Co-profiling of in situ RNA-protein interactions and transcriptome in single cells and tissues

The human genome encodes at least 1,500 RNA-binding proteins (RBPs)^1,2, which are fundamental regulators of RNA biology, orchestrating processes such as splicing, localization, translation and degradation of mRNAs and numerous non-coding RNAs. RBPs are indispensable for critical physiological functions, including cellular differentiation, embryonic development, neuronal activity, immune surveillance and aging^3,4,5,6,7. Furthermore, disruptions in RBP–RNA interactions are increasingly being implicated in diseases like cancer, neurodegenerative disorders, cardiovascular diseases and metabolic syndromes⁸, rendering them promising therapeutic targets. For example, G3BP1 is aberrantly expressed in various cancers and has been shown to promote the proliferation and metastasis of cancer cells^9,10. To fully understand the regulatory roles of RBPs in both normal and disease conditions, it is essential to characterize dynamic RBP–RNA interactions across diverse contexts with high resolution at both the molecular and cellular levels.

RNA immunoprecipitation (RIP) and cross-linking immunoprecipitation (CLIP) are foundational techniques for studying RNA–protein interactions within cells. These techniques use specific antibodies to pull down an RBP along with its associated RNA targets^11,12,13. Advanced CLIP variants, such as iCLIP¹⁴ and eCLIP¹⁵, have improved the resolution of RBP binding sites, but they remain labor-intensive and prone to non-specific interactions in low-complexity libraries, and require large sample inputs. These limitations impede their use with low-input or rare samples and in large-scale parallel analyses. Recent innovations, such as antibody-directed reverse-transcription-based techniques (for example, RT&Tag¹⁶ and ARTR-seq¹⁷), have enabled the capture of RBP-interacting transcripts from low-input samples. However, they cannot achieve single-cell resolution or resolve isoform-specific targets. Moreover, none of the above methods allow parallel transcriptome analysis in the same cellular context, making it impossible to directly correlate RBP binding with regulation of gene expression.

Two recently developed techniques, TRIBE¹⁸ and STAMP¹⁹, employ ectopically expressed ADAR or APOBEC deaminase–RBP fusions to mark RNA targets without enrichment, preserving full transcriptome information. Although these IP-free techniques enable substrate discovery from low-input material, including single cells, their reliance on genetic manipulation limits their use in primary cells and clinical samples. Furthermore, ectopic fusion proteins might introduce artifacts or alter endogenous RBP function, and their lack of temporal resolution hinders the study of dynamic processes. Therefore, a versatile and user-friendly method is urgently needed to profile the transcriptome and RBP–RNA interactome with high temporal resolution, minimal input and compatibility with tissue samples.

In this study, we present MAPIT-seq, an approach that uses an antibody-targeted editing strategy to comprehensively profile the RBP–RNA interactome in situ alongside transcriptome data from the same fixed sample. MAPIT-seq is applicable for studying any RBP with a suitable antibody. We validate MAPIT-seq across RBPs with diverse functions, and reveal the weak RNA-binding abilities of PRC2 components in situ. Additionally, we investigate binding profiles of G3BP1 and define its regulatory role in perinatal neural development using mouse embryonic brain tissues. Furthermore, we optimize MAPIT-seq to achieve single-cell and isoform resolution. The paired single-cell transcriptome data allow us to reveal G3BP1’s cell-cycle-stage-specific function and its opposing regulatory effects on distinct target-gene subsets. In summary, MAPIT-seq is a versatile dual-omics platform, offering a unique opportunity that directly links RBP binding to gene-expression outcomes in tissue sections and single cells.

Design and optimization of MAPIT-seq

MAPIT-seq utilizes an antibody-directed RNA-editing strategy for in situ investigation of endogenous RBP targets (Fig. 1a). The key reagent of MAPIT-seq is a recombinant protein, produced in insect cells (Methods), that fuses protein A/G (pAG) with two RNA deaminases: the human ADAR2 deaminase domain with the E488Q substitution²⁰ (hADAR2dd) and rat APOBEC1 (ref. ¹⁹) (rAPOBEC1). Because hADAR2dd and rAPOBEC1 prefer editing on different RNA substrates, we incorporated both enzymes in MAPIT-seq to enhance the sensitivity and accuracy of target detection.

The MAPIT-seq workflow began with the fixation of cells using formaldehyde (FA), preserving dynamic and weak RBP–RNA interactions in their native contexts (Fig. 1a and Methods). Following cell permeabilization, samples were sequentially incubated with a primary antibody specific to the RBP of interest, and a secondary antibody to recruit the rAPOBEC1–pAG–hADAR2dd fusion protein to specific RBP-binding sites on RNA. After washing off the unbound fusion protein, the in situ RNA deamination began with Zn²⁺-containing buffer. The samples were then directly processed for RNA extraction, library preparation and sequencing. To identify RBP targets, we used an in-house analysis pipeline that includes a two-round unique mapping process, fine-tuning alignments, single-nucleotide variants (SNVs) calling and differential editing analysis (Extended Data Fig. 1a and Methods).

To initially assess the feasibility of MAPIT-seq, we generated HeLa cell lines expressing either FLAG or YTHDF2–FLAG, the latter being a well-characterized N⁶-methyladenosine (m⁶A) reader protein. We performed MAPIT-seq in these cells to quantify both A-to-G and C-to-U editing events near YTHDF2 PAR–CLIP peaks²¹. In YTHDF2–FLAG-expressing cells, a substantial increase of editing events was observed for the anti-FLAG MAPIT compared with the IgG control. By contrast, anti-FLAG-treated control cells exhibited editing levels comparable to those of the IgG control (Fig. 1b). Furthermore, the editing events were specifically enriched around GLORI-identified m⁶A sites²² that overlapped with PAR–CLIP peaks²¹, but not in random regions (Fig. 1c). Collectively, these results confirm that MAPIT-seq specifically and efficiently identifies YTHDF2-binding regions, with editing signals dependent on both the target RBP and specific antibody.

Next, we optimized several key parameters of MAPIT-seq in HeLa and HEK293T cells using endogenous G3BP1, an RBP with extensive published RNA-interactome data. We first screened a range of mild formaldehyde-fixation conditions and selected 0.5% formaldehyde for subsequent experiments, because it best preserved the transcriptome features of untreated samples²³ (Fig. 1d,e, Extended Data Fig. 1b,c and Supplementary Table 1) while exhibiting markedly higher editing activity than TRIBE-ID²⁴ (Extended Data Fig. 1d). Furthermore, among tested fusion protein configurations, rAPOBEC1–pAG–hADAR2dd was selected on the basis of its robust signals (Fig. 1f). The appropriate concentration range was determined to be 18–36 μg ml⁻¹ for this construct (Extended Data Fig. 1e and Methods). In addition, the inclusion of the secondary antibody was essential for optimal performance (Fig. 1b and Extended Data Fig. 1f). With these optimizations, we detected >80% of all potential editing events across all genes using 10 million reads (Extended Data Fig. 1g). For low-abundance RNAs (bottom third, <8.2 transcripts per million (TPM), average 3.60 TPM), reliable results were achieved at 12 million reads (Extended Data Fig. 1g,h). Biological replicates demonstrated high reproducibility in both transcriptome (R = 0.98) and editing (R = 0.94) profiles, underscoring the method’s robustness for dual-omics profiling (Extended Data Fig. 2a).

Validating MAPIT-seq through comparison with established methods

We then checked editing signals at the gene level and found abundant A-to-G and C-to-U editing events on several known G3BP1 targets identified by PAR-CLIP²⁵ in anti-G3BP1 samples, but not in IgG controls (Fig. 2a). To define G3BP1-interacting RNA targets transcriptome-wide, we applied a set of stringent criteria (Extended Data Fig. 1a, Supplementary Table 2 and Methods). In brief, we considered transcripts as RBP targets based on an editing fold enrichment > 2, MAPIT score > 0.5 and statistical significance. The editing fold enrichment and MAPIT score quantifies the ratio and the difference of the cumulative editing rate of all editing sites in each gene between anti-RBP–MAPIT and IgG–MAPIT samples. To minimize false negatives, we combined A-to-G and C-to-U edits to calculate the MAPIT score for each gene.

Reassuringly, the targets identified showed significant overlap with G3BP1 PAR–CLIP results²⁵ in HEK293T (Fig. 2b). Moreover, MAPIT-seq preferentially uncovers high-confidence targets (Fig. 2c). Here, the robustness of MAPIT-seq was demonstrated by the nearly identical editing patterns obtained from two distinct anti-G3BP1 antibodies (Fig. 2d and Supplementary Table 2). We also validated that the dual-editor approach substantially increased sensitivity, identifying approximately 20–30% more reported targets than did either single editor alone (Extended Data Fig. 2b,c and Supplementary Table 3). Consistently, the majority of G3BP1 targets measured by the dual editor were cross-validated by multiple independent approaches, including cross-linking and immunoprecipitation sequencing (CLIP-seq) and its variants, RIP-seq and HyperTRIBE^{24,25,26,27,28} (Extended Data Fig. 2c).

We then compared results from MAPIT-seq and CAP-seq²⁹, a proximity-dependent RNA-labeling method. We found that MAPIT scores were well correlated with fold enrichment determined by G3BP1 CAP-seq in HEK293T cells (Extended Data Fig. 2d). We also performed G3BP1 MAPIT-seq in HeLa cells and observed a significant overlap between the targets identified by our method and those identified by ARTR-seq¹⁷ (Extended Data Fig. 2e and Supplementary Table 4).

