Fluorouridine labeling enables spatiotemporal mapping of transcriptomes and enhancers in vivo Authors: Yuxuan Feng, Yiwei Zhang, Heyuan Lian, Zhenning Wang, Jieming Wang, Li Wu, Charles J David, Fei Xavier Chen, Harihar Basnet Published: 2026 ABSTRACT AbstractTransient RNA-labeling tools can reveal transcriptional dynamics with high temporal precision, yet most approaches remain largely confined to cell culture. Here, we report a 5-fluorouridine (5-FUd)-based transient transcriptomics method, 5FU-seq, that surpasses standard strategies by enabling higher metabolic incorporation, greater sensitivity, and robust detection across a broad spectrum of transcripts. Applying 5FU-seq to the murine brain uncovered stimulus-dependent transcriptional networks and revealed a dynamic regulatory program associated with Nrf2 regulation in the hippocampus during anesthesia-induced unconsciousness. We further engineered a transgenic mouse line for cell type-specific transient transcriptomics in situ, enabling simultaneous mapping of gene expression and active enhancer activity with high spatiotemporal resolution. Benchmarking in myeloid cells responding to inflammatory cues against FACS-based RNA-seq and ATAC-seq confirmed the robustness of this dual chemical-genetic toolkit. Together, this work establishes a broadly applicable platform for integrated transcriptomic and epigenetic profiling in complex tissues.Subject terms: Gene expression analysis, Epigenetics Introduction Transcriptional plasticity underlies key cellular processes, including differentiation, metabolic adaptation, immune defense, and stress responses. Approximately 70-80% of the mammalian genome is transcribed into diverse RNA classes1,2 and recent work has uncovered critical roles for previously understudied transcripts in regulatory element establishment, immune activation, stem cell renewal, and immunotherapy3–6. Yet, the functions and regulation of many transcriptional units remain poorly defined, partly due to the lack of methods that enable their study in complex biological contexts. Metabolic labeling of nascent RNAs with synthetic uridine derivatives provides a powerful means to map transcriptional activity independent of RNA stability7–10, but its in vivo application has remained limited. To address this issue, we previously developed Flura-seq, a cytosine deaminase (CD) and uracil phosphoribosyltransferase (UPRT)-based metabolic RNA-labeling method that selectively labels RNAs in enzyme-expressing cells upon 5-fluorocytosine (5FC) treatment11. However, this method is currently limited to cell transplantation models and has been optimized primarily for protein-coding transcripts. In this study, we established a versatile 5-fluorouridine (5-FUd)-based RNA labeling strategy that enables robust transcriptional profiling in vitro, dynamic stimulus-dependent transcriptional network mapping in vivo, and simultaneous resolution of gene expression and enhancer activity in situ within defined cell types under physiological or pathological conditions. Results Transcriptomic incorporation of different uridine derivatives in cells A crucial consideration in selecting a nucleoside analog for RNA labeling is its efficient and faithful incorporation into RNA. Thio- and bromo-uridine derivatives are commonly used in metabolic RNA labeling methods; therefore, we first compared their incorporation rates with that of 5-FUd in different cell lines. Because thio, bromo, and fluoro groups differ in atomic size and reactivity (Fig. 1a), these nucleoside derivatives are expected to show distinct incorporation efficiencies during transcription. We examined three cell lines of different origins: MDA-MB-231 (MDA231), a human breast adenocarcinoma line; DLD1, a human colorectal cancer line; and E0771, a murine breast adenocarcinoma line. Cells were treated with each analog under identical conditions (500 μM, 2 h), and the relative abundance of labeled nucleosides in RNA was quantified by mass spectrometry (MS). To ensure efficient ionization of the thio group, 4-SU–treated samples were first alkylated before MS analysis (Supplementary Fig. 1a). The alkylation reaction was highly efficient, achieving nearly 100% conversion (Supplementary Fig. 1b). Incorporation of BrU and 5-FUd was measured directly by MS. Incorporation rates of different nucleoside derivatives into nascent RNA. a Schematic diagram showing the chemical structures of 5-fluorouridine (5-FUd), 4-thiouridine (4-SU), and 5-bromouridine (BrU), the workflow of each method, and their applications. IP: Immunoprecipitation. b, c Incorporation efficiency of 5-FUd, 4-SU, or BrU in nascent RNAs. b Indicated cells were treated with 500 μM of the uridine derivatives for 2 hours, and the levels of each uridine derivative, relative to unmodified uridine, were quantified by mass spectrometry (MS) (n = 2). c DLD1 cells were treated with 500 μM 5-FUd, 4-SU, or BrU for the indicated times, and incorporation of each labeled nucleoside relative to unmodified uridine was quantified as in (b) (n = 2). Unexpectedly, 5-FUd exhibited a significantly higher incorporation rate than BrU or 4-SU (Fig. 1b), likely due to the relatively small size of fluorine and the limited bioreactivity of the fluorine tag. Although incorporation efficiencies varied across cell lines, 5-FUd consistently showed superior incorporation (Fig. 1b). We further assessed short-term incorporation (2.5-10 min) and again observed that 5-FUd incorporation was highest under all conditions tested (Fig. 1c). Beyond its superior incorporation efficiency, 5-FUd also offers advantages over commonly used synthetic uridine derivatives in terms of versatility for applications in cell-type-specific in situ transcriptomics and potential clinical applicability (Fig. 1a). While robust incorporation of 5-FUd into nascent RNAs is a major advantage for efficient transcriptomic applications, the clinical use of 5-fluorouracil (5FU) as a chemotherapeutic agent12 warrants systematic evaluation of potential toxicity from 5-FUd treatment that could alter cellular transcriptomic states. First, we assessed toxicity under our experimental conditions ( < 15 min for the majority of transient transcriptomic profiling) by testing whether increasing concentrations of 5-FUd induced transcriptomic alterations, given that 5FU-related toxicity is dose-dependent13. We observed no appreciable transcriptional differences across RNA classes between cells treated with 1 mM or 4 mM 5-FUd for 15 min (Supplementary Fig. 2a). To directly evaluate the effect of the upper-limit experimental dose (2 mM 5-FUd) on transcription, we also examined RNA