Yuefan Wang

Target-directed immunotherapies, including TCR-T cell therapies, targeted antibody therapy, and therapeutic vaccines, rely on the identification of tumor-specific antigens (TSAs). However, classical immunopeptidomics is severely constrained by an inability to distinguish passively presented self-ligands of HLAs from functionally recognized by T cells. To resolve these challenges, we developed immunopeptidomics for receptor-associated antigens, a platform capable of sequentially isolating immunopeptidome and their functions with immune cell receptors from a single clinical tissue specimen. Applied to a cohort of 30 renal cell carcinoma (RCC) specimens (18 primary tumors and 12 paired normal adjacent tissues), this approach generated expansive presentation maps encompassing 38,848 unique HLA-I and 26,961 unique HLA-II peptides. Functional receptor captures (IgG- and TCR-bound) achieved up to 5.3-fold higher tumor-specific peptide (TSP) purity ratios compared to direct baseline HLA isolation. Furthermore, cross-fractional co-identification validated recurrent, high-priority unmutated targets (NNMT, TMBIM6, RPL3, VIM, and ENPP3) alongside tumor-specific phosphopeptide hubs and macrophage-restricted N-glycopeptides (CD68). Integrating companion global proteomics uncovered a fundamental regulatory divergence between MHC classes: HLA-I presentation throughput scales linearly (R = 0.883) via a STAT1-orchestrated transcriptional program driving the antigen presentation machinery (APM), whereas HLA-II presentation is uncoupled from parental protein abundance (R = 0.302) due to asynchronous downstream endolysosomal processing and proteolytic over-cleavage kinetics. Together, this platform establishes a micro-input framework that unifies physical presentation with native functional characterization and microenvironmental regulation to accelerate target discovery in low-TMB human cancers.

An LC/MS-based deep multi-omics platform for integrated molecular characterization from a single sample.

Background: Pancreatic cancer is an aggressive malignancy with poor survival. Most patients are diagnosed at advanced or metastatic stages because early disease is often asymptomatic and effective screening tools are lacking. We evaluated a three-marker model comprising serum CA19-9 in combination with the plasma proteins ITIH3 and CEACAM1 for pancreatic ductal adenocarcinoma (PDAC) detection. Methods: Matched plasma and serum samples were collected from 649 participants (250 PDAC cases and 399 controls). Plasma proteins were enriched using high-surface area magnetic covalent organic framework (COF) polymers. Serum CA19-9 was measured using the Tosoh Bioscience immunoassay. The marker panel was trained using a radial-based SVM with repeated 10-fold cross-validation using a set-aside training sample set. The derived model along with a fixed cutoff corresponding to 95% sensitivity in training samples were independently validated using a blinded sample set. Results: In the independent blinded validation, the combined panel of serum CA19-9 with plasma ITIH3 and CEACAM1 achieved an AUC of 0.917 indicating that the three-marker panel maintained strong performance in distinguishing PDAC from controls. At the prefixed threshold, the three-marker panel had a specificity of 53.3% (95% CI: 46.8–59.7%), significantly outperforming CA19-9 alone at 14.5% (95% CI: 10.4–19.7%). Conclusions: In independently blinded validation, combining plasma ITIH3 and CEACAM1 with serum CA19-9 substantially improved diagnostic performance for PDAC, achieving high specificity while maintaining 95% sensitivity compared with serum CA19-9 alone. These findings support further validation of this three-marker panel as a potential PDAC monitoring and detection approach in larger, multicenter studies.

The National Institute of Standards and Technology (NIST) is developing analytical methods to characterize extracellular vesicles (EVs) to support the urgent need for standardized EV reference materials (RMs). This study used orthogonal techniques, cryogenic electron microscopy (Cryo-EM), particle tracking analysis (PTA), asymmetrical flow field-flow fractionation (AF⁴), and microfluidic resistive pulse sensing (MRPS), to evaluate particle size distributions (PSDs) and particle number concentrations (PNCs) of human mesenchymal stem cells (MSCs) and LNCaP prostate cancer cell EVs. Proteomic profiles were assessed by mass spectrometry (MS), and microRNA (miRNA) content of LNCaP EVs was evaluated by small RNA-seq at two independent laboratories. A commercial green fluorescent protein exosome served as a control, except in Cryo-EM, proteomic, and miRNA analyses. Cryo-EM, regarded as the gold standard for morphological resolution, served as PSD reference. PSDs from all methods skewed larger than Cryo-EM, with MRPS closest, AF⁴ most divergent, and PTA intermediate with broader distributions. All techniques reported broad PSDs (30 nm to >350 nm) with PNCs decreasing with increasing particle size, except for AF⁴. Quantitative discrepancies in PNCs reached up to two orders of magnitude across methods and cell sources. MS identified global and EV-specific proteins, including syntenin-1 and tetraspanins CD9, CD63, and CD81. RNA-seq revealed notable inter-laboratory variation. These findings highlight the variability across measurement platforms and emphasize the need for reproducible methods to support NIST’s mission of developing reliable EV reference materials.

