We began by repurposing SiR as a photocatalyst. Visible-light photocatalysts exert their catalytic ability through light-induced electron transfer (ET) or energy transfer (EnT). Pioneering works by Macmillan, Rovis, Oslund, and Fadeyi have demonstrated NIR-light activated PL through ET-mediated nitrene generation. In photodynamic therapy (PDT), light-induced EnT between the photosensitizer and molecular oxygen has been explored to generate singlet oxygen for therapeutic purposes. This EnT pathway has similarly been applied in the development of light-activated PL strategies. SiR has been used as a photocatalyst for in situ generation of tetrazines via photooxidation. However, it remains uncertain whether SiR, optimized for fluorescence imaging, can generate sufficient singlet oxygen for PL. To investigate, 10 μM SiR-CA was reacted with 10 μM Halo protein, forming a covalent SiR-CA-Halo complex that turned on SiR fluorescence (Supplementary Fig. 1a). Addition of biotinylated perfluorinated aniline probe B1 (Supplementary Fig. 1b), followed by irradiation with NIR light (660 nm) resulted in Halo protein biotinylation confirmed by western blot analysis (Supplementary Fig. 1c). Substitution of B1 with biotinylated aniline BA increased labeling efficiency, whereas alkylamine B2 reduced it (Supplementary Fig. 1c). The generation of singlet oxygen was verified by adding the singlet oxygen fluorescence sensor SOSG to the reaction. Irradiation with 660 nm LED light increased SOSG fluorescence, indicating singlet oxygen formation (Supplementary Fig. 1d). The intensity of SOSG fluorescence is consistent with the labeling efficiency (Supplementary Fig. 1e). The addition of a singlet oxygen quencher, Vitamin C (NaVc), reduced both SOSG fluorescence and labeling efficiency, further confirming singlet oxygen involvement in Halo protein biotinylation (Supplementary Fig. 1d, e). Notably, the fluorogenic nature of SiR-CA minimizes singlet oxygen generation upon irradiation of SiR-CA alone (Supplementary Fig. 1d). Increasing the concentration of Dioxane in HO decreases the ability of SiR to generate singlet oxygen (Supplementary Fig. 2a, b), confirming that the fluorescent form of SiR is contributing to the photocatalytic activity. To demonstrate that this fluorogenicity reduces nonspecific background labeling, BSA was labeled with either SiR-CA or a non-fluorogenic NIR photosensitizer, Methylene Blue (MB). SiR-CA labeled BSA only in the presence of Halo protein, whereas MB labeled BSA regardless of Halo protein presence (Fig. 1c). In addition, we measured the singlet oxygen quantum yield of SiR-CA in free and Halo-bound states, although the singlet oxygen quantum yield of SiR-CA bound with Halo (Ф = 0.00064) was lower than free SiR-CA (Ф = 0.00075), it is still capable of generating sufficient labeling signals due to its higher absorbance (Supplementary Fig. 1f, g). A noncovalent version of SiR-CA (SiR-no Cl) was synthesized and the compound failed to label BSA in the presence of Halo protein, demonstrated the importance of covalent conjugation between SiR-CA and Halo protein for singlet oxygen generation (Supplementary Fig. 3a-d). Various nucleophilic probes containing alkyne handles (Supplementary Fig. 1h), enabling click chemistry and subsequent LC-MS/MS analysis, were screened to further improve the labeling efficiency. All probes but AP achieved efficient labeling (Supplementary Fig. 1i). For the labeling dynamics, while labeling could be detected upon 5 min light exposure, we observed an increase of labeling band intensity with longer irradiation time (Fig. 1d, e).

Encouraged by the initial results, labeling conditions in cellulo were optimized. Nucleophilic probes containing alkyne handles validated in vitro were first screened. HEK293T cells transiently expressing Flag-Halo-NLS (Nuclear Localization Signals) were treated with 1 μM SiR-CA and 500 μM nucleophilic probes, then irradiated with 660 nm LED light (50 mW/cm²) for 30 min. Following cell lysis, proteins labeled with an alkyne handle underwent Cu(I)-catalyzed azide-alkyne cycloaddition (CuAAC) with Biotin-N and were analyzed by western blot. Among the tested probes, 3-ethynylaniline (3-EA) exhibited the highest labeling efficiency (Fig. 2a). Thus, 3-EA was selected as the nucleophilic probe for subsequent experiments.

