Supplementary Figure 1: Bone metastasis immunosuppressive microenvironment. Supplementary Figure 2: Gating strategy for RNA sequencing. Supplementary Figure 3: Differentially expressed genes during breast cancer bone metastasis progression. Supplementary Figure 4: Metastasis-infiltrating T cells are profoundly altered by metastatic progression. Supplementary Figure 5: Immune Checkpoint Receptors (ICRs) are expressed by metastasis-infiltrating T cells. Supplementary Figure 6: Targeting TIGIT results in activation of T cell mediated anti-tumor immunity. Supplementary Figure 7: (a) Spinal Cord Compression occurrence in treated mice. P-values were calculated using two-sided Chi square test. Supplementary Figure 8: TIGIT is highly expressed in human bone metastasis. Supplementary Table 1: Patient cohort data. Supplementary Table 2: Primers sequences.

Pathology foundation models (PFMs) have advanced rapidly in recent years and support training classifiers for a range of histopathology tasks. However, their robustness across hospitals remains limited: performance often degrades when training a classifier on data from one hospital and evaluating it on another target hospital. We address this challenge by fine-tuning PFMs with a local maximum mean discrepancy (LMMD) objective that applies to two settings: domain adaptation, where unlabeled target-hospital data is available, and domain generalization, where target-hospital data is unavailable at all. Experiments at both the patch- and slide-level show consistent improvements across multiple PFMs and tasks.

Abstract Background: Immune checkpoint blockade (ICB) has transformed cancer therapy, yet only a subset of patients benefit, underscoring the need for robust, generalizable biomarkers capable of identifying tumors with sufficient immune activation. Although numerous immune-related gene signatures and transcriptomic approaches have been proposed, most lack pan-cancer robustness and fail to generalize across independent cohorts. To address this gap, we derived TIME_ACT, an unsupervised 66-gene tumor immune activation (“hotness”) signature that integrates immune infiltration, inflammatory activity, and spatial TIL density. This signature accurately predicts tumor hotness across diverse cancer types, forming a robust pan-cancer measure of immune activation. Methods: TIME_ACT was evaluated across 22 publicly available pre-treatment ICB transcriptomic cohorts (n=1,416) spanning seven cancer types. To enable prediction from pathology, TIME_ACT scores were inferred from H&E slides using Path2Omics, a recently published deep-learning model trained on TCGA, and validated in nine new histopathology cohorts from different medical centers and populations (n=459). Results: TIME_ACT accurately identifies immune-hot tumors across eight TCGA cancer types (AUC 0.96-1.00) and two external datasets (AUC 0.96-0.98). TIME_ACT genes were consistently enriched in T cells, B cells, and dendritic cells at single-cell resolution. Spatial analyses shows that TIME_ACT-high regions co-localize with lymphocyte-dense niches adjacent to the tumor epithelium. Across 22 transcriptomic ICB cohorts, TIME_ACT robustly predicted response (mean AUC 0.74; mean OR 5.49). Importantly, histopathology-inferred TIME_ACT scores achieve strong performance across nine multi-centric cohorts (mean AUC 0.73; mean OR 5.07), matching transcriptomic-based prediction levels and importantly, outperforming direct slide-based models. Conclusions: TIME_ACT is a robust pan-cancer tumor immune activation signature that predicts ICB response directly from the readily available tumor pathology slides. Upon further prospective testing and validation, it offers an exciting new way for further democratizing precision immunotherapy. Citation Format: Danh-Tai Hoang, Sumit Mukherjee, Sumeet Patiyal, Lipika R. Pal, Tiangen Chang, Sumona Biswas, Saugato Rahman Dhruba, Amos Stemmer, Arashdeep Singh, Abbas Yousefi-Rad, Tien-Hua Chen, Binbin Wang, Denis Marino, Wonwoo Shon, Yuan Yuan, Mark Faries, Omid Hamid, Karen Reckamp, Barliz Waissengrin, Beatriz Ornelas, Keluo Yao, Pen-Yuan Chu, Lisa Ley, Dilara Akbulut, Nourhan El Ahmar, Sabina Signoretti, David Alexander Braun, Joo Sang Lee, Hyunjeong Joo, Hyungsoo Kim, Arsen Osipov, Robert A. Figlin, Jair Bar, Iris Barshack, Chi-Ping Day, Sridhar Hannenhalli, Karine Sargsyan, Andrea B. Apolo, Kenneth D. Aldape, Muh-Hwa Yang, Michael B. Atkins, Ze’ev A. Ronai, Eytan Ruppin. Pan-cancer prediction of response to immune checkpoint blockade from histopathology [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 3999.

