Abstract
The inherent heterogeneity of cancer leads to varied responses to treatment and underscores the need for the development of precision medicine. Single-cell proteomics is crucial for deciphering intra- and inter-tumor heterogeneity. Mass cytometry and other antibody-based technologies, which enable simultaneous quantification of >100 proteins per cell, have been widely applied in cancer studies. Concurrently, liquid chromatography-tandem mass spectrometry-based single-cell proteomics has dramatically improved protein coverage from approximately 1,000 to >6,500 proteins per cell, driven by enhanced sample preparation, instrumentation, and throughput. These high-throughput capabilities now empower large-scale, protein-level analyses, which have uncovered detailed maps of tumor heterogeneity. The resulting insights deepen our understanding of tumor biology and provide new opportunities for guiding precision cancer medicine.
keywords
- Mass spectrometry
- single-cell protein analysis
- single-cell proteomics
- tumor heterogeneity
- precision medicine
Introduction
Proteomics has been widely applied in cancer research, which has enabled the discovery of key regulators and pathways in clinical samples, animal models, and cultured cells, and elucidated the molecular mechanisms underlying tumorigenesis and progression. However, most studies rely on bulk analysis, which provides population-level averages across all cells in a biological system, thereby obscuring proteomic heterogeneity among individual cells. The intra- and inter-tumor heterogeneity of tumors drives the variable responses to therapies. To address such challenges, precision medicine is proposed, aiming to deliver personalized therapies based on specific biological signatures of individual patients. Notably, single-cell level analysis is the cornerstone to reveal heterogeneity within tumors as well as variability in tumor cellular composition across patients.
While technologies, such as single-cell genomics, single-cell transcriptomics, and spatial transcriptomics, are now well-established and provide extensive genomic and transcriptomic-level information, single-cell protein analysis remains a challenge1–4. Although liquid chromatography-tandem mass spectrometry (LC-MS/MS) has served as the standard method for proteomics, the application to single cells was initially limited by sensitivity constraints and the minimal protein content in individual cells. Consequently, early single-cell protein studies primarily relied on antibodies. Advances in multiplexing for antibody-based methods, such as the use of metal ions5 and DNA barcodes6, allow simultaneous detection of >100 proteins at the single-cell level. Antibodies can also be applied to tissue sections, yielding spatially resolved protein mapping at the single-cell resolution7. Recently, with advances in LC-MS/MS technology, thousands of proteins can now be quantified from nanogram-scale samples, a breakthrough that has further accelerated the development of single-cell proteomics (SCP). Since the first SCP workflows were introduced8–10, a variety of technologies leveraging state-of-the-art instruments and methodologies have emerged. These technologies optimize key procedures, such as single-cell isolation, sample preparation, liquid chromatography, and mass spectrometry, ultimately increasing the number of proteins detectable in a single cell to >6,50011. These paradigms differ fundamentally in detection chemistry, analytical depth, throughput, and contextual information. Antibody-based platforms rely on predefined panels, enabling high-throughput and multiplexed yet targeted protein detection. In contrast, MS-based approaches provide broad, hypothesis-free proteome coverage without prior target selection. MS-based methods generally achieve substantially higher protein coverage per cell, whereas antibody-based technologies offer superior throughput and scalability. Spatial resolution introduces an additional dimension of complexity because imaging mass cytometry and spatial MS workflows incorporate tissue context, enabling microenvironment-aware proteomic profiling. Integrative strategies combining laser capture microdissection with nano-scale proteomic workflows further bridge spatial information and deep proteome coverage.
SCP enables more refined cell classification using high-dimensional datasets, while identifying cellular and molecular signatures associated with pathologic abnormalities, patient stratification, and drug response prediction. In-depth protein analysis may also uncover molecular mechanisms underlying drug resistance and facilitate the discovery of novel therapeutic targets. In this review, advances in single-cell protein analysis are comprehensively summarized, including antibody-based approaches, LC-MS/MS-based SCP, and spatially resolved proteomic technologies, with a focus on the applications in cancer research and precision oncology. Importantly, the majority of the current SCP platforms still face issues, such as insufficient throughput. This is particularly critical because different research goals impose distinct technical priorities. For example, studies aimed at biomarker discovery or cohort-level stratification often prioritize analytical throughput and reproducibility, whereas mechanistic investigations of tumor heterogeneity or signaling demand maximal proteome depth, ideally with single-cell resolution.
Antibody-based protein analysis in single cells
Antibody-based techniques provide high specificity and sensitivity through the detection of targeted epitopes. These methods can be broadly categorized into two classes: flow cytometry-based technologies; and multiplexed imaging technologies (Table 1). In conventional flow cytometry, fluorophore-conjugated antibodies are used to label specific surface or intracellular proteins and individual cells suspended in fluid flow past laser detectors where emitted fluorescence signals are recorded12,13. Spectral overlap typically limits simultaneous detection to ~10 proteins, which constrains multiplexing capacity. Recent advances in spectral flow cytometry have further expanded multiplexing capacity by capturing the full emission spectra of fluorophores and applying computational spectral unmixing algorithms. This strategy reduces signal overlap and increases the number of detectable markers compared to conventional fluorescence-based flow cytometry, while maintaining high throughput.14 Mass cytometry (e.g., CyTOF) overcomes spectral limitations by using antibodies conjugated to lanthanide or other transition metal isotopes5. Because labeled single cells pass through an inductively coupled plasma (ICP) ionization source, bound metal tags are ionized and quantified using time-of-flight (TOF)-MS, providing highly multiplexed, quantitative protein measurements at the single-cell level. Mass cytometry routinely detects 40–50 proteins with a theoretical capacity exceeding 100 markers. However, these methods require tissue dissociation, which disrupts the tissue spatial architecture and prevents analysis of cellular localization and cell-cell interactions.
Antibody-based single-cell proteomic technologies
Multiplexed imaging technologies enable in situ protein detection while preserving the spatial organization of tissues. These methods utilize antibodies labeled with metal isotopes, fluorophores, or DNA oligonucleotides, allowing multiplexed and spatially resolved protein analysis15. Metal isotope-conjugated antibodies are used to stain tissue sections in imaging mass cytometry (IMC), followed by laser ablation and TOF-MS to map the spatial distribution of approximately 30–40 proteins simultaneously at a 1-μm resolution7,16. Multiplexed ion beam imaging (MIBI) replaces the laser with a rasterized ion beam, achieving higher resolution (200–300 nm)17,18. Recently, the expand and compress hydrogels (ExPRESSO) platform has further enhanced this approach by integrating with physical tissue expansion, improving resolution and signal detection while remaining compatible with IMC and MIBI19. These methods provide key advantages, such as high multiplexing capacity without spectral overlap, minimal background signal, and simultaneous detection of all markers. Nevertheless, the methods face challenges, such as the need for vacuum-compatible sample preparation, reliance on specialized instrumentation, and slow image acquisition for large tissue areas.
Fluorescence-based multiplexed imaging is supported by widely available commercial antibodies and conventional microscopy systems. However, spectral overlap inherently limits the number of detectable markers. To overcome this constraint, cyclic imaging methods, e.g., tissue-based cyclic immunofluorescence (t-CyCIF) and iterative bleaching extends multiplexity (IBEX), have been developed20–24. These techniques use iterative rounds of staining, imaging, and signal inactivation, using chemical bleaching or antibody removal to sequentially detect >60 proteins on the same tissue section. While these cyclic fluorescence methods are practical and cost-effective, meticulous optimization is required to minimize signal carryover and tissue damage across successive cycles. Hyperplexed immunofluorescence imaging (HIFI) utilizes a thiol-based elution strategy for gentle antibody removal, enabling detection of >45 markers per tissue while preserving tissue integrity25. More recently, scission-accelerated fluorophore exchange (SAFE) has extended fluorescence multiplexing to live cells and fresh tissues by using a bio-orthogonal click-to-cut reaction for rapid fluorophore removal, enabling longitudinal imaging in living samples26.
