Citations

Abstract Background: Recent studies indicate a correlation between increased stromal composition in tumors and resistance to immune checkpoint inhibitors (ICIs) across various cancer types. Specifically, the extracellular matrix (ECM) within the stromal compartment significantly influence the tumor microenvironment, affecting cancer cell proliferation and immune cell recruitment. However, identifying and quantifying the ECM at the collagen fiber level in whole slide images (WSIs) is challenging due to its inherent heterogeneity and complexity. In this work, we utilized statistical analysis and a deep learning (DL)-based approach to identify and assess collagen in the ECM, examining its characteristics in relation to the ICIs treatment response in gastric cancer. Methods: We analyzed 116 WSIs of hematoxylin and eosin (H&E) stained slides before ICIs treatment from three institutions. Responders were defined as those with a complete response or partial response (N=18); those with stable disease (SD) (N=34) or progressive disease (PD) (N=64) were considered non-responders (N=98). For non-responders, post-treatment H&E slides were available for 21 patients; SD (N=5) and PD (N=16). We applied a DL-based second harmonic generation (SHG) synthesis algorithm to tumor regions identified by a tumor detection model. We then calculated various collagen properties, including volume, fiber orientation and alignment, waviness, straightness, and fiber thickness, based on collagen fiber centerlines extracted from the generated SHG images. An attention-based multiple instance learning (ABMIL) model was trained to automatically predict the treatment response solely based on collagen features. The dataset of 116 pre-treatment slides was divided into training and validation subsets using stratified 3-fold cross-validation. The area under the curve (AUC) was used to assess the performance of the trained models. Results: Collagen fiber features were significantly associated with ICIs response. Notably, ICIs responders have a decrease in collagen waviness and fiber orientation variation (p-value<0.05, by Mann-Whitney U test). Among non-responders, the PD group showed a statistical decrease in waviness after treatment (p-value<0.05, by Mann-Whitney U test). The ABMIL model achieved a mean AUC score of 0.823, with a standard deviation of 0.023. Conclusion: This study provides a comprehensive analysis of collagen in the ECM, and its properties in relation to gastric cancer ICIs response within Tumor Immune Microenvironment. We also developed a model using collagen features and demonstrated high accuracy in predicting treatment response. Citation Format: Minji Kim, Ekaterina Redekop, Yoonho Choi, Sung Hak Lee, Jae-Ho Cheong, Yu Aoki, Kohei Shitara, Tae Hyun Hwang. AI-driven analysis of collagen characteristics in the tumor immune microenvironment predicts immune checkpoint inhibitors treatment responsiveness in gastric cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 4262.

Abstract Lung cancer is the leading cause of cancer-related deaths worldwide. Existing therapeutic options have limited efficacy, underscoring the critical need for the identification of new actionable therapeutic targets. We previously identified the Transmembrane Serine Protease TMPRSS11B as a novel gene that promotes the transformation of human bronchial epithelial cells in vitro and induces tumorigenesis in vivo. TMPRSS11B promotes extracellular release of Basigin, an obligate chaperone of the lactate monocarboxylate transporters MCT1/4, enhancing lactate export and glycolytic metabolism, and thereby promoting tumorigenesis. Importantly, TMPRSS11B is frequently overexpressed in human lung squamous cell cancers (LSCC) and high expression is associated with poor patient survival. Moreover, as a cell surface protein and enzyme, TMPRSS11B represents a tractable target for therapeutic intervention. To investigate whether TMPRSS11B activity impacts the host immune system and the tumor microenvironment (TME), we evaluated the effect of Tmprss11b depletion in a syngeneic LSCC mouse model, KLN205. Tmprss11b loss of function in KLN205 LSCC cells significantly reduced tumor burden in immunocompetent mice and triggered an accumulation of CD4+ T cells. We are currently extending these studies by investigating the effect of Tmprss11b loss of function using CRISPR/Cas9 editing in the Rosa26LSL-Sox2-IRES-GFP; Nkx2-1 fl/fl; Lkb1fl/fl (SNL) mouse model of LSCC. Tumor burden analysis and pH-sensitive fluorescent imaging are ongoing to assess changes in lactate accumulation. Moreover, RNA FISH analysis and spatial transcriptomics revealed that Tmprss11b expression is highly enriched in lung squamous tumors compared to lung adenocarcinomas and normal lung in SNL mice. Spatial analysis of the transcriptome is currently ongoing to assess the effects of Tmprss11b expression on the TME. Collectively, these studies will elucidate the molecular mechanisms through which TMPRSS11B promotes tumor growth and alters the tumor microenvironment in LSCC. Citation Format: Hari Shankar Sunil, Barrett L. Updegraff, Jingfei Zhu, Lisa Thomas, Bret M. Evers, Jean R. Clemenceau, Isabel Barnfather, John D. Minna, Ralph J. Deberardinis, Trudy G. Oliver, Tae Hyun Hwang, Kathryn A. O'Donnell. Transmembrane serine protease TMPRSS11B promotes tumorigenesis in lung squamous cell carcinoma through enhanced lactate export and modulation of the tumor microenvironment [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 1793.

