A Brief Progress in Methods for Deciphering Protein–Protein Interaction Networks

Abstract

Protein–protein interactions (PPIs) are fundamental regulators of cellular function and disease. Systematic mapping of the interactome is essential for identifying therapeutic targets and advancing drug design, a pursuit that has driven significant innovation to capture the spatiotemporal regulation of PPIs in vivo. This review summarizes this methodological revolution. We outline foundational, first-generation techniques—yeast two-hybrid and co-immunoprecipitation—which established frameworks for binary interaction mapping and static network generation, especially when integrated with mass spectrometry. The discussion then pivots to second-generation methods, including proximity-dependent labeling and advanced imaging, which enable the capture of PPIs within their native, dynamic cellular contexts. We provide a comparative analysis of these techniques, detailing their principles, strengths, and limitations. The review concludes with a practical framework for method selection and a perspective on emerging frontiers—such as spatial proteomics and single-cell interactomics—that are poised to further decode the evolving interactome. This concise overview serves as a strategic guide for specialists adopting new techniques and a broader audience integrating network-level data into their research.

1. Introduction

In living organisms, proteins do not act in isolation but form intricate, dynamic interaction networks to precisely regulate cellular functions. Protein–protein interactions (PPIs) therefore serve as the fundamental molecular basis of life activities. By analyzing the architecture and dynamics of these PPI networks, we can infer the roles of uncharacterized proteins and elucidate their functions within the broader biological system [1]. This systems-level view enables applied research: comprehensive mapping of PPI networks can decode the complexity of signaling pathways and identify disease-related protein modules or key nodes [2].

These dysregulated or critical nodes, in turn, become prime targets for therapeutic intervention. A mechanistic understanding of specific PPIs allows researchers to rationally design small-molecule inhibitors or activators, which can lead to drugs with superior therapeutic efficacy. This rational, target-driven pipeline has already proven successful, with breakthrough targeted drugs against the B-cell lymphoma 2 (Bcl-2) family and the Mouse Double Minute 2 Homolog (MDM2)-p53 interaction serving as prominent examples [3]. Venetoclax is a landmark Bcl-2 inhibitor that functions by blocking the anti-apoptotic Bcl-2/Bak interaction, thereby promoting apoptosis in cancer cells [4]. Approved by the U.S. Food and Drug Administration (FDA) in 2016, it holds dual significance: it is not only the first-in-class Bcl-2 protein inhibitor but also represents the first approved small-molecule drug designed to directly challenge a protein–protein interaction [5]. It recorded an annual sale of 1.2 billion United States Dollar (USD) in 2024.

Another prominent class of PPI-targeted drugs focuses on the MDM2-p53 interaction, a master regulator of apoptosis. The field was catalyzed in 1996 when Kussie et al. reported the co-crystal structure of the complex, revealing a well-defined p53-binding pocket on MDM2 [6]. This structural insight established the direct feasibility of designing high-affinity small molecules to disrupt this interaction, sparking extensive drug discovery efforts in oncology. This rationale has translated into robust clinical pipelines. To date, nine small-molecule MDM2 inhibitors—including RG7112, Idasanutlin and Milademetan—have advanced to clinical trials, demonstrating the tractability of this target [7]. These case studies—from Bcl-2 inhibition to MDM2-p53 targeting—collectively underscore a fundamental principle: PPI research is not merely an academic pursuit but a crucial conduit for understanding biological systems, signal transduction, and disease mechanisms, directly enabling the development of novel, mechanism-based therapeutics.

The systematic study of PPIs began over four decades ago with low-throughput biochemical techniques such as protein affinity chromatography and coimmunoprecipitation. While foundational, these “first-generation” methods were constrained by scale, identifying interactions individually rather than systematically.

A transformative shift occurred in the late 1990s with the integration of mass spectrometry, particularly liquid chromatography–tandem mass spectrometry (LC-MS/MS). This advancement propelled the field from “single validation” to “large-scale screening,” enabling the mapping of genome-wide interaction networks and the discovery of numerous previously unknown PPIs [8].

Yet, these static network maps, while comprehensive, could not capture the essence of living systems. They are inadequate in explaining critical dynamic processes, such as the temporal progression of signaling cascades or network remodeling during cell differentiation. Consequently, the frontier of PPI research has undergone a second major shift: from “static network mapping” to “dynamic mechanism analysis.” The core objective is now to capture the spatiotemporal dynamics and underlying molecular mechanisms of PPIs within their native, physiological contexts [9].

This review provides a concise overview of recent methodological progress in protein–protein interaction research. We systematically introduce the principles, key advantages, inherent limitations, and illustrative application cases of current techniques, offering a practical framework for selecting appropriate experimental and computational strategies. Furthermore, we discuss emerging trends and future directions in the field, aiming to provide the insights needed to advance the next generation of dynamic and context-aware PPI studies.

2. Methods for Screening PPIs

Screening techniques form the foundation for constructing protein interaction networks. They can rapidly identify potential interaction partners of target proteins, enable large-scale mining of unknown interaction relationships, and provide data support for building initial interaction network frameworks (Figure 1).

Figure 1.

