppress proliferative capacity, and inhibit the migration of RCC cells in vitro. IDMDE-mediated demethylation at TRAF7 C1798 attenuated oncogenic phenotypes in renal cell carcinoma. A The m5C level changed under different treatment conditions using the ABA-inducible (IDMDE system) or non-inducible m5C demethylation system in A-498 cells. B The mRNA and (C) protein expression levels of TRAF7 in A-498 cells under the same treatment as (A). D CCK-8 assay revealed the proliferative activity of A-498 cells under the same treatment as (A). E Flow cytometer measured the apoptotic cell rates of A-498 cells under the same treatment as (A). F Transwell assay revealed the migration ability of cells was significantly inhibited under the treatment of ABA-inducible (IDMDE system) or non-inducible m5C demethylation system. Three independent experiments were presented as mean ± SD (n = 3). Statistical analysis was performed using one-way ANOVA followed by Tukey’s multiple comparisons test (A, B, E, F), and two-way analysis of variance (D), defined as ns. not significant, *P < 0.05 and ***P < 0.001 IDMDE-mediated m5C demethylation for in vivo RCC targeted therapy Prior studies on m5C methylation editors have predominantly focused on in vitro models due to the challenges of achieving efficient and specific editing in vivo. To address this bottleneck, as illustrated in Fig. 7A, we performed an in vivo evaluation of IDMDE in RCC xenografts, aiming to demonstrate its therapeutic potential for precision oncology applications. BALB/c nude mice bearing A-498 tumors were divided into three groups: (1) IDMDE + ABA (AAV9-delivered IDMDE targeting TRAF7 C1798 with intraperitoneal ABA administration), (2) IDMDE + PBS (IDMDE without ABA induction), and (3) the AAV9 control. Two rounds of intratumoral AAV9 injections followed by six doses of ABA or PBS (administered every three days) revealed that the IDMDE + ABA treatment induced significant tumor suppression, whereas the IDMDE + PBS and AAV9 control groups exhibited progressive tumor growth (Fig. 7B-E). Throughout the treatment period, no significant difference in body weight or abnormal manifestations was observed in the IDMDE + ABA group compared with the control groups (Additional file 1: Fig. S7A), indicating the minimal systemic toxicity of ABA. In addition, we examined the hepatic and renal functions via mouse blood analysis before sacrifice, and no significant dysfunction were observed (Additional file 1: Fig. S7B). To further characterize the in vivo biosafety of ABA, we harvested major vital organs (including the heart, liver, spleen, lung, and kidney) for histopathological evaluation. No obvious organ damage or histopathological changes were detected in these organs from the IDMDE + ABA group, which further supported the favorable in vivo safety of ABA (Additional file 1: Fig. S7C). These results were consistant with the literature reporting that ABA, as a plant hormone, shows no significant toxicity in animals [34]. Further analysis of subcutaneous tumor specimens confirmed the capacity of IDMDE to drive site-specific m5C remodeling in vivo. m5CRIP-qPCR demonstrated an approximately 85% reduction in the TRAF7 C1798 methylation level in the IDMDE + ABA group (Fig. 7F), accompanied by concurrent decreases in TRAF7 mRNA and protein levels (Fig. 7G, H). These coordinated molecular changes confirmed the sequence-specific demethylation and downstream transcriptional silencing achieved by IDMDE. Additionally, TRAF7 downregulation coincided with a reduced tumor proliferation marker Ki-67, supporting the functional link between m5C editing and oncogenic pathway suppression. Although the role of TRAF7 in RCC progression has not been fully validated clinically, preclinical studies suggestted its involvement in tumorigenic pathways, and our findings highlighted RNA epigenetic-targeted editing as a strategy that could enhance therapeutic precision in advanced RCC. AAV9-mediated delivery of IDMDE targeting TRAF7 C1798 demethylation suppressed renal cell carcinoma progression in vivo. A Schematic of the AAV9-based delivery of the IDMDE system and targeted demethylation of renal cell carcinoma in vivo. B and C Representative tumor morphology in A-498 xenograft models across experimental cohorts: AAV-Ctrl, IDMDE-untreated (+ PBS) and treated (+ ABA) groups (n = 5/group). D Tumor volume of different treatment groups versus therapeutic process. E Excised tumor weight of different treatment groups. F The m5C level and (G) mRNA level of xenograft tumor tissues changed under different treatment conditions. H Representative images of H&E staining, Ki-67 and TRAF7 immunohistochemical staining demonstrating decreased Ki-67 and TRAF7 protein levels in IDMDE-treated (+ ABA) group subjected to AAV-Ctrl and IDMDE-untreated (+ PBS) groups. Scale