The study began with developing an efficient chemical probe for RNA labelling under physiological conditions (Fig. 1a). Our recent investigations have focused on the protein labelling capabilities of QM compounds, which could be precisely controlled through a bio-orthogonal photocatalytic reaction. Intriguingly, certain QM compounds have also been demonstrated reactivity towards nucleosides. These findings imply the possibility of developing QM-based probes for RNA labelling. Therefore, we designed a series of QM-derived probes incorporating an ultraviolet (UV) removable protection group (MNP) for UV activation and a biotin tag for immunoblotting detection (Fig. 1b). These probes included both classical oxygen-QMs and our newly introduced thio-QMs, which we have recently developed as efficient protein labelling warheads. We initially evaluated the reactivity of these probes using a streptavidin dot blot assay with a 1,000-nt RNA template produced by in vitro transcription. Interestingly, QM-1 displayed markedly higher labelling efficiency compared with its analogues, showcasing a distinct performance contrast to their efficiency in protein labelling (Fig. 1c).

To identify the specific nucleoside targeted by QM-1, we conducted labelling reactions with single-stranded DNA composed of defined nucleotides. Dot-blot analysis indicated cytidine as the dominant reactive nucleoside (Fig. 1d and Extended Data Fig. 1a). This selectivity was further confirmed through liquid chromatography tandem mass spectrometry (LC-MS/MS) analysis of nucleosides digested from labelled RNA, which showed a dominant labelling adduct on cytidine with QM-1 (Fig. 1e). A small-molecule reaction between cytidine and QM-1 analogues resulted in the formation of a direct N-C bond formation between the N of cytidine and the methide carbon of the QM (Fig. 1f and Extended Data Fig. 1b-d). This suggests a potential hydrogen bond-assisted conjugation pathway during cytidine labelling, in line with previous studies. In contrast, the protein labelling by the thio-QM probably favour a nucleophilic addition pathway owing to the easier formation of the reactive zwitterion resonance of the thio-QM. These distinct reaction pathways may explain the observed preferential affinity of QM-1 for RNA and QM-4 for proteins, respectively (Extended Data Fig. 1e).

Encouraged by the efficient labelling with QM-1, we next aimed to develop a photocatalytic version of the RNA labelling system to achieve spatiotemporal precision in living systems (Fig. 1g). To this end, QM-1 was designed and synthesized, incorporating a para-azidobenzyl group (PAB) in place of MNP. This group undergoes photocatalytic deprotection under blue LED light in the presence of an iridium catalyst (Fig. 1h). Dot-blot analysis confirmed that purified HeLa cell RNA could be biotinylated by QM-1 in the presence of iridium catalyst and blue LED irradiation in vitro, displaying increased labelling intensity with higher light power (Fig. 1i). As expected, the labelling intensity was correlated with the catalyst loading, illustrating the capability of this strategy for spatial control of RNA labelling in living systems through targeted catalyst deployment (Fig. 1j). Pulse-chase labelling experiments utilizing light switching further verified the temporal control achievable by LED light (Fig. 1k), proving the successful development of a photocatalytic chemistry-enabled RNA labelling system.

Encouraged by the successful in vitro labelling with CAT-seq, our focus shifted to subcellular-resolved RNA labelling in living cells (Fig. 2a and Extended Data Fig. 2a). Mitochondria emerged as a prime target organelle because of (1) their critical and unique, independent transcriptional system, which makes them an intriguing subject for studying RNA dynamics, (2) their relatively lower RNA abundance compared with the nucleus, which serves as a reference, providing an advantageous context for evaluating the subcellular specificity of our labelling approach and (3) prior observations suggesting that the iridium catalyst itself can localize to mitochondria, eliminating the need for additional catalyst modifications. It has been proved that the photocatalyst can target to mitochondria (Extended Data Fig. 2b). However, labelling signals were also detected in the nucleus (Extended Data Fig. 2c). Further enrichment following cell lysis and subsequent quantitative polymerase chain reaction with reverse transcription (RT-qPCR) analysis of the labelled RNA showed insufficient mitochondrial specificity (Extended Data Fig. 2d). To address the probe diffusion issue, we redesigned the labelling probe by introducing a mitochondria-trapping triphenylphosphine moiety (QM-2; Fig. 2b). Satisfactory mitochondria specificity was achieved using the new probe QM-2 (Pearson's R = 0.82; Fig. 2c and Extended Data Fig. 2e-h) without mitochondrial metabolism perturbation (Supplementary Fig. 1).