polymerase II (Pol II) occupancy by chromatin immunoprecipitation sequencing (ChIP-seq). Among ~30,000 peaks detected peaks, fewer than 100 genome-wide, and only 14 out of ~10,000 in the promoter peaks showed statistically significant differences (Supplementary Fig. 2b; Supplementary Data 1), indicating minimal transcriptomic impact of 5-FUd under our experimental conditions. In addition, we also investigated the potential cytotoxicity associated with 5-FUd labeling by assessing stress granule formation marked by G3BP1 foci14. As expected, robust stress granule formation was observed following treatment with the cytotoxic agent sodium arsenite14, whereas no stress granules were detected in two independent cell lines (MDA231 and DLD1) treated with 5-FUd under the experimental conditions (Supplementary Fig. 2c–f). These results indicate minimal cytotoxicity of 5-FUd under the conditions used in our study. Comparing different nascent RNA sequencing methods Next, we compared the performance of the three RNA labeling methods in capturing nascent transcripts using DLD1 cells ( ~ 4 million cells), during 10 min of RNA labeling. First, we examined the different classes of RNAs captured by each method. Indeed, 5-FUd labeling (5FU-seq) captured the largest number of protein-coding mRNAs, noncoding RNAs, as well as previously unannotated transcripts (Fig. 2a). The performance of 4SU-seq was relatively close for protein-coding genes; however, 5FU-seq significantly outperformed both 4SU-seq and BrU-seq for different classes of noncoding and unannotated transcripts (Fig. 2a, Supplementary Fig. 3a–l). Interestingly, the distribution of aligned reads to different genomic regions showed distinct patterns across the methods, with 5FU-seq showing the highest intronic alignment rate (Supplementary Fig. 3m), suggesting greater coverage of nascent transcripts, as newly synthesized RNAs are expected to be unprocessed with intact introns9. Consistently, 5FU-seq exhibited the highest intron-to-exon ratios for genes detected by each method, as well as for those commonly detected by all three methods (Supplementary Fig. 3n, o), further demonstrating its robustness. Comparison of 4-SU-, BrU-, and 5-FUd-based sequencing methods. a Number of transcriptional units (TUs), including mRNAs, non-coding RNAs, and unannotated transcripts, detected by each method. b Intronic signal relative to exons. Average intron coverage is shown relative to the maximum signal in the last 50 bp upstream of the 5′ splice site. A representative UCSC browser track is shown. c Experimental design for measuring background RNA pulled down by TT-seq and 5FU-seq. d Proportion of reads mapped to the human genome (background) in TT-seq and 5FU-seq. A representative UCSC browser track shown. e Intronic and exonic read alignment after labeling with 4-SU or 5-FUd for the indicated times in min (mins). f Total reads mapped to the human genome under the conditions in (e). A representative UCSC browser track shown. g 5FU-seq performed with the indicated cell numbers in millions (M). Alignment to exons and introns is shown. (n = 2). For visualization, data were normalized to counts per million (CPM). The 4SU-based RNA labeling method was further improved by introducing a fragmentation step that removes the 5′ portion of nascent RNA lacking 4-SU, termed transient transcriptome sequencing (TT-seq)9. To test comparability, we also performed 5FU-seq after RNA fragmentation and compared it with TT-seq. Consistent with the previous report9, fragmentation greatly improved intronic coverage relative to exons in TT-seq from 3% to 54% (Fig. 2b). In comparison, fragmentation increased intronic coverage in 5FU-seq from 15% to 86% (Fig. 2b). We also assessed the signal-to-noise ratio of each method by measuring the level of pull-down of unmodified RNA (background). Unlabeled human RNAs were mixed with mouse spike-in RNAs labeled with either 5-FUd or 4-SU, and the proportion of human RNAs was quantified (Fig. 2c). More than 45% of the reads aligned to human (background) in TT-seq, compared to fewer than 15% in 5FU-seq (Fig. 2d), demonstrating the low inherent noise of 5FU-seq. Additionally, we compared the performance of the two methods under various RNA labeling conditions. A minimum of 5 min of RNA labeling is required for TT-seq9,15; hence, we examined intronic coverage after 2.5-10 min of labeling. Interestingly, 5FU-seq achieved similar intronic coverage after just 2.5 min of labeling compared to TT-seq at 10 min (Fig. 2e). Likewise, the relative mapping to human versus mouse spike-ins increased proportionally with RNA labeling time in 5FU-seq, whereas TT-seq exhibited a higher baseline human mapping (Fig. 2f), indicating a better signal-to-noise ratio in 5FU-seq compared to TT-seq. We next evaluated the performance of 5FU-seq with small cell numbers. Typically, 15-50 million cells are used in standard TT-seq experiments15. We performed 5FU-seq with 0.16-4 million cells and compared the transient transcriptome data. Each condition yielded approximately 70% mapping to introns, showed similar duplication rates, and detected a comparable number of genes when down sampled (Fig. 2g; Supplementary Fig. 4a, b), indicating robust performance even with relatively small input. We also tested different 5-FUd labeling concentrations (0.5-4 mM). In all conditions, approximately 60% of reads mapped to introns (Supplementary Fig. 4c). However, overall alignment to human RNA relative to spike-ins improved significantly at >1 mM (Supplementary Fig. 4d), suggesting that optimal performance is achieved at >1 mM 5-FUd. Finally, we assessed the effect of single versus double elution on 5FU-seq performance. Interestingly, double elution decreased intronic alignment with low input (0.16 million cells) but increased it with higher input (4 million cells) (Supplementary Fig. 4e). Thus, single elution may be preferable under low-input conditions. Collectively, these data demonstrate the robust performance of 5FU-seq compared to standard TT-seq under both normal and challenging conditions. Consciousness-related dynamic transcriptional networks in the brain Having validated a robust 5-FUd incorporation into nascent RNAs, enabling short-term transcriptomic profiling in vivo (Supplementary Fig. 5a), we next tested its ability to capture rapid, stimulus-dependent transcriptional changes in the brain. This highly dynamic organ responds within minutes to diverse physical and chemical stimuli. While immediate-early genes such as c-Fos are induced within 30–90 min16, it is unclear whether broader transcriptional networks are activated on similar timescales. We