Proteomics in bone malignancies: from bulk profiling to single-cell and ultra-low-input technologies Authors: Jianming Chen, Jiaqiao Luo, Ning Ni, Jinhao Zhang, Lei Jiang, Hongbao Wu, Shenghao Wen, Ninghao Hu, Jiaran Li, Gang Deng, Shuxiang Song, Xiaomeng Shi Published: 2026 ABSTRACT AbstractBackgroundBone malignancies, particularly high-grade primary bone sarcomas, remain clinically challenging due to early dissemination, marked heterogeneity, and limited progress in systemic therapies for metastatic or relapsed disease. While genomics and transcriptomics have clarified structural complexity and transcriptional programs, they provide only an indirect view of the functional machinery that ultimately drives invasion, immune escape, and therapy resistance. Proteomics and phosphoproteomics offer a complementary and often non-redundant layer by reporting protein abundance, pathway activity states, and actionable targets at the execution level. However, the strength of proteomics evidence is currently uneven across bone malignancy entities, with osteosarcoma providing the most mature cohort-scale multi-omics context, while emerging cohort-scale proteomics is increasingly available for selected entities such as Ewing sarcoma.Main bodyIn this Review, we synthesize recent advances in mass spectrometry–based proteomics for bone malignancies, using osteosarcoma as the primary exemplar where cohort-scale proteomics / phosphoproteomics-integrated studies are most mature, while selectively incorporating evidence from other bone tumors and skeletal metastasis contexts. To address evidence heterogeneity explicitly, we highlight where conclusions are supported by cohort-scale tumor proteomics/proteogenomics versus where evidence remains exploratory, model-driven, or cross-entity extrapolations). We summarize key results from bulk tissue proteomics, circulating proteomics, extracellular vesicle (EV) and secretome profiling, and highlight how these data have revealed recurrent biological axes including extracellular matrix remodeling and cell–matrix signaling, metabolic rewiring and stress-response programs, and immune/stromal contexture. We further discuss how multi-omics integration refines molecular subtyping, links bulk signatures to functional dependencies, and supports biomarker prioritization with translational intent (e.g., secretome-informed circulating candidates). We additionally provide a practical comparison of proteomics workflows (label-free DDA/DIA, isobaric labeling, phosphoproteomics, top-down, and low-input/single-cell methods) and summarize key barriers to clinical implementation (pre-analytics, standardization, QC governance, and assay/regulatory considerations).ConclusionsFinally, we outline emerging single-cell and ultra-low-input proteomics technologies and propose a staged roadmap for their implementation in bone tumor research, emphasizing feasibility atlases, integrated multi-omics cohorts, and translation into targeted assays and clinically deployable risk models. Collectively, available evidence suggests that proteomics may become an increasingly important pillar for precision medicine in bone malignancies by bridging molecular alterations to actionable functional states and cellular mechanisms. Real-world translation will require fit-for-purpose study designs, harmonized SOPs and multicenter benchmarking, and disciplined down-selection of discovery signatures into validated targeted assays that demonstrably add value beyond existing clinical predictors.Keywords: Osteosarcoma, Bone sarcoma, Proteomics, Phosphoproteomics, Proteogenomics, Biomarkers, Single-cell proteomics, Liquid biopsy, Extracellular vesicles, Data-independent acquisition (DIA) Background Scope, evidence hierarchy, and literature search Scope and evidence hierarchy. Although this Review is framed as “bone malignancies”, the depth of proteomics and phosphoproteomics evidence is currently most mature in osteosarcoma, particularly in cohorts that integrate proteomics/phosphoproteomics with genomics and transcriptomics. To avoid over-generalization, we explicitly distinguish: (i) findings supported by cohort-scale tumor proteomics/proteogenomics in specific entities; (ii) findings supported mainly by small-n clinical series or model systems; and (iii) concepts extrapolated from adjacent contexts such as skeletal metastasis biology. Recent cohort-scale proteomics in Ewing sarcoma further indicates that the evidence landscape is evolving and warrants periodic reassessment. Scope boundary and labeling, as can be seen in Fig. 1. To reduce ambiguity between primary bone sarcomas and skeletal metastasis contexts, we use osteosarcoma-centric evidence as the main anchor for cohort-scale proteomics/phosphoproteomics, and we explicitly label when evidence comes from (i) other primary bone malignancies (e.g., Ewing sarcoma, chondrosarcoma, chordoma) or (ii) metastatic bone disease/literature used mainly for microenvironment and liquid-biopsy concepts. Conceptual framework for proteomics in bone malignancies: how specimen types (tumor tissue, blood/plasma/serum, EV/secretome) map onto proteomics layers (bulk proteome, phosphoproteome, circulating proteomics, EV/secretome profiling, multi-omics integration, and low-input/single-cell) and then onto translational outputs (subtyping, prognostic risk models, biomarker panels, pharmacodynamic monitoring, and trial stratification). Failure modes/constraints: pre-analytical handling, decalcification effects, batch variability, small-cohort overfitting, assay validation challenges (rounded box at bottom) Literature search (narrative review). We performed a narrative literature search focusing on mass spectrometry–based proteomics, phosphoproteomics, proteogenomics, extracellular vesicle/secretome proteomics, and single-cell/ultra-low-input proteomics in bone malignancies. Searches were performed in PubMed and Google Scholar using combinations of the terms “osteosarcoma”, “Ewing sarcoma”, “chondrosarcoma”, “chordoma”, “bone metastasis”, “proteomics”, “phosphoproteomics”, “proteogenomics”, “extracellular vesicles”, “secretome”, “DIA”, and “single-cell proteomics”. Priority was given to (i) cohort-scale clinical tumor proteomics/proteogenomics studies, (ii) studies with orthogonal validation or clear translational endpoints (diagnosis/prognosis/monitoring/target nomination), and (iii) methodological standards and benchmarking studies relevant to harmonization and reproducibility. Clinical and biological burden of bone malignancies Primary malignant bone tumors are uncommon but highly aggressive neoplasms that account for a disproportionate burden of morbidity and mortality in children, adolescents, and young adults. Osteosarcoma is the most frequent primary bone malignancy, followed by Ewing sarcoma and chondrosarcoma, with a bimodal age distribution in which osteosarcoma and Ewing sarcoma peak during adolescence and young adulthood, whereas conventional chondrosarcoma predominantly affects middle-aged and older individuals [1]. In high-income countries, primary bone and joint cancers constitute a small fraction of total cancer incidence. However, they result in substantial years of life lost, as they typically strike otherwise healthy young individuals and entail intensive therapies with significant long-term sequelae [2]. Despite considerable progress in multimodal management since the introduction of multi-agent chemotherapy in the 1970s and 1980s, survival outcomes for many patients with bone malignancies have plateaued. For osteosarcoma, widely cited population-based and cooperative group data indicate 5-year overall survival (OS) rates of approximately 60–70% for patients with localized disease, but only 20–30% for those who present with or subsequently develop metastatic disease, most commonly to the lungs [3, 4]. Recent analyses confirm that, after an initial improvement with modern chemotherapy, long-term survival has remained largely unchanged over the past several decades, and late mortality among childhood osteosarcoma survivors remains substantial [5]. Similar patterns are observed for Ewing sarcoma, in which metastatic or relapsed disease continues to carry a poor prognosis despite intensification of systemic therapy and advances in supportive care [1]. Beyond primary bone sarcomas, metastasis to the skeleton is a frequent and clinically significant complication of common epithelial cancers such as breast, prostate, and lung cancer. Bone metastases are a major cause of pain, pathological fractures, spinal cord compression, and hypercalcemia, and they substantially impair quality of life and functional independence. In breast and prostate cancer, the skeleton is often the first and predominant metastatic site, and the presence of bone metastases is a key determinant of treatment choices and prognosis [6]. When primary and secondary bone tumors are considered together, bone malignancies are best viewed as systemic diseases in which tumor-intrinsic programs intersect with bone-resident stromal biology, immune contexture, and organ-specific metastatic tropism. This review will focus on high-grade primary bone sarcomas (using data-rich osteosarcoma as the primary model) to explore these interactions through the lens of proteomic and proteogenomics evidence. Clinically oriented multi-modal studies have also begun to clarify how immune contexture shapes outcome and may constrain therapeutic responsiveness in osteosarcoma, providing early references for tumor immune microenvironment (TIME)-aware stratification [7]. As an overview, Fig. 2 summarizes the immuno-genomic landscape and TIME heterogeneity in osteosarcoma. Reference atlas of TIME heterogeneity in osteosarcoma used in this Review to motivate microenvironment-aware stratification and to contextualize limited activity of unselected immunotherapy. Reproduced from Wu et al. [7]; see the original article for full panel descriptions Limitations of conventional diagnostic and therapeutic approaches The conventional diagnostic paradigm for bone malignancies relies on a combination of imaging and histopathology. Radiographs, computed tomography (CT), magnetic resonance imaging (MRI), bone scintigraphy, and positron emission tomography (PET) provide complementary information on tumor location, extent, matrix production, and metastatic spread, while biopsy specimens enable definitive histological classification and grading. However, even with state-of-the-art imaging and expert pathology review, a multitude of crucial clinical queries remain challenging to address. First, radiological and histological features often fail to capture early micrometastatic dissemination, minimal residual disease, or subtle biological differences that determine relapse risk. Second, conventional diagnostic categories do not adequately reflect the molecular heterogeneity that is now recognized within ostensibly uniform entities such as high-grade osteosarcoma [8, 9]. Therapeutically, the backbone of treatment for high-grade osteosarcoma and Ewing sarcoma has remained a combination of neoadjuvant chemotherapy, wide surgical resection whenever feasible, and adjuvant chemotherapy, occasionally supplemented by radiotherapy in selected settings. The standard regimens, built around agents such as high-dose methotrexate, doxorubicin, cisplatin, and ifosfamide, were established decades ago and have not been fundamentally transformed since. Attempts to improve outcomes by adding targeted agents or immune checkpoint inhibitors have so far yielded modest and often inconsistent benefits in unselected populations, highlighting the absence of robust predictive biomarkers and biologically informed patient stratification [10]. In metastatic and relapsed disease, salvage options are limited, responses are typically short-lived, and survival remains disappointingly poor. A growing body of work suggests that these therapeutic limitations are intimately related to the biological complexity of bone malignancies. Genomic studies have revealed highly rearranged genomes with widespread chromosomal instability, copy-number alterations, and structural variants; however, these alterations often do not translate into clear, recurrent, and actionable drivers, and they provide limited direct readouts of pathway activation, proteostasis stress, and microenvironment-coupled functional states that ultimately shape therapeutic response [8, 9]. At the same time, the tumor microenvironment—including osteoblasts, osteoclasts, mesenchymal stromal cells, endothelial cells, and diverse immune subsets—plays a key role in disease progression, therapeutic resistance, and organ-specific metastasis, yet is incompletely captured by standard diagnostic approaches. Together, these observations underscore the need for comprehensive, functionally oriented molecular profiling strategies that can disentangle tumor-intrinsic and microenvironmental determinants of outcome. Why proteomics for bone malignancies? In the era of high-throughput sequencing, much attention has been devoted to genomics and transcriptomics as primary tools to dissect cancer biology. While these layers have yielded invaluable insights, they provide only an indirect view of the functional machinery that executes cellular programs. Proteins, rather than nucleic acids, mediate most structural and catalytic functions, encode the activity state of signaling pathways, and constitute the majority of actionable drug targets. Post-translational modifications, protein–protein interactions, and dynamic changes in protein abundance can profoundly influence tumor behavior yet are often poorly predicted from DNA and RNA alone. Accordingly, contemporary integrative frameworks increasingly treat proteomics as the functional layer that links genotype and transcriptional programs to actionable phenotypes. Mass spectrometry (MS)-based proteomics has undergone rapid technological evolution, with advances in instrumentation, acquisition strategies (data-dependent and data-independent acquisition, isobaric labeling), and sample preparation enabling deep and relatively quantitative coverage of the proteome from limited clinical material. Recent osteosarcoma-focused reviews have summarized how MS-based proteomics has been used to nominate candidate biomarkers, map dysregulated signaling networks, and interrogate chemoresistance and metastasis [11, 12]; in contrast, this Review broadens the scope to bone malignancies as a spectrum (high-grade primary bone sarcomas plus selected skeletal metastasis contexts where proteomic or proteogenomics evidence exists) and emphasizes proteomics-as-function across specimen types (tissue, circulation, EV/secretome) and integration modes (proteogenomics and pathway-state inference), with an explicit forward-looking focus on low-input and single-cell proteomics as the enabling step toward clinically feasible profiling under bone-tissue constraints. These studies span tumor tissue, serum and plasma, extracellular vesicles, and tumor-derived secretomes, collectively illustrating how proteomic profiling can capture both cell-intrinsic alterations and tumor–host crosstalk [13]. Earlier work integrating proteomic and genomic data in bone tumors already hinted at the potential of this approach to refine disease classification and highlight biologically coherent subgroups [14, 15]. More recently, multi-omics efforts have explicitly incorporated proteomic and phosphoproteomic data alongside genomic, transcriptomic, and epigenomic layers to delineate molecularly distinct osteosarcoma subtypes with different prognoses and putative therapeutic vulnerabilities [8, 9]. These integrative studies underscore that proteomic information adds non-redundant value beyond DNA and RNA, for example by resolving pathway activation states, capturing metabolic rewiring, and revealing immune and stromal components that are not easily inferred from bulk transcriptomes alone. At the same time, continuing improvements in sensitivity and throughput are progressively lowering the input requirements for MS-based workflows, setting the stage for ultra-low-input and, ultimately, single-cell proteomics. In this Review, we summarize recent advances in proteomics for bone malignancies, using osteosarcoma as the primary exemplar while drawing selective parallels to other bone tumors and skeletal metastasis contexts. We review bulk proteomics across tumor tissue, circulation, and EV/secretome specimens, and then discuss how proteomics and phosphoproteomics contribute