The study, led by Dennis Friedel, Rhaissa Ribeiro da Silva, Felix Sahm and Philipp Sievers of University Hospital Heidelberg together with collaborators across the German Cancer Research Center, set out to answer a deceptively simple question: can proteomic and phosphoproteomic profiling of ordinary archived tissue recover biologically meaningful signaling information that complements established molecular diagnostics? The researchers focused on ten cases of IDH-wildtype glioblastoma, the most aggressive adult brain tumor, split evenly between five tumors carrying EGFR amplification — one of the most frequent and well-characterized genomic alterations in this disease — and five without it. Every tumor had already undergone complete diagnostic workup, including DNA methylation profiling on the Infinium EPIC array with classification by the Heidelberg Brain Tumor Classifier and targeted sequencing of 220 CNS tumor genes, providing a rigorous genomic and epigenomic reference against which the new protein-level data could be judged.

The technical achievement underlying the study lies in how the tissue was prepared. Formalin fixation and paraffin embedding have long been the nemesis of proteomics, because cross-linking and wax infiltration make proteins difficult to extract. The Heidelberg team applied an optimized plate-based workflow that dispensed with conventional xylene deparaffinization altogether. Tiny tissue punches were placed in a 96-well plate with sodium dodecyl sulfate and magnetic beads, then processed in a BeatBox tissue homogenizer, where an alternating magnetic field drives the beads to mechanically shatter tissue and surrounding paraffin. Four cycles of ten minutes of bead-based disruption alternated with twenty minutes of heating at 95 degrees Celsius liberated the proteins, which were then cleaned up and digested using an automated single-pot solid-phase-enhanced sample preparation, or SP3, workflow on a liquid-handling robot, with 55 micrograms of protein input per sample.

Phosphopeptide enrichment was performed with iron-NTA cartridges on an automated platform before analysis on an Orbitrap Exploris 480 mass spectrometer. Global proteomes were measured using a 90-minute data-independent acquisition method, while phosphoproteomes used data-dependent acquisition, with rigorous quality controls including an MCF7 reference sample processed alongside the tumors and regular injections of a HeLa peptide standard to monitor instrument stability. The results were striking: global profiling identified a median of 7,647 proteins per tumor, with 5,417 proteins detected in all ten samples. The phosphoproteomic analysis reached a total of 13,467 phosphosites, with a cohort median of 5,457 confidently localized sites per sample, and 1,233 phosphosites reproducibly found in every single tumor. Comparable intensity distributions across samples confirmed that the quantification was robust and consistent.

To interpret phosphorylation correctly, the researchers made a crucial methodological choice: phosphosite intensities were adjusted by subtracting the measured abundance of the corresponding protein from the matched global proteome. This protein-abundance adjustment helps ensure that differences in phosphosite levels reflect genuine changes in phosphorylation state rather than simply more or less of the underlying protein being present. With this adjustment in place, the team asked whether EGFR amplification left a detectable signature on the phosphoproteome — effectively using a known genomic alteration as a biological benchmark for the new technology.

The answer was yes, but with an important nuance. Principal component analysis of the adjusted phosphoproteomic data showed only partial separation between the two groups, with EGFR amplification status significantly associated with the first principal component, which accounted for 29.9 percent of total variance. Differential analysis identified 443 phosphosites meeting predefined exploratory criteria, and among the sites enriched in EGFR-amplified tumors were two phosphorylation sites on the EGFR receptor itself, Y1110 and Y1197, which remained enriched even after adjustment for protein abundance — a biologically coherent finding given the known signaling role of the amplified receptor. Kinase-substrate enrichment analysis, which infers kinase activity from the coordinated behavior of annotated substrate sites, revealed higher activity of EGFR together with SRC-family kinases including YES1, FYN, SRC, LCK and LYN in the amplified tumors, while kinases such as PRKACA, PRKG1, MAPKAPK2 and PDPK1 showed lower inferred activity in that group.