To validate MAPIT-seq’s specificity, we applied it in G3BP1-knockdown HeLa cells. Editing events around ARTR-seq peaks¹⁷ decreased markedly upon G3BP1 knockdown (Extended Data Fig. 2f,g and Supplementary Table 5). Additionally, MAPIT-seq-identified G3BP1 targets were significantly downregulated in G3BP1-knockdown cells compared with their levels in control cells (Extended Data Fig. 2h), suggesting that G3BP1 positively regulates RNA stability in HeLa cells. Altogether, these results highlight the reliability of MAPIT-seq in capturing genuine RBP–RNA interactions.

MAPIT-seq uncovers recognition motifs across multiple RBPs

After benchmarking MAPIT-seq against established methods, we evaluated its general applicability across multiple canonical RBPs, including PTBP1 (ref. ³⁰), SERBP1 (ref. ³¹), YTHDF2 (ref. ²¹), RBFOX2 (ref. ¹⁵) and PUM1 (refs. ^32,33). Consistent with observations for G3BP1, targets identified by MAPIT-seq significantly overlapped with those detected by CLIP-based methods^{15,21,30,31,32}, underscoring its broad applicability (Fig. 2e and Supplementary Tables 6–10). A recently developed method, in situ sensitive capture of RNA–protein interactions in biological environments (INSCRIBE)³⁴, uses an APOBEC1-fused nanobody guided by a primary antibody to label RBP targets. When applied to RBFOX2, MAPIT-seq exhibited substantially higher signal-to-noise ratios (SNRs) of editing events around eCLIP peaks¹⁵ than did INSCRIBE³⁴ (Fig. 2f). In addition, MAPIT-seq captured more validated RBFOX2 substrates (Extended Data Fig. 2i) and consistently mirrored the cellular transcriptome contexts (Extended Data Fig. 2j).

Next, we evaluated whether MAPIT-seq could effectively detect RBP binding motifs, focusing on PTBP1, YTHDF2, RBFOX2 and PUM1. We first utilized the flagging areas of RNA-editing enrichment (FLARE)³⁵ pipeline to identify high-confidence ‘edit clusters’ marked by both C-to-U and A-to-G editing events (Methods). Using 43,951 PTBP1–MAPIT edit clusters, motif analysis by HOMER³⁶ revealed significant enrichment of known CU-rich motifs (Fig. 2g; Supplementary Table 11 shows top ten motifs across various lengths). Transcripts containing more CU-rich sequences exhibited higher MAPIT scores (Extended Data Fig. 3a), suggesting that the MAPIT score could potentially quantify RBP–RNA interaction strength. Similarly, we found enrichment of the conserved GGAC motif for the endogenous YTHDF2 (ref. ²¹), the canonical UGCAUG motif for RBFOX2 (ref. ¹⁵) and the pumilio recognition element (PRE) motif UGUANAUA for PUM1 (ref. ³²) (Fig. 2g and Supplementary Table 12). Notably, MAPIT-seq edit clusters consistently showed two- to sixfold enrichment of consensus sequences over permuted control clusters in the same genes (Extended Data Fig. 3b–e), comparable to the performance of STAMP or CLIP. These data clearly demonstrate that MAPIT-seq reliably identifies RBP-binding motifs.

We further selected RBFOX2 and PUM1, two RBPs with well-defined recognition sequences, for an in-depth evaluation of MAPIT-seq resolution. Edit clusters of RBFOX2–MAPIT converged into a distinct peak within a 200-nucleotide (nt) window flanking the canonical UGCAUG motif (Fig. 2h), demonstrating a sharper enrichment than that uncovered by STAMP¹⁹. Of note, a substantial portion of RBFOX2–MAPIT edit clusters localized closely around the UGCAUG motif and aligned with eCLIP peaks¹⁵ (Extended Data Fig. 3f–h). Furthermore, MAPIT-seq edit clusters accurately pinpointed eCLIP-identified RBFOX2-binding regions, centering around canonical motifs on known RBFOX2 targets such as SNX12, HMGA2 and KLHL20 (Fig. 2i). Similar results were observed for PUM1–MAPIT: edit clusters concentrated within 100 nt around either side of the core PRE motif UGUANA, and profiles properly resembled those from CLIP-seq³² (Extended Data Fig. 4a–e). In summary, although single-nucleotide resolution remains unattainable, MAPIT-seq reliably delineates sequence-specific binding regions of RBPs, with performance comparable to that of STAMP^19,33 although still below CLIP-seq resolution.

Evaluating functional relevance of MAPIT-seq editing signals

Next, we characterized the genomic distribution of edit clusters from five MAPIT-seq libraries above to capture RBP-specific binding patterns (Supplementary Tables 13–17). As expected for canonical splicing factors RBFOX2 and PTBP1 (ref. ¹⁷), MAPIT-seq edit clusters were predominantly enriched in introns (83.5% for RBFOX2 and 88.4% for PTBP1) (Fig. 2j and Extended Data Fig. 5a). We further generated RNA maps by analyzing the distribution of editing signals around exons regulated upon knockdown of these RBPs. For RBFOX2, edits were markedly enriched in introns immediately downstream of exons inactivated upon RBFOX2 knockdown, compared with levels in constitutive exons (Extended Data Fig. 5b). For PTBP1, edits predominantly localized near exons activated upon its knockdown (Extended Data Fig. 5b). These patterns closely align with published CLIP data^15,37, supporting their position-dependent regulatory models^38,39.

For mRNA-associated RBPs (YTHDF2, PUM1 and SERBP1)^17,31,32, MAPIT-seq edit clusters were mainly localized within the 3′ untranslated region (3′ UTR) and coding sequence (CDS) (Extended Data Fig. 5a). Consistent with YTHDF2’s role as an m⁶A reader, YTHDF2–MAPIT scores strongly correlated with m⁶A content determined by GLORI²² (Extended Data Fig. 5c). Furthermore, metagene analysis revealed prominent editing enrichment near stop codons, mirroring the reported m⁶A distribution on mRNAs¹⁷ (Extended Data Fig. 5d).

In summary, these analyses demonstrate that MAPIT-seq faithfully captures RBP-binding patterns aligned with known functions, underscoring its capability to reveal functionally relevant RBP–RNA interactions across diverse contexts.

Reevaluating PRC2–RNA interactions using MAPIT-seq

PRC2, a histone methyltransferase catalyzing the trimethylation of histone H3 on lysine 27 (H3K27me3), has been suggested to interact with X inactive specific transcript (XIST) and other long non-coding RNAs (lncRNAs)⁴⁰, suggesting that these RNAs have a broad role in chromatin regulation. However, a recent study has presented contradictory evidence, challenging the notion of PRC2–RNA binding in cells³⁰. Our MAPIT-seq approach offers an alternative method to address this controversy, because it detects RBP–RNA interactions in situ, minimizing the artifacts typically introduced during cell lysis. We performed MAPIT-seq on three core components⁴¹ of PRC2 (EED, EZH2 and SUZ12), along with CHTOP, a chromatin-associated protein with known roles in RNA processing and RNA export⁴², and PTBP1 (ref. ³⁹). As expected, CHTOP and PTBP1 displayed strong proximity to numerous RNAs (Fig. 3a,b and Supplementary Tables 18–21). By contrast, MAPIT-seq identified only one RNA, XIST, consistently located in proximity to EED, EZH2 and SUZ12 (Fig. 3a,c). These findings suggest that PRC2 components might generally have no RNA-binding abilities in cells; however, certain RNAs, such as XIST, might still interact with PRC2 under specific conditions^40,43, potentially serving important biological functions.

Profiling the G3BP1–RNA interactome in embryonic mouse brain

Existing methods for profiling RBP–RNA interactomes are inadequate for studying rare tissues and patient samples, particularly because they fail to capture transcriptome data simultaneously. Here, we tested MAPIT-seq in mouse tissues by conducting G3BP1 MAPIT-seq on both fixed and fresh frozen sections of mouse embryonic day 10.5 (E10.5) embryos, with a slightly modified protocol (Extended Data Fig. 6a, Supplementary Table 22 and Methods). Editing events were enriched in fresh frozen but not fixed frozen tissues (Extended Data Fig. 6b). Additionally, transcriptome and editing profiles derived from fresh frozen samples showed strong correlation between two continuous sections from the same embryo (Extended Data Fig. 6c,d). In accordance, transcriptome profiles generated by MAPIT-seq closely matched RNA sequencing (RNA-seq) results from untreated embryo sections (Extended Data Fig. 6d). On the basis of these findings, we decided to use fresh frozen sections for subsequent experiments.

Although G3BP1 is ubiquitously expressed across the entire embryo⁴⁴, we were surprised to find that G3BP1-bound RNAs were predominantly related to neuronal growth and organization in E10.5 embryo sections (Fig. 4a and Supplementary Table 23). These results align strongly with previous functional studies in G3bp1^−/− mouse embryos⁴⁴, which showed severe defects in the brain but not other organs. We then extended MAPIT-seq to fresh frozen sections of embryonic mouse brains at E12.5 and E16.5 (Supplementary Table 22). Transcriptome principal component analysis (PCA) of MAPIT-seq demonstrated high reproducibility and clearly separated two developmental stages (Extended Data Fig. 6d). Further gene expression analysis confirmed that the neurogenic-to-gliogenic transition occurred between E12.5 and E16.5 (ref. ⁴⁵), as the expression of radial glia lineage genes (for example, Shh and Nes) gradually reduced while the expression of oligodendrocyte precursor cell lineage (for example, Oligo1 and Pdgfra), committed OPC lineage (for example, Neu4 and Mag) and glioblast lineage (for example, Tnc and Pla2g7) genes increased from E12.5 to E16.5 (Fig. 4b).