to multi-omics subtyping, prognostic modeling, and pathway-state inference. We highlight translational opportunities and the major barriers to deployment, including cohort heterogeneity, specimen constraints in bone (scarcity, decalcification, and matrix effects), and pre-analytical variability in liquid biopsies. Finally, we outline why ultra-low-input and single-cell proteomics are enabling next steps to resolve tumor–immune–stroma heterogeneity and to distill bulk signatures into targeted, clinically deployable assays. Bulk proteomics in bone malignancies A practical comparison of proteomics workflows and clinical suitability Because the bone malignancy literature spans heterogeneous proteomics platforms, we provide a concise comparison of major MS-based approaches in Table 1, focusing on strengths/limitations and translational fit. Table 1Comparison of major MS-based proteomics workflows used in bone malignancies and their translational suitabilityWorkflowTypical strengthsKey limitations / failure modesNear-term clinical use-fitLabel-free DDABroad discovery; simpler sample prep; widely availableMissingness; stochastic IDs; batch sensitivity across centersDiscovery; verification needs targeted assaysLabel-free DIAHigher completeness and reproducibility; better cross-run consistencyLibrary dependence/standardization; computational complexityBest candidate for scalable discovery + longitudinal monitoring cohortsIsobaric labeling (TMT/TMTpro)Multiplexing; reduced run-to-run variance; deep proteome coverageRatio compression; batch design critical; costCohort discovery under controlled design; translation via targeted assaysPhosphoproteomicsPathway-state readout; kinase activity inference; druggable signalingPre-analytical sensitivity; enrichment bias; higher input needsMechanism and target nomination; pharmacodynamic readouts in trialsTop-down proteomicsProteoform/PTM-resolved biology in select size rangesLower throughput; limited depth; demanding instrumentationMechanistic niche applications (selected PTM/proteoform questions)Ultra-low-input / spatial microproteomicsWorks with scarce/FFPE/LCM regions; bridges bulk to cell statesSample losses; batch effects; needs strict QCFeasibility atlases; subtype anchoring; panel down-selectionSingle-cell MS proteomicsCell-state resolution; maps bulk signatures to cellular sourcesMissingness; batch correction risks; limited direct clinical readinessMechanism discovery; distillation into deployable targeted panels Comparison of major MS-based proteomics workflows used in bone malignancies and their translational suitability Tissue proteomics: tumor–normal contrasts and proteome-informed subtyping Early proteomic work in bone malignancies focused on differential protein expression between tumor tissue and matched or unmatched non-malignant counterparts, with the aim of discovering candidate biomarkers and gaining insight into dysregulated pathways. Using two-dimensional gel electrophoresis and mass spectrometry, followed by validation by immunohistochemistry or immunoblotting, these studies identified a range of proteins involved in extracellular matrix (ECM) remodeling, cytoskeletal organization, oxidative stress, and energy metabolism as recurrently altered in osteosarcoma, Ewing sarcoma, and chondrosarcoma [14]. Although technically constrained by limited proteome coverage and semi-quantitative readouts, they established the feasibility of proteomic profiling in bone tumors and highlighted the potential of proteins such as osteopontin, ezrin, and various matrix metalloproteinases as clinically relevant markers. Fueled by advances in high-resolution LC–MS/MS and refined sample preparation methods, tissue-based proteomics has broadened its scope to include shotgun and label-based quantitative strategies, thereby incorporating the capacity to profile thousands of proteins from either FFPE or fresh-frozen specimens. In osteosarcoma, several groups have applied untargeted shotgun proteomics to characterize the tumor proteome and its variation across histological subtypes or clinical subgroups. In a recent comparative study, Alaa and colleagues analyzed FFPE samples from pediatric osteosarcoma representing four histological subtypes and identified a robust proteomic signature that discriminated between subtypes while being largely independent of conventional clinical variables [16]. Enriched pathways included focal adhesion, ECM–receptor interaction, and PI3K–Akt signaling, underscoring the central role of cell–matrix crosstalk and pro-survival pathways in shaping osteosarcoma heterogeneity. These data complement earlier reports of subtype differences in osteosarcoma protein expression and provide a rationale for subtype-specific therapeutic strategies. Phosphoproteomics further enables inference of kinase activity states and signaling rewiring that are often weakly predicted by RNA abundance alone, supporting mechanism-oriented subtype interpretation and target prioritization [17–19]. Indeed, sarcoma phosphoproteomics studies have demonstrated that global tyrosine-phosphorylation profiling can reveal activated kinase landscapes and help nominate candidate driver tyrosine kinases with therapeutic implications [17, 20]. Because histologic response to neoadjuvant chemotherapy is a core clinical endpoint in osteosarcoma, beyond subtype comparisons, tissue proteomics has been employed to investigate associations between protein expression patterns and chemotherapy response. Several studies have stratified osteosarcoma specimens according to histological response (e.g., Huvos grade) and compared protein abundance profiles between good and poor responders. Although the identified protein panels differ across cohorts, recurring themes include differential regulation of proteins linked to DNA damage response, apoptosis, mitochondrial function, and stress signaling [11]. Such analyses typically rely on log2-fold change () and moderated statistical tests to identify significantly deregulated proteins: where and denote the mean normalized intensity for protein i in tumor and control groups, respectively, and \varepsilon is a small constant to avoid division by zero. Downstream pathway enrichment and network analyses then help contextualize the observed changes in terms of biological processes. A complementary line of investigation has explored the low-molecular-weight proteome using top-down mass spectrometry. Yang et al. developed a top-down LC–MS workflow to profile intact proteins in canine osteosarcoma cell phenotypes, focusing on proteins in the 1–30 kDa range and demonstrating that even modest increases in proteoform resolution can uncover molecular differences between cell states that are difficult to capture with peptide-based approaches [21]. Although conducted in a veterinary model, this work illustrates the feasibility of top-down proteomics in bone tumors and suggests that intact-protein analysis may be particularly useful for detecting post-translationally modified proteoforms relevant to signaling and drug response. Tissue proteomics is increasingly interpreted in the light of genomic and transcriptomic alterations. Large-scale multi-omics studies have revealed pervasive chromosomal instability and copy-number changes in osteosarcoma, often accompanied by relatively modest overlap between mRNA and protein-level alterations [8, 9]. In this context, tissue proteomics can refine molecular subtyping by identifying protein-defined clusters that better correlate with clinical outcome than transcript-based groupings alone. By integrating multi-omics data, Jiang et al. defined osteosarcoma subtypes characterized by distinct prognoses and immune microenvironment features, where protein and phosphoprotein data crucially contributed to the subtype classification and thereby helped interpret their specific signaling dependencies [8]. Similarly, Tang and co-workers described proteomic clusters with differing enrichment of immune and stromal signatures, further supporting the notion that proteome-level information adds a non-redundant layer to existing classification schemes [9]. Figure 3 provides a schematic overview of typical tissue-based proteomic workflows in bone malignancies, from tissue acquisition and processing to LC–MS acquisition and bioinformatic analysis. Schematic overview of bulk tissue proteomics workflows in bone malignancies. Representative steps include sample acquisition (surgical resection or biopsy), tissue processing (decalcification if needed, deparaffinization for FFPE, protein extraction), digestion and peptide clean-up, LC–MS/MS acquisition (DDA or data-independent acquisition (DIA)), and downstream statistical and pathway analysis integrating clinical annotations Key bone-specific constraints (e.g., decalcification and matrix effects) and cross-study reproducibility issues are synthesized in section “Current challenges to translation in bone malignancy proteomics”. Beyond osteosarcoma: proteomics evidence in Ewing sarcoma, chondrosarcoma, and chordoma A recurring reviewer concern is that “bone malignancies” reviews can over-index on osteosarcoma because it has the most mature proteomics-integrated multi-omics cohorts. Here we provide concise, evidence-graded snapshots of other major primary bone malignancies. Ewing sarcoma (emerging cohort-scale tumor proteomics) Historically, proteomics in Ewing sarcoma has been smaller and more model-driven; however, recent work is moving toward cohort-scale clinical tumor proteomics. Notably, a deep MS-based proteomic analysis reported profiling of 170 tumor samples from 74 patients spanning primary, relapsed, and metastatic disease, enabling proteome-informed stratification and microenvironment-aware interpretation. This study provides a concrete benchmark that the entity-level evidence map is evolving and supports more balanced cross-entity synthesis [22]. Proteomics can also connect to therapeutic development in Ewing sarcoma through “surfaceome” prioritization. A surface and global proteome analysis identified ENPP1 and other surface proteins as actionable immunotherapeutic targets in Ewing sarcoma models, illustrating a practical route from discovery proteomics to target nomination under a defined translational question [23]. Chondrosarcoma (question-driven proteomics and diagnostic dilemmas) Compared with osteosarcoma, unbiased cohort-scale tumor proteomics in chondrosarcoma is still limited and often organized around specific clinical bottlenecks. A recurring practical dilemma is distinguishing enchondroma from low-grade (grade 1) chondrosarcoma, where treatment strategy differs substantially. Proteomics has been used to nominate immunohistochemistry-ready candidates for this differential diagnosis, providing a concrete example of “proteomics-for-decision” beyond osteosarcoma [24]. Chordoma (cohort-scale proteogenomics as a bone-malignancy benchmark) Chordoma provides a notable non-osteosarcoma exemplar where proteogenomics and phosphoproteomics have already reached cohort scale. Large-cohort integrative studies demonstrate that proteomics-defined subtypes and pathway-activity inference can generate actionable hypotheses not readily recoverable from DNA/RNA alone, serving as a benchmark for what “translationally credible” cohort-scale bone tumor proteogenomics can look like [25]. Implication for this Review Across entities, cross-malignancy generalization is most defensible at the level of pathway programs (ECM/adhesion, stress–metabolism, immune/stromal contexture) rather than entity-specific drivers. Where entity-level proteomics is thin, we explicitly label conclusions as exploratory and highlight the validation steps needed for translation. Circulating and extracellular vesicle proteomics in bone tumors Given the morbidity associated with repeated invasive sampling of bone lesions, non-invasive or minimally invasive biomarkers detectable in blood are highly attractive for diagnosis, risk stratification, and disease monitoring in bone malignancies. Liquid biopsy has emerged as a promising diagnostic and monitoring strategy across oncology by enabling the detection of tumor-derived components in blood, including circulating cell-free DNA (cfDNA), circulating tumor DNA (ctDNA), proteins, and extracellular vesicles. Although cfDNA-based approaches are not yet routinely applied in primary bone malignancies, advances in cfDNA detection technologies—such as droplet digital PCR (ddPCR), next-generation sequencing (NGS) with molecular barcoding, and microarray-based assays—have markedly improved the sensitivity for detecting low-abundance tumor-derived DNA fragments. Notably, the short half-life of circulating cfDNA (approximately 16 minutes to 2.5 hours) allows real-time assessment of tumor burden and treatment response. In a cross-species oncologic context, cfDNA-based monitoring has demonstrated clinical utility in enzootic bovine leukosis, where circulating viral cfDNA reflects tumor clonality and disease progression, highlighting the broader translational potential of liquid biopsy strategies for rare malignancies with limited tissue accessibility [26]. Recent sarcoma-focused liquid-biopsy reviews emphasize that blood-based assays are particularly attractive for rare, heterogeneous sarcomas, but also highlight recurring constraints including limited cohort sizes, platform heterogeneity, and the need for longitudinal validation in clinically defined use-cases [13, 27]. MS-based proteomics of serum, plasma, and other biofluids has therefore emerged as a complementary approach to tissue-based analyses. Wan-Ibrahim et al. reviewed early genomics and proteomics studies that identified candidate circulating protein biomarkers for osteosarcoma, Ewing sarcoma, and chondrosarcoma, including osteopontin, alkaline phosphatase, and several acute-phase proteins [14]. While many of these candidates have yet to be translated into clinically adopted assays, they have shaped the search space for contemporary discovery efforts. More recent studies have taken advantage of high-resolution LC–MS/MS and improved depletion and fractionation strategies to interrogate the circulating proteome in greater depth. Sirikaew et al. summarized a series of serum and plasma proteomics studies in osteosarcoma, showing that multiple independent cohorts converge on alterations in complement and coagulation cascades, ECM components, and metabolic enzymes when comparing patients with healthy controls or benign bone lesions [11]. These studies frequently couple differential expression analysis with receiver operating characteristic (ROC) curve evaluation to assess the diagnostic performance of individual proteins or multi-marker panels in distinguishing malignant from non-malignant conditions. This approach, however, is often limited by small sample sizes and a lack of independent validation. Recent osteosarcoma-focused liquid-biopsy studies have extended beyond bulk plasma proteins to EV-enriched fractions and tumor-conditioned secretomes, aiming to improve tumor specificity and longitudinal monitoring performance [27–30]. A representative advance is the integration of proteomic and transcriptomic profiling to identify circulating biomarkers with robust prognostic value. Lacinski and colleagues combined label-free quantitative proteomics of osteosarcoma cell line secretomes, tandem mass tag (TMT)-based proteomics of FFPE tumor specimens, and bulk RNA sequencing to nominate apo-transcobalamin-II (APO-TCN2) as a candidate biomarker of overall survival in osteosarcoma [31]. Subsequent analysis of serum samples demonstrated that APO-TCN2 levels were associated with patient outcome after adjusting for clinical covariates, suggesting that a subset of proteomically defined biomarkers may be amenable to translation into blood-based assays. In parallel, there is increasing interest in extracellular vesicles (EVs), including exosomes and microvesicles, as carriers of tumor-derived proteins that may be more stable and more reflective of tumor biology than bulk plasma proteins [32]. While comprehensive EV proteomics in bone malignancies is still in its infancy, several studies have profiled EV-enriched fractions from osteosarcoma cell lines or patient plasma, identifying enrichment of proteins linked to cell adhesion, migration, and immune modulation [11, 33]. In principle, EV proteomics might help distinguish active disease from remission, identify minimal residual disease, or provide early indications of metastatic spread, but rigorous