Proximity labeling (PL) has emerged as a powerful technology to capture biomolecular interactions in living systems, gaining new insights into various biological processes including protein–protein interactions (PPIs), RNA–protein interactions, drug–target interactions, and cell–cell interactions. While many of these molecular recognition events are traditionally difficult to resolve due to their transient, dynamic, and spatially confined nature, PL coupled to multiple omics techniques offers an attractive solution to this key challenge. Ever since its initial technical establishment in the early 2000s, a multitude of PL strategies and platforms have been developed over the past two decades [1, 2, 3, 4, 5, 6, 7].

The principle of PL involves the use of a promiscuous enzyme or a chemical catalyst to generate reactive intermediate species within the proximity of the protein of interest (POI), and then achieve covalent labeling of the neighboring molecules. These labeled molecules typically bearing biotinylation tags can be subsequently enriched via affinity purification and analyzed by mass spectrometry or RNA sequencing. Of note, several high-quality reviews have been published in the last two years to provide an extensive update of the recent development of PL methods and probes as well as the application to profiling diverse biomolecular interaction networks [∗8, ∗9, ∗10].

Complementary to these previous reviews, in this short review, we focus on the recent advancements of PL-based research mostly documented from 2023 to 2025 and discuss the innovation of enzymatic tools, the strategies for targeting endogenous POI or organelles, and the determination of the labeling radius of PL.

Proximity labeling can be broadly categorized into enzyme-based and chemical catalyst-based approaches. The latter, catalyst-based PL utilizing small molecules or metal complexes to generate reactive species has been comprehensively reviewed by both MacMillan group and Chen group [8,10]. In this section, we aim to highlight the recent development of new enzymatic tools and systems for PL in the past two years (Figure 1a).

In 2023, Tanner et al. employed directed evolution to create a new G-quadruplex peroxidase-mimicking DNAzyme, mSBDZ-X-3, which exhibits advantages such as facile synthesis, enhanced stability, and improved resistance to hydrolysis [11]. Of note, this enzyme still requires the presence of hydrogen peroxide (H2O2) for catalytic activity.

To circumvent the requirement for H2O2, several novel PL enzymes have been developed in the past two years, which fall into two major categories according to their activation mechanism: light-dependent and light-independent. In 2023, Muir and et al. introduced the Light-induced Interactome Tagging (LITag) system by integrating an engineered light–oxygen–voltage (LOV) domain into the POI [12]. This LOV module has a molecular weight similar to the engineered flavin-binding protein miniSOG (∼12 kDa) [13] and offers superior temporal resolution, enabling labeling within 1∼3 s of blue light exposure. It is noteworthy that, although this LITag fusion can be targeted to the nucleolus, it fails to effectively label nucleolar proteins when using biotin-phenol (BP) as the substrate. Later, Ting et al. developed LOV-TurboID, which incorporates a light-sensitive LOV domain into the conventional PL enzyme TurboID [3]. This design enables spatiotemporal control via low-intensity blue light activation, and significantly reduces the labeling background in a biotin-rich environment such as neurons. However, LOV-TurboID shows barely any activity within the secretory compartments and demonstrates relatively low targeting efficiency toward the mitochondrial matrix [14].

In regard to the light-independent new PL enzymes, Hamachi et al. discovered a bacterial tyrosinase from Bacillus megaterium (BmTyr) in 2024, which is H2O2-free and enables rapid (≤10 min) and low-background protein labeling [15]. Recently, Qin et al. also explored BmTyr in labeling plasma proteins in vivo and labeling region-specific proteomes in the mouse brain to demonstrate its improved biocompatibility compared to existing PL enzymes [16]. Moreover, Ting et al. introduced, in a bioRxiv preprint, a novel PL enzyme termed LaccID which was obtained through directed evolution of an ancestral fungal laccase [17]. Unlike the widely used peroxidase tools (horseradish peroxidase [HRP], APEX, and APEX2) that all oxidize aromatic substrates in the presence of peroxide, LaccID relies on molecular oxygen to catalyze single-electron oxidation of aromatic substrates, thus eliminating the usage of H2O2. Notably, LaccID is only active when expressed on the cell surface, making it particularly suited to mapping cell surface proteomes via fusion protein expression. Owing to its relatively low enzymatic activity, LaccID currently requires 1–2 h of labeling to yield detectable signals, awaiting future improvement for capturing rapid changes in proteome dynamics [18]. Given that both BmTyr and LaccID rely on molecular oxygen to catalyze substrate oxidation, their labeling efficiency may be compromised under hypoxic conditions.