We then quantified the number of editing events and found enrichment in G3BP1–MAPIT compared with IgG–MAPIT in mouse brain sections (Fig. 4c), albeit with a lower SNR than that in cultured cells (Extended Data Fig. 1e). This discrepancy was presumably due to flash-freezing effects on RBP–RNA interactions or inherent tissue heterogeneity. Moreover, G3BP1–MAPIT uncovered specific edits on previously identified targets microtubule-associated protein tau (Mapt)⁴⁶ and Cadm2 (ref. ⁴⁷) (Fig. 4d). Overall, we identified 395 and 925 G3BP1 targets in E12.5 and E16.5 mouse brain sections, with 82 shared genes (Fig. 4e and Supplementary Table 24). These targets also significantly enriched G3BP1-binding motifs determined by HITS–CLIP⁴⁷ (Extended Data Fig. 6e), confirming that MAPIT-seq retains its specificity for detecting RBP binding in tissue sections.

A previous study on G3bp1^−/− mice found that G3BP1 promoted expression of several selected genes at E12.5 but inhibited the expression of the same targets at E15.5 and E17.5 (ref. ⁴⁴). This raises an intriguing hypothesis that G3BP1 could have distinct roles in regulating mRNA turnover at different developmental stages. Consistent with this hypothesis, targets with stronger G3BP1 binding showed increased RNA abundance at E12.5, whereas the opposite trend was observed at E16.5 (Extended Data Fig. 6f). Furthermore, targets from the E12.5 brain were enriched for terms such as axonogenesis and glial-cell differentiation, whereas those from the E16.5 brain were associated with terms such as dendrite development and neuron projection organization (Fig. 4f and Extended Data Fig. 6g). Notably, pathways related to synapse organization, structure and activity were enriched among G3BP1 targets at both developmental stages, aligning with the synaptic plasticity defects in G3bp1^−/− mouse brains⁴⁸ (Fig. 4f and Extended Data Fig. 6g). Together, these results demonstrate that MAPIT-seq can be effectively applied to tissue sections and uncover the temporal dynamics of RBP function in vivo.

Development and optimization of scMAPIT-seq

Co-profiling of RBP–RNA interactome and the transcriptome at single-cell resolution in non-genetically-engineered systems is highly valuable for dissecting cell-type- and cell-stage-specific RBP regulation under physiologically relevant conditions. However, such a method is still lacking. Here, we assessed whether MAPIT-seq can detect in situ RBP–RNA interactome at the single-cell level. First, G3BP1–MAPIT libraries, generated from 500 to 50,000 HeLa cells, all yielded reliable editing events (Extended Data Fig. 7a). We combined MAPIT-seq with fixed and recovered intact single-cell RNA (FRISCR)⁴⁹ to perform plate-based scMAPIT-seq of G3BP1. A focused examination of known G3BP1 targets revealed clear editing signals, both in individual cells and in an aggregation of 16 cells (Fig. 5a), demonstrating MAPIT-seq’s ability to identify RBP targets in individual cells. Notably, more edited sites were observed in scMAPIT-seq than in bulk MAPIT-seq (Fig. 5a).

A major hurdle in increasing throughput for scMAPIT-seq was the formaldehyde fixation process, which entails a complicated reverse cross-linking step, making it incompatible with most high-throughput single-cell RNA-seq (scRNA-seq) platforms. To overcome this issue, we tested two alternative fixatives compatible with scRNA-seq: methanol and dithiobis (succinimidyl propionate) (DSP). Gene expression profiles from methanol- and DSP-fixed G3BP1–MAPIT libraries showed high correlations with untreated and formaldehyde-fixed samples (Extended Data Fig. 7b). Notably, DSP–MAPIT yielded a substantially higher SNR around PAR–CLIP peaks²⁵ than did both methanol–MAPIT and formaldehyde–MAPIT (Extended Data Fig. 7c). Consistently, G3BP1 targets identified by DSP-MAPIT were similar to those identified by formaldehyde-based MAPIT and PAR-CLIP²⁵ (Extended Data Fig. 7d and Supplementary Table 25). Furthermore, plate-based scMAPIT-seq with DSP fixation successfully preserved transcriptome integrity in single cells (Extended Data Fig. 7e). Encouraged by these results, we chose DSP fixation for high-throughput scMAPIT-seq.

Next, we combined scMAPIT-seq with the 10x Genomics single-cell workflow (Methods). In total, we captured 3,400 G3BP1–MAPIT cells and 3,945 IgG–MAPIT cells after quality control (Supplementary Table 26). These two samples showed no differences in the number of genes and counts, and exhibited highly similar gene expression profiles, yet were completely separated by their editing profiles (Extended Data Fig. 7f,g). Dual editing events were enriched on G3BP1 targets in both the single-cell and the aggregated G3BP1–MAPIT dataset relative to IgG–MAPIT (Fig. 5b,c). Furthermore, aggregating signals across all individual cells revealed that scMAPIT-seq achieved sensitivity comparable to that of bulk MAPIT-seq and specifically marked down G3BP1-binding regions, showing strong concordance with ARTR-seq data¹⁷ (Fig. 5c,d and Supplementary Table 27). These data support high-throughput scMAPIT-seq’s ability to effectively uncover RBP targets and profile gene expression concurrently at the single-cell level.

Cell-cycle-phase-specific regulation of G3BP1

Leveraging the dual-omics capability of scMAPIT-seq, we next identified cell-cycle-phase-specific G3BP1 targets. Cells were annotated as being in the G1, S or G2/M phase on the basis of their transcriptomic profiles (Fig. 5e). Despite substantial variations in G3BP1 targets across these phases (Fig. 5f), we observed conserved roles for these targets in essential processes throughout the cell cycle, such as RNA splicing, localization and ribosome biogenesis (Extended Data Fig. 8a).

To further explore the temporal dynamics of G3BP1–RNA interactions, we focused on differential interactions across the cell cycle phases. Targets with differential interactions were categorized into four distinct clusters (Extended Data Fig. 8b). Notably, the majority of these differentially interacting G3BP1 targets (61.1%, 916/1,499) exhibited peak binding during the G2/M phase and were enriched in pathways closely related to mitosis (Extended Data Fig. 8b,c). In addition, targets with the highest G3BP1 binding strength in the G1 and S phases were enriched in pathways potentially associated with corresponding cell cycle progression (Extended Data Fig. 8c).

To understand how G3BP1 regulates its target expression during the cell cycle, we calculated Pearson’s correlation coefficients between G3BP1 binding strength and RNA abundance across different phases (Supplementary Table 28). This analysis identified 141 genes with significant positive correlations and 94 with significant negative correlations between G3BP1 binding strength and target expression throughout the cell cycle (Fig. 5g and Extended Data Fig. 8d). To validate the functional impact of G3BP1 binding, we examined the expression of these targets in G3BP1-knockdown HeLa cells. Positively correlated targets were significantly downregulated, while negatively correlated targets were significantly upregulated following G3BP1 knockdown (Fig. 5h and Extended Data Fig. 2d), supporting a regulatory role for G3BP1 in controlling the expression of these genes during cell cycle progression.

Positively correlated targets were primarily enriched in processes related to mitotic nuclear division and chromosome segregation, exemplified by genes such as CENPA, CENPK, SGO1 and SGO2 (Fig. 5i and Extended Data Fig. 8e–g). Notably, no specific Gene Ontology (GO) terms were associated with negatively correlated targets. To further investigate the mechanisms underlying G3BP1’s differential regulation, we analyzed previously reported regulatory elements, including AU-rich elements (ARE)⁵⁰ and m⁶A modifications²². We found that G3BP1 targets exhibited significantly higher ARE content but lower m⁶A content than did non-targets (Fig. 5j,k). Additionally, positively correlated targets generally showed higher ARE and m⁶A levels among all G3BP1 targets (Fig. 5j,k). By contrast, there was no evidence linking ARE or m⁶A content with G3BP1-mediated negative regulation (Fig. 5j,k).

Together, these data demonstrate the value of scMAPIT-seq in uncovering cell-state-specific RBP–RNA interactions in mixed cell populations. Additionally, our results reveal G3BP1’s opposing regulatory roles in distinct target groups, along with insights into potential underlying mechanisms.

Long-read MAPIT-seq reveals isoform-specific G3BP1 binding

Current RBP interactome profiling methods have mostly failed to capture isoform-specific RBP interactions. To bridge this gap, we aimed to enhance the resolution of MAPIT-seq to identify RBP binding specific to individual transcript isoforms. We subjected both formaldehyde- and DSP-fixed MAPIT-seq samples to the PacBio HiFi sequencing platform. DSP-G3BP1-MAPIT samples yielded longer read lengths (Extended Data Fig. 9a) and higher editing signals compared to formaldehyde fixed samples (Extended Data Fig. 9b), leading us to proceed with DSP-MAPIT data for further analysis.

When comparing long-read MAPIT-seq with short-read MAPIT-seq, we observed strong concordance in target gene identification between the two datasets (Fig. 6a, Extended Data Fig. 9c and Supplementary Table 29). A closer examination revealed that G3BP1 exhibited isoform-specific binding across several genes, including YTHDF2, GDAP2 and ING3 (Fig. 6b and Extended Data Fig. 9d). Notably, these isoform-specific interactions were consistently confirmed through RIP–quantitative polymerase chain reaction (qPCR) validation (Fig. 6c).