bar = 50 μm. Data were presented as mean ± SD. Statistical analysis was performed using one-way ANOVA followed by Tukey’s multiple comparisons test (E–G), and two-way analysis of variance (D), defined as ns. not significant, and ***P < 0.001 Development of photo-responsive IDMME/IDMDE for spatiotemporal epigenetic editing Utilizing the optogenetic strategy, we integrated the IDMDE system with photocaging technology to develop a light-controllable chemically induced m5C editing platform that enables dual temporal and spatial control. The core design involved substituting free ABA with 4,5-dimethoxy-2-nitrobenzyl (DMNB)-caged ABA, which was cleaved via UV light to release bioactive ABA, thereby initiating the functional activation of the IDMDE system (Fig. 8A) [35]. ABA-DMNB was synthesized as previously reported and confirmed via LC–MS analysis (Fig. 8B). Design and validation of light-inducible m5C demethylation/methylation system. A Schematic of the photoactivatable m5C demethylation system. Photo-caged ABA-DMNB required UV irradiation to release active ABA, enabling recruitment of ABI-TET2CD for site-specific m5C demethylation. B Chemical structure of ABA-DMNB with molecular weight 459.4 Da, as determined by LC–MS analysis. C The demethylation effect of IDMDE system targeting TRAF7 C1798 under ABA-DMNB treatment for different UV irradiation time was evaluated by m5CRIP-qPCR. D-G m5C RIP-qPCR analysis of TRAF7 C1798 (D), PBX1 C6935 (E), HDGF C1052 (F), SLC7A11 C313 (G) methylation levels under ABA or ABA-DMNB treatment with/without UV irradiation. Three independent experiments were presented as mean ± SD (n = 3). Statistical analysis was performed using one-way ANOVA followed by Tukey’s multiple comparisons test, defined as ns. not significant, **P < 0.01, ***P < 0.001 To establish a framework for spatiotemporally controlled RNA editing, we first examined whether ABA-DMNB could confer spatial control over PYL-ABI interactions. A-498 cells co-expressing EGFP-PYL and NES-ABI-NES fusion constructs were treated with DMSO, free ABA, or ABA-DMNB with or without UV irradiation (Additional file 1: Fig. S8). In the absence of bioactive ABA (the DMSO and ABA-DMNB without UV groups), EGFP-PYL exhibited a diffuse distribution throughout both the cytoplasm and the nucleus. Upon the generation of bioactive ABA, either by direct addition (free ABA) or via the photo-uncaging of ABA-DMNB (+ UV), EGFP-PYL underwent rapid nuclear exclusion, demonstrating that UV irradiation triggered localized ABA release and subsequent PYL-ABI interaction (Additional file 1: Fig. S8). These results confirmed that ABA-DMNB enabled light-controlled activation of ABA signaling with spatial precision. Having established spatial controllability, we next characterized the temporal dynamics of this photocaged system. A UV exposure time-course experiment revealed that ABA-DMNB-mediated m5C editing was irradiation-dependent, with m5C levels decreasing progressively upon prolonged UV irradiation and reaching a minimum after 120 s of exposure (Fig. 8C). We further validated this photocaged system in both targeted m5C demethylation (ABI-TET2CD) and methylation (ABI-NSUN2-1) platforms. Upon UV irradiation of ABA-DMNB, the released bioactive ABA effectively activated both systems, inducing significant m5C demethylation at the TRAF7 and PBX1 loci (Fig. 8D, E) and m5C methylation at the HDGF and SLC7A11 loci (Fig. 8F, G), with efficiencies comparable to those achieved with free ABA treatment. These findings demonstrated that integrating the light-activatable ABA-DMNB enabled the construction of a photo-controllable m5C editing system, achieving enhanced spatiotemporal precision compared to chemical ligand-only approaches, while offering potential for dose-dependent regulation. Discussion Traditional approaches for manipulating RNA m5C methylation rely on altering the expression of upstream m5C writers or erasers [12, 36]. However, the global overexpression of methyltransferases can lead to non-specific m5C modifications across non-target mRNAs, rRNAs, and tRNAs, thereby disrupting the functional and mechanistic analysis of specific transcripts. These approaches also face significant challenges in elucidating the functions of individual m5C sites, which not only limits their utility as research tools but also raises concerns about their therapeutic safety. While subsequent efforts, such as those by Zhang et al., developed CRISPR-CasRx-based single-transcript editing tools for RNA m5C methylation, these systems remain limited by the absence of reversible control elements and suboptimal editing efficiency [21]. To address these limitations, we developed a precision RNA m5C editing platform by integrating dCasRx with CIP systems. This strategy spatially and temporally restricts NSUN2/TET2 