Supplementary Figure S8. Myeloid cell transcriptional diversity in response to mutant-selective and broad spectrum RAS inhibitors.

CAR27-28ζ NK cell serial killing against Raji cells.

Abstract Background: Colorectal cancer (CRC) metastases frequently recur due to minimal residual disease (MRD) and persisting micrometastases after therapy. However, the spatial and molecular features underlying micrometastatic persistence and CRC recurrence remain poorly defined. Study design and methods: We performed integrative spatial multi-omics profiling—including Visium spatial transcriptomics (ST), Visium HD ST, laser-capture microdissection with whole-genome sequencing (LCM-WGS), and PhenoCycler-Fusion multiplex imaging—across 49 tumors from 19 patients with paired primary CRC, liver (CLiM), and lung (CLuM) metastases. The analysis encompassed 341,328 Visium spots and approximately 3.8 million Visium HD bins. Non-negative matrix factorization (NMF) was applied to identify conserved and distinct spatial metaprograms across CLiM, CLuM, and primary CRC using Visium ST datasets. For Visium HD ST data, StarDist-SMURF segmentation was used to transform subcellular bins into single-cell-level data. Cross-modality alignment and Jaccard similarity analyses integrated spatially resolved DNA, RNA, and protein profiles across both corresponding and independent tissue blocks, enabling multi-layer characterization of tumor evolution and microenvironmental organization. Results: Spatial phylogenetic and molecular analyses delineated distinct evolutionary trajectories of primary and metastatic CRC, revealing early clonal divergence and stem-like phenotypes in liver micrometastases (CLiMi) across DNA, RNA, and protein levels. Spatial profiling uncovered stromal interactions in both CLiM and CLuM, with macrophages enriched in CLiM and lymphocytes predominating in CLuM. Micrometastases exhibited pronounced immunosuppression and T cell exhaustion, potentially mediated by PGE2/PTGES2-PTGER4 and NECTIN2/3-TIGIT signaling interactions. A CLiMi-specific six-gene signature predictive of micrometastasis was identified and validated, correlating with disease-free survival (DFS) and MRD-DFS in the MDACC cohort (n = 117), and with DFS and overall survival (OS) in TCGA (n = 610) and GSE17538 (n = 232). Conclusions: Our integrative spatial multi-omics analysis provides a comprehensive atlas of CRC micrometastases, revealing their evolutionary and immune landscapes. These findings illuminate the molecular and spatial determinants of micrometastatic persistence and identify potential therapeutic vulnerabilities for preventing CRC recurrence. Citation Format: Yang Liu, Akshaya S. Jadhav, Yuwen Pan, Jianlong Liao, Isha Khanduri, Yunhe Liu, Riham Katkhuda, Wei Lu, Kyung Serk Cho, Tieling Zhou, Baohua Sun, Mei Jiang, Sharia D. Hernandez, Idania Carolina Julio, Patrick Brennan, Guangsheng Pei, Kai Yu, Yibo Dai, Tian Chu, Fuduan Peng, Khaja Khan, Saxon Rodriguez, Ling Xia, Youming Guo, Alicia Mejia, Zhiming Tong, Sean W. Barnes, Ou Shi, Shreeya Indulkar, Alaa Mohamed, Natalie Wall Fowlkes, Timothy Newhook, Yun Shin Chun, Van K. Morris, David G. Menter, Dadi Jiang, Jean-Nicolas Vauthey, Ruoyan Li, Humam Kadara, Luisa M. Solis Soto, Scott Kopetz, Linghua Wang, Dipen M. Maru. Spatial multi-omics dissection of colorectal cancer micrometastasis [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 6116.