Abstract Many cancer therapies show strong preclinical activity yet fail clinically due to inadequate experimental models. Conventional 2D cultures lack physiological relevance because they cannot reproduce the complex tumor-stromal-immune interactions or the biomechanical cues that shape tumor behavior and therapeutic response. This limitation is particularly pronounced in aggressive tumors, where patient heterogeneity and dynamic microenvironmental interactions drive treatment outcomes, or in rare tumors, where data on response to available therapies is scarce. To address this translational gap, we developed two patient-derived 3D platforms: (1) 3D tumoroids generated from the dissociated tumor tissues and co-cultured with matched peripheral blood mononuclear cells (PBMC), enabling tumor-immune-stromal interactions, and (2) 3D-bioprinted constructs formed using two bioinks: one incorporating tumor and tumor-microenvironment (TME) cells, and the other containing endothelial cells and pericytes to create perfusable vascular channels flowing PBMC and drugs. We are validating the ability of these high-throughput 3D models to recapitulate patient-specific tumor biology and predict responses to chemotherapy, immunotherapy, and targeted therapies. Their predictive performance is being evaluated in an IRB-approved clinical study (SMC-9417-22) involving 80 patients across seven cancer types. To guide personalized therapy selection, we integrate standard-of-care and investigational drugs with AI-derived treatment matches generated by ENLIGHT-DP (Pangea Biomed), a deep-learning platform that infers gene expression from tumor HandE slides and combines them with proprietary predictive biomarkers to produce individualized drug-response scores. AI-prioritized treatments are reviewed with oncologists and then tested on 3D platforms. Preliminary evidence suggests a correlation between the 3D tumoroid models and clinical outcomes. Notably, in a case of mucosal melanoma, standard therapies failed both clinically and ex vivo, whereas ENLIGHT-DP screening identified regorafenib, which demonstrated potent activity in the 3D model. Compassionate-use treatment led to a durable clinical response lasting nearly 12 months. In another metastatic melanoma case harboring an ALK rearrangement (identified via Tempus sequencing and prioritized by ENLIGHT-DP), lorlatinib demonstrated significant efficacy ex vivo and produced a sustained clinical response in the patient for more than 6 months at the time of this writing, with near-complete responses of visceral and brain metastases. Together, these patient-derived 3D models, integrated with AI-based drug prioritization, provide a robust platform for functional precision oncology, enabling personalized drug screening, reducing ineffective treatments, and bridging the gap between preclinical modeling and clinical response. Citation Format: Anshika Katyal, Anne Krinsky, Opal Avramoff, Yulia Liubomirski, Gal Dinstag, Omer Tirosh, Ranit Aharonov, Tuvik Beker, Iris Barshack, Shaked Lev-Ari, Shirly Grynberg, Ronnie Shapira-Frommer, Ronit Satchi-Fainaro. Patient-derived 3D tumor models integrated with AI-driven treatment matching for target discovery and personalized therapy [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 2 (Late-Breaking, Clinical Trial, and Invited Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(8_Suppl):Abstract nr LB334.

Context.—:Eosinophilic esophagitis (EoE) is a chronic, immune-mediated condition with overlapping histologic features with gastroesophageal reflux disease and other esophagitis subtypes. Current diagnostic criteria based on eosinophil density and basal zone hyperplasia lack specificity, necessitating more reliable tissue biomarkers. Objective.—:To evaluate the diagnostic utility of signal transducer and activator of transcription 6 (STAT6) immunohistochemistry (IHC) in distinguishing EoE from histologic mimics in esophageal biopsies. Design.—:STAT6 IHC was performed on 208 esophageal biopsy specimens, including cases of EoE, treated EoE, reflux esophagitis, other esophagitis, and normal mucosa. A blinded gastrointestinal pathologist scored basal epithelial nuclear staining. Diagnostic performance was assessed using predefined and optimized thresholds. Correlations with eosinophil density and basal zone hyperplasia were analyzed using Spearman correlation and multivariable linear regression. Results.—:Among 151 well-characterized cases, a 50% nuclear staining threshold yielded 95.6% sensitivity and 97.2% specificity for EoE. An optimized 32.5% threshold achieved 100% sensitivity and 95.3% specificity. STAT6 expression strongly correlated with eosinophil density (ρ = 0.80) and basal zone hyperplasia (ρ = 0.69), but was not entirely explained by either, indicating STAT6 reflects distinct T helper type 2-mediated epithelial activity. Residual staining was observed in some treated EoE cases despite histologic remission. Conclusions.—:STAT6 IHC is a robust, specific biomarker for EoE that offers high diagnostic accuracy and reflects underlying T helper type 2-driven epithelial signaling. Its application may improve diagnostic precision and support future efforts in disease monitoring and therapeutic response assessment.