exposed mice to leptin17, avertin, or isoflurane, followed by 5-FUd injection and harvest brain after 30 min (Fig. 3a). As expected, 5FU-seq robustly detected expression of a wide spectrum of RNAs in the murine brain under homeostasis (Fig. 3b). Interestingly, in response to stimuli, protein-coding transcripts were dominated by down-regulation, with no overlap between leptin and anesthetic treatments (Fig. 3c, d; Supplementary Fig. 5b; Supplementary Data 2). Furthermore, avertin and isoflurane shared transcriptional changes, suggesting a signature linked to consciousness (Fig. 3c, d). A deeper analysis of genes significantly altered during unconsciousness identified clear down-regulation of several immediate-early response genes (Fig. 3e). Interestingly, multiple genes involved in antioxidant defense, including Nfe2l2 (Nrf2), the master regulator of antioxidant defense18, and some genes involved in regulation of circadian rhythm were downregulated by both anesthetics (Fig. 3f; Supplementary Fig. 5c). In contrast, leptin treatment led to transcriptional downregulation of several genes with established roles in neuronal activity (Supplementary Fig. 5d). As expected, total RNA-seq detected minimal changes ( < 10 genes) (Supplementary Fig. 5e). We also examined whether the transcriptional alterations occurred earlier upon avertin treatment, and found that many of the transcriptional shifts appeared within 15 min of avertin exposure (Supplementary Fig. 5f). Dynamic transcriptional networks in the brain in response to stimuli. a Schematic of the experimental design (Created in BioRender. Feng, Y. (2026) https://BioRender.com/zhnxqii) to identify acute transcriptional responses to Leptin, Avertin, and Isoflurane in the brain. b Classes of RNAs (transcriptional units, TUs) and numbers detected by 5FU-seq after 30 min labeling. c, d Overlap of c, up- and d, downregulated genes across treatments (fold change > 1.5, P < 0.05) (n = 3). e, f Heatmap showing transcriptional alterations in (e), immediate-early response genes, and f antioxidant genes in vehicle-treated negative control (NC) and upon avertin or isoflurane treatment. Genes with significant change in expression by either treatment were plotted. Nfe2l2 and Ier3 expression under indicated treatments were visualized in UCSC Genome Browser. g RNA-FISH of Nfe2l2 in mouse brain sections in vehicle treated (Control) or after 30 mins Avertin or Isoflurane treatment; hippocampus region is outlined (dotted line). Scale bar, 100 μm (n = 3). h Effect of NAC pretreatment on the duration of unconsciousness induced by avertin (n = 6). The P value was calculated using an unpaired two-tailed t test. i Heatmap of different classes of differentially expressed RNAs (fold change > 1.5, P < 0.05) in at least one comparison of Avertin/Isoflurane/Leptin versus control. For visualization, data were normalized to counts per million (CPM). To gain spatial insight into the regulation of Nrf2 expression in the brain during unconsciousness, we performed RNA-FISH using probes against Nrf2. Intriguingly, Nrf2 was found to be down-regulated specifically in the hippocampus region after both anesthetic treatments (Fig. 3g). To investigate whether antioxidants functionally modulate consciousness, we pretreated mice with N-acetylcysteine (NAC), a commonly used dietary antioxidant supplement, and assessed its effect on the duration of avertin-induced unconsciousness. To our surprise, NAC-pretreated mice regained consciousness ~25% faster than control mice (Fig. 3h), supporting a functional role for antioxidant pathways in modulating consciousness. We also examined the alterations in other classes of RNAs during the anesthetic-induced unconsciousness. Notably, both anesthetics significantly altered the expression of multiple RNA classes, whereas leptin had minimal effects on these RNA classes (Fig. 3i; Supplementary Fig. 5g). Collectively, these findings establish 5FU-seq as a robust method to resolve rapid, stimulus-dependent transcriptomic programs across RNA classes. Application of 5FU-seq to identify cell type-specific transient transcriptome in situ A tissue is composed of different cell populations with unique gene expression patterns. Currently, the most widely used approach to define the transcriptome of a specific cell-type involves breaking the tissue into single cells through mechanical and chemical dissociations and sorting the cells by FACS before making the RNA library. To examine the potential benefit of adopting 5FU-seq over FACS-sorted RNA-seq for cell-type-specific transcriptome, we used a lung metastasis xenograft model of RFP-expressing MDA-MB-231 cells that allows us to rigorously determine the extent of transcriptional artifacts introduced by FACS (Fig. 4a). Using this assay, RNA-seq was performed using RNA isolated directly from the total lung harboring the human cells, and then human reads were bioinformatically assigned, or the cells were first sorted by FACS and then sequenced (Fig. 4a). Intriguingly, we found >8000 genes were differentially expressed between unprocessed cells and FACS-sorted cells (Fig. 4b), underscoring the magnitude of FACS-induced perturbations and the need for robust in situ transcriptomic tools to accurately profile gene expression in vivo. Identification of myeloid cell-specific transient transcriptomes in the lung. a Schematic showing the experimental design (Created in BioRender. Feng, Y. (2026) https://BioRender.com/mktmgvv) to assess transcriptional artifacts generated by FACS in a xenograft lung metastasis assay. b Volcano plot of differentially expressed genes (fold change > 1.5, P < 0.05; n = 3) in human cells from RNA extracted directly from lung metastases versus RNA from sorted human cells. FDR, False Discovery Rate. c Schematic (Created in BioRender. Feng, Y. (2026) https://BioRender.com/ayjj2wk) showing the generation of Cre-inducible URCfl/fl mice. d Classification of RNA classes (transcriptional units, TUs) and numbers detected in myeloid cells in mouse lung after 90 min of RNA labeling via 5FC injection in URCfl/fl;Lyz2-Cre mice. e Schematic (Created in BioRender. Feng, Y. (2026) https://BioRender.com/qmybfvf) showing the experimental design to profile LPS-regulated transcriptomes in myeloid cells using 5FU-seq and FACS. f Volcano plot of differentially expressed genes (fold change > 1.5, P < 0.05; n = 3) in myeloid cells upon LPS treatment, as detected by 5FU-seq and FACS. Overlap between g upregulated and h downregulated genes in lung myeloid cells upon LPS treatment as identified by 5FU-seq and FACS-sorted RNA-seq. i Gene Ontology (GO) analysis of genes upregulated upon