Foundation models have emerged as powerful feature extractors in computational pathology. However, they typically omit mechanisms for leveraging the global spatial structure of tissues and the local contextual relationships among diagnostically relevant regions—key elements for understanding the tumor microenvironment. Multiple instance learning (MIL) remains an essential next step following the foundation model, designing a framework to aggregate patch-level features into slide-level predictions. We present EAGLE-Net, a structure-preserving, attention-guided MIL architecture designed to augment prediction and interpretability. EAGLE-Net integrates multi-scale absolute spatial encoding to capture global tissue architecture, a top-K neighborhood-aware loss to focus attention on local microenvironments, and background suppression loss to minimize false positives. We benchmarked EAGLE-Net on large pan-cancer datasets, including three cancer types for classification tasks (10,701 slides) and seven cancer types for survival prediction (4,172 slides), using three distinct histology foundation backbones (REMEDIES, Uni-V1, Uni2-h). Across tasks, EAGLE-Net achieved up to 3% higher classification accuracy and the top concordance indices in 6 of 7 cancer types, producing smooth, biologically coherent attention maps that aligned with expert annotations and highlighted invasive fronts, necrosis, and immune infiltration. These results position EAGLE-Net as a generalizable, interpretable framework that complements foundation models, enabling improved prognostic and diagnostic performance.

Colorectal cancer (CRC) metastases frequently recur due to minimal residual disease (MRD) and persistent micrometastases after therapy. Here, we performed spatial multimodal profiling using spot-level and high-resolution spatial transcriptomics, multi-regional whole-genome sequencing following laser-capture microdissection, and high-plex protein imaging to map 49 tumors from 19 patients, encompassing paired primary CRC and matched liver (CLiM) and lung (CLuM) metastases. Phylogenetic reconstruction revealed that liver micrometastases (CLiMi) arose from early clonal divergences and harbored a stem-like, quiescent state consistent with metastatic dormancy. Spatially, we uncovered distinct stromal barriers: macrometastases were encapsulated by myofibroblasts, whereas micrometastases were surrounded by immunosuppressive niches characterized by T cell exhaustion and distinct ligand-receptor signaling networks. Notably, we identified a CLiMi-specific six-gene signature associated with MRD status, disease-free survival, and chemotherapy resistance across multiple independent cohorts. These findings elucidate the spatial evolutionary landscape of CRC metastases and provide tissue-based spatially validated biomarkers for surveillance and therapeutic targeting.

(Related to Fig.6G) The list of differentially abundance proteins between CAR27-ζ vs. CAR27-28.

Supplementary Figure S15. Clinical implications of mucinous lineage differentiation in human KRAS-mutant NSCLC cohorts

CAR27-28ζ NK cells show polyfunctionality.

Supplementary Figure S3. Acute effects of active RAS inhibition in the KP2 and KL2 KrasG12C-mutant lung adenocarcinoma models

Expression of CD70 at the protein and mRNA levels in different tumor types.

Supplementary Figure S10. MAPK pathway modulation in immune and stromal cell subsets in response to active RAS inhibition.

Supplementary Figure S4. Acute inhibition of active RAS in KRASG12C-mutant lung adenocarcinoma.

CD28 co-stimulation enhances phosphorylation of the CAR CD3ζ signaling endodomain.