The performance of various NIR photocatalysts were subsequently evaluated. Consistent with prior results, SiR-CA exhibited a Halo protein-dependent labeling pattern (Fig. 2b, lines 2 and 5). In contrast, photocatalyst MB demonstrated Halo-independent labeling (Fig. 2b, lines 3 and 6). JF635, a derivative of SiR, exhibited reduced labeling efficiency relative to SiR-CA, probably because it is more likely to shift to the nonfluorescent spirolactone form (Supplementary Fig. 4a). Furthermore, we employed Halo mutant (D105A) deficient in chloroalkane binding and the noncovalent version of SiR-CA (SiR-no Cl) as controls to confirm that SiR-CA's photoactivation depends on covalent conjugation to Halo (Fig. 2c). The specificity of SiR-CA was preliminarily assessed using confocal microscopy (Supplementary Fig. 4b). Following PL, cells were fixed, clicked with Biotin-N via click reaction and stained with Streptavidin-FITC to visualize biotinylated proteins. Colocalization (Pearson correlation coefficient (PCC) = 0.98) was observed between the 647 nm channel (Halo-SiR) and the 488 nm channel (Streptavidin-FITC), confirming labeling specificity. The cytotoxicity of SiR-CA and MB were assessed using the cell viability assay. SiR-CA exhibited minimal cytotoxicity, whereas MB showed a concentration-dependent cytotoxic effect (Supplementary Fig. 4c).

With the optimal photocatalyst and probe for cellular PL identified, we further optimized the labeling conditions (Supplementary Fig. 4d, e), and characterized the organelle-specific proteomes using SeeID. U2OS or HEK293T cells transiently expressing Halo-tagged nuclear localization signals (NLS), nuclear export signals (NES), endoplasmic reticulum (ER) marker Sec61β, mitochondrial outer membrane protein TOMM20, lysosomal membrane protein LAMP1, and plasma membrane-targeting motifs (CAAX, PDGFR-TM) were subjected to confocal imaging and the optimized PL protocol (Fig. 2d). Western blot analysis revealed distinct labeling patterns with varying intensities across samples (Fig. 2e). Correct localization of Halo-tagged organelle markers was confirmed by confocal imaging (Fig. 2f). Confocal microscopy demonstrated strong colocalization between labeled proteins (TAMRA labeled signals) and Halo-tagged organelle markers (PCC = 0.83-0.94), confirming the high spatial specificity of SeeID across all tested organelles (Fig. 2f).

The performance of SeeID was benchmarked through proteomic profiling of the local proteome at the endoplasmic reticulum (ER) membrane, a widely studied compartment for evaluating the spatial specificity of PL methods (Fig. 3a). HEK293T cells expressing the Halo-Sec61β fusion protein (SeeID-ERM) were treated with 1 μM SiR-CA and 500 μM 3-EA, followed by irradiation with 660 nm LED light at 50 mW/cm² for 30 min. A negative control lacking HaloTag expression was included, along with a spatial reference using SeeID-NES to nonspecifically label all cytosolic proteins. Biotinylated proteins were captured with streptavidin beads, digested with trypsin on-beads to release peptides, and labeled with three isotopes ("light," "medium," and "heavy") via dimethyl labeling before pooling and LC-MS/MS analysis. Peptides were identified, and dimethyl labeling ratios were quantified using MaxQuant. Proteins with at least two peptides were quantified based on their median ratios. In total, 1652 proteins were quantified across three biological replicates, which demonstrated high correlation (Supplementary Fig. 5a and Supplementary Data 1).

