Swanson, Kristin R.

Abstract Glioblastoma (GBM) is the most aggressive and most common primary malignant brain tumor in adults, with a poor median survival time of 15 months. One of the key challenges in treating GBM is its highly heterogeneous nature, with multiple distinct subtypes that have been shown to occur on both inter- and intra-patient levels. Three main classifications, known as classical/proliferative, mesenchymal and proneural have become commonly demonstrated phenotypes. The cell cycle is a fundamental and highly conserved process that controls faithful division of cells; dysregulation of the cell cycle is known to be a key driver in many cancers. However, how the cell cycle is differently regulated between these subtypes has not been well classified in vivo. We investigate these three GBM subtypes using a recently published single nucleus RNAseq (snRNAseq) data set. We compare cell cycle regulation/dysregulation among these three subtypes using Tricycle, an R/Bioconductor package that utilises dimension reduction via principal component analysis and transfer learning to infer cell cycle position from any snRNAseq data set. We find that the classical GBM subtype has the highest proportion of actively dividing cells (cells in: S/G2/M phases), while the mesenchymal and proneural subtypes have a very low proportion of actively dividing cells. This supports the idea of a proliferation-migration dichotomy between GBM subtypes. We use this proportion of actively proliferating cells to calibrate a minimal spatiotemporal mathematical model for GBM tumor growth that accounts for the differences in cell cycle regulation between these three GBM subtypes.

Abstract High-grade glioma continues to have dismal survival owing in part to its intra- and inter-patient heterogeneity. Standard clinical protocol collects tumor samples with the aim of providing or confirming a diagnosis and determining the status of a few key genes (e.g. IDH1, MGMT). However, this protocol is unable to capture the diversity within tumor regions, immune expression or normal cell abundances that play key roles in the development of the disease. To overcome this, during surgery we collect image-localized multi-regional biopsies to characterize this disease heterogeneity. Data collection is ongoing, and we currently have 202 samples from 58 patients with available bulk RNA-Seq. With a single-cell reference dataset from Columbia University, we used CIBERSORTx, a deconvolution method, to predict relative abundances of 7 normal, 6 glioma, and 5 immune cell subpopulations for each sample. We used Cox proportional hazard models with first-order statistics (mean, minimum, maximum) of abundances within patients to determine whether these are prognostic, then used TCGA RNA-Seq data to validate these findings. We found that one glioma proneural subtype was significantly beneficial for patient survival relative to other glioma subtypes across all statistics (and showed significance in TCGA). Proliferative and mesenchymal glioma subtypes also showed significance for one or two of the calculated statistics. Oligodendrocyte progenitor cell abundances were consistently significantly beneficial for patient survival in our cohort, while an increased abundance of abnormal (reactive) astrocytes is associated with poor prognosis. We also ran these analyses within invasive margin and core tumor regions to determine location-specific population abundances associated with patient survival. In conclusion, understanding the in vivo diversity of cellular subpopulations within high grade glioma is important for treatment stratification and patient benefit.

Abstract Identification of key phenotypic regions such as necrosis, contrast enhancement, and edema on magnetic resonance imaging (MRI) is important for understanding disease evolution and treatment response in patients with glioma. Manual delineation is time intensive and not feasible for a clinical workflow. Automating phenotypic region segmentation overcomes many issues with manual segmentation, however, current glioma segmentation datasets focus on pre-treatment, diagnostic scans, where treat-ment effects and surgical cavities are not present. Thus, existing automatic segmenta-tion models are not applicable to post-treatment imaging that is used for longitudinal evaluation of care. Here, we present a comparison of three-dimensional convolutional neural networks (nnU-Net architecture) trained on large temporally defined pre-treatment, post-treatment, and mixed cohorts. We used a total of 1563 imaging timepoints from 854 patients curated from 13 different institutions as well as diverse public data sets to understand the capabilities and limitations of automatic segmenta-tion on glioma images with different phenotypic and treatment appearance. We as-sessed the performance of models using Dice coefficients on test cases from each group comparing predictions with manual segmentations generated by trained techni-cians. We demonstrate that training a combined model can be as effective as models trained on just one temporal group. The results highlight the importance of a diverse training set, that includes images from the course of disease and with effects from treatment, in the creation of a model that can accurately segment glioma MRIs at multiple treatment time points.