To understand the mechanisms underlying G3BP1’s differential binding to isoforms of the same gene, we analyzed the molecular characteristics of isoforms with strong versus weak G3BP1 binding. We found that protein-coding isoforms exhibited significantly stronger G3BP1 binding compared with non-coding RNA and intron-retained isoforms (Fig. 6d). Among the protein-coding isoforms, we found a positive correlation between isoform length and G3BP1 binding (Fig. 6e), largely driven by differences in the 3′ UTR and CDS (Fig. 6f). These findings highlight MAPIT-seq’s ability to dissect isoform-specific RBP interactions, shedding light on the molecular features that drive G3BP1’s selective binding to particular isoforms.

In this study, we introduce MAPIT-seq, a dual-omics profiling platform, and demonstrate its effectiveness in characterizing a broad range of RBPs in mammalian cells. Compared with existing methods, MAPIT-seq offers several advantages (Extended Data Fig. 10), making it a powerful tool for investigating post-transcriptional regulation across diverse biological contexts. Because MAPIT-seq performs deamination in fixed cells, it enables high temporal resolution, providing detailed insights into dynamic RBP–RNA interactions over time. Its adaptability for small tissue samples, such as mouse embryonic brains, makes it particularly valuable for investigating RBP regulation mechanisms in rare clinic samples. Additionally, its streamlined, time-efficient and scalable protocol supports parallel analysis of multiple RBPs across various samples, paving the way for large-scale investigations in clinical settings.

Investigating the RBP–RNA interactome at single-cell resolution has long been considered a difficult task¹⁷, let alone performing concurrent transcriptome profiling. LACE-seq provides single-cell RNA binding data, but it operates at low throughput and does not provide concurrent transcriptome information⁵¹. Although STAMP¹⁹ and agoTRIBE⁵² enable high-throughput identification of RBP–RNA interactions at single-cell resolution, both methods rely on ectopic expression of the deaminase–RBP fusion protein, restricting their applicability in primary cells or tissues and limiting temporal resolution. By contrast, MAPIT-seq overcomes these limitations, allowing high-throughput single-cell profiling of in situ RBP targets alongside the transcriptome. By integrating single-cell transcriptome information, scMAPIT-seq can precisely identify RBP-binding preferences across distinct cell subpopulations. This integration further enables the analysis of correlations between RBP binding and RNA abundance during dynamic processes, such as differentiation and stress responses. Future efforts should prioritize the application of scMAPIT-seq to tissue samples, facilitating comprehensive interrogation of cell-type-specific RBP functions during development and disease progression.

A recently published method, INSCRIBE, has also achieved in situ RNA editing in non-genetically-modified cells³⁴. However, MAPIT-seq presents several potential advantages over INSCRIBE. First, it employs two distinct classes of deaminases, thereby minimizing potential substrate biases. Second, it consistently exhibits a higher SNR than does INSCRIBE when targeting the same RBP in the same cell type. Although the exact reasons remain unclear, possible contributing factors include the use of secondary antibodies in MAPIT-seq, potentially increasing local enzyme concentrations, and stringent washing conditions (300 mM salt) that minimize non-specific interactions. Third, MAPIT-seq is compatible with a broad range of commercially available antibodies, whereas INSCRIBE requires customized nanobodies specifically engineered for each antibody³⁴. Fourth, MAPIT-seq features a shorter in vitro deamination step (3 h at 30 °C) than does INSCRIBE (16 h at 37 °C)³⁴, potentially better preserving RNA integrity. Finally, MAPIT-seq successfully enables single-cell resolution and simultaneous co-profiling of the transcriptome and RBP–RNA interactome, which have not yet been achieved by INSCRIBE³⁴.

Collectively, these strengths establish MAPIT-seq as a versatile and robust dual-omics tool capable of integrating post-transcriptional regulation with gene expression at single-cell resolution.

Ethics statement

All animal experiments were approved by Institutional Animal Care and Use Committee (IACUC) of Peking University, which are accredited by the AAALAC (Association for Assessment and Accreditation of Laboratory Animal Care International).

Plasmid construction

For ectopic expression of FLAG–YTHDF2, the complementary DNA encoding human full-length YTHDF2 was cloned into the piggyBac vector with an amino-terminal triple FLAG-tag (3×FLAG). An empty vector with 3×FLAG was also constructed as a control. For recombinant deaminase expression, genes encoding rAPOBEC1 and eukaryotic codon-optimized protein A/G (pAG) were synthesized by Tsingke. The cDNA for the hADAR2 deaminase domain was cloned and mutated (E488Q) through PCR-based site-directed mutagenesis. These fragments, XTEN linkers, and a 6×His tag were assembled into the pFastbac backbone with an N-terminal twin-strep tag for purification, using the ClonExpress Ultra One Step Cloning Kit (C115, Vazyme). Four constructs were generated by combining pAG with two deaminases: pFastBac-rAPOBEC1-pAG-hADAR2dd, pFastBac-rAPOBEC1-hADAR2dd-pAG, pFastBac-rAPOBEC1-pAG, and pFastBac-hADAR2dd-pAG (sequences in Supplementary Table 30).

Cell culture

HeLa (CCL-2, ATCC) and HEK293T (CRL-3216, ATCC) cells were cultured at 37 °C with 5% CO₂ in high-glucose Dulbecco’s modified Eagle’s medium (DMEM, SH30243.01, Hyclone) supplemented with 10% (vol/vol) FBS (900-108, GeminiBio) and 1% (vol/vol) penicillin–streptomycin (15140163, Gibco). The insect cells Spodoptera frugiperda (Sf21, B821-01, Invitrogen) and Trichoplusia ni (Hi5, B855-02, Invitrogen) were cultured in a non-humidified shaker at 27 °C, 110 r.p.m. in SIM-SF (MSF1, Sino Biological) and SIM-HF medium (MHF1, Sino Biological).

Cell transfection

Stable HeLa cell lines expressing 3×FLAG–YTHDF2 or 3×FLAG were generated by transfecting 1 × 10⁵ cells at 60–80% confluency with 0.2 µg pBase and 0.8 µg piggyBac plasmids using jetPRIME transfection reagent (468 PT-114-75, Polyplus-transfection) and culturing for 2 days, followed by 130 µg ml⁻¹ hygromycin B (cat. no. 10843555001, Roche) selection for one week. For G3BP1 knockdown, 2 × 10⁵ HeLa cells were plated in a well of 6-well plate and cultured for 12 h to reach 50% confluency before transfection. Small interfering RNAs (siRNAs) were synthesized by GenePharma. Cells were transfected with 50 nM control (5′-UUCUCCGAACGUGUCACGUTT-3′), which has no homology with mammalian genes, or G3BP1-specific (5′-UCAACAUGGCGAAUCUUGGTG-3′) siRNAs using jetPRIME. Three days after transfection, cells were collected for western blotting and MAPIT-seq experiments. To knock down RBFOX2 and PTBP1, 1 x 10⁵ HEK293T cells were plated in a well of 12-well plate and transfected with siRNAs from Hippobio. The sequence for RBFOX2 siRNA is 5′-CGGGUUCGUAACUUUCGAGAAdTdT-3′. The sequence for PTBP1 siRNA is 5′-GCGUGAAGAUCCUGUUCAAUAdTdT-3′. Cells were collected for RNA-seq 2 days after transfection.

Expression and purification of deaminase–pAG fusion proteins

Deaminase–pAG fusion proteins rAPOBEC1–pAG–hADAR2dd, rAPOBEC1–hADAR2dd-pAG, rAPOBEC1–pAG and hADAR2dd–pAG were expressed and purified using a baculovirus expression system. The recombinant pFastBac plasmids were transformed into DH10Bac (Biomed, BC112) to generate bacmids. Bacmid DNA was transfected into Sf21 cells using X-tremeGENE HP DNA Transfection Reagent (06366236001, Roche) and incubated for 4 days. Low-titer recombinant baculoviruses were collected and amplified in Sf21 cells to produce high-titer viruses. Next, the baculoviruses were used to infect Hi5 cells at a 1:200 ratio and cultured for 48–60 h. Cells were lysed in lysis buffer containing 50 mM Tris-HCl, pH 8.0 and 200 mM NaCl. Lysates were sonicated at 4 °C with SCIENTZ sonicator set to a 7 s on and 5 s off cycle at 300 W for 20 min. After sonication, the lysates were centrifuged at 16,000 g, 4 °C, for 40 min. The deaminase–pAG fusion proteins were then purified using Strep-Tactin beads (SA053, Smart Lifesciences). In brief, the supernatant was incubated with Strep-Tactin beads at 4 °C for 1 h. After washing the beads with 30 volumes of lysis buffer, the deaminase–pAG fusion proteins were eluted with lysis buffer containing 5 mM desthiobiotin. Further purification was performed by a Superdex 200 Increase column (GE Healthcare). Deaminase–pAG fusion proteins were eluted in a buffer containing 25 mM HEPES, pH 7.5, and 150 mM NaCl and then supplemented with 5% glycerol. The aliquots of deaminase–pAG fusion proteins were snap-frozen in liquid nitrogen and stored at –80 °C.

Antibodies

Anti-G3BP1 (1:1,000, 13057-2-AP, Proteintech), anti-GAPDH (1:1,000, ET1601-4, Huabio) antibodies and donkey anti-rabbit IgG secondary antibody (1:10,000, 926–32213, LI-COR) were used for western blot. For MAPIT-seq, the following primary antibodies were used: Anti-Flag (1:100, F1804, Sigma), mouse IgG (1:40, sc-2025, Santa Cruz Biotech), rabbit IgG (1:100, 2729S, CST), anti-G3BP1 (1:100, ab56574, Abcam), anti-G3BP1 (1:100, 13057-2-AP, Proteintech), anti-PTBP1 (1:100, MABE986, Millipore), anti-YTHDF2 (1:50, 24744-1-AP, Proteintech), anti-SERBP1 (1:100, A303-938A, BETHYL), anti-RBFOX2 (1:100, A300-864A, BETHYL), anti-PUM1 (1:40, ab92545, Abcam), anti-EED (1:80, 85322, CST), anti-EZH2 (1:80, 5246, CST), anti-SUZ12 (1:80, 3737, CST) and anti-CHTOP (1:80, Invitrogen, PA544307). The following secondary antibodies were used: rabbit anti-mouse (1:100, ab46450, Abcam) and guinea pig anti-rabbit (1:100, ABIN101961, Antibodies Online). For RIP–qPCR, mouse IgG (1:20, sc-2025, Santa Cruz Biotech) and anti-G3BP1 (1:50, ab56574, Abcam) were used.