longitudinal studies and standardized isolation protocols will be required to realize this potential [34]. Consistent with this promise, a pilot circulating-exosome proteomics study in a spontaneous osteosarcoma clinical model reported that a small panel of exosomal proteins could discriminate osteosarcoma cases from controls, supporting the feasibility of EV-derived protein signatures for disease classification and monitoring [28]. Community standards have been updated to further strengthen EV study design, separation, characterization, and reporting, which is particularly important for cross-cohort comparability in EV proteomics [35]. Beyond classical blood-based biomarkers, mass spectrometry-based “liquid biopsy” approaches that integrate proteomic, metabolomic, and nucleic acid measurements from bodily fluids are being explored across oncology [36, 37]. Although bone tumor–specific applications remain limited, these developments further underscore the importance of circulating proteomics as a key component of non-invasive disease monitoring in the future. To avoid repetition across specimen types, a consolidated discussion of cohort heterogeneity, pre-analytical variables, and validation pitfalls is provided in section “Current challenges to translation in bone malignancy proteomics”. Secretome proteomics and pre-metastatic niche remodeling The tumor secretome—the ensemble of proteins actively released or shed by tumor cells into the extracellular milieu—plays a central role in local and systemic tumor–host interactions. In osteosarcoma, secretome proteomics has been used to characterize factors that promote invasion, angiogenesis, immune evasion, and formation of the pre-metastatic niche. Building on in vitro proteomics of conditioned media from osteosarcoma cell lines, Almeida et al. recently combined secretome profiling, functional assays, and in vivo models to provide supporting evidence that osteosarcoma-derived secretome components actively remodel the lung microenvironment prior to overt metastasis [38]. In this study, the authors applied label-free quantitative LC–MS/MS to conditioned media from osteosarcoma cell lines with differing metastatic potential, identifying a panel of secreted proteins enriched in the highly metastatic phenotype. Functional and in vivo experiments demonstrated that this secretome induced vascular permeability, ECM remodeling, and immune cell recruitment in the lung, thereby creating a pre-metastatic niche that facilitated subsequent metastatic colonization [38]. Among the candidates, EFEMP1 (EGF-containing fibulin-like extracellular matrix protein 1) emerged as a key mediator, with genetic or pharmacological perturbation of EFEMP1 attenuating lung remodeling and reducing metastatic burden. Notably, EFEMP1 and other secretome-derived candidates were detectable in plasma samples and could be correlated with clinical features, bridging the gap between in vitro mechanistic work and potential translational applications. Secretome proteomics also intersects with EV research, as many secreted proteins are packaged into vesicles rather than being released in soluble form. Distinguishing vesicle-associated from soluble secreted proteins, for example by combining size-exclusion chromatography, ultracentrifugation, and density-gradient separation with proteomic analysis, may help refine our understanding of how specific cargoes are trafficked and delivered to recipient cells. Moreover, secretome studies frequently identify proteins of stromal or immune origin, reflecting the bidirectional nature of tumor–microenvironment communication in co-culture or organotypic models. From a systems perspective, secretome datasets can be integrated with intracellular proteomics and transcriptomics to construct multi-layered interaction networks. For example, proteins overrepresented in the secretome of highly metastatic osteosarcoma cells may correspond to ligands whose receptors are enriched in lung endothelial cells, fibroblasts, or resident immune cells, thereby defining candidate ligand–receptor pairs involved in pre-metastatic niche formation. Such integrative analyses may become more mechanistically precise as single-cell transcriptomic (and, ultimately, single-cell proteomic) atlases of target organs mature, enabling ligand–receptor links and downstream response programs to be assigned to specific recipient cell types and states, helping to distinguish early niche-priming events from later bystander changes. Proteomics of drug resistance and therapy response Beyond diagnostic applications, plasma proteomics has also been explored for response prediction; notably, a clinical study reported that baseline plasma proteome patterns could stratify osteosarcoma patients by chemotherapy response [39]. Resistance to systemic therapy remains a major obstacle to curing patients with bone malignancies. In osteosarcoma, standard chemotherapy regimens achieve high rates of radiological and pathological response in some patients, while others exhibit primary resistance or rapidly relapse despite initial responses. Proteomics has been applied to both preclinical models and clinical specimens to dissect mechanisms of resistance and to identify biomarkers predictive of therapy response. At the preclinical level, MS-based proteomics has been used to compare chemotherapy-sensitive and -resistant osteosarcoma cell lines, sometimes generated by stepwise exposure to drugs such as doxorubicin or cisplatin. These studies consistently implicate alterations in redox homeostasis, mitochondrial metabolism, chaperone networks, and DNA repair pathways in acquired resistance [11, 40]. For instance, resistant cells frequently exhibit upregulation of antioxidant enzymes and heat-shock proteins, as well as changes in enzymes involved in glycolysis and the tricarboxylic acid (TCA) cycle, suggesting a rewiring of bioenergetic and stress-response programs. While most of these findings are descriptive, they generate hypotheses about vulnerabilities that could be exploited with combinatorial therapies targeting mitochondrial function or proteostasis. Mechanistically, quantitative phosphoproteomics in osteosarcoma cells has shown that IL-6 stimulation can rewire signaling programs associated with chemoresistance, with pathway analyses highlighting MAPK-related signaling as a prominent axis in this induced resistant state [18]. Clinically, proteomic analyses of diagnostic biopsies or resection specimens stratified by response category have sought to define protein signatures associated with good versus poor histological response. In aggregate, these studies indicate that baseline expression of proteins linked to proliferation, cell cycle control, and immune regulation may influence chemosensitivity [11]. Recent multi-omics analyses have further suggested that mitochondrial function and cuproptosis-related pathways are associated with osteosarcoma prognosis and immune landscape [41]. Although the latter work primarily leveraged transcriptomic data, the implicated pathways are inherently proteomic in nature, and their activity is more directly reflected at the protein and post-translational modification levels. An emerging paradigm is to integrate proteomics with other data types to build risk models or “signatures” that can stratify patients according to predicted response or survival. Multi-omics analyses by Jiang et al. and Tang et al. identified osteosarcoma subtypes characterized by distinct immune infiltration patterns and metabolic profiles, with implications for differential sensitivity to chemotherapy and targeted agents [8, 9]. Proteomic data contributed to the identification of pathway-level differences—such as DNA repair, oxidative phosphorylation, and ECM–receptor interaction—between subtypes. These findings resonate with broader oncology efforts in which proteomics-derived scores or pathway activity metrics are being explored as biomarkers for immunotherapy or targeted therapy responsiveness. Table 2 summarizes representative bulk proteomics studies in bone malignancies and related models, highlighting sample types, analytical strategies, and main contributions. Table 2Representative bulk proteomics studies in bone malignancies and selected clinical/proteogenomics milestones that benchmark cohort scale, clinical proximity, or translational readinessStudySample typeProteomic approachMain contributionBulk proteomics in bone malignancies (discovery and mechanism)Wan-Ibrahim et al. 2015 [14]Mixed bone tumors (tissue, serum)Multiple MS-based workflowsOverview of genomic and proteomic biomarker candidates for osteosarcoma, Ewing sarcoma, and chondrosarcoma; discussion of translational challenges.Sirikaew et al. 2022 [11]Osteosarcoma tissue and biofluidsShotgun LC–MS/MS, label-free and labeledComprehensive review of osteosarcoma proteomics across tissue, serum/plasma, EV, and secretome studies; highlights recurring pathways and candidate markers.Alaa et al. 2023 [16]Pediatric osteosarcoma FFPE tissueShotgun LC–MS/MSProteomic signature discriminating histologic subtypes; enrichment of focal adhesion/ECM and PI3K–Akt signaling [42] programs.Yang et al. 2023 [21]Canine osteosarcoma cell phenotypesTop-down MS (intact proteins)Proteoform-level profiling of low-molecular-weight proteins; illustrates added resolution for PTM/proteoform differences.Almeida et al. 2023 [38]Osteosarcoma cell secretome and mouse lungsLabel-free LC–MS/MS (conditioned media and tissues)Secretome-driven pre-metastatic lung remodeling; EFEMP1 and related mediators nominated as mechanistically grounded candidates.Lacinski et al. 2024 [31]Osteosarcoma secretome, FFPE tissue, and serumLabel-free + TMT LC–MS/MS integrated with RNA-seqAPO-TCN2 prioritized by multi-layer evidence and evaluated in serum as a survival-associated candidate biomarker.Cohort-scale integration and clinical/proteogenomics benchmarksJiang et al. 2022 [8]Osteosarcoma tissue cohortMulti-omics incl. proteomics / phosphoproteomicsMolecular subtypes with distinct prognosis and immune landscape; proteomic layers inform pathway-level dependencies beyond RNA.Tang et al. 2023 [9]Osteosarcoma tissue cohortMulti-omics incl. proteomicsMulti-omics portrait and subtype classification highlighting immune/stromal heterogeneity reflected in proteomic clusters.Wu et al. 2020 [7]Clinical osteosarcoma specimensMulti-modal profiling (incl. protein-level assays)Clinically oriented immuno-genomic stratification clarifying TIME constraints and potential reasons for limited ICI responsiveness in unselected OS.Zhang et al. 2024 [25]187 skull-base chordoma tumoursProteogenomics with phosphoproteomicsLarge-cohort integrative study demonstrating feasibility of proteogenomics at scale in bone malignancies; identifies proteomics-defined subtypes and actionable hypotheses.Burns et al. 2023 [43]321 soft tissue sarcoma tumoursLarge-cohort tumour proteomicsProteome-defined modules transcending histology and independently prognostic for metastasis; practical benchmark for mesenchymal cancer cohort size and validation.MISEV2018 [44]Extracellular vesicle studies (general oncology)Community guideline / reporting standardsMinimal information standards that improve EV study interpretability and cross-cohort reproducibility; directly relevant to EV proteomics in bone tumors.MISEV2023 [35]Extracellular vesicle studiesUpdated community guideline/reporting standardsUpdated minimal information standards to improve EV interpretability, separation/characterization rigor, and reproducibility; relevant to EV proteomics pipelines. Representative bulk proteomics studies in bone malignancies and selected clinical/proteogenomics milestones that benchmark cohort scale, clinical proximity, or translational readiness Taken together, bulk proteomics studies in bone malignancies have progressed from cataloguing differential abundance to generating clinically interpretable hypotheses. Across specimen types, three consistent themes emerge: (i) extracellular matrix remodeling and cell–matrix signaling recur as dominant axes that align with invasion and metastatic competence; (ii) mitochondrial metabolism, redox control, and proteostasis frequently associate with chemotherapy response phenotypes; and (iii) circulating, EV-associated, and secretome-enriched proteins provide a mechanistically anchored route to liquid biopsy candidates, but remain highly sensitive to pre-analytical variability and cohort confounding. These themes motivate a shift from single-marker discovery toward fit-for-purpose panels with targeted verification and prespecified clinical use cases. These bulk-level programs raise the next translational question: how protein-level pathway states integrate with DNA/RNA layers to yield stable molecular subtypes and actionable hypotheses, which we discuss in the following section. Multi-omics integration and molecular subtyping Multi-omics integration in bone malignancies: evidence hierarchy and subtype axes The rapid maturation of high-throughput sequencing has made genomics and transcriptomics routine in cancer research, including in bone malignancies. Here we take a bone-malignancy-wide view of multi-omics integration, while using osteosarcoma as the primary anchor because it currently has the densest body of cohort-level studies that incorporate proteomics/phosphoproteomics and link molecular programs to clinically annotated outcomes. Evidence for other primary bone malignancies is more heterogeneous: Ewing sarcoma is often best covered at the level of fusion-centric transcriptional programs and epigenetic states; chondrosarcoma studies frequently emphasize cartilage/ECM biology and IDH-linked metabolic rewiring; and chordoma provides a notable example where proteogenomics has already reached cohort scale. In metastatic bone disease, multi-omics is rarely organized around a single “bone metastasis” entity in the sarcoma literature, but the systemic niche and monitoring needs make it a useful comparative context for microenvironmental interpretation and liquid-biopsy concepts [1, 6]. To make this evidence hierarchy explicit, Table 3 summarizes where multi-omics depth is currently deepest and what clinical questions each stratum most directly supports. Throughout, we emphasize that cross-entity generalization is most credible at the level of pathway programs and microenvironmental states, whereas entity-specific drivers (e.g., EWSR1 fusions in Ewing sarcoma or cartilage-lineage programs in chondrosarcoma) often require disease-tailored interpretation. Table 3Entity-level evidence map for multi-omics and proteomics in bone malignancies. Osteosarcoma is used as the main exemplar in Section 3 because cohort-level proteomics/phosphoproteomics-integrated studies are most mature in this entity, whereas evidence for other bone malignancies is currently more heterogeneousEntity/contextTypical multi-omics depthProteomics rolePrimary clinical/biological questionsOsteosarcoma (primary focus)Cohort multi-omics incl. proteomics / phosphoproteomicsPathway-state readout; subtype refinement; TIME/stroma anchoringPrognostic stratification; chemotherapy response biology; pathway dependencies; immune/stromal programs [7–9].Other primary bone sarcomas (overview level)Uneven; often smaller or layer-limitedTarget nomination and mechanistic anchoring when availableTransferability of OS-derived integration patterns; entity-specific biology; hypothesis generation under data sparsity [1, 14].Chordoma (bone malignancy exemplar beyond OS)Proteogenomics at cohort scale (selected settings)Subtype definition; actionable hypothesis generationDemonstrates feasibility of cohort-scale proteogenomics in bone malignancies; actionable dependencies [25].Metastatic bone disease (systemic context)Typically not tumor-entity multi-omics centered in sarcoma literatureBiomarker/monitoring concepts; microenvironmental interpretationSkeletal complications, systemic monitoring needs, and niche biology motivating liquid biopsy concepts [6]. Entity-level evidence map for multi-omics and proteomics in bone malignancies. Osteosarcoma is used as the main exemplar in Section 3 because cohort-level proteomics/phosphoproteomics-integrated studies are most mature in this entity, whereas evidence for other bone malignancies is currently more heterogeneous Two osteosarcoma studies illustrate the current integration pattern in the most proteomics-rich setting, and chordoma provides a useful cohort-scale comparative exemplar beyond osteosarcoma [25]. Jiang and colleagues