Methods for screening PPIs. (a) Yeast two-hybrid (Y2H) reconstitutes a transcription factor from bait- and prey-fused fragments to detect interactions via reporter gene activation in yeast. (b) Co-IP/AP-MS isolates protein complexes from lysates using antibodies or tags against the bait, followed by MS to identify interactors. (c) Co-fractionation MS (CF-MS) separates native protein complexes by biochemical fractionation and infers their composition through correlated elution profiles in MS. (d) Cross-linking MS (XL-MS) stabilizes protein interactions via chemical cross-linking, enriches cross-linked complexes, and identifies linked peptides by MS. (e) Proximity ligation assay (PLA) uses antibody-conjugated DNA probes that form a circularizable template upon target proximity; rolling-circle amplification generates a localized fluorescent signal for in situ interaction imaging. (f) Proximity labeling (PL) employs a bait-fused enzyme (e.g., BioID, APEX) to biotinylate neighboring proteins in live cells, which are then enriched and identified by MS to map proximal interactomes.

2.1. Yeast Two-Hybrid (Y2H)

2.1.1. Core Principles and Mechanisms

The Y2H system leverages the modular nature of eukaryotic transcriptional activators, which consist of a DNA-binding domain (BD) and an activation domain (AD). Individually, these domains cannot initiate transcription; however, when brought into spatial proximity by the interaction of “bait” (BD-fusion) and “prey” (AD-fusion) proteins, they reconstitute a functional unit that triggers reporter gene expression (Figure 1a) [8,9,10,11]. This enables the detection of protein–protein interactions (PPIs) within the physiological environment of living yeast cells [8,9,11].

2.1.2. Scaling the Interactome

Y2H is pivotal for mapping large-scale interaction networks. Early landmark studies in S. cerevisiae by Uetz et al. and Ito et al. identified thousands of putative interactions, providing essential functional context for previously uncharacterized proteins [12,13]. The methodology has since evolved to human systems: Weimann et al. used Y2H-seq to map methyltransferase networks [14], while Rolland et al. scaled the framework to identify over 53,000 binary interactions, creating a comprehensive reference map of the human interactome [15,16].

2.1.3. Overcoming Limitations: Membrane Proteins and False Positives

Traditional Y2H is limited by the requirement for nuclear localization, making it unsuited for membrane proteins. To solve this, Split-Ubiquitin technology was proposed [17], leading to the DUALmembrane and MYTH systems for high-throughput membrane screening [18]. These concepts were later adapted for human cells via the MaMTH assay [19]. Other variations include:

• Signaling-based systems: Sos/Ras recruitment systems that trigger growth cascades at the membrane [20].

• Visualization: BiFC and luminescence-based reporters for spatial resolution.

• Data Integrity: To combat false positives, Gu Y et al. introduced Y2H-in-frame-seq, combining traditional screening with NGS to reduce false positives by 60% while maintaining high sensitivity [21,22].

2.2. Co-Immunoprecipitation (Co-IP)

2.2.1. Principles and Advantages

Co-IP is a cornerstone technique for studying protein–protein interactions (PPIs) by using specific antibodies to precipitate target proteins and their associated complexes (Figure 1b) [23]. As it is performed in cell lysates, it preserves native protein conformations and post-translational modifications, avoiding artifacts common in in vitro methods. This high physiological relevance makes Co-IP a “gold standard” for PPI validation [24]. Furthermore, when specific antibodies are available, the workflow is rapid and bypasses the need for gene cloning or heterologous expression.

2.2.2. Limitations and Optimization

The primary constraint of Co-IP is its absolute dependence on high-quality, specific antibodies [25]. Low-affinity antibodies lead to inefficient enrichment and high background noise. Additionally, the method often struggles to distinguish between direct and indirect interactions [26], and transient complexes may be lost during washing steps. While optimizations—such as the refined buffers and conditions proposed by Burckhardt et al. (2021)—improve reproducibility, the fundamental challenges of antibody dependency remain [27].

2.2.3. Clinical and Mechanistic Applications

Co-IP is vital for analyzing endogenous interactions in disease contexts [8]. In 2022, Liliang Shen et al. used an anti- phosphoglycerate dehydrogenase (PHGDH) antibody in T24 cells to demonstrate that PHGDH binds to the RNA-binding protein protein poly(rC)-binding protein 2 (PCBP2), inhibiting its degradation. This upregulates the ferroptosis inhibitor solute carrier family 7 member 11 (SLC7A11), driving malignant progression in breast cancer [28]. Similarly, Xingwu Liu et al. utilized Co-IP in colorectal cancer to show that Ubiquitin specific peptidase 10 (USP10) promotes oxaliplatin resistance and DNA repair by stabilizing XPA binding protein 2 (XAB2), highlighting the USP10/XAB2/Annexin A2 (ANXA2) axis as a potential therapeutic target [29].

2.3. Affinity Purification–Mass Spectrometry (AP-MS)

2.3.1. Principles and Tagging Strategies

AP-MS identifies protein interactors by fusing an affinity tag to a bait protein, enabling targeted enrichment followed by mass spectrometric analysis (Figure 1b) [30]. This approach replaces bait-specific antibodies with standardized tags, reducing background noise and enabling systematic protein–protein interaction mapping. Small epitope tags such as FLAG or HA minimize functional interference, whereas larger tags like GFP facilitate live-cell imaging. To improve purification specificity, Tandem Affinity Purification employs sequential dual-tag enrichments [31], while proximity-labeling methods like Proximity-dependent biotin identification (BioID) capture weak or transient interactions often lost in conventional pull-down assays [32].