The SeeID-ERM labeling sample was compared against both the negative control and the spatial reference. True-positive ERM proteins were significantly more enriched than false-positive proteins (e.g., mitochondrial matrix or non-secretory proteins) in both comparisons, demonstrating the high spatial specificity of SeeID (Fig. 3b). SeeID-labeled ERM proteins were defined as those significantly enriched in both comparisons (p < 0.05), yielding a set of 284 high-confidence ERM proteins (Fig. 3c). Gene Ontology (GO) analysis of these proteins revealed significant enrichment for terms related to the endoplasmic reticulum (ER) (Fig. 3d). Additionally, 90% of SeeID-labeled proteins were previously annotated as secretory pathway proteins (Fig. 3e). This specificity was comparable to miniSOG-ERM labeling (94%) and TurboID (87%), and significantly exceeded that of APEX2 (82%). In a comparative analysis of sub-secretory specificity, SeeID exhibited the highest ER specificity (Fig. 3f). Most labeled ER proteins were confirmed as ERM proteins, with levels comparable to those observed using other PL enzymes (Supplementary Fig. 5b). These results collectively demonstrate that SeeID exhibits excellent spatial specificity.

Previous studies suggest that miniSOG generates singlet oxygen to oxidize histidines in adjacent proteins, a mechanism that is likely shared by SeeID. To investigate this mechanism, the superTOP-ABPP platform was used to identify modification sites of SeeID-ERM and compare them with miniSOG-ERM labeling in parallel (Supplementary Fig. 5c). Streptavidin blotting revealed that SeeID-ERM, activated by NIR light, achieved higher labeling efficiency than miniSOG-ERM under blue light (Supplementary Fig. 5d). Alkyne-labeled peptides were enriched from lysates using agarose beads functionalized with azide groups and acid-cleavable linkers, then identified via LC-MS/MS and analyzed using MSFragger. This approach identified two significant mass shifts on histidines: +272 Da for the 3-EA and 2-oxo-histidine adduct and +290 Da after hydrolysis, observed in both labeling methods (Supplementary Fig. 5e). Notably, SeeID labeled more histidines than miniSOG, consistent with the increased labeling observed in streptavidin blots (Fig. 3g and Supplementary Data 2). Furthermore, an analysis of five ERM proteins with known topologies showed that SeeID labeling sites were predominantly located in regions facing the cytosol (Fig. 3h), which is in line with the N-terminus Halo-tag facing the cytosolic surface.

We further compared SeeID to the more closely related LUX-MS method that uses 590 nm light activated thiorhodamine (ThioR) for PL. We first compared SiR and ThioR in the carboxylic acid form. SiR exhibits weak absorption in phosphate-buffered saline (PBS) but shows significantly enhanced absorption in the presence of 0.1% SDS, indicating a structural transformation into a fluorescent zwitterionic form and confirming its strong fluorogenic capability (Supplementary Fig. 6a). In contrast, although ThioR also shows increased absorption in 0.1% SDS, it already exhibits high absorption in PBS, demonstrating that it lacks significant fluorogenicity (Supplementary Fig. 6b). We synthesized ThioR-CA for parallel camparison with SiR-CA (Supplementary Fig. 6c). As expected, ThioR-CA was not fluorogenic (Supplementary Fig. 6d-f). Although the covalent binding with Halo resulted in a slight increase in singlet oxygen generation by ThioR-CA (Supplementary Fig. 6f), the in vitro labeling results demonstrated that the efficacy of ThioR-CA in labeling BSA was independent on Halo protein (Supplementary Fig. 6g). Additionally, we compared the labeling efficiency of Thio-CA and SiR-CA in cells. Thio-CA labeled proteins on cell surface as efficiently as SiR-CA (Supplementary Fig. 6h). However, nonspecific background labeling was observed in the absence of Halo protein. Furthermore, Thio-CA was unable to achieve efficient intracellular protein labeling (Supplementary Fig. 6h). Overall, these findings demonstrate that SeeID functions as a genetically encoded photosensitizer with enhanced efficiency and specificity.