activity to ABA-inducible states, enabling both chemical and optical control over m5C modification with high spatiotemporal precision. Furthermore, by optimizing the effector domains, our platform achieved site-specific modification with editing efficiencies exceeding previous benchmarks. Validation through m5CRIP-qPCR and m5C DRS-seq in PBX1-targeted RCC models demonstrated good specificity, with minimal detectable m5C alterations at homologous off-target transcripts. Taken together, this platform represents a substantial improvement over conventional approaches. The advent of high-throughput sequencing has unveiled over 170 distinct RNA modifications [37], with m6A being the most abundant and extensively studied. While other modifications, including m1A and m5C, exhibit lower stoichiometry, they play critical regulatory roles in gene expression. CRISPR-based systems, such as dCas13b/d fusion proteins guided by specific gRNAs, have enabled targeted m6A addition or removal [38, 39]. DNA-editing tools such as CRISPR-Cas9 and CRISPR-CasΦ have established well-developed technical platforms and are widely applied in bioengineering and therapeutic strategies [40]. The well-known Cas9 system has been optimized to recognize and act on single-stranded RNA by introducing short DNA oligonucleotide segments that bind to target RNA sequences, and targeted therapeutic tools for bladder cancer have been developed based on this approach [18]. Compared with the modified Cas9 RNA editing system, the Cas13d ribonuclease exhibits much higher targeting specificity with negligible off-target effects. Within the Cas13d family, CasRx is a smaller effector protein with highly specific cleavage activity. The catalytically inactive dCasRx effector can efficiently bind to RNA without altering mRNA or protein levels, making dCasRx fusion proteins a flexible RNA-binding module for targeting specific RNA sequences. However, dCasRx-based gene editing systems have been scarcely investigated for therapeutic applications. For instance, the cellular translation enhancement system formed by fusing dCasRx with eIF4G enables the treatment of kidney stones by targeting ferroptosis [15, 41]; CRISPR/dCasRx fused with SINEB2 inhibits bladder cancer growth and lung metastasis by efficiently enhancing the translation of tumor suppressor genes [16]; and the fusion of CasRx crRNA with an antisense hammerhead ribozyme achieves RNA knockdown to inhibit bladder cancer proliferation and metastasis [17]. Regarding RNA methylation, Xie et al. developed the REMOVER tool using dCas13b-ALKBH3 fusions to eliminate m1A methylation on MALAT1 transcripts [20]. Despite these advances, existing tools lack reversible control and temporal precision. To address this, light-sensitive proteins such as CIBN/CRY2PHR and chemically inducible systems such as ABA-PYL/ABI could be integrated with CRISPR dCas13 platforms for the spatiotemporal regulation of genetic and epigenetic editing. However, analogous tools for reversible m5C editing remain underdeveloped. Unlike constitutively active m5C editors like dCasRx-NSUN2 fusions, our platform introduced reversible and drug-dependent m5C modulation by coupling dCasRx with CIP systems. ABA-inducible NSUN2 recruitment enabled precise spatiotemporal control, overcoming the irreversibility inherent to traditional CRISPR-based editors. In RCC, our system demonstrated programmable functional versatility, as small-molecule-inducible m5C methylation of SLC7A11 transcripts regulated ferroptosis, whereas redesigning the gRNA and effector modules to target TRAF7 enabled m5C demethylation-dependent apoptosis. Crucially, CIP-dependent m5C editing validated by quantitative m5CRIP-qPCR and DRS-seq revealed that < 5% of enriched fragments exhibited off-target perturbations in IDMDE. While transcriptome-wide sequencing for IDMME was not conducted in this study, several lines of biochemical and structural evidence supported the conclusion that IDMME maintained low off-target effects. Literature indicates that the full-length NSUN2 possesses a PUA domain essential for RNA recognition and localization, with a dissociation constant Kd of approximately 340 nM [42]. By removing this domain, the RNA-binding affinity of the NSUN2-1 (residues 66–430) was significantly attenuated (with Kd increasing to 3.58 μM), thereby minimizing its ability to bind non-target RNA independently. In contrast, the ABA-induced interaction between PYL10 and ABI1 exhibits a much higher affinity (Kd ≈ 0.02 μM) [43]. This ensures that the recruitment of the catalytic core to the target site is predominantly driven by the dCasRx-gRNA complex rather than the intrinsic affinity of the methyltransferase itself. Furthermore, off-target effects mediated by gRNA sequence similarity were found to be negligible. Taken