LPS treatment in myeloid cells, as identified by 5FU-seq (left) and RNA-seq (right) (n = 3). P value was calculated using one-tailed hypergeometric test with Benjamini–Hochberg adjustments for multiple comparisons. A unique advantage of 5FU-seq is its potential for cell-type-specific applications in mouse models via cell-type-restricted expression of CD-UPRT coupled with 5-fluorocytosine (5FC) treatment11. Accordingly, we generated Lox-STOP-Lox-HA-UPRT-RFP-CD (URCfl/fl) transgenic mice that express URC in a Cre-dependent manner (Fig. 4c). To confirm cell-type specificity, we examined HA-Uprt expression across five tissues from URCfl/fl mice by FACS. HA-Uprt was undetectable in all tissues tested in URCfl/fl mice and became detectable only after cell-type-specific Cre expression, as exemplified by URCfl/fl;Lyz2-Cre mice, in which Cre is restricted to myeloid cells (Supplementary Fig. 6a,b). Next, we tested the feasibility of cell-type-specific 5FU-seq in URC;Lyz2-Cre mice by administering 5FC for 90 min. Immunoprecipitation of the labeled RNA isolated from whole lung using an anti-BrdU antibody recovered >12,000 protein-coding genes and large numbers of all major noncoding RNA classes (Fig. 4d), demonstrating robust in situ performance of 5FU-seq in a defined cell-type. Because myeloid cells rapidly respond to lipopolysaccharide (LPS)19, we next examined the acute transcriptional response to LPS (90 min) in URC;Lyz2-Cre mice. We first verified the physiological response to LPS by measuring canonical LPS-induced proteins in serum by ELISA and monitoring body temperature. As expected, multiple cytokines (Il1, Il6, Ifnγ, and Tnfα) known to be induced by LPS19 were elevated, and a hypothermic response was observed (Supplementary Fig. 6c, d), confirming systemic responsiveness under our experimental conditions. In addition, 5FC treatment did not alter the expression of LPS-induced genes (Supplementary Fig. 6e). We then performed 5FU-seq to define the in situ myeloid transcriptomic response to LPS. 5FU-seq identified 2,105 upregulated and 1,330 downregulated genes during 90 min of RNA labeling in LPS-treated mice (Fig. 4f). In parallel, we isolated RFP⁺ myeloid cells by FACS (Supplementary Fig. 6b), and performed RNA-seq to identify differentially expressed genes. Surprisingly, FACS-based RNA-seq detected 441 upregulated and 3,257 downregulated genes (Fig. 4f). Upregulated genes showed substantial overlap between methods, whereas downregulated genes overlapped minimally ( ~ 3%) (Fig. 4g, h). Notably, 5FU-seq identified an additional 1,811 LPS-induced genes missed by FACS, including key pathways involved in macrophage activation (Fig. 4g, i). A correlation plot of these 1,811 LPS-upregulated genes, as detected by 5FU-seq versus RNA-seq, revealed genes with both higher and lower basal expression (Supplementary Fig. 6f), indicating both qualitative and quantitative advantages of 5FU-seq for detecting stimulus-induced transcriptional programs. We also performed Gene Ontology (GO) analyses to identify unique or common programs captured by each method (Supplementary Fig. 6g–i). Finally, we also assessed the response to LPS at an earlier time point (30 min), with RNA labeling during the final 15 min. As expected, a smaller gene set was regulated within this time window: 281 genes were upregulated and 8 genes were downregulated (Supplementary Fig. 6j, k; Supplementary Data 3). In contrast, FACS-based RNA-seq detected only 14 upregulated genes and 19 downregulated genes (Supplementary Fig. 6l; Supplementary Data 4). Collectively, these results highlight the robustness of 5FU-seq for identifying stimulus-dependent gene programs in specific cell types in situ. Application of 5FU-seq to identify cell-type-specific active enhancers in situ Another unique benefit of 5FU-seq is its ability to simultaneously identify active enhancers bound by transcription factors (TFs), as TF-bound enhancers transcribe enhancer RNAs (eRNAs)20. Currently, identifying active enhancers in complex tissues requires separate assays. For example, cells of interest are first isolated by FACS, and potential enhancers are then identified using methods, such as ATAC-seq or ChIP-seq. In contrast, 5FU-seq can directly identify TF-bound active enhancers by detecting eRNAs. Furthermore, ATAC-seq only reveals whether an enhancer is open, but cannot distinguish between active and poised enhancers. We performed a head-to-head comparison of ATAC-seq and 5FU-seq to identify dynamic enhancers in an in situ LPS treatment assay (Fig. 5a). In our analysis, each biological triplicate was required to have at least one read covering annotated enhancer regions21 to be considered detected. Using 5FU-seq, ~100,000 enhancers were detected in both control and LPS-treated samples, ~19,000 were unique to control, and ~17,000 were unique to LPS-treated samples (Fig. 5b). By contrast, ATAC-seq identified ~86,000 open enhancers common to both conditions, ~4000 unique to control, and ~8000 unique to LPS-treated samples (Fig. 5c). As expected, transcription factor (TF) footprinting22 identified 39 differentially enriched TFs upon LPS treatment in the ATAC-seq data (Supplementary Data 5), including multiple AP-1 family members that are activated during immune stimulation23. Among these enhancers, ~8,400 showed differential expression between control and LPS-treated samples (fold change >1.5, P < 0.05) by 5FU-seq, whereas ATAC-seq detected only 49 statistically significant enhancers (Fig. 5d). We confirmed that the higher number of differential active enhancers identified by 5FU-seq was not due to differences in sequencing depth (Supplementary Data 6). This striking difference likely arises because ATAC-seq measures DNA accessibility, which may not change substantially in already open chromatin during transcriptional activation, whereas changes in eRNA expression are cumulative and therefore more pronounced. Identification of myeloid cell specific active enhancers in the lung. a Experimental design (Created in BioRender. Feng, Y. (2026) https://BioRender.com/d2vsrcz) to profile LPS-regulated enhancers in myeloid cells using 5FU-seq and ATAC-seq. b, c Active enhancers detected by b, 5FU-seq and c, ATAC-seq in control and LPS-treated lungs. Each enhancer was considered detected in 5FU-seq if it had > 1 read per replicate. Representative examples of LPS-upregulated enhancers are shown, with zoomed-in regions (left) (n = 3). d Differential enhancers identified by DESeq2 (fold change >1.5, P < 0.05) for 5FU-seq and DiffBind (FDR < 0.05) for ATAC-seq; overlapping enhancers are shown. e Differential eRNAs were mapped to the nearest