Nature Cancer

Genomics-guided precision oncology has improved survival in cancer entities with actionable mutations but cannot capture oncogenic signaling that manifests at the protein level. Here, we report a prospective, real-world pan-cancer study profiling proteomes and phosphoproteomes of 1,998 tumor samples from adults and children with rare or advanced cancers enrolled in the German precision oncology programs DKFZ/NCT/DKTK MASTER, CATCH and INFORM and their molecular tumor boards (MTBs). We developed tumor proteome activity status (TOPAS) scores for 46 clinically relevant kinases, an immune activity score capturing antigen presentation and T-cell activation and identified therapeutically targetable cell-surface proteins for 94% of patients. These readouts enhance MTB recommendations by exposing actionable non-genomic kinase activity, refining interpretation of oncogenic genome alterations, and highlighting cell-surface treatment options. Three proof-of-concept analyses indicate clinical utility including kinase activity-stratified pazopanib response in sarcoma, immune activity score-tracked checkpoint-inhibitor outcomes pan-cancer, and a phosphoproteomic biomarker distinguishing EGFR-inhibitor response in chordoma.

Hematologic malignancies are diagnosed through a fragmented, sequential workup of morphology, immunophenotyping, cytogenetics, and molecular testing that can take days to weeks and is unavailable at many centers. DNA methylation profiling has transformed central nervous system tumor diagnosis, yet hematologic classifiers have remained confined to narrow acute leukemia panels. Here we present Lamprey, a deep-learning methylation classifier spanning 86 hematologic malignancy entities, trained on a reference cohort of 8,544 patients and deployed directly from nanopore sequencing. A depth-aware training framework allows confident classification from the first minutes of a run. Against blinded integrated reference diagnoses across retrospective, external, and prospective cohorts, Lamprey exceeded 98% accuracy among classified cases. Lamprey reaches a confident call within minutes, and cost as little as $82 per sample. Lamprey consolidates a sequential diagnostic workup into a single, rapid, same-day molecular readout.

The role of comprehensive genomic profiling for therapeutic decision-making is established in high-risk pediatric cancers, but its utility in rare and diagnostically challenging tumors is unclear. Here we report 123 non-high-risk patients enrolled in the Australian ZERO Childhood Cancer Program for diagnostic uncertainty, clinician request to address a specific molecular query, or other rare tumors. Comprehensive multi-omic profiling led to a change in diagnosis in 17.9% (22/123) of patients, with overall diagnostic utility in 35% (43/123). Molecular queries were resolved in 97.6% (40/41). Multi-omic results informed conventional management in 20.3% (25/123). Precision-guided therapy was recommended in 67.5% (83/123), and administered in 36.1% (30/83), with an objective response or prolonged (>6 months) stable disease in 88.9% of evaluable cases (16/18). Findings were confirmed in an independent cohort from the Canadian KiCS program (n=41). In conclusion, in rare and diagnostically challenging pediatric tumors, multi-omic profiling improved diagnostic accuracy and informed clinical management, supporting its integration into routine care.

Cancer outcomes vary widely between individual patients, each accumulating an irregular record of treatments, diagnoses, measurements, and complications. Current prognostic models reduce this complexity into a single snapshot, focus on narrow clinical settings, and rarely generalize across hospitals. Here we introduce Chronicle, an explainable transformer that learns from entire patient trajectories, predicts diverse clinical outcomes throughout the disease course while capturing both short- and long-term temporal dependencies. Trained on 53.7 million longitudinal data points from 51,711 patients spanning 67 cancer types, Chronicle operates natively on irregular data without imputation and jointly predicts eight endpoints within a flexible framework adaptable to additional outcomes. Chronicle outperformed cross-sectional models for overall survival prediction (C-index 0.84 vs 0.76-0.79), stratified patients more accurately than established prognostic systems, including TNM stage, and predicted seven adverse event and transfusion endpoints (AUC 0.80-0.92). Applied without retraining to 69,341 patients in Germany, Switzerland, and the United States, Chronicle generalized across healthcare systems and improved further with local fine-tuning. Integrated explainability traced each risk update to patient-specific clinical factors, revealing distinct temporal persistence of prognostic information, with relevance half-lives ranging from weeks for therapies to nearly one year for baseline characteristics. These findings demonstrate that learning from hospital-wide patient trajectories enables interpretable and continuously updated predictions, providing a scalable framework to support individualized treatment decisions.