together, these results suggest that IDMME may exhibit only minimal non-specific background editing. In general, our architecture establishes a toolkit for reversible and site-restricted epigenetic engineering, achieving spatiotemporal precision with low off-target interference for RCC therapy. The efficacy and specificity of RNA m5C editors critically depend on two design principles according to our experimental findings: (1) minimizing non-catalytic domains to reduce steric hindrance and off-target interactions, and (2) enabling transient enzymatic recruitment to avoid persistent RNA–protein binding. To address the former challenge, we engineered dCasRx-fused methyltransferase/demethylase constructs that were trimmed to 20–50% of their original lengths while retaining the catalytic domains. This structural refinement significantly enhanced editing efficiency and reduced off-target methylation events. For the latter objective, our chemically inducible system leveraged ABA-dependent dimerization to transiently recruit NSUN2 to target RNAs, a strategy that contrasts with conventional CRISPR-based RNA editing tools often plagued by prolonged dCasRx-RNA complex accumulation. Time-controlled activation not only mitigated potential steric hindrance at the RNA sites but also aligned with emerging innovations in RNA modification engineering. For example, studies utilizing the PhyB-PIF3 photo-reversible dimerization system have demonstrated that pulsatile enzyme recruitment improves specificity by preventing protein aggregation [44]. Such spatiotemporal control is particularly advantageous in the context of tumors, where sustained artificial RNA–protein interactions could dysregulate endogenous RNA processing networks. The integration of the DMNB-caged ABA system significantly expands the utility of our m5C editing platform by introducing spatiotemporal control that is unattainable through global chemical induction alone. This photo-activation strategy provides several distinct biological and therapeutic advantages, including the ability to restrict editing activity to specific tissue regions or cell populations and the facilitation of investigation into the impact of rapid, transient RNA methylation dynamics on biological functions. Furthermore, while chemical induction relies on passive diffusion, photocaging offers the potential for higher precision through targeted illumination. Although our current validation remains in vitro, our preliminary experiments demonstrated that the UV intensity and irradiation duration within the working range did not induce significant cellular toxicity, and the DMNB-caged ABA compound itself exhibited low toxicity at functional concentrations under light exposure. However, several hurdles for the translation of this technology in vivo must be considered, including the limited tissue penetration of ultraviolet (UV) light, the potential toxicity of caging byproducts, and the risk of off-target photorelease due to caging instability. Future advancements to overcome these limitations may include the development of caging groups responsive to long-wavelength light, such as near-infrared (NIR), to enhance light penetration and safety, or the use of minimally invasive optical fiber devices as previously reported [45], for localized irradiation in deep tissues. Gene-editing therapy for RCC has not been widely applied to date, primarily due to the irreversible alterations made to genetic material. In contrast, our IDMME/IDMDE system targets RNA modifications and incorporates control modules, thus enabling the regulation of RCC cell fate with low toxicity and holding potential application value. In vitro validation using RCC cell models stably expressing IDMME/IDMDE confirmed that this system can alter cell fate by targeting of different genes, thereby achieving the goal of precise treatment for RCC. For instance, it can modulate the ferroptosis pathway by stabilizing the expression of SLC7A11 mRNA through the upregulation of m5C level. Alternatively, it can reduce the stability of TRAF7 mRNA by downregulating the TRAF7 m5C level, thereby inducing RCC cell apoptosis. These processes can be controlled by ABA treatment or light irradiation, enabling safe and controllable tumor therapy. To the best of our knowledge, existing RNA m5C editing strategies have been limited to in vitro systems, with no successful application to in vivo models reported prior to this study. Our study demonstrated spatiotemporally controlled m5C methylation editing in subcutaneous tumors through an ABA-inducible system. By coupling ABA-dependent regulation with targeted m5C modification of oncogenic transcripts, we achieved precise modulation of RNA methylation levels in vivo. This approach avoids the genotoxic risks associated with traditional DNA-centric