Beyond these single-enzyme-mediated PL methods, researchers have established cascade reaction-based PL systems to enhance spatial selectivity. For instance, in 2024, Pan et al. found that, in the presence of superoxide dismutase and under blue light illumination, the singlet oxygen photosensitizing protein-3 can convert molecular oxygen into H2O2, thereby activating APEX2-mediated PL without the need of exogenous H2O2. This method exhibited higher spatiotemporal resolution (labeling time <10 s), higher efficiency, and flexible applicability compared to conventional approaches, supporting the proteomic study of organelle contact sites and cell–cell interface [19] (Figure 1b). In 2025, Ju et al. developed the two-level spatially localized proximity labeling (P2L) system, by incorporating a galactose oxidase (GAO) that oxidizes galactose to generate H2O2, prior to the proximity labeling step. By separately targeting proteins and glycans with GAO and HRP, P2L allows for distinguishing cells with varying glycosylation levels within a heterogeneous cell population [20] (Figure 1c). As cascade PL strategies involve multiple biological components whose expression levels and subcellular localization must be precisely coordinated, improper control of these factors may lead to nonspecific labeling or reduced labeling efficiency.

In addition to enhanced spatial precision, PL enzymes activated by endogenous signals within physiological environment offer unique advantages. For example, in 2023, Weerapana et al. leveraged endogenous reactive oxygen species (ROS) as a source of H2O2 to activate APEX-mediated PL and thus monitor oxidative events specifically occurring within ROS hotspots [21]. In 2024, Ju et al. developed stress-activated PL by anchoring HRP to stressors (e.g. fungal mimic or live fungi). Endogenous H2O2 released by host cells such as macrophages in response to the stressor triggers the PL reaction, dynamically recording stress levels associated with cell–cell interactions [22] (Figure 1d). These approaches eliminate the need for external H2O2 supplementation, reduce the cellular toxicity, and enhance the physiological relevance. Furthermore, Ingolia et al. developed an engineered biotin ligase, Cal-ID, which senses local calcium ion fluctuation via calmodulin. Elevated Ca2+ concentrations triggers Cal-ID to biotinylate nearby proteins, providing spatially resolved biochemical recording of Ca2+ signaling and neuronal activity [23] (Figure 1e). Nevertheless, these environment-responsive PL systems would be restrained by the availability of endogenous signals in the cellular contexts.

The majority of current PL approaches rely on genetic engineering of the POI to be fused with a PL enzyme and exogenous expression of the fusion protein in cells or tissues. As noted previously, these approaches could suffer from perturbing the structure, activity, and trafficking of the native POI and encountering technical obstacles when dealing with hard-to-transfect cells, animal models, or clinical samples [8]. To in part address this key challenge, new strategies have been developed recently which can be classified into three categories (Figure 2).