Background: Ovarian cancer (OC) exhibits substantial heterogeneity in response to platinum-based chemotherapy, resulting in variable clinical outcomes and frequent recurrence. Current biomarkers, including serum CA-125 kinetics and BRCA mutational status, incompletely predict therapeutic response. We investigated whether patient-derived organoids (PDOs) could functionally stratify chemotherapy sensitivity and better reflect patient-specific clinical behaviour. Methods: Twenty patients with OC treated between January 2024 and May 2026 were included, from whom fourteen PDO lines were successfully established. Eight PDOs with robust low-passage expansion and comprehensive longitudinal follow-up underwent functional profiling against carboplatin, paclitaxel, olaparib, and doxorubicin. Drug responses were assessed using half-maximal inhibitory concentration (IC50) and area under the curve (AUC) analyses and integrated with radiological response, serum CA-125 kinetics, BRCA status, and progression-free survival (PFS). Results: Clinical outcomes varied considerably despite similar platinum-taxane regimens. Although post-treatment CA-125 reduction was associated with prolonged PFS, neither CA-125 kinetics nor BRCA mutational status consistently predicted therapeutic response. PDO-guided functional stratification segregated tumours into four clinically relevant platinum-taxane response phenotypes: dual-sensitive, platinum-sensitive/taxane-resistant, platinum-resistant/taxane-sensitive, and dual-resistant. These functional categories closely mirrored radiological response, CA-125 normalisation, and disease progression patterns. PDOs exhibiting low IC50 and AUC values were associated with durable clinical benefit, whereas resistant PDOs tracked with persistent disease and early recurrence. Conclusions: PDO-guided functional stratification captures clinically meaningful therapeutic heterogeneity in OC and complements conventional biomarkers by directly measuring tumour-specific drug susceptibility. Prospective integration of PDO testing may facilitate patient-specific therapeutic selection and support functional precision oncology approaches in ovarian cancer.

The intensification of neoadjuvant therapy for early triple-negative breast cancer (eTNBC) - through the addition of carboplatin to standard chemotherapy and the incorporation of pembrolizumab - has markedly improved prognosis in recent years. However, this escalation carries a substantial risk of toxicity, and not all patients require the full regimen to achieve benefit. Realizing individualized treatment strategies will therefore depend on prognostic and predictive biomarkers that can forecast treatment response and long-term outcome. In the present study, we interrogated public gene expression datasets to develop transcriptomic signatures predicting response to neoadjuvant treatment and risk of recurrence. To validate these signatures, we used the Multiplex8+ platform for spatially informed comprehensive transcriptomic profiling in a real-world, multicenter, retrospective cohort of 590 patients diagnosed with eTNBC and treated with neoadjuvant chemotherapy with or without immunotherapy. The diagnostic Multiplex8+ test uses H&E and multiplexed RNA-FISH to guide the selection of specific tumor areas for the whole transcriptome sequencing and signature analysis. In the real-world cohort, the Multiplex8+ signatures were associated with both response and prognosis, remaining highly significant in multivariable models that included clinical parameters. The signatures were complementary to established biomarkers such as stromal tumor-infiltrating lymphocytes. These findings warrant prospective integration of the signatures into risk-stratified clinical trials to support future de-escalation and escalation strategies, enabling a better balance of efficacy, toxicity, cost, and drug availability.