Thereafter, we focussed on investigating the internal mechanism and validation in living cells. RT-qPCR analysis after biotin enrichment demonstrated high specificity for mitochondrial genes MTCO1 and MTND1 compared with cytosolic mRNA references (Fig. 2d and Supplementary Fig. 2a,b). Also, compared with a non-photocatalytic labelling system, CAT-seq can capture more mtRNA, displaying the strength of this photocatalytic-based assay (Supplementary Fig. 2c). To further unveil the composition of the labelled RNA, we used CAT-seq for biotin tagging and enrichment followed by high-throughput sequencing. Notably, we observed evident and robust enrichment of mitochondrial RNA with ideal reproducibility across two biological replicates (62.9% versus 1.54% specificity; Fig. 2e and Extended Data Fig. 3a), which remarkably surpasses the 6.08% specificity achieved by conventional mitochondria isolation methods (Extended Data Fig. 3b) and underscores the sufficient specificity of CAT-seq for subcellular RNA studies. CAT-seq was also compared with other established enzymatic proximity labelling tools APEX-seq and CAP-seq, showing similar performance (Extended Data Fig. 3c-g). The overall coverage of mitochondrial transcripts in the biotin immunoprecipitation (IP) samples demonstrated a noteworthy augmentation compared with the cellular transcriptome (input; Fig. 2f). Specifically, all 13 mRNAs and 2 rRNAs encoded by the mitochondrial genome were notably enriched with high reproducibility across two biological replicates (Fig. 2g). Given the complexity of nucleic acids in living systems, such as DNA and structured RNA, we next investigated their labelling efficiency and found that only exposed cytidine residues can be labelled (Supplementary Figs. 3 and 4). To further unveil the impact of RNA structure and composition on CAT-seq, we also analysed the cytidine proportions in mitochondrial mRNAs and rRNAs. This analysis demonstrated that CAT-seq achieves unbiased amplification of mitochondrial transcripts (Supplementary Figs. 5 and 6).

In addition, we also observed a notable enrichment of mitochondrial tRNAs over their nuclear counterparts, which are often challenging targets for existing labelling methods owing to their short sequence and complicated secondary structure (Fig. 2h).

These results collectively demonstrate the successful development of the chemical method CAT-seq for achieving robust and specific labelling of diverse mitochondrial RNAs in living cells, opening exciting avenues for in situ exploration of the dynamics of mitochondrial RNAs and their distinct roles in cellular processes (Fig. 2i).

M1 macrophages play a pivotal role in the immune system, orchestrating potent responses against invading pathogens. Their polarization state greatly impacts the intensity and immunogenicity of these responses. One intriguing aspect of M1 polarization is the observed decrease in oxidative phosphorylation (OXPHOS) levels, indicating alterations in the complex machinery within mitochondria -- the powerhouse of the cell. Despite its importance, the mechanisms underlying this metabolic remodelling process remain largely uncharacterized. Notably, the mitochondrial genome encodes 13 crucial subunits of OXPHOS complexes, hinting at potential changes in the mitochondria during this process. Leveraging the unique advantage of CAT-seq for hard-to-transfect cells, we aimed to examine the mitochondrial RNA dynamics during macrophage M1 polarization.

Hence, we employed CAT-seq to investigate polarized RAW 264.7 cells triggered by lipopolysaccharide (LPS), a potent inducer of M1 polarization (Fig. 3a and Extended Data Fig. 4a). The overall sequencing alignment demonstrated that the mitochondrial transcriptome could be effectively captured with over 40% alignment rates, regardless of 24-h LPS stimulation or not (Fig. 3b and Extended Data Fig. 4b). The alignment with the mitochondrial genome further proved that the RNA captured by CAT-seq predominantly originated from mitochondrial transcripts (Extended Data Fig. 4c). To delve deeper into the regulatory mechanisms, we conducted a more comprehensive analysis of the mitochondrial transcriptome in RAW 264.7 cells across varying durations of LPS exposure. The samples subjected to different duration of LPS stimulation exhibited dynamic changes in the mitochondrial transcriptome, as characterized by principal component analysis (PCA), which revealed a transcriptional shift trace from 0 to 24 h (Fig. 3c). Upon closer examination, mitochondrial-encoded mRNAs were found to have a more substantial impact than tRNAs and rRNAs on these changes (Extended Data Fig. 4d).