2.3.2. Global Interactome Mapping

The Krogan laboratory significantly advanced AP-MS by constructing comprehensive protein interaction networks in yeast [33,34]. Applying this method, their team identified 332 high-confidence interactions between viral proteins and human host factors, revealing 66 druggable targets and validating several FDA-approved compounds with antiviral activity [35].

2.3.3. Limitations and Computational Optimization

Key challenges of AP-MS include high false-positive rates from non-specific binding, the difficulty in distinguishing direct from indirect interactions, and potential artifacts from protein overexpression. Mali et al. demonstrated that the choice of solid-phase resin significantly influences capture efficiency and complex stability [36]. To address these issues, robust computational frameworks have been developed. Tools such as SAAINT (https://github.com/tommyhuangthu/SAAINT, accessed on 10 February 2026) and MiST (https://mistdb.com/, accessed on 10 February 2026) apply probabilistic models to score interactions based on specificity and reproducibility [37,38]. Additionally, the CRAPome database (https://reprint-apms.org/?q=chooseworkflow, accessed on 10 February 2026) catalogs common background contaminants to facilitate systematic false-positive filtering [39].

2.3.4. Spatiotemporal Innovations

Our research group has pioneered the integration of AP with high-resolution quantitative proteomics to map dynamic signaling networks. In a 2013 Nature study, Yong Zheng et al. utilized the adaptor protein SHC adaptor protein 1 (Shc1) as a temporal probe within the EGF receptor network, revealing its role as a molecular switch that transitions from early mitogenic signaling to late cytoskeletal reorganization [40]. Follow-up work combined subcellular fractionation with targeted mass spectrometry to demonstrate that Shc1 anchors a mobile, “receptor-free” cytosolic subcomplex [41]. To further enhance reliability, Mei et al. integrated Tandem Mass Tag multiplexing with AP-MS, using quantitative comparison across samples to statistically distinguish genuine biological interactions from non-specific background [22].

2.4. Chemical Cross-Linking Mass Spectrometry (XL-MS)

2.4.1. Principles and Historical Evolution

XL-MS is a structural biology technique that employs chemical cross-linkers to covalently connect amino acid residues within spatial proximity (a few nanometers) (Figure 1d). Following enzymatic digestion and LC-MS/MS, identifying these residue pairs provides distance constraints for modeling protein interaction sites and the 3D architecture of complexes [42]. The field evolved from using bifunctional reagents like dimethyl adipimidate and glutaraldehyde in the 1960s to stabilize complexes for SDS-PAGE [42,43,44,45]. The major breakthrough was the integration of mass spectrometry; pioneering work by Young et al. established the modern workflow by deriving structural constraints from intramolecular cross-links [46,47,48]. Subsequently, Rappsilber and colleagues demonstrated its power by mapping the topology of the large, non-crystalline yeast Nucleoporin NUP84 nuclear pore sub-complex [48].

2.4.2. Bioinformatics and Scalability

The advancement of XL-MS has relied heavily on bioinformatics to interpret complex “chimeric” spectra [49]. Following foundational work by Back et al., specialized software such as pFind and the dedicated search engine pLink (developed by Dong Mengqiu’s laboratory and the pFind team) set benchmarks for accurate peptide matching [50,51,52]. Modern innovations, including MS-cleavable cross-linkers and search engines like Scout (which utilizes machine learning for high-speed analysis), have further enhanced the efficiency and scalability of XL-MS data processing [52].

2.4.3. Unique Advantages

XL-MS excels at capturing transient interactions and dynamic protein structures that elude traditional methods. By providing spatial distance restraints, it serves as a robust tool for integrative structural modeling, particularly for large, heterogeneous protein complexes that defy conventional structural analysis.

2.4.4. Cross-Linkers

The efficacy of Cross-linking Mass Spectrometry (XL-MS) is fundamentally dictated by the chemical properties of the cross-linker, specifically its reactivity, arm length, and fragmentability. Current studies predominantly employ homobifunctional amine-reactive linkers, such as the lipophilic Disuccinimidyl suberate(DSS) and its water-soluble derivative Bis(sulfosuccinimidyl)suberate(BS3), to establish spatial constraints between lysine residues for structural modeling [53]. To mitigate the computational challenges of identifying cross-linked peptides within complex proteomes, MS-cleavable linkers like Disuccinimidyl sulfoxide (DSSO) and Urea crosslinker-C4-arm, NHS ester (DSBU) have emerged as the gold standard; these reagents undergo predictable fragmentation during Tandem mass spectrometry (MS/MS) to produce distinct reporter ions that simplify peptide identification [54]. Furthermore, sensitivity is significantly enhanced by specialized “enrichable” linkers such as PhoX, which features a phosphonic acid group for selective Immobilized metal ion affinity chromatography (IMAC) purification, or “zero-length” linkers like Ethyldimethylaminopropyl Carbodiimide (EDC), which facilitate the direct coupling of carboxyl-to-amine interfaces for maximum spatial resolution [55,56]. Together, this diverse chemical toolkit enables a transition from basic interactome mapping to the high-resolution structural and spatiotemporal analysis of native protein complexes [42,57].

