Artificial intelligence (AI) has achieved expert-level performance in medical imaging, but its clinical translation requires a reliable and interpretable system design. Decision-level hierarchical cascaded pipelines are widely used in medical AI, yet their benefits and limitations require further investigation. We compared decision-level hierarchical cascaded and parallel architectures in medical imaging and revealed a trade-off between error vulnerability and diagnostic transparency. Based on these observations, we developed an error-aware framework integrating automated error detection, adaptive correction, and confidence-guided human-AI collaboration. The framework was evaluated on over 10,000 patient cases, including 6,103 colorectal pathology and 7,164 cervical cytology samples, with additional validation in endoscopic and ophthalmological imaging. Decision-level hierarchical cascaded systems, although susceptible to error amplification, were more aligned with hierarchical clinical reasoning and enabled more transparent decision pathways. The proposed framework mitigated error propagation through automated detection and correction. Combined with confidence-guided human-AI collaboration, it achieved absolute accuracy gains of 16.12 and 10.11 percentage points in colorectal pathology and cervical cytology, respectively, compared with parallel systems, with consistent gains across other imaging tasks. Our study highlights the trade-off between error propagation and interpretability in hierarchical cascaded AI systems and demonstrates that error-aware design can improve their robustness and clinical applicability, contributing to the development of more reliable and human-centered medical AI.

BackgroundApproximately 20% of patients with stage II colorectal cancer (CRC) experience tumor relapse despite standard surgical treatment. Histopathological analysis holds promise for postsurgical risk stratification and guiding adjuvant chemotherapy (ACT) decisions. The aim of this study was to use deep learning to extract explainable tissue biomarkers from whole-slide images.Methods and findingsIn this retrospective cohort study, we developed and validated SurvFinder, an interpretable deep learning framework designed to autonomously identify tissue-based risk biomarkers from hematoxylin and eosin (H&E)-stained slides. The framework aims to support individualized risk stratification and explore associations with treatment outcomes. The present study included 6,950 H&E slides from 1,604 patients with stage II CRC across four independent cohorts in China. Patients were enrolled from 2012 to 2018 and followed for a minimum of 24 months. The primary outcome of the study was relapse-free survival (RFS). Our analyses identified tertiary lymphoid structures (TLSs) as critical prognostic features in stage II CRC. The multi-view integration of TLS characteristics by SurvFinder consistently demonstrated superior predictive and prognostic accuracy across four multicenter datasets (AUROC with 95% confidence interval [CI]: 0.827 [0.789,0.864], 0.805 [0.749,0.860], 0.805 [0.748,0.861], and 0.712 [0.621,0.804]), surpassing traditional clinical prognostic parameters (hazard ratio [HR]: 8.23, 95% CI: 5.43-12.47; p < 0.001). Using explainable AI (XAI) methods, we ensured model transparency and identified key TLS features-such as their location at the tumor periphery and their maturity state-as significant factors influencing prognosis and the efficacy of adjuvant therapy. The retrospective design without prospective validation and real-world clinical deployment is the main limitation of this study.ConclusionsTogether, these results highlight the potential utility of deep learning-based histopathological analysis for automated risk stratification in stage II CRC. In particular, our findings support the relevance of TLSs as a histological biomarker with potential implications for personalizing ACT decisions.

BTAF1, an ATP-dependent remodeler of the TBP-DNA complex, is frequently mutated in gastric cancer. However, its role in DNA repair and therapeutic relevance remains largely undefined. Here, we show that BTAF1 knockout leads to accumulation of double-strand breaks (DSBs) by impairing DNA end-resection process of homologous recombination (HR) repair, thereby sensitizing cells to genotoxic agents both in vitro and in vivo. Mechanistically, BTAF1 prevents ubiquitin-mediated degradation of MRE11, maintaining its protein stability, promoting DNA end resection and HR, and consequently enhancing cellular resistance to DNA-damaging stress. Notably, the interaction between BTAF1 and MRE11 is dynamically regulated by PARP1-mediated PARylation of BTAF1 during the DNA damage response. Loss of BTAF1 also increases chemosensitivity in gastric cancer xenograft and organoid models. Clinically, high BTAF1 expression correlates with poor prognosis in gastric cancer patients receiving neoadjuvant chemotherapy. Collectively, our findings identify BTAF1 as a critical regulator of HR repair through stabilization of MRE11 and propose BTAF1 as a potential biomarker for predicting response to genotoxic chemotherapy. BTAF1 is frequently mutated in gastric cancer, yet its role in DNA repair remains unclear. BTAF1 promotes homologous recombination and chemoresistance in gastric cancer by stabilizing MRE11. BTAF1 loss impairs DNA end resection, sensitizing cells to genotoxic agents. The BTAF1-MRE11 interaction is regulated by PARP1 mediated PARylation.

Background and Purpose To assess the efficacy and safety of cytoreductive surgery combined with hyperthermic intraperitoneal chemotherapy (CRS-HIPEC) in patients with inadvertently morcellated uterine leiomyosarcoma (uLMS). Materials and Methods We conducted a retrospective analysis using prospectively collected data from the HIPEC PROGRAM Registry. Eligible patients had undergone CRS-HIPEC for inadvertently morcellated uLMS. Progression-free survival (PFS) and overall survival (OS) were using Kaplan-Meier methods. Cox proportional hazards regression was performed to identify prognostic factors. Results Nineteen patients were included, with a median interval from morcellation to CRS-HIPEC (TM-CRS) of 22 days (range: 11–51). Following CRS, all patients received HIPEC with a combination of docetaxel (75 mg/m2) and gemcitabine (1000 mg/m2). The median follow-up time was 40 months (range: 9–66 months). The 6-month, 1-year, 2-year, 3-year, and 5-year PFS rates were 94.7%, 89.5%, 84.2%, 84.2%, and 84.2%, respectively. The corresponding OS rates were 100%, 94.7%, 89.2%, 89.2%, and 89.2%, respectively. TM-CRS was the only factor significantly associated with recurrence risk. Receiver operating characteristic analysis identified 23 days as the optimal cutoff, with patients treated within this interval demonstrating significantly longer PFS. One patient experienced a grade 3 surgical site infection; no other serious adverse events were reported. Conclusions CRS-HIPEC is a safe and potentially effective treatment for inadvertently morcellated uLMS. Survival benefits may be greater when CRS-HIPEC is performed within 23 days of morcellation.

Intraoperative pathology is pivotal to precision surgery, yet its clinical impact is constrained by diagnostic complexity and the limited availability of high-quality frozen-section data. While computational pathology has made significant strides, the lack of large-scale, prospective validation has impeded its routine adoption in surgical workflows. Here, we introduce CRISP, a clinically oriented foundation model developed on over 100,000 frozen sections from ten medical centers, specifically designed to provide Clinically-oriented Robust Intraoperative Support for Pathology (CRISP). CRISP was comprehensively evaluated on more than 15,000 intraoperative slides across nearly 100 retrospective diagnostic tasks, including benign-malignant discrimination, key intraoperative decision-making, and pan-cancer detection, etc. The model demonstrated robust generalization across 6 institutions, 14 tumor types, and 24 anatomical sites-including previously unseen sites and rare cancers. In a prospective cohort of over 3,000 patients, CRISP sustained high diagnostic accuracy under real-world conditions, directly informing surgical decisions in 92.6% of cases. Human-AI collaboration further reduced diagnostic workload by 35%, avoided 105 ancillary tests and enhanced detection of micrometastases with 87.5% accuracy. Together, these findings suggest that CRISP represents a clinically oriented approach to AI-driven intraoperative pathology, with the potential to support surgical decision-making and facilitate the translation of computational methods into clinical practice. CRISP, a vision-based pathology foundation model developed exclusively from frozen section slides, supports treatment decision-making throughout the surgical workflow with superior performance to current foundation models and extensive validation, including in a prospective cohort.

Liver metastasis is the primary cause of mortality in colorectal cancer (CRC) patients. To decipher the underlying mechanisms, we performed single-cell RNA sequencing (scRNA-seq) on paired primary colorectal tumors, adjacent tissues and liver metastases from three CRC liver metastasis (CRLM) patients, alongside colorectal tumors and adjacent tissues from three non-metastatic CRC patients. Our analysis revealed a significant enrichment of Enolase 2-expressing (ENO2⁺) cancer cells in CRLM patients compared to their non-metastatic counterparts. Functional characterization, supported by bioinformatics and murine models, demonstrated that ENO2⁺ cancer cells exhibit enhanced epithelial-mesenchymal transition (EMT) and are critical drivers of CRLM. Mechanistically, the ENO2 protein directly binds to macrophage migration inhibitory factor (MIF) within cancer cells, stabilizing MIF by inhibiting its C-terminus of Hsc70-Interacting Protein (CHIP)-mediated ubiquitination and degradation. This ENO2-MIF interaction activates MIF signaling, fostering robust tumor cell-macrophage crosstalk that promotes M2 macrophage polarization, which is validated by spatial transcriptomics showing the colocalization of ENO2⁺ cancer cells and M2 macrophages. Crucially, both organoid and in vivo models confirmed that ENO2 in CRC cells is essential for inducing M2 macrophage polarization via the MIF pathway, thereby facilitating liver metastasis. Knockout of ENO2 significantly suppressed tumor growth and liver metastasis in mouse models. An inhibitor of the ENO2-MIF interaction, pyrithioxin, can effectively reduce the burden of liver metastasis in mice. Collectively, our findings identify ENO2 as a key driver of CRLM by stabilizing MIF to orchestrate M2 macrophage polarization, highlighting the ENO2-MIF axis as a promising therapeutic strategy for CRLM.

Metabolic reprogramming is a hallmark of cancer that promotes tumor progression and immune evasion. Here, we identify a NIPAL1-driven metabolic-epigenetic circuit in esophageal squamous cell carcinoma (ESCC) that facilitates tumor growth and suppresses antitumor immunity. Mechanistically, NIPAL1 recruits the tyrosine kinase HCK to phosphorylate LDHA at Y10, enhancing glycolysis and lactate production. Lactate accumulation promotes p300-mediated histone H3K18 lactylation (H3K18la), which transcriptionally activates NIPAL1 expression, establishing a self-sustaining NIPAL1-HCK-p-LDHA-lactate-p300-H3K18la loop. This axis functions independently of NIPAL1's canonical magnesium transporter activity and promotes immune escape by impairing CD8+ T cell function. Pharmacological inhibition of HCK or p300 disrupts this loop and restores antitumor immunity, sensitizing tumors to anti-PD-1 therapy. Clinically, expression of NIPAL1, p-LDHA (Y10), and H3K18la correlates with response to immune checkpoint blockade. Our findings reveal a previously unrecognized NIPAL1-HCK-H3K18la signaling loop that integrates tumor metabolism to immune regulation, offering promising targets to improve immunotherapy efficacy in ESCC.

Purpose: Cytology is a cornerstone of pathologic diagnosis. However, the use of artificial intelligence (AI) models for cytology-based diagnostics remains constrained by limited data availability and stringent privacy regulations. This study aims to develop COIN, a controllable cytology image generation foundation model, to address these challenges by synthesizing high-quality cytology images to enhance AI diagnostics and support clinical applications.Experimental Design: The COIN model was trained on a large-scale dataset of 112,226 cytology image-report pairs from 16 anatomic sites. Using diagnostic textual reports, it generates high-fidelity cytology images with morphologically and semantically coherent features. Expert cytologists evaluated the generated images for anatomic and diagnostic authenticity. The model's utility was assessed through data augmentation experiments, AI model training under data-scarce conditions, and content-based image retrieval applications.Results: Expert evaluations confirmed the high anatomic and diagnostic fidelity of the images generated by COIN. When used for data augmentation, COIN significantly improved the performance of diagnostic AI models across various tasks. Under data-scarce conditions, models trained exclusively on COIN-generated images demonstrated effective generalization to real-world datasets. Furthermore, COIN supported content-based image retrieval, offering a novel tool for case referencing and clinical decision support.Conclusions: COIN represents a robust and privacy-preserving framework for scalable cytology data generation. Its ability to synthesize realistic images and enhance AI diagnostics highlights its broad applicability in computational pathology, providing a valuable tool to accelerate the development and implementation of AI-based diagnostic solutions.

Esophageal squamous cell carcinoma (ESCC) is a highly lethal cancer with limited therapeutic options and frequent relapse. RNA acetylation has emerged as a crucial regulator of tumor biology. In this study, we identify circTHBS1 as an oncogenic circRNA that promotes autophagy and drives ESCC progression. Mechanistically, circTHBS1 interacts with the N-acetyltransferase 10 (NAT10), preventing its ubiquitin-mediated proteasomal degradation and thereby increasing global N4-acetylcytidine (ac4C) modification. Rab8a is identified as a key ac4C target, where ac4C modification at a conserved CGCCAG motif within the coding sequence facilitates NAT10 binding and enhances translation. Disruption of this modification impairs Rab8a-driven autophagy and suppresses tumor progression. Collectively, our findings reveal a circTHBS1/NAT10-ac4C/Rab8a axis that reprograms translational control and autophagy, providing new mechanistic insights and potential therapeutic targets for ESCC.

Transposable elements (TEs), which constitute nearly half of the human genome, have long been regarded as genomic "dark matter". However, their reactivation in tumor cells, resulting in the production of TE-chimeric transcripts (TCTs), has emerged as a potential driver of cancer progression. The complexity and full extent of these transcripts remain elusive, largely due to the limitations of short-read next-generation sequencing technologies. These methods have struggled to comprehensively capture the diversity and structure of TCTs, particularly those involving short interspersed nuclear elements (SINEs) or closely co-transcribed TEs. Leveraging full-length cDNA sequencing technology based on nanopore sequencing platform, we developed a customized pipeline for identifying and quantifying TCTs in 19 lung adenocarcinoma (LUAD) cell lines. The short-read RNA-seq dataset from a LUAD corhort ( 200 tumor samples) was employed to validate the identified TCTs and explore their association with tumor progression. To assess the functional roles of a specific TCTs, cell migration and cell proliferation assays were performed. We uncovered 208 unique TCT candidates in the LUAD cell lines. Our approach allowed for the identification of cryptic promoters and terminators within non-transposing TEs. Notably, we identified a chimeric transcript involving MIR_HKDC1, which appears to play a significant role in the progression of LUAD. Furthermore, the expression of these TCTs were associated with poor clinical outcomes in a cohort of LUAD patients, suggesting their potential as novel biomarkers for both LUAD progression and prognosis. Our study underscores the application of long-read sequencing to unravel the complex landscape of TCTs in LUAD. We provide a comprehensive characterization of TCTs in LUAD, exploring their potential regulatory roles in cancer progression. These findings contribute to a deeper understanding of the genomic intricacies underlying cancer, and offer new directions for the development of targeted therapies and personalized treatment strategies for LUAD. This research highlights the potential of TCTs as both biomarkers and therapeutic targets in the oncogenesis, offering new insights into the interplay between transposable elements and gene regulation in cancer.

BACKGROUND:Ferroptosis, a regulated form of cell death driven by lipid peroxidation and iron dysregulation, plays a critical role in tumor suppression. The expression of certain genes determine ferroptosis sensitivity and may serve as biomarkers to identify patients who could benefit from ferroptosis-promoting therapies. The role of Methyltransferase-like (METTL) family proteins in colorectal cancer (CRC) tumorigenesis has gained increasing attention, yet the functional relevance of METTL21D (also known as VCPKMT) remains largely unexplored. METHODS:CCK8 assays, colony formation assays, and xenograft tumor experiments were used to investigate the role of VCPKMT in proliferation and ferroptosis sensitivity in CRC. RNA sequencing, IP-MS, and RT-qPCR were used to identify the relevant genes of VCPKMT. Co-IP and subcellular fractionation assays, western blotting and immunofluorescence (IF) staining were used to explore the underlying molecular mechanisms of actions of VCPKMT. RESULTS:Herein, we identified that VCPKMT is downregulated in CRC tissues and lower VCPKMT expression correlates with ferroptosis resistance and poorer prognosis. Mechanistically, VCPKMT-induced VCP methylation facilitates its nuclear translocation, a process that enhances VCP's interaction with HDAC1. This interaction promotes the degradation of HDAC1 via the ubiquitin-proteasome pathway, leading to reduced expression of FTH1, a key inhibitor of ferroptosis. Combined targeting of HDAC1 and ferroptosis synergistically suppressed the growth of tumors with low VCPKMT expression. CONCLUSIONS:Our findings reveal a novel role for VCPKMT as a regulator of ferroptosis sensitivity in CRC, highlighting the VCPKMT-VCP-HDAC1 axis as a potential therapeutic target for CRC treatment.

[This corrects the article DOI: 10.1186/s40824-022-00251-z.].

[This corrects the article DOI: 10.7150/thno.20942.].

Cancer-associated fibroblasts (CAF) are pivotal constituents of the tumor microenvironment that significantly influence cancer aggressiveness through the secretion of various factors. A more detailed characterization of the specific secretions exclusive to CAFs that drive tumor progression could identify potential targets to perturb this intracellular cross-talk. In this study, we identified latent TGFβ-binding protein 2 (LTBP2) as a unique protein secreted exclusively by esophageal squamous cell carcinoma (ESCC) CAFs that promotes metastasis and chemoresistance. LTBP2 exerted its oncogenic effects by interacting with integrin α6β4, which serves as a functional receptor, and thereby activating Src signaling in ESCC cells. Notably, targeting LTBP2 with specific antagonistic antibodies markedly increased the susceptibility of ESCC cells to chemotherapeutic agents. These findings highlight the pivotal role of LTBP2 as a crucial mediator of CAF-induced cancer cell aggression and introduce it as a promising target to enhance chemotherapeutic efficacy in ESCC. SIGNIFICANCE:CAF-secreted LTBP2 binds integrin α6β4 and activates Src signaling to drive metastasis and chemoresistance in esophageal cancer, highlighting LTBP2 as a key regulator of CAF-mediated tumor progression that can be therapeutically targeted.

Intraoperative pathology is pivotal to precision surgery, yet its clinical impact is constrained by diagnostic complexity and the limited availability of high-quality frozen-section data. While computational pathology has made significant strides, the lack of large-scale, prospective validation has impeded its routine adoption in surgical workflows. Here, we introduce CRISP, a clinical-grade foundation model developed on over 100,000 frozen sections from eight medical centers, specifically designed to provide Clinical-grade Robust Intraoperative Support for Pathology (CRISP). CRISP was comprehensively evaluated on more than 15,000 intraoperative slides across nearly 100 retrospective diagnostic tasks, including benign-malignant discrimination, key intraoperative decision-making, and pan-cancer detection, etc. The model demonstrated robust generalization across diverse institutions, tumor types, and anatomical sites-including previously unseen sites and rare cancers. In a prospective cohort of over 2,000 patients, CRISP sustained high diagnostic accuracy under real-world conditions, directly informing surgical decisions in 92.6% of cases. Human-AI collaboration further reduced diagnostic workload by 35%, avoided 105 ancillary tests and enhanced detection of micrometastases with 87.5% accuracy. Together, these findings position CRISP as a clinical-grade paradigm for AI-driven intraoperative pathology, bridging computational advances with surgical precision and accelerating the translation of artificial intelligence into routine clinical practice.

Neoadjuvant immunochemotherapy (nICT) has significantly improved the treatment of locally advanced esophageal cancer (EC), yet accurately identifying patients' response remains a major challenge. In this study, we introduce eSPARK, a multimodal framework designed to integrate routinely available clinical data for informed decision-making in nICT treatment for EC. The model is developed using 344 patients from three independent regions, each with pre-treatment-paired computed tomography (CT) imaging and pathological slides, and postoperative pathological complete response (pCR) outcomes. By incorporating cytological semantic information, eSPARK demonstrates superior generalizability, outperforming single-modality models and achieving robust predictive accuracy across multicenter datasets. Additionally, a multi-scale interpretability module identifies several biomarkers, including the neutrophil-to-lymphocyte ratio (NLR) in the tumor microenvironment, associated with nICT response. Our findings underscore the potential of eSPARK as a powerful tool for personalized therapeutic decision-making in locally advanced EC and its broader implications for advancing precision oncology through multidisciplinary data integration.

BACKGROUND:Intraductal papillary mucinous neoplasms (IPMNs) are known precancerous lesions whose malignant transformation to pancreatic ductal adenocarcinoma (PDAC) worsens prognosis. Current experimental models inadequately elucidate the molecular mechanisms driving this progression. METHODS:We performed spatial transcriptomics (ST) on three fresh tissue samples from the same patient including normal pancreas, high-grade IPMN, and invasive PDAC. We used Spatial inferCNV to profile the evolutionary clone trajectory. We utilize SpaCET to identify the tumor region and adjacent tumor microenvironment (TME) feature variation, as well as its crosstalk between tumor cells by CellChat. RESULTS:Our findings identified the master transcript factors and critical signaling pathways that might be involved in the progression of IPMN to invasive PDAC. Moreover, both IPMN and PDAC harbored the ELF3, MYC, and KLF4 amplification. We identified a significant heterogeneity among PDAC tissue with seven distinct subclones showing diverse functions, such as hypoxia, oxidative phosphorylation, and epithelial-mesenchymal Transition. Compared to IPMN, PDAC showed immune landscape remodeling: CD4+ T cells and dendritic cells depleted, while immunosuppressive M2 macrophages increased. In addition, cancer-associated fibroblasts (CAFs), particularly myofibroblastic CAFs, were enriched adjacent to invasive PDAC. CONCLUSIONS:Our study integrates multiple computational approaches for spatial transcriptomics to identify ELF3/MYC/KLF4 amplifications during IPMN-to-PDAC progression, as well as 7 heterogeneous subclones with functions including hypoxia and EMT, alongside immune-stromal landscape remodeling and tumor-adjacent myofibroblastic CAF enrichment.

Metabolic dysregulation has been implicated as a key factor in colorectal cancer (CRC) initiation, however, the underlying driving forces and mechanisms remain poorly understood. Herein, transcriptome profiling of paired early-stage CRCs and adenomas identifies Nudix hydrolase 13 (NUDT13) as a critical suppressor. Elevated NUDT13 expression impedes the proliferation of CRC cells under hypoxic conditions and markedly inhibits CRC initiation by upregulating PKM1. Mechanistically, NUDT13 directly binds and stabilizes PKM1 protein by reducing its poly ADP-ribosylation (PARylation), which is catalyzed by PARP1 at E275/D281/E282/E285/D296, thereby inducing an oxidative phosphorylation (OXPHOS) phenotype in CRC cells. Moreover, spatiotemporal knockout of Nudt13 enhances intestinal tumorigenesis in mice, which can be significantly suppressed by PARP1 inhibitor Olaparib. Notably, residues E245/E248/E249 within the Nudix box motif of NUDT13 are essential for PKM1 PARylation, and a mimic peptide derived from this motif is sufficient to stabilize PKM1 protein and robustly inhibit CRC tumorigenesis. Collectively, this study reveals a previously unknown PARylation-dependent mechanism that regulates PKM1 protein stability and switches the metabolic pathway of CRC cells, providing a promising target for CRC treatment.