The new technique, called spatial NT-seq, was described in Nature Neuroscience by Qi Qiu, Hongjie Zhang, Zijie Xia and colleagues in the laboratories of Hao Wu, Zhaolan Zhou, Hongjun Song, Guo-li Ming and Erica Korb. The method addresses a long-standing technical gap. Standard single-cell and spatial transcriptomics measure total RNA abundance, which reflects only the balance between synthesis and degradation. Metabolic RNA labeling can distinguish newly transcribed RNA from pre-existing pools, but existing approaches that combine labeling with single-cell sequencing have been largely restricted to cultured cells. Applying them to living animals, particularly to the brain with its diverse cell types and dense tissue architecture, has been a persistent challenge.

The researchers began by systematically validating a transgenesis-free labeling strategy based on 4-thiouridine, or 4sU, a cell-permeable nucleoside analog that cells incorporate into RNA as they synthesize it. When injected into mice, 4sU crosses both the fetal–maternal and blood–brain barriers and labels nascent transcripts in vivo. The team benchmarked this approach against the widely used TU-tagging method, which requires transgenic expression of a Toxoplasma enzyme to convert a related analog into a form that can be incorporated into RNA. In mice engineered to express this enzyme specifically in excitatory forebrain neurons, both 4sU and 4-thiouracil labeled new RNA robustly in the targeted neurons, with highly concordant transcriptome-wide measurements. Critically, 4sU also efficiently labeled RNA in cell types that lacked the transgene, whereas the transgene-dependent analog did not. This means 4sU achieves labeling efficiency comparable to the genetic method but with far broader applicability and minimal cytotoxicity.

Across more than 26,000 cortical cells from postnatal mice, the researchers showed that labeled RNA fractions increased almost linearly with labeling duration, were reproducible across biological replicates and cell types, and plateaued at a defined dose. Liquid chromatography–tandem mass spectrometry confirmed that cortical 4sU concentrations rose for two hours after injection and then stabilized. The labeling minimally perturbed cell-type composition, mapping rates, or library complexity, and had little effect on mRNA splicing. To validate that the measurements reflected intrinsic RNA processing rather than local variability in 4sU availability, the team compared labeling-based new-to-total RNA ratios with labeling-free unspliced-to-total ratios derived from RNA splicing. The significant correlations between these independent measures suggested that the labeling approach captures genuine cell-type-specific RNA turnover.

With in vivo labeling validated, the team built spatial NT-seq by combining the transgenesis-free 4sU strategy with in situ chemical recoding on the widely used Visium spatial transcriptomics platform. The key chemical step, an optimized SLAM reaction performed directly within tissue sections, covalently attaches a carboxyamidomethyl group to the thiol in 4sU. This modified base is then transformed into a cytosine analog during reverse transcription, inducing thymine-to-cytosine substitutions specifically in newly synthesized RNA. Because the reaction occurs inside the tissue section, the spatial architecture is preserved, and modified reverse transcription and second-strand synthesis reactions allow efficient capture of both new and pre-existing transcriptomes from the same section.

Benchmarking on postnatal mouse brains showed that spatial NT-seq achieved a 14.6-fold increase in thymine-to-cytosine conversion compared to standard Visium while preserving library complexity, alignment rates, and spatial cluster distributions. Clustering of over 10,000 spatial spots identified 22 distinct spatial clusters, annotated using the Allen Common Coordinate Framework and validated by Cell2location-based integration with a whole-brain single-nucleus RNA-seq reference. The method also recapitulated the higher turnover rates of transcription factors relative to housekeeping genes observed by single-cell sequencing, whereas splicing-based analysis failed to robustly capture spatial new RNA profiles, likely because shorter or intron-less genes are sparsely covered by intronic reads.

Applying spatial NT-seq to map RNA turnover across the postnatal mouse brain revealed pronounced regional heterogeneity. Most regions showed similar new RNA fractions, averaging 25.7 percent, but the dentate gyrus stood out dramatically at 44.8 percent, with cortical layer 2/3 and CA1 also elevated. These differences were unlikely to be driven by local 4sU availability, since vascular density is comparable or lower in the dentate gyrus and the probability of thymine-to-cytosine substitutions in newly synthesized transcripts was comparable across regions. The regional differences therefore reflect intrinsic regulation of RNA metabolism. Notably, key components of major RNA degradation pathways — including nonsense-mediated decay, N6-methyladenosine-mediated decay, the CCR4–NOT deadenylase complex, and the microRNA pathway — were upregulated in the dentate gyrus relative to other regions, as was Uck1, a rate-limiting enzyme in the pyrimidine salvage pathway that supports new RNA synthesis.

To investigate how spatially distinct RNA turnover contributes to activity-dependent gene regulation, the team applied spatial NT-seq to adult mice subjected to electroconvulsive stimulation, a clinically effective treatment for refractory depression. Mice received three hours of 4sU labeling and were collected at zero, thirty, and one hundred twenty minutes post-stimulation. Integrated analysis of over 44,000 spatial spots across sixteen brain sections identified eighteen major spatial clusters. Immediate early genes such as Fos and Egr1 were rapidly and reproducibly detected in newly synthesized RNA, but not in old RNA, within the dentate gyrus across biological replicates. Labeling-based RNA velocity analysis revealed early gene activation within thirty minutes, demonstrating the high temporal resolution of the approach. By contrast, splicing-based analysis failed to resolve new and old RNA dynamics for these genes and was less effective at capturing activation signatures across brain regions.