Most high-dimensional studies of tumor-immune interactions focus on metastatic models, limiting insight into how immune remodeling in primary tumors shapes metastatic competence. Here, integrating single-cell RNA sequencing, CyTOF, and functional studies across metastatic (4T1) and non-invasive (EMT6) triple-negative breast cancer (TNBC) murine models, we define tumor state-specific immune programs that distinguish metastatic competence. Tumors with metastatic capacity uniquely drive early bone marrow expansion of CXCR2 neutrophils, which infiltrate primary tumors acquiring a CXCL2-producing phenotype that promotes EMT-associated cancer stem cell (CSC) plasticity. This program depends on TGF-{beta}/CEBPD-mediated induction of S100A9. Elevated CXCL2, together with G-CSF, establishes a feed-forward circuit that drives systemic neutrophil mobilization and recruitment to distant organs, where neutrophil-derived S100A8/A9 (calprotectin) promotes MET-driven CSC outgrowth and metastatic colonization. Clinically, gene signatures associated with CXCR2 neutrophils predict poor survival in TNBC patients, whereas monocyte/macrophage (CX3CR1) and T cell activation signatures correlate with improved outcomes. S100A9 ablation disrupts this cascade and enhances immunotherapy responsiveness, defining a TGF-{beta}/S100A9/CXCR2 axis linking immune remodeling, CSC plasticity and metastasis. HighlightsO_LIMetastatic TNBC engages a TGF-{beta}/C/EBP{delta}/S100A9 axis that expands CXCR2 neutrophils C_LIO_LINon-invasive EMT6 tumors retain a CX3CR1 monocyte/macrophage and T-cell landscape C_LIO_LICXCR2+ neutrophils in pre-metastatic niches suppress T cell response while promoting tumor cell proliferation C_LIO_LIS100A9 loss redirects myelopoiesis and potentiates anti-PD-L1 in TNBC models C_LI In BriefAlkan et al. dissect how tumor state programs the myeloid compartment in TNBC. Metastatic 4T1 tumors uniquely engage a TGF-{beta}/C/EBP{delta}/S100A9 axis driving CXCR2 neutrophil expansion and CXCL2/G-CSF-dependent systemic mobilization, coupling immune remodeling to EMT/MET cancer-stem-cell plasticity, while S100A9 loss restores CX3CR1 myeloid identity and unlocks checkpoint-inhibitor responsiveness.

Triple-negative breast cancer (TNBC) exhibits pronounced molecular heterogeneity, yet the majority of published transcriptomic prognostic signatures suffer from limited reproducibility and have not achieved clinical translation. We systematically benchmarked 62 published TNBC prognostic signatures across 6 independent cohorts (n=1,357) using a unified analytical framework spanning multiple scoring algorithms, survival endpoints, and threshold strategies. While 17 signatures demonstrated consistent univariate prognostic associations, only 4 remained independently prognostic after adjustment for clinicopathological variables, and none achieved robustness across all analytical conditions-underscoring the fragility of existing classifiers. Leveraging genes with concordant survival associations across all 6 discovery cohorts, we identified a reproducible 15-gene directionally concordant gene set (DCGS) signature and distilled it into MetaSig-EFS, a 13-gene prognostic model optimized using a cohort-aware DeepSurv framework. MetaSig-EFS demonstrated robust cross-cohort generalizability, achieving validation concordance indices of 0.89 in GSE19615 and 0.69 in JBordet, with corresponding 3- and 5-year time-dependent AUROCs of 0.89 and 0.96 in GSE19615 and 0.80 and 0.71 in JBordet, respectively, while retaining independent prognostic value across established TNBC molecular sub-typing systems. Leveraging the TAHOE-100M transcriptomic perturbation atlas, we systematically prioritized candidate therapeutics through large-scale drug repurposing, identifying Paclitaxel as the top-ranked compound, followed by Venetoclax and Tucatinib. Together, these findings provide biologically informed and clinically actionable therapeutic hypotheses that extend the translational utility of our validated prognostic framework for TNBC risk stratification.

Cluster of differentiation 47 (CD47) blockade promotes tumor cell phagocytosis; however, transitioning this initial innate immune event into durable anti-cancer adaptive immunity and tumor control remains a critical challenge. To address this translational gap, we have engineered aCD47 - CpG, a novel immune-stimulating antibody conjugate (ISAC) coupling a CD47-blocking antibody (aCD47) to a Toll-like receptor 9 agonist, unmethylated CpG, to enable synchronized deployment of co-stimulatory signals during tumor engulfment. This ISAC, aCD47-CpG, but not the parent aCD47, reprograms macrophages toward an anti-tumor M1-like phenotype, enhances antigen cross-presentation, and promotes robust CD8+ T cell priming. We found that systemically administered aCD47-CpG drove macrophage-dependent tumor suppression in a human lymphoma xenograft model. In parallel, the treatment in immunocompetent syngeneic models of lymphoma and highly immunosuppressive triple-negative breast cancer resulted in profound tumor regression, metastasis inhibition, and durable anti-cancer immune memory. These effects were accompanied by remodeling of the immunosuppressive tumor microenvironment toward an immune-permissive state, characterized by M2-to-M1 macrophage repolarization, intratumoral infiltration of cytotoxic and memory T cells, and depletion of regulatory T cells. This spatiotemporally coordinated immunotherapy platform offers a promising strategy to bridge innate and adaptive immunity, thereby advancing the translational potential of CD47-targeted cancer therapies.