performed a cohort analysis integrating genomic and transcriptomic layers with proteomics / phosphoproteomics to define molecular subtypes with distinct prognosis and immune landscape, highlighting that protein and phosphorylation layers add non-redundant information beyond RNA and enable more direct inference of pathway activation [8]. In a complementary direction, clinically oriented immunogenomic stratification studies clarified how immune contexture constrains outcomes and may shape therapeutic responsiveness in osteosarcoma [7]. These cohort-scale subtype studies also motivate intervention-oriented workflows that connect molecular states to drug-response hypotheses, which we discuss explicitly below as a translational output of multi-omics [45]. Although proteomics-integrated cohort studies are most mature in osteosarcoma, the cross-layer narrative is partially transferable to other primary bone malignancies when framed at the level of pathway programs rather than entity-specific drivers. In Ewing sarcoma, multi-omics efforts are often organized around fusion-linked transcriptional and epigenetic states, yet they repeatedly converge on downstream hallmarks that are inherently proteomic—cell-cycle/DDR control, lineage and differentiation programs, and stress-adaptation circuits. In chondrosarcoma, integration commonly highlights cartilage-lineage ECM remodeling and IDH-associated metabolic rewiring, reinforcing that pathway-state readouts (and ultimately protein-level effectors) provide a shared language for comparative synthesis even when deep proteomics layers are limited. Figure 4 highlights an example of proteomics-integrated multi-omics subtyping linked to prognosis in osteosarcoma. Table 4 summarizes selected proteomics-integrated multi-omics studies in osteosarcoma and their translational outputs. Example of proteomics-integrated multi-omics subtyping in osteosarcoma, included here to illustrate how integrative clusters connect pathway programs (immune activation/suppression, HRD, and MYC-linked programs) to overall survival. Reproduced from Jiang et al. [8]; see the original article for full panel-level details Table 4Proteomics-integrated multi-omics studies in osteosarcoma (selected)StudyCohort sizeLayersKey subtype axes / programsTranslational outputJiang et al. 2022 [8]Reported in studyGenomics + transcriptomics + proteomics / phosphoproteomicsPathway-level subtype programs (e.g., ECM/adhesion, DDR/cell cycle, metabolic states) with immune/stromal differencesPrognostic stratification; pathway activity readouts enabling target/combination hypothesesWu et al. 2020 [7]Reported in studyImmunogenomic / multi-modal profiling (layer mix varies by cohort component)TIME-aware stratification and immune constraints relevant to ICI responsivenessTIME-informed risk/response framing; rationale for combination strategiesZhang et al. 2023 [45]Reported in studyMulti-omics integration + functional screening (as designed)Subtype-conditioned dependencies and drug-response programsCandidate targets and synergistic combinations for follow-up validation Proteomics-integrated multi-omics studies in osteosarcoma (selected) Proteogenomics profiling is also beginning to scale beyond osteosarcoma in selected bone malignancies. A prominent example is a cohort-scale integrative proteomics/phosphoproteomics analysis of 187 skull-base chordomas, which delineated proteomics-defined subtypes and generated actionable pathway-dependency hypotheses [25]. For review purposes, it is useful to distinguish three modes of multi-omics integration. Early integration concatenates features from different layers (e.g., DNA + RNA + protein) into a unified feature matrix, followed by clustering or supervised learning. Intermediate integration applies integrative latent-variable methods (e.g., matrix factorization or multi-view learning) to learn shared representations that capture cross-omics covariation while reducing dimensionality. Late integration performs analyses within each layer and integrates results at the level of pathways, modules, or networks. The osteosarcoma literature has most often relied on late-to-intermediate integration, where pathway enrichment, module analysis, and latent subtype models are combined with clinical correlation [8, 9]. This is arguably appropriate given typical sample sizes, the high dimensionality of omics data, and the need for interpretability in translational contexts. In rare bone malignancies where cohort sizes are often small relative to feature dimensionality, late integration (pathway/module-level synthesis within each omics layer followed by cross-layer triangulation) usually provides the most interpretable and transportable conclusions. Intermediate integration (e.g., latent-factor models such as MOFA+) can be valuable when multiple layers are available for a moderate-n) cohort, because it separates shared versus layer-specific signals while reducing dimensionality. Early (feature-level) integration is most defensible when sample size and external validation capacity are sufficient, because it is most vulnerable to overfitting and to “analysis degrees of freedom” (normalization, missingness handling, batch correction, and feature selection). Across all modes, we therefore emphasize: (i) locked endpoints and covariates, (ii) stability checks (bootstrap/CV), (iii) explicit batch/QC governance, and (iv) external validation whenever feasible, especially before proposing deployable subtype calls or risk-score cutoffs [46–48]. An overview of these integration strategies and their outputs is shown in Fig. 5. Conceptual schematic of multi-omics integration in bone malignancies (osteosarcoma as an exemplar). A representative bone malignancy workflow integrates genomics (mutations/SCNAs/structural variants), transcriptomics (bulk RNA-seq; optionally single-cell RNA-seq for deconvolution and cell-state priors), and proteomics/phosphoproteomics (pathway activity, druggable targets). Integration can proceed via late (pathway/module), intermediate (latent factor), or early (feature-level) strategies, culminating in molecular subtypes and clinically actionable hypotheses for stratified therapy and biomarker development Proteomics-driven subtypes and prognostic signatures To clarify why proteomics adds non-redundant value beyond RNA-based profiling in bone malignancies, Table 5 compares the two approaches across several dimensions directly relevant to biological interpretation and clinical translation. Table 5Proteomics/Phosphoproteomics versus transcriptomics: practical differences for bone malignancy translationDimensionProteomics / phosphoproteomicsTranscriptomicsPathway activationDirectly reports protein abundance and phosphorylation states; supports kinase activity inference and pharmacodynamic monitoring.Often indirect; mRNA does not reliably predict protein activity/PTMs.Post-transcriptional regulationCaptures translation/degradation and PTM effects (execution layer).Does not directly capture translation, degradation, or PTMs.Secretome/EV biologyMeasures exported proteins and EV cargo directly; relevant for liquid biopsy candidates.Requires inference; secreted protein abundance may not follow mRNA.Clinical assay translationNatural bridge to targeted MS (PRM/SRM) and protein panels; many drug targets are proteins.Mature clinical pipelines exist for RNA panels but may miss activity-state biomarkers.Tissue constraints in boneSensitive to pre-analytics, decalcification chemistry/time, and matrix effects; low-input methods mitigate scarcity.Often more tolerant of low input but still affected by tumor purity and batch effects. Proteomics/Phosphoproteomics versus transcriptomics: practical differences for bone malignancy translation A central question in osteosarcoma multi-omics is whether proteomic information merely recapitulates transcriptomic patterns or provides distinct biological and clinical value. Across cancers, global mRNA–protein correlation is typically modest, reflecting regulation at translation, degradation, and post-translational modification levels [47]. In osteosarcoma, this gap is amplified by extensive genomic instability, variable tumor cellularity, and prominent microenvironmental contributions in bulk specimens. Consequently, proteomics may enhance molecular stratification by highlighting pathway activity states and functional effectors that are not reliably inferred from RNA alone. In cohort-level studies, proteomics and phosphoproteomics contribute to subtype definition in two main ways. First, they enable more direct assessment of pathway activation (e.g., PI3K–AKT, MAPK, cell-cycle kinases) using phosphorylation patterns and protein abundance. Second, they can identify proteomic modules that cut across histological categories and align more tightly with clinical outcomes. While analogous work is more mature in soft tissue sarcomas, where large-scale proteomics has revealed proteome-defined modules predictive of distant metastasis, the same conceptual framework is directly relevant to osteosarcoma and other bone sarcomas [43]. These findings reinforce a broader point about multi-omics rather than any single layer: clinically meaningful stratification in sarcomas is often most stable when subtype axes are supported by orthogonal evidence across RNA, protein, and pathway-activity readouts. In practice, cross-layer concordance helps distinguish true biology from cohort- or platform-specific artifacts, while layer-specific discordance can be informative by flagging post-transcriptional regulation, microenvironmental admixture, or state-dependent signaling that RNA-only clustering under-resolves. Sarcoma-scale benchmarking further clarifies what “clinically usable” looks like in this setting: sufficiently sized cohorts, reproducible module/subtype calls, and validation demonstrating incremental prognostic value beyond histology and transcriptomic signatures [43]. A practical output of multi-omics stratification is the derivation of prognostic “signatures” or risk scores. In osteosarcoma, multiple studies have proposed gene expression–based prognostic models linked to mitochondrial function, redox balance, and immune contexture. While many of these signatures are transcriptomics-centered, they frequently implicate processes that are intrinsically proteomic, such as oxidative phosphorylation and regulated cell death programs. For example, Jia et al. constructed a cuproptosis- and mitochondria-linked prognostic framework that illustrates why cross-layer interpretation matters even when the model is trained on transcriptomes: they first used scRNA-seq to resolve osteosarcoma cellular heterogeneity (30 clusters) and then distilled prognostically relevant mitochondrial/cuproptosis-associated genes into a bulk-cohort subtype and risk model [41]. Specifically, 24 prognosis-associated candidates were used for consensus clustering, yielding two molecular subgroups with divergent pathway programs (including enrichment of G2M checkpoint and oxidative phosphorylation in one subtype) [41]. They then derived a compact three-gene Cox signature (MDK, P4HA1, PTN) with an explicit risk equation and strong time-dependent ROC performance in training and validation cohorts (e.g., 1-year/5-year AUC of 0.922/0.862 in training; 0.689/0.792 in testing) [41]. Importantly for “precision” relevance, the same risk score was associated with an interpretable immune/stromal shift: higher risk correlated with lower T cells and multiple effector/innate subsets, lower immune score but higher stromal score, and downregulation of multiple immune-checkpoint genes, alongside TIDE-based differences consistent with altered immunotherapy responsiveness [41]. The authors further linked the signature to predicted chemotherapy sensitivity and even explored drug–target plausibility via molecular docking for P4HA1 with first-line agents [41]. Taken together, this kind of study is most convincing when its “RNA-derived” states are read through a proteomics lens: mitochondrial programs, cuproptosis machinery, ECM/stroma balance, and immune suppression are ultimately executed at the protein and PTM level, making proteomic and phosphoproteomic layers natural anchors for mechanistic validation and for translating signatures into deployable assays. A commonly used formulation for prognostic scoring in clinical oncology is a Cox-model-based linear predictor: where represents the normalized abundance (or activity proxy) of feature k and denotes its fitted coefficient. In a multi-omics setting, may include transcriptomic features (e.g., gene expression), proteomic features (e.g., protein abundance), and phosphoproteomic features (e.g., kinase substrate activity scores). A key methodological challenge is to avoid overfitting when K is large relative to cohort size, a common scenario in osteosarcoma. Hence, robust model development typically requires (i) careful feature preselection guided by biology (e.g., pathway membership), (ii) penalized regression or dimension reduction, and (iii) independent validation on external cohorts. The osteosarcoma multi-omics field is moving in this direction but remains constrained by limited sample availability, motivating data-sharing, harmonized processing pipelines, and prospective multi-center initiatives. In short, for rare entities such as osteosarcoma, the credibility of any multi-omics signature is determined less by the elegance of the model than by disciplined feature control, transparent validation, and transportability across cohorts and platforms. Notably, proteomics-driven stratification is not restricted to tumor-intrinsic biology. Protein markers can report immune cell abundance, activation states, and cytokine network engagement, thereby providing a more proximal readout of the tumor immune microenvironment (TIME) than bulk transcriptomics in some contexts. This is particularly relevant for osteosarcoma, where clinical outcomes appear to be shaped not only by tumor cell–intrinsic programs but also by immune infiltration, stromal remodeling, and osteoclast/osteoblast dynamics within the bone niche [8, 9]. Proteomics can also capture extracellular matrix remodeling and secreted factors that influence metastatic dissemination, offering a bridge from subtype classification to metastatic risk prediction. Integrating proteomics with genomics and transcriptomics for precision medicine The clinical promise of integrating proteomics with genomics and transcriptomics is best understood through a precision-medicine lens: multi-omics integration is valuable insofar as it refines diagnosis, improves risk stratification, or identifies actionable therapeutic vulnerabilities. Proteogenomics, broadly defined as the joint analysis of genomic/transcriptomic alterations with proteomic and phosphoproteomic readouts, has already demonstrated substantial impact in several epithelial cancers by connecting genetic changes to pathway activity and therapeutic response [47, 49]. For osteosarcoma and other bone malignancies, where canonical driver mutations are scarce and pathway dysregulation is often mediated by copy-number changes, structural variants, and epigenetic deregulation, proteogenomics may be particularly informative. Several practical examples illustrate where multi-omics can outperform single-layer approaches. First, genomic alterations do not always translate into pathway activation. A copy-number gain in a receptor tyrosine kinase locus may not produce a corresponding increase in protein abundance or phosphorylation if post-transcriptional regulation or negative feedback dominates. Conversely, a pathway can be hyperactivated through microenvironmental cues or upstream signaling without a clear genomic driver, which may be evident only at the phosphoproteome level. Second, transcript-derived immune deconvolution can be confounded by tumor purity and stromal admixture; proteomic detection of immune markers and cytokine networks can provide independent evidence for immune phenotypes. Third, drug response often depends on protein state rather than gene state: kinase inhibitors, proteasome inhibitors, and metabolic drugs ultimately act on proteins and protein complexes, making proteomic readouts arguably closer to pharmacologic mechanisms. Moving from discovery to clinical utility requires translation of multi-omics knowledge into deployable assays and decision frameworks. At present, routine multi-omics profiling for osteosarcoma patients is not yet feasible in most clinical settings due