Abstract Background: Immune checkpoint inhibitors (ICIs) targeting PD-1 or PD-L1 have dramatically improved survival of patients with gastric cancer (GC). However, recent clinical trials, including KEYNOTE-059, KEYNOTE-061, and ATTRACTION-2, show that the objective response rate (ORR) for ICIs is still less than 20% in patients with advanced GC. Therefore, predicting who will respond to ICIs in advance is essential to make better treatment plans for GC patients. There are potential biomarkers to predict ICI response, but most of them can only be applied to a limited population of patients. For example, Kim ST., et al., (PMID: 30013197) show that patients with MSI-high or EBV-positive responded well to ICIs, but these subtypes account for less than 30 percent of GC patients (PMID:25079317). In this study, we propose a prediction model of ICI response using the 32 GC-specific genes that we identified from mutation data (PMID: 26635139) and showed their close association with survival and ICI response in GC (PMID: 35140202; PMID: 36508166). In addition, since the number of available GC-ICI samples is small, we design our prediction model in Bayesian and ensemble learning frameworks which are robust to small sample size problems. Methods: We collected mRNA sequencing data from GC-ICI treated patients from our institutions and publications (PMID: 30013197; PMID: 35254400; PMID: 33846173). We used 108 samples previously included in our previous work (PMID: 36508166) as the training data and 27 samples (13 responders and 14 non-responders) from (PMID: 3384617) as the test data. We built and trained a classifier to separate responders from non-responders using the expressions of the 32 genes. We proposed to use a Deep Gaussian process (GP) model, a multi-layered extension of a GP that is itself a Bayesian method. We optimized the parameters of each GP model in variational inference. To implement ensemble learning, we trained multiple DGP models with different initializations of the model parameters and data splitting (training/validation sets). For each training run, the best model was selected based on the prediction performance on the validation set, measured by AUC. The final prediction for a test sample was made by combining the prediction results of all the DGP models by taking the average. Results: We implemented the DGP-based prediction model using PyTorch in Python, and the codes will be publicly available on the GitHub repository (https://github.com/hwanglab/GC_ICI_response_prediction). We achieved a prediction performance of 0.813 AUC on the test data (sensitivity = 0.769 and specificity = 0.714). Conclusion: We propose the prediction model of ICI response, which can handle small sample size problems in the framework of Bayesian and ensemble learning. The experimental results using the GC-ICI data demonstrate that the 32 genes can be used as predictive markers for ICI response in GC. Citation Format: Sunho Park, Minji Kim, Changjin Hong, Tae Hyun Hwang. A deep Gaussian process (DGP)-based prediction model for immunecheckpoint inhibitor (ICI) response in gastric cancer (GC) [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 2316.

Abstract Background: Microsatellite instability-high (MSI-H) tumors respond well to immune checkpoint blockade and oftentimes not to chemotherapeutics. Despite value of determining MSI status to guide therapy, many patients’ samples are not untested. Deep learning models may be able to predict MSI status by analyzing hematoxylin and eosin (H&E) stained whole-slide images (WSIs). Validation of this tool will provide another tool to increase MSI testing. However, to be deployed into clinical routine care, prediction models need be externally validated in large and diverse patient cohorts. In addition, these models should provide uncertainty of prediction to help clinicians make informed decisions. Results: In this study, we introduce a novel Bayesian prediction model using whole-slide images (WSIs) and deep Gaussian processes (DGPs) in a weakly supervised learning context. Extensively tested on large datasets (n=3,146) from multiple centers, our model not only demonstrated superior performance but also showed that incorporating uncertainty into predictions significantly improves MSI prediction accuracy, achieving AUROC of >0.91 in colon cancer and >0.94 in gastric cancer. Additionally, an in-depth analysis within the gastric cancer cohort indicated a significant correlation between predicted MSI-H regions and immunotherapy response. Patients who responded to immune checkpoint inhibitors had a higher proportion of MSI-H regions (P-value < 0.05). This correlation was also observed in patients categorized as MSS but responsive to ICIs, highlighting the potential of our model to identify MSI-H patterns not detected through standard laboratory tests. This advancement could lead to more precise patient selection for ICI therapy, enhancing personalized treatment strategies in gastrointestinal cancer. Citation Format: Sunho Park, Minj Kim, Ji-Youn Sung, Jeong Hwan Park, Sung Hak Lee, Sam C. Wang, Jaeho Cheong, Tae Hyun Hwang. Explainable AI model incorporating uncertainty estimation for enhanced MSI-H prediction and immunotherapy response in gastrointestinal cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 7385.