Pancreatic ductal adenocarcinoma (PDAC) is driven by oncogenic KRAS in roughly 90% of cases, and KRAS-pathway inhibition has finally become clinically active. Durable benefit, however, will require identifying the adaptive and baseline vulnerabilities that shape response to KRAS inhibition. Two resistance mechanisms have been proposed separately in the literature - receptor-tyrosine-kinase bypass of KRAS, and dependence on the adhesion kinase FAK - but whether they are one target class or two, and which should partner a KRAS inhibitor, is unresolved. We integrate public perturbation, dependency, and survival data to nominate them as mechanistically separable candidate combination partners. Two findings define the separation. First, KRAS loss increases ERBB2/3 receptor expression. This appeared in both an inducible genetic KRAS-extinction model and, independently, in five PDAC lines treated with pharmacological KRAS-G12C/D inhibitors, while MAPK output collapsed as expected. The signal was clearest for ERBB2 and in the genetic model; in the small pharmacological cohort the effect was modest and its confidence intervals crossed zero, so we treat ERBB2/3 up-regulation as a candidate adaptive response - ERBB2-dominant and ERBB3-compatible - not a proven resistance mechanism. Second, focal adhesion kinase (FAK/PTK2) is the top-ranked standing druggable dependency within the KRAS/Src/RTK/adhesion network we examined (essential in 58% of pancreatic lines), yet it is not induced by KRAS shutdown. FAK dependency is present at baseline and, in DepMap, is statistically independent of a line's KRAS dependency (Spearman rho = +0.05, n.s.) - a genuinely standing vulnerability rather than a KRAS-rebound effect. The candidate adaptive response and the standing dependency are not positively co-regulated across the perturbed lines (pooled Spearman rho = -0.43, but n = 8 and n.s., so this cannot by itself establish independence); we therefore treat them as separable on mechanistic grounds - each nominated by different data and engaged by a different drug - rather than as statistically demonstrated independent programs. A Src-centered signaling-landscape analysis associates patient prognosis with the coordinated invasion-and-RTK program these nodes organize, rather than with any single transcript; this program remains prognostic after adjustment for a conventional EMT/stromal signature, which does not (Src-neighborhood per-standard-deviation OS hazard ratio 1.9, p = 3 x 10^-5; EMT signature null on adjustment). Together, these results motivate a concrete, testable hypothesis: that FAK inhibition (a standing dependency) and ERBB inhibition (a candidate induced adaptive response) are separable candidate partners for a KRAS inhibitor, best evaluated as distinct arms of a biomarker-stratified platform. They also clarify why single-agent Src inhibition - a non-oncogene dependency tested as monotherapy, without a KRAS backbone, in advanced rather than micro-metastatic disease - was not positioned to surface either mechanism. No protein-level, phospho-signaling, or combination-response validation is performed here; all findings are computational nominations that require experimental validation before any clinical inference.

Follicular lymphoma (FL) is an incurable, prototypical relapse-remitting cancer, implying the existence of therapy-persistent cells that survive frontline treatment and seed disease recurrence1-3. However, these persister cells remain difficult to study directly in patients because immediate post-treatment sampling is ethically and practically challenging. Using a genetically defined mouse model that allows sampling of persistent cells immediately after frontline R-CHOP therapy, we prospectively isolate and functionally define relapse-founding cancer persister cells (CPC). The CPC is an IgM memory-like B-cell with high germinal center re-entry capacity. This state represents a discrete component of a heterogeneous residual pool indicating that residual disease is polytypic and that relapse potential may depend on which cells persist rather than on residual tumour burden alone. By integrating mouse CPC with human FL datasets, we show that an analogous transcriptional programme is detectable at diagnosis and is enriched in patients with inferior clinical outcome across independent cohorts4,5. These findings support the concept that relapse risk is linked to a conserved, genotype-agnostic CPC programme present before therapy. To explore therapeutic vulnerabilities, we developed a scalable in-vitro platform that models the CPC-like state and used it to identify sensitivity to histone deacetylase inhibition. Romidepsin and panobinostat killed CPC-like cells in-vitro, and decreased therapy-persistent cells after R-CHOP treatment in-vivo and in patient-derived organoids. Together, these data define a tractable CPC state in FL, with a validated clinical readout and an immediately testable therapeutic entry point, opening CPC-directed strategies for durable FL control.