Despite these advancements, notable technical limitations persist. First, selecting an appropriate cross-linker remains largely empirical, as variations in reactivity, arm length, and specificity can lead to non-specific cross-linking, compromising result accuracy and introducing false positives. Second, data analysis remains inherently complex, requiring specialized software, advanced algorithms, and bioinformatics expertise, all of which place high demands on researchers’ technical skills and experience. In 2023, Zhang et al. proposed a strategy based on chemical cross-linking and hierarchical analysis to analyze the conformational changes and functional regulatory mechanisms of proteins in vivo [58]. In 2022, Jian-Hua Wang et al. developed a 3,4-Dihydroxyphenyl L-Alanine (DOPA) cross-linker. By sampling at different time points, performing cross-linking reactions, and combining with mass spectrometry analysis, they tracked the changes in cross-linking sites during protein unfolding, thereby revealing the dynamic evolution process of protein structure [59].

2.5. Co-Fractionation Mass Spectrometry (CF-MS)

2.5.1. Principles and Technical Advantages

CF-MS is based on the principle that interacting proteins co-migrate during biochemical fractionation (Figure 1c). By quantifying protein abundance across eluted fractions with high-resolution mass spectrometry and applying correlation analysis to co-elution patterns, physical associations can be inferred [60]. Unlike affinity purification mass spectrometer (AP-MS), CF-MS requires no antibodies or exogenous tags, thereby preserving native protein states and avoiding functional interference [61]. Additionally, the use of multiplexed mass tags enables quantitative comparisons across different biological conditions or cell types [62,63], making CF-MS a powerful method for mapping global protein interactomes.

2.5.2. Limitations and Optimization Strategies

The technique faces several challenges, including false positives from non-specific co-elution, overlapping elution profiles due to limited separation resolution, and the static nature of the data [1,61,63,64]. Recent methodological innovations have directly addressed these issues. For example, Wei Liu et al. introduced the XL-CoFrac-Q-MS workflow, which employs cross-linkers such as Disuccinimidyl sulfoxide (DSSO) to fix spatial proximity in situ prior to cell lysis [62]. Their accompanying software, ChromaQuant, enables accurate peak quantification to effectively filter noise-driven false positives [62]. In parallel, Pierre C. Havugimana et al. developed the multiplexed Co-Fractionation MS (mCF/MS) platform, which integrates tandem mass tag (TMT) multiplexing with wide-window size-exclusion chromatography (SEC) and the Synchronous Precursor Selection-MS3 (SPS-MS3) acquisition strategy [63]. This approach achieved baseline separation for over 95% of protein complexes. Using the CoFracNet algorithm, the team performed high-resolution interactome comparisons between mammary epithelial cells and breast cancer cells, demonstrating the platform’s capability for differential complex analysis [63].

2.5.3. Broad Applications in Complex Systems

Continued methodological refinements have significantly expanded the applicability of CF-MS across diverse biological systems [64]. McWhite et al. constructed the first pan-plant protein complex atlas by analyzing 13 plant species, identifying 13,287 conserved complexes that span major organelles and the cytoskeleton [65]. In neuroscience, Zilocchi et al. established an optimized CF-MS workflow for neuronal samples, identifying 217 known and 109 novel mitochondrial complexes and characterizing their assembly states in mammalian primary neurons and brain tissues [66]. For dynamic signaling studies, Wei Liu et al.) combined surface-charge-based fractionation with phosphotyrosine (pTyr) enrichment to track functional protein assemblies [60]. Using their ComplexQuant algorithm, they identified 216 pTyr-associated complexes that disassemble during T cell exhaustion and linked 78 of these directly to programmed cell death protein 1 (PD-1) expression levels, providing mechanistic insight into immune dysfunction [60].

2.6. Proximity-Based Analysis Methods

Proximity analysis techniques are a class of interaction screening methods based on molecular proximity, capable of capturing transient and weak interactions in living cells with subcellular spatial resolution. The technology comprises two core branches: one focuses on proximity-enhanced reactions (PERs) for signal amplification and visualization/quantification, including Proximity Ligation Assay (PLA), Proximity Extension Assay (PEA), and Proximity Proteolysis Assay (PPA); the other centers on Proximity Labeling (PL) for covalent labeling and interactome identification. Both rely on proximity-triggered reactions but differ in technical goals, mechanisms, and applications, together forming a core system for studying non-nucleic acid molecular interactions.

2.6.1. Proximity-Enhanced Reactions

1. Proximity ligation assay (PLA)

Proximity Ligation Assay (PLA) enables in situ interaction detection through antibody recognition and DNA signal amplification. Two primary antibodies target the proteins of interest, followed by secondary antibodies conjugated with complementary DNA probes. When the proteins are within 40 nm, the probes ligate to form a circular DNA template, which is amplified via Rolling Circle Amplification (RCA) to generate long DNA concatemers (Figure 1e). Fluorescent probes then hybridize to these concatemers, producing localized fluorescent signals for visualization and quantification of single-molecule interactions [67].