The joint analysis of newly synthesized and pre-existing transcriptomes within the same spatial spot revealed that the dentate gyrus was the most ECS-responsive region, exhibiting significant changes in both new and old RNA levels. The researchers classified activity-regulated genes by their temporal patterns: seventy-four early-response genes, including Fos, Egr1 and Btg2, peaked within thirty minutes and often encode transcription factors that drive the activation of late-response genes; two hundred ninety-one late-response genes, including Nptx2, Bdnf and Stmn4, were linked to synaptic functions. The analysis also identified sixty-six genes with progressively reduced new RNA levels, suggesting transcriptional repression, and fifty-two genes likely regulated through ECS-induced mRNA decay. Some genes appeared co-regulated through both mechanisms, while others showed opposing effects between transcriptional activation and post-transcriptional degradation, which may serve to buffer expression levels. Comparison with curated disease gene sets showed that transcriptionally activated genes were enriched for those associated with major depressive disorder, mood disorder and schizophrenia, whereas genes regulated through old RNA decay were enriched for bipolar disorder-related genes.

Beyond the spatial map of turnover, the team developed a computational framework called RNAkinetoScope to estimate transcriptome-wide RNA synthesis and degradation rates at single-cell or single-spot resolution. By leveraging a k-nearest neighbor graph to aggregate newly synthesized and pre-existing transcriptomes from neighboring cells or spots, the framework improves steady-state RNA kinetics inference while preserving intracluster heterogeneity. The analysis revealed an average RNA half-life of 4.6 hours across cortical cell types, consistent with previous estimates from cultured cells, but with transcriptome-wide half-lives spanning more than two orders of magnitude, from less than one hour to over one hundred hours. The team also identified distinct RNA kinetics regulatory strategies across gene modules: cooperative modules showed negative correlations between synthesis and degradation rates, neutral modules showed weak or no correlation, and destabilizing modules showed positive correlations. Genes regulated by different strategies were associated with distinct cellular functions, with fast-turnover modules enriched for neuron-specific transcription factors and slow-turnover modules preferentially encoding synapse-localized proteins.

Finally, the researchers applied machine learning to their in vivo RNA half-life data to identify the sequence features and post-transcriptional regulators that shape mRNA stability across spatial and cellular contexts. A least absolute shrinkage and selection operator regression model, trained on features including basic mRNA properties, codon frequencies, three-prime untranslated region motifs, microRNA target scores and RNA-binding protein binding scores, showed strong correlations between predicted and observed half-lives across cell types and spatial regions. The model recovered known relationships, such as the positive correlation of open reading frame exon junction density with stability and the negative association of three-prime UTR length. It also uncovered known regulatory motifs, including the destabilizing AU-rich element and the stabilizing cleavage and polyadenylation signal. Notably, the dentate gyrus showed distinct codon regulatory profiles relative to cortical layers, suggesting brain region-specific modulation of codon optimality-mediated mRNA degradation. The model revealed that most microRNAs destabilize mRNAs, with brain-enriched microRNAs such as miR-124 and miR-137 showing strong destabilizing effects predominantly in excitatory neurons, while other microRNAs exhibited dentate gyrus-specific effects. RNA-binding proteins showed a balanced mix of stabilizing and destabilizing effects, with predicted effects of well-characterized proteins largely aligning with prior in vitro studies.

Together, this integrated framework — combining transgenesis-free RNA labeling, spatial co-detection of newly synthesized and pre-existing RNAs, and predictive modeling — provides a spatially resolved view of RNA turnover kinetics that the authors describe as an in vivo timescope. The findings suggest that enhanced RNA turnover in the dentate gyrus may allow its cells to rapidly adjust protein output from existing transcript pools, supporting cellular plasticity during adult neurogenesis and responses to external stimuli such as learning and electroconvulsive stimulation. The approach should be broadly adaptable for investigating RNA turnover and stability regulation across diverse in vivo model organisms, ex vivo human tissue systems and in vitro human organoid models, and future integration with emerging single-cell resolution spatial transcriptomics platforms and multiple-pulse labeling using orthogonal nucleoside analogs will further illuminate the regulatory interplay between RNA synthesis and decay across complex tissue environments.

Subject of Research: Spatial mapping of RNA turnover kinetics in the mouse brain

Article Title: Spatial mapping of RNA turnover kinetics in the mouse brain

Article References: Qiu, Q., Zhang, H., Xia, Z., Gao, W., Leu, J. I., Liang, D., Li, Y., Li, F., Su, Y., Feierman, E., Van Horn, E., Ming, G.-L., Korb, E., Song, H., Zhou, Z., & Wu, H. (2026). Spatial mapping of RNA turnover kinetics in the mouse brain. Nature Neuroscience. https://doi.org/10.1038/s41593-026-02420-y

Image Credits: AI Generated

DOI: 10.1038/s41593-026-02420-y

Keywords: spatial transcriptomics, RNA turnover, metabolic labeling, neuroscience, dentate gyrus, electroconvulsive stimulation, machine learning, mRNA stability, mouse brain, Nature Neuroscience, Spatial, mapping