Moreover, DNA-barcoded antibodies provide an alternative strategy to overcome the limitation of spectral overlap. Among these DNA-barcoded antibodies, co-detection by indexing (CODEX) has become a widely adopted platform, utilizing antibodies conjugated with unique DNA barcodes. Fluorescently labeled complementary oligonucleotides are sequentially hybridized, imaged, and stripped, allowing detection of up to 100 protein markers without cumulative tissue damage6,27–29. Recent efforts have focused on integrating CODEX data with transcriptomic and genomic information for multi-omic spatial analysis30. Similarly, immunostaining with signal amplification by exchange reaction (Immuno-SABER) used iterative DNA hybridization for scalable multiplexing, incorporating a signal amplification mechanism that significantly boosts detection sensitivity31. While DNA-based methods offer high multiplexing potential, the imaging throughput is constrained by the sequential nature of hybridization cycles.
Despite these technical advances, effective panel design and antibody validation remain key bottlenecks limiting broader adoption. To address these challenges, the organ mapping antibody panels (OMAP) initiative was established as a community-driven, standardized repository of rigorously validated antibody panels optimized for tissue-specific multiplexed imaging applications32. Designed to support methods, such as CODEX, IBEX, and IMC, OMAP are tailored to organ-specific anatomic structures and cell types, aiming to improve reproducibility and comparability across studies. Nevertheless, while antibody-based strategies have transformed single-cell phenotyping, the reliance on predefined targets inherently restricts discovery potential. Achieving a systems-level understanding of regulatory networks at the single-cell resolution ultimately requires comprehensive and untargeted proteome profiling beyond antibody-based detection.
MS-based protein analysis in single cells
MS-based approaches represent the most comprehensive strategies for unbiased proteome analysis in single cells, offering untargeted and proteome-wide detection. One of the earliest approaches used matrix-assisted laser desorption/ionization (MALDI), which enables soft ionization of peptides and proteins directly from single cells with minimal sample handling33,34. In addition, by rastering a laser across matrix-coated tissue, MALDI-mass spectrometry imaging (MSI) generates ion maps that reflect the distribution of proteins, lipids, and metabolites with spatial resolution to the subcellular level35,36. However, it remains challenging for MALDI-based strategies in SCP applications largely due to insufficient sensitivity for protein analysis.
To meet the demands of large-scale proteome profiling at the single-cell level, the proteomics community has sustained efforts to advance LC-MS/MS-based methodologies. Recent advances in sample preparation, ultrasensitive MS detection, and quantitative acquisition strategies have collectively enabled the detection of thousands of proteins per cell with the potential to capture post-translational modifications (PTMs) and protein variants37–40. Notably, these improvements have extended the scope of SCP to spatially resolve protein profiling in tissues. In the following sections, the key developments in sample processing, detection technologies, and quantification strategies that have propelled the field toward increasingly comprehensive and spatially informed single-cell analyses are discussed.
Sample preparation for SCP of cell suspensions
MS-based SCP presents unique challenges in sample preparation because the minimal protein content in individual cells requires highly efficient cell lysis, digestion, and peptide recovery to minimize sample loss and maximize proteome coverage. Early studies, such as single cell proteomics by mass spectrometry (SCoPE-MS), involved manually isolating individual live cells under a microscope using micropipettes, followed by lysis and digestion in relatively large volumes (~10 μL)8. Although technically straightforward, this approach suffered from sample dilution and adsorption losses. Over the past few years, SCP sample preparation has evolved substantially towards automated and nanoliter-scale pipelines, leveraging microfabricated chips, digital microfluidics, and automated dispensing systems to minimize sample loss, while enabling parallel preparation of hundreds to thousands of single cells.
Microfabricated chip-based platforms use micro- and nano-fabrication techniques to miniaturize reaction spaces (Figure 1A). These platforms can be broadly categorized into open microfabricated platforms and closed microfluidic chip platforms. Open microfabricated platforms utilize patterned nanowells or hydrophobic surfaces to confine nanoliter-scale liquid droplets, relying on surface tension to maintain reaction integrity for single-cell processing in an open environment. These methods require high-precision external dispensing systems to accurately position single-cell suspensions and reagents onto the chip surface or into nanowells. Early methods, such as the oil-air-droplet (OAD) chip, rely on manually dispensing nanoliter droplets containing single cells and reagents into oil-covered nanowells with the oil layer reducing evaporation and preventing cross-contamination9. In addition, nanodroplet processing in one pot for trace samples (nanoPOTS) used microfabricated hydrophilic nanowells on glass chips to confine 200-nL droplets, supporting low-loss lysis, digestion, and peptide recovery directly on-chip10. The subsequent nested nanoPOTS platform further improves miniaturization, reducing reaction volumes to 30 nL, while increasing nanowell density to enhance throughput and reaction efficiency41. The nanodroplet processing in an open microreactor (nPOP) platform introduces an alternative strategy, utilizing fluorocarbon-coated glass slides as open reaction substrates42,43. Nanoliter droplets containing individual cells and reagents are precisely dispensed onto patterned hydrophobic regions using the cellenONE system, enabling massively parallel sample processing in thousands of spatially arranged droplets. Reactions proceed in open droplets without the need for physical barriers. Similarly, the proteoCHIP EVO 96 platform uses a microfabricated PTFE chip with conical low-adsorption nanowells arranged in a 96-well format. Single cells and reagents are dispensed using the cellenONE system with the subsequent proteolysis occurring in these open nanowells. The use of low-retention materials and optimized geometry further minimizes sample losses44. Together, open microfabricated platforms offer a balance of miniaturization, parallel processing capacity, and operational simplicity, enabling high-throughput SCP workflows.
Sample preparation workflows for LC-MS/MS-based SCP. (A) Microfabricated chip-based workflow. Suspended cells are precisely dispensed onto open platforms, such as fluorocarbon-coated glass slides used in the nPOP workflow or hydrophilic nanowelled glass chips used in the nanoPOTS workflow. Alternatively, cell suspensions are loaded directly into closed microfluidic chips. Following in situ digestion on or within the chip, peptide solutions are transferred into autosampler vials for subsequent LC-MS/MS analysis. (B) Chip-Tip workflow. The CellenOne instrument precisely dispenses suspended cells into wells of the ProteoCHIP EVO 96, followed by programmatic addition of reagents for protein digestion. After digestion, peptides are centrifugally transferred from the chip to an Evotip for desalting and cleaning, simultaneously serving as the loading device for the Evosep One LC system. (C) One-pot and One-Tip sample preparation workflow. Cells are directly dispensed into LC-compatible containers for streamlined sample preparation. (D) Spatial proteomics workflow. LCM enables precise isolation of phenotype-specific cells from tissue sections for subsequent sample processing. Tissue expansion techniques enhance resolution by physically increasing sample volume. AM-DMF-SCP, active matrix digital microfluidic chip for single-cell proteomics; FACS, fluorescence-activated cell sorting; FF, fresh frozen; FFPE, formalin-fixed paraffin-embedded; LCM, laser capture microdissection; LC-MS, liquid chromatography-mass spectrometry; nanoPOTS, nanodroplet processing in one pot for trace samples; sciProChip, single-cell integrated proteomics chip; SCP, single-cell proteomics. Figure created using Adobe Illustrator.
Closed microfluidic chip platforms confine sample processing within sealed microchannels or chambers, using valves or electrowetting to enable automated, contamination-free workflows. Pre-engineered channel geometries, valve arrays, or electrode grids autonomously manage single-cell capture, reagent handling, and droplet partitioning, ensuring high precision and reproducibility. For example, pneumatically actuated valves control sequential reagent delivery, reaction isolation, and waste removal in integrated proteomics chip (iProChip). Single cells are size-selectively trapped in wedge-shaped microchambers, then lysed, digested, and desalted on-chip before LC-MS/MS analysis45. The active matrix digital microfluidic chip for the single-cell proteomics (AM-DMF-SCP) platform adopts a different architecture based on large-scale addressable electrode arrays. In a closed oil phase, nanoliter droplets containing single cells and reagents are manipulated via electrowetting, enabling fully programmable droplet movement, merging, and dispensing without physical valves or channels46. Overall, closed microfluidic platforms integrate capture, fluid handling, and proteomic processing within a controlled environment, providing an automated solution for SCP.