To interrogate protein–protein interactions for an endogenous POI in living systems, both ligand-directed PL and antibody/lectin-directed PL methods have been established for specific targets. Ligand-directed PL leverages different classes of ligands (aptamer, small molecule, peptide, and protein) which are tethered to a PL enzyme or a photocatalyst. Upon treatment of cells expressing the native POI, the ligand–catalyst conjugate binds to the POI and triggers the labeling reaction in close proximity in the presence of the substrate (Figure 2). In 2023, the Ding group conjugated cell-specific tagging aptamers with HRP to achieve selective and covalent modification of target cells [24]. Subsequently, the Tanner group combined DNA aptamers that target cancer cell surface markers with DNAzymes to label proteins proximal to the cell membrane antigens in cancer cells [11] (Figure 2a). In the meantime, the MacMillan group linked small-molecule ligands to iridium (Ir)-based photocatalysts, leveraging their photocatalytic PL platform, μMap, for the identification of protein targets and ligand-binding sites [25]. In 2024, the Hamachi group tethered small-molecule ligands to the PL enzyme, BmTyr, so as to map the proximal proteomes of endogenous neurotransmitter receptors in the live mouse brain, providing the proof-of-concept for in vivo labeling [15]. Subsequently, the same group developed the PhoxID strategy by anchoring a small-molecule photosensitizer to the neurotransmitter receptor to allow proximal proteome profiling in the live mouse brain at different developmental stages, further demonstrating the feasibility of in vivo PL in live animals [26] (Figure 2b).

Unlike most of the above mentioned studies in which the ligand module in the PL probe were binders of POI, we conjugated an agonist, the GLP-1 peptide, to the APEX2 enzyme so as to map the endogenous cell membrane interactomes for the activated GLP-1 receptor in two different cell types, revealing protein interaction landscapes and dynamics distinct from those previously profiled in less physiological systems [27] (Figure 2c). In addition, the Saeed group conjugated the photocatalyst Ir to the SARS-CoV-2 spike protein to map the interactome engaged by the viral protein and host cell surface proteins [28] (Figure 2d).

Antibody/lectin-directed PL achieves labeling of the endogenous interactome by utilizing a target-specific antibody or lectin. For example, in 2020, the MacMillan group introduced the μMap platform which conjugates Ir photocatalysts to antibodies for labeling of antibody-binding targets and their neighboring proteome, revealing proteins surrounding programmed death ligand 1 (PD-L1) in live lymphocytes [5]. Later in 2024, the Wells group refined spatial resolution of photo-PL by site-specifically introducing Ir into primary antibodies, enabling distance-calibrated PL [29] (Figure 2e). Moreover, the Wells group developed the MultiMap platform, in which the photocatalyst Eosin Y is conjugated to specific antibodies, to activate multiple probes with distinct labeling radii and enable profiling of the immune synaptic proteome [30]. The Seath group and Fan group also exploited the antibody-based photo-PL strategy to map the epidermal growth factor receptor (EGFR) interactome and tumor–immune interface, respectively, in primary tumor tissues [31,32]. In parallel to the photo-PL strategy, the Zou group utilized the HRP enzyme-conjugated antibodies specific for the axon initial segment (AIS) protein NFASC to allow AIS-restrained PL, thereby revealing dynamic changes in the AIS proteomes during neuronal development [33] (Figure 2f). Additionally, MacMillan et al. used photocatalyst Ir-conjugated IgG-opsonized beads to label phagocytic interfaces, capturing cell surface proteomic snapshots during phagocytosis [34] (Figure 2g). Finally, Huang and Mao exploited glycan–lectin interactions by conjugating lectins (Galectin-3 or Siglec-9) with APEX2 so as to label proteins possibly associated with specific glycans [35,36] (Figure 2h).

While most antibody- or lectin-directed PL approaches target cell surface proteins, the strategy is also applicable to intracellular proteins. For example, the Killinger group employed antibody-directed PL for the alpha-synuclein (α-syn) target, enabling the analysis of α-syn-engaged proximal proteomics in brain slices from patients with neurodegenerative diseases [37].