to cost, tissue constraints, and the need for standardized pipelines. However, several developments suggest a plausible route forward. One is the growing incorporation of proteomic evidence into precision oncology workflows and molecular tumor boards, where proteomics can complement genomic testing by reporting pathway activity and druggable protein targets [50]. Another is the emergence of reference materials and benchmarking strategies designed to improve cross-platform comparability of DNA, RNA, protein, and metabolite measurements, which will be critical if multi-omics models are to be deployed across centers [51]. Finally, osteosarcoma-specific reviews have emphasized that multi-omics is likely to be most impactful when coupled to well-defined clinical questions (e.g., predicting chemotherapy response; identifying patients likely to benefit from immunotherapy combinations) and supported by functional validation in patient-derived models [45, 52]. Despite these opportunities, several limitations remain. Osteosarcoma cohorts are typically small, and sample processing is heterogeneous across studies, limiting reproducibility. Decalcification and variable tissue handling can affect proteome integrity, and tumor cellularity differs widely among biopsies and resections. Data integration itself introduces additional degrees of freedom, and different computational choices can yield different subtype solutions. For these reasons, the field would benefit from community efforts to harmonize sample processing, adopt transparent reporting standards, and prioritize external validation. In this context, multi-omics integration should be viewed less as a one-off classification exercise and more as an iterative process: subtype models should evolve as cohorts grow, technologies improve (particularly phosphoproteomics and low-input proteomics), and clinical annotations become richer. Overall, multi-omics studies in osteosarcoma have begun to converge on a coherent narrative: molecular subtypes are defined by pathway programs involving cell-cycle control, DNA damage response, ECM remodeling, metabolic rewiring, and immune contexture; proteomic layers provide essential functional information; and integrative approaches can generate actionable hypotheses for stratified therapy. These advances set the stage for the next step in the field—moving from bulk multi-omics toward spatially resolved and single-cell proteomic measurements that can directly interrogate tumor heterogeneity and microenvironmental interactions, which we discuss in later sections. Translational outputs of multi-omics: biomarkers and therapeutic hypotheses Multi-omics subtyping becomes clinically meaningful only when it yields deployable outputs. In bone malignancies, the most tractable outputs fall into two categories: (i) biomarker candidates and fit-for-purpose panels for diagnosis, prognosis, and monitoring; and (ii) subtype-conditioned therapeutic hypotheses, including targets and combinations that can be tested in perturbation models. Proteomics has shaped bone tumor biomarker research along a pragmatic continuum: discovery in tumor tissue or model systems; prioritization by orthogonal evidence (e.g., secretome detectability, pathway plausibility, reproducibility across cohorts); verification using targeted measurements; and, ultimately, translation into assays compatible with clinical workflows. In bone malignancies, candidate biomarkers can be grouped into three broad strata. First, tissue-level markers measurable by immunohistochemistry (IHC) or targeted proteomics can support biological stratification and may complement histopathology. Second, circulating biomarkers in serum or plasma offer a minimally invasive route for diagnosis, prognostic risk stratification, and longitudinal monitoring. Third, extracellular vesicle (EV)-associated and secretome-derived candidates are increasingly viewed as mechanistically informed biomarkers, potentially closer to metastatic behavior and tumor–host communication than bulk plasma proteins [11, 14]. Table 6 summarizes proteomics-prioritized candidates with translational relevance, emphasizing evidence type, assay tractability, and current limitations. Table 6Proteomics-prioritized biomarkers/targets in bone malignancies: specimen, validation status, intended use, and incremental value (selected examples)CandidateSpecimenValidation statusIntended useIncremental value and key limitationsAPO-TCN2Serum; tumor FFPE; secretomeDiscovery + multi-layer support; serum evaluation reportedPrognosis / risk stratificationStrength: secretome + tumor proteomics + transcriptomic concordance supports detectability and biological anchoring. Limitation: needs independent cohorts, prespecified cutoffs, and timing-aware control for systemic inflammation and treatment effects [31].EFEMP1Secretome; lung niche; potentially plasma/EVMechanistic functional evidence in models; clinical translation earlyMetastasis risk / monitoring hypothesisStrength: mechanistically linked to pre-metastatic niche remodeling. Limitation: requires longitudinal human cohorts capturing metastasis timing and comparison against clinical predictors (stage, necrosis rate, imaging) [38].EV/secretome panels (adhesion/ECM proteins)EV-enriched plasma fractions; conditioned mediaDiscovery-heavy; validation heterogeneousDiagnosis / monitoringStrength: potentially higher tumor specificity than bulk plasma proteins. Limitation: EV isolation and quantification are major sources of batch effects; requires harmonized SOPs consistent with community guidance and multicenter replication [35].ENPP1 (Ewing sarcoma surfaceome)Tumor models / PDX; surfaceome-enriched fractionsProteomics-based target nomination with supporting datasetsTarget discovery (immunotherapy)Strength: illustrates a clear translational question (“actionable surface targets”) and an assayable protein target class. Limitation: clinical expression heterogeneity and on-target/off-tumor risk need systematic evaluation in patient tissues and normal panels [23]. Proteomics-prioritized biomarkers/targets in bone malignancies: specimen, validation status, intended use, and incremental value (selected examples) From proteomic states to therapeutic hypotheses: what is already closest to trial design? Direct proteomics-stratified interventional trials remain uncommon in primary bone malignancies, but proteomics has begun to contribute in trial-relevant ways by prioritizing actionable targets and “assayable” biomarkers. For example, in Ewing sarcoma, combined surfaceome and global proteome analyses have been used to nominate immunotherapy-relevant cell-surface targets (including ENPP1) as actionable hypotheses, illustrating how proteomics can move beyond descriptive subtyping toward target prioritization with clearer therapeutic intent. In parallel, newer cohort-scale proteomic profiling across primary versus metastatic settings helps define which pathway programs are conserved versus context-specific, which is essential if biomarkers or targets are to generalize across disease stages [22, 23]. Among recent osteosarcoma-focused efforts, apo-transcobalamin-II (APO-TCN2) represents a notable example of how integrated proteomics can nominate a biomarker candidate that remains measurable in clinically accessible specimens. Lacinski et al. integrated (i) label-free proteomics of osteosarcoma secretomes, (ii) TMT-based proteomics of FFPE tumor specimens, and (iii) transcriptomic evidence to prioritize APO-TCN2 as a survival-associated marker, followed by evaluation in patient serum [31]. This multi-stage reasoning is instructive: secretome evidence increases the a priori plausibility of detectability in blood; tissue proteomics grounds the association in tumor biology rather than systemic inflammation alone; and cross-layer concordance (RNA/protein) improves confidence that the signal is not a single-cohort artifact. While further independent validation remains necessary, this study exemplifies a tractable path from discovery to an assay-ready candidate. Secretome-derived markers also highlight how proteomics can bridge mechanistic biology and translational potential. Almeida et al. combined secretome proteomics with functional and in vivo experiments to show that osteosarcoma-derived secretome actively remodels the lung microenvironment before overt metastasis, implicating EFEMP1 among key mediators [38]. Although EFEMP1 is not yet a clinical biomarker with established cutoffs, its appeal lies in biological anchoring: a candidate linked to pre-metastatic niche formation plausibly associates with metastatic risk, and it may be measurable in circulation or EV-enriched fractions. In translational terms, such mechanism-linked candidates can be prioritized for targeted verification (e.g., PRM/SRM or immunoassays) in longitudinal cohorts designed to capture relapse and metastatic progression. Despite these advances, the bulk of reported circulating osteosarcoma biomarkers remain in early phases. A systematic review of liquid-biopsy applications in bone and soft-tissue sarcomas similarly concluded that current evidence is frequently limited by small and heterogeneous cohorts, reinforcing that prospective, use-case-driven validation is essential before clinical deployment [30]. Reviews summarizing serum/plasma proteomics frequently identify recurring pathway themes—complement cascades, coagulation, ECM remodeling, and metabolic shifts—but cohort sizes are often small, and the boundary between tumor-derived signals and confounding systemic responses (inflammation, tissue injury, infection, perioperative changes) is not always clean [11]. For this reason, the most actionable near-term strategy is typically not to chase single-protein winners, but to build fit-for-purpose panels that are verified by robust quantitative assays and evaluated with pre-specified statistical plans in independent cohorts. In practice, the point at which many candidate markers fail is the transition from discovery-grade significance to a fit-for-purpose assay and a demonstrable decision impact in clinically realistic cohorts; the major steps and cross-cutting requirements of this translation pathway are summarized in Fig. 6. A pragmatic translation pathway for proteomics biomarkers in bone malignancies: from discovery to verification, assay development, clinical utility assessment, and prospective trials. The lower row highlights cross-cutting requirements that frequently determine whether a biomarker can move beyond publication-level significance In principle, proteomics is well aligned with therapeutic discovery because most drug targets are proteins and because pathway activity is often better captured at the proteome and phosphoproteome levels than at the transcriptome level. In osteosarcoma, multi-omics studies have repeatedly implicated dysregulation of cell-cycle control, DNA damage response, ECM–receptor interaction, and metabolic rewiring, offering hypotheses for both tumor-intrinsic vulnerabilities and microenvironmental co-dependencies [8, 9]. Proteomics can sharpen these hypotheses by identifying (i) which pathways are actually active (e.g., kinase signaling inferred via phosphorylation), (ii) which proteins are abundant enough to be practical targets, and (iii) whether candidate targets co-vary with resistance phenotypes or immune states. While osteosarcoma currently provides the most mature proteomics-integrated multi-omics examples, translational reasoning should not be confined to a single entity. Across Ewing sarcoma, chondrosarcoma, chordoma, and rarer primary bone tumors, datasets are heterogeneous in cohort size, measured layers, and clinical annotation, with deep proteomics/phosphoproteomics often missing [1, 14]. Under these constraints, the most actionable cross-entity outputs are typically derived at the level of pathway programs and microenvironmental states rather than single recurrent drivers. Indeed, even when proteomics is absent, multi-omics conclusions frequently converge on proteomics-executed hallmarks such as ECM remodeling, adhesion/cytoskeletal programs, stress adaptation, and metabolic rewiring, which can be used as a shared “translation language” across bone malignancies. A practical synthesis strategy is therefore to separate (i) transferable motifs supported by multi-layer evidence in osteosarcoma and at least one other entity (even if via RNA/protein mismatch) from (ii) entity-specific hypotheses that remain provisional. Where unbiased proteomics is not available, triangulation by orthogonal proxies becomes central: emphasizing module-level comparisons, pathway activity scores, or targeted protein panels as intermediate outputs that can later be verified by targeted MS or immunoassays. This approach avoids overfitting small-n cohorts into fragile subtype taxonomies while still yielding deployable outputs (e.g., fit-for-purpose panels, pathway-state readouts, and candidate combination hypotheses) that are robust under data sparsity. Chordoma provides a useful benchmark beyond osteosarcoma: cohort-scale proteogenomics and phosphoproteomics have demonstrated that proteome-defined subtype structure and pathway-activity inference can be feasible at scale in a bone malignancy setting and can generate actionable hypotheses not easily derived from DNA/RNA alone [25]. Together, these considerations justify an osteosarcoma-centered narrative while keeping translational outputs explicitly portable across bone malignancies at the level of shared programs and rigor-aware evidence hierarchies. In practice, osteosarcoma target discovery is constrained by biology: extensive genomic instability and heterogeneous signaling states often yield multiple, context-dependent dependencies rather than a single dominant “driver”. As a result, proteomics-informed therapy is likely to be most impactful when embedded in strategies that explicitly account for heterogeneity, such as (i) subtype-informed therapy selection, (ii) combination regimens designed to target parallel survival pathways, or (iii) adaptive strategies guided by longitudinal biomarkers. Integrated studies combining multi-omics with functional screening illustrate this direction by converting descriptive subtypes into actionable hypotheses—for example, nominating synergistic drug combinations conditioned on subtype-specific pathway activity [45]. From a translational standpoint, the strongest candidates are those supported by convergent evidence: pathway activation at the phosphoproteome level, consistent expression across independent cohorts, and functional dependence in orthogonal models (e.g., patient-derived cells or xenografts). Proteomics also provides a bridge to immuno-oncology and bone niche biology. Osteosarcoma subtypes with distinct immune infiltration patterns suggest that immune contexture influences outcome and, potentially, response to immunomodulatory therapies [8, 9]. Proteomics can contribute by quantifying immune markers, cytokine networks, and ECM remodeling signals that shape immune cell trafficking and activation. In bone malignancies, where osteoclast–osteoblast dynamics and stromal remodeling are integral to disease progression, this capability may be particularly valuable for identifying rational combinations (e.g., pathway inhibition plus immune modulation, or anti-resorptive strategies coupled to anti-tumor therapies). Nonetheless, moving from proteomic associations to treatment decisions requires disciplined validation, ideally through prospective studies where proteomic hypotheses are tested as stratification tools or pharmacodynamic readouts. Looking forward, single-cell and ultra-low-input proteomics should help resolve subtype-conditioned dependencies and therapy-induced resistant niches that bulk profiles blur, thereby tightening the link between pathway-state readouts and actionable target/combination choices. In particular, mapping kinase activity states and stress-response programs to specific tumor and microenvironmental cell states can clarify which vulnerabilities are likely to generalize and which are niche- or treatment-state contingent. However, translating multi-omics subtypes and proteomics-derived biomarker hypotheses into deployable clinical assays is constrained by cohort design, specimen handling, platform variability, and assay-validation requirements; we consolidate these barriers in the next section. Current challenges to translation in bone malignancy proteomics Proteomics translation in oncology often stalls