While microfabricated chip-based platforms can perform all key protein processing steps in “one pot,” efficient recovery of digested peptides remains crucial for subsequent LC-MS/MS analysis. This step may lead to additional sample loss through evaporation, surface adsorption, or incomplete recovery in open and closed formats. Peptides are transferred from the proteoCHIP EVO 96 into C18-packed Evotips using centrifugation for direct injection into LC-MS/MS systems in the Chip-Tip workflow (Figure 1B), while the nanoPOTS workflow incorporates a dedicated autosampler combined with a Dry-Extract-Load strategy11,44,47. Single cells are processed directly in standard 384-well plates or Evotips in one-pot and One-Tip workflows (Figure 1C), enabling low-loss lysis and digestion, while integrating seamlessly with automated liquid handling systems and commercial autosamplers, thereby supporting scalable, reproducible analyses48,49. Similarly, other platforms use packed tips, insert tubes, or capillaries to confine the entire workflow within a single closed reaction space50–53. Together, these systems highlight the value of streamlined device architectures in minimizing sample loss and leveraging automation to boost analysis throughput in SCP.
Some microfluidics-based systems focus on multifunctional solutions for upstream single-cell preparation and manipulation. These platforms emphasize deterministic single-cell capture, reaction confinement, and scalable automation, addressing key challenges shared across single-cell analytical workflows. For example, the T cell exhaustion and reactive oxygen species analyzer (TEROSA) integrates deterministic single-cell trapping with parallel microchamber architectures, enabling high-throughput and reproducible functional interrogation of individual T cells under tightly controlled microenvironments. This system enables multi-parametric analysis of oxidative stress and immune exhaustion dynamics with low technical variability by simultaneously monitoring mitochondrial superoxide levels, surface PD-1 expression, and extracellular hydrogen peroxide secretion at the single-cell level54. Similarly, the SMART microfluidic platform enables high-throughput single-cell capture, long-term dynamic culture, and multi-dose drug perturbation through integrated on-chip concentration gradient generators55. SMART supports efficient capture and longitudinal tracking of thousands of single cells using a self-limiting microchamber array optimized for hematopoietic cells, facilitating the analysis of cellular heterogeneity, clonal evolution, and drug resistance dynamics within a single experiment. Although these microfluidics-based platforms are not directly coupled to MS-based proteomic readouts, the engineering principles provide important design foundations for the development of next-generation SCP workflows that aim to balance sensitivity, robustness, and scalability.
Sample preparation for spatial proteomics
Spatial proteomics deciphers tissue heterogeneity by mapping protein expression in spatial context and was awarded as 2024 Method of the Year by Nature Methods56. Formalin-fixed, paraffin-embedded (FFPE) tissues have been extensively used in clinical pathology and represent a valuable resource for MS-based spatial proteomics pipelines57. A widely adopted strategy involves first distinguishing cellular phenotypes through histologic staining, such as hematoxylin and eosin (H&E), immunofluorescence (IF), or immunohistochemistry (IHC), followed by laser capture microdissection (LCM) to precisely isolate individual cells or cell clusters for downstream proteomic analysis. For example, the previously introduced nanoPOTS technology has been successfully adapted to accommodate H&E-stained fresh frozen (FF) single-cell samples obtained by LCM58.
Deep visual proteomics (DVP) stands out as the most integrated and advanced approach for spatial single-cell proteomics, combining high-resolution tissue imaging, artificial intelligence (AI)-driven cell phenotyping, and ultra-sensitive MS analysis to enable spatially resolved proteome profiling directly from tissue sections (Figure 1D)59. The workflow begins with whole-slide imaging of FFPE or FF tissues, followed by deep learning-based cell segmentation and classification using morphologic features and multiplexed IF markers. Targeted cell populations are then isolated via automated, high-precision laser microdissection using a coordinate-registered system that ensures exact spatial correspondence. Dissected cells are then transferred into low-retention plates for streamlined sample preparation, including thermal lysis, enzymatic digestion, and peptide cleanup. The resulting peptides are analyzed using nanoLC-MS/MS for deep and reproducible protein quantification. The single-cell implementation, single-cell DVP (scDVP), further scales this workflow to hundreds of spatially resolved single cells per experiment, achieving proteome depth of up to 4,300 proteins from one-third of a single cell60,61. Importantly, scDVP supports spatial projection of proteomic data onto tissue maps, allowing in situ correlation between molecular phenotypes and histologic context. scDVP have been applied across a wide range of biomedical studies, including high-resolution mapping of tumor-immune cell interactions, discovery of spatially confined immune cell subsets, validation of organoid models, and construction of human protein atlases59–63.
Another strategy, expansion proteomics (ProteomEx), improves spatial resolution and sampling flexibility through tissue expansion64. This approach applies hydrogel embedding and isotropic expansion to physically magnify formalin-fixed tissues, enabling manual microdissection at 0.61 nL μm volume resolution. In addition, filter-aided expansion proteomics (FAXP) achieves higher resolution (0.042 nL μm volume resolution) by enabling proteome-scale analysis from a single nucleus dissected in situ65. This method combines hydrogel-based expansion with high-temperature homogenization, LCM-based single-cell nucleus segmentation, and miniaturized “in-tip” digestion. Using this pipeline, >2,300 proteins were consistently quantified from individual nuclei in FFPE brain sections. FAXP preserves protein localization and enables histologic annotation through pre-expansion staining, enabling cell-type-resolved spatial proteome mapping across cortical layers.
Although not all spatial proteomics approaches achieve single-cell resolution, largely due to insufficient analysis throughput, spatial proteomics approaches can provide critical insights into tissue-level proteome heterogeneity and microenvironment structure. A recently developed multimodal pipeline [spatial and cell-type proteomics (SCPro)], combines multiplexed IHC, automated LCM, and highly sensitive ion exchange-based protein aggregation capture (iPAC) to achieve spatially resolved proteomic profiling from as few as 60 to as many 100 cells per region. SCPro applies spatial deconvolution algorithms to infer the composition and distribution of immune and stromal cell subsets in the tumor microenvironment (TME) by incorporating a cell-type reference atlas built from fluorescence activated cell sorting (FACS) proteomes of 14 major cell populations66. Another promising approach, microscaffold-assisted spatial proteomics (MASP), enables near-cellular resolution proteome mapping through micro-compartmentalization. A 3D-printed microscaffold with ~400-μm micro-wells is placed on FFPE sections mounted on polydimethylsiloxane and the tissue is partitioned into spatially registered micro-specimens via controlled pressurization, followed by microscale surfactant-assisted extraction and digestion67. 3D imaging of solvent-cleared organs profiled by MS (DISCO-MS) extends spatial proteomics to 3D tissues by combining solvent-based optical clearing, light-sheet imaging, AI-guided region selection, and robotic microdissection or biopsy with ultra-sensitive MS68. In summary, spatial proteomics advances are now reshaping our ability to understand tissue heterogeneity and cellular organization at unprecedented resolution, laying the groundwork for spatially informed precision medicine.
Protein quantification strategies in SCP
Quantification methods applied in SCP can be categorized into label-free quantification (LFQ) and isotopic labeling with the latter further divided into isobaric and non-isobaric formats. LFQ measures protein abundance by comparing precursor or fragment ion intensities across multiple LC-MS runs. The emergence of advanced MS instruments, such as Orbitrap Astral (Figure 2A) and timsTOF (Figure 2B), has significantly improved the sensitivity and throughput of LFQ39,40,69,70. When combined with robust DIA-based quantification, e.g., narrow-window data-independent acquisition (nDIA) or DIA parallel accumulation serial fragmentation (diaPASEF), LFQ can now support the quantification of >5,000 proteins per cell with decent reproducibility71,72. Advanced retention time alignment, feature matching, and missing value imputation algorithms further improve data completeness for SCP studies. The improvements in MS acquisition speed have increased the processing capacities from <10 to >120 single-cell proteomes per day using short LC gradients (Table 2)9,11. However, sample throughput remains a critical bottleneck because many biological experiments require profiling thousands-to-millions of individual cells, therefore demanding substantial instrument time and limiting broader applications of SCP.