These results encouraged us to further investigate the connection between mitochondrial RNA dynamics and macrophage M1 polarization. We first confirmed the phenomenon of OXPHOS downregulation during M1 polarization (Fig. 3d). However, a deeper analysis of the changes in mitochondrial RNA abundance showed no dramatic shifts (log fold change >0.58, P < 0.01; Extended Data Fig. 4e) and only a subtle increase in most components of mitochondrial mRNA during LPS stimulation, as visualized in the heatmap (Fig. 3e). These changes were insufficient and conflicting to account for the dramatic downregulation of OXPHOS, suggesting that transcriptional alterations have little impact on this process. Given the central role of mitochondria in OXPHOS activity, we considered that other factors, such as RNA modifications, could be involved in this process. We therefore separated the enriched RNA samples into small-length ( < 200 nt) and long-length ( > 200 nt) groups, and analysed several commonly appearing modifications in mitochondria (Fig. 3f and Extended Data Fig. 4f). Specifically, an obvious decrease in two tRNA modifications, 5-carboxymethylaminomethyl (τmU) and 5-methylcytosine (mC), was observed across the small-length RNA groups, implying the potential critical role of RNA modifications, beyond mere RNA abundance, in the decrease of OXPHOS during macrophage M1 polarization.

To further elucidate the molecular pathway within mitochondria during this process, we sought to uncover the underlying reasons for the decrease in RNA modification. Given the executive function of proteins in altering RNA modifications, we explored the mitochondrial protein dynamics by taking advantage of our previously developed photocatalytic protein profiling technique (CAT-S). In accordance with the observed OXPHOS decrease, there was a general downregulation in the mitochondrial proteome during the M1 polarization (Fig. 3g). Notably, visible reductions were observed in the writer proteins for the two most downregulated RNA modifications, writer protein of mC (NSUN2) and writer protein of τmU (GTPBP3), responsible for mitochondria tRNA mC and τmU, respectively, as shown in the volcano plot (Fig. 3g). To verify the protein changes detected by MS/MS, we performed western blot analysis for these two writer proteins (Fig. 3h,i). Interestingly, while GTPBP3 showed a decrease in both whole cell and mitochondria fractionation, NSUN2 levels remained unchanged in the whole cell but decreased in the mitochondria fraction, highlighting the complexity of the biomacromolecule network in this process and the necessity for subcellular-resolved dynamic investigations (Fig. 3h,i). These results indicated the writer protein alterations as the possible underlying cause of the observed changes in RNA modifications. Additionally, the observed reductions in protein and modification levels may reflect a non-linear correlation, consistent with findings from some previous research on other RNA modifications.

Previous reports have proved that specific modifications on tRNAs are involved in the regulation of mitochondrial RNA translation, so we are now interested in investigating their relationship with the M1 polarization process in macrophage using CAT-seq. As we observed a decrease in NSUN2/GTPBP3-mediated tRNA modifications and no evident change in the mitochondrial transcription level during macrophage M1 polarization, we wondered whether such effects contribute to the reduction in OXPHOS activity through a mitochondrial RNA translational regulation pathway. To validate this, we first knocked down NSUN2 and GTPBP3 in RAW 264.7 cells using shRNAs (Extended Data Fig. 4g). We indeed observed a marked downregulation of OXPHOS activity that mirrored the observed phenotype in LPS-stimulated RAW 264.7 cells (Fig. 3i), suggesting the involvement of these two tRNA modifications in OXPHOS activity regulation during macrophage M1 polarization.