We then explored whether SeeID could reveal unknown PPI. The Kirsten rat sarcoma viral oncogene homolog (KRAS) gene is one of the most frequently mutated oncogenes in cancer. Somatic mutations in KRAS results in hyper-activation of downstream MAPK and PI3K-Akt signaling pathways. Despite extensive efforts in studying KRAS and its effector proteins, limited success has been achieved to develop therapeutic strategies targeting KRAS effector proteins. Discovery of new KRAS interacting proteins could provide new insights into KRAS biology as well as potential new strategies targeting KRAS mutant cancers. To test whether SeeID could capture unknown KRAS interacting proteins, we stably expressed Halo-KRAS in HeLa cells, and Halo-KRAS in KRAS mutant SW1573 cells. Cells expressing Halo-NES were used as spatial reference control. The correct localization of Halo-KRAS on the plasma membrane was confirmed by confocal microscopy (Supplementary Fig. 7a). Cells expressing the corresponding Halo-KRAS were treated with 1 µM SiR-CA and 500 µM 3-EA, followed by 660 nm LED light irradiation at 50 mW/cm² for 30 min. After cell lysis, the biotinylated proteins were then enriched and subjected to LC-MS/MS analysis (Fig. 4a). To improve peptide coverage, reproducibility and quantitative accuracy, samples were processed with label-free quantitative method and data were acquired using DIA (data-independent acquisition). 391 and 118 proteins were enriched in HeLa and SW1573 cells respectively, with 82 proteins identified in both cell lines (Fig. 4b, c and Supplementary Data 3). The enriched proteins in two cell lines exhibits certain differences. We speculate that it might due to variations in the proteins interacting with distinct KRAS mutants and inherent differences in the intrinsic proteomes of the two cell lines. Of the 427 total enriched KRAS interacting proteins, 215 proteins were reported to associate with KRAS from the BioGRID database, and 212 proteins were newly identified by SeeID (Fig. 4d). According to GO analysis, the enriched proteins were mainly involved in plasma membrane, vesicle and endomembrane organelle, which was consistent with the localization of KRAS (Supplementary Fig. 7b). Reactome pathway analysis showed that these proteins were involved in known signaling pathways related to RAS regulation, including receptor tyrosine kinase signaling, Rho GTPase signaling, MAPK signaling, and transport of small molecules (Supplementary Fig. 7c). Canonical KRAS interacting proteins (e.g. ARAF, RHOA, RICTOR, AFDN), as well as proteins identified by previous KRAS proximity labeling efforts (e.g. EGFR, CD44, BSG and ITGB1) were recovered by SeeID, confirming the reliability of our method. For the newly identified 20 proteins both in HeLa and SW1573, we focused on AXL, OSBPL5, and PHLDA1, which were not included in the BioGRID database. AXL has been reported to be involved in the resistance mechanism of anti-EGFR drugs in wild-type RAS patients, and the resistance of KRAS-mutant tumor toward KRAS inhibitors. OSBPL5 and other OSBPL family members were found to be up-regulated in Pancreatic ductal adenocarcinoma (PDAC) patient tissue. PHLDA1 was reported to mediate drug resistance in receptor tyrosine kinase (RTK) driven cancer, and act as an oncogene to promote glioma progression and recurrence. To further validate the proximity of KRAS to these three proteins, Flag-tagged candidate proteins were co-transfected with HA-tagged KRAS variants in HEK293T cells, followed by proximity ligation assay (PLA) and co-immunoprecipitation. PLA revealed that these proteins produced significant fluorescent signal with either KRAS-WT or KRAS-G12C compared to Flag-NLS control (Fig. 4e). In addition, the interaction of these proteins with KRAS was confirmed by co-immunoprecipitation using exogenously expressed proteins (Supplementary Fig. 7d). However, co-immunoprecipitation of endogenous OSBPL5, AXL, or PHLDA1 was not successful.