The acceleration of automated scientific discovery has been fundamentally bottlenecked by the epistemic gap between the semantic reasoning of large language models (LLMs) and the complex, non-linear reality of mammalian biology. While recent multi-agent frameworks have achieved autonomous hypothesis generation and in vitro experimental analysis, they frequently lack the rigorous statistical constraints required for multi-scale clinical translation. Furthermore, while algorithmic clinical digital twins successfully forecast biological states, they often rely on opaque latent spaces, sacrificing mechanistic interpretability for predictive accuracy. Here, we introduce the Multi-Scale Autonomous Discovery Engine (Octopus), a neuro-symbolic framework that unites a fully localised, privacy-preserving multi-agent swarm with regularised predictive algorithmic environments. Rather than stopping at isolated cellular assays, the system autonomously prioritises therapeutic hypotheses against in vitro CRISPR dependency data (CCLE), traces feature attribution cascades using XGBoost SHAP vectors, and orthogonally translates emergent vulnerabilities in silico to predict in vivo mammalian tumour trajectory (PDX) and human overall survival (Marisa). In a fully unsupervised sweep of colorectal cancer transcriptomes, the pipeline autonomously prioritised Insulin-like Growth Factor 2 (IGF2) as a significant candidate vulnerability to 5-Fluorouracil resistance. The discovery maintained significance after rigorous Benjamini-Hochberg false discovery rate correction (q = 0.0292, Log-Rank p = 0.0007) and successfully predicted significant in vivo tumour volume shrinkage in an independent mouse cohort (Mixed-Effects LMM p = 0.0373). By bridging agentic hypothesis generation with statistically bounded clinical survival, this framework establishes a verifiable, local paradigm for the automated computational prioritisation of biomedical discoveries.

Effective therapies for high-risk medulloblastoma (MB), particularly MYC-driven Group 3 (G3) MB, remain elusive due to limited druggable mutations, poor blood-brain barrier (BBB) penetration, and rapid resistance. We developed SINBA (Synergy Inference by Data-driven Network-Based Bayesian Analysis), a systems biology framework that computationally prioritizes synergistic, BBB-permeable drug combinations by identifying hidden drivers sustaining oncogenic programs. Integrating MB-specific networks, transcriptomic data, and drug-gene interactions, SINBA nominated 32 candidates, of which 19 were experimentally validated as synergistic. Through iterative prioritization and experimental refinement, the MEK inhibitor mirdametinib and p38 inhibitor regorafenib emerged as the top brain-penetrant pair, suppressing G3 MB progression and extending survival in xenograft and immunocompetent models, with efficacy enhanced by low-dose radiation. Single-cell analysis revealed selective targeting of developmental origins and immune reprogramming. These findings establish SINBA as a computationally assisted discovery framework for clinically actionable combinations in high-risk MB.

Targeting oxidative phosphorylation (OXPHOS) represents an attractive therapeutic strategy in acute myeloid leukemia, which exhibits exceptional dependence on mitochondrial respiration compared to normal hematopoietic cells. However, clinical attempts to exploit this vulnerability have been limited by on-target toxicity to healthy tissue. Here, we comprehensively compare the cellular consequences of inhibiting distinct nodes of the electron transport chain in AML. We demonstrate that selective inhibition of the F1 subunit of ATP synthase with EB2023 (ammocidin A) delivers an energetic stress to AML cells without the profound redox stress that characterizes complex I inhibition, preventing NAD/NADH imbalance and allowing continued TCA cycling. Further, the duration of OXPHOS inhibition is transient in nature in vivo, a finding revealed through pharmacokinetic and serial pharmacodynamic monitoring of AMPK phosphorylation accompanied by OPA1-mediated mitochondrial structural remodeling that primes AML cells for BCL2 inhibitor synergy. EB2023 in combination with venetoclax demonstrates potent anti-AML activity across cell lines and patient-derived xenograft models at doses that spare normal hematopoietic progenitors and avoid the neuropathy and sustained detrimental systemic metabolic rewiring in healthy tissues associated with prior efforts to target OXPHOS. These findings establish F1-selective ATP synthase inhibition as a clinically actionable therapeutic strategy in AML and establish the duration of OXPHOS inhibition as a critical and previously underappreciated determinant of therapeutic index.