PLA amplifies signals through DNA amplification (e.g., RCA), enabling detection of low-concentration protein interactions or modifications, including transient and weak interactions, and is suitable for micro-sample analysis. It converts protein interaction data into fluorescent signals, allowing in situ visualization in cells or tissues to clarify interaction locations. However, the method depends on high-quality specific antibodies; antibody specificity, affinity, and penetration directly affect results, with poor performance risking false positives or negatives. Non-specific antibody binding, oligonucleotide hybridization, or DNA ligation may also produce background noise. Additionally, consumables such as antibodies, oligonucleotide probes, and fluorescent reagents are costly, and some require fresh preparation, increasing experimental expense.

Recent advancements in PLA focus on improving signal amplification efficiency and detection throughput. Multi-color PLA probes have been developed to enable simultaneous detection of multiple protein interactions in the same cell. In 2006, Masoud Vedad et al. combined PLA with immunofluorescence to simultaneously capture protein localization and interaction information, enhancing subcellular-level analysis [68]. In 2020, Chen Y et al. introduced a microfluidic PLA chip, increasing throughput while reducing sample and reagent consumption, making it suitable for clinical micro-samples [69].

PLA is widely used to validate and localize known protein interactions. In 2021, Felipe Melo-González et al. applied PLA to analyze the dynamic proximity between Cluster of Differentiation 3 (CD3) and Zeta chain of T cell receptor-associated protein kinase 70 Gene (Zap70) during T cell activation, revealing immune signaling mechanisms [70]. In 2013, Nair DP et al. used PLA to locate interaction sites between Postsynaptic Density Protein 95 (PSD95) and N-methyl-D-aspartic acid receptor (NMDA) receptors in neuronal synapses, clarifying synaptic plasticity mechanisms [71]. In 2024, Yinghui Qiu et al. enhanced standard PLA by introducing an engineered ~40 nt “splint probe” after antibody-probe binding. This approach enabled analysis of HER2–p95HER2 interaction heterogeneity in mouse breast cancer tissues and real-time tracking of EGF-induced EGFR–Grb2 translocation between the plasma membrane and early endosomes [68].

• 2.Proximity Extension Assay (PEA)

PEA is a multiplexed, highly sensitive quantitative technique designed for biomarker screening in precision medicine, enabling the detection of low-abundance non-nucleic acid targets in complex samples like serum or cell lysates. It employs pairs of affinity probes—such as antibodies or aptamers—each conjugated with unique oligonucleotides that recognize distinct epitopes of the target molecule. When both probes bind to the target, their oligonucleotides are brought into close proximity, hybridize via complementary ends, and are extended by DNA polymerase to form a quantitative real-time PCR (qPCR) template. This template is subsequently quantified via real-time PCR or high-throughput sequencing. PEA can simultaneously quantify over 100 proteins from just 1 µL of sample with femtomolar sensitivity, high reproducibility, and strong anti-interference capability, making it well-suited for large-scale clinical profiling [72,73,74,75].

• 3.Proximity Proteolysis Assay (PPA)

PPA is an amplification-free method based on enzymatic cleavage and signal release, avoiding nucleic acid contamination and allowing real-time monitoring of dynamic interactions in living cells or nucleic acid-sensitive systems. The system typically uses target-binding domains fused to protease components: either split inactive protease fragments that reassemble upon target binding, or a full protease paired with a domain carrying its specific cleavage sequence. Target-driven proximity activates the protease, which then cleaves an added reporter substrate—often a FRET-based peptide—resulting in a measurable fluorescence change. This approach offers low background, rapid response, and suitability for tracking live-cell dynamics without nucleic acid interference [76,77,78,79].

2.6.2. Proximity Labeling (PL)

Proximity labeling involves fusing a target protein with an engineered enzyme—such as biotin ligase mutants (BioID/TurboID) or peroxidase (APEX2)—that catalyzes covalent biotin tagging of neighboring molecules. The biotinylated proteins are then enriched via streptavidin beads and identified by mass spectrometry, enabling mapping of proximal interactomes and spatial proteomics without reliance on high-quality antibodies or in vitro purification (Figure 1f). This approach excels at capturing transient or weak interactions but faces challenges including non-specific labeling, high background, and potential cellular toxicity [80,81,82,83,84].

Recent advances have addressed these limitations. Branon TC et al. (2018) engineered TurboID and miniTurbo for rapid, efficient biotinylation within 10 min, significantly improving labeling speed and reducing toxicity compared to earlier variants [84]. The Ting lab enhanced peroxidase-based labeling by developing APEX2, which offers higher catalytic efficiency and broader applicability [81]. Subsequent optimizations extended APEX2 to various organelles and subcellular structures [85], while APEX-Seq enabled transcriptome-wide mapping of RNA spatial localization in mammalian cells, revealing isoform-specific distribution patterns [80,81].

In a complementary approach to general proximity labeling, Tian Ruijun’s group developed Tris-succinimidyl trioxane (TSTO), an MS-cleavable homotrifunctional cross-linker that covalently links three proximal lysine residues. This method enables the structural “triangulation” of higher-order protein complexes and over-comes computational challenges in identifying multi-peptide cross-links, providing precise evidence of multimeric assembly architectures [86]. Combined with earlier NHS-biotin-diazirine probes for enrichment, this chemical toolkit enhances the resolution of interactome mapping for both basic and clinical research [87].