CONTEXT:Pancreatic ductal adenocarcinoma arising in carriers of germline BRCA1 or BRCA2 mutations represents a molecularly defined subgroup characterized by defective homologous recombination DNA repair, yet its histologic phenotype remains incompletely characterized. OBJECTIVE:To characterize the histologic spectrum of resected pancreatic ductal adenocarcinomas associated with germline BRCA mutations and to describe recurring morphologic features that may reflect underlying DNA-repair deficiency. DESIGN:A retrospective cohort study of surgically resected pancreatic ductal adenocarcinomas from patients with confirmed germline BRCA1 or BRCA2 mutations treated at a tertiary referral center. All hematoxylin-eosin-stained slides were jointly reviewed, and architectural, cytologic, stromal, and peritumoral features were assessed semiquantitatively. RESULTS:Thirty-two tumors with residual carcinoma were evaluable. Tumors demonstrated marked morphologic heterogeneity, including vacuolated morphology (21/32, 65.6%), clear-cell areas (12/32, 37.5%), large-duct pattern (8/32, 25.0%), and micropapillary growth (9/32, 28.1%). A high gland-to-stroma ratio was present in 21 of 32 cases (65.6%), accompanied by stromal hyalinization (23/32, 71.9%) and myxoid change (28/32, 87.5%). Marked nuclear pleomorphism (29/32, 90.6%) and loss of polarity (30/32, 93.8%) were common. Peritumoral lymphoid follicles were identified in 20 of 32 tumors (62.5%). CONCLUSIONS:BRCA-associated pancreatic ductal adenocarcinomas demonstrate recurrent morphologic patterns, including increased glandularity, stromal hyalinization, myxoid stromal change and architectural heterogeneity. These observations provide a framework for future comparative studies investigating morphologic correlates of homologous recombination deficiency in pancreatic cancer. Recognition of these patterns may support consideration of germline or somatic DNA-repair gene testing in appropriate clinical settings.

Immune checkpoint inhibitors (ICIs) have improved cancer outcomes; however, many patients fail to respond, highlighting the need for novel targets. HVEM (Herpes Virus Entry Mediator) is an immune regulator with both inhibitory and stimulatory functions, making it a promising therapeutic candidate. We have developed Anti-4CB1, a fully human monoclonal antibody (mAb) that selectively blocks HVEM interactions with BTLA and CD160. Its activity was evaluated in-vitro using human tumor-infiltrating lymphocytes (TILs), peripheral blood mononuclear cells (PBMCs), and M1 macrophages, as well as in ex-vivo patient-derived tumor samples and in-vivo transgenic and humanized mouse models. HVEM expression was also assessed in serum and tumor tissues. Anti-4CB1 enhanced T-cell activation and cytotoxicity, evidenced by increased tumor cell killing, upregulation of activation markers (41BB, CD107a), and elevated IFNγ and TNFα secretion. It also promoted macrophage-mediated phagocytosis. In ex-vivo analyses of 49 patient-derived tumor samples, Anti-4CB1 increased cytotoxicity in 28.5% of cases, including samples unresponsive to anti-PD1. In-vivo, Anti-4CB1 demonstrated significant anti-tumor activity as monotherapy and showed enhanced efficacy in combination with anti-PD1. Additionally, higher tumor HVEM expression correlated with improved response to checkpoint blockade, while elevated soluble HVEM levels were associated with reduced responsiveness to Anti-HVEM. Anti-4CB1 enhances both adaptive and innate anti-tumor immunity and shows activity in anti-PD1 resistant settings. These findings support its potential as a novel therapeutic agent and suggest HVEM as a predictive biomarker for immunotherapy response.