Live-cell imaging provides spatial-temporal information on protein dynamics, while PL gives a snapshot of protein interactome at single time point. We thought to develop a method where bulk live-cell imaging events could be analyzed directly by PL. PINK1/Parkin signaling pathway governs mitochondrial quality control via mitophagy. We have previously discovered a small molecule, BL-918, that triggers PINK1 accumulation and Parkin translocation to initiate PINK1/Parkin-mediated mitophagy. To study the BL-918 triggered Parkin translocation process, we treated HeLa cells stably expressing Halo-Parkin with 20 µM BL-918 and tracked the translocation of Parkin by confocal microscopy. Meanwhile, additional dishes with the same cells were prepared for PL. Live-cell imaging showed that Parkin was ubiquitously expressed in the cytosol before BL-918 treatment (Fig. 5a). After 2 h of BL-918 treatment, Halo-Parkin showed an aggregation pattern, formed vesicle-like structures, and began to translocate to the morphologically altered mitochondria. The correlation became more obvious after 4 h of drug treatment (Fig. 5a, PCC = 0.44, 0.82, 0.85 at 0, 2, 4 h). In contrast, the localization of Halo-NES was maintained in the cytoplasm after BL-918 treatment (Supplementary Fig. 8a). PL was performed simultaneously at these time points (0 h, 2 h, 4 h) to capture the dynamic proteomic changes of Halo-Parkin as we track the translocation process by live-cell imaging (Fig. 5b). After PL, the samples were first verified by immunostaining then subjected to MS analysis. Consistent with the live cell imaging results, immunofluorescence results of fixed cells after a click reaction with TAMRA-N showed that the TAMRA-labeled signal accumulates from being scattered in the cytoplasm to damaged mitochondria, similar to the signal of Halo-SiR (Supplementary Fig. 8b). The interactome of Parkin in this dynamic process was then analyzed by LC-MS/MS using Halo-NES as a spatial control. 269 proteins interacting with Halo-Parkin before BL-918 treatment, and 250 and 93 proteins after BL-918 treatment for 2 h and 4 h, respectively, were obtained (Supplementary Fig. 8c and Supplementary Data 4). Cell viability analysis revealed no significant reduction at 2 h of BL-918 treatment, and the decreased amount of proteins enriched at 4 h might result from the lower cell viability at this time point (Supplementary Fig. 8d). Proteome data revealed a significantly differentiated Parkin interactome upon drug treatment. Proteins enriched prior to treatment were spread across the nucleoplasm, organelle lumen and nucleus. And the proteins that interact with PARKIN were engaged in processes such as chromatin organization, regulation of primary metabolic pathways, and modulation of RNA metabolic activities (Supplementary Fig. 8e). While the mitochondrial membrane was the primary location of proteins enriched after BL-918 treatment (Fig. 5c), and the interacting proteins were involved in small molecule metabolism, assembly of mitochondrial respiratory chain complex, oxidative phosphorylation and mitochondrial transmembrane transport (Supplementary Data Fig. 8e). PARKIN substrates during mitochondrial depolarization have been systematically explored, but a proximity-labelling-based approach to identify PARKIN substrates has not been previously reported. Among the proteins enriched based on SeeID labeling, we identified 44 proteins that were previously characterized as substrates of PARKIN. The ion intensity of known Parkin interacting proteins was similarly increased with BL-918 treatment. These proteins include PKN1(PINK1), FIS1, HK1, PARK7, MTOR, RIPK1, PKM located in the mitochondria, and TBC1D15, PSMC1, VCP, RAD23A in the cytoplasm (Fig. 5d and Supplementary Data 5). For the identification of previously unknown interacting proteins of Parkin, we focused on the effectors involved in the mitochondrial protein degradation pathway. 13 proteins in this pathway were identified by SeeID. Among which, 9 of 13 were identified by previous affinity mass spectrometry or proximity labeling, and 4 proteins have not been reported, including ECI1, BDH1, OXSM, ATP5F1B (Fig. 5e). Interestingly, quantitative analysis indicated that 29 out of the 45 proteins associated with the proteasome complex has increased (Ratio >1.5) in ion intensity after 2 h BL-918 treatment (Fig. 5f). This enrichment was not due to the overall increased protein expression, as the proteasome complex was not enriched in the Halo-NES group. A decrease of proteasome complex enrichment was observed after 4 h. Although this observation is in line with previous reports indicating that Parkin plays a role in recruiting proteasomes to depolarized mitochondria and facilitating the degradation of ubiquitin-tagged mitochondrial proteins, to the best of our knowledge, the dynamic spatial interplay between the proteasome complex and Parkin during mitophagy has not been discovered before. We identified an enrichment of proteins associated with mitochondrial fusion, specifically MIGA1 and MIGA2 (Supplementary Data 4), which aligns with the morphological changes in mitochondria observed through microscopy. Furthermore, we noted an enrichment of proteins related to fatty acid metabolism (including ACAD10, EHHADH, SLC27A2, ACADVL, CBR4, and ACOT8) and oxidative phosphorylation process (such as NDUFA11, NDUFB7, NDUFAF1, NDUFA12, NDUFS5, DNAJC30, and NIPSNAP2) following mitochondrial depolarization (Supplementary Data 4). Although we did not explore the specific mechanisms in depth, these results suggests that PARKIN may be involved in these processes. These findings demonstrated the dynamic protein interactions during mitophagy and showcased the potential of imaging-guided PL for studying dynamic protein interactomes.