3. Methods for Validating PPIs

The interaction networks constructed by screening techniques often have problems such as false positives or indirect interactions. Validation techniques are needed to confirm the key interaction relationships in the network, clarify the direct associations between nodes, and improve the reliability and accuracy of the interaction network (Figure 2). Fluorescence Resonance Energy Transfer (FRET) is one of the core technologies for PPI validation, which can accurately verify direct protein interactions in real time in living cells, providing key evidence for the precise construction of interaction networks.

Figure 2.

Methods for validating PPIs. It determines whether the bait and prey proteins interact by detecting the non-radiative energy transfer from the donor fluorophore to the acceptor fluorophore.

3.1. Fluorescence Resonance Energy Transfer (FRET)

Fluorescence Resonance Energy Transfer (FRET) refers to the phenomenon where when two fluorescent chromophores are sufficiently close, exciting the donor molecule with light of a specific wavelength can transfer energy to the adjacent acceptor molecule through dipole–dipole interaction, causing a change in its energy and generating fluorescence, thereby realizing resonant energy transfer (Figure 2). This characteristic allows FRET to accurately capture close-range interactions between molecules, making it an important tool for analyzing the dynamic relationships of biomolecules.

The most prominent feature of FRET in biological research is that it can perform real-time monitoring in living cells, thereby observing the dynamic changes and interactions of biomolecules under normal physiological conditions [88]. Moreover, this technology can perform multi-color labeling, providing possibilities for the research of complex biological systems. However, the limitations are also relatively obvious. Firstly, it is necessary to conjugate fluorescent substances on the molecules to be detected for labeling, and this labeling process may affect the normal functions of the molecules [88,89,90]. Secondly, the intensity of the FRET signal is limited by distance; effective energy transfer can only be detected when the distance between the donor and the acceptor is within the range of 1–10 nm, which limits its application in a wider distance range. Thirdly, poor spectral overlap between the donor and acceptor or other background fluorescence in complex biological samples may cause fluorescence interference, affecting the reliability of experimental results [91]. Fourthly, each molecule has its specific spectral characteristics and biocompatibility, making it difficult to meet all research needs [88].

To further break through the application bottlenecks of traditional FRET and meet the needs of protein interaction validation in different research scenarios, researchers have developed a variety of derived technologies based on traditional FRET, including single-molecule Fluorescence Resonance Energy Transfer (smFRET), Time-Resolved Fluorescence Resonance Energy Transfer (TR-FRET), Fluorescent Protein FRET (FP-FRET), and FRET Biosensors. These derived technologies have made targeted optimizations for the limitations of traditional FRET, greatly expanding the application scope and accuracy of FRET technology in protein interaction validation (Table 1).

3.1.1. Single-Molecule Fluorescence Resonance Energy Transfer (smFRET)

smFRET is a single-molecule-level quantitative analysis technology based on the FRET phenomenon. It abandons the averaging processing of group molecular signals by traditional FRET and can directly observe the dynamic process and heterogeneity characteristics of interactions between individual protein molecules [92]. Its advantages lie in the single-molecule level resolution, which can accurately capture kinetic details such as conformational transitions, binding and dissociation rates in protein interactions, and can distinguish the differential performance of different molecular individuals in interactions. In terms of application scenarios, smFRET is widely used in studies that require precise capture of individual molecular behaviors, such as analyzing the allosteric regulation mechanism of enzymes, the assembly and dissociation process of protein complexes, and the dynamic interactions of membrane proteins [93].

In 2023, Maus H et al. site-specifically introduced the donor dye Alexa Fluor 488 and the acceptor dye Alexa Fluor 594 at the N-terminus (Gly10) of Non-structural protein 2B (NS2B) and the C-terminus (Lys184) of Non-structural protein 3 (NS3) in Zika Virus (ZIKV) NS2B-NS3pro, respectively. Single-protein molecules were diluted to picomolar concentration using a microfluidic chip, and the FRET efficiency (E_FRET) trajectories of >10,000 single molecules were recorded in real time using Total Internal Reflection Fluorescence Microscopy (TIRF). This revealed the differential regulatory mechanisms of allosteric inhibitors and competitive inhibitors on the conformational dynamics of ZIKV NS2B-NS3 protease [94]. In 2023, Gobert Heesink et al. systematically revealed the residue-level inter-regional and intra-regional dynamic interaction network in α-synuclein, a typical intrinsically disordered protein, by integrating smFRET experiments with multi-scale Molecular Dynamics (MD) simulations [95].

3.1.2. Time-Resolved Fluorescence Resonance Energy Transfer (TR-FRET)

The core innovation of TR-FRET is the introduction of a time-delayed detection strategy. After exciting the fluorescent donor, the signal is not collected immediately but after a delay. This design can effectively eliminate interference signals such as autofluorescence of biological samples and background fluorescence generated by non-specific binding, significantly improving the signal-to-noise ratio [93,96]. Compared with traditional FRET, TR-FRET has stronger anti-interference ability, and is particularly suitable for protein interaction validation in complex biological samples such as cell lysates, tissue homogenates, and serum. In addition, this technology has quantitative analysis capabilities and is often used as a precise validation method for positive interactions after high-throughput screening. It is also widely used in research scenarios such as quantitative detection of the interaction intensity between drugs and target proteins, and evaluation of the regulatory effect of small-molecule compounds on protein interactions [96].