Circulating tumor cells (CTCs) and immune cells form dynamic multicellular ecosystems in blood, but their spatial organization and clinical relevance have not been systematically characterized. We developed the Cell and Cluster Identification Program (CCIP), an artificial intelligence-based framework that analyzes routine multiplex immunofluorescence blood scans to segment cells, identify CTCs and five immune lineages with high accuracy, and quantify multicellular clusters and tumor-immune interactions. Applying CCIP to 2,693 blood scans from 1,399 patients, we profiled over 60 million cells (>7 million multi-cell clusters) and linked imaging-derived features to patient outcomes. Correlated with circulating-tumor DNA mutation burdens, a 14-feature image model predicted overall survival in breast cancer, outperformed clinicopathologic variables and CTC enumeration, and generalized to prostate cancer. Prognostic imaging signatures were also associated with therapy response-related progression-free survival as well as with single-cell RNA sequencing-derived immune suppression states, connecting circulating tumor-immune architecture with systemic immune dysfunction.

Circulating cell-free DNA (cfDNA) offers a minimally invasive lens into temporal tumor evolution. However, the accurate quantification of clonal composition from cfDNA remains challenging, particularly in low tumor fraction (TF) settings. Existing liquid biopsy deconvolution frameworks are frequently constrained by their reliance on bulk tissue references, simplified copy-number assumptions, and incomplete representations of clonal architecture, which collectively compromise sensitivity and bias evolutionary inferences. To address these limitations, we developed cfClone, a Bayesian framework that integrates single-cell whole-genome sequencing (scWGS) derived clonal structures with cfDNA whole-genome sequencing data to enable high-resolution, tissue-informed clonal tracking. Notably, while cfClone inherently leverages genomic instability, we demonstrate that cfClone achieves accurate TF estimates and circulating tumor DNA (ctDNA) detection even in malignancies with limited copy-number variant (CNV) burden. We validate cfClone in low and high CNV burden cases using simulated data derived from real patient data, establishing sensitive detection thresholds across a range of aneuploidy levels. By jointly modeling local copy-number alterations and allele-specific signals via Bayesian model selection and Markov chain Monte Carlo (MCMC) sampling, the algorithm yields uncertainty-aware estimates of clonal prevalence and TF. Applied to longitudinal clinical cohorts, cfClone reconstructs real-time evolutionary trajectories and uncovers clonal selection driving therapeutic resistance, including the de novo detection of emergent clonal populations. Github link: https://github.com/Roth-Lab/cfclone

Computational tools for phylogenetic inference of early tumor evolution in real time are currently lacking. We present LACHESIS, a standardized R package and Shiny app that times early and most recent common ancestors of individual tumors from whole-genome sequencing data. LACHESIS automates mutational signature-aware molecular clock modeling for trajectory reconstruction and evolution-based risk stratification. We validate its utility for childhood and adult malignancies, providing a broadly applicable pan-cancer workflow.

Activating HER2 mutations are significantly enriched in both primary and metastatic invasive lobular breast cancer (ILC), with large public datasets of primary breast tumors linking them to a worse prognosis in ILC. Despite their oncogenic role, no FDA-approved therapies currently target HER2-mutant breast cancers. While the HER2-directed antibody-drug conjugate (ADC) trastuzumab deruxtecan (T-DXd) has shown efficacy in HER2-mutant non-small cell lung cancer, its activity in HER2-mutant ILC remains unknown. Using the Caris real-world database, one of the largest cohorts with survival data in advanced breast cancers, we confirmed that HER2 mutations are more prevalent in advanced ILC than in invasive breast cancer of no special type (NST) tumors, are associated with worse survival in both histologies, yet predict improved response to T-DXd across subtypes, highlighting the need for mutation-directed, histology-informed therapies. Using endogenous HER2-mutant ILC cell lines (UACC3133-S310F, BCK4-L755S) and CRISPR-engineered isogenic ILC models with clinically relevant HER2 mutations (S310F, V777L), we found these mutations drive HER2/HER3 hyperactivation and downstream signaling, conferring increased sensitivity to HER2 tyrosine kinase inhibitors (TKIs) and T-DXd. Mechanistically, HER2 mutants showed enhanced receptor ubiquitination, internalization, and lysosomal degradation upon T-DXd treatment, explaining the observed drug sensitivity. While combining T-DXd with neratinib or the HSP90 inhibitor ganetespib yielded synergistic effects in long-term growth assays, accompanied by increased HER2 ubiquitination, the concurrent hyperactivation of HER3 in HER2-mutant cells suggested that co-targeting HER3 could provide an effective alternative strategy. Accordingly, HER2-mutant ILC exhibited enhanced sensitivity to the HER3-directed ADC patritumab deruxtecan (P-DXd) or LJM716, a HER3-targeting antibody. We further uncovered a previously unrecognized mechanism of P-DXd beyond HER3 ligand blockade and payload delivery: P-DXd promotes HER2/HER3 association, increases HER2 ubiquitination, and enhances T-DXd internalization, resulting in potent synergy with T-DXd. Mechanistically, we identified HER3 extracellular domains I and II as essential for P-DXd binding and for mediating P-DXd-induced HER2/HER3 association, establishing a structural basis for this activity. In vivo, both T-DXd and P-DXd suppressed UACC3133 and BCK4 xenograft growth, with combination therapy trending toward greater efficacy and prevented regrowth of tumors. Extending these findings beyond HER2-mutant ILC, combination treatment with T-DXd and P-DXd demonstrated synergistic activity across multiple breast cancer models, including (i) HER2-amplified NST patient-derived organoids (PDOs) harboring hotspot HER2 mutations, (ii) HER2-wild-type NST PDOs with clinically intrinsic or acquired T-DXd resistance, and (iii) isogenic HER2-mutant ILC PDOs with experimentally induced resistance after prolonged T-DXd exposure. Collectively, these findings support HER2 as an actionable target in HER2-mutant ILC and position T-DXd-based regimens, particularly in combination with HER3 inhibition, as a promising therapeutic strategy for this underserved patient population.