Abstract Image-based mathematical modeling is emerging as an advantageous tool to integrate into the care routine for patients with brain tumors. These tools can be used for a variety of tasks including predicting tumor growth, indicating spatial genomic alterations, and guiding radiation therapy. Often, the methods to produce these maps are developed in silos, using cohorts of retrospective patients and switching to a per patient map calculation can be an afterthought. Automatic pipelines are needed to enable creating mathematical maps in a clinically-relevant workflow. To this end, we developed a pipeline to create an image-based mathematical map of edema abundance based on T2-weighted (T2W) magnetic resonance imaging that includes preprocessing steps, abnormality and tissue type segmentation, brain extraction, map generation, and conversion between image file formats. The python-based pipeline was designed to take in T1-weighted (T1W), T1W with gadolinium contrast (T1Gd), T2W, and fluid attenuated inversion recovery (FLAIR) DICOM images for a single patient as input and convert them to NIfTIs for downstream steps. The pipeline preprocessed images by correcting bias field fluctuations using the N4 algorithm and normalizing intensities. The T1W, T1Gd, and FLAIR are then registered to the T2W. T1Gd and FLAIR are used in a deep learning algorithm to delineate the brain and abnormality including enhancing tumor, necrosis, and peripheral edema. Normal tissue segmentations are performed using the statistical parametric mapping (SPM) segmentation routine. The segmentations and preprocessed images are all utilized in the edema mathematical model. The resulting edema map is then reformatted as a DICOM and output for further use by clinicians and clinical imaging systems. Our future goals include integrating our pipeline in a clinical workstation and sending final maps to our hospital’s picture archiving and communication system (PACS) to allow for deployment of these models for patient care in a clinically efficient timeline.

Abstract BACKGROUND Treatment of high-grade glioma (HGG) remains a challenge. Several groups have developed ex vivo drug sensitivity assays to evaluate susceptibility of the tumor tissue to specific chemotherapeutics. This tool, in addition to molecular analysis of tumors might provide another layer of confidence in choosing treatments for patients with high-grade gliomas. METHODS 3D PredictTM platform (Kiyatec, Greenville, SC, USA) was tested at three academic medical centers in the United States. Twelve patients with newly diagnosed and recurrent HGGs were included. Tumor spheroids created ex vivo were exposed to twelve pre-selected drugs. Responses were reported as: no response, moderate response and response. We analyzed: demographics, histopathology, stage of disease, success rate and results of the ex vivo assay and clinical implications. RESULTS Majority of patients were women (67%). Median age was 58.5. Histology: GBM-83%, HGGs-17%. Eighty three percent of the samples represented recurrent disease. Sixty percent represented first recurrence while the remaining 40% represented 2nd or 3rd recurrence. Eight out of 12 tissue samples (67%) were successfully tested ex vivo. In six cases (75%), the information was used for immediate clinical decision making. In 2 cases, the decision was deferred until the next progression of the disease. The drugs chosen for treatment based on assay results included: olaparib (the assay was positive for response to rucaparib), temozolomide, lomustine, carboplatin and irinotecan. CONCLUSIONS This platform is easy to use in a standard clinical setting, the success rate for creating viable spheroids for drug sensitivity testing is high and the information obtained can be useful in clinical practice. Addition of other agents to the assay, correlation with molecular testing of the tumor tissue and attempt to assess response to drug combinations should be considered in further development of this technology. Prospective clinical trial is needed to validate this approach for routine clinical use.