Western blotting analysis

For western blot, 5 × 10⁵ HeLa cells transfected with siRNA were collected, and G3BP1 and GAPDH were detected using respective antibodies at 4 °C overnight, with secondary antibodies at room temperature (RT) for 1 h. The membranes were imaged using Odyssey (LI-COR), and relative protein levels were quantified using ImageJ (v1.49).

MAPIT-seq

A full step-by-step protocol for MAPIT-seq has been deposited in the protocols.io repository⁵³.

For cell fixation and ConA-binding, 5×10²–5×10⁵ HeLa or HEK293T cells were trypsinized and washed to create single-cell suspensions. Cells were fixed with 0.5% formaldehyde at RT for 5 min and then quenched with 125 mM glycine at RT for 5 min. Cells underwent three rounds of DPBS washes before cells being centrifuged at 450g and 4 °C for 5 min. Cells were then incubated with 10 μl concanavalin A-coated beads (ConA beads, BP531, Bangs Laboratories) on a roller at RT for 15 min. Samples were then placed on a magnetic stand to remove the supernatant and washed twice. Afterwards, cells were transferred to a 0.2-ml tube prewashed with PBS supplemented with 1% BSA to minimize cell loss.

For cell permeabilization and incubation with antibodies and deaminases, cell-coupled ConA beads were resuspended in 50 μl antibody buffer 1 (1 mM PMSF, 1×protease inhibitor cocktail, 1 U μl⁻¹ RiboLock RNase Inhibitor (EO0382, ThermoFisher), 0.01% Digitonin, 1 mM DTT, 2 mM EDTA and 0.1% BSA in DPBS) pre-mixed with primary antibodies (1:100) and incubated with rotation at 4 °C for 3 h. The beads were washed once with DPBS on a roller at 4 °C for 5 min and incubated with secondary antibody at a 1:100 ratio in 50 μl antibody buffer 2 (1 mM PMSF, 1× protease inhibitor cocktail, 1 U μl⁻¹ RiboLock RNase Inhibitor, 0.01% Digitonin, and 1 mM DTT in DPBS) at 4 °C for 1 h. After a 5-min wash in DPBS, 1 μg pAG–deaminases in 50 μl incubation buffer (20 mM HEPES, 300 mM NaCl, 1 mM PMSF, 1×protease inhibitor cocktail, 1 U/μl RiboLock RNase Inhibitor, 0.01% Digitonin and 1 mM DTT) was added and incubated at 4 °C for 1 h.

For the deamination reaction, the unbound pAG–deaminases were removed by washing beads twice with 100 μl wash buffer (20 mM HEPES, 300 mM NaCl and 0.005% digitonin) on a roller at 4 °C for 5 min each time. The thoroughly washed beads were then resuspended in deamination buffer (15 mM HEPES pH 7.9, 60 mM KCl, 15 mM NaCl, 5% glycerol, 0.5 mM DTT and 0.1 μM ZnCl₂, 1 × protease inhibitor cocktail, 1 U μl⁻¹ RiboLock RNase Inhibitor) and incubated at 30 °C for 3 h.

After deamination, the buffer was removed using a magnetic stand for RNA extraction. Samples were resuspended in Proteinase K digestion buffer (10 mM Tris-HCl pH 8.0, 100 mM NaCl, 0.5% SDS, and 1 mM EDTA) with 0.2 mg/ml Proteinase K (AM2546, Invitrogen). The mixture was digested at 56 °C for 1 h. For large samples (> 100,000 cells), total RNA was extracted following the standard TRIzol protocol (Invitrogen). For small samples (≤ 50,000 cells), mRNA was isolated with Oligo d(T)25 Magnetic Beads (S1419S, NEB) and then subjected to Smart-seq2 protocol⁵⁴, as detailed in plate-based scMAPIT-seq.

Then, 500 ng RNA was used to generate RNA-seq libraries using the VAHTS Universal V6 RNA-seq Library Prep Kit for Illumina (NR604, Vazyme). cDNA libraries were assessed using Qubit and Agilent 2100 Bioanalyzer and then subjected to the Illumina NovaSeq 6000 or DNBSEQ-T7 platform for 150-nt paired-end sequencing (Novogene or GenePlus). This protocol utilizes random hexamer primers for reverse transcription to generate RNA-seq libraries.

MAPIT-seq of tissue sections, single-cell MAPIT-seq and long-read MAPIT-seq were performed with minor modifications. Tissue sections were fixed with 0.2% formaldehyde, and MAPIT-seq was conducted on the slides. For scMAPIT-seq, cells were fixed with 1 mM DSP, collected by centrifugation at 650g and subjected to the Smart-seq2 protocol⁵⁴ or 10x Genomics Chromium platform after deamination. Long-read MAPIT-seq adopted Kinnex full-length RNA kit (103-072-000, Pacbio) and Pacbio Revio instrument for circular consensus sequencing to generate HiFi full-length reads. 10x Genomics v3.1 and long-read RNA-seq libraries were constructed and sequenced by Annoroad Gene Technology. Details for these MAPIT-seq variants are available in the Supplementary Methods.

Mice and preparation of tissue slides

Animals were bred and maintained under specific-pathogen-free (SPF) conditions at the institutional animal facility at Peking University. They were kept on a 12-h dark–light cycle under a temperature of 20–25 °C and humidity of 30–70%, and were provided with food. C57BL/6J mice were allowed to mate overnight, with embryonic day 0.5 (E0.5) designated by the presence of vaginal plugs the following morning. Pregnant mice were euthanized by cervical dislocation, and embryos at indicated stages were dissected and embedded in optimal cutting temperature compounds (OCT; 4583, Sakura). Intact E10.5 embryos were processed as fixed frozen (4% PFA fixation, sucrose dehydration) or fresh frozen (snap-frozen in liquid nitrogen). E10.5 embryos were sagittally sectioned. Tissues of E12.5 and E16.5 were processed as fresh frozen and coronally sectioned. All tissues were sectioned at 10-μm thickness using a Cryostat (Leica), mounted on poly-l-lysine-coated glass slides (188105 W, Citotest), and stored at –80 °C.

RIP–qPCR analysis

Ultraviolet (UV) cross-linking RIP was performed as described in our previous study⁵⁵. Specifically, 1×10⁶ HEK293T cells were plated in a 10-cm dish and grown for 3 days. Protein G Dynabeads (cat. no. 10003, Invitrogen) were incubated with mouse IgG or anti-G3BP1 antibody in dilution buffer (50 mM Tris-Cl, pH 7.4, 150 mM NaCl, 1 mM EDTA, 0.1% Triton X-100) at 4 °C overnight. Cells were cross-linked with 254 nm UV (400 mJ cm⁻²) in a CL-1000 UV Crosslinker (UVP), lysed with lysis buffer (50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1% TRITON X-100, 5% glycerol, 1 mM DTT, 1 mM PMSF, 1 × protease inhibitor cocktail, 0.4 U μl⁻¹ RiboLock RNase Inhibitor) at 4 °C for 1 h, and centrifuged at 4 °C, 12,000g for 20 min. The supernatant was incubated with the antibody-bound beads at 4 °C for 6 h. Beads were washed five times with 0.5 ml IP200 buffer (20 mM Tris-Cl, pH 7.4, 200 mM NaCl, 1 mM EDTA, 0.3% Triton X-100, 5% glycerol) at 4 °C for 5 min each time, and digested with proteinase K at 55 °C for 1 h before RNA extraction. RNA was reverse transcribed using the HiScript III Q RT SuperMix (Vazyme, R323). qPCR was performed with the SYBR Green Master Mix (Vazyme, Q141) on ABI StepOnePlus Real-Time PCR System (Applied Biosystems). The relative fold enrichment of G3BP1 targets cells was calculated as 2^{(Ct(input) − Ct(IP))}, and normalized to IgG control. qPCR primers were synthesized by Tsingke; sequences are provided in Supplementary Table 31.

MAPIT-seq data processing

For bulk MAPIT-seq, 150-bp paired-end reads were quality-checked by FastQC (v0.11.9), trimmed by Trim Galore (v0.6.7), and filtered to remove abundant RNA types using BWA-MEM⁵⁶ (v0.7.17). Reads were aligned to the human (hg38) or mouse (mm10) genome using a two-round mapping strategy with HISAT2 (ref. ⁵⁷) (v2.2.1) and BWA-MEM⁵⁶, guided by GENCODE annotations. PCR duplicates were removed, exon-junction reads were split, and base quality was recalibrated using GATK⁵⁸ (v4.5.0.0). RNA editing variants were called using GATK HaplotypeCaller⁵⁹, excluding known SNPs, and gene expression was quantified by featureCounts⁶⁰ (v2.0.1) and normalized to TPM. The bulk MAPIT-seq data processing adopted a protocol similar to HyperTRIBE²⁰, TRIBE-ID²⁴ and STAMP^19,33.