Significance:Multiplexed biomarker assessment is increasingly required in oncologic diagnostics, yet tissue depletion from sequential immunohistochemistry (IHC) and molecular testing limits routine pathology. Many multiplex immunofluorescence platforms rely on iterative staining or complex instrumentation, restricting scalability and clinical adoption. Aim:The aim is to develop and clinically evaluate a hyperspectral multiplexed fluorescence imaging platform for simultaneous detection of multiple diagnostic biomarkers on a single formalin-fixed paraffin-embedded (FFPE) tissue section potentially compatible with routine workflows. Approach:A Fourier transform hyperspectral system incorporating a monolithic Sagnac interferometer was engineered for single-shot, full-spectrum acquisition without filter switching or iterative staining. A seven-biomarker lung cancer panel was applied in a single staining and imaging workflow, and custom spectral reconstruction and unmixing generated spatially resolved biomarker maps. Clinical performance was assessed retrospectively through blinded comparison with single-plex chromogenic IHC. Results:The system achieved submicron spatial resolution with minimal spectral cross-talk, enabling robust seven-channel imaging. Single-section analysis showed complete diagnostic concordance with reference IHC across the evaluated cohort of non-small cell lung cancer and other thoracic specimens and revealed biomarker co-localization, including tumor-associated PD-L1. Conclusions:This platform enables high-content diagnostic profiling from a single FFPE tissue section while conserving tissue and preserving spatial information, supporting further evaluation of its potential integration into routine clinical pathology workflows.

center dot Context.-Eosinophilic esophagitis (EoE) is a chronic, immune-mediated condition with overlapping histologic features with gastroesophageal reflux disease and other esophagitis subtypes. Current diagnostic criteria based on eosinophil density and basal zone hyperplasia lack specificity, necessitating more reliable tissue biomarkers. Objective.-To evaluate the diagnostic utility of signal transducer and activator of transcription 6 (STAT6) immunohistochemistry (IHC) in distinguishing EoE from histologic mimics in esophageal biopsies. Design.-STAT6 IHC was performed on 208 esophageal biopsy specimens, including cases of EoE, treated EoE, reflux esophagitis, other esophagitis, and normal mucosa. A blinded gastrointestinal pathologist scored basal epithelial nuclear staining. Diagnostic performance was assessed using predefined and optimized thresholds. Correlations with eosinophil density and basal zone hyperplasia were analyzed using Spearman correlation and multivariable linear regression.Results.-Among 151 well-characterized cases, a 50% nuclear staining threshold yielded 95.6% sensitivity and 97.2% specificity for EoE. An optimized 32.5% threshold achieved 100% sensitivity and 95.3% specificity. STAT6 expression strongly correlated with eosinophil density (q = 0.80) and basal zone hyperplasia (q = 0.69), but was not entirely explained by either, indicating STAT6 reflects distinct T helper type 2-mediated epithelial activity. Residual staining was observed in some treated EoE cases despite histologic remission. Conclusions.-STAT 6 IHC is a robust, specific biomarker for EoE that offers high diagnostic accuracy and reflects underlying T helper type 2-driven epithelial signaling. Its application may improve diagnostic precision and support future efforts in disease monitoring and theraassessment.

Non-small cell lung cancer (NSCLC) patient management relies on molecular analysis to determine eligibility for targeted therapy. Furthermore, neoadjuvant immunotherapy is primarily suitable in the absence of specific genomic alterations. However, significant challenges remain, including suboptimal molecular testing and patients being assigned to non-optimal treatment strategies. Here, we present AI classifiers for the identification of EGFR, ALK, BRAF and MET alterations directly from hematoxylin and eosin (H&E)-stained tissue using CanvOI 1.1, a digital pathology foundation model. Their performance was evaluated on an independent validation dataset of 968 NSCLC samples. The classifiers achieved AUCs of 0.87 for EGFR, 0.96 for ALK, 0.88 for BRAF and 0.83 for MET. Moreover, they demonstrated high accuracy in identifying cases lacking alterations. Our results highlight the potential of deep-learning tools for the detection of NSCLC biomarkers and specifically the identification of tumors without EGFR or ALK driver alterations, supporting more informed clinical decision-making.