Data acquisition modes in LC-MS/MS-based SCP. (A) Orbitrap Astral MS and nDIA acquisition. Orbitrap Astral MS integrates quadrupole, Orbitrap, and astral analyzers to enable asymmetric track lossless analysis. The nDIA mode combines the unbiased nature of DIA with the high selectivity of DDA, effectively addressing co-elution interference in traditional DIA and missing value issues in DDA. (B) timsTOF MS and PASEF-based acquisition. The timsTOF platform couples TIMS with a TOF analyzer. The timsTOF enables PASEF technology, which accumulates and separates ions based on ion mobility, maximizing ion utilization and simplifying spectra. When combined with DIA, timsTOF provides high spectral acquisition rates and near-complete ion coverage. DDA, data-dependent acquisition; DIA, data-independent acquisition; LC-MS/MS, liquid chromatography-tandem mass spectrometry; PASEF, parallel accumulation-serial fragmentation; SCP, single-cell proteomics; m/z, mass-to-charge ratio; TIMS, trapped ion mobility spectrometry; TOF, time-of-flight. Figure created using Adobe Illustrator.
LC-MS/MS-based single-cell proteomic technologies
Isobaric labeling strategies, such as tandem mass tags (TMT), offer a solution for multiplexed single-cell quantification. In this format, peptides from different samples are chemically tagged with isotopic reagents that generate distinct reporter ions upon fragmentation. These methods enable multiplexed sample pooling prior to LC-MS analysis and thus increase analysis throughput78. A pioneering application of this approach is the SCoPE-MS workflow. The innovation of this workflow lies in the carrier channel design, in which peptides from hundreds of cells are pooled into one TMT channel to boost peptide detectability across all multiplexed samples, allowing the identification and quantification of ~1,000 proteins per single cell (Figure 3A)8. Building upon this finding, SCoPE2 introduces an automated sample preparation workflow with improved peptide labeling chemistry, optimized data-dependent acquisition (DDA), and enhanced data processing pipelines, thereby increasing proteome depth to 1,500–2,000 proteins per cell73. Prioritized SCoPE (pSCOPE) addressed the stochastic nature of conventional DDA by using a multi-tier prioritization strategy, which consistently targets thousands of informative peptides across single cells, enhancing sensitivity for low-abundance proteins74. However, in these workflows, the carrier proteome may introduce biases that affect both identification and quantification. Specifically, high carrier-to-sample ratios can lead to ion interference, dynamic range compression, and isotopic impurities. In addition, the carrier proteome influences the detection of proteins in single-cell channels, creating a dependence on the composition of the pooled carrier sample79,80. Overall, while isobaric labeling strategies may not match LFQ in terms of proteome coverage, isobaric labeling strategies offer advantages in multiplexing capacity and analytical throughput. For example, the combination of TMTpro 35-plex with the Orbitrap Astral Zoom mass spectrometer enables the analysis of >2,000 single cells per day, providing a solution for large-scale, population-level studies requiring extensive single-cell coverage81.
Protein quantification strategies in LC-MS/MS-based SCP. (A) Isobaric labeling for multiplexed quantification. Individual cells are isolated, lysed, and digested into peptides in SCoPE-MS workflow; peptides from each single cell channel, a 5-cell reference channel, and a 50–200-cell carrier channel are labeled with unique TMT tags. All TMT-labeled samples are pooled and analyzed in a single LC-MS/MS run, in which reporter ion intensities enable relative quantification of proteins across single cells. (B) Non-isobaric labeling for multiplexed quantification. Single-cell samples are processed and labeled with non-isobaric mTRAQ mass tags in the plexDIA workflow (Δ0/Δ4/Δ8). Labeled peptides are pooled and analyzed in a single LC-MS/MS run, in which peptides are distinguished by discrete m/z shifts at MS1 levels. Single-cell digests are labeled with dimethyl tags and combined with a reference channel (uniform cell pool) to enhance signal stability in the mDIA workflow. CID, collision-induced dissociation; HCD, higher-energy collisional dissociation; LC-MS/MS, liquid chromatography-tandem mass spectrometry; mDIA, multiplexed data-independent acquisition; mTRAQ, mass differential tags for relative and absolute quantification; m/z, mass-to-charge ratio; SCP, single-cell proteomics; TMT, tandem mass tags. Figure created using Adobe Illustrator.
Moreover, isotopic labeling strategies based on metabolic incorporation or chemical tagging are integrated with DIA to achieve high-throughput SCP. For example, plexDIA utilizes mTRAQ mass tags (Δ0/Δ4/Δ8) in conjunction with DIA-MS to perform parallel analysis of multiple single-cell samples in a single LC-MS run. Unlike isobaric labeling, the labeled peptides are distinguished by discrete m/z values, allowing sample-specific identification and quantification at the MS1 and MS2 levels (Figure 3B). plexDIA can quantify 1,000 proteins per cell using a 5-min gradient, thus enabling scalable and time-efficient analysis75,82. Similarly, multiplex-DIA (mDIA) used dimethyl chemical labeling, and when combined with Lys-N digestion, supports up to five sample channels. The inclusion of a reference channel enhances signal stability and quantitative accuracy. mDIA achieves a median of 2,377 proteins per single cell with maximum coverage surpassing 4,000 proteins76. Complementary to these chemical labeling approaches, single-cell pulsed stable isotope labeling by amino acids in cell culture (SC-pSILAC) used metabolic labeling with heavy amino acids (e.g., Lys8/Arg10) to quantify steady-state protein abundance and turnover dynamics77. Viable cells are pulse-labeled with heavy isotopes in this strategy, enabling dual-channel quantification of newly synthesized vs. pre-existing proteins. SC-pSILAC has enabled quantification of ~4,000 dual-labeled proteins per single cell, revealing key biological insights, such as cell-cycle states, drug responses, and differentiation trajectories. Together, plexDIA, mDIA, and SC-pSILAC form a complementary set of stable isotope-based strategies that expand the depth, throughput, and biological dimensionality of SCP.
Continuous advances in MS instrumentation, data acquisition modes, and quantification strategies have collectively driven rapid improvements in SCP performance. Proteome depth and analytical throughput have increased dramatically over the past several years (Figure 4A, B). These developments have greatly expanded the applicability of SCP, enabling increasingly comprehensive characterization of cellular heterogeneity in complex biological systems. Taken together, these quantification strategies provide complementary solutions for achieving deep and scalable single-cell proteome profiling. Indeed, the optimal approach should be selected according to the sample type and experimental requirements, such as proteome depth and analytical throughput.
Evolutionary trajectory of LC-MS/MS-based SCP technologies. (A) Evolution of proteome depth in LC-MS/MS-based SCP technologies. The color of the dots indicates the mass spectrometry platform used. The edge color of each dot denotes the data acquisition mode: red, DDA; orange, dia-PASEF; yellow, DIA. (B) Evolution of analytical throughput in LC-MS/MS-based SCP technologies. Dot colors indicate the quantification strategies used. Red represents TMT labeling, green represents LFQ, and yellow represents other labeling-based quantification approaches, including mDIA, plexDIA, and SC-pSILAC. DDA, data-dependent acquisition; DIA, data-independent acquisition; dia-PASEF, data-independent acquisition parallel accumulation-serial fragmentation; FAXP, filter-aided expansion proteomics; LC-MS/MS, liquid chromatography-tandem mass spectrometry; LFQ, label-free quantification; mDIA, multiplexed data-independent acquisition; MS, mass spectrometry; nPOP, nano-ProteOmic sample Preparation; nanoPOTS, nanodroplet processing in one pot for trace samples; OAD Chip, open access droplet chip; PiSPA, picoliter-scale sample preparation for proteomic analysis; pSCoPE, prioritized single-cell proteomics by mass spectrometry; SCoPE, single-cell proteomics by mass spectrometry; SCP, single-cell proteomics; SC-pSILAC, single-cell pulsed stable isotope labeling by amino acids in cell culture; TIMS-TOF, trapped ion mobility spectrometry-time-of-flight; TMT, tandem mass tags. Figure created using Adobe Illustrator.