Application of SeeID in vivo will allow the study protein interactomes in animal models. To demonstrate the utility of SeeID in vivo, we aimed to label the mouse brain. Halo-tagged PDGFR-TM protein (Halo-TM) was used to target SiR to the plasma membrane. Biotin-conjugated probe BA was used as the probe to capture the proximal proteins (Supplementary Fig. 9a). The ability of SeeID to label mouse brain was first demonstrated on acute brain slices. The packaged AAV virus carrying Halo-TM (AAV-Halo-TM) was stereotactically injected to the hippocampus regions of C57BL/6 J male mice. After 2 weeks of virus expression in mice, the brains were extracted under anesthesia, submerged in artificial cerebrospinal fluid (ACSF), and cut into 300 µm slices. Brain slices containing hippocampal regions were incubated with 1 µM SiR-CA and 500 µM BA in ACSF for 30 min on ice, followed by labeling with 660 nm LED light at an intensity of 100 mW/cm for 45 min on ice (Fig. 6a). Immunofluorescence demonstrated robust labeling signals in the hippocampus, while the labeling was not observed without Halo-TM expression, SiR-CA or irradiation (Fig. 6b). Western blot results further confirmed the labeling specificity (Fig. 6c).

After confirming SeeID could be utilized for brain slice labeling, we moved to PL in vivo. AAV-Halo-TM was stereotactically injected into the left (L) and right (R) visual cortex region of mice. Labeling of mouse cortical tissues was performed after 3 weeks of in vivo expression. BA with DMSO or BA with SiR-CA were delivered to the left or right cortical regions of mice at the same site where the virus was introduced by stereotactic injection. Mice were subsequently labelled by 660 nm LED irradiation at 50 mW/cm for 30 min while being anaesthetized (Fig. 6d). Two mice brains after irradiation were dissected after perfusing for subsequent immunofluorescence staining analysis. The cortical tissues of other mice brains were isolated and homogenized to extract proteins for western blot and mass spectrometry analysis. Immunofluorescence results showed a SiR-CA dependent labeling of up to 2 mm tissue depth (Fig. 6e), demonstrating the specificity of our labeling protocol and the advantage of NIR for potential deep tissue labeling. The observation was further confirmed by western blot: the labeling of the right side of the brain tissue of each mouse (with SiR-CA) was significantly higher than that of the left side (without SiR-CA) (Supplementary Fig. 9b).

We next performed mass spectrometry analysis on the above extracted tissue proteins and determined the specificity of in vivo labeling. A total of 60 labeled proteins were enriched when the proteins were filtered in all three biological replicates with Ratio (SiR-CA/DMSO) > 1.5 and p value < 0.05 (Fig. 6f and Supplementary Data 6). Of the 60 enriched proteins, ~62% proteins are associated with membrane, vesicle and extracellular matrix (Fig. 6g), e.g. neurotransmitter receptors (Gabra2, Gpr158, Gria1), neural cell adhesion molecule (Ncam1) and vesicular trafficking regulator (Syt5) (Fig. 6h). In summary, our experiments demonstrated the utility of SeeID for in vivo PL, and the use of NIR permits spatiotemporally controlled deep tissue labeling, which has not been achieved by PL methods in the literature.