We next analysed the mitochondrial translation level by employing O-propargyl-puromycin (OPP) labelling of nascent proteins in RAW 264.7 cells, with or without LPS treatment, followed by mitochondrial isolation and nascent protein detection. As expected, a notable global reduction in mitochondria translation level was observed in response to LPS treatment, aligning with our protein dynamic results (Fig. 3g,j). This suggests that LPS-induced OXPHOS downregulation stems from translational inhibition within mitochondria, rather than protein degradation. Furthermore, both NSUN2 and GTPBP3 knockdown cells showed a notable decrease in nascent protein synthesis efficiency, mirroring the effects of LPS treatment. This indicates that tRNA modification changes mediated by these writer proteins influence mitochondrial translation during macrophage M1 polarization (Fig. 3k,l). Indeed, previous studies have highlighted the critical role of GTPBP3 and τmU modifications in mitochondrial translation in HEK293T cell lines, further validating the capability of our CAT-seq technique in deciphering these mechanisms. However, our experiments (Extended Data Fig. 4h,i) and prior reports indicate that NSUN2 knockdown has a minimal impact on mitochondrial translation in HEK293T cells, highlighting the importance of cell type-specific differences and warranting further detailed biological investigation. In the context of macrophage M1 polarization, the observed reduction in mitochondrial tRNA stability owing to decreased mC and τmU modifications provides further support for this mechanism. It is plausible that reduced levels of writer proteins lead to a decrease in tRNA modifications, resulting in tRNA instability, which subsequently lowers mitochondrial translation efficiency and downregulates OXPHOS activity during M1 polarization (Fig. 3m).

Another fascinating advantage of our chemical approach is its capability to study intact primary living samples directly, eliminating the need for genetic manipulation. Considering the often-precious resources of primary samples and the desire to maximize information retrieval from limited samples, we explored whether CAT-seq could be orthogonal to a protein labelling method in the same sample. Such an orthogonal labelling strategy would not only conserve sample material but also minimize the unavoidable variances between different sample groups (Fig. 4a). The major challenge in this endeavour was achieving simultaneously orthogonal labelling of RNA and protein. Nonetheless, our earlier observation of distinct reactivity preferences among different QMs towards RNA and protein presented a unique opportunity (Figs. 1cand 4a and Extended Data Fig. 1e). Coupled with the potential of our decaging strategy, which is capable of releasing diverse QMs by the same photocatalytic chemistry, we envisioned a promising and valuable opportunity for developing a synchronous RNA and protein labelling approach.

To this end, we employed probe QM-3, which features a reactive unit (thio-QM with double fluorine substitution) tailored in our reported protein labelling tool CAT-S and a distinctive alkyne handle to differentiate it from the biotin handle on the RNA probe QM-2 (Fig. 4a). In vitro tests using BSA as a model protein with QM-1 and QM-4 confirmed the preferential protein labelling of QM-4 (Supplementary Fig. 7a). Subsequently, we evaluated orthogonal labelling with both dot-blotting and immunoblotting assays. To our delight, 1,000-nt template RNA was primarily labelled by QM-1, while QM-4 showed a stronger tendency towards protein labelling (Fig. 4b). Transitioning to living cells, only HeLa cells treated by QM-3 displayed protein labelling via the biotin click reaction, and negligible RNA labelling was detected with QM-3 (Fig. 4c). RT-qPCR of the enriched RNA further confirmed that the selective mitochondrial RNA labelling was unaffected by the synchronous labelling approach (Fig. 4d). Finally, immunofluorescent imaging of the synchronously labelled cells revealed good colocalization of both biotin and alkyne immunofluorescence with mitochondrial marker TOMM20, demonstrating successful simultaneous labelling of both RNA and proteins within mitochondria (Fig. 4e).

To comprehensively evaluate our newly developed synchronous labelling method, we conducted high-throughput sequencing for RNA and tandem mass spectrometry for protein analysis. Notably, the enriched RNA profiles from both CAT-seq and synchronous labelling approaches displayed excellent correlation across all mitochondria RNAs (Fig. 4f). Volcano plot analysis further confirmed the enrichment of mitochondrial RNA components (Fig. 4g) and provided accurate coverage of the mitochondria genome (Fig. 4h). For protein analysis, tandem mass spectrometry identified a total of 504 enriched proteins using the alkyne handle. Among these proteins, 378 (75.0% specificity) were identified as mitochondrial on the basis of MitoCarta3.0 annotations (Fig. 4i and Supplementary Fig. 7b). Together, by taking advantage of the unique QM chemistry and photocatalytic decaging strategy, we successfully developed a synchronous labelling system, which offers a unique platform for synchronized analysis of subcellular RNA and protein within the same sample, minimizing potential biases from sample heterogeneity and alterations.