Profiling PTMs and proteoforms in single cells
Understanding protein function extends beyond measuring protein abundance and hinges on decoding molecular variants, especially the structural and functional diversity introduced by PTMs83. Modifications, such as phosphorylation, acetylation, and glycosylation, regulate protein activity, stability, and subcellular localization, generating immense proteomic complexity from a limited genomic repertoire84–86. While PTM analysis is well-established in bulk proteomics, detecting these modifications at single-cell resolution remains profoundly challenging due to sample scarcity and limited detection sensitivity.
Direct PTM detection without enrichment is becoming increasingly feasible with recent advances in SCP. For example, pasefRiQ, which employs timsTOF-MS combined with a carrier channel to enhance peptide signal, enables systematic identification of diverse PTMs, including phosphorylation, acetylation, methylation, and succinylation in individual cancer cells. This method also achieves high sequence coverage, allowing precise localization of PTM sites on key regulatory proteins87. Mun et al.88 integrated diaPASEF with customized spectral libraries, further improving the detection of multiple PTMs under ultra-low input conditions. The Chip-Tip platform has also enabled phosphoproteome profiling in single HeLa cells11. In addition, pSCoPE apply real-time and multi-tiered precursor selection, enabling deeper and more consistent quantification of biologically relevant, low-abundance PTM peptides without enrichment74.
PTMs, together with genetic variation and alternative splicing, collectively define proteoforms, the true functional units of the proteome. To this end, single-cell proteoform imaging mass spectrometry (scPiMS) couples nanospray desorption electrospray ionization (nano-DESI) with individual ion mass spectrometry (I2MS) to directly image intact proteoforms of abundant proteins within surface-fixed cells without the need for protease digestion or labeling89,90.
AI-powered SCP: data analysis and beyond
AI has become an integral component of SCP, supporting analytical processes from image-based cell delineation to proteome-level inference and functional interpretation. Unlike bulk proteomics or transcriptomics, SCP data are characterized by extreme sparsity, stochastic detection, and heterogeneous noise structures, particularly in MS-based workflows. As a result, the most impactful applications of AI in SCP are not generic transfers of existing models, but purpose-built adaptations that address the physical and statistical constraints unique to single-cell protein measurements. One of the most important applications of AI in SCP lies in the segmentation and identification of individual cells within complex tissue images, especially in spatially resolved workflows. Accurate delineation of individual cells within complex tissues is a prerequisite for downstream proteomic analysis, yet manual annotation is labor-intensive and poorly scalable. AI-powered image analysis tools, such as CellPose, Mesmer, and nucleAIzer, use deep learning algorithms to accurately define cell boundaries and structures, reducing manual workload and improving reproducibility91–93. In addition, AI models can classify cells based on protein expression and spatial context. By integrating local marker expression with cell neighborhood information, SpaGFT and CELESTA can distinguish cell types even in noisy or poorly labeled datasets94,95. Other models, such as CellSighter and MAPS, operate directly on image data to provide automated cell-type predictions with pathologist-level accuracy96,97. These advances are particularly valuable for analyzing the TME, inflammatory conditions, and immune responses across large clinical cohorts.
SCP faces a fundamentally different set of challenges at the proteomic data level once cells are defined and isolated. MS-based single-cell measurements suffer from pervasive missing values caused by low protein copy numbers, ion suppression, and stochastic precursor sampling. Unlike transcriptomic dropout, missingness in SCP is strongly protein- and acquisition-dependent, complicating downstream analysis and limiting the direct applicability of RNA-derived computational frameworks. AI-based approaches have therefore been adapted to explicitly model proteomics-specific noise and uncertainty. Machine learning methods, such as LightGBM, have demonstrated strong performance in imputing missing protein expression, particularly when spatial neighborhood features are incorporated, as shown in large-scale breast cancer tissue studies98. In parallel, deep generative models, such as scVI and total VI originally developed for scRNA-seq, have been adapted to learn batch-corrected latent representations that facilitate cross-experiment integration of protein and multimodal data99,100. While these tools address specific challenges, single-cell proteomic analysis faces broader, intertwined issues. The scPROTEIN framework integrates peptide uncertainty estimation and cell embedding generation to simultaneously tackle quantification uncertainty, data denoising, batch correction, and the production of proteomic-specific embeddings, which demonstrate effectiveness in key tasks, including cell clustering, batch correction, annotation, clinical analysis, and spatial data exploration101. As SCP studies scale toward larger cohorts and clinical applications, reproducibility and standardization have become critical considerations. AI contributes not only through modeling but also by enabling modular and transparent workflow infrastructures. Modular ecosystems, such as SCPline and the tidyomics ecosystem, provide end-to-end solutions for pre-processing, normalization, visualization, and integration of SCP and other omics data, allowing researchers to build transparent and standardized analytical pipelines102,103. These frameworks reflect a shift from ad hoc data processing toward standardized, AI-enabled SCP ecosystems.
Beyond data processing, AI is being increasingly used to integrate proteomics with other modalities, such as transcriptomics and spatial imaging. In many settings, direct feature correspondence between modalities is weak or incomplete, limiting the effectiveness of traditional alignment strategies. AI models, such as MaxFuse and MatchCLOT, overcome this limitation by leveraging graph-based alignment, contrastive learning, and optimal transport to match cells based on structural and relational similarity rather than one-to-one feature overlap104,105. These methods are particularly important for reconstructing spatial organization and identifying cell states that are shaped by molecular and positional cues. AI is also proving useful for detecting spatial microenvironments and cellular niches. By combining molecular features with neighborhood structures, tools like CellCharter and SpaGCN reveal hidden patterns in tissue architecture that may reflect disease progression or immune regulation106,107. Furthermore, AI models, such as PINNACLE and GIANT, can predict protein function and interaction networks by incorporating spatial context, cell-cell communication, and tissue-specific characteristics107–109. These models help uncover functional roles of under-characterized proteins and support the development of new hypotheses for biological research and clinical applications.
Single-cell protein analysis and SCP in elucidating cancer heterogeneity
The diversity of cancer-related biological questions places fundamentally different demands on SCP technologies. While some studies aim to profile thousands of cells to identify robust protein biomarkers or define population-level heterogeneity, other studies focus on resolving deep proteomic states within rare cell subsets to elucidate regulatory mechanisms and therapeutic vulnerabilities. Accordingly, the SCP platforms discussed in the preceding sections are deployed in cancer research with distinct priorities, ranging from high-throughput multiplexed workflows optimized for cohort-scale analysis to depth-oriented, low-input strategies designed to maximize proteome coverage in individual cells.
Single-cell protein analysis facilitates tumor subtyping and biomarker discovery
Single-cell protein analysis techniques, including mass cytometry (e.g., cyTOF), IMC, MIBI, CODEX, and cyclic imaging methods [e.g., multiplexed immunofluorescence (MxIF) and multiplexed immunohistochemistry (MxIHC)] have been widely applied to the study of clinical samples across various cancer types (Table 3). By utilizing antibody panels targeting cancer or immune cells, the heterogeneity within tumors and the immune microenvironments has been characterized. The high-dimensional protein expression data generated by these techniques enable the classification of cell types into more refined subpopulations, yielding insights into the frequencies and functional states within clinical samples. Multiplexed imaging of tissue sections further adds a spatial dimension, capturing the distribution of cell populations across the tissue landscape. Through these single-cell techniques, specific cellular or molecular signatures have been identified that correlate with tumor subtyping, patient stratification, and treatment response or resistance. In addition, some studies have uncovered therapeutic vulnerabilities, providing novel drug targets (Figure 5).
