Tumor Immune MicroEnvironment Facility

Division Of Basic Sciences - Nci

Investigators

Linked publications & trials

Abstract

We added 1 more patient to our cohort of 6 patients having local advanced prostate cancer and undergoing radical prostatectomy (RP). These patients were treated with neoadjuvant PROSTVAC and Nivolumab (NCT 02933255). We investigated the immune modulation in TIME after treatment using multiplex immunofluorescence (Opal Method). FFPE sections from matched pre-treated prostate biopsies and post-treated RP samples were stained with our validated T cell panel1. Combination immunotherapy significantly increased CD4+ T cells and CD8+ T cells densities in the invasive margin, intratumoral and the benign compartments. 6/7 and 5/7 patients showed more than 2Xincrease of CD4 and CD8 T cells in the TIME, respectively, in at least one of the three compartments, showing more effect that Prostvac alone. Increased proliferative indices in CD4+ and CD8+ T cells were also seen after treatment. Tregs were present in low frequencies in TIME (maximum of 12 cells/mm2) with no significant changes. Moreover, a significant drop in tumor cell Ki67 after treatment suggests that the combination may control tumor growth. The combination of Neoadjuvant Prostvac and nivolumab was associated with increased immune cell infiltration in a cohort of early prostate cancer patient. This work was presented as a poster at SITC in November, 2021. We are developing new IF-multiplex panels for mouse tissues to support the preclinical team at the center or immuno-oncology (CIO). The preclinical team is treating mice bearing different tumor models with different drugs in solo and in combination. To look at the effect of these drugs on TIME and correlate we are developing 6 multiplex panels using IF, each of these panels include 6 markers and DAPI. We have a collaboration with Dr. Michael Lenardo's laboratory to explore the immune modulation in patients with inflammatory bowel disease (IBD) and having GPR15 mutations. GPR15, a G protein-coupled receptor originally identified as a high affinity coreceptor for HIV/SIV, was recently shown to home lymphocytes to murine colon and developing epidermis. The effect of Gpr15 deficiency on murine colitis varied in different mouse models but the studies demonstrated the critical role of GPR15 in homing Treg and Teff cells to the large intestine lamina propria (LILP). In addition, the physiological and pathogenic roles of GPR15 in human have not been determined. Dr Lenardo's group has identified 3 families with GPR15 mutations and severe early onsets of IBD. In order to look at the effect of this mutation on the immune cells present in colon of these patients, we developed 5 multiplex fluorescence panels using the opal technology. The panels we developed are the following: Panel 1 (CD8, KI67, GRANZYME B, PRO-CASPASE 3 AND PAN-CK), PANEL 2 (CD3, GPR15, GPR15L, CD4, PAN-CK), PANEL 3 (CD4, FOXP3, PRO-CASPASE 3, KI67, PAN-CK), PANEL 4 (CD4, FOXP3, KI67, CLEAVED CASPASE 3, CD20, PAN-CK,) and panel 5 (CD3, GRANZYME B, CD8, CD56, CD4). Our studies showed that the GPR15 mutations led to impaired GPR15L-induced signal transduction and T cell migration, perturbing the normal T cell distribution in the patient small intestine and colon. We also identified a unique group of CD56+GZMB+ cells that was upregulated in our patient intestine. This population was not observed in ulcerative colitis, Crohn's disease, and healthy controls without GPR15 mutations, indicating these cells may be unique to the GPR15-deficient IBD etiology. Finally, we showed that loss of GPR15 in primary NK cells significantly upregulated GZMB expression and cytotoxicity. Thus, our study not only demonstrates GPR15's role in maintaining lymphocyte homeostasis but also uncovers the role of NK cells in gut inflammation, which has not been studied to the best of our knowledge. A manuscript is in preparation to publish this data. We have a collaboration with Dr. George Zeki from NFL and Drs. Shaban Mohammed and Faisal Mahmood from Harvard Medical School and the Division of Computational Pathology at the Brigham and Women's Hospital. During the first seven months of the award of the grant : Artificial Intelligence Based Analysis of Tumor Immune MicroEnvironment (TIME) in Patients Treated with Immune Therapy Agents , we have successfully accomplished several tasks, in terms of data generation and image processing, related to our first AIM: "Stratify the tumor immune cell patterns in Multiplex Immunofluorescence MxIF images using unsupervised learning approaches based on Convolution Neural Networks and Graph Convolution Neural Networks and GCNN". For this grant, we are using pre and post treated tumor tissue section from 42 patients treated with Bintrafusp alfa. These tissue sections were immuno-stained using our validated panels using multiplex fluorescence assays and then scanned. For every patient, three fluorescence panels were performed. Each panel contains six multiplex immunofluorescent markers and DAPI. For every section, we select regions of interest (ROIs) that we scan using 40X objectives the clinical protocol requires the scanning of selected regions of interest (ROIs) crops that are about 1500 * 1500 * 7 images from the whole slide. Additionally, for the purpose of the LDER (Laboratory Director Exploratory Research) grant, we are also scanning all the crops available in the section which takes more significant scanning time. We have implemented a standard operating procedure to collect the relevant metadata for every patient, section, and panel. This metadata includes the time of acquisition, tissue type, patient response, and type of the section (tumor/biopsy). With the increasing number of images, sections, and panels, these metadata would be crucial as develop, analyze, and evaluate the machine learning models generated by this research. We also defined and implemented a data and metadata anonymization and sharing protocols. These protocols enabled us to share the generated data with our external collaborators from Brigham and Women's hospital and to automate the data sanitation and transfer process to be able to track and process thousands of files generated during the period of the grant. In summary, the data generation is moving forward with a pace that allows us to move forward with the image processing tasks related to AIM1. Another aim in this project is to define a new method for an Automated prediction/generation of a marker using a combination of other markers can help to reduce the time and cost to prepare a MxIF panel. As a proof of concept, Dr Mohammed has developed a deep learning method that takes a set of protein markers (CD4, CD8, CD56, CD68, CD163, and DAPI) and predicts the Cytokeratin (CK) marker. This method uses an autoencoder architecture to first down sample the set of the input markers into a latent representation, then synthesize the Cytokeratin marker. He trained and evaluated the proposed method on an in-house pilot dataset of 1300 images (1024x1024 pixels). We provided images from this panel without CK and with CK to be used as ground truth. The proposed method has shown reliable performance both in terms of quantitative and qualitative results. The structural similarity between generated and the real CK markers is 98%. We further compared the CK+ and CK- cells in generated and the real CK markers. The proposed method achieves average precision and F1 score (which is a parameter to compare the performance of 2 classifiers) of 0.87 and 0.75, respectively. Now we are generated new set of images from patient's tissues stained with our va *TRUNCATED*