genes and compared with LPS-regulated genes detected by 5FU-seq. Representative UCSC Browser examples are shown, with zoomed-in enhancer regions (left). Annotated enhancers are shown by yellow rectangles. f, g Motif analysis of f, up- and g, downregulated enhancers identified by 5FU-seq using HOMER (P < 0.05); five representative enriched motifs are shown. P values were calculated using the binomial distribution. h Comparison of different transcriptomic profiling methods for cells in complex tissues. For visualization, data were normalized to counts per million (CPM). To examine cis-regulatory relationships of differentially expressed eRNAs and LPS-regulated protein-coding genes, we assessed the expression correlation between eRNAs and their neighboring genes. Indeed, 75% of upregulated eRNAs were adjacent to upregulated genes, and over 50% of downregulated eRNAs were adjacent to downregulated genes (Fig. 5e). To identify potential transcription factors (TFs) activated by LPS, we performed motif analysis using HOMER on both up- and downregulated enhancers. This analysis revealed substantially more differential TF motifs by 5FU-seq compared to ATAC-seq (Fig. 5f, g; Supplementary Data 7, 8), many of which are known to function in inflammatory responses. In summary, 5FU-seq is a simple yet powerful and versatile chemogenetic tool that offers unique advantages over existing RNA-profiling approaches, in terms of an improved signal-to-noise ratio due to high incorporation and competitive elution, robustness due to high incorporation rate, low experimental complexity due to fewer steps, and precise temporal control with as little as 15 min of RNA labeling in vivo and 2.5 min in vitro (Fig. 5h). Importantly, 5FU-seq enables in situ application to simultaneously profile gene expression and active enhancers, which provides important information about the epigenetic landscape of specific cell types residing within their native tissue microenvironment (Fig. 5h). Discussion In this study, we used 5-FUd, a nucleoside analog that exhibits more robust incorporation into RNA compared to other analogs (4-SU, BrU), to profile the nascent RNA transcriptome. With this sensitive tool, we were able to capture all major RNA types with improved performance relative to current standard methods (4SU-seq, BrU-seq, TT-seq). This enhancement enabled profiling of acute, dynamic in vivo transcriptional networks encompassing mRNA, eRNA, tRNA, and other RNA species in response to multiple stimuli in the brain. We established an unprecedented lower limit for bulk nascent transcriptomics, 2.5 min of RNA labeling and 0.16 million cells, enabling the mapping of acute transcriptional responses and the study of rare cell populations. Although we were able to perform 5FU-seq with as few as 10,000 cells, the number of detected genes decreases as input cell numbers decline. Furthermore, we generated a versatile transgenic mouse that expresses yeast CD, an enzyme absent in mammals, and UPRT, exclusively in Cre-positive cells, allowing precise, cell-type-specific transcriptomic profiling while simultaneously assessing the epigenetic landscape in situ. We demonstrated the robustness of this mouse line by capturing dynamic transcriptional networks, epigenetic landscapes, and regulated TF motifs through head-to-head comparisons with standard FACS- and ATAC-seq-derived results in Lyz2-Cre mice stimulated with LPS. RNA-FISH–based spatial transcriptomics methods such as MERFISH24, are powerful tools for mapping the transcriptome in tissue sections; however, a major limitation is the restricted number of genes that can be probed simultaneously. Using 5FU-seq, we identified a highly dynamic and immediate transcriptional response associated with the Nrf2-dependent antioxidant program, which could be spatially mapped to the hippocampus by RNA-FISH as early as 30 min after induction of unconsciousness. Given the robustness of 5FU-seq and its applicability to any cell type in situ, it can serve as a complementary tool for candidate gene prioritization in spatial transcriptomics. Another major advantage of 5-FUd-based RNA labeling is its potential application in clinical studies. 5FU is a clinically approved drug for treating various cancer types, and it could potentially be used in human subjects to evaluate the efficacy and toxicity of therapeutic drugs by identifying transcriptional response signatures associated with these outcomes. Such applications could pave the way for personalized cancer treatment and are currently under active investigation. Methods Animal studies All animal experiments were performed in accordance with a protocol (24-HB1) approved by the Tsinghua University Institutional Animal Care and Use Committee. Mice were maintained under specific pathogen-free conditions with 12/12 dark/light cycle, 20–26 °C, and 30–70% humidity. For all experiments, 6-8-week-old mice were used. For the brain-related experiments, 6-week-old mice were purchased from Beijing Hfk Bioscience Co., Ltd. To measure the transcriptomic changes in the murine brain during unconsciousness, 2.5% Avertin (12 μl/g) was given to the mice by intraperitoneal (IP) injection for anesthesia. In the isoflurane experiment, mice were kept in a chamber supplied with isoflurane. The unconsciousness was determined by softly pinching the mice on their feet. For the Leptin experiment, 3 mg/kg Leptin (Novorprotein) was injected intraperitoneally. C57BL/6J-Gt(ROSA)26Sortm1(CAG-LSL-HA/Uprt-T2A-RFP-IRES-CD) (URCfl/fl) mouse line was generated by Model Organism, Shanghai, through zygotic injection and CRISPR/Cas9-mediated homologous recombination of a targeting vector in the ROSA26 locus. LPS challenge was performed by IP injection of 2.5 mg/kg of LPS (Solarbio). For transient transcriptomic labeling of RNA in mice, 200 μl of 5-FUd (125 mg/ml, Acmec) was injected through the tail vein. For myeloid cell RNA labeling, 500 μl of 12.5 mg/ml 5-fluorocytosine (5FC, Macklin) was injected intraperitoneally. For consciousness-related experiments, mice were anesthetized by intraperitoneal injection of 2.5% Avertin (230 mg/kg). Anesthesia induction was quantified using the loss of the righting reflex (LORR) assay. After injection, each mouse was placed supine on a flat surface and assessed at regular intervals by gently positioning the animal on its back; LORR was defined as the first time point at which the mouse failed to spontaneously return to a prone position within 30 s. Recovery was defined as the return of the righting reflex (RORR), scored as the first time point at which the mouse spontaneously returned to a prone position within 30 s, and the duration of anesthesia was recorded from LORR to RORR. For NAC intervention, NAC (250 mg/kg) was prepared in PBS and administered by intraperitoneal injection 1 h before Avertin injection. Mass spectrometry for RNA incorporation analysis Cells were treated with either 4-SU (MedMol), BrU (Selleck) or 5-FUd, and labeled RNAs were extracted using TRIzol (ThermoFisher) or RNA isolater Total RNA Extraction Reagent (Vazyme). 