For 10x Genomics scMAPIT-seq, reads were aligned to the human genome (hg38) using STARsolo⁶¹ (v2.7.11b). Low-quality cells were filtered using Scanpy⁶² (v1.10.1) and Seurat⁶³ (v5.1.0). UMAP was used for dimensionality reduction. Cell cycle phases were assigned through Seurat’s CellCycleScoring, and each phase was subdivided into early and late stages on the basis of UMAP distribution. For editing analysis, reads from selected cells were extracted by pysam (v0.22.1), deduplicated by UMI-tools⁶⁴ (v1.1.4), and split into per-cell BAMs BAMtools (v2.5.2) using the ‘CB’ tag. In each cell, RNA editing sites were identified by REDItools2 (ref. ⁶⁵).

For long-read MAPIT-seq, full-length non-chimeric (FLNC) reads were generated using IsoSeq refine (v4.0.0) to remove polyA tails and concatemers. FLNC BAMs were converted to FASTQ using bedtools (v2.31.1), and aligned to the human genome (hg38) with Minimap2 (ref. ⁶⁶) (v2.28) using splice-aware parameters and GENCODE v40 annotation. IsoQuant⁶⁷ (v3.4.1) was used to assign reads to isoforms on the basis of exon–exon/intron structures, filter ambiguous mappings and quantify expression. Uniquely assigned reads were annotated with transcript and gene IDs in bam files using pysam (v0.22.1), and split into transcript-level BAMs with BAMtools (v2.5.2). RNA editing was detected for each isoform using REDItools2 (ref. ⁶⁵). Details for bulk, single-cell and long-read MAPIT-seq can be found in the Supplementary Methods.

Differential editing analysis and target identification

A-to-G and C-to-U editing sites were annotated using bedtools⁶⁸ (v2.31.1) with GENCODE (Human v40 or Mouse vM25) for the gene annotation and UCSC RepeatMasker⁶⁹ for the repetitive element annotation, with low-coverage sites (<10 reads) excluded. To quantify editing enrichment for a MAPIT-seq sample, we describe a fraction formula:

Here, ‘total editing events’ represents the number of all C-to-U or A-to-G events detected across the entire sample, and ‘total mapped bases’ is calculated by the SAMtools stat command⁷⁰ (v1.20), indicating sequencing depth. This metric quantifies the average number of edits normalized to per aligned million bases in the sample, which serves as a coarse-grained measure to compare overall deamination activity between experimental and control groups. The SNR was defined as:

At the individual base level, the editing rate is calculated as:

This represents the fraction of reads showing an edit at a specific genomic position i, analogous to allele frequency in variant calling. To evaluate the editing enrichment of each transcript, we defined an editing index as the cumulative editing rates across all bases in a transcript or a given genome window:

where N is the number of mapped bases in a transcript or a given genome window. The MAPIT score was defined as the difference between editing indices of a given transcript in anti-RBP and IgG samples:

To identify RBP targets, a Wilcoxon signed-rank test was performed on the ‘editing index’ within 50-nt continuous and non-overlapping windows of transcripts between RBP–MAPIT and IgG–MAPIT samples by package scipy (v1.13.0), and the P values were adjusted by Benjamini–Hochberg correction. As a universal default in bulk MAPIT-seq, transcripts with editing fold enrichment > 2, MAPIT score > 0.5, P value < 0.001, and adjusted P < 0.05 were identified as RBP targets. For frozen embryonic sections, a lower threshold (editing fold enrichment > 1.5, MAPIT score > 0.4, P value < 0.05) was used.

In Figs. 1b,f and 2f and Extended Data Fig. 1b, we applied the same ‘number of editing events per million bases’ formula. Both ‘total editing events’ and ‘total mapped bases’ were specifically constrained to RBP binding peaks identified by CLIP, allowing a region-specific assessment of editing enrichment.

Details for CLIP, RIP and CAP-seq data analysis are available in the Supplementary Methods.

Identification of edit clusters and RBP-binding motifs

C-to-U and A-to-G editing sites were identified using the SAILOR workflow¹⁹ (https://github.com/YeoLab/FLARE/tree/master/workflow_sailor). High-confidence sites (confidence score > 0.5 (C-to-U) or > 0.9 (A-to-G); editing fraction < 0.7) were used as input to FLARE³⁵ (https://github.com/YeoLab/FLARE) to define edit clusters. Clusters detected in both replicates were intersected, and MAPIT high-confidence clusters were defined as overlapping C-to-U and A-to-G clusters within 100 nt. These clusters were extended to 300 nt by bedtools (v2.31.1) and subjected to de novo motif discovery using HOMER³⁶ (v4.11). For more details, see the Supplementary Methods.

Splicing map analysis

For si-PTBP1 and si-RBFOX2 RNA-seq, trimmed reads were aligned to the human genome (hg38) using STAR (v2.7.11b) with ‘–twopassMode Basic’. Differential alternative splicing (AS) events were identified from the resulting BAM files using rMATS (v4.2.0), with included and excluded exons defined as those with |IncLevelDifference | > 0.1 and FDR < 0.1. CLIP-derived RBP splicing maps were generated using RBP-Maps⁷¹ in default ‘Plotting peaks’ mode (–peak). Genomic coordinates of native cassette exons and constitutive exons were obtained from the RBP-Maps GitHub repository. For MAPIT-seq data, splicing-map-like plots were generated by computing the difference in editing rates between RBP–MAPIT and IgG–MAPIT and mapping them to genomic coordinates of included and excluded exons, native cassette exons and constitutive exons using the ‘computeMatrix’ and ‘plotProfile’ functions in deepTools⁷² (v3.5.6).

Metagene analysis

High-confidence editing sites were identified by modeling A-to-G or C-to-U edits as a Poisson process. Sites were filtered (baseline P < 0.01) on the basis of the proportion of YTHDF2–MAPIT edits significantly exceeding the IgG–MAPIT background. Meta-distributions of editing events were plotted using the R package Guitar (v2.18.0).

Clustering analysis of G3BP1 scMAPIT-seq signals

To track the changing pattern of G3BP1 binding during cell cycle progression, MAPIT scores were used to represent binding strength. Only genes in the top 50% of s.d. of MAPIT scores were considered, and MAPIT scores were scaled by z-score before clustering. Fuzzy c-means clustering was calculated by the ‘mfuzz’ function from the Mfuzz⁷³ package (v.2.62.0) with Euclidean distance as the clustering option.

Differentially expressed genes and function enrichment analysis

Differentially expressed genes (DEGs) were identified by DESeq2 (ref. ⁷⁴) (v1.36.0) in R software (v4.2.0). Gene ENSEMBL IDs were converted to Entrz IDs with package org.Hs.eg.db (v3.15.0) or org.Mm.eg.db (v3.15.0). GO enrichment analysis was performed using functions ‘enrichGO’ with package clusterProfiler⁷⁵ (v4.7.1.3). GO terms with adjusted P value < 0.05 were defined as significantly enriched.

Data visualization and statistical analysis

Package Seaborn (v0.13.2) and Matplotlib (v3.9.0) in python (v3.10.14) were used for generation of heatmaps, line plots, scatter plots, box-violin plots, Venn diagrams, genome browser tracks, density plots and box plots. For IGV-like genome browser tracks, the coverage of all bases (left y axis) and the A-to-G and C-to-U editing rates (right y axis) were extracted for visualization. PCA was conducted by package sklearn (v1.5.1). Unless otherwise specified, statistical comparisons between groups were conducted using two-tailed Wilcoxon rank-sum tests, and enrichment significance between RBP target sets was assessed using two-tailed Fisher’s exact test by scipy (v1.13.0). P values and fold enrichments are shown with each Venn diagram.

Reporting summary

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

Sequencing data are deposited in the Gene Expression Omnibus under accession code GSE278418. Reads were mapped to the reference genome (hg38 and mm10 for human and mouse samples) downloaded from GENCODE (https://www.gencodegenes.org/human/release_40.html; https://www.gencodegenes.org/mouse/release_M25.html). Previously published data are available under accession numbers GSE168943 (G3BP1 CLIP-seq)²⁶, PRJNA533136 (G3BP1 PAR-CLIP)²⁵, GSE49339 (YTHDF2 PAR-CLIP)²¹, GSE226161 (G3BP1 ARTR-seq)¹⁷, GSE207005 (G3BP1 RIP-seq, HyperTRIBE and TRIBE-ID)²⁴, GSE230717 (G3BP1 eCLIP)²⁷, GSE171008 (G3BP1 iCLIP)²⁸, GSE223295 (G3BP1 CAP-seq)²⁹, GSE156015 (SERBP1 eCLIP)³¹, GSE77633 (RBFOX2 iCLIP)¹⁵, GSE77629 (RBFOX2 eCLIP)¹⁵, GSE240014 (RBFOX2 INSCRIBE)³⁴, GSE155649 (RBFOX2 STAMP)¹⁹, GSE110519 (PUM1 CLIP)³², GSE216334 (PUM1 STAMP)³³, GSE210563 (GLORI in HeLa)²², GSE230846 (PTBP1 iCLIP)³⁷ and GSE253477 (CHTOP and PTBP1 CLAP-seq)³⁰. Source data are provided with this paper.

Codes for processing MAPIT-seq data are available in the following GitHub repository: https://github.com/WangLabPKU/MAPIT-seq.

1. Gerstberger, S., Hafner, M. & Tuschl, T. A census of human RNA-binding proteins. Nat. Rev. Genet. 15, 829–845 (2014).

   CAS  PubMed  PubMed Central  Google Scholar 

2. Hentze, M. W., Castello, A., Schwarzl, T. & Preiss, T. A brave new world of RNA-binding proteins. Nat. Rev. Mol. Cell Biol. 19, 327–341 (2018).