Abstract Background: Immune checkpoint inhibitors (ICIs) have improved survival in non-small-cell lung cancer (NSCLC), yet only a subset of patients benefits, and biomarkers like PD-L1 remain limited. Here, we introduce a deep learning-based pathomics framework that utilizes routine H&E-stained slides to predict therapeutic response and survival outcomes in ICI-treated metastatic NSCLC. Methods: The study included 797 ICI-treated NSCLC patients from MD Anderson, with external validation in 280 patients from Mayo Clinic, Gustave Roussy, and the Phase III ICI-naïve LUSC Lung-MAP S1400I trial receiving nivolumab with or without ipilimumab. Path-IO (Pathology-Driven Immunotherapy Optimization) comprised of four major steps: (1) pathologist-verified tissue classifier segmented WSIs into eight compartments—Background, Bronchi, Immune, Lung, Necrosis, Stroma, Tumor, and Vessel—and was validated on TCGA and CPTAC dataset; (2) a survival prediction module generating patient-level risk scores in the MD Anderson cohort and validated across external datasets; (3) integration of Path-IO risk scores with radiomics and clinical features to improve prognostic accuracy; and (4) biological interpretability analyses correlating Path-IO risk immune contexture from multiplex immunofluorescence and transcriptomic signatures from NanoString profiling. Results: Path-IO effectively stratified patients into high and low-risk groups with significant survival differences. In the MD Anderson cohort, it achieved HR = 2.11 (p < 0.001) and HR = 2.51 (p < 0.001) for OS in the discovery and validation sets, and HR = 2.34 (p < 0.001) and HR = 1.87 (p < 0.001) for PFS, respectively. In the multicenter Phase III Lung-MAP S1400I trial, Path-IO achieved HR = 1.78 (p = 0.016) for OS and HR = 2.76 (p = 0.006) for PFS. In external validation, it predicted outcomes in the Gustave Roussy (OS = 1.97, p = 0.003; PFS = 1.51, p = 0.046) and Mayo Clinic (OS = 2.46, p = 0.007; PFS = 2.45, p = 0.027) cohorts. Path-IO outperformed PD-L1 and remained an independent predictor in multivariate analysis (p < 0.001). It enabled data-driven frontline selection between anti-PD-1 monotherapy and chemo-immunotherapy, offering guidance toward more personalized treatment decisions. Integration with radiomics and clinical features further improved predictive accuracy (OS 0.63→0.75; PFS C-index 0.58→0.70), while high Path-IO risk scores correlated with immune-cold phenotypes identified by multiplex immunofluorescence and NanoString transcriptomics. Conclusions: Path-IO demonstrates that deep learning models rooted in histopathologic architecture can generate interpretable and biologically informed survival predictions in NSCLC treated with ICIs. By integrating pathology, radiology, and clinical data, Path-IO provides complementary predictive value beyond established biomarkers such as PD-L1. Citation Format: Rukhmini Bandyopadhyay, Linghzi Hong, Mihaela Aldea, Shenduo Li, Lodovica Zullo, Frank R. Rojas, Maliazurina B. Saad, Maricel C. Marin, Muhammad Waqas, Jiexin Zhang, Eman Showkatian, Claudio A. Arrechedera, Xiaoyu Han, Yuliya Kitsel, Sherif Ismail, Muhammad Aminu, Bo Zhu, Carol C. Wu, Brett W. Carter, Joe Y Chang, Zhongxing Liao, Maria R. Ghigna, Davide Soldato, Hai T. Tran, Xiuning Le, Tina Cascone, Bingnan Zhang, Haniel A. Araujo, Mehmet Altan, Simon Heeke, David Jaffray, Don L. Gibbons, Ara Vaporciyan, J Jack Lee, Neda Kalhor, Cara Haymaker, Ignacio Wistuba, John V. Heymach, Yanyan Lou, Natalie Vokes, Luisa M. Solis Soto, Jianjun Zhang, Jia Wu. Path-IO: A deep learning pathomics framework for personalized immunotherapy selection and outcome prediction in metastatic non-small cell lung cancer [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 4003.

Supplementary Figure S2. Models and mechanisms of acquired resistance to RAS inhibitors