2 μg RNA from each sample was used as an input. RNAs were first denatured at 95 °C for 10 min before hydrolyzing at 37 °C for 2 hours in a buffer containing 15 mM sodium acetate (pH 5.3), 6 mM zinc chloride and 0.5 μl nuclease P1 (NEB). The reaction was terminated by adding 250 mM ammonium bicarbonate. 0.5 units of alkaline phosphatase was added and incubated at 37 °C for 1 h. The product was centrifuged at 13,000g for 10 min, and the supernatant was collected for mass spectrometry analysis using Liquid Chromatography-Tandem Mass Spectrometry (LC-MS/MS). Mass Spectrometry experiments were performed with two biological replicates. Cell culture All cell lines used in this study were grown in DMEM medium (Gibco) supplemented with 10% FBS (Gemini) at 37 °C and 5% CO2. Immunofluorescence staining Cells were seeded in 24-well plates at 50–60% confluency and treated with different concentrations of 5-FUd or sodium arsenite (500 μM; NaAsO₂, TCI) for the indicated times. Following treatment, cells were washed once with PBS and fixed in 1% paraformaldehyde for 15 min at room temperature. After washing with PBS, cells were permeabilized with 0.2% Triton X-100 for 10 min and blocked in 3% BSA for 1 h. Cells were then incubated with a G3BP1 polyclonal antibody (Proteintech, 3057-2-AP; 1:500) overnight at 4 °C, washed three times with PBS, and then incubated with an Alexa Fluor 488-conjugated anti-rabbit secondary antibody (1:500 in 3% BSA) for 1 h at room temperature. After three additional washes with PBS, nuclei were stained with DAPI (100 ng/ml; Solarbio) for 10 min. Coverslips were mounted using antifade mounting medium (Solarbio). Images were acquired using an Olympus FV3000 confocal microscope with 40× and 100× objectives. Transient transcriptome sequencing 80-90% confluent cell cultures in 10 cm culture plates were treated with 2 mM 5-FUd, 500 μM 4-SU or 500 μM BrU (or at indicated concentrations). Cells were harvested in 2 ml TRIzol, and total RNAs were extracted following the standard RNA extraction protocol. For each sample, 5% of spike-in RNA from 5-FUd-treated E0771 cells was used. RNAs were fragmented in 250 μl of 0.2 M NaOH solution on ice for 14 min. The reaction was neutralized by adding 250 μl of Tris-HCl buffer. Labeled nascent RNA was immunoprecipitated using Protein G Dynabeads (Invitrogen) pre-conjugated with 1 μg anti-BrdU antibody (Abcam) through overnight incubation at 4 °C. Each RNA sample was incubated with the antibody-bead complex in 0.5× Binding buffer (0.5× Sodium Chloride-Sodium Phosphate-EDTA (SSPE), 0.025% Tween 20) at 4 °C for 2 h on a rotator. After incubation, the beads were washed three times with 0.5× Binding buffer, three times with Wash buffer B (1× SSPE, 0.05% Tween 20), once with Wash buffer C (TE with 50 μg glycogen), and finally with TE buffer. The bound RNA was eluted in 60 μl of 100 μg/mL BrdU (Selleck) for 45 min on a shaker at 4 °C, and subsequently purified using VAHTS RNA Clean Beads. TT-seq25 and BrU-seq8 were done following standard published protocols. Sequencing libraries were constructed using VAHTS Universal V8 RNA-seq Library Prep Kit for MGI and sequenced utilizing an Illumina HiSeq PE150 platform (Novogene Technology, Beijing, China). For in vivo tissue labeling, 125 mg/ml 5-FUd was given through intravenous injection for short-term labeling (15 or 30 min) as described above. Cell-type-specific nascent RNA was labeled by IP injection of 500 μl 5FC (12.5 mg/ml) as described above. Organs were harvested in 2 ml TRIzol, homogenized with a tissue grinder, and total RNA was extracted using the standard protocol. 5% 5-FUd-labeled human RNA from MDA231 cells was added as spike-in. Chromatin Immunoprecipitation sequencing (ChIP-seq) 80–90% confluent cells were crosslinked using 1% formaldehyde for 10 min at room temperature. Crosslinking was terminated by dropwise addition of glycine to 125 mM, followed by a further 10 min incubation. Cells were collected by scraping and washed with 2.5 ml Buffer A (10 mM HEPES, pH 7.9; 10 mM KCl; 1.5 mM MgCl₂; 340 mM sucrose; 10% glycerol; 5 mM 2-mercaptoethanol; 1× protease inhibitor cocktail (CWBio, CW2200S); 1× phosphatase inhibitor cocktail (CWBio, CW2383S)). For spike-in normalization, 5% murine E0771 cells were included and processed in parallel. Cell membranes were lysed by adding 3 ml Buffer A supplemented with 0.2% Triton X-100 and incubated on ice for 10 min. Nuclei were isolated by centrifugation (1300 × g, 5 min, 4 °C), resuspended in 1 ml Buffer A, and further purified by layering onto a 3 ml sucrose cushion (10 mM HEPES, pH 7.9; 30% sucrose; 1.5 mM MgCl₂), followed by centrifugation (1300 × g, 12 min, 4 °C) to clean up nuclei. The nuclei were resuspended in lysis buffer (150 mM NaCl; 1% Triton X-100; 20 mM Tris, pH 8.0; 0.1% SDS) and subjected to sonication for 6 min (12 min total, 15 s ON/15 s OFF). Sonicated chromatin was precleared with 25 µl Protein G magnetic beads (Invitrogen, 10003D) for 1 h at 4 °C with rotation. The precleared material was then incubated overnight at 4 °C with 5 µg of the indicated antibody (Pol II [mAb], Active Motif, 39097). Antibody-chromatin complexes were captured by adding 35 µl protein G-conjugated Dynabeads and rotating for 4 h at 4 °C. Beads were washed three times using wash buffer (1% Triton X-100; 50 mM Tris, pH 8.0; 10% glycerol) with stepwise increases in NaCl concentration (150 mM, 300 mM, and 400 mM NaCl), and then rinsed once with Tris-EDTA (TE) buffer. Bound chromatin was eluted in elution buffer (1% SDS in TE) at 37 °C for 45 min with shaking. Crosslinks were reversed by incubating samples overnight at 65 °C, after which DNA was purified using a DNA purification column (Zomanbio, ZPV202-1). Purified DNA was end-repaired and A-tailed with the VAHTS Repair Module for ONT (Vazyme, TS21-01) following the manufacturer’s protocol. Sequencing adapters (Vazyme, N323-01) were ligated using Quick T4 DNA Ligase (NEB, M2200L). Libraries were then PCR-amplified for 18 cycles with KAPA HiFi HotStart ReadyMix (Roche, KK2692) and sequenced on an Illumina HiSeq PE150 platform (Novogene, Beijing, China). RNA-FISH Brains were harvested and fixed in 4% formaldehyde for 24 h and dehydrated in 30% sucrose buffer