In 2020, Ville Eskonen et al. developed a high-sensitivity Time-Resolved Fluorescence Resonance Energy Transfer (TR-FRET) platform based on single-peptide probes without antibodies, realizing the multiplex quantitative detection of various cysteine-specific post-translational modifications (Cys-PTMs) for the first time, including S-nitrosylation (SNO), S-glutathionylation (SSG), S-sulfonation (SSO₃H), and S-palmitoylation [97]. In 2023, Yue et al. reported a revolutionary homogeneous TR-FRET serological detection platform for the specific and quantitative detection of IgG antibodies targeting any antigen in unpretreated human serum, plasma, and whole blood samples, achieving clinical-grade serological diagnosis with “1 μL sample, 1 h completion, and one-click adaptation to new antigens” [98].

3.1.3. Fluorescent Protein FRET (FP-FRET)

FP-FRET uses fluorescent proteins (such as Green Fluorescent Protein GFP, Yellow Fluorescent Protein YFP, Cyan Fluorescent Protein CFP, etc.) as donors and acceptors for FRET. Fluorescent proteins are directly fused and expressed with target proteins through genetic engineering technology, eliminating the need for in vitro fluorescent group conjugation labeling [93]. This characteristic fundamentally avoids the interference of exogenous fluorescent labeling on the native structure and function of proteins, enabling in situ, non-interfering real-time monitoring of protein interactions under the physiological environment of living cells [99]. However, FP-FRET also has certain limitations. Due to the relatively fixed spectral overlap, quantum yield, and other characteristics of fluorescent proteins themselves, their detection sensitivity for weak protein interactions is slightly lower than that of traditional FRET with chemical fluorescence labeling, and they are more suitable for verifying medium-to-high intensity intracellular protein interactions [21,88].

Fluorescence Resonance Energy Transfer (FRET) technology combined with Fluorescent Proteins (FPs) has become the gold standard method for studying protein–protein interactions, enzyme activity, and small-molecule dynamic changes in living cells [88]. Its core advantages include no need for separation, non-invasiveness, real-time performance, spatial resolution, and the ability to effectively eliminate non-specific interference such as concentration, optical path, and excitation light fluctuations through ratiometric imaging, significantly improving quantitative accuracy and signal-to-noise ratio. In 2011, Hesam Shahravan et al. constructed a Cyan Fluorescent Protein (CFP)-6×Activator Protein 1 (AP1)-Yellow Fluorescent Protein (YFP) biological probe and used FP-FRET technology to monitor the interaction between proteins and DNA in real time [100].

3.1.4. FRET Biosensors

FRET biosensors are integrated detection tools. Their design core is to fuse a donor fluorophore, an acceptor fluorophore, and a specific “sensing module” (usually a protein domain, peptide fragment, etc., that can specifically recognize or bind to target interaction molecules) [101]. When the target protein interacts, it triggers a conformational change in the sensing module, thereby changing the distance or relative orientation between the donor and acceptor fluorophores, ultimately leading to a significant change in FRET signal intensity [102]. The prominent advantage of this technology is that it can not only verify the existence of protein interactions but also quantify in real time the intensity and duration of interactions, and simultaneously correlate and detect upstream and downstream regulatory signals of interactions (such as the effect of phosphorylation and ubiquitination modifications on protein interactions) [103].

In applications, FRET biosensors are widely used for dynamic tracking of protein interactions in intracellular signaling pathways, such as the interaction activation process between kinases and substrates, and the regulation of protein interactions in G protein-coupled receptor signaling pathways [104]. In 2018, Michelle L et al. used genetically encoded FRET biosensors as spatiotemporally resolved tools in living cells to systematically reveal the dynamic organization mechanism of G protein-coupled receptor (GPCR) signals in subcellular compartmentalization [105]. In 2019, Alessandra Ghigo et al. established the core law of cyclic Adenosine Monophosphate (cAMP) compartmentalization based on a genetically encoded biosensor constructed on the principle of Fluorescence Resonance Energy Transfer (FRET) [106]. Its functional realization is highly dependent on subcellular spatial compartmentalization, rather than the traditional cognition of uniform cytoplasmic diffusion [106].

Table 1.

Comparison of FRET Techniques.

Technique	Principle	Advantages	Limitations	Applications	References
smFRET	Single-molecule FRET imaging	Single-molecule resolution, captures dynamics.	Complex operation, high cost.	Enzyme allostery, assembly/dissociation.	[92,93]
TR-FRET	Delayed signal acquisition	Anti-interference, suitable for complex samples	Requires specific fluorophores, costly.	Drug–target assays, serological diagnostics.	[93,96]
FP-FRET	Genetically encoded fluorescent protein fusions	Non-invasive, live-cell monitoring	Low sensitivity, poor detection of weak interactions	Live-cell protein/DNA interactions.	[88,93]
FRET Biosensors	Sensor module-triggered signal changes	Quantitative analysis, spatial resolution	Sensor specificity critical, environment-sensitive.	Signaling pathway tracking, GPCR studies.	[101,102,103]

4. Toward Dynamic Structures and Functions: Future Directions for Protein Network Analysis

Through established screening and validation technologies, multi-dimensional raw data on protein–protein interactions (PPIs) can now be generated. Low-throughput methods yield precise, validated data for specific interactions, high-throughput technologies produce large-scale interaction networks, and dynamic techniques capture the spatiotemporal regulation of PPIs.