Abstract Loss-of-function mutations in RBM10, encoding a protein involved in the regulation of alternative splicing, are observed in ∼10% of non-squamous NSCLC and are enriched in tumors harboring activating mutations in KRAS, yet little is known about how RBM10 inactivation promotes lung cancer progression or influences response to standard-of-care systemic therapies. Here, we elucidated the cooperative interplay between oncogenic KRAS activation and RBM10 loss in NSCLC pathogenesis using a novel genetically engineered mouse model with conditional RBM10 deletion, as well as multiple isogenic syngeneic allograft models that faithfully recapitulate RBM10-deficient human lung adenocarcinoma. We found that loss of RBM10 accelerates KRAS-mutant NSCLC progression by fostering the establishment of a tolerogenic, myeloid cell-rich tumor immune microenvironment (TIME). Mechanistically, RBM10 loss promoted R-loop accumulation and chronic DNA damage signaling that engaged the non-canonical TRAF6-STING pathway in a cGAS-independent manner, leading to sustained NF-κB activation. We identified several cytokines and chemokines, canonical targets of NF-κB signaling, such as IL-1β, IL-6, TNFα, and MCP-1, that were upregulated in RBM10-deficient cells. This NF-κB-driven secretome promoted the development of an inflamed, TIME characterized by accumulation of suppressive myeloid cell subsets - most notably monocytes and M2-like macrophages, and dysfunctional tumor infiltrating lymphocytes (TILs) thereby fostering immune evasion and cancer progression. Furthermore, we exploited the RNA-seq database of human lung adenocarcinoma from The Cancer Genome Atlas (TCGA) and again found that RBM10 loss was significantly linked to impaired DNA damage response, upregulation of the HALLMARK_TNFA_SIGNALING_VIA_NF-κB, and enrichment of M2-macrophages. Strikingly, targeting the CSF1/CSF1R axis with an anti-CSF1R antibody significantly curtailed the growth of RBM10-deficient tumors in syngeneic immunocompetent models and synergized with anti-PD-1 therapy to promote tumor regression. In conclusion, our findings uncovered a novel critical role of RBM10 inactivation in driving immune escape and PD-1 inhibitor resistance in KRAS-mutant NSCLC and suggest a potential therapeutic strategy by co-targeting suppressive myeloid cells to improve cancer immunotherapy for patients bearing KRAS;RBM10 co-mutated tumors. Citation Format: Minh Truong Do, Teng Zhou, Mhd Yousuf Yassouf, Richard Lee, Obada E. Ababneh, Yanhua Tian, Leticia B. Rodriguez, Jayanthi Gudikote, Haniel A. Araujo, Stephanie T. Schmidt, Jing Wang, Frank R. Rojas Alvarez, Luisa M. Solis Soto, Marcelo V. Negrao, Alexandre Reuben, Don L. Gibbons, Jianjun Zhang, John V. Heymach, Ferdinandos Skoulidis. RBM10 loss promotes KRAS-mutant non-small cell lung cancer immune tolerance and PD-1 inhibitor resistance via a non-canonical STING/ NF-κB axis [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 6989.