for 48 h. 16 μm brain sections were obtained by cryosectioning and RNA-FISH was performed following a published protocol26. Nfe2l2 probe sequences used in the experiments were as follows: GGCACTATCTAGCTCCTCCAAAAAAAAAAAGACGCTAATCAACCTGGT; CGGCTGAATTGGGAGGAATTAAAAAAAAAAGACGCTAATCAACCTGGT; CTGGTTTGCTGCAGAGAGTAAAAAAAAAAAGACGCTAATCAACCTGGT; GCTCCATGTCCTGCTCTATGAAAAAAAAAAGACGCTAATCAACCTGGT; GCGAGCTGAGCTTCATTGAAAAAAAAAAAAGACGCTAATCAACCTGGT; GGCATCATCAGTGGAGAGGAAAAAAAAAAAGACGCTAATCAACCTGGT; GCCGTTCTGTTTGACACTTCAAAAAAAAAACTCGATCTGGACACTCTT; GTGTCTGTCTGGATGTGCTGAAAAAAAAAACTCGATCTGGACACTCTT; GGAACAAGGAACACATTGCCAAAAAAAAAACTCGATCTGGACACTCTT; TGGAAAATAGCTCCTGCCAAAAAAAAAAAACTCGATCTGGACACTCTT; TCTCCTGCGTATATCTCGGAAAAAAAAAAACTCGATCTGGACACTCTT; TGAGTCTAAGGAGGTCAGCTAAAAAAAAAACTCGATCTGGACACTCTT. Fluorescence-activated cell sorting and ATAC-seq Tissues were harvested and minced into small pieces, followed by enzymatic digestion in 4 ml dissociation buffer (DMEM/F12 supplemented with 10% FBS, 1 mg/ml collagenase V (Solarbio, Cat#C8170), 200 U/ml DNase I (Solarbio, Cat#D8071), and 1 ml TrypLE (Gibco)) at 37 °C with gentle rotation. Digested tissues were passed through a 70-μm cell strainer, and cells were pelleted by centrifugation at 600 g for 10 min. Red blood cells were lysed twice using ACK buffer (150 mM NH4Cl, 10 mM KHCO3, 0.1 mM Na2EDTA, pH 7.2–7.4) for 5 min. Cells were then resuspended in 500 μl FACS buffer containing 5 μl DAPI for cell sorting. A total of 10,000 cells were collected for ATAC-seq and libraries were prepared following the protocol from Hyperactive ATAC-Seq Library Prep Kit for Illumina (Vazyme). For surface staining, cells were incubated with Fc receptor blocker (TruStain FcX™ PLUS (anti-mouse CD16/32; BioLegend, Cat#156604)), live/dead cell stain (Thermo, Cat#L34967), and fluorophore-conjugated antibodies in FACS buffer (2% FBS in PBS) for 30 mins at 4 °C, followed by two washes. For intracellular staining, cells were fixed and permeabilized using a fix/perm kit (BD, Cat#554714) for 20 min at 4 °C. HA was detected with anti-HA antibody (Cell Signaling Technology, Cat#3724 L) diluted 1:200 in perm/wash buffer and incubated for 30 mins at 4 °C. After two washes, cells were resuspended in perm/wash buffer containing 1:500 AF488-conjugated goat anti-rabbit secondary antibody and stained for 30 mins, then washed twice and resuspended in perm/wash buffer for flow cytometric analysis. ELISA Blood was collected via retro-orbital sinus puncture. After clotting at room temperature for 1.5 , samples were centrifuged at 1000 × g for 15 min at 4 °C, and serum was collected. Cytokine levels were quantified using a double-antibody one-step sandwich ELISA (Yuanxin Bio, IL-1β:YX-091201; IL-6:YX-091206; IFNγ: YX-138532; TNFα: YX-138183). Briefly, standards and diluted samples (10 μl sample plus 40 μl sample diluent per well) were added to antibody-precoated 96-well plates, followed by 100 μl HRP-conjugated detection antibody and incubation at 37 °C for 60 min. After five washes with diluted wash buffer, 50 μl each of substrates A and B were added and incubated at 37 °C for 15 min in the dark, and the reaction was stopped with 50 μl stop solution. Absorbance at 450 nm was measured within 15 min, and cytokine concentrations were calculated from a linear standard curve generated by plotting OD values against standard concentrations, taking into account the 5-fold sample dilution. qPCR primersGeneForwardReverseIl1 βTGTGTAATGAAAGACGGCACACTCAAACTCCACTTTGCTCTTGAIl6AGACAAAGCCAGAGTCCTTCAGTTAGGAGAGCATTGGAAATTGG RNA-seq data analysis Raw paired-end reads were first trimmed to remove adapters and low-quality reads using fastp (v0.23.227). For samples with spike-in, the filtered reads were then aligned to the hybrid human GRCh38 (hg38) and mouse GRCm38 (mm10) genomes28 using HISAT2 (v2.2.1) with the parameters “--no-mixed” and “--no-discordant”29, otherwise, only the human or mouse genome was used as the reference. Only alignments that are unique proper-pair alignments (-q 2; -f 99, -f 147, -f 83, -f 163) were selected using SAMtools30. Reads mapped to the ENCODE blacklist regions31 were further filtered out using BEDTools (v2.30.032). RNA-seq signal was normalized to the library size (CPM) and coverage profiles were generated using deeptools (v3.5.133). For visualization purposes, normalized BigWig files for replicates were merged. ATAC-seq data analysis Raw reads were processed as in RNA-seq analysis. Trimmed reads were aligned to the mouse genome reference (GRCm38) using Bowtie2 (v2.5.334). Only alignments that are unique proper-pair alignments (-q 30; -f 99, -f 147, -f 83, -f 163) were selected using SAMtools (v1.1730), and PCR duplicates were removed using Picard (v2.26.10). Reads mapped to the ENCODE blacklist regions31 were further filtered out using BEDTools (v2.30.032). ATAC-seq peaks were called using MACS2 (v2.2.7.135) using the parameter ‘macs2 callpeak --bdg -f BAMPE -g mm --keep-dup all --nolambda’. The R DiffBind package (DBA_DESEQ2) was used to identify ATAC-seq peaks that are differentially accessible upon LPS treatment. ChIP-seq data analysis Raw reads were preprocessed as in RNA-seq analysis. Trimmed reads were aligned to the hybrid human GRCh38 (hg38) and mouse GRCm38 (mm10) genomes using Bowtie2 (v2.5.334). Only alignments that are unique proper-pair alignments (-q 30; -f 99, -f 147, -f 83, -f 163) were selected using SAMtools (v1.1730), and PCR duplicates were removed using Picard (v2.26.10). Reads mapped to the ENCODE blacklist regions31 were further filtered out using BEDTools (v2.30.032). ChIP-seq peaks were called using MACS2 (v2.2.7.135) using the parameter ‘macs2 callpeak --bdg -f BAMPE -g hs --keep-dup all’. The R DiffBind (v3.12.0) package was used to identify ChIP-seq peaks that are differential upon 5FU treatment, with normalization using spike-in tracks from reads mapped to the mouse genome. Peaks were further annotated using the R ChIPseeker package36 using DBA_DESEQ2. De novo transcriptome assembly and TU annotation To characterize the nascent transcriptome, StringTie (v2.1.737) de novo assembly was performed on RNA-seq data. Reads from biological replicates were merged as input, and StringTie was run with the parameters -f 0.5 -g 200 -c 2 -j 3 –rf38. The featureCounts tool from the Subread package (v2.0.1) was used to quantify transcript expression with the -O option to count reads on overlapping TUs, and transcripts with at least 1 transcript per million (TPM) in all replicates were kept for subsequent analysis. The GffCompare utility (v0.12.639) was used to compare the assembled transcripts with protein-coding genes from the GENCODE (v46 