Rhabdoid tumors are very aggressive rare pediatric cancers with poor survival affecting very young children. They are characterized by the bi-allelic loss of SMARCB1 or SMARCA4, which is suspected to occur prenatally. However, their genomic evolution is not well understood. Here we assembled the largest cohort of whole-genome sequenced rhabdoid tumors to date, comprising 97 tumors from 88 children. We discovered that, in 42% of cases, the bi-allelic inactivation of the driver gene occurred via a Copy Number Neutral-Loss of Heterozygosity (CN-LOH). We exploited these CN-LOH events and the steady accumulation of age-related mutations in the tumor genomes to estimate the age of donors at the time of occurrence of the driver event and at the time of emergence of the clonal expansion. Across all cases with CN-LOH, the loss of the driver gene occurred very early during prenatal development. However, the clonal expansion that ultimately gave rise to the tumor occurred at different times during infancy, even several years after the acquisition of the founder event. These results indicate that probably other factors, besides the genetic driver event, are required to promote rhabdoid tumorigenesis.

Background: Medulloblastoma (MB) is the most common malignant brain tumour in children, and aggressive subgroups are frequently driven by the oncoproteins MYC or MYCN. Direct therapeutic targeting of MYC/MYCN has been challenging because of their intrinsically disordered protein structures. The aim of this study was to determine whether novel SE486-11 analogues (UNSW-SCs) can therapeutically target MYC/MYCN-driven MB. Methods: The anticancer activity of UNSW-SCs was assessed in MB cell lines with differential MYC/MYCN expression. Target engagement was evaluated using surface plasmon resonance and drug affinity responsive target stability assays. Blood-brain barrier penetration, MYC/MYCN protein degradation, cell cycle effects, apoptosis, DNA damage, and synergy with histone deacetylase (HDAC) inhibitors were examined. Therapeutic efficacy was evaluated in murine models of MYC- and MYCN-driven human MB. Results: UNSW-SCs showed potent anticancer activity, with preferential selectivity toward MB cells expressing high MYC/MYCN levels and IC50 values ranging from 0.22 to 1.18 M. The lead molecule, UNSW-SC-22, directly bound MYC, crossed the blood-brain barrier, and achieved a brain-to-plasma ratio of 1.44 at peak concentrations. UNSW-SC-22 induced MYC/MYCN-dependent cytotoxicity associated with enhanced proteasomal degradation, cell cycle arrest, apoptosis, and DNA damage. Combined treatment with HDAC inhibitors further reduced MYC/MYCN protein levels, increased DNA damage, and enhanced apoptosis. In vivo, UNSW-SC-22, either alone or with entinostat, significantly suppressed intracranial tumour growth and prolonged survival. Conclusions: UNSW-SC-22 is a brain-penetrant MYC/MYCN-targeting molecule with potent preclinical activity in MYC/MYCN-driven MB, supporting its development as a monotherapy or combination strategy with HDAC inhibition.