That nuance — substantial overlap between the groups despite the clear genomic difference — may be the study’s most conceptually significant message. Two tumors can carry the same driver alteration yet differ markedly in how that alteration is expressed at the level of downstream signaling. Genomics establishes that a driver is present; phosphorylation patterns reveal whether and how its pathway is actually being used. The authors argue that this makes phosphoproteomics an orthogonal functional layer that complements, rather than duplicates, genomic and epigenomic classification. The global proteome told a parallel story: EGFR protein itself was more abundant in amplified tumors, E2F target programs were enriched in that group, while non-amplified tumors showed enrichment of stress-, immune- and microenvironment-associated programs including hypoxia, interferon-gamma response, complement and apoptosis signaling.

To show how this might work in practice, the team built a prototype research-use-only report for one representative EGFR-amplified tumor, generated entirely from that single sample without reference to the rest of the cohort. The report combined analytical quality-control metrics with pathway-level enrichment and representative phosphosites, grouped into therapeutically relevant signaling categories such as EGFR/ERBB, RAS–RAF–MEK–ERK/MAPK, PI3K–AKT–mTOR and VEGF-associated signaling, with sites including EGFR Y869 and Y1197 supporting the pathway-level findings. The authors stress that this report is a demonstration of a potential reporting framework, not a validated diagnostic or therapeutic readout, and implies nothing about drug sensitivity.

The broader context makes the approach timely. Large-scale proteogenomic studies of glioblastoma have already shown that signaling states undergo substantial post-translational remodeling during tumor evolution, meaning functional state cannot be inferred from genomic alterations alone. Elsewhere in oncology, phosphoproteomic profiling has identified patient-specific drug targets in cholangiocarcinoma and defined clinically distinct subtypes with actionable vulnerabilities in hepatocellular carcinoma. In neuro-oncology, the N2M2/NOA-20 umbrella trial has already demonstrated the clinical value of functional pathway assessment, selecting glioblastoma patients with activated mTOR signaling — judged by phospho-mTOR staining — for treatment with temsirolimus. Multiplexed phosphoproteomics could in principle extend such strategies by assessing many signaling pathways simultaneously, though the authors caution that prospective validation and clinically applicable thresholds will be required.

Significant hurdles remain before this technology reaches the clinic. The study cohort comprised only ten tumors, so individual phosphosites and inferred kinase activities must be treated as hypothesis-generating rather than candidate biomarkers. Bulk tissue analysis cannot resolve the spatial and subclonal heterogeneity that defines glioblastoma, and the protein-abundance adjustment, while informative, does not constitute a direct measurement of phosphorylation stoichiometry. Kinase-substrate enrichment infers activity from annotated substrate behavior rather than measuring enzymatic activity directly. Analytical complexity, infrastructure requirements and the cost of mass spectrometry will likely confine the approach to selected clinical and translational settings for now, and standardization of tissue processing, pipelines, thresholds and reporting will be essential for broader implementation. Nevertheless, the core demonstration stands: routine FFPE archive tissue, long valued primarily for morphology and DNA, also preserves a readable record of the tumor’s active signaling life. Integrating that functional layer with genomic and epigenomic diagnostics could eventually give neuro-oncologists a more complete picture — connecting who a tumor is with what it is actually doing.

Subject of Research: Proteomic and phosphoproteomic profiling of archived FFPE glioblastoma tissue to complement genomic and epigenomic CNS tumor diagnostics

Article Title: Functional proteomic and phosphoproteomic profiling of routine FFPE CNS tumor tissue complements genomic and epigenomic characterization