Abstract AIMS High-grade glioma continues to have dismal survival with current standard-of-care treatment, owing in part to its intra- and inter-patient heterogeneity. Typical diagnostic biopsies are taken from the dense tumor core to determine the presence of abnormal cells and the status of a few key genes (e.g. IDH1, MGMT). However, the tumor core is typically resected, leaving behind possibly genetically, transcriptomically and/or phenotypically distinct invasive margins that repopulate the disease. As these remaining populations are the ones ultimately being treated, it is important to know their compositional differences from the tumor core. We aim to identify the phenotypic niches defined by the relative composition of key cellular populations and understand their variation amongst patients. METHOD We have established an image-localized research biopsy study, that samples from both the invasive margin and tumor core. From this protocol, we currently have 202 samples from 58 patients with available bulk RNA-Seq, collected between Mayo Clinic and Barrow Neurological Institute. Using a single-cell reference dataset from our collaborators at Columbia University, we used CIBERSORTx, a deconvolution method, to predict relative abundances of 7 normal, 6 glioma, and 5 immune cell states for each sample. RESULTS We find that these cell state abundances connect to patient survival and show regional differences. For example, proneural glioma states were higher in invasive regions, whereas proliferative and mesenchymal states were higher in the tumor core. CONCLUSIONS Our analysis demonstrates a need to characterize the residual tissue following glioma resection to better understand the recurrent disease.

Abstract BACKGROUND Treatment of recurrent high-grade glioma remains a clinical challenge. There are very few agents active in this disease and short of enrollment in a clinical trial, the choice of therapy for a patient at the time of recurrence remains challenging. Several groups, including our own have developed ex vivo drug sensitivity assays to evaluate susceptibility of the tumor tissue to specific chemotherapeutics. To our knowledge, there is only one commercially available platform from the US based company Kiyatec. This tool, in addition to the molecular analysis of the tumor might provide another layer of confidence in choosing treatments for patients with recurrent high-grade gliomas. MATERIAL AND METHODS We present the data from our initial experience with the 3D PredictTM platform (Kiyatec, Greenville, SC, USA). We used the platform for 6 patients with recurrent high-grade gliomas undergoing tumor resection at the time of recurrence. We followed the company protocol for obtaining, preserving and shipping the tissue. RESULTS Out of the 6 specimens shipped for analysis, 4 assays were completed with full results available for analysis and clinical use. In two cases, the ex vivo tissue expansion was unsuccessful and drug testing was not completed. For the remaining 4 cases, spheroids created from the expansion of the harvested tissue were exposed to 12 drugs (temozolomide, lomustine, procarbazine, abemaciclib, carboplatin, dabrafenib, etoposide, everolimus, irinotecan, osimertinib, rucaparib and trametinib). We have observed different levels of sensitivity to several agents graded as: NO RESPONSE, MODERATE RESPONSE and RESPONSE. In 3 cases the results were used for clinical decision making and 2 patients were started on temozlomidede and carboplatin respectively and remain on treatment. In one case where sensitivity to etoposide was detected, we were not able to proceed with therapy due to insurance denial for the chosen drug. CONCLUSION In our experience, this platform is easy to use in a standard clinical setting, the success rate for creating viable spheroids for testing is high and the information derived from the drug sensitivity testing can be useful in clinical practice. Large, prospective clinical trial is need to validate this approach for routine clinical use.

Abstract Glioblastoma is a uniquely challenging and aggressive cancer that is practically considered uniformly fatal. This aggressiveness is driven by heterogeneity between and within patients and the diffuse invasion of tumor cells deep into the normal appearing brain tissue surrounding the frank tumor abnormality. Mathematical neuro-oncology (MNO) is a burgeoning field that seeks to bring together individual patient data (like patient imaging, tissue biopsies etc) to tune mathematical models that accurately predict and quantify response to treatments for each patient. These mathematical models use both mechanistic “weather forecasting” methods and artificial intelligence pattern recognition methods. Ultimately, these models form the basis of modern "precision medicine" approaches to tailor therapy in a patient-specific manner. These models can be used to overcome the limited ability of imaging to accurately detect tumor cells, improve prognostic predictions, stratify patients, and assess treatment response in silico in the construction of effective clinical trials and treatment protocols, thus accelerating the pace of clinical research. This presentation will focus on the growing translation of MNO to clinical neuro-oncology highlighting burgeoning insights into sex differences in tumor incidence, outcomes, and response to therapy. Specifically, I will discuss 1) use of our 4,500+ patient cohorts of 100,000+ annotated serial MR imaging data to develop digital twins accurately predict response and outcomes in patients and 2) our unique collection of 1000+ image-localized biopsies across nearly 200 patients to detect sex-distinct patterns of mapping between regional imaging features and underlying regional tumor biology. We will demonstrate the value of these novel combined resources in predicting response to treatment for EGFR-targeted therapies in 3 clinical trials and differentiating response to novel dendritic cells vaccines and CAR T-cell therapies in 2 additional clinical trials. Taken together, these complimentary approaches provide a platform to incorporate inter- and intra- tumoral heterogeneity into accurate predictive models that advance precision oncology for glioblastoma patients. Citation Format: Kristin R. Swanson. Composing each patient’s equation: Unraveling brain tumor complexity by integrating multi-regional image-localized biopsies. [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 SY30-03.

