Global analysis of protein turnover dynamics in single cells

      A human proteome distribution atlas for tissue-specific plasma proteome dynamics

      Precision proteogenomics reveals pan-cancer impact of germline variants

      Proteogenomic characterization of non-functional pancreatic neuroendocrine tumors

      AI-driven predictive biomarker discovery with contrastive learning to improve clinical trial outcomes

      UK Biobank plasma proteomics and genomics reveal asthma risk and heterogeneity

      Deep Visual Proteomics maps proteotoxicity in a genetic liver disease

      A cerebrospinal fluid synaptic protein biomarker for prediction of cognitive function in Alzheimer’s disease

      Microbiome–metabolome dynamics associated with impaired glucose control and responses to lifestyle changes

        Plasma-to-tumour tissue integrated proteomics using nano-omics for biomarker discovery in glioblastoma

        DIA-BERT: pre-trained end-to-end Transformer models for enhanced DIA proteomics data analysis

Although single-cell proteomics (SCP) has made remarkable progress, it remains largely confined to the single dimension of protein abundance. Researchers from the group of Jesper V. Olsen at the University of Copenhagen, together with the group of Zilu Ye at the Institute of Systems Medicine, Chinese Academy of Medical Sciences / Suzhou Institute of Systems Medicine, used pulsed stable isotope labeling by amino acids in cell culture (pSILAC) to achieve, for the first time at the single-cell level, the simultaneous measurement of protein abundance and turnover dynamics (SC-pSILAC). By optimizing the single-cell proteomics workflow, they successfully detected dual SILAC labeling of approximately 4,000 proteins in individual HeLa cells, reproducing known biological processes. The researchers performed a large-scale SC-pSILAC time-series analysis of the spontaneous differentiation of human induced pluripotent stem cells (iPSCs), covering six time points over two months and more than 1,000 single cells. The protein turnover kinetics showed coordinated regulation of protein complexes and core histone turnover during differentiation, clearly separating proliferating from non-proliferating cell populations. At the same time, correlation analysis between cell diameter and the abundance of specific proteins revealed that the expression of histones and certain cell-cycle proteins is not governed by cell volume. The SC-pSILAC technique offers a multi-dimensional perspective on protein dynamics for single-cell biology research.

▼

A human proteome distribution atlas for tissue-specific plasma proteome dynamics

The plasma proteome maintains a dynamic equilibrium through the bidirectional passage of proteins to and from surrounding organs and cells. To quantify the extent to which different organs and cells shape the plasma proteome in health and disease, the research group of Johan Malmström at Lund University in Sweden developed a mass spectrometry-based proteomics strategy to trace the tissue origin of proteins detected in human plasma. The study first built a large-scale human proteome atlas covering 18 vascularized organs and 8 highly abundant blood cell types, and integrated it with existing RNA and protein atlas databases, thereby objectively defining protein–organ associations across the whole proteome and enabling traceable, reproducible quantification of organ-specific proteins in plasma. Validation across six independent patient cohorts (sepsis, pancreatitis, myocardial injury and others) showed that this resource can effectively measure quantitative changes in disease-specific, organ-enriched protein groups. The strategy can be extended to other diseases and provides a new research route for dissecting the mechanisms behind plasma proteome dynamics.

▼

Numerous studies have confirmed the critical role of germline genomics in cancer risk assessment and in the design of personalized treatment strategies. Yet the impact of germline variants on the cancer proteome and on the post-translational modification (PTM) landscape — particularly in regulating oncogenic signaling pathways and in cancer initiation and evolution — remains poorly understood. The group of Li Ding at Washington University in St. Louis, working with several research institutions, systematically dissected the influence of germline variants on the cancer proteome by analyzing 1,064 patient samples spanning 10 cancer types. The team developed a novel “precision peptidomics” approach, mapping 337,469 coding germline variants onto peptide sequences in patient mass spectrometry data, and combined it with protein database resources to reveal the potential effects of these variants on post-translational modifications, protein stability, allele-specific expression and protein structure. The study found that rare pathogenic variants and common germline variants in cancer genes can alter proteomic features, including variants affecting protein abundance and structure, as well as variants in kinases (ERBB2 and MAP2K2) that regulate phosphorylation. Precision peptidomic analysis predicted protein stability-disrupting events in signal regulatory protein alpha (SIRPA) and glial fibrillary acidic protein (GFAP), associated respectively with immune regulation and glioblastoma diagnosis; genome-wide association studies also identified millions of SNPs and quantitative trait loci linked to the expression or levels of thousands of proteins; and polygenic risk scores were significantly associated with distal effects of risk variants. Using high-throughput precision peptidomics, this study elucidates the substantial contribution of germline genetic variation to cancer heterogeneity.