Article References: Friedel, D., da Silva, R. R., Ahmed, I. A., Wolff, B. M., Neuerburg, A., Irsevic, R., Ahmadi, S., Jayavelu, A. K., Kulozik, A. E., Witt, O., Pfister, S. M., Krieg, S. M., Wick, W., von Deimling, A., Reuss, D. E., Sahm, F., & Sievers, P. (2026). Functional proteomic and phosphoproteomic profiling of routine FFPE CNS tumor tissue complements genomic and epigenomic characterization. Acta Neuropathologica, 152(1), Article 42. https://doi.org/10.1007/s00401-026-03091-6

Image Credits: AI Generated

DOI: 10.1007/s00401-026-03091-6

Keywords: phosphoproteomics, proteomics, FFPE tissue, glioblastoma, EGFR amplification, mass spectrometry, molecular diagnostics, CNS tumors, DNA methylation profiling, kinase signaling, precision oncology, neuropathology

The team began with a strikingly simple question: if you look at the complete inventory of proteins in a failing left ventricle, do hearts that failed for different reasons actually look different from one another? To answer it, they used high-resolution mass spectrometry to quantify 3,788 proteins in cardiac tissue samples collected from patients undergoing advanced care, alongside non-failing control hearts. The patients spanned ischemic cardiomyopathy, dilated and hypertrophic cardiomyopathy, arrhythmogenic cardiomyopathy, myocarditis, adult congenital heart disease and non-ischemic dilated cardiomyopathy, among other diagnoses. Statistical models adjusted for age, sex and batch effects compared protein abundance in each etiology group against controls.

The first major finding was one of convergence. When patients were grouped by clinical diagnosis, the vast majority of significantly altered proteins were shared across etiologies rather than unique to any single one. Upset plots and enrichment analyses showed that common changes dwarfed diagnosis-specific ones, and functional enrichment mapped all the heart failure groups onto a shared landscape: metabolic reprogramming, altered contractile machinery, extracellular matrix remodeling and inflammatory signaling appeared everywhere. In other words, by the time the heart reaches end-stage failure, the terminal proteomic state looks remarkably similar regardless of what started the decline. This convergence echoes earlier single-nucleus RNA sequencing work, which likewise found common transcriptional signatures across cardiomyopathies, and it helps explain why a one-size-fits-all therapeutic approach has historically achieved only incremental gains.

But convergence at the disease endpoint was only half the story. When the researchers abandoned clinical labels altogether and let an unbiased computational method, non-negative matrix factorization, sort patients purely by proteomic similarity, the cohort split cleanly into two groups. Group 1 and Group 2, comprising 59 and 63 patients with reduced ejection fraction respectively, cut across every clinical etiology: a patient with ischemic disease and a patient with myocarditis could sit in the same molecular camp, while two patients with the same diagnosis could fall into opposing ones. The clustering was robust, reproducible in principal component analyses, and independent of demographic confounders.

What distinguished the two groups was not the cause of disease but the character of the heart’s remodeling response. Group 2 hearts were enriched for proteins derived from activated fibroblasts, smooth muscle cells and pro-inflammatory myeloid cells, signatures consistent with aggressive scar formation and immune infiltration. Overlaying the protein signatures onto a published atlas of 15 cardiac cell types from single-cell transcriptomics revealed that Group 2 mirrored the cellular states observed in dilated cardiomyopathy, including expansions of disease-associated cardiomyocyte populations, activated fibroblast states FB5 through FB9, and pro-inflammatory macrophages, monocytes and dendritic cells. Group 1, by contrast, retained a profile closer to non-failing tissue, relatively enriched for cardiomyocyte contractile and metabolic proteins. The subgrouping was independently validated in bulk RNA sequencing data from 198 heart failure samples in a public cohort, where the same protein signature reproduced the same molecular split at the transcript level.

The researchers then descended one level deeper into biology by profiling the phosphoproteome, the layer of phosphate tags that flip proteins on and off and transmit signals through kinases. From the same tissues they quantified 3,014 phosphosites, and the phosphorylation patterns reproduced the proteomic stratification, confirming that the two subgroups differ not only in which proteins they contain but in which signaling circuits are actually running. Critically, phosphoproteomics is where drug discovery lives: kinases, phosphatases and their downstream effectors are among the most pharmacologically tractable targets in medicine.