Abstract Intrahepatic cholangiocarcinoma (iCCA) is an aggressive type of biliary tract cancer (BTC) with a rising incidence and limited treatment options. Previously established cell-surface targets of antibody-based therapies are not common in iCCA; for example, HER2 amplification occurs in <5% of patients. Knowledge of targetable membrane-bound proteins is a critical bottleneck in the field. Here, we performed a proteogenomic analysis that revealed FGFR2b as a dominant cell-surface isoform in iCCA and explored the association of its expression with genomic alterations, expression pathways, and splicing regulation. We developed a comprehensive end-to-end platform to search for cell surface targets, including alternative splicing variants. Briefly, we perform de novo assembly of RNA-sequencing data from BTC samples (n=79), the majority iCCA (n=67), to include unannotated alternative isoforms using stringtie. Using fragpipe, we searched for novel proteins from in silico translation of these assembled transcripts in cell surface protein enriched mass spectrometry data from 15 cancer cell lines. Splice junction expression from detected isoforms was normalized and compared. Using recount3, we also assessed expression of these splice junctions across 10,415 TCGA samples and 19,081 GTEx samples. An optimized IHC protocol was established using the Roche anti-FGFR2b mouse monoclonal antibody (FPR2-D). A unique glycosylated peptide from a FGFR2 alternative transcript was detected in the cell-surface mass spectrometry data, and we identified this alternative isoform to be FGFR2b. In our institutional cohort of 79 BTC samples, FGFR2b was the predominant FGFR2 isoform for 88.6% (70/79) of patients, which was further confirmed in 34 BTC samples from TCGA (88.2%; 30/34). Across 33 TCGA tumor types, BTC had the highest average expression of FGFR2b. FGFR2b had limited expression in normal tissues from GTEx. Patients with FGFR2 fusions had significantly higher FGFR2b expression than FGFR2c (p=0.025). FGFR2b expression negatively correlated with epithelial-to-mesenchymal transition and positively correlated with the epithelial splicing factor ESRP1, suggesting that ESRP1-driven FGFR2b expression is a hallmark of a more differentiated subtype of iCCA. In our institutional iCCA samples, 31.6% (6/19) of patients had >10% 2+/3+ staining. Strikingly, all patients with FGFR2 fusions (4/4; 100%) were positive for FGFR2b. These results indicate that FGFR2b is highly expressed in iCCA, especially in patients with FGFR2 fusions, and highlight it as a compelling therapeutic target. Notably, the positivity rate of 31.6% is higher than that reported for gastric cancer (∼16%), a setting in which multiple FGFRb-targeted therapeutic campaigns are currently underway. Our findings provide a strong rationale for the repurposing and evaluation of these FGFR2b-targeting agents in iCCA in prospective clinical trials. Citation Format: Nakul M. Shah, Beatriz Alvarado-Hernandez, Xinyue Chen, Quentin Kimana, Felicity Namayanja, Wei Lu, Khaja B Khan, Juan Gallegos, Sangeeta Goswami, Lawrence N. Kwong, Luisa M. Solis Soto, Milind M. Javle, Sachet A. Shukla. FGFR2b is a highly prevalent and actionable cell surface target in intrahepatic cholangiocarcinoma [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 3075.

Supplementary Figure S1. Activity of RAS inhibitors in immune-competent pre-clinical models of KRASG12C-mutated NSCLC.

Background:The use of immune checkpoint inhibitors (ICIs) has led to a paradigm change in cancer management. Many patients may have inherent primary resistance to ICIs or develop secondary resistance after initial response. The impact of using novel therapeutic combinations of checkpoint blockade (avelumab) with immune stimulating agonists such as anti-OX40 and/or anti-4-1BB on the tumor microenvironment and modulation of the immune response is an intriguing strategy to evaluate how these agents interact and whether the hypothetical rationale for combinations can be translated into augmentation of anti-tumor immunity in solid tumors. Methods:We performed whole exome sequencing (WES), bulk RNAseq, multiplex immunofluorescence (mIF) and chromogenic immunohistochemistry (IHC) on tumor tissue and flow cytometry of the peripheral blood to study longitudinal changes following the combination of avelumab with utomilumab (a 4-1BB agonist) (arm A), PF-04518600 (an OX40 agonist) (arm B), utomilumab and PF-04518600 (arm C) and utomilumab and radiotherapy (arm D) in phase I/II study (NCT03217747). Results:We observed low tumor mutation burden (TMB < 6) (median: 1.88), alteration of RTK-RAS, TP53, PI3K and WNT pathways across the cohorts. Mutations in TP53, TTN and KRAS (mostly p.G12C, p.G12D) genes and copy number variations (CNV) were found in PIK3CA, CCNE1 and KRAS. Interferon gamma signaling pathway was enriched early on-treatment in tumors from patients with colorectal and pancreatic cancers in arm C. Patients deriving clinical benefit (CR/PR/SD ≥ 4 months) displayed higher T-cell frequencies at baseline (p = 0.0157), C1D15 (p = 0.0086), and C3D15 (p = 0.0070) than patients without clinical benefit. Conclusions:Our findings, though limited, highlight genomic differences between histologic subsets and outcome as well as the need for combination strategies that drive the recruitment and/or priming of anti-tumor T cells and address low immune permissive tumor states in patients with advanced solid tumors. Clinical trial registration:This clinical trial was registered on clinicaltrials.gov NCT03217747.

(Related to Fig.6F) The list of differentially abundance proteins between CAR27-ζ vs. CAR27-28ζ.