▼

Pancreatic neuroendocrine neoplasms (PanNENs) are the second most common epithelial tumors of the pancreas, and both incidence and prevalence have risen markedly over the past 40 years. Among them, non-functional pancreatic neuroendocrine tumors (NF-PanNETs) are the most common, with a high rate of distant metastasis and limited treatment options. Research teams from Fudan University Shanghai Cancer Center, Peking University Cancer Hospital, the Shanghai Institute of Materia Medica of the Chinese Academy of Sciences, Shanghai Changhai Hospital and the CAS Center for Excellence in Molecular Cell Science collaborated on an in-depth study of NF-PanNETs, performing integrated genomic, transcriptomic, proteomic and phosphoproteomic analysis of 108 NF-PanNETs to achieve comprehensive molecular characterization. The study not only clarified the functional mechanisms of genomic driver alterations in NF-PanNETs, but also revealed the mediators of MEN1 variants through conditional Men1 knockout mouse experiments. Protein biomarkers built with machine-learning models were confirmed as independent prognostic factors and validated in external independent cohorts. Proteomic and phosphoproteomic stratification identified four subtypes with distinct molecular features, immune microenvironments and clinicopathological characteristics. Drug screening using patient-derived tumor organoids identified cyclin-dependent kinase 5 (CDK5) and the calcium voltage-gated channel alpha 1D subunit (CACNA1D) as a broad-spectrum target and a subtype-specific target respectively, findings that were validated in vivo in xenograft models. Through multi-omics integration, this study maps the complete molecular landscape of NF-PanNETs, offering new insights into their biological mechanisms and potential therapeutic targets.

▼

AI-driven predictive biomarker discovery with contrastive learning to improve clinical trial outcomes

Modern clinical trials can capture tens of thousands of clinical genomics measurements for every participant. Compared with prognostic markers, the discovery of predictive biomarkers remains challenging. To address this, a research team at AstraZeneca proposed a neural network framework based on contrastive learning — the Predictive Biomarker Modeling Framework (PBMF) — to explore potential predictive markers in an automated, systematic and unbiased way. Through retrospective analysis of real-world clinical genomics datasets, especially immuno-oncology trial data, PBMF successfully identified a class of biomarkers that predict which patients derive significant benefit from immunotherapy. The results show that this framework can discover predictive and biologically interpretable markers from early clinical data, providing retrospective guidance for phase III trials. Patients carrying this predictive marker showed survival gains of up to 15% compared with the original trial results. PBMF offers a general, efficient and robust strategy for predictive biomarker discovery, bringing actionable insights to personalized treatment and clinical decision-making.

▼

UK Biobank plasma proteomics and genomics reveal asthma risk and heterogeneity

Although hundreds of asthma-associated genetic loci have been identified, translating them into mechanistic understanding remains challenging. By analyzing germline genetic variation, plasma proteomic profiles and phenotypic data from thousands of UK Biobank participants, a Genentech research team resolved asthma-related gene regulation and protein abundance changes, and identified potential drug targets and blood biomarkers. Using genetic causal inference, the researchers identified biomarkers and effector proteins linked to asthma risk and heterogeneity. A total of 609 proteins were associated with asthma status, of which 269 remained after adjustment for body mass index and smoking. Analysis of genetically predicted protein levels identified 70 proteins with potentially causal roles in asthma risk, including known drug targets as well as proteins without prior genetic evidence for asthma (such as GCHFR, TDRKH and CLEC7A). The genetic architecture of these causal proteins provided supporting evidence for a Toll-like receptor (TLR1)–interleukin (IL-27) related asthma pathway. Finally, the study found causal relationships between proteins and features of asthma biological heterogeneity, such as TSPAN8 and neutrophil count. These results show that integrating biobank-scale genetics with plasma proteomics provides a new framework for identifying therapeutic targets and mechanisms underlying disease risk and heterogeneity.