   CAS  PubMed  Google Scholar 

3. Cookson, M. R. Aging—RNA in development and disease. Wiley Interdiscip. Rev. RNA 3, 133–143 (2012).

   CAS  PubMed  Google Scholar 

4. Fu, M. & Blackshear, P. J. RNA-binding proteins in immune regulation: a focus on CCCH zinc finger proteins. Nat. Rev. Immunol. 17, 130–143 (2017).

   CAS  PubMed  Google Scholar 

5. Darnell, R. B. RNA protein interaction in neurons. Annu. Rev. Neurosci. 36, 243–270 (2013).

   CAS  PubMed  PubMed Central  Google Scholar 

6. Hao, J., Duan, F. F. & Wang, Y. MicroRNAs and RNA binding protein regulators of microRNAs in the control of pluripotency and reprogramming. Curr. Opin. Genet. Dev. 46, 95–103 (2017).

   CAS  PubMed  Google Scholar 

7. Ye, J. & Blelloch, R. Regulation of pluripotency by RNA binding proteins. Cell Stem Cell 15, 271–280 (2014).

   CAS  PubMed  PubMed Central  Google Scholar 

8. Gebauer, F., Schwarzl, T., Valcarcel, J. & Hentze, M. W. RNA-binding proteins in human genetic disease. Nat. Rev. Genet. 22, 185–198 (2021).

   CAS  PubMed  Google Scholar 

9. Dou, N., Chen, J., Yu, S., Gao, Y. & Li, Y. G3BP1 contributes to tumor metastasis via upregulation of Slug expression in hepatocellular carcinoma. Am. J. Cancer Res. 6, 2641–2650 (2016).

   CAS  PubMed  PubMed Central  Google Scholar 

10. Wang, Y. et al. G3BP1 promotes tumor progression and metastasis through IL-6/G3BP1/STAT3 signaling axis in renal cell carcinomas. Cell Death Dis. 9, 501 (2018).

    PubMed  PubMed Central  Google Scholar 

11. Gagliardi, M. & Matarazzo, M. R. RIP: RNA immunoprecipitation. Methods Mol. Biol. 1480, 73–86 (2016).

    CAS  PubMed  Google Scholar 

12. Ule, J. et al. CLIP identifies Nova-regulated RNA networks in the brain. Science 302, 1212–1215 (2003).

    CAS  PubMed  Google Scholar 

13. Licatalosi, D. D. et al. HITS-CLIP yields genome-wide insights into brain alternative RNA processing. Nature 456, 464–469 (2008).

    CAS  PubMed  PubMed Central  Google Scholar 

14. Konig, J. et al. iCLIP reveals the function of hnRNP particles in splicing at individual nucleotide resolution. Nat. Struct. Mol. Biol. 17, 909–915 (2010).

    PubMed  PubMed Central  Google Scholar 

15. Van Nostrand, E. L. et al. Robust transcriptome-wide discovery of RNA-binding protein binding sites with enhanced CLIP (eCLIP). Nat. Methods 13, 508–514 (2016).

    PubMed  PubMed Central  Google Scholar 

16. Khyzha, N., Henikoff, S. & Ahmad, K. Profiling RNA at chromatin targets in situ by antibody-targeted tagmentation. Nat. Methods 19, 1383–1392 (2022).

    CAS  PubMed  PubMed Central  Google Scholar 

17. Xiao, Y. et al. Profiling of RNA-binding protein binding sites by in situ reverse transcription-based sequencing. Nat. Methods 21, 247–258 (2024).

    CAS  PubMed  PubMed Central  Google Scholar 

18. McMahon, A. C. et al. TRIBE: hijacking an RNA-editing enzyme to identify cell-specific targets of RNA-binding proteins. Cell 165, 742–753 (2016).

    CAS  PubMed  PubMed Central  Google Scholar 

19. Brannan, K. W. et al. Robust single-cell discovery of RNA targets of RNA-binding proteins and ribosomes. Nat. Methods 18, 507–519 (2021).

    CAS  PubMed  PubMed Central  Google Scholar 

20. Xu, W., Rahman, R. & Rosbash, M. Mechanistic implications of enhanced editing by a HyperTRIBE RNA-binding protein. RNA 24, 173–182 (2018).

    CAS  PubMed  PubMed Central  Google Scholar 

21. Wang, X. et al. N⁶-methyladenosine-dependent regulation of messenger RNA stability. Nature 505, 117–120 (2014).

    PubMed  Google Scholar 

22. Liu, C. et al. Absolute quantification of single-base m6A methylation in the mammalian transcriptome using GLORI. Nat. Biotechnol. 41, 355–366 (2023).

    CAS  PubMed  Google Scholar 

23. Mas-Ponte, D. et al. LncATLAS database for subcellular localization of long noncoding RNAs. RNA 23, 1080–1087 (2017).

    CAS  PubMed  PubMed Central  Google Scholar 

24. Seo, K. W. & Kleiner, R. E. Profiling dynamic RNA-protein interactions using small-molecule-induced RNA editing. Nat. Chem. Biol. 19, 1361–1371 (2023).

    CAS  PubMed  PubMed Central  Google Scholar 

25. Meyer, C., Garzia, A., Morozov, P., Molina, H. & Tuschl, T. The G3BP1-family-USP10 deubiquitinase complex rescues ubiquitinated 40S subunits of ribosomes stalled in translation from lysosomal degradation. Mol. Cell 77, 1193–1205.e5 (2020).

26. He, X., Yuan, J. & Wang, Y. G3BP1 binds to guanine quadruplexes in mRNAs to modulate their stabilities. Nucleic Acids Res. 49, 11323–11336 (2021).

    CAS  PubMed  PubMed Central  Google Scholar 

27. Street, L. A. et al. Large-scale map of RNA-binding protein interactomes across the mRNA life cycle. Mol. Cell 84, 3790–3809 (2024).

    CAS  PubMed  Google Scholar 

28. Nabeel-Shah, S. et al. SARS-CoV-2 nucleocapsid protein binds host mRNAs and attenuates stress granules to impair host stress response. iScience 25, 103562 (2022).

    CAS  PubMed  Google Scholar 

29. Ren, Z., Tang, W., Peng, L. & Zou, P. Profiling stress-triggered RNA condensation with photocatalytic proximity labeling. Nat. Commun. 14, 7390 (2023).

    PubMed  PubMed Central  Google Scholar 

30. Guo, J. K. et al. Denaturing purifications demonstrate that PRC2 and other widely reported chromatin proteins do not appear to bind directly to RNA in vivo. Mol. Cell 84, 1271–1289.e12 (2024).

    CAS  PubMed  PubMed Central  Google Scholar 

31. Su, H. et al. Photoactive G-quadruplex ligand identifies multiple G-quadruplex-related proteins with extensive sequence tolerance in the cellular environment. J. Am. Chem. Soc. 143, 1917–1923 (2021).

    CAS  PubMed  Google Scholar 

32. Sternburg, E. L., Estep, J. A., Nguyen, D. K., Li, Y. & Karginov, F. V. Antagonistic and cooperative AGO2–PUM interactions in regulating mRNAs. Sci. Rep. 8, 15316 (2018).

    PubMed  PubMed Central  Google Scholar 

33. Lin, Y. et al. RNA molecular recording with an engineered RNA deaminase. Nat. Methods 20, 1887–1899 (2023).

    CAS  PubMed  PubMed Central  Google Scholar 

34. Liang, Q. et al. High-sensitivity in situ capture of endogenous RNA–protein interactions in fixed cells and primary tissues. Nat. Commun. 15, 7067 (2024).

    CAS  PubMed  PubMed Central  Google Scholar 

35. Kofman, E., Yee, B., Medina-Munoz, H. C. & Yeo, G. W. FLARE: a fast and flexible workflow for identifying RNA editing foci. BMC Bioinformatics 24, 370 (2023).

    PubMed  PubMed Central  Google Scholar 

36. Heinz, S. et al. Simple combinations of lineage-determining transcription factors prime cis-regulatory elements required for macrophage and B cell identities. Mol. Cell 38, 576–589 (2010).

    CAS  PubMed  PubMed Central  Google Scholar 

37. Nabeel-Shah, S. et al. C2H2-zinc-finger transcription factors bind RNA and function in diverse post-transcriptional regulatory processes. Mol. Cell 84, 3810–3825 (2024).

    CAS  PubMed  Google Scholar 

38. Yeo, G. W. et al. An RNA code for the FOX2 splicing regulator revealed by mapping RNA–protein interactions in stem cells. Nat. Struct. Mol. Biol. 16, 130–137 (2009).

    CAS  PubMed  PubMed Central  Google Scholar 

39. Xue, Y. et al. Genome-wide analysis of PTB–RNA interactions reveals a strategy used by the general splicing repressor to modulate exon inclusion or skipping. Mol. Cell 36, 996–1006 (2009).

    CAS  PubMed  PubMed Central  Google Scholar 

40. Zhao, J. et al. Genome-wide identification of polycomb-associated RNAs by RIP-seq. Mol. Cell 40, 939–953 (2010).

    CAS  PubMed  PubMed Central  Google Scholar 

41. Almeida, M., Bowness, J. S. & Brockdorff, N. The many faces of Polycomb regulation by RNA. Curr. Opin. Genet. Dev. 61, 53–61 (2020).

    CAS  PubMed  PubMed Central  Google Scholar 

42. Viphakone, N. et al. Co-transcriptional loading of RNA export factors shapes the human transcriptome. Mol. Cell 75, 310–323 (2019).