genome editing systems like CRISPR-Cas9 while maintaining high controllability, which is a critical advantage for future clinical applications. Building on these results, we aim to extend this platform to transgenic in vivo models for non-oncogenic diseases, thereby broadening the scope of RNA epigenetic editing in therapeutic contexts. Conclusions Conclusively, we developed chemically inducible CRISPR-dCasRx systems, IDMME and IDMDE, which enable spatiotemporally controlled m5C methylation editing in RCC. By integrating ABA-mediated proximity induction with RNA-targeting CRISPR technology, this platform achieved efficient, precise, reversible, and time-dependent manipulation of m5C modifications at disease-relevant transcripts with negligible off-target effects. Methods Plasmid construction The dCasRx-based m5C editor was engineered by fusing either methyltransferase NSUN2 or demethylase TET2CD to the C-terminus of dCasRx (RRID:Addgene_109050) via a SV40 NLS (P-K-K-K-R-K-V)-A-A-A-HA (Y-P-Y-D-V-P-D-Y-A)-A-S-G-S-G linker. Catalytically inactive variants were generated by site-directed mutagenesis: dCasRx-NSUN2-1Mut (C271A/C321A) and dCasRx-TET2CD Mut (H1382Y/D1384V). All NSUN2 truncations and TET2CD constructs were synthesized and cloned into the dCasRx backbone. For chemical induction, PYL domains (RRID:Addgene_38247) were cloned and replaced NSUN2 position in the dCasRx vector (HA-tagged). ABI modules (RRID:Addgene_38247) were co-cloned with NSUN2-1 or TET2CD into a Myc-tagged pCDH backbone. Target-specific gRNAs were designed using NCBI BLAST (https://blast.ncbi.nlm.nih.gov) to minimize off-target binding in human transcripts. gRNA cassettes were inserted into the dCasRx scaffold via BbsI restriction sites. gRNA sequences were cataloged in Additional file 2: Table S1. Cell culture, transfection and treatment HEK-293T cells (ATCC Cat# CRL-3216, RRID:CVCL_0063) were cultured in Dulbecco's Modified Eagle Medium (Servicebio, #G4515), 769-P (ATCC Cat# CRL-1933, RRID:CVCL_1050) and A-498 (RRID:CVCL_1056) cell lines in RPMI-1640 (Servicebio, #G4535). All media contained 10% fetal bovine serum (Cellmax, #SA211.02) and 1% penicillin/streptomycin (Servicebio, #G4003), with incubation at 37 °C and 5% CO2. All cell lines were authenticated by short tandem repeat (STR) profiling. To prevent mycoplasma contamination, cells underwent prophylactic clearance with Plasmocin® treatment (InvivoGen, #ant-mpt) at 1/1000 concentration, followed by maintenance at 1/10,000. Mycoplasma-free status was confirmed pre-experiment using MycoBlue™ Mycoplasma Detector (Vazyme, #D101-02). Cells were seeded in 10 cm culture dishes and incubated overnight until reaching 70%−80% confluency. Transfection was performed using Lipofectamine 3000 with 20 µg plasmids (dCasRx-PYL: ABI-NSUN2/TET2: gRNA = 3:3:2), followed by 24 h of incubation. Afterward, 100 µM ABA (Aladdin, #A100492) or DMSO was added, and cells were further incubated for 24 h before collection for subsequent experiments. RT-qPCR Total RNA was extracted from the samples using Ultrapure RNA Kit (CoWin Biosciences, #CW0581M), following the manufacturer's protocol as previously described [18]. For reverse transcription, 1 µg of RNA from each sample was converted to cDNA using the RT SuperMix for qPCR with gDNA Eraser (Servicebio, #G3337). The resulting cDNA was then mixed with specific primers and 2 × Universal Blue SYBR Green qPCR Master Mix (Servicebio, #G3326). Quantitative PCR amplification was conducted on a LightCycler 480 (Roche, USA). All assays were performed in triplicate, and primers used for RT-qPCR were listed in Additional file 2: Table S2. Western blot Western blot analysis was performed as previously described [46]. Briefly, proteins were extracted using RIPA buffer (Beyotime, #P0013B) supplemented with protease inhibitors (Servicebio, #G2007&G2008) and separated by SDS-PAGE. After transferring to PVDF membranes (Millipore, #ISEQ00010/IPVH00010), blocking was carried out with 5% skim milk for 1 h at room temperature. Membranes were incubated overnight at 4 °C with the following primary antibodies: anti-HA-Tag (Cell Signaling Technology Cat# 3724, RRID:AB_1549585, 1:1000), anti-Myc-Tag (ABclonal Cat# AE070, RRID:AB_2863795, 1:10,000), anti-HDGF (Abclonal, A0589, 1:1000), anti-SLC7A11 (Abcam Cat# ab307601, RRID:AB_3094570, 1:1000), anti-KAT7 (Proteintech Cat# 13,751–1-AP, RRID:AB_2266703, 1:3000), anti-TRAF7 (Huabio, ER60690, 1:2000), anti-GAPDH (ABclonal Cat# AC002, RRID:AB_2736879, 1:10,000). Membranes were then probed with HRP-conjugated anti-mouse/rabbit secondary antibodies (Proteintech, #SA00001-1/SA00001-2, 1:2000) for 1 h at room temperature. Protein signals were detected using a ImageQuant™ 800 (Amersham, Sweden). Immunofluorescence assay HEK-293T cells were seeded and cultured on 20 mm diameter