The OncotypeDX 21-gene assay guides adjuvant chemotherapy decisions in early-stage, hormone receptor-positive, HER2-negative breast cancer, but cost and turnaround time limit access. This study presents a deep learning-based approach for predicting OncotypeDX recurrence scores directly from hematoxylin and eosin-stained whole slide images. Our approach leverages a deep learning foundation model pre-trained on 171,189 slides via self-supervised learning, which is fine-tuned for our task. The model was developed and validated using five independent cohorts, out of which three are external. On the two external cohorts that include OncotypeDX scores, the model achieved an AUC of 0.836 and 0.817, and identified 22% and 16.3% of the patients as low-risk with sensitivity of 0.97 and 0.97 and negative predictive value of 0.97 and 0.96, showing strong generalizability despite variations in staining protocols and imaging devices. Kaplan-Meier analysis demonstrated that patients classified as low-risk by the model had a significantly better prognosis than those classified as high-risk, with a hazard ratio of 4.1 (P < 0.001) and 2.0 (P < 0.01) on the two external cohorts that include patient outcomes. This artificial intelligence-driven solution offers a rapid, cost-effective, and scalable alternative to genomic testing, with the potential to enhance personalized treatment planning, especially in resource-constrained settings.

BACKGROUND:Histological assessment of mucosal biopsies in patients with ulcerative colitis (UC) can determine the activity and extent of disease and assess response to treatment. However, its widespread adoption is limited by the time required for the advanced GI specialty training to handle and review histopathology digital images, inter- and intra-observer variability, and cost associated with the interpretation of data. Artificial intelligence, and specifically machine learning‒driven medical image processing, have emerged to help standardise and automate histopathologic assessments. METHODS:In this global study conducted with participation of 38 sites in 19 countries (global roll-out phase), we collected histopathologic slides prepared from biopsy samples from patients with UC to train an AI model to recognise various cell types and assign a disease activity score based on the Nancy histological index (NHI). Results were compared with findings from a previous iteration of the machine learning model (pilot roll-out phase). RESULTS:In total, 850 tiles were analysed and used for training, validation, and testing. Model quality, assessed using the Nancy metric, improved from 61.50% in the pilot roll-out phase to 74.82% in the current global roll-out phase. Cell detection quality (F1-score metric) also increased from 27.50% (pilot roll-out) to 58.80% (global roll-out). CONCLUSIONS:In this global roll-out, the quality of the AI model was significantly improved for both NHI scores and cell detection. Further development and implementation of the model at the participating international sites continues and may lead to a valuable and scalable tool for the analysis of disease activity in UC.

Over the last two decades, the diagnosis and treatment of breast cancer patients have improved considerably. However, brain metastases remain a major clinical challenge and a leading cause of mortality. Thus, a better understanding of the pathways involved in the metastatic cascade is essential.To this end, we have investigated the reciprocal effects of astrocytes and breast cancer cells, employing traditional 2D cell culture and our unique 3D multicellular tumouroid models.Our findings revealed that astrocytes enhance the proliferation, migration and invasion of breast cancer cells, suggesting a supportive role for astrocytes in breast cancer outgrowth to the brain. Elucidating the key players in astrocyte-breast cancer cells crosstalk, we found that CCL2 is highly expressed in breast cancer brain metastases tissue sections from both patients and mice. Our in vitro and in vivo models further confirmed that CCL2 has a functional role in brain metastasis. Given their aggressive nature, we sought additional immune checkpoints for rationale combination therapy. Among the promising candidates were the adhesion molecule P-selectin, which we have recently shown to play a key role in the crosstalk with microglia cells and the co-inhibitory receptor PD-1, the main target of currently approved immunotherapies. Finally, combining CCL2 inhibition with immunomodulators targeting either PD-1/PD-L1 or P-selectin/P-Selectin Ligand-1 axes in our human 3D tumouroid models and in vivo presented more favourable outcomes than each monotherapy.Taken together, we propose that CCL2-CCR2/CCR4 is a key pathway promoting breast cancer brain metastases and a promising target for an immunotherapeutic combination approach. Brain metastasis in breast cancer remains a significant clinical challenge. Israeli Dangoor et al. investigate the pathways driving metastasis, focusing on the interaction between astrocytes and breast cancer cells, and propose a new combination immunotherapy approach targeting the CCL2 and PD-1/P-selectin axes.

Primers and Antibodies used in this paper.