Abstract Traditional cancer research methodologies predominantly focus on 2D tumor modeling, including spatial profiling and AI-driven analysis at the 2D level. To advance beyond these limitations, we implemented Holotomography (HT), a 3D, label-free optical imaging method. This technique, leveraging refractive index contrasts, allows for enhanced visualization of tumor tissues, organoids, and cells in three dimensions. Using serial H&E sections and unstained slides, we created detailed 3D tumor models in gastric cancer, providing insights into tumor, immune, and stromal composition, as well as cell-cell interactions and extracellular matrix characteristics. AI algorithms further enhanced our 3D tumor modeling, offering novel perspectives on the initiation and progression of tumor, immune, and stromal components at the subcellular level. Analyzing gastric tumors treated with immune checkpoint inhibitors before, during, and after treatment revealed dynamic changes in spatial composition over time, offering unique insights into the evolution of the 3D tumor ecosystem. Additionally, by integrating spatial transcriptomic and in situ molecular imaging data, we developed a novel Multimodal 3D Tumor Model. This model not only uncovers subcellular-level molecular characteristics but also elucidates cellular interactions during treatment, suggesting potential biomarkers and therapeutic approaches. Lastly, by constructing 3D cellular interaction networks incorporating true 3D interactions, we provided an in-depth view of cellular communication within the tumor microenvironment. This pioneering approach, representing the first Multimodal 3D Tumor Model at the subcellular level, offers novel insights into gastric cancer's spatial evolution, potentially revolutionizing our understanding and treatment of gastric cancer. Citation Format: Yoon Ho Choi, Minji Kim, Katya Redekop, Inyeop Jang, Jean R. Clemenceau, Isabel Barnfather, Sumanth Reddy Nakkireddy, Changjin Hong, Sunho Park, Ji-Youn Sung, Jeong Hwan Park, Shitara Kohei, Aoki Yu, Sung Hak Lee, Sam Wang, Jaeho Cheong, Tae Hyun Hwang. AI-driven multimodal 3D tumor modeling with spatial molecular insights for predicting ICI response in gastric cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 7400.

686 Background: Immune checkpoint inhibitors (ICI) have been used to treat advanced muscle invasive and/or metastatic bladder cancer. However, ICIs are only effective in 30-40% of cases. In light of potentially significant IRAE’s, tools to predict individual patient level benefits are desirable. AI technology offers the opportunity to apply computational methods to predict outcomes from readily available digitized pathology slides. We investigated the utility of an AI platform integrating computational biomarkers based on morphological characterization of digitized H&E slide whole images (WSI) to predict response to ICI. Methods: We analyzed H&E-stained whole slide images (WSI) of bladder tumor tissue collected from transurethral resection of bladder tumor (TURBT) from 116 patients with advanced or metastatic bladder cancer at a single institution between 2015-2020. Adjacent multiplex IHC stained specimens were analyzed myeloid and lymphoid marker panels. 20 patients from the overall cohort treated with ICI were designated ‘responders’ (complete response, partial response, stable disease) and 20 patients were identified as non-responders (progressive disease) based on ICI best clinical/radiographic response. WSIs were divided into small non-overlapping image patches. These image tiles were processed into multiple AI encoder models to extract morphological features. The tile morphological features, represented by high dimensional vectors, were combined by an aggregation model to represent the whole slide morphology and then used to classify the patients into responders vs non-responders to immunotherapy. Area-under-the-receiver operating characteristic (AUC) was used to measure the performance of response prediction. Results: Our method shows AUC of 0.708 at classification of the patients (n=40) into responder and non-responder groups. With a cutoff that identifies the top 50% patients (n=20) of high probability of responding to immunotherapy predicted by our models, 75% of them are responders. In the bottom 50% patients (n=20) of the low scores, 75% of them are non-responders. . Multiplex IHC (myeloid and lymphoid panel) was performed using adjacent sections and were characterized yielding deeper mechanistic insight into the inflammatory populations driving predictive morphologic patterns. Conclusions: By applying innovative AI morphological analysis on patient WSIs, we generated a computational biomarker that shows promising ability to predict patient response to immunotherapy in bladder cancer. This approach merits future prospective validation and additional patient level analysis in broader cohorts. [Table: see text]