coverslips following transfection, and once adhered, were washed once with pre-chilled PBS before being fixed with 4% paraformaldehyde for 15 min. Following fixation, the cells were washed three times with PBS, permeabilized with PBS + 0.2% Triton-X100 (PBST) for 15 min at room temperature, and then blocked in 5% goat serum for 30 min. The cells were subsequently incubated overnight at 4 °C with HA-tag antibody (Cell Signaling Technology Cat# 3724, RRID:AB_1549585, 1:1000) and Myc-tag antibody (ABclonal Cat# AE070, RRID:AB_2863795, 1:200). On the following day, the cells were washed three times with PBST and incubated with fluorochrome-conjugated secondary antibodies at room temperature in the dark for 1 h, followed by DAPI staining for 15 min. Finally, the coverslips were mounted on slides with anti-fade reagent, and images were captured using a confocal laser scanning microscope. Co-immunoprecipitation (co-IP) The treated cells were lysed using IP lysis buffer, and the supernatant was collected for BCA quantification. Equal amounts of protein were incubated overnight at 4 °C with washed anti-HA magnetic beads (MCE, #HY-K0201) using a rotating mixer, while 10% of the protein lysate was saved as input. After washing the beads three times with wash buffer, 50 μL of 1 × SDS-PAGE loading buffer was added to the beads, which were then heated at 95 °C for 5 min. Following this, the beads were separated, and the supernatant was collected into a new EP tube for subsequent SDS-PAGE analysis. m5C RNA immunoprecipitation qPCR (m5CRIP-qPCR) A total of 100 µg of RNA was extracted and fragmented using Fragmentation Buffer at 94 °C for 3 min to generate ~ 200 nt fragments, and the fragmentation reaction was stopped with EDTA. The fragmented RNA was then extracted, and one-eighteenth of the sample was set aside as the input group. The remaining RNA was divided equally into IP and IgG groups. Each group was incubated overnight at 4 °C with 4 µg of IgG (BersinBio) or m5C antibody (Abcam Cat# ab214727, RRID:AB_2802117), followed by the addition of Protein A/G magnetic beads and incubation for 1 h at 4 °C on a vertical shaker. The RNA-antibody-bead complexes were digested in 200 µl of elution buffer at 55 °C for 30 min, followed by TRIzol-chloroform extraction and ethanol precipitation for purification. The RNA was then quantified by RT-qPCR. The immunoprecipitation (IP) enrichment ratio was calculated as the percentage of transcript in the IP relative to the input, with the IgG background subtracted. All experiments were performed in triplicate. Primers used in the m5C-RIP assay are listed in Additional file 2: Table S3. RIP-qPCR Equal amounts of IgG (ABclonal Cat# AC005, RRID:AB_2771930) and Myc-tag (ABclonal Cat# AE070, RRID:AB_2863795) antibodies were incubated with washed Protein A/G Agarose beads at 4 °C for 2 h. Ten million cells were lysed using Lysis Buffer, and 10% of the total lysate was saved as input. The remaining lysate was divided into two equal parts, each of which was incubated with antibody-pre-bound Protein A/G Agarose beads (IgG or Myc) at 4 °C overnight. Following incubation, the Protein A/G Agarose beads were washed gently four times with NT2 wash buffer, and 10% of the beads were used for protein elution and subsequent SDS-PAGE analysis. The remaining beads were incubated with elution buffer at 55 °C for 30 min to elute and purify the bound RNA, which was then analyzed using RT-qPCR analysis. Targeted bisulfite sequencing (TBS-seq) of RNA m5C methylation analysis To analyze RNA m5C modification, 1 μg of total RNA was subjected to bisulfite treatment using the EZ RNA Methylation™ Kit (Zymo Research, USA) according to the manufacturer's instructions. The bisulfite-converted RNA was eluted and reverse transcribed into cDNA. The resulting cDNA was then amplified using the high-fidelity PCR enzyme KOD -Multi & Epi-® (TOYOBO, Japan). After verification of the PCR products by agarose gel electrophoresis, the amplicons were submitted to the Hi-TOM platform for subsequent sequencing [47]. Primers used for TBS-seq were listed in Additional file 2: Table S4. Nanopore direct RNA sequencing (DRS-seq) DRS-seq was conducted by Seqhealth company (China). Polyadenylated RNA was isolated from total RNA (1 mg) using a GenElute™ mRNA Miniprep Kit (Sigma-Aldrich, #MRN70). Sequencing libraries were prepared following the Oxford Nanopore DRS protocol (SQK-RNA002) and loaded onto FLO-MIN106D flow cells for 72-h runs on a MinION device. Raw signals were base-called using Guppy (v3.1.5) and converted to single-read FAST5 files via the ont_fast5_api (v3.0.2). Reads were aligned to the reference genome of homo sapiens (GRCh38_release110) using Tombo (v1.5.1). For m5C detection, Tombo’s detect_modifications command was executed with the all-context alternate model (“-alternate-bases 5mC”). Tombo’s alternative_model algorithm compared raw signal deviations (± 5 bp windows) to a built-in reference model. Statistical scores were computed per read: reads with scores > 2.5 were classified as unmethylated (UM), < − 2.5 as methylated (M), and intermediate scores as unclassified (UC). Methylation scores were calculated based on the following equation: and final scores and coverage were extracted using Tombo’s text_output function. mRNA stability assay Cells were treated with 5 µg/ml actinomycin D (MCE, #HY-17559) post-transfection, and total RNA was extracted at specific time points (0, 2, 4, and 6 h). RT-qPCR was performed as previously described to calculate the mRNA half-life for assessing mRNA stability, with GAPDH used for normalization. All experiments were conducted in triplicate. CCK-8 assay A total of 3000 cells were resuspended in 200 µL of complete medium and seeded into a 96-well plate. After the cells adhered to the surface, the supernatant was discarded, and medium containing 10% CCK-8 (GlpBio, #GK10001) was added. The absorbance at 450 nm was measured after the cells were incubated in the dark for 2 h. This procedure was repeated every 24 h, with absorbance readings taken a total of four times throughout the experiment. Colony formation assay 500 cells were resuspended in 2 mL of complete medium, seeded into a 6-well plate, and incubated for 2 weeks. Once the majority of the colonies reached more than 50 cells, the cells were fixed with paraformaldehyde for 30 min, followed by staining with 0.1% crystal violet for 30 min. After several washes with PBS, images were captured to analyze the number of colonies. Transwell assay The Transwell insert (8 µm pore size, Biofil, China) was placed into a 24-well plate containing complete medium supplemented with 10% FBS, and 3 × 104 cells were resuspended in 200 µL of serum-free medium and added to the upper chamber. After 24 h of incubation, the insert was fixed in paraformaldehyde for 30 min, followed by staining with 0.1% crystal violet for 30 min. The number of migrated cells was then analyzed by capturing images of the insert under a microscope. Cell apoptosis assay The cell supernatant and cells were collected 48 h post-transfection after digestion with EDTA-free trypsin and washed once with cold PBS. The cells were then resuspended in pre-chilled 1 × binding buffer and incubated with 5 µL of Annexin V-FITC and 5 µL of PI (Servicebio, #G1511) at room temperature in the dark for 15 min. Apoptosis was assessed by flow cytometry using the CytoFlex LX (Beckman Coulter). Tumor xenografts Male BALB/c nude mice (4 weeks old) were purchased from Gempharmatech (Jiangsu, China) and were housed at the Laboratory Animal Research Center. All experiments involving animals adhered to the guidelines established by the Institutional Animal Care and Use Committee (IACUC) of Zhejiang Chinese Medical University (Reference Number: 20250331–31). To induce tumor formation, 5 × 10⁶ A-498 cells were resuspended in a mixture of 100 µL PBS and Matrigel (1:1 ratio) and subcutaneously injected into the axillary region of the mice. Once the tumor volume reached approximately 100 mm3, the tumor-bearing mice were randomly divided into three groups: AAV-Ctrl, AAV-dCasRx-PYL + ABI-TET2CD + PBS, and AAV-dCasRx-PYL + ABI-TET2CD + ABA. The AAV-dCasRx-PYL and AAV-ABI-TET2CD vectors were packaged into AAV serotype 9 viruses by Beijing Qingke Biotechnology Co., Ltd. Mice were intratumorally injected with 1 × 1011 vp AAV every 3 days for a total of two injections. One week after the first AAV injection, 100 μL of PBS or 100 μL of ABA was administered by intraperitoneal injection. At the end of the experiment, mice were euthanized, and subcutaneous tumor tissues were measured and collected for subsequent analysis. Tumor volume was calculated using the formula: Volume (mm3) = 0.5 × length × width2. Histopathological evaluation of tumor sections was performed using H&E, TRAF7, and Ki-67 staining. Meanwhile, blood samples were collected from each group of nude mice for serum biochemical analysis of liver (AST, ALT, GGT) and kidney (UREA, CREA, UA) function markers. Major organs (heart, liver, spleen, lungs, and kidneys) were harvested for H&E staining to evaluate the in vivo biosafety profile of ABA. Statistical analysis All experimental data are presented as the mean ± standard deviation (SD), based on at least three independent replicates, and statistical analysis was performed using GraphPad Prism 10. Statistical comparisons were made using two-tailed Student’s t-test, one-way analysis of variance (ANOVA), or two-way analysis of variance. Statistical significance was defined as *P < 0.05, **P < 0.01, and ***P < 0.001, with ns indicating no significant difference. Supplementary Information Additional