By integrating protein abundance, phosphosite regulation and prior knowledge networks of kinases, transcription factors and upstream regulators, the team identified selectively activated, druggable signaling networks confined to one molecular subgroup. Among the circuits implicated were RhoA and ROCK2 signaling, components of the actin cytoskeleton such as the Arp2/3 complex, and mitogen-activated protein kinase pathways including ERK1/2 and p38, alongside regulatory nodes such as glycogen synthase kinase 3 and RAF1. The researchers cross-referenced these targets against the Drug Repurposing Hub, ChEMBL, Open Targets and Mendelian randomization evidence for causal heart failure associations, filtering for known cardiac toxicity, and assembled a shortlist of candidate therapeutic entry points that would apply only to the molecularly defined patients whose networks show that activation. This is precision cardiology at the level of the cell’s control systems rather than its genome.

Perhaps the most clinically consequential result came from the UK Biobank. The remodeling-associated protein signature, originally discovered in end-stage explanted hearts, was tested against plasma proteomic data from an entirely independent cohort of heart failure patients who are, on average, at much earlier stages of disease and treated in the community rather than the transplant center. The signature stratified these patients into subgroups and predicted survival trajectories, with accelerated failure time modeling showing that the molecular group assignments carried prognostic information. This extends the relevance of the tissue-based findings far beyond the operating theater: a blood test grounded in the same biology may eventually help identify which ambulatory patients carry a fibrotic, inflammatory molecular phenotype and which do not.

The implications for drug development are considerable. Heart failure trials have long been plagued by modest average effect sizes, and a plausible explanation is that molecularly heterogeneous populations dilute real benefits among patients whose disease is not driven by the targeted pathway. If a candidate anti-fibrotic or kinase inhibitor benefits only Group 2 biology, enrolling an unstratified population would bury the signal. Stratifying trials by these protein-defined subgroups could sharpen effect estimates, reduce required sample sizes and rescue otherwise promising mechanisms. The prognostic signal in plasma also raises the prospect of patient stratification with routine blood draws rather than myocardial biopsy, something no clinician would contemplate for diagnosis in most heart failure settings.

The study also leaves open questions that will shape the next phase of work. The cohorts are dominated by end-stage disease and reduced ejection fraction, so the behavior of the subgrouping in heart failure with preserved ejection fraction remains untested. The two groups describe biology at a moment in time, and longitudinal studies will need to establish whether patients transition between molecular states as disease evolves, or whether subgroup membership is fixed from the outset. And while the plasma findings demonstrate prognostic value, translating them into a clinically deployable test will require targeted assays and prospective validation. Still, the message of the work is unambiguous: the failing heart’s molecular interior tells a story that diagnosis codes cannot, and listening to that story, in both tissue and blood, may finally make it possible to match heart failure therapies to the biology of the individual patient rather than the label on the chart.

Subject of Research: Molecular subgrouping of human heart failure by proteomic and phosphoproteomic profiling of cardiac tissue

Article Title: Cardiac proteomic and phosphoproteomic profiling defines clinically relevant molecular subgroups in human heart failure

Article References: Hamed, O., Reitz, C. J., Kuzmanov, U., Gramolini, S., Hu, M., Halpern, J., Billia, F., & Gramolini, A. O. (2026). Cardiac proteomic and phosphoproteomic profiling defines clinically relevant molecular subgroups in human heart failure. Nature Cardiovascular Research. https://doi.org/10.1038/s44161-026-00880-w

Image Credits: AI Generated

DOI: 10.1038/s44161-026-00880-w

Keywords: heart failure, proteomics, phosphoproteomics, cardiac remodeling, precision medicine, biomarkers, UK Biobank, fibrosis, kinase signaling, cardiomyopathy, left ventricle, mass spectrometry