Application workflow of SCP in cancer research. SCP method generates high-dimensional protein expression data from clinical samples, which facilitates cell-type classification, patient stratification and precision medicine. SCP, single-cell proteomics. Figure created using Adobe Illustrator.
Summary of cancer researches utilizing antibody-based SCP method
The molecular classification of tumors based on protein biomarkers has significantly enhanced the clinical treatment outcome compared to the conventional morphology-based classification. For example, patients with breast cancer are now stratified based on the expression of biomarkers, including estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor-2 (HER2). Although such stratifications have improved treatment decision-making, patient responses still vary within each subtype. Tumor and immune cells from different subtypes of breast tumors (luminal A, luminal B, luminal B-HER2+, HER2+, and triple negative) have been classified into distinct clusters using mass cytometry, revealing the heterogeneity in TME among tumors. Wagner et al.110 achieved tumor ecosystem-based tumor classification and characterized 7 distinct tumor groups as well as 36 tumor singletons from 144 tumor samples using the frequencies of epithelial, T cell, and myeloid clusters. The treatment of localized prostate cancer is based on clinicopathologic information, including the Gleason score, prostate-specific antigen levels, stage, and patient age. However, this grading system sometimes failed to characterize metastatic phenotype. De Vargas Roditi et al.111 clustered the single cells from patients with intermediate- or high-grade localized prostate cancer using mass cytometry and characterized 13 patient groups based on the proportion of cellular meta-clusters, providing more precise stratification beyond the current grading system. Non-small-cell lung cancer (NSCLC) is categorized into the following three primary subtypes with distinct treatments and prognoses: lung adenocarcinoma (LUAD); lung squamous cell carcinoma (LUSC); and large-cell carcinoma. Desharnais et al.112 characterized the differences of spatial organization of immune cells from a clinically matched cohort of LUAD and LUSC patients using imaging mass cytometry, elucidating the distinctive immune landscapes and cellular interactions within the two histologic subtypes of lung cancer. In addition, other studies using mass cytometry or IMC have also identified cellular or molecular signatures (e.g., cell subpopulations) that exhibit strong correlations with prognosis and patient outcomes in various cancers, including breast cancer113, lung cancer114–116, liver cancer117,118, pancreatic cancer119, renal cancer120, endometrial cancer121, esophageal cancer122, glioblastoma123, and lymphoma124. The single-cell protein analysis in tumors and the TME may facilitate an understanding of distinct patterns in each tumor subtype and inform the design of precision therapies tailored by the cellular composition or spatial organization.
The cellular or molecular signatures of characterized in the patients showing response to specific therapy (responders) holds the potential as predictive biomarkers, facilitating therapy decision-making in cancer treatment. Using mass cytometry, Tislevoll et al.125 reported that a reduction of extracellular-signal-regulated kinase (ERK) 1/2 and p38 mitogen-activated protein kinase (MAPK) phosphorylation in the myeloid cell compartment 24 h post-chemotherapy could predict 5-year overall survival in a cohort of acute myeloid leukemia patients. In addition, mass cytometry and multiplexed imaging have been applied to explore the predictive biomarkers for the response to immune-checkpoint blockade (ICB) therapies. For example, in patients with stage IV melanoma, the frequency of CD14+ CD16− HLA-DRhi monocytes was shown to be elevated in anti-programmed death 1 (PD-1) immunotherapy responders compared to non-responders and served as a strong predictor of progression-free and overall survival after anti-PD-1 immunotherapy126. In another study involving melanoma, EOMES+ CD69+ CD45RO+ effector memory T cell phenotype was significantly more elevated in responders to combined anti-PD-1 and anti-cytotoxic T lymphocyte-associated antigen-4 (CTLA-4) immunotherapies compared to non-responders127. In head and neck cancer, the decrease of progenitor exhausted CD8+ T cell frequency in uninvolved lymph nodes and the increase of circulating proliferating intermediate-exhausted CD8+ T cell frequency were shown to correlate with responses to anti-PD-L1 therapy128. In triple-negative breast cancer, the fractions of proliferating CD8+ TCF1+ T and MHCII+ cancer cells were predictors of the response to ICB therapies129. Potential predictive biomarkers for response to immunotherapy were also identified using mass cytometry and IMC in other studies involving cancers, including melanoma130–132, lung cancer133–137, lymphoma124, colorectal cancer138, liver cancer117,139–141, esophageal cancer142,143, and breast cancer144. These predictive biomarkers detected by single-cell protein analysis allow for the selection of patients who are more likely to respond and may provide potential alternative therapeutic options for non-responders.
These single-cell protein analysis techniques also facilitated the identification of novel targets for establishing new therapies or overcoming resistance of current therapies. Through mass cytometry and single-cell RNA sequencing, a unique population of CD73hi macrophages was identified in glioblastoma multiforme that persists after anti-PD-1 treatment. The absence of CD73 in a murine model was proven to improve survival, suggesting CD73 as a potential immunotherapeutic target145. In colorectal cancer, a population of CD16+ neutrophils that specifically accumulate in CRC tumor tissues were shown by IMC to have pro-tumor activity by disturbing natural killer (NK) cells. CD16-knockout in a neutrophil cell line reversed the pro-tumor activity of neutrophils and restored NK cell cytotoxicity146. In NSCLC, a highly proliferative, overactivated, and apoptotic dysfunctional CD8+ tumor-infiltrating lymphocytes subset was identified by mass cytometry and the abundance of CD8+ tumor-infiltrating lymphocytes was shown to be associated with resistance to anti-PD-1 therapy147. In HER2-positive breast cancer, fibroblast-induced immune barriers were identified by IMC as the mechanism underlying TCbHP regimen resistance (combined docetaxel, carboplatin, trastuzumab, and pertuzumab). Docetaxel, carboplatin, pyrotinib, and inetetamab (neoPICD) therapy disrupted fibroblast barriers and enhanced of immune cell function, overcoming resistance to anti-HER2 therapies148. In addition, mass cytometry has also been used to identify predictive biomarkers for immune-related adverse events149,150.
In summary, mass cytometry and multiplexed imaging techniques support the advances in precision cancer medicine by guiding rational therapy selection, optimizing personalized treatment regimens, and ultimately improving patient survival and quality of life.
LC-MS/MS-based SCP characterizes the proteome heterogeneity in cancer cell lines and tissue sections
Antibody-based SCP technologies are more widely applied in clinical cancer research owing to high throughput, well-suited for large-cohort investigations focused on dozens of biomarkers. In contrast, LC-MS/MS-based SCP confers superior proteomic depth for unbiased global protein expression profiling yet is currently limited by relatively low throughput. This technical constraint restricts the present application scope primarily to mechanistic investigations for precision cancer medicine (Table 4). For example, LC-MS/MS-based SCP has enabled the exploration of intrinsic proteomic heterogeneity within cancer cell lines162,163 and characterized the heterogeneous responses of cancer cells to drug treatment over time164, underscoring the considerable potential for anti-cancer drug discovery and mechanistic exploration of therapeutic efficacy.
Summary of cancer researches utilizing LC-MS-based SCP method
Végvári et al.151 studied the time-resolved responses of human A549 lung adenocarcinoma single cells to anti-cancer drugs (methotrexate, camptothecin, and tomudex) using the SCoPE-MS workflow. Time-course single-cell proteomic data revealed that drug-treated cancer cells diverged into subpopulations oriented toward survival versus the subpopulations undergoing cell death. Orsburn152 identified two distinct populations in the pancreatic cancer cell lines responding differently to the KRASG12D inhibitor, MRTX1133. Reduction of total KRAS abundance was observed in one population of cells treated with the drug but not in the other population. In addition, SCP was applied to the H358 cancer cell line treated with the histone deacetylase inhibitor, mocetinostat, and identified single cell heterogeneity through the elevated level of histone acetylation after treatment153. Xu et al.154 characterized the interactions between NK and chronic myelogenous leukemia cells using single cell-pair proteomics developed based on the PiSPA workflow.