Encouraged by the successful development of our synchronous labelling tool, we next focussed on investigating the mitochondria-related biological process in primary cells. Recent discoveries underscore the importance of mitochondrial translation in the maintenance of the cytotoxic capabilities of T cells, emphasizing the critical role of mitochondrial RNA and protein T cell biology. Herein, we were interested in examining the molecular dynamics during the transformation process of naive T lymphocytes (NTLs) into cytotoxic T lymphocytes (CTLs) using intact primary T cells from mice. CD8 CTLs are pivotal in adaptive immune response, with augmented mitochondrial respiration and OXPHOS being essential during their activation and target-cell killing processes. The properties of primary CD8 NTLs and CTLs hinder traditional enzyme-based labelling strategy for subcellular multi-omics profiling. Thus, the underlying molecular dynamics of mitochondrial remodelling during NTL-to-CTL transformation remains largely unexplored. Leveraging the advantage of our synchronous labelling tool, which requires no genetic manipulation, we focussed on deciphering the multi-omics variation during the primary T lymphocyte transformation (Fig. 5a).

We isolated primary NTLs from the spleen of wild-type C57BL/6N mice and then directly subjected them to synchronous labelling. The labelled RNAs and proteins were enriched via biotin and alkyne handles, respectively. We confirmed the sufficient specificity of the mitochondrial RNA enrichment for primary NTLs through RT-qPCR detection (Fig. 5b). Then, the western blot test assay verified the efficiency of protein labelling within the same sample (Fig. 5c), and LC-MS/MS analysis showed high mitochondrial specificity of the enriched proteins (60.3% specificity; Fig. 5d).

Next, we derived CTLs from the NTLs by incubating them for 5-day post-activation, during which they acquired cytolytic activity and upregulated CD69 (Fig. 5e and Supplementary Fig. 8a). Following the synchronous labelling, the enriched RNAs and proteins underwent RNA sequencing and tandem MS/MS identification, respectively. PCA revealed an identical transcriptional shift between NTLs and CTLs (Supplementary Fig. 8b), indicating the transcriptional changes during this transformation process. Through comprehensively analysis of the sequencing results, we observed a global upregulation (log fold change >0.58, P value <0.01) in the expression of mitochondrial RNAs during the NTL-to-CTL transformation, particularly mRNAs (Fig. 5f and Supplementary Fig. 8c). To further confirm this, we examined several mitochondrial mRNAs using RT-qPCR; all the tested mRNA showed higher expression level in CTLs (Fig. 5g), and selected mitochondrial-encoded proteins NADH-ubiquinone oxidoreductase chain 5 (MTND5) and cytochrome c oxidase subunit 2 (MTCO2) showed upregulation in western blot (WB) analysis (Fig. 5h,i). To understand the potential regulatory mechanism, we analysed the proteomics results and noticed the remarked upregulation of two proteins, FAKD4 and TEFM (Fig. 5j), which has been reported to stabilize mitochondrial mRNA species or increase mitochondrial RNA polymerase processivity, either of which would increase the expression of mitochondria RNA.

Finally, as the protein translated by mitochondria-encoded RNA is responsible for the oxidative respiratory chain and directly related to mitochondrial oxidative phosphorylation, we then examined the OXPHOS during the NTL-to-CTL transformation process. We observed an evident increase of OXPHOS level in CTLs compared with NTLs (Fig. 5k). This phenotype aligns well with the observed increase in mitochondrial RNA expression. Collectively, the subcellular-resolved molecular dynamics during the NTL-to-CTL transformation process was thus investigated, revealing the involvement of mitochondria-specific transcription-related regulation on OXPHOS (Fig. 5l). While more research is needed to uncover the detailed mechanisms, our findings underscore the effectiveness of our chemical approach in the studies of intact primary living samples and also provide unprecedented molecular insights into the primary living T cells of mice.