not because discovery studies fail to produce statistically significant associations, but because signals do not survive the handoff to executable assays and decision workflows under real-world variability across sites, timepoints, and care pathways. This translation gap is recognized even in common cancers, yet it is amplified in primary bone malignancies by small and heterogeneous cohorts, specimen constraints (scarce, decalcified, pathology-prioritized material), and strong cellular admixture in bulk samples. Below, we organize recurring barriers into a practical “challenge map” spanning clinical study design, bone-specific specimen chemistry, analytical variability, computational fragility, and last-mile deployment. Key barriers that hinder translation are summarized in Fig. 7. Challenge map for translating proteomics in bone malignancies. Barriers cluster into (i) clinical/cohort design (endpoints, timing, confounding, multicenter accrual), (ii) specimen constraints (scarce biopsies, decalcification, ECM/mineral matrix, tumor purity/spatial mixing), (iii) analytical variability (pre-analytical handling, platform effects, QC governance), (iv) computational/statistical fragility (missingness, batch correction degrees of freedom, overfitting), and (v) deployment constraints (fit-for-purpose assays, turnaround time, cost, regulatory/workflow integration) Clinical and cohort-design challenges: endpoints, timing, and confounding A core barrier is that clinically meaningful endpoints are often difficult to align across studies. Bone malignancy cohorts may report overall survival, event-free survival, relapse, metastasis-free survival, histologic response categories, or composite endpoints, and these are not interchangeable. Moreover, sampling timepoints vary widely (diagnosis, pre-operative, on-therapy, post-operative, surveillance), and many proteins in blood or tissue track treatment effects (inflammation, tissue injury, perioperative state) as strongly as they track tumor biology. Without treatment-aware stratification, time-matched controls, and pre-specified covariates, biomarkers can inadvertently become proxies for care pathways rather than tumor states. These design issues are challenging even in common cancers; in rare entities such as osteosarcoma they are exacerbated by limited accrual, heterogeneous presentation (age, site, stage), and the practical need to pool multi-institutional material. Rigorous measures—blinding, randomized run order, explicit power considerations, and confounder-aware modeling—are still underused in discovery studies but largely determine whether signals replicate [48]. For monitoring-type biomarkers in particular, cross-sectional comparisons are often insufficient; longitudinal designs with prespecified sampling windows and clinically actionable decision points are typically required. Specimen constraints unique to bone tumors: scarcity, decalcification, and matrix effects Bone tumors create specimen-level constraints that directly shape what is measurable and what is reproducible. Diagnostic biopsies may be small and clinically prioritized for pathology, leaving limited material for discovery proteomics. Surgical resections can contain variable proportions of osteoid-rich matrix, hemorrhage, necrosis, and reactive stroma. Decalcification—often unavoidable—can reduce peptide yield and induce protein-specific loss or modification, while mineralized and ECM-dense matrices bias extraction toward abundant structural proteins and can suppress detection of low-abundance signaling mediators and PTMs. Clinical bone-metastasis pathology studies also show that decalcification procedures can alter immunohistochemistry and molecular pathology results, reinforcing that bone-specific pre-processing is a first-order determinant of downstream biomarker reproducibility [53]. Decalcification chemistry and duration can materially alter peptide recovery and induce protein-class–specific loss or modification, introducing systematic bias in downstream differential abundance and PTM analyses [53–55]. Empirically, decalcification chemistry and duration have protocol-specific effects: EDTA-based protocols and short formic-acid decalcification tend to better preserve immunohistochemistry and downstream molecular readouts than strong-acid or prolonged acid decalcification, which can degrade analytes and increase false-negative results [54]. This issue is even more acute for signaling readouts: phosphoproteins can change rapidly after tissue procurement, and practical preservation strategies that stabilize phosphorylation states outside snap-freezing workflows have been proposed to improve clinical feasibility of phosphoproteomics [55]. In bulk tissue, compositional variability introduces a specific failure mode: an apparent prognostic signal may track tumor purity or stromal fraction rather than tumor-intrinsic biology. Translation-ready studies therefore benefit from either (i) quantifying and adjusting for tumor purity and major cell-type fractions (histology-informed estimates or orthogonal marker panels), or (ii) explicitly positioning the biomarker as a microenvironmental readout with a pre-specified clinical use case. Without this discipline, “replication failure” may reflect uncontrolled spatial/cellular mixing rather than absence of biology. Pre-analytical and analytical variability: from tubes to platforms Body-fluid proteomics is highly sensitive to pre-analytical variables: collection tubes, clotting time, delayed centrifugation, hemolysis, freeze–thaw cycles, and storage duration [56]. Importantly, major sources of variability in quantitative body-fluid proteomics can be grouped into pre-analytical and analytical factors, and ignoring them can introduce systematic error that obscures true biological differences [57]. In controlled evaluations of plasma handling, time-to-processing has been identified as a particularly influential variable for measured plasma protein levels, underscoring the need to lock and record processing timelines in multicenter studies [58]. In multi-center studies, minor deviations can introduce batch effects larger than true biological differences, motivating harmonized SOPs, documentation of handling variables, and inclusion of reference materials [37, 59, 60]. Large community efforts such as the HUPO Plasma Proteome Project have similarly emphasized that specimen collection and handling choices can measurably shape plasma proteomic readouts, motivating harmonized protocols as a prerequisite for reliable biomarker discovery and verification [61]. In blood-based proteomics, pre-analytical variables (tube type, clotting time, delayed processing, hemolysis, and freeze–thaw cycles) can introduce variance comparable to or exceeding disease effects, underscoring the need for harmonized SOPs and recorded handling metadata [57, 58, 61–63]. For tissue proteomics, variable ischemic time, fixation conditions, and FFPE crosslinking can differentially affect protein classes and PTMs, complicating cross-cohort comparability. Comprehensive discussions of plasma/serum proteomics further underscore that these preanalytical variables should be measured, recorded, and controlled whenever possible, because they can dominate variance in downstream quantitative proteome analyses [62]. Analytically, platform choices (DDA vs. DIA), library strategies, and peptide/protein inference rules can materially shift results. The key translational point is not that one choice is universally superior, but that cross-center reproducibility requires explicit QC governance: shared reference samples, repeated measures, transparent reporting of instrument settings, and locked data-processing pipelines. Computational and statistical fragility: degrees of freedom under small-n Proteomics results can be highly sensitive to upstream choices (normalization, missing-value handling, batch correction, protein inference) and downstream choices (feature selection, multiple testing, cutpoint optimization, model form). In small-n bone tumor studies, these degrees of freedom can generate apparently “strong” markers that are unstable under resampling or fail external transport. Translation-grade work benefits from locked analysis plans (pre-specified covariates and endpoints), stability checks (bootstrap or cross-validation), and reporting of effect sizes with uncertainty rather than only p-values. Cross-platform reproducibility and multicenter harmonization are therefore not optional add-ons but prerequisites for credibility. Broader clinical proteomics efforts have shown that multicenter benchmarking is feasible and materially improves transportability [64]; bone malignancies, where single centers rarely have sufficient volume, stand to benefit disproportionately from such designs. Assay development and deployment: fit-for-purpose performance and workflow integration Discovery-grade LC–MS/MS is not synonymous with a clinical assay. Deployment requires defined analytical performance (LOD/LOQ, precision, linearity, stability, matrix robustness), interpretable reporting (reference ranges, locked cutoffs), and operational fit (turnaround time, cost, sample volume). Targeted MS (PRM/SRM) offers a practical bridge from discovery to verification and can outperform immunoblotting in quantitative accuracy for well-chosen peptides, but it demands careful peptide selection, stable isotope-labeled standards, and rigorous QC [65, 66]. Immunoassays can scale more easily but hinge on antibody specificity and can be vulnerable to isoform ambiguity and lot-to-lot variability. From an implementation standpoint, translating a discovery signal into a clinical-grade readout usually requires (i) analytically validated assays (precision, LOD/LOQ, linearity, stability, and matrix robustness), (ii) cross-laboratory reproducibility (often multi-center ring trials or shared reference materials), and (iii) governance aligned with local accreditation and, where applicable, in vitro diagnostic pathways. Because these requirements are jurisdiction- and workflow-dependent, we emphasize a “fit-for-purpose” framing: define the intended use (diagnosis vs monitoring vs pharmacodynamics), lock pre-analytics and cutoffs, and then validate the assay within that use-case rather than optimizing retrospectively on convenience cohorts [64, 67]. Even technically solid assays may fail without demonstrated clinical utility. Utility requires incremental value beyond standard predictors (stage, metastasis status, surgical margins, histologic response) and a plausible action linked to results (surveillance intensity, early intervention triggers, trial stratification, pharmacodynamic monitoring). In practice, the “last mile” often demands prospective designs where assay outputs are tied to pre-specified decisions and evaluated against clinically meaningful outcomes. Table 7 summarizes common failure modes in proteomics translation for bone malignancies and practical, fit-for-purpose mitigation strategies discussed in this section. Table 7Translation checklist for proteomics studies in bone malignancies: common failure modes and practical mitigationsBarrier classCommon failure modePractical mitigation (fit-for-purpose)Endpoint & timingMixed endpoints/timepoints; treatment effects confounded with tumor signalPre-specify endpoint and sampling window; treatment-aware stratification; longitudinal designs for monitoring use-casesSpecimen constraintsScarce/decalcified tissue; variable tumor purity; ECM/mineral biasRecord decalcification chemistry/time; purity/cell-fraction adjustment or explicit microenvironmental framing; QC sentinelsPre-analyticsTube/processing variation dominates blood signalsHarmonized SOPs; record handling metadata; hemolysis indicators; reference materialsPlatform & pipelinePlatform-specific artifacts; flexible processing yields irreproducible hitsShared reference samples; locked pipelines; transparent reporting; multicenter benchmarkingStatistics & modelingOverfitting; unstable cutpoints; optimistic AUCsStability checks (bootstrap/CV); effect sizes with uncertainty; external validation; locked cutoffsDeploymentAssay not scalable or not actionableDefine intended clinical workflow; analytical validation plan; decision impact framing; prospective evaluation Translation checklist for proteomics studies in bone malignancies: common failure modes and practical mitigations Single-cell and ultra-low-input proteomics as enabling technologies Taken together, the most persistent bottlenecks in bone malignancies are tightly coupled to what bulk measurements cannot resolve: (i) cellular admixture and spatial mixing (tumor purity; immune/stromal composition), (ii) clinically decisive intra-tumor heterogeneity (rare resistant or metastatic clones), and (iii) state-dependent signaling transitions under therapy that are averaged out in cohort-scale bulk profiles. These limitations are not merely analytical inconveniences; they determine whether a candidate biomarker is mechanistically interpretable, whether it generalizes across variable specimen composition, and whether it can guide decisions under realistic sampling constraints (small biopsies, decalcified material, pathology-prioritized tissue). Figure 8 illustrates how single-cell and ultra-low-input proteomics can bridge bulk associations to mechanism-anchored translation. From bulk associations to mechanism-anchored translation. Bulk proteomics can nominate pathway programs and candidate panels but is confounded by cellular admixture and spatial mixing. Single-cell and ultra-low-input proteomics can deconvolve tumor–immune–stroma programs at the protein/PTM level, map bulk signatures to specific cell states and niches, and reveal therapy-induced resistant states that shape response and relapse As a result, several “hard” problems highlighted above naturally motivate the next methodological step: single-cell and ultra-low-input proteomics, which can (i) deconvolve tumor–immune–stroma programs at the protein and PTM level, (ii) link cohort-scale bulk signatures to specific cell states and microenvironmental niches, and (iii) enable meaningful profiling when tissue is scarce or compromised. In the next section, we therefore discuss the technologies and emerging workflows that make single-cell and ultra-low-input proteomics increasingly practical for bone malignancies, and how they can convert bulk discoveries into clinically deployable, mechanism-anchored assays. Single-cell and ultra-low-input proteomics in bone malignancies: technologies and future directions Overview of emerging single-cell and ultra-low-input proteomics technologies Evidence boundary note. While the enabling technologies for ultra-low-input and single-cell proteomics have advanced rapidly, direct clinical-scale applications in primary bone malignancies remain early and often limited to feasibility or cross-cancer methodological validation. We therefore separate (i) near-term, already practical steps in bone-tissue constraints (ultra-low-input bulk MS, small cell pools, and targeted panels distilled from bulk signatures) from (ii) medium-term single-cell MS aspirations, and we explicitly flag where inferences rely on broader oncology evidence rather than bone-malignancy-specific cohorts. The case for moving beyond bulk proteomics is particularly strong in bone malignancies. High-grade osteosarcoma and related sarcomas are characterized by marked intra-tumoral heterogeneity, complex immune and stromal microenvironments, and a propensity for early dissemination. Bulk proteomics and multi-omics studies have already revealed molecular subtypes, pathway programs, and immune contexture states, but by design they report averaged signals across mixed cell populations [8, 9]. This averaging risks diluting signals from rare but clinically decisive subpopulations, such as drug-tolerant persisters, invasive or metastasis-primed cells, and stem-like compartments [68]. In parallel, bone tumor biopsies are often small, heavily mineralized, and clinically prioritized for diagnostic histopathology, leaving only limited residual material for research. Together, these constraints motivate proteomics workflows that can extract informative protein-level measurements from hundreds, tens, or even single cells. Fig. 9 provides an end-to-end workflow of MS-based single-cell proteomics, which we use as a reference map for the technologies discussed below. End-to-end workflow of MS-based single-cell proteomics. Reproduced (or adapted) from