LC-MS/MS-based SCP has also been used to determine the molecular mechanism underlying drug resistance. Woo et al.155 characterized the single cell proteome of cisplatin-resistant prostate cancer PC3 cells and identified three distinct proteomic sub-clusters within the drug-resistant cell (DRC) population, as well as key proteins related to adaptability, resistance, and metastasis risks. The three proteomic sub-clusters exhibited unique proteomic profiles and functional adaptations to the drug treatment, potentially corresponding to varying degrees of malignancy and resilience against therapeutic interventions. One cluster was enriched in pathways associated with translational and metabolic processes, the second cluster was characterized by the proteins related to stress-response mechanisms and metabolic flexibility, and the last cluster was marked by a highly aggressive and invasive phenotype. The proteomic analysis also revealed several surface proteins and transcription factors in the DRCs that may be related to invasiveness and become potential therapeutical targets. Fan et al.156 incorporated microscopic imaging and microfluid single-cell manipulation system as described in PiSPA to accurately pick up cisplatin-resistant A549 cells and conducted in-depth proteomic analysis of the single cells. Fan et al.156 identified subpopulations of drug-resistant A549 cells characterized by morphology and proteomics pattern. One of the subpopulations maintaining a round geometric shape with no vacuolated morphology was observed and showed high expression of the proliferation marker, Ki67, and upregulation of mitosis-associated proteins, indicating long-term drug-resistance potential. The other subpopulation showed high vacuolization morphology, low Ki67 expression, and downregulation of cell cycle-related proteins, potentially undergoing future cell death. In addition, Yang et al.46 characterized the proteomes of lung cancer NCI-H1975 cells and cells resistant to a third-generation EGFR inhibitor, ASK120067. The SCP analysis revealed that vimentin (VIM) was upregulated in drug-resistant cells, corresponding to activation of the epithelial-mesenchymal transition (EMT) pathway. Yang et al.46 also discovered a significant positive correlation between drug resistance and VIM expression for most EGFR inhibitors in non-small cell lung adenocarcinoma cell lines. Su et al.157 revealed the key proteins contributing to abnormal DNA damage response dynamics and resistance to ionizing radiation in U2OS osteosarcoma cells using microscopy and MS-based SCP. These SCP studies have delivered more accurate protein expression profiles and identified key pathways and proteins underlying drug response or resistance, as well as factors that might be misinterpreted in bulk population analyses.
Mund et al.59 recently identified disease-specific protein signatures in a salivary gland acinic cell carcinoma tissue using the DVP spatial proteomics workflow and profiled spatially resolved proteome of distinct cellular phenotypes in a primary melanoma tissue. A comparison between normal-appearing acinar cells and acinic cell carcinoma in salivary gland carcinoma resulted in the identification of potential protein markers, four of which (CNN1, SRC, CK5, and FASN) were proven to clearly distinguish between normal and cancer tissues by IHC. CD146+ SOX10+ melanoma cells were classified into 5 classes in primary melanoma with a defined spatial distribution: class 1, melanoma in situ; class 2, predominantly tumor; class 3, cells of the TME; class 4, enriched in CD146-high regions; and class 5, enriched in CD146-low regions. The proteomic analysis revealed the pathways that changed in a spatial manner as normal melanocytes transiting to invasive melanoma. Kabatnik et al.158 profiled the proteomes of epithelial colon and stromal cells in colorectal adenomas (CRAs) and identified potential protein markers linked to recurrence. The proteomic comparison between stromal and epithelial cells from high-grade dysplasic regions of the tissue revealed three potential protein markers (DMBT1, MARCKS, and CD99) that were linked to recurrence and stratified the CRA cohort, thus may potentially avoid overtreatment and frequent surgical removals of non-malignant colorectal polyps.
Spatial proteomics also provides valuable information for clinical treatment. For example, Kabatnik et al.159 applied DVP to bladder, prostate, seminal vesicles, and lymph nodes from a patient with signet ring cell carcinoma (SRCC) who responded more effectively to pembrolizumab, a PD-1 inhibitor, than conventional chemotherapies. The proteomic analysis of signet ring cells revealed significant alterations in DNA damage response pathways, including upregulation of the ATR signaling axis, and the enrichment of immune-related protein signatures, consistent with the observed clinical response. In addition, Schweizer et al.63 integrated cell-type resolved spatial proteomics and transcriptomics to reveal the molecular landscape of epithelial and stromal cells during tumor progression from non-invasive serous borderline tumors (SBTs) to metastatic low-grade serous ovarian cancer (LGSC). Representative histologic cases from patients with SBT, micropapillary SBT (SBT-MP), and primary LGSC (LGSC-PT), and the corresponding metastases (LGSC-Met), were analyzed. The single-cell proteome data revealed the upregulation of folate-receptor α (FOLR1) and cyclin-dependent kinases CDK4/6 in LGSC. Combined treatment with the inhibitors significantly reduced tumors in vivo, offering a promising therapeutic strategy for LGSC cases resistant to conventional therapies used for high-grade serous ovarian cancer.
The recently developed multiplexed-imaging-powered DVP (mipDVP) combined 14- to 22-plex tissue imaging and DVP, enabling the identification of 21 cell types and in-depth proteomic analysis. Zheng et al.160 profiled the proteomes of distinct cell populations in human colorectal and tonsil cancer using mipDVP. In a colorectal tumor, as a cold tumor case, mipDVP revealed spatial compartmentalization of immunosuppressive macrophages and distinct functional states of T cells in different tumor compartments. In a tonsil tumor, as a hot tumor case, proteomic analysis uncovered significant tumor cell heterogeneity influenced by proximity to tumor-infiltrating lymphocytes. Upregulation of hypoxia response pathways were detected in CD45R− CD45RO− tumor-infiltrating T cells, indicating an adaptation to hypoxic tumor regions. In addition, Hu et al.161 applied the spatial proteomics method parallel-flow projection and transfer learning across omics data (PLATO) to human breast cancer tissues and profiled the protein expression at the whole-tissue level with a spatial resolution of 25 μm. Using a frozen breast cancer sample from a patient diagnosed with HER2+, ER (70%), and PR− molecular features, two subtypes of proteomes were identified: HER2+, ER−, and PR− (tumor subtype 1); and HER2+, ER+, and PR− (tumor subtype 2). Tumor subtype 1 was enriched in the epithelial-mesenchymal transition, suggesting a more aggressive tumor phenotype, while tumor subtype 2 was enriched in estrogen signaling pathways.
Although single-cell DVP has been developed, researchers in the aforementioned studies typically conducted proteomic analyses on samples with an equivalent of 50–200 cells per replicate to acquire enough proteins for functional analysis and to reduce instrument time. The throughput of LC-MS/MS-based SCP is still a major bottleneck for studies involving large cohorts. Nevertheless, compared to antibody-based strategies, LC-MS/MS-based SCP holds the potential to offer more detailed insights into the molecular mechanisms that underpin precision cancer medicine.
Future perspectives
Since the advent of proteomics back in 1990s, there has been a sustained effort in this field to profile the proteome at the single-cell level. In 2025 Ye et al.11 achieved label-free quantification of >6,500 proteins in a single cell, reaching the protein coverage approaching conventional bulk proteomics. Nevertheless, the application of SCP remains largely confined to cell lines with limited studies on clinical samples. A major limitation is the insufficient analytical throughput, which restricts scalability. Together, current SCP methodologies constitute a flexible experimental toolkit and the choice of platform necessitates carefully weighing the trade-offs between throughput, depth, spatial resolution, and reproducibility. Early discovery and translational studies benefit from scalable, high-throughput approaches, whereas detailed interrogation of tumor evolution, signaling plasticity, and therapeutic resistance relies on deep, high-resolution single-cell proteomic profiling. Recognizing and explicitly aligning these trade-offs with biological questions is essential for the effective application and continued maturation of SCP. In addition, although PTMs are now detectable in single cells without enrichment, the detection depth remains inadequate for comprehensive profiling and advancing PTM studies at single-cell resolution demands significantly more sensitive technologies. Therefore, breakthroughs in detection sensitivity and analytical throughput are essential to expand the biomedical applications of SCP.