Abstract High grade gliomas (HGG) are aggressive primary brain malignancies typified by diffuse invasion, genetic heterogeneity, and a universally fatal outcome. MRI-defined contrast-enhancing (CE) tumor burden serves as the clinical standard that guides maximal surgical resection and post-therapy response assessment. However, HGGs also comprise an invasive non-enhancing (NE) tumor margin that extends beyond the CE core and harbors the cells that contribute to recurrence. Sampling restrictions have hindered the comprehensive study of these NE HGG cell populations driving tumor progression. Herein, we present an integrated multi-omic analysis of 313 spatially matched multi-regional CE and NE tumor biopsies from 68 HGG patients, performing whole exome and RNA sequencing of both IDH wild-type and IDH mutant HGGs. We report spatially restricted molecular profiles in IDH-mutant HGG, highlighting a concern for sampling bias given the importance of molecular diagnosis and prognostication in IDH-mutant HGG. Regardless of IDH status, we found that NE tumor regions harbored the highest proportion of private mutations, which reflects an increased development of regional genomic complexity in infiltrative tumor. The multiregional genomic profiling of our IDH wild-type HGG cohort reveals that EGFR and NF1 somatic alterations occur as mutually exclusive events in 98.7% of tumors. However, we resolved rare low allele frequency co-alterations of EGFR and NF1 within the NE region. We find this co-occurrence enriched in recurrent tumors, pointing to the early emergence of NF1 inactivation in the NE regions. We constructed genomic models predictive of recurrent disease from both NE and CE regions, which highlight the occurrence of clonal EGFR copy number alterations and NF1 loss as clonal or subclonal events, respectively, emphasizing the regional and temporal complexity of well-studied canonical driver alterations. We detailed the spatially unique acquisition of multiple distinct EGFR alterations giving rise to intra-tumoral EGFR mosaicism, a challenge in the implementation of EGFR directed therapies. Our study also identified two transcriptomic clusters delineated by the significant overrepresentation of neuronal (NEU) and glycolytic/plurimetabolic (GPM) pathway-based functional states in the NE region. NE regions of the NEU subtype harbor the greatest proportion of private mutations, suggesting these infiltrative tumor cells accumulate alterations without clonal expansion. GPM populations conversely displayed a less branched phylogeny and were transcriptionally enriched in immune cell signatures. This phenotypic dichotomy between GPM and NEU populations supports the growing body of evidence that invasive GBM cells either take on a neuronal phenotype for active invasion or a more metabolic phenotype involving interaction with astrocytes, other glial cells, and infiltrating immune cells. Citation Format: Mylan R. Blomquist, Leland S. Hu, Fulvio D'Angelo, Taylor M. Weiskittel, Francesca P. Caruso, Shannon P. Fortin Ensign, Christopher Sereduk, Gustavo De Leon, Lee Curtin, Javier Urcuyo, Ashlynn Gonzalez, Ashley Nespodzany, Teresa Noviello, Jennifer M. Eschbacher, Kris A. Smith, Peter Nakaji, Bernard R. Bendok, Richard S. Zimmerman, Chandan Krishna, Devi Patra, Naresh Patel, Mark Lyons, Kliment Donev, Maciej Mrugala, Alyx Porter, Anna Lasorella, Kristin R. Swanson, Michele Ceccarelli, Antonio Iavarone, Nhan L. Tran. Multiregional sampling of high grade glioma identifies regional biologic signatures [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 1507.