for human and vM23 for mouse) annotation. A single representative transcript per protein-coding gene was selected using the “knownCanonical” table from the UCSC database (https://hgdownload.soe.ucsc.edu/goldenPath/hg38/database/knownCanonical.txt.gz for human; https://hgdownload.soe.ucsc.edu/goldenPath/mm10/database/knownCanonical.txt.gz for mouse). TUs with classification codes ‘=’ ‘c’, ‘k’, ‘m’, ‘n’, ‘j’, ‘e’ were annotated as mRNA, and those mapped to the same protein-coding gene were merged. Remaining TUs with codes ‘x’, ‘i’ or ‘u’ were selected40 to identify ncRNAs based on their overlap with annotations calculated using BEDTools (intersect, v2.30.032), requiring that either 50% of the TU or 50% of the annotation is covered; otherwise, they were classified as unannotated TUs. Common ncRNA annotations (e.g., lncRNA) were taken from GENCODE (excluding the biotype of “protein-coding”); annotations for enhancers, tRNA and transposable elements were retrieved from the UCSC Table Browser. Enhancer: (Group: Expression and regulation; Track: ENCODE cCREs); tRNA: (Group: Genes and Gene Predictions; Track: tRNA Genes); transposable elements: (Group: Repeats; Track: RepeatMasker). PROMPTs were defined as TUs within 1 kb from 5’ end and on the antisense strand of protein-coding gene annotations. Differential expression analysis Protein-coding gene and ncRNA quantification were performed using featureCounts. Differential expression analysis was performed using the DESeq2 R package (v1.32.041), and size factors for each sample were calculated using the “estimateSizeFactors” function in DESeq2 and protein-coding gene counts were used as the library size for both genes and ncRNAs42. GO analysis Gene Ontology enrichment analysis was performed using clusterProfiler R package (v4.10.143) and genome-wide annotation for mouse org.Mm.eg.db. Assignment of reads to the human and mouse genome GENCODE (v46 for human and vM23 for mouse) annotated protein-coding genes and the “knownCanonical” table from the UCSC database were used to select a single representative transcript per gene and define exons and introns. Noncoding RNA genes overlapping regions of protein-coding genes, regardless of the strand, were removed. Regions that are not annotated as protein-coding nor noncoding RNA genes were defined as intergenic. Read counts for exons, introns, noncoding RNAs, and intergenic regions were calculated with featureCounts. Calculation of intron-to-exon ratio The coverage of introns with respect to exons for each protein-coding gene was obtained as the ratio of total intronic signal to total exonic signal (both in RPK)9. For genome-level calculation, all introns were rescaled to 1 kb using the computeMatrix from deeptools (v3.5.133) and the intron-to-exon ratio was obtained as the average intronic signal relative to the maximum signal in the last 50 bp upstream of the 5’ splice site. TF footprinting analysis The TF footprinting analysis of the ATAC-seq peaks was performed using TOBIAS (v.0.17.3)44. Briefly, mouse TF motifs were downloaded from the HOCOMOCO V11 databases22 (HOCOMOCOv11_core_MOUSE_mono_meme_format.meme). BAM files from ATAC-seq data were processed using the TOBIAS ‘ATACorrect’ function to account for Tn5 insertion bias, followed by the ‘ScoreBigwig’ and ‘BINDetect’ functions to estimate differentially bound TF binding sites. All TFs with -log10 (p-value) above the 95% quantile or differential binding scores below the 5th or above the 95th percentile (top 5% in each direction) were considered differential binding TFs. Motif analysis Motif enrichment in enhancers was conducted using HOMER (v5.145) (findMotifsGenome.pl -mknown = HOCOMOCOv11_core_MOUSE_mono_homer_format_0.001.motif -size given)46. Statistical analyses The statistical tests performed are described in the figure legends. For all analyses, “n” refers to biological replicates. For cell culture experiments, biological replicates are independently processed samples starting from cell plating. For mouse experiments, each mouse represents one biological replicate. For the experiments with technical replicates, the mean of the technical replicates was used as one data point. Supplementary information Supplementary Information Description of Additional Supplementary Files Supplementary Data 1-8 Transparent Peer Review file Source data Source Data Acknowledgements We thank Tsinghua University School of Basic Medical Sciences Bioinformatics Core Services for providing the server for sequencing data analysis. We thank Tsinghua University Technology Center for Protein Science for performing mass spectrometry, and the Biomedical Analysis Center in Tsinghua University for FACS analysis. Author contributions H.B., F.X.C., and Y.F. conceived the project, designed the experiments and wrote the manuscript with contributions from C.J.D. and Y.Z. Y.Z. performed the bioinformatics analysis. Y.F., H.L., Z.W., and J.W. performed experiments. L.W. contributed a mouse line. H.B. supervised the project. Peer review Peer review information Nature Communications thanks Alessandro Gardini, and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available. Funding This work was supported by start-up funds from Tsinghua University and Peking University-Tsinghua Center for Life Sciences, and Tsinghua University Initiative Scientific Research Program to HB. Data availability The data supporting the findings of this study are available from the corresponding authors upon request. The sequencing data generated in this study have been deposited in NCBI under Bioproject accession number PRJNA1230698. Source data for the figures and Supplementary Figs. are provided as a Source Data file. Source data are provided with this paper. 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. These authors contributed equally: Yuxuan Feng, Yiwei Zhang. Contributor Information Fei Xavier Chen, Email: fei.chen@sibcb.ac.cn. Harihar Basnet, Email: hbasnet@tsinghua.edu.cn. Supplementary information The online version contains supplementary material available at 10.1038/s41467-026-74725-2. References Associated Data Supplementary Materials Supplementary Information Description of Additional Supplementary Files Supplementary Data 1-8 Transparent Peer Review file Source Data Data Availability Statement The data supporting the findings of this study are available from the corresponding authors upon request. The sequencing data generated in this study have been deposited in NCBI under Bioproject accession number PRJNA1230698. Source data for the figures and Supplementary Figs. are provided as a Source Data file. Source data are provided with this paper.