▼

Protein misfolding diseases such as alpha-1 antitrypsin deficiency (AATD) pose a serious threat to human health, yet their cellular pathogenesis remains poorly understood. To dissect this process in depth, the group of Matthias Mann at the Max Planck Institute of Biochemistry, together with the group of Pavel Strnad at University Hospital RWTH Aachen, used mass spectrometry-based spatial proteomics and machine learning to map the molecular landscape of liver tissue from AATD patients. The researchers combined Deep Visual Proteomics (DVP) with single-cell analysis to resolve pseudotemporal molecular events of hepatocyte stress across fibrosis stages in intact patient biopsy tissue. Using as little as one third of the material from a single cell in formalin-fixed paraffin-embedded (FFPE) tissue, they reached a detection depth of up to 4,300 proteins. The dataset revealed peroxisome upregulation occurring before the canonical unfolded protein response, which may hold value for clinical intervention. Single-cell proteomics data showed that the accumulation of alpha-1 antitrypsin is largely cell-autonomous, with minimal stress signaling between hepatocytes. Combining proteomic data with AI-guided image phenotyping across multiple disease stages, the researchers identified a late-stage hepatocyte phenotype characterized by globular protein aggregates and elevated TNFSF10/TRAIL levels, bearing a distinct proteomic signature. This phenotype may represent a critical stage of disease progression. The study not only provides new insights into AATD pathogenesis, but also establishes a powerful methodological framework for high-resolution in situ proteomic analysis of complex tissues, setting a new standard for revealing the molecular mechanisms of protein misfolding diseases in human tissues at the single-cell level.

▼

The rate of cognitive decline varies substantially among patients with Alzheimer’s disease (AD). Although hallmark pathological biomarkers such as amyloid beta (Aβ) and tau have greatly improved pathology-based diagnosis, they explain only 20–40% of the variance in AD-related cognitive impairment (CI). To identify new markers of cognitive impairment, researchers including the group of Tony Wyss-Coray at Stanford University performed cerebrospinal fluid (CSF) proteomic analysis of 3,397 participants from six major prospective AD case-control cohorts. The study found that synapse-related proteins showed the strongest association with cognitive impairment, independent of Aβ and tau. Using machine learning, they derived the CSF YWHAG:NPTX2 synaptic protein ratio. In Aβ-positive and phosphorylated tau-positive (A+T1+) individuals, this ratio explained more of the variance in cognition than CSF pTau181:Aβ42, tau PET, CSF neurofilament, growth-associated protein 43 and neurogranin. CSF YWHAG:NPTX2 increases during normal aging, and in carriers of autosomal dominant AD mutations it changes as early as 20 years before estimated symptom onset. In terms of cognitive prognosis, after 15 years of follow-up and adjustment for multiple factors, CSF YWHAG:NPTX2 predicted progression from A+T1+ cognitively normal to mild cognitive impairment (hazard ratio per standard deviation increase = 3.0, P = 7.0 × 10-4) and from A+T1+ mild cognitive impairment to dementia (hazard ratio per SD increase = 2.2, P = 8.2 × 10-16). The researchers also developed a plasma proteomic signature of cognitive impairment that partially reproduced the CSF YWHAG:NPTX2 findings across 13,401 samples. This study confirms CSF YWHAG:NPTX2 as a strong prognostic biomarker for predicting cognitive resilience and the development of AD, highlights the potential of plasma proteomics as an alternative to CSF testing, and further validates synaptic dysfunction as a core driver of AD dementia.