Abstract Although the majority of HPV (+) oropharyngeal squamous cell carcinoma (OPSCC) patients exhibit a favorable prognosis, approximately 20% face recurrent or metastatic disease, posing management challenges. Therefore, the identification of robust prognostic markers for risk stratification is essential. Our study focused on the comprehensive characterization of intratumor heterogeneity within the tumor immune microenvironment (TME) in both primary tumors and metastatic lymph nodes. Utilizing computational approaches, we integrated hematoxylin and eosin (H&E) and 11 adjacent immunohistochemistry (IHC)-stained slides to investigate the cellular composition and functional characteristics across different regions within the TME. Based on these detailed immune characteristics, we classified patients into specific immune subtypes. Our integrative TME analysis indicated that patients with high tumor-infiltrating lymphocytes (TIL), increased CD8 + levels, and reduced CD163 + cell counts within their primary tumors were likely to have a more favorable progression-free survival outcome. The integration of H&E and IHC data enhances the functional characterization of immune phenotypes of the TME with biological interpretability and improves patient stratification in HPV(+) OPSCC.

Abstract Differences in clonality, genomic and epigenetic alterations are known to occur between primary Glioblastoma (pGBM) and recurrent (rGBM), but this has been insufficient to yield a promising therapeutic approach. Recent studies show that the GBM tumor microenvironment (TME), composed of reactive glial cells, microglial subtypes, neuronal activity, myeloid cells, lymphocytes, etc., plays a crucial role in the invasiveness and evolution of the glioma cells. Nevertheless, how the spatial architecture and interplay among these cells facilitates GBM resurgence is largely unknown. To this point, we developed a pipeline for characterizing the TME changes using single nucleus and spatial transcriptomics (ST). We retrospectively collected matched, fresh-frozen, IDH-wildtype pGBM and rGBM samples (n=14, 7 patients) from the Mayo Clinic Neuroscience Biobank. Single nucleus RNA sequencing was performed using the 10x Chromium platform. Matched FFPE blocks from 8 samples were collected from the Mayo Clinic pathology department for 4 of the 7 patients. A region of interest was identified from the corresponding histopathology slide by a pathologist, avoiding areas of necrosis and hemorrhage. The region was collected from the block using a 5mm biopsy punch. The tissue was re-paraffinized, and a 5µm thick section was used to perform ST with the 10x Visium for FFPE platform. We performed transcriptomic and image texture-based clustering of capture dots along with differential gene expression analysis. Based on gene expression, we identified and characterized conserved regions across samples. We used single nucleus data to infer cell composition. Spatial correlation and autocorrelation statistics were calculated on features, such as gene expression, gene set scores, cell type prediction, and antibody capture. The features’ relative spatial changes were compared between pGBM and rGBM. Our analyses showed high heterogeneity within and across samples, but we found higher molecular and textural heterogeneity in rGBM samples compared to pGBM. We found higher androgen response activity (p < 0.001) presents diffusely (I: 0.09) throughout rGBM tissue but localized to mesenchymal-like regions in pGBM (I: 0.15). KRAS signaling signature was decreased in pGBM (p < 0.01) and increased in rGBM (p < 0.01). This increase was significantly overexpressed in cluster 0 for rGBM (p=0.02), which is an area characterized by neural progenitor signature. No significant difference in KRAS or EGFR magnitude or distribution of expression was present for either GBM type, indicating an independent mechanism for pGBM. We established a pipeline for assessing the spatio-temporal changes of the GBM TME. We found evidence indicating differences in the type of growth signaling, the cells bing targeted, and their distribution across the TME between pGBM and rGBM. Our work suggests how ST can help us better understand the intricate dynamics of GBM. Citation Format: Jean R. Clemenceau, Paola Suarez Meade, Mark E. Jentoft, Alfredo Quiñones-Hinojosa, Tae Hyun Hwang. Spatial transcriptomic landscape of the primary and recurrent glioblastoma tumor microenvironment [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 144.