In the context of subcellular or organelle proteomics, previous genetic engineering-based PL has resulted in the successful proteome profiling for different compartments or subcellular structures including mitochondria [38], nuclear lamina [1], plasma membrane [39], endoplasmic reticulum membrane [3], centriole–cilium interface [40], primary cilia [41], etc. Nevertheless, such strategies are challenging to implement in hard-to-transfect cells. In recent years, researchers have developed chemical photocatalysts for targeting organelles such as the mitochondria [42,43], endoplasmic reticulum [44], and lysosomes [45], thereby enabling specific and in situ labeling of native intracellular compartments (Figure 2i–l). Notably, the Fan group introduced LysoCat, a lysosome-specific catalyst that overcomes challenges posed by the acidic environment and low protein abundance of lysosomes, to enable profiling of lysosomal proteome dynamics in live cells [45].

Historically, the labeling radius of the PL enzyme BioID is estimated to be ∼10 nm using the rigid nuclear pore complex as a molecular ruler [46], and electron microscopy imaging suggests a labeling radius of ∼20 nm for another enzyme APEX [47]. However, a precise measurement of the labeling radius by different PL tools had been lacking until recently. In 2022, the MacMillan group first utilized stimulated-emission depletion microscopy to determine labeling radii of PL reactions on bovine serum albumin-coated coverslips [48]. They found that the peroxidase-based phenoxy radical system exhibited a labeling radius of 269 ± 41 nm (using biotin-tyramine as the substrate), whereas the iridium catalyst-based μMap platform exhibited a much smaller radius of 54 ± 12 nm (using biotin-diazirine as the substrate). Moreover, they reported that longer-lived intermediates, such as nitrenes, led to a wider labeling radius (119 ± 33 nm) compared to carbenes. Additionally, slower-diffusing substrates such as PEG24-PhN3 exhibited a reduced radius (80 ± 28 nm) compared to PEG3-PhN3 (119 ± 33 nm). These results provide valuable guidance for building multi-range labeling platforms.

In 2024, the Shi group established proximity labeling expansion microscopy (PL-ExM) to enable super-resolution imaging on standard confocal microscopes. Using this technique, they found that HRP-catalyzed labeling produced higher spatial precision than APEX2, with mitochondrial labeling diameters of 0.56 μm vs 0.97 μm, respectively [49]. In another study, Chen et al. utilized DNA nanostructure platforms to precisely measure the labeling radii of TurboID and APEX2, and revealed their labeling mechanisms with in vitro assays. While it is conventionally proposed that labeling occurs via diffusion of reactive intermediates, their findings indicated distinct mechanisms: TurboID predominantly operates in a contact-dependent manner, with a labeling range of ≤6 nm; in contrast, APEX2 mediates labeling both through direct contact (approximately 10 nm) and a diffusion-based mechanism, with a labeling radius extending up to 500 nm [50].

To fine-tune the proximity labeling radius, researchers have exploited various strategies. For example, chemical modification of substrates can generate intermediates with tunable half-lives, while the use of different labeling mechanisms—such as contact-dependent enzymes versus contact-independent enzymes [8] or chemical catalysts—offers additional control. Among these, substrate-based modulation of labeling radius has been extensively investigated. The Tian group reported that in APEX2-mediated PL, the use of the BN2 probe yields a smaller labeling radius compared to BP, making it more suitable for capturing cytosolic protein complexes [51]. The Chen group demonstrated the capability of quinone methide substrates with second-scale half-lives in micrometer-scale labeling radii for spatially resolved analysis of cellular organization [52]. Additionally, the Wells group uncovered the labeling resolution using Eosin Y catalysis to follow the order of diazirine (high), aryl azide (medium), and phenol (low), providing a useful reference for selecting substrates to achieve the desired spatial precision. Using the integrated multiscale PL proteomics platform, they mapped local and distal protein networks at and between cell surfaces, enabling systematic reconstruction of the cell surface interactome, revealing horizontal signaling partners, and uncovering potential immunotherapeutic targets [30].

Despite the development of multiple approaches for measurement and control of the labeling radius, variations in the experimental settings hinder direct comparison between different studies. Future efforts would be expected in building uniform cellular models and standardized instrument platforms, to guide the rational design and optimization of PL tools for precisely capturing biomolecular interactions.