▼

Type 2 diabetes (T2D) is a complex disease shaped by both genetic and environmental factors, including the gut microbiome. Recent research has revealed substantial heterogeneity in the pathophysiological mechanisms and subtypes of T2D and prediabetes. Researchers are therefore increasingly turning to multi-omics approaches to identify individualized risk factors for T2D and its related complications, such as obesity, acute coronary syndrome, ischemic heart disease and heart failure. To investigate microbial metabolites associated with the phenotypic heterogeneity of prediabetes and T2D, teams from Fudan University and the University of Gothenburg in Sweden performed metabolomic analysis of two Swedish population cohorts (n = 1,167), identifying more than 500 blood metabolites associated with impaired glucose tolerance, about one third of which were linked to altered gut microbiota. The results indicate that metabolic disturbances in microbiome–metabolome dynamics may be potential mediators of impaired glucose homeostasis — for example, the potential interaction between the gut bacteria Hominifimenecus microfluidus and Blautia wexlerae via hippurate production. Short-term lifestyle interventions (such as diet and exercise) can specifically modulate microbiome-associated metabolites. The study points to the microbiome–metabolome axis as a modifiable target in T2D management, with integrated lifestyle interventions offering the prospect of better health outcomes.

▼

Glioblastoma (GB) is one of the most lethal brain cancers, and patient survival has barely improved over the past two decades. Although multi-cancer early detection technologies have advanced, the use of liquid biopsy in brain tumors remains at an early stage, and no clinically approved biomarkers are available for early detection or disease monitoring. Nanoparticle protein coronas hold considerable potential for disease monitoring and biomarker discovery. In this study, nano-omics technology was used to systematically resolve glioblastoma-associated protein changes in plasma and tumor tissue, and the value of biomarker discovery was validated in mouse models and patient cohorts. The integrated nano-omics workflow developed here is the first to systematically link systemic (plasma) and local (tumor tissue) protein changes during GB progression. Mass spectrometry analysis of nanoparticle biomolecular coronas in GL261 tumor-bearing mice at different stages of GB progression showed that the plasma proteome is already altered even at low tumor burden, with more than 30% overlap between the GB-specific plasma and tumor tissue proteomes. Analysis of paired plasma and surgically resected tumor samples from patients with high-grade glioma confirmed the clinical applicability of this nano-omics workflow. Cross-species correlation analysis identified 48 candidate GB biomarkers, involving pathways such as actin cytoskeleton organization, focal adhesion, platelet activation, leukocyte migration, amino acid biosynthesis, carbon metabolism and phagosomes. The nano-omics approach proposed here offers a new avenue for biomarker discovery in the early detection and monitoring of central nervous system diseases, laying the groundwork for subsequent clinical validation.

▼

Data-independent acquisition mass spectrometry (DIA-MS) is playing an increasingly important role in quantitative proteomics. The DIA-BERT software tool developed in this study is the first to apply pre-trained artificial intelligence (AI) models based on the Transformer architecture to DIA proteomics data analysis. The model is trained end to end, eliminating the cumbersome feature extraction and representation learning steps of conventional methods: for each precursor ion, it directly extracts fragment ion peak group information, integrates it with spectral library information such as fragment ion intensities, and feeds these raw data straight into the model. Leveraging the powerful modeling capacity of convolutional neural networks and Transformer networks, the model autonomously learns to evaluate the consistency of peak shapes within a peak group and the correlation between observed peak intensities and spectral library intensities. This integrated training process allows the model to optimize representation learning and scoring model learning simultaneously, significantly improving identification sensitivity. The DIA-BERT identification model was trained on 276 million high-quality peptide precursors extracted from existing DIA-MS files, and the quantification model on 34 million peptide precursors from synthetic DIA-MS files. Tested on five human cancer sample sets (cervical, pancreatic, muscle sarcoma, gallbladder and gastric cancer), it increased the average number of protein identifications by 51% and peptide precursor identifications by 22% compared with the current mainstream tool DIA-NN, while maintaining high quantitative accuracy. This study confirms the great potential of pre-trained models and synthetic datasets for enhancing DIA proteomics analysis.

▼