    CAS  PubMed  PubMed Central  Google Scholar 

43. Zhao, J., Sun, B. K., Erwin, J. A., Song, J. J. & Lee, J. T. Polycomb proteins targeted by a short repeat RNA to the mouse X chromosome. Science 322, 750–756 (2008).

    CAS  PubMed  PubMed Central  Google Scholar 

44. Zekri, L. et al. Control of fetal growth and neonatal survival by the RasGAP-associated endoribonuclease G3BP. Mol. Cell. Biol. 25, 8703–8716 (2005).

    CAS  PubMed  PubMed Central  Google Scholar 

45. La Manno, G. et al. Molecular architecture of the developing mouse brain. Nature 596, 92–96 (2021).

    PubMed  Google Scholar 

46. Atlas, R., Behar, L., Elliott, E. & Ginzburg, I. The insulin-like growth factor mRNA binding-protein IMP-1 and the Ras-regulatory protein G3BP associate with tau mRNA and HuD protein in differentiated P19 neuronal cells. J. Neurochem. 89, 613–626 (2004).

    CAS  PubMed  Google Scholar 

47. Martin, S. et al. Preferential binding of a stable G3BP ribonucleoprotein complex to intron-retaining transcripts in mouse brain and modulation of their expression in the cerebellum. J. Neurochem. 139, 349–368 (2016).

    CAS  PubMed  Google Scholar 

48. Martin, S. et al. Deficiency of G3BP1, the stress granules assembly factor, results in abnormal synaptic plasticity and calcium homeostasis in neurons. J. Neurochem. 125, 175–184 (2013).

    CAS  PubMed  Google Scholar 

49. Thomsen, E. R. et al. Fixed single-cell transcriptomic characterization of human radial glial diversity. Nat. Methods 13, 87–93 (2016).

    CAS  PubMed  Google Scholar 

50. Fallmann, J., Sedlyarov, V., Tanzer, A., Kovarik, P. & Hofacker, I. L. AREsite2: an enhanced database for the comprehensive investigation of AU/GU/U-rich elements. Nucleic Acids Res. 44, D90–D95 (2016).

    CAS  PubMed  Google Scholar 

51. Su, R. et al. Global profiling of RNA-binding protein target sites by LACE-seq. Nat. Cell Biol. 23, 664–675 (2021).

    CAS  PubMed  Google Scholar 

52. Sekar, V. et al. Detection of transcriptome-wide microRNA-target interactions in single cells with agoTRIBE. Nat. Biotechnol. 42, 1296–1302 (2024).

    CAS  PubMed  Google Scholar 

53. Cheng, Q.-X. et al. MAPIT-seq protocol V.1. protocols.io https://doi.org/10.17504/protocols.io.q26g79ekkvwz/v1 (2025).

54. Picelli, S. et al. Full-length RNA-seq from single cells using Smart-seq2. Nat. Protoc. 9, 171–181 (2014).

    CAS  PubMed  Google Scholar 

55. Li, Y.-P. et al. A TRIM71 binding long noncoding RNA Trincr1 represses FGF/ERK signaling in embryonic stem cells. Nat. Commun. 10, 1368 (2019).

    PubMed  PubMed Central  Google Scholar 

56. Li, H. Aligning sequence reads, clone sequences and assembly contigs with BWA-MEM. Preprint at https://arxiv.org/abs/1303.3997 (2013).

57. Kim, D., Paggi, J. M., Park, C., Bennett, C. & Salzberg, S. L. Graph-based genome alignment and genotyping with HISAT2 and HISAT-genotype. Nat. Biotechnol. 37, 907–915 (2019).

    CAS  PubMed  PubMed Central  Google Scholar 

58. McKenna, A. et al. The Genome Analysis Toolkit: a MapReduce framework for analyzing next-generation DNA sequencing data. Genome Res. 20, 1297–1303 (2010).

    CAS  PubMed  PubMed Central  Google Scholar 

59. Poplin, R. et al. Scaling accurate genetic variant discovery to tens of thousands of samples. Preprint at bioRxiv https://doi.org/10.1101/201178 (2017).

60. Liao, Y., Smyth, G. K. & Shi, W. featureCounts: an efficient general purpose program for assigning sequence reads to genomic features. Bioinformatics 30, 923–930 (2014).

    CAS  PubMed  Google Scholar 

61. Kaminow, B., Yunusov, D. & Dobin, A. STARsolo: accurate, fast and versatile mapping/quantification of single-cell and single-nucleus RNA-seq data. Preprint at bioRxiv https://doi.org/10.1101/2021.05.05.442755 (2021).

62. Wolf, F. A., Angerer, P. & Theis, F. J. SCANPY: large-scale single-cell gene expression data analysis. Genome Biol. 19, 15 (2018).

    PubMed  PubMed Central  Google Scholar 

63. Hao, Y. et al. Dictionary learning for integrative, multimodal and scalable single-cell analysis. Nat. Biotechnol. 42, 293–304 (2024).

    CAS  PubMed  Google Scholar 

64. Srivastava, A., Malik, L., Smith, T., Sudbery, I. & Patro, R. Alevin efficiently estimates accurate gene abundances from dscRNA-seq data. Genome Biol. 20, 65 (2019).

    PubMed  PubMed Central  Google Scholar 

65. Flati, T. et al. HPC-REDItools: a novel HPC-aware tool for improved large scale RNA-editing analysis. BMC Bioinformatics 21, 353 (2020).

    CAS  PubMed  PubMed Central  Google Scholar 

66. Li, H. Minimap2: pairwise alignment for nucleotide sequences. Bioinformatics 34, 3094–3100 (2018).

    CAS  PubMed  PubMed Central  Google Scholar 

67. Prjibelski, A. D. et al. Accurate isoform discovery with IsoQuant using long reads. Nat. Biotechnol. 41, 915–918 (2023).

    CAS  PubMed  PubMed Central  Google Scholar 

68. Quinlan, A. R. & Hall, I. M. BEDTools: a flexible suite of utilities for comparing genomic features. Bioinformatics 26, 841–842 (2010).

    CAS  PubMed  PubMed Central  Google Scholar 

69. Fernandes, J. D. et al. The UCSC repeat browser allows discovery and visualization of evolutionary conflict across repeat families. Mob. DNA 11, 13 (2020).

    PubMed  PubMed Central  Google Scholar 

70. Li, H. et al. The sequence alignment/map format and SAMtools. Bioinformatics 25, 2078–2079 (2009).

    PubMed  PubMed Central  Google Scholar 

71. Yee, B. A., Pratt, G. A., Graveley, B. R., Van Nostrand, E. L. & Yeo, G. W. RBP-Maps enables robust generation of splicing regulatory maps. RNA 25, 193–204 (2019).

    CAS  PubMed  PubMed Central  Google Scholar 

72. Ramirez, F. et al. deepTools2: a next generation web server for deep-sequencing data analysis. Nucleic Acids Res. 44, W160–W165 (2016).

    CAS  PubMed  PubMed Central  Google Scholar 

73. Kumar, L. & Futschik, M. E. Mfuzz: a software package for soft clustering of microarray data. Bioinformation 2, 5–7 (2007).

    PubMed  PubMed Central  Google Scholar 

74. Love, M. I., Huber, W. & Anders, S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 15, 550 (2014).

    PubMed  PubMed Central  Google Scholar 

75. Wu, T. et al. clusterProfiler 4.0: a universal enrichment tool for interpreting omics data. Innovation 2, 100141 (2021).

    CAS  PubMed  PubMed Central  Google Scholar 

Download references

Author notes

1. These authors contributed equally: Qi-Xuan Cheng, Gang Xie.

Authors and Affiliations

1. State Key Laboratory of Gene Function and Modulation Research, Institute of Molecular Medicine, College of Future Technology, Peking University, Beijing, China

   Qi-Xuan Cheng, Shuangjin Ding, Yi-Xia Wu, Fei-Fei Duan, Zi-Li Wan & Yangming Wang

2. Beijing Advanced Center of RNA Biology (BEACON), Peking University, Beijing, China

   Qi-Xuan Cheng, Gang Xie, Jie Wang, Shuangjin Ding, Yi-Xia Wu, Ming Shi, Zi-Li Wan & Yangming Wang

3. Academy for Advanced Interdisciplinary Studies, Peking University, Beijing, China

   Gang Xie, Jie Wang, Junyu Xiao & Yangming Wang

4. School of Life Sciences, Peking University, Beijing, China

   Xiangyu Zhang & Junyu Xiao

5. Peking-Tsinghua Center for Life Sciences, Peking University, Beijing, China

   Ming Shi & Junyu Xiao

6. College of Biological Sciences, China Agricultural University, Beijing, China

   Jing-Jia Wei

Contributions

Y.W. conceived and designed the study. Q.-X.C. designed and performed all experiments except the expression and purification of deaminase–pAG fusion protein, assisted by J.W., S.D., Y.-X.W., M.S., F.-F.D., Z.-L.W. and J.-J.W. G.X. conducted all bioinformatics analysis. X.Z. and J.X. performed the deaminase–pAG fusion protein expression and purification experiments. Q.-X.C., G.X. and Y.W. wrote the paper with input from all other authors. All authors discussed the results and commented on the paper.

Corresponding author

Correspondence to Yangming Wang.

Competing interests

Y.W., Q-X.C., G.X., X.Z. and J.X. are the inventors of a patent on protein A–protein G–deaminase fusion proteins and their applications in sequencing (CN2024116327110, patent pending), whose value could be affected by this paper. The other authors declare no competing interests.

Peer review information

Nature Methods thanks Faraz Mardakheh, Jernej Ule and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available. Primary Handling Editor: Lei Tang, in collaboration with the Nature Methods team.

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions