Supplementary Figure S5. Gene expression profiles across different RNA clusters in this study cohort.

High-grade endometrial carcinoma exhibits marked histologic diversity, yet its molecular basis and the potential contribution of transcriptomic phenotyping to molecular classification remain incompletely understood. To address this, we performed whole-exome sequencing and RNA sequencing on 81 high-grade endometrial carcinomas, including serous, clear cell, grade 3 endometrioid, and carcinosarcoma. Tumors were assigned to molecular subtypes (POLE-ultramutated, microsatellite instability-high (MSI-H), TP53-mutated (TP53-mut), and no specific molecular profile (NSMP), based on The Cancer Genome Atlas and ProMisE frameworks. Transcriptomic phenotypes were identified by unsupervised clustering of gene expression and analyzed in relation to histology, molecular subtypes, immune-related gene expression, and clinical outcomes. In this context, substantial discordance was observed among TP53 mutation status, p53 immunohistochemistry (IHC), and copy number-based classification in non-POLE/non-MSI-H tumors. Transcriptomic clustering identified three phenotypic groups linked to cell differentiation status: glandular/luminal, ciliated, and epithelial-mesenchymal transition-like (EMT-like). These phenotypes transcended molecular subtype boundaries. For example, TP53-mut tumors were distributed across both glandular/luminal and EMT-like phenotypes. The glandular/luminal phenotype was associated with elevated antigen presentation (e.g., HLA expression) and immune-related signaling, whereas the EMT-like phenotype, frequently observed in carcinosarcoma, was linked to stemness and metastatic potential. TP53-mut and NSMP were associated with poor prognosis in high-grade endometrial carcinoma, whereas the glandular/luminal phenotype was associated with better outcomes than the EMT-like phenotype, an effect largely influenced by carcinosarcoma prevalence. Transcriptomic phenotypes complement molecular subtypes in high-grade endometrial carcinoma, enhancing biological resolution and capturing clinically relevant heterogeneity. These results underscore persistent challenges of current molecular classification approaches, supporting the need for integrative strategies in high-grade disease. SIGNIFICANCE:High-grade endometrial carcinomas show marked heterogeneity. We demonstrate substantial discordance among TP53 mutation status, p53 IHC, and copy number-based assignment, alongside transcriptomic phenotypes spanning molecular subtypes. These findings provide a foundation for developing improved classification approaches to better interpret the heterogeneous molecular landscape of high-grade disease.

5582 Background: Administration of 5-aminolevulinic acid (5-ALA) results in tumor-selective accumulation of protoporphyrin IX, enabling visualization of malignant lesions as red fluorescence under blue light illumination. Although 5-ALA–based photodynamic diagnosis (ALA-PDD) is well established for non-muscle invasive bladder cancer, its role in gynecologic malignancies remains insufficiently characterized. We conducted an investigator-initiated clinical study to evaluate the efficacy and safety of ALA-PDD in ovarian, fallopian tube, and primary peritoneal cancers. Methods: From October 2024 to May 2025, patients scheduled for cytoreductive surgery or exploratory laparotomy were enrolled. After exclusion of one patient due to a change in treatment strategy, 22 patients received oral 5-ALA hydrochloride (20 mg/kg) 3–8 hours prior to surgery. Diagnostic performance for detecting peritoneal dissemination under blue light (BL) was compared with that under conventional white light (WL), assessing sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). Statistical tests were performed using McNemar’s-test (sensitivity and specificity) or Chi-square test (PPV and NPV). Safety was evaluated based on the incidence and severity of adverse events. Results: Overall sensitivity was comparable between BL and WL (88.8% vs 80.9%), whereas BL demonstrated significantly higher specificity (67.3% vs 51.9%, p = 0.0229), indicating fewer false-positive findings. Notably, in patients undergoing interval debulking surgery (IDS; n = 9), BL significantly outperformed WL across all diagnostic metrics, including sensitivity, specificity, PPV, and NPV (all p < 0.05). These findings suggest that ALA-PDD enables more accurate intraoperative discrimination between malignant and non-malignant lesions following neoadjuvant chemotherapy. All patients experienced adverse events, predominantly Grade ≤2. One case of postoperative ileus was reported; however, no causal relationship with 5-ALA was identified, and no clinically relevant safety signals emerged. Conclusions: ALA-PDD appears to enhance intraoperative diagnostic accuracy compared with conventional white light inspection and demonstrates an acceptable safety profile in ovarian, fallopian tube, and primary peritoneal cancers. In IDS patients, ALA-PDD may serve as a clinically meaningful adjunct to achieve precise cytoreduction and support surgical decision-making in the neoadjuvant chemotherapy setting. Clinical trial information: jRCTs031240345.

Myoepithelioma-like tumor of the vulvar region (MELTVR) is a rare SMARCB1-deficient mesenchymal neoplasm of adult women that can mimic malignant vulvar sarcomas, particularly epithelioid sarcoma. Although loss of SMARCB1/INI1 expression is a defining feature, the comprehensive genomic landscape of MELTVR remains poorly characterized. We report two cases of MELTVR and performed integrated histopathologic, immunophenotypic, and molecular analyses, including whole-exome sequencing (WES) with copy number assessment and targeted RNA-based fusion testing using the Archer FusionPlex Sarcoma panel. Histologically, both tumors consisted of relatively uniform epithelioid to short spindle cells in solid nests and cords within focal myxoid stroma, with complete loss of INI1 and positivity for smooth muscle markers and focal ER/EMA expression. Genomic profiling demonstrated a quiet molecular background in both cases, with low tumor mutation burden (0.45 and 1.03 mut/Mb) and no pathogenic SNVs/indels in major cancer-associated genes. One case showed a focal homozygous deletion of the SMARCB1 locus at 22q11.2, whereas the other case exhibited INI1 loss without detectable SMARCB1 mutation or copy number loss, suggesting heterogeneous mechanisms of inactivation. CDKN2A copy number remained neutral in both tumors. No canonical sarcoma-associated gene rearrangements, including EWSR1, FUS, PLAG1, or NR4A3, were identified. Together with a review of previously reported cases, these findings support MELTVR as an SMARCB1-inactivated neoplasm with low genomic complexity and highlight the diagnostic value of NGS-based profiling in excluding malignant mimics and preventing overtreatment.

Differences in chromosome CN status by molecular subtype. A, Box plots showing the length of genomic regions with CNAs across different molecular subtypes or histologic subtypes. Statistical comparisons were performed using ANOVA followed by Tukey post hoc analysis. B, Box plots showing the number of amplified genes across different molecular subtypes or histologic subtypes. ANOVA P values are provided for each comparison. C, Heatmap showing the CN signatures of various chromosomal regions across different molecular subtypes. D, Kaplan–Meier curves for PFS according to CN status. E, Kaplan–Meier curves for PFS according to CN status for TP53-mut and NSMP patients. P values were calculated using the log-rank test. ANOVA, analysis of variance; APOBEC, apolipoprotein B mRNA-editing enzyme, catalytic polypeptide; G1/G2, grades 1–2 endometrioid; LOH, loss of heterozygosity.

SMARCA4- deficient undifferentiated uterine sarcoma (SDUS) is a rare, highly aggressive SWI/SNF-deficient malignancy that can morphologically overlap with undifferentiated endometrial carcinoma (UDEC). Although prior targeted sequencing studies have suggested that SDUS often has a "quiet genome," exome-wide corroboration remains limited. We report a 23-yr-old woman who presented with a >15 cm uterine mass and underwent hysterectomy with bilateral salpingo-oophorectomy, followed by rapid peritoneal dissemination and para-aortic metastases within 2 mo. The disease was refractory to gemcitabine/docetaxel and the patient died 6 mo postoperatively. Histologically, the tumor consisted of sheets of monomorphic undifferentiated epithelioid cells with rhabdoid features, focal phyllodes-like architecture entrapping benign endometrial glands and stromal hyalinization, and no identifiable carcinomatous component. Immunohistochemistry demonstrated complete loss of SMARCA4 (BRG1) and SMARCA2 (BRM) with retained SMARCB1 (INI1), a wild-type p53 pattern, intact mismatch repair protein expression, and negativity for claudin-4 and SOX2. Whole-exome sequencing with copy-number analysis of paired tumor and nontumor tissue identified biallelic SMARCA4 inactivation (somatic frameshift deletion c.3717del, p.W1239fs; VAF 0.89) with loss of heterozygosity at 19p13.2, microsatellite stability, and low tumor mutational burden (0.70 mut/Mb), without additional canonical driver mutations; notably, no TP53 mutation was detected despite hemizygous 17p loss. A focal copy-number gain involving the BRCA1 locus was also identified; however, its biological significance remains uncertain. This case provides exome-wide support for the "quiet genome" concept in SDUS and underscores its biological distinction from UDEC. Comprehensive genomic profiling may assist diagnosis in this aggressive malignancy.

Supplementary Figure S3. Representative results of immunohistochemical staining.

Early detection of epithelial ovarian cancer (EOC) remains a major clinical challenge. Although serum tumor markers are widely used for detection, their diagnostic performance remains limited. We previously developed a comprehensive serum glycopeptide spectrum analysis (CSGSA) approach that integrates tumor marker measurements and enriched glycopeptides (EGPs) using convolutional neural networks. In this study, we evaluated whether a two-step LightGBM framework incorporating cancer antigen 125 (CA125), human epididymis protein 4 (HE4), cancer antigen 72 − 4 (CA72-4), and EGPs could improve the diagnostic specificity and projected positive predictive value (PPV) for EOC detection compared with conventional biomarker-based approaches. The study included 553 patients with EOC and 1,144 non-EOC controls (healthy individuals or patients with benign conditions). Serum levels of CA125, HE4, and CA72-4 were measured along with 1,712 EGPs. Diagnostic models were developed using machine learning algorithms and evaluated for accuracy, area under the receiver operating characteristic curve (ROC-AUC), PPV, and negative predictive value (NPV). The highest diagnostic performance was achieved using a two-step classification framework. First, patients were stratified into high-, intermediate-, and low-risk groups based on tumor markers and age. Second, the intermediate-risk group was reclassified using a model incorporating EGP-derived features. Among the evaluated algorithms, LightGBM achieved the best performance, yielding a prevalence-adjusted (projected) PPV of 18.7

Integration of transcriptomic phenotypes with molecular and histologic classifications in high-grade endometrial carcinoma. A, The t-SNE plot shows the distribution of samples across different molecular subtypes. B, The t-SNE plot shows the distribution of samples across different histologic types. C, The relationship between molecular subtypes and phenotypes (RNA clusters) is shown in association with histologic subtypes (distinguished by color). Each colored block represents a specific histologic subtype: blue, serous carcinomas; green, clear cell carcinomas; light green, clear endometrioid or clear serous; red, carcinosarcoma; purple, EMG3; and gray, other. t-SNE, t-distributed stochastic neighbor embedding.

Supplementary Figure S9. Analysis of immunological status in the TCGA dataset.

Supplementary Table S9. Gene set enrichment analysis statistics for representative pathways.

Clustering of high-grade endometrial carcinomas based on gene expression profiles. A, Principal component analysis plot of transcriptome expression showing the distribution of samples in association with their histologic type. B, The t-SNE plot shows the results of hierarchical clustering, indicating RNA cluster classifications. C, Box plots of gene expression levels for selected genes across RNA clusters. The y-axis represents the log2(normalized count + 1) of gene expression. P values are provided for comparisons based on ANOVA followed by Tukey post hoc analysis. D and E, GSEA plots showing representative pathways with nominal enrichment in different RNA clusters. Corresponding nominal P values and FDR q-values are listed in Supplementary Table S9. D, Enrichment scores for cluster 3 compared with clusters 1 and 2. E, Enrichment scores for cluster 1 compared with clusters 2 and 3. F, Box plots showing the distribution of phenotypic scores across RNA clusters. G, Box plots showing the distribution of phenotypic scores across histologic subtypes. ANOVA, analysis of variance; CS, carcinosarcoma; EM, endometrioid; G1/G2, grades 1–2 endometrioid; N/A, not available; NC, normalized count; PC, principal component; t-SNE, t-distributed stochastic neighbor embedding.

Abstract Patient-derived xenografts (PDXs) are used in translational research and are particularly valuable for studying rare cancers such as sarcomas. Although it is critical to understand how accurately they reflect the genetic and phenotypic features of the original tumor, comprehensive analyses that access changes in both single-nucleotide variants (SNVs) and copy number variants (CNVs) during the establishment and passaging of PDXs remain scarce. In addition, as previous studies have focused on those derived from early-stage tumors or major cancer types, it remains unclear how genomic features are preserved in PDXs derived from advanced, treatment-resistant sarcomas.Here we performed whole-exome sequencing, RNA-seq and DNA methylation analysis of 144 sarcoma PDXs established from 39 patients, along with their matched primary tumors when available. The cohort was enriched for highly recurrent cases (n = 29) and included not only major sarcoma types, such as osteosarcoma (n = 11), but also ultra-rare sarcomas, including alveolar rhabdomyosarcoma (n = 4) and CIC-rearranged sarcoma (n = 2). PDXs were successfully established from a wide range of molecular subtypes, including fusion-driven tumors, TP53-mutated cases, and hypermutated types associated with microsatellite instability. Targetable genomic alterations, including amplifications of CCNE1 or MDM2 and deletions of MTAP, were identified in 55% of cases. Several PDXs harbored mutations associated with resistance to molecular targeted agents and immunotherapies.The vast majority of somatic alterations detected in the primary tumors were retained in the PDXs (97% for SNVs, 99% for CNVs). Moreover, mutational signature analysis suggested that mutations detected exclusively in PDXs likely represent low-frequency mutations already present in the original tumors. Extrachromosomal DNAs involving oncogenes such as MYC, CDK4, and MDM2 were also identified in 28% of PDXs and were retained in 83% of cases during serial PDX passaging. Despite overall genomic stability, clonal architecture changed during establishment and passaging, with highly diverse patterns, and clonal shifts were observed in 53%, 65%,71% of cases based on SNVs, CNVs, and both combined, respectively. Interestingly, clonal shifts were observed in all models derived from diagnostic samples (n = 9) and primary tumors (n = 8), which was much more frequent than in models derived from relapsed or metastatic tumors (62%). In addition, gene expression and methylation clustering revealed distinct clusters within each cancer type and individual cases, suggesting that key tumor characteristics were preserved during PDX establishment and passaging.Our data reveal the genetic and phenotypic stability of sarcoma-derived PDX models during their establishment and passaging, and provide new insights into their clonal evolution, which we believe will contribute to the clinical utility of PDX models. Citation Format: Kentaro Gosho, Shigehiro Yagishita, Kaito Mimura, Takahiro Ebata, Nobuyuki Kakiuchi, Kotoe Katayama, Seiya Imoto, Yoshitaka Narita, Akihiro Yoneda, Mitsuya Ishikawa, Yukihide Kanemitsu, Kan Yonemori, Akira Kawai, Shun-ichi Watanabe, Ayumu Arakawa, Toshikazu Ushijima, Seishi Ogawa, Akinobu Hamada, Kenichi Yoshida. Multi-omics analysis revealed clonal evolution of sarcoma patient-derived xenografts [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 6080.

Supplementary Figure S8. Immunological features of tumors according to phenotypic status.

Sarcomas are rare, heterogeneous tumors with poor prognoses, underscoring the urgent need for novel therapies. Liposomal eribulin (E7389-LF) was developed to improve the efficacy and safety profile of eribulin; however, its preclinical evaluation in sarcomas remains limited. This study aimed to evaluate the clinical applicability of E7389-LF. E7389-LF was comprehensively evaluated and compared with eribulin using 12 patient-derived xenograft (PDX) models representing six sarcoma subtypes. Antitumor activity was assessed across five treatment groups, including independently scaled, clinically relevant regimens of eribulin (0.47 mg/kg) and E7389-LF (0.77 mg/kg). Pharmacokinetic analyses were performed using liquid chromatography-tandem mass spectrometry and fluorescence immunostaining. In addition, correlations between antitumor activity and clinical efficacy were descriptively investigated. Treatment with E7389-LF (0.77 mg/kg) induced tumor regression in 7 of 12 models (58%), compared with 4 of 12 models (33%) treated with eribulin (0.47 mg/kg). Notable responses were observed in rhabdomyosarcoma, Ewing sarcoma, and leiomyosarcoma models. No apparent weight loss was observed in either treatment group. Pharmacokinetic analyses indicated increased systemic and intratumoral drug exposure with E7389-LF. In three PDX models derived from patients treated with eribulin, tumor responses reflected the corresponding clinical outcomes, providing preliminary evidence for the translational relevance of these models. This study demonstrates that E7389-LF exerts antitumor activity broadly comparable to eribulin and exerts antitumor activity in several models across sarcoma PDX models, with favorable tolerability and improved intratumoral drug retention. These findings support the clinical development of E7389-LF for patients with sarcoma who have limited treatment options.

Differential molecular, immunologic, and clinical features among transcriptomic clusters in high-grade endometrial carcinoma. A, Box plot showing the density of CD8+ T cells in MSS and MSI-H samples. B, Box plots showing the density of CD8+ T cells across different histologic subtypes and RNA clusters. C, Box plots showing the PD-L1 staining across different histologic subtypes and RNA clusters. D, Box plots showing the expression of HLA-A and HLA-B across different RNA clusters. E, Scatter plots showing the correlation between glandular/luminal score and the expression of HLA-A and HLA-B. Spearman rank correlation coefficients and P values are indicated. F, Kaplan–Meier curves for PFS according to RNA clusters. G, Kaplan–Meier curves for PFS according to RNA clusters for TP53-mut patients and NSMP patients. P values were calculated using the log-rank test. H, Kaplan–Meier curves for PFS according to RNA clusters for patients with serous and clear cell carcinoma. P values were calculated using the log-rank test. I, Box plots showing the distribution of phenotypic scores across RNA clusters in the TCGA dataset. J, Box plots showing the distribution of phenotypic scores across histologic subtypes in the TCGA dataset. K, Kaplan–Meier curves for PFS of MSI-H, TP53-mut, and NSML patients in the TCGA dataset. L, Kaplan–Meier curves for PFS of the same patient groups in the TCGA dataset, stratified by RNA clusters. NC, normalized count.

Supplementary Table S2. Significance thresholds used for GSEA and ssGSEA analyses.