However, these data remain fragmented and technically limited: high-throughput datasets are prone to false positives, and dynamic data are often obscured by noise [107]. Systematic data integration and analysis are therefore essential to transform raw data into biologically coherent models of protein interaction logic.

Such integration is critical not only for interpreting existing data but also for establishing the foundation to study more complex protein functions. A key emerging direction is the study of protein machines: multi-subunit complexes whose assembly, activity, and regulation depend on dynamic, specific, and often stimulus-responsive interactions between subunits [108]. These sophisticated systems may undergo conformational changes, couple to energy-dependent processes, and be modulated by ligands and reversible post-translational modifications [109].

Accurately revealing the assembly and regulatory mechanisms of protein machines requires the integration and optimization of low-throughput, high-throughput, and dynamic approaches. Advancing from data collection to mechanistic insight—especially in the study of protein machines—depends on building more powerful integrative analysis systems. Innovations such as AlphaFold are poised to play a pivotal role in this process by providing structural context for interaction data [110]. Already, systematically integrated PPI data can provide a preliminary interaction framework for many protein machines, marking a promising step toward understanding their full complexity.

Future advancements in protein interaction network analysis will focus on four core technological frontiers.

1. Multi-technology and multi-omics integration. This involves synthesizing high-throughput screening (e.g., AP-MS, CF-MS), precise biophysical validation (e.g., FRET), and structural analysis (e.g., cryogenic Electron Microscopy (cryo-EM)) with genomic and transcriptomic data. This integrative paradigm was exemplified in 2023, when Francis et al. combined XL-MS, co-elution MS, and AI structure prediction (AlphaFold2/Multimer) to systematically model the dynamic protein complexes of Bacillus subtilis, shifting structural proteomics toward physiological mechanisms [111].

2. AI-assisted prediction and network refinement. Artificial intelligence is revolutionizing both predictive modeling and experimental data processing. AI tools like AlphaFold are being used to predict interaction interfaces and complex structures while simultaneously filtering false positives from high-throughput datasets to enhance network accuracy. Recent benchmarks include a comprehensive human PPI map constructed using deep learning models and experimental cross-validation [112]. The advent of AlphaFold 3, capable of predicting atomic structures for diverse biomolecular complexes (proteins, nucleic acids, ligands), has further elevated interaction modeling to unprecedented resolution [113].

3. In vivo dynamic tracking. The ultimate frontier is the real-time observation of network dynamics in living systems. This requires the advancement of proximity labeling, biosensors, and fluorescence imaging technologies suitable for intact tissues and organisms to reveal how interaction networks remodel during physiological and pathological processes.

4. PPI database integration and application. As the core carrier for data integration and sharing, PPI databases serve as a critical cornerstone for advancing in-depth protein network analysis. Current mainstream databases, such as STRING (https://cn.string-db.org/, accessed on 10 February 2026) and BioGRID (https://thebiogrid.org/, accessed on 10 February 2026), possess distinct features: The STRING database covers thousands of species, integrates multi-dimensional data including experimentally validated results and genomic co-evolution information, and provides confidence scores, thereby facilitating the rapid construction of comprehensive PPI network frameworks. BioGRID focuses on curating experimentally validated PPI data, encompassing outcomes from various techniques such as yeast two-hybrid and Co-IP, with detailed annotations and high reliability.

In conclusion, the evolution of PPI technologies is fundamentally shifting our understanding of cellular regulation from linear pathways to dynamic networks. Moving forward, the iterative refinement of AI tools like AlphaFold and their deeper integration with multi-dimensional omics data will further illuminate the assembly and regulation of complex protein machines. This progress will directly accelerate the translation of molecular mechanisms into therapeutic target discovery, expanding the frontiers of both basic and applied life science research.

Abbreviations

The following abbreviations are used in this manuscript:

PPIs	Protein–protein interactions
LC-MS/MS	Liquid chromatography–tandem mass spectrometry
Y2H	Yeast Two-Hybrid
CO-IP	Co-Immunoprecipitation
AP-MS	Affinity Purification–Mass Spectrometry
XL-MS	Chemical Cross-Linking Mass Spectrometry
CF-MS	Co-Fractionation Mass Spectrometry
PLA	Proximity ligation assay
PEA	Proximity extension assay
PPA	Proximity extension assay
PL	Proximity labeling
FRET	Fluorescence Resonance Energy Transfer

Author Contributions

X.Y. conceived and designed this project. X.Y. and W.C. wrote the draft of the manuscript. L.W. and Y.Z. revised the manuscript. X.Y. prepared and edited the tables and figures. All authors have read and agreed to the published version of the manuscript.

Data Availability Statement

No new data were created or analyzed in this study. Data sharing is not applicable to this article.

Conflicts of Interest

The authors declare no conflicts of interest.

Funding Statement

This work was supported in part by National Natural Science Foundation of China (32560242), Natural Science Foundation of Jiangxi Province (20242BAB26092), granted to Y.Z., Ganzhou City Science and Technology Plan (2023PNS26873) and Ganzhou Key R&D Program Project (GZ2024YU121) granted to L.W.