Momenzadeh et al. [69] Current single-cell and ultra-low-input proteomics (ULIP) technologies fall into several broad classes (Table 8). First, carrier-based isobaric labeling strategies, typified by SCoPE-MS and its successor SCoPE2, use tandem mass tag (TMT) labeling to multiplex individual cells (or small cell pools) together with a “carrier” channel containing bulk material; this boosts peptide identifications while preserving quantitative information for single-cell channels [70, 75]. For example, SCoPE2 has been used to quantify on the order of thousands of proteins per cell across large numbers of single immune cells, enabling proteome-level dissection of macrophage/monocyte heterogeneity with matched transcriptomic context [75, 76]. Subsequent refinements, including miniaturized sample preparation, optimized chromatographic gradients, and improved ion optics, have enabled quantification of thousands of proteins per cell in adherent and suspension cell types. More recent data-independent acquisition (DIA) variants, such as plexDIA and related methods, adapt DIA to multiplexed single-cell samples and have demonstrated substantial gains in throughput and proteome coverage at the single-cell level [71]. Methodological refinements such as offset-mass carrier strategies have been proposed to improve quantitative performance in multiplexed single-cell proteomics, helping mitigate ratio compression and enhance proteome-wide quantification across single-cell channels [77]. Table 8Representative classes of single-cell and ultra-low-input proteomics technologies relevant to bone malignanciesTechnology classTypical input scaleKey technical featuresRepresentative implementationsPotential relevance for bone tumorsCarrier-based isobaric single-cell MSSingle cells + carrier; ∼100–1,000 cells per runTMT/TMTpro multiplexing; carrier channel to boost IDs; DDA or DIA readoutSCoPE-MS, SCoPE2, plexDIA-based workflows [69–71]Dissect osteosarcoma intratumoral heterogeneity and drug-tolerant subpopulations in limited biopsies.Microfluidic chip- and droplet-based ULIPTens to hundreds of cells; down to single cellsNano- to picoliter-scale processing; integrated lysis/digestion/cleanup; reduced lossesnanoPOTS, digital microfluidics platforms, droplet-based microreactors coupled to MS [72, 73]Efficient processing of small, decalcified osteosarcoma samples; potential coupling to FACS or microdissection.Affinity-based single-cell proteomics– cells per panel; single-cell readoutAntibody panels, metal-tag or oligo-tag labeling; flow or imaging readoutMass cytometry (CyTOF), imaging mass cytometry, CITE-seq [74]High-dimensional profiling of tumor-infiltrating immune cells, osteoclast/osteoblast lineages, and checkpoint expression.Ultra-low-input bulk MS∼10–1,000 cells; nanogram protein amountsMiniaturized lysis and digestion (SP3/autoSP3); high-sensitivity LC–MS; often DIAULIP workflows on timsTOF/Orbitrap platforms [69]Spatially resolved proteomes from small bone lesions, LCM regions, and pre-metastatic niches in lung and marrow. Representative classes of single-cell and ultra-low-input proteomics technologies relevant to bone malignancies A second class of ULIP technologies exploits microfluidics and miniaturized reactors to minimize sample losses during lysis, digestion, and cleanup [78, 79]. Chip-based platforms such as nanoPOTS and related nanoliter-scale processing workflows confine all sample handling to picoliter-to-nanoliter volumes, thereby increasing digestion efficiency and recovery from very low input [69, 78]. Droplet-based and digital microfluidic systems extend this concept by encapsulating single cells or small cell pools in droplets or on programmable electrode arrays, allowing controlled reagent delivery, incubation, and transfer directly upstream of LC–MS analysis [72, 73]. Recent nanoliter-scale processing approaches further enable paired single-cell proteomics and transcriptomics from the same cells at increased throughput, providing a practical bridge for joint cell-state definition and cross-modal validation in scarce clinical specimens [80]. These platforms are particularly attractive for bone tumor research because they can, in principle, process limited clinical biopsies or fine-needle aspirates with high recovery, and they are compatible with integration of upstream cell-sorting strategies based on surface markers or fluorescence reporters. Recent single-cell nanodroplet processing pipelines further support the practical value of nanoliter-scale sample handling by reducing adsorption loss and enabling end-to-end integration from single-cell preparation to LC–MS analysis [81]. A third set of approaches focuses on high-content functional proteomics of single cells using affinity-based technologies such as mass cytometry (CyTOF), imaging mass cytometry, and CITE-seq, which couples oligonucleotide-barcoded antibodies to single-cell RNA sequencing [74]. Although these methods do not rely on unbiased MS-based peptide identification, they provide highly multiplexed measurements of pre-selected proteins at single-cell resolution and can be tightly integrated with transcriptomic readouts. For bone malignancies, where immune and stromal interactions are critical, such technologies offer a complementary route to interrogate immune checkpoints, cytokine receptors, osteoclast/osteoblast markers, and signaling nodes that are difficult to capture comprehensively by unbiased MS alone. Finally, ultra-low-input bulk proteomics occupies an intermediate position between true single-cell and classical bulk experiments. Here, microgram-to-nanogram amounts of protein from small tissue regions, laser capture microdissected (LCM) areas, or small cell pools are processed using optimized surfactant-assisted lysis and solid-phase-enhanced sample preparation (SP3/autoSP3), coupled to high-sensitivity instruments such as timsTOF or Orbitrap Eclipse [69]. For decalcified osteosarcoma resections or small metastatic deposits in lung and bone marrow, ULIP can already deliver spatially resolved proteomes at near-cellular resolution without the full complexity of single-cell workflows. Taken together, these technology classes offer a palette of options that can be tailored to the practical realities of bone tumor specimens—from small core biopsies to surgically resected lesions and metastatic deposits. The central challenge is not the absence of methods, but rather the deliberate matching of technological capabilities to concrete biological questions, a theme that becomes more explicit when considering potential applications in bone tumor biology. In addition to tissue-based single-cell analyses, liquid biopsy approaches such as cfDNA profiling offer complementary opportunities for real-time, minimally invasive monitoring of tumor dynamics, particularly in bone malignancies where repeated tissue sampling is challenging [26]. Potential applications in bone tumor biology Single-cell and ultra-low-input proteomics provide a natural extension of bulk and multi-omics work in osteosarcoma and related malignancies. Bulk analyses already suggest the existence of molecular subtypes driven by pathway programs involving cell-cycle control, DNA damage response, ECM remodeling, and immune contexture [8, 9]. Single-cell proteomics can, in principle, resolve how these programs are distributed across distinct cellular compartments and whether small, functionally specialized subpopulations dominate clinical behavior. One immediate application is the dissection of intra-tumoral heterogeneity in osteosarcoma. Single-cell proteomics could separate proliferative clusters from invasive, metastasis-primed, or drug-tolerant populations based on pathway activity markers (e.g., phospho-kinase signatures, DNA damage and repair proteins, oxidative stress regulators). Compared with single-cell RNA sequencing, which has already revealed transcriptional heterogeneity in many solid tumors, protein-level measurements report the executed state of signaling pathways and may better capture post-translational regulation that underpins chemoresistance and metastatic competence [69]. For example, a small fraction of cells with heightened PI3K–AKT and focal adhesion signaling—pathways already implicated by bulk proteomics in osteosarcoma [16]—might exhibit disproportionate metastatic potential or resistance to standard MAP regimens. A second, equally compelling area is the characterization of the tumor immune microenvironment (TIME) and broader bone niche. Bone sarcomas evolve within a complex ecosystem that includes T cells, myeloid-derived suppressor cells, macrophages, osteoclasts, osteoblast lineage cells, endothelial cells, and mesenchymal stromal cells. Immunogenomic studies have already linked immune contexture to prognosis and potential immunotherapy responsiveness in osteosarcoma [7, 8]. Single-cell functional proteomics—via CyTOF, CITE-seq, or MS-based scProteomics—could refine this picture by quantifying surface checkpoint molecules, intracellular signaling nodes, cytokine production capacity, and activation/exhaustion markers across immune and stromal subsets. Such data would directly inform combination strategies that unite immune checkpoint blockade or macrophage reprogramming with agents targeting osteoclast/osteoblast dynamics or ECM remodeling. Third, single-cell and ULIP proteomics can deepen understanding of pre-metastatic niche formation, a process that is particularly relevant to osteosarcoma, where lung is the dominant metastatic site. Secretome and bulk proteomics have already implicated osteosarcoma-derived factors such as EFEMP1 in remodeling the lung microenvironment before detectable metastases [38]. Extending this logic, spatially resolved ULIP and single-cell proteomics in lung and bone marrow could reveal which resident cell types—for example, specific endothelial subsets, fibroblast lineages, or immune populations—respond first to tumor-derived secreted proteins and how their proteomic state evolves during niche formation. Combining such measurements with longitudinal sampling in preclinical models would help distinguish early, causally implicated changes from late, bystander alterations. These application scenarios are not mutually exclusive. A single, well-designed study could, for instance, profile osteosarcoma biopsies and matched lung metastases with ULIP or scProteomics, integrate the resulting cell-state map with bulk subtypes and secretome signatures, and link subpopulations to clinical variables such as chemotherapy response or relapse timing. What differentiates single-cell proteomics from existing modalities is not only resolution but also its proximity to functional execution, an essential consideration when the aim is to identify genuine drivers of metastasis, immune evasion, or drug resistance rather than surrogate correlates. Beyond mammalian cell models, single-cell proteomics has also begun to demonstrate cell-type/state-resolved stress-response mapping in complex tissues, supporting the broader feasibility of applying scProteomics to heterogeneous, matrix-rich specimens once sample-processing constraints are addressed [82]. Integration with existing bone tumor omics datasets Bone tumor research already benefits from a growing reservoir of bulk genomic, transcriptomic, and proteomic datasets, alongside emerging single-cell RNA-seq and spatial transcriptomics maps of sarcomas and bone niches. Instead of viewing single-cell and ULIP proteomics as standalone techniques, it is more productive to treat them as additional layers in a multi-omics integration framework that leverages existing data. At the single-cell level, joint analysis of scRNA-seq and single-cell or ultra-low-input proteomics enables more reliable cell-state definition and annotation. Conceptually, a common strategy is to embed cells into a shared low-dimensional space that captures covariation across multiple modalities. Methods such as MOFA+ (multi-omics factor analysis), totalVI, and weighted nearest neighbor (WNN) analysis in Seurat exemplify this family of approaches and have been successfully applied to multi-modal single-cell data in other disease settings [46, 83, 84]. In the context of osteosarcoma, scRNA-seq could provide broad coverage of transcriptional programs and cell types, while scProteomics contributes higher-fidelity measurements of pathway activity and surface marker expression for key subsets (e.g., stem-like tumor cells, exhausted T cells, immunosuppressive macrophages). Integrative latent-factor models would then align these modalities to refine cluster boundaries and annotate intermediate or transitional states that might be obscured in any single modality. At the bulk level, ULIP and single-cell data can be used to deconvolve and interpret signatures derived from large osteosarcoma cohorts. For example, proteomics-defined subtypes characterized by ECM–receptor interaction or oxidative phosphorylation enrichment [8, 9] can be linked back to specific cellular sources by correlating bulk signature scores with the abundance of cell states inferred from scProteomics and scRNA-seq. This “cell-of-origin” mapping helps to clarify whether a poor-prognosis signature reflects, for instance, expansion of a particular stromal cell type, emergence of a drug-tolerant tumor subclone, or a coordinated reprogramming of multiple compartments. Conceptually, this is analogous to combining bulk transcriptomic signatures with cell-type reference atlases, but with the added advantage that key pathway activities are measured directly at the protein level. Integration frameworks can also be extended to coupling bulk secretome and EV proteomics with single-cell datasets. Ligand–receptor analyses based on scRNA-seq already seek to infer cell–cell communication networks; adding secretome and EV cargo profiles provides direct evidence of which ligands are actually produced and exported by tumor cells and stromal components [11, 38]. In osteosarcoma, combining tumor-cell scProteomics with EV and secretome data would facilitate construction of interaction networks that connect metastasis-primed tumor subclones to responsive endothelial, fibroblast, or immune subpopulations in distant organs. Technical and practical challenges specific to bone malignancies Applying single-cell and ultra-low-input proteomics in bone malignancies raises a set of challenges that are partly shared with other solid tumors but are amplified by the physical and biological peculiarities of bone tissue. These can be grouped into specimen-related, technology-related, and clinical/practical considerations. On the specimen side, the mineralized and fibrous nature of bone poses obvious barriers to single-cell dissociation. Decalcification procedures, particularly those involving strong acids, can compromise protein integrity and epitope preservation; yet decalcification is often unavoidable in routine pathology workflows. Enzymatic and mechanical dissociation optimized for bone marrow and trabecular bone may not directly translate to dense cortical or tumor-infiltrated bone, and cell viability can rapidly deteriorate during prolonged processing. Moreover, tumor cells are intermingled with osteoblasts, osteoclasts, stromal cells, immune infiltrates, and vascular elements, making it challenging to enrich specific compartments without bias. For retrospective studies that rely on FFPE material, spatially resolved ULIP based on LCM may circumvent some dissociation issues, but at the cost of losing viable single cells for functional assays. From a technological perspective, all single-cell and ULIP proteomics workflows involve trade-offs between sensitivity, throughput, and quantification accuracy. Carrier-based TMT strategies can suffer from ratio compression and limited dynamic range for very low-abundance proteins; DIA-based single-cell methods demand sophisticated data analysis and robust spectral libraries; and microfluidic platforms require meticulous control of surface chemistry to avoid adsorption losses [69, 72]. Automation of sample handling is improving but remains less mature than in single-cell RNA sequencing, increasing susceptibility to batch effects and operator variability. Data analysis pipelines for scProteomics are also evolving rapidly, with no consensus yet on best practices for quality control, normalization, batch correction, and differential abundance testing at the single-cell level. Notably, replication-focused analyses of single-cell proteomics data emphasize that batch