In this study, we generated label-free data-independent acquisition (DIA)-based liquid chromatography (LC)-mass spectrometry (MS) proteomics data from 261 renal cell carcinomas (RCC) and 195 normal adjacent tissues (NAT). The RCC tumors included 48 nonclear cell renal cell carcinomas (non-ccRCC) and 213 ccRCC. A total of 219,740 peptides and 11,943 protein groups were identified, with 9,787 protein groups per sample on average. We adopted a comprehensive approach to select representative samples with different mutations, considering histopathological, immune, methylation, and non-negative matrix factorization (NMF)-based subtypes, along with clinical characteristics (gender, grade, and stage) to capture the complexity and diversity of ccRCC tumors. We identified a protein signature containing 55 proteins that distinguish RCC tumors from NATs. Furthermore, a protein signature containing 39 proteins that differentiate different RCC tumor subtypes was also identified. Our findings offer an extensive perspective of the proteomic landscape in RCC, illuminating specific proteins that serve to distinguish RCC tumors from NATs and among various RCC tumor subtypes.

Liquid chromatography (LC) tandem mass spectrometry (MS/MS) is one of the widely used proteomic techniques to study the alterations occurring at the protein level as well as post-translational modifications (PTMs) of proteins that are relevant to different physiological or pathological statuses. The mass spectrometric analysis of peptides digested from proteins (bottom-up proteomics) has emerged as one of the major approaches for proteomics. In this approach, proteins are first cleaved by one or more proteases into peptides for MS analysis, and peptides with PTMs are further enriched, followed by the LC-MS/MS analysis. To achieve a reproducible and quantitative proteomic characterization, a well-established protease digestion and PTM peptide enrichment protocol is critical. In this study, we developed AUTO-SP, a sample preparation platform providing automated protocols for BCA analysis, protein digestion, and PTM enrichment for protein and PTM analyses. We utilized patient-derived xenograft (PDX) breast cancer tumor tissues (basal-like and luminal subtypes) to demonstrate the efficacy of AUTO-SP. The protein amount was quantified, and proteins were further digested by using AUTO-SP for each PDX sample. Based on the data-independent acquisition (DIA)-MS data, we observed that samples of the same breast cancer subtypes were highly correlated (≥0.98). Additionally, >25,000 phosphopeptides and >14,000 ubiquitinated peptides were identified in the PDX samples when using AUTO-SP for PTM enrichment, while unique pathways were enriched from the differentially expressed ubiquitinated peptides of basal-like and luminal subtypes. AUTO-SP demonstrated its efficacy to provide a reliable and reproducible sample preparation procedure for MS-based proteomic and PTM analyses.

Liquid chromatography (LC) and mass spectrometry (MS) are two critical components in proteomics. Advances in methods for both LC and MS have significantly enhanced protein identification and quantifications of limited amounts of proteins, particularly at the picogram-to-nanogram level of proteins. In this study, we explored various LC conditions and MS platforms to optimize protein identification and quantification using data-independent acquisition (DIA). Our investigation focused on evaluating the sensitivity for protein identification, reproducibility of quantification, and robustness across multiple models, specifically focused on analyzing proteins at pico- to nanogram levels, with an emphasis on single-cell proteomics. We further applied our approach for the proteomic analysis of HeLa single cells. Overall, we identified and quantified over 6300 proteins at the single-cell level amount of peptides with a coefficient of variation (CV) of less than 20%, and detected up to 5000 proteins from isolated single HeLa cell samples. Finally, we analyzed docetaxel-treated and nontreated PC3 cells to reveal proteome changes at the single-cell level. This study provides a comprehensive technical evaluation for LC-MS methods in protein identification and quantification for analytical applications involving single-cell proteomics from the picogram to nanogram level of proteins.

Machine learning applications in biomedicine such as omics data analysis are frequently hindered by datasets that are small, high-dimensional, and affected by batch effects across different patient cohorts. To address these challenges, we introduce TabSyM, a modular generative pipeline that synthesizes high-quality, task-relevant data to improve predictive modeling. TabSyM integrates three key stages: it extends a diffusion-based model (TabDDPM) to generate new omics data, employs a novel task-aware sampling mechanism guided by Bayesian optimization to select the most informative synthetic samples, and uses a Multi-Domain Adversarial Network (MDAN) to align data distributions for cross-cohort generalization. We validated our pipeline on a challenging, real-world task of predicting 3-year survival in gastric cancer patients from high-dimensional scRNA-seq data across five cohorts. The full TabSyM pipeline achieved a 30.2% AUROC improvement over the best tree-based models and an 11.5% AUROC gain over leading automated machine learning frameworks. Furthermore, the generative and sampling components are model-agnostic and can substantially boost the performance of classical models like XGBoost independently. These results establish that combining generative modeling with task-aware sampling and domain adaptation provides a robust and effective strategy for overcoming critical data limitations in biomedical tabular data analysis.