Another challenge for SCP is data accuracy, completeness, and reproducibility among experiments. AI-based tools are expected to facilitate signal denoising, missing value imputation, and biological interpretation. The AI virtual cell (AIVC) concept integrates prior knowledge, structural information, and dynamic states into predictive multimodal models for in silico simulation. Domain-specific large language models trained on extensive biological datasets promise to accelerate SCP by supporting cross-omics knowledge transfer, literature-data integration, and end-to-end workflows from protein function prediction to therapeutic design165,166. Future progress will require the convergence of SCP with other omics for comprehensive cell-state modeling and the development of standardized, large-scale SCP databases. Coordinated advances in technology, data resources, and AI modeling will drive SCP toward dynamic multi-scale proteome analysis to transform basic research and clinical translation.
Despite progress in single-cell proteomics, the clinical translation faces persistent practical barriers. Most SCP strategies require cell suspensions and thus often depend on fresh samples. DVP holds considerable promise by addressing two critical clinical bottlenecks: DVP enables standardized experimental workflows and data analysis pipelines to enhance inter-laboratory reproducibility; and DVP is compatible with FFPE tissues, the most abundant form of archived clinical specimens. However, DVP still faces significant challenges with high costs from specialized equipment, such as LCM, high-resolution imaging tools, high-end mass spectrometer, and high demand for instrument time. In addition, while scDVP has been developed, achieving true single-cell resolution remains challenging because attaining deep, reproducible proteome coverage from single cells of FFPE samples poses significant difficulties.
In parallel, the integration of single-cell RNA sequencing (scRNA-seq) with SCP represents a critical frontier for resolving tumor heterogeneity and functional cell states. Over the past decade, scRNA-seq has matured into a scalable technology, enabling transcriptome-wide profiling of hundreds of thousands of cells across tissues, disease states, and developmental contexts. Recent methodologic advances have further expanded its compatibility with fixed and archived samples, facilitating application to clinically relevant specimens167,168. To date, the most prevalent mode of combining scRNA-seq with proteomic readouts relies on a sequential strategy, in which scRNA-seq is first used to identify transcriptionally altered genes or cell populations of interest, followed by antibody-based techniques, such as flow cytometry, multiplex immunofluorescence, or IMC for downstream protein-level validation. While effective for hypothesis testing, this approach measures different cells and cannot resolve RNA-protein discordance within the same cell. Emerging platforms, such as single-cell simultaneous transcriptome and proteome (scSTAP), begin to bridge this gap by enabling joint RNA and protein profiling from individual cells50. However, a fundamental imbalance persists because scRNA-seq has achieved orders-of-magnitude higher throughput and robustness, whereas MS-based SCP remains constrained by analytical sensitivity, acquisition speed, and reproducibility. Consequently, current integrative strategies often prioritize transcriptomic discovery at scale with SCP providing deeper but lower-throughput functional validation. Future progress requires narrowing this throughput gap by developing computational frameworks that integrate sparse proteomic and transcriptomic data and enabling coordinated multi-omic measurements from fixed clinical specimens. As these challenges are addressed, SCP will increasingly complement scRNA-seq by anchoring transcriptional states to protein abundance, post-translational regulation, and pathway activity, ultimately enabling a more complete and mechanistically grounded understanding of tumor ecosystems.
In addition, MS-based methods, emerging single-molecule protein sequencing technologies, including nanopore detection, DNA-PAINT, and fluorescence-based approaches, aim to revolutionize proteomic analysis by directly decoding full-length amino acid sequences along with modifications. These techniques promise new opportunities for high-throughput, single-molecule-resolution proteomic profiling at the cellular scale, potentially circumventing current limitations in detection sensitivity and throughput that constrain conventional SCP169–173.
The current understanding of tumor heterogeneity predominantly relies on single-cell genomic and transcriptomic approaches yet these nucleic acid-centric methods capture only partial biological reality. The integration of SCP will transform cancer research and precision oncology by delivering spatially resolved protein profiles to complete our molecular understanding of tumor ecosystems. This proteome-level resolution may enable the discovery of biologically significant cell states, reveal novel protein biomarkers for diagnosis, treatment response prediction, and prognosis, identify druggable signaling hubs through pathway network mapping, and guide rational combination therapies against co-evolving resistance mechanisms. As SCP converges with other omics, SCP will generate unified functional models of tumor evolution, ultimately accelerating protein-informed patient stratification and shifting precision oncology from correlative genetics to actionable proteome-guided clinical interventions.
Conflict of interest statement
No potential conflicts of interest are disclosed.
Author contributions
Conceived the manuscript: Ruibing Chen, Yi Liu and Yingdong Du.
Wrote the original draft of the manuscript: Yi Liu, Yingdong Du.
Reviewed and edited the manuscript: Ruibing Chen and Ran Su.
- Received August 14, 2025.
- Accepted March 18, 2026.
- Copyright: © 2026, The Authors
This work is licensed under the Creative Commons Attribution-NonCommercial 4.0 International License.
References
- 1.↵
- 2.
- 3.
- 4.↵
- 5.↵
- 6.↵
- 7.↵
- 8.↵
- 9.↵
- 10.↵
- 11.↵
- 12.↵
- 13.↵
- 14.↵
- 15.↵
- 16.↵
- 17.↵
- 18.↵
- 19.↵
- 20.↵
- 21.
- 22.
- 23.
- 24.↵
- 25.↵
- 26.↵
- 27.↵
- 28.
- 29.↵
- 30.↵
- 31.↵
- 32.↵
- 33.↵
- 34.↵
- 35.↵
- 36.↵
- 37.↵
- 38.
- 39.↵
- 40.↵
- 41.↵
- 42.↵
- 43.↵
- 44.↵
- 45.↵
- 46.↵
- 47.↵
- 48.↵
- 49.↵
- 50.↵
- 51.
- 52.
- 53.↵
- 54.↵
- 55.↵
- 56.↵
- 57.↵
- 58.↵
- 59.↵
- 60.↵
- 61.↵
- 62.
- 63.↵
- 64.↵
- 65.↵
- 66.↵
- 67.↵
- 68.↵
- 69.↵
- 70.↵
- 71.↵
- 72.↵
- 73.↵
- 74.↵
- 75.↵
- 76.↵
- 77.↵
- 78.↵
- 79.↵
- 80.↵
- 81.↵
- 82.↵
- 83.↵
- 84.↵
- 85.
- 86.↵
- 87.↵
- 88.↵
- 89.↵
- 90.↵
- 91.↵
- 92.
- 93.↵
- 94.↵
- 95.↵
- 96.↵
- 97.↵
- 98.↵
- 99.↵
- 100.↵
- 101.↵
- 102.↵
- 103.↵
- 104.↵
- 105.↵
- 106.↵
- 107.↵
- 108.
- 109.↵
- 110.↵
- 111.↵
- 112.↵
- 113.↵
- 114.↵
- 115.
- 116.↵
- 117.↵
- 118.↵
- 119.↵
- 120.↵
- 121.↵
- 122.↵
- 123.↵
- 124.↵
- 125.↵
- 126.↵
- 127.↵
- 128.↵
- 129.↵
- 130.↵
- 131.
- 132.↵
- 133.↵
- 134.
- 135.
- 136.
- 137.↵
- 138.↵
- 139.↵
- 140.
- 141.↵
- 142.↵
- 143.↵
- 144.↵
- 145.↵
- 146.↵
- 147.↵
- 148.↵
- 149.↵
- 150.↵
- 151.↵
- 152.↵
- 153.↵
- 154.↵
- 155.↵
- 156.↵
- 157.↵
- 158.↵
- 159.↵
- 160.↵
- 161.↵
- 162.↵
- 163.↵
- 164.↵
- 165.↵
- 166.↵
- 167.↵
- 168.↵
- 169.↵
- 170.
- 171.
- 172.
- 173.↵
In this issue
Jump to section
Related Articles
Cited By...
- No citing articles found.