correction and data missingness are recurrent, high-impact challenges that can materially affect downstream biological conclusions if not handled with explicit, stability-aware strategies [85]. Because scProteomics data are characterized by higher missingness and stronger run-level effects than bulk proteomes, robust QC governance, reporting standards, and stability-aware statistics are essential to avoid overconfident differential calls [76, 80, 82, 85]. Clinically, cost, complexity, and sample throughput currently confine single-cell and ULIP proteomics to specialized research settings rather than routine diagnostics. Bone sarcomas are rare, so assembling cohorts of sufficient size to draw robust conclusions requires multi-center collaboration and harmonized protocols. Even when high-quality scProteomics datasets are generated, translating their insights into clinically actionable tests often involves down-selection to smaller panels of proteins measurable by more scalable techniques (e.g., targeted MS, immunohistochemistry, or multiplexed immunofluorescence), followed by prospective validation. These steps mirror the translational barriers discussed for bulk proteomics biomarkers, but the added cost and complexity of single-cell workflows raise the bar further for demonstrating added value. Despite these challenges, early feasibility studies in other solid tumors indicate that single-cell and ULIP proteomics can be successfully implemented on clinical specimens and can reveal biologically and clinically meaningful heterogeneity that is invisible to bulk assays [69, 71]. The key for bone malignancies will be to prioritize focused questions and carefully designed pilot studies that establish technical feasibility and generate clear hypotheses for follow-up, rather than attempting immediate large-scale surveys. A roadmap for implementing single-cell and ultra-low-input proteomics in bone tumor research A pragmatic roadmap for deploying single-cell and ultra-low-input proteomics in bone tumor research over the next 5–10 years can be conceptualized in three overlapping phases (Fig. 10). Rather than prescribing specific technologies, this roadmap emphasizes aligning methodological choices with biological questions and integrative analysis plans. Conceptual roadmap for implementing single-cell and ultra-low-input proteomics in bone tumor research. Phase I focuses on feasibility and exploratory atlases from limited specimens; phase II builds integrated multi-omics cohorts linking cell states to bulk subtypes and secretome/EV networks; phase III distils key findings into targeted assays and translational applications Phase I: feasibility and exploratory atlases The initial phase would focus on small, carefully curated cohorts of osteosarcoma and related bone malignancies, for example 10–20 primary tumors and a limited number of matched lung or bone marrow metastases. The primary aims would be to (i) establish robust protocols for specimen processing (fresh vs. frozen, dissociation vs. LCM), (ii) benchmark one or two ULIP/scProteomics workflows on bone-derived material, and (iii) generate exploratory maps of tumor, immune, and stromal cell states at protein resolution. In this phase, technical replication, cross-validation with scRNA-seq and immunohistochemistry, and detailed reporting of pre-analytical variables would be more important than cohort size. Early findings might include the identification of recurrent proteomic cell states (e.g., metastasis-primed tumor cells or immunosuppressive macrophage subsets) that can be provisionally linked to bulk proteomic subtypes or clinical variables. Phase II: integrated multi-omics cohorts Once feasibility is established, the next phase would involve assembling small-to-medium-sized multi-omics cohorts in which a subset of patients contributes genomic, transcriptomic, bulk proteomic, and ULIP/scProteomics data. The objective here is twofold: to refine molecular subtypes by explicitly incorporating cell-state information, and to anchor bulk signatures in cellular sources. Analytical frameworks such as MOFA+, WNN, and related integrative methods would be used to derive joint latent factors that capture shared and modality-specific variation [46, 83]. For osteosarcoma, such cohorts could update existing subtype models by, for example, distinguishing tumors dominated by proliferative tumor-cell programs from those driven by stromal or immune modules, and clarifying which cell states underpin high-risk signatures identified in bulk datasets [8, 9]. In parallel, secretome and EV proteomics could be integrated to map ligand–receptor networks linking tumor subclones to pre-metastatic niche remodeling in lung and bone marrow [38]. Phase III: translational refinement and targeted assays The final phase would concentrate on translating single-cell and ULIP-derived insights into tractable biomarkers and therapeutic hypotheses. Protein panels that consistently mark high-risk cell states or immunologically “cold” niches could be distilled into targeted assays—for example, multiplexed immunohistochemistry or targeted MS panels—that are compatible with routine clinical specimens. These assays could then be evaluated prospectively as prognostic tools, minimal residual disease markers, or pharmacodynamic readouts in clinical trials testing intensified chemotherapy, targeted agents, or immunotherapy combinations. In parallel, scProteomics-informed hypotheses about actionable dependencies (e.g., pathway activation in specific subclones, immune checkpoint expression patterns) could inform the design of small, biomarker-enriched early-phase trials in osteosarcoma and related sarcomas. In summary, single-cell and ultra-low-input proteomics should be viewed not as an isolated technology push but as a logical next step in a broader proteogenomics program for bone malignancies. When coupled to existing bulk and multi-omics resources, and implemented through staged, question-driven studies, these approaches have the potential to clarify the cellular underpinnings of osteosarcoma heterogeneity, refine risk stratification, and nominate mechanistically grounded biomarkers and targets for future translational efforts. Conclusions and future perspectives In this Review, we argue for a function-first view of bone malignancy biology in which proteomics and phosphoproteomics serve as the most direct bridge from molecular alterations to executed pathway states and clinically actionable decisions. We propose a practical framework that connects three layers of evidence—bulk tissue programs, circulation/secretome-derived signals, and multi-omics subtype context—to the translational “last mile” of assay design and workflow integration. Finally, we outline a staged roadmap in which ultra-low-input and single-cell proteomics are deployed explicitly to overcome the dominant failure modes of bulk profiling (cellular admixture, state-dependent signalling, and cohort variability), and to distil cell-state insights into deployable panels for risk stratification, monitoring, and trial stratification. What is now actionable: a proteomics-to-decision toolkit Rather than asking whether proteomics is “useful” for bone malignancies, the more relevant question in the current era is what it already enables as a decision-oriented toolkit. Across cohorts and specimen types, three deliverables have become sufficiently consistent to treat as working building blocks. First, pathway-program readouts from bulk tissue proteomics have converged on a small set of recurrent axes—ECM/adhesion and focal-adhesion signalling, PI3K–AKT activity, mitochondrial metabolism, and stress-response programs—that repeatedly organize osteosarcoma heterogeneity and align with outcome and immune contexture [8, 9, 16]. This matters because it moves the field from protein lists to interpretable, modular “states” that can, in principle, be targeted or monitored. Second, circulating proteomics has matured into a panel problem rather than a single-marker problem: independent studies often rediscover complement/coagulation cascades, ECM fragments, and metabolic markers in serum/plasma, suggesting that the reproducible unit is a fit-for-purpose multi-marker panel tied to a defined use case (diagnosis, prognosis, surveillance) [11, 14]. Third, secretome and EV studies have sharpened what “mechanistic plausibility” can look like for biomarkers and targets. Secreted mediators that remodel distant organs prior to overt metastasis offer anchors for niche biology and for monitoring concepts, with EFEMP1 providing a concrete example of a mediator tied to pre-metastatic lung remodeling [38]. In the same spirit, APO-TCN2 illustrates a practical prioritisation logic—secretome detectability, tumour proteome support, and RNA concordance—followed by evaluation in patient serum [31]. These examples are still uncommon, but they demonstrate an important point: proteomics can generate hypotheses that survive contact with clinically accessible matrices. Limitations of the current evidence and of this Review This Review is constrained by the underlying evidence base. First, bone malignancy proteomics cohorts are often small, clinically heterogeneous (age, stage, anatomic site, treatment), and processed with non-uniform pre-analytics, limiting direct comparability across studies. Second, “bone-tissue constraints” (scarcity, decalcification, and matrix effects) introduce systematic biases that are incompletely reported in many publications, which can inflate irreproducibility and complicate meta-synthesis. Third, for non-osteosarcoma entities (e.g., Ewing sarcoma, chondrosarcoma, chordoma) the depth of proteomics/phosphoproteomics layers is uneven, so cross-entity generalizations are most credible at the level of pathway programs rather than entity-specific drivers. Finally, as a narrative review, our literature search was designed to capture representative, clinically relevant and methodologically instructive studies rather than exhaustively enumerating all reports; we therefore emphasize evidence hierarchies and reproducibility criteria to help readers interpret conclusions under data sparsity. Why translation still stalls: three bottlenecks that define the next decade If the toolbox exists, why is proteomics not yet a routine pillar in sarcoma care? Across the literature synthesised here, the translation gap is less about discovery and more about three bottlenecks that repeatedly break the chain between “significant” and “usable.” Bottleneck 1: averaging hides the decisive biology. Bulk measurements blur tumour–immune–stroma mixing, dilute PTM-driven pathway states, and miss rare resistant or metastatic subclones—exactly the features that determine relapse and therapy failure in high-grade sarcomas. This is why pathway themes can be reproducible yet still fail to yield stable decision thresholds across cohorts. Bottleneck 2: real-world variability dominates weakly anchored signals. In blood, perioperative timing, inflammation, infection, and treatment effects can rival tumour-derived variance; in tissue, decalcification chemistry, ECM/mineral matrix effects, necrosis, and purity shifts can overwhelm biological differences. Without harmonised handling and disciplined covariate modelling, replication failure is often unsurprising rather than informative. Bottleneck 3: clinical utility is not guaranteed by biological truth. Even robust signals must be packaged into assays with defined analytical performance and embedded in workflows where a result changes an action—surveillance intensity, stratification for combinations, or pharmacodynamic monitoring. Large sarcoma proteome series and cohort-scale chordoma proteogenomics show what “usable” evidence tends to require: adequate cohort size, reproducible module calls, and validation that adds value beyond histology and transcriptomics [25, 43]. Seen this way, the field’s next step is not simply “more omics,” but more resolution where bulk fails, and more discipline where variability dominates. Future perspectives: a roadmap from cell states to deployable assays Single-cell and ultra-low-input proteomics are best viewed not as an optional upgrade, but as a direct response to the bottlenecks above. As sensitivity, microfluidic handling, and low-input workflows improve, protein- and PTM-level profiling is moving closer to resolving the units that actually govern outcome: cell states, niches, and treatment-conditioned dependencies [69, 72, 73]. For bone malignancies, where specimens are scarce and often decalcified, the “low-input” dimension is as important as the “single-cell” dimension. A practical roadmap can be framed around three translational milestones—each with a clear output that connects back to the bulk era rather than replacing it. Milestone 1: map bulk programs to cell-state mechanisms. The near-term win is not a complete atlas, but a linkage: which tumour-cell states and microenvironmental compartments instantiate the bulk ECM/adhesion, metabolic–stress, and immune/stromal programs; and which phosphorylation-defined pathway states mark actionable dependencies. Milestone 2: identify and track resistant niches. Therapy-induced drug-tolerant states and microenvironmental sanctuaries are likely to explain why subtype-conditioned vulnerabilities fail to generalise. Single-cell and microproteomic approaches can define which niches persist under treatment and which secreted/EV signals report their emergence, creating a mechanistic bridge from target discovery to longitudinal monitoring. Milestone 3: distil cell-state knowledge into clinical-grade panels. The endpoint is not single-cell MS in every hospital, but signatures that can be measured robustly—targeted MS panels, IHC multiplexes, or composite scores—that improve risk models, stratify trials, or serve as pharmacodynamic readouts. Delivering these milestones will require multi-institutional cohorts, bone-aware sampling protocols, and joint design by clinicians, pathologists, and proteomics engineers. The encouraging point is that the field already has a scaffold: bulk proteomics defines reproducible programs, secretome/EV studies provide mechanistic candidates, and multi-omics integration supplies subtype and pathway context. If higher-resolution proteomics is deployed explicitly to repair the three bottlenecks above, proteomics has a realistic path to reshaping how bone malignancies are classified, monitored, and treated over the next decade. Acknowledgements Not applicable. Abbreviations Data-dependent acquisition Data-independent acquisition Extracellular matrix Extracellular vesicles Formalin-fixed paraffin-embedded Immune checkpoint inhibitor Liquid chromatography tandem mass spectrometry Osteosarcoma Post-translational modification Single-cell proteomics Tumor immune microenvironment Tandem mass tag Ultra-low-input proteomics Author contributions Jianming Chen conceived and designed the study, performed data curation and formal analysis, and drafted the initial manuscript. Jiaqiao Luo, Ning Ni and Jinhao Zhang contributed to methodology development, conducted data analysis, and participated in manuscript revisions. Lei Jiang, Hongbao Wu and Shenghao Wen assisted with data curation and manuscript review. Ninghao Hu and Jiaran Li provided resources and contributed to manuscript revisions. Gang Deng and Xiaomeng Shi assisted with data curation, prepared visualizations, and reviewed the manuscript. Shuxiang Song contributed to the study’s conceptualization, supervised the project, and revised the manuscript critically for important intellectual content. All authors read and approved the final manuscript and agree to be accountable for all aspects of the work. Funding This work was funded by 2023Z161 Ningbo Key Research and Development Program and Zhejiang Provincial Medical and Health Science and Technology Program Project 2024kY1616. Data availability Not applicable. No new datasets were generated or analyzed in this review. Declarations Ethical approval and consent to participate Not applicable. This article is a narrative review of publicly available literature and does not involve new studies with human participants, identifiable data, or animal experiments performed by the authors. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Jianming Chen, Email: cjmdn01710@163.com. Shuxiang Song, Email: shxsong@bjmu.edu.cn. References Associated Data Data Availability Statement Not applicable. No new datasets were generated or analyzed in this review.