file 1: Supplementary Figure S1-S8. Additional file 2: Supplementary Tables S1-S4. Sequences of gRNAs and primers for RT-qPCR, m5C-RIP-qPCR, and TBS-seq. Additional file 3. Uncropped gels for western blotting. Authors’ contributions Jinming Xu: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Validation, Visualization, Writing—original draft, Writing—review & editing. Jiaju Xu: Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Supervision, Visualization, Writing—original draft, Writing—review & editing. Congcong Cao: Methodology, Supervision. Yichang Shu: Data curation. Letian Shen: Data curation. Xuhong He: Methodology. Qi Huangfu: Resources. Chengfang Sun: Methodology. Weikai Wang: Data curation. Jingchao Wei: Funding acquisition. Ming Cai: Funding acquisition. Bohan Wang: Funding acquisition, Supervision. Aolin Li: Project administration, Supervision. Yuchen Liu: Methodology, Project administration, Supervision. Jiaming Wen: Funding acquisition, Project administration, Resources, Supervision, Writing—review & editing. All authors read and approved the final manuscript. Peer review information Chunlong Xu, Veronique van den Berghe and Wenjing She were the primary editors of this article and managed its editorial process and peer review in collaboration with the rest of the editorial team. The peer-review history is available in the online version of this article. Funding This work was supported by the National Key R&D Program of China (Grant No. 2021YFA0911600), Shenzhen Science and Technology Program (Grant No. RCJC20221008092723011), National Natural Science Foundation of China (Grant No. 82371630), Zhejiang Provincial Natural Science Foundation (Grant Nos. ZCLQN25H1606 and LZ26H160001), China Postdoctoral Science Foundation (Grant No. 2023M733083), National Natural Science Foundation of China (Grant Nos. 82370762, 82370750 and 82200850), Key Project of Science and Technology Department of Zhejiang Province (Grant No. 2022C03127), and Shenzhen Special Fund for medical research (Grant No. A2503021). Data availability The sequences of gRNAs and primers for RT-qPCR, m5C-RIP-qPCR, and TBS-seq are provided in Additional file 2, and the uncropped gels for western blotting are provided in Additional file 3. Details of the published datasets, repository information, and accession numbers/dataset DOIs are listed below. The raw sequence data reported in this paper have been deposited in the Genome Sequence Archive (Genomics, Proteomics & Bioinformatics 2025) in National Genomics Data Center (Nucleic Acids Res 2025), China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences (GSA-Human: HRA018841) that are publicly accessible at https://ngdc.cncb.ac.cn/gsa-human/browse/HRA018841 [48]. The plasmids generated in this study have been deposited in Addgene database (258,124, 258,125, 258,126, 258,127) [49]. The images for quantification in this study have been deposited on Figshare with 10.6084/m9.figshare.32408247 [50]. Declarations Ethics approval and consent to participate All the animal experiments were approved by the Experimental Animal Ethics and Ethics Committee of Zhejiang Chinese Medical University (approval No. 20250331–31; Date: 31 March 2025), operating in accordance with the ethical guidelines set by the International Council for Laboratory Animal Science (ICLAS). Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Jinming Xu, Jiaju Xu and Congcong Cao contributed equally to this work. Contributor Information Aolin Li, Email: ozai188@163.com. Yuchen Liu, Email: liuyuchenmdcg@163.com. Jiaming Wen, Email: wenjiaming@zju.edu.cn. References Associated Data Data Citations Supplementary Materials Additional file 1: Supplementary Figure S1-S8. Additional file 2: Supplementary Tables S1-S4. Sequences of gRNAs and primers for RT-qPCR, m5C-RIP-qPCR, and TBS-seq. Additional file 3. Uncropped gels for western blotting. Data Availability Statement The sequences of gRNAs and primers for RT-qPCR, m5C-RIP-qPCR, and TBS-seq are provided in Additional file 2, and the uncropped gels for western blotting are provided in Additional file 3. Details of the published datasets, repository information, and accession numbers/dataset DOIs are listed below. The raw sequence data reported in this paper have been deposited in the Genome Sequence Archive (Genomics, Proteomics & Bioinformatics 2025) in National Genomics Data Center (Nucleic Acids Res 2025), China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences (GSA-Human: HRA018841) that are publicly accessible at https://ngdc.cncb.ac.cn/gsa-human/browse/HRA018841 [48]. The plasmids generated in this study have been deposited in Addgene database (258,124, 258,125, 258,126, 258,127) [49]. The images for quantification in this study have been deposited on Figshare with 10.6084/m9.figshare.32408247 [50].