Abstract Advancements in state-of-the-art molecular profiling techniques has resulted in better understanding of pediatric cancers and their drivers. Conversely, it also became apparent that pediatric cancers are much more heterogeneous than previously thought. Many new types and subtypes of pediatric cancers have been identified with distinct molecular and clinical characteristics. However, for most newly recognized entities there is no specific treatment available yet. The ITCC-P4 consortium is a collaboration between many academic centers across Europe and several pharmaceutical companies involved in preclinical testing, with the overall aim to establish a sustainable platform of >400 molecularly well-characterized PDX models of high-risk pediatric cancers and to use them for in vivo testing of novel mechanism-of-action based treatments. Currently, 340 models are fully established, including 87 brain and 253 non-brain tumor models, together representing different tumor types both from primary (113) and relapsed (92)/metastatic disease (42). 252 of these models have been fully molecularly characterized, representing 18 pediatric cancer entities and 43 different subtypes. Using low coverage whole-genome and whole exome sequencing, somatic mutation calling, DNA copy number, transcriptome analysis and methylation profiling we have observed that the molecular profile of most PDX models closely mimics their original tumors. Clonal evolution of somatic variants was only observed in some PDX-tumor pairs or so between disease states. Somatic copy number variant analysis highlights specific alterations for instance MYB, MYC, MYCN, NTRK3, PTEN loss differently distributed between PDX-patient tumor pairs in high-grade gliomas. Overall, our results show that we have established >250 PDX models of solid pediatric cancers, that well represents the disease spectrum and that is currently being used for in vivo testing of standard of care drugs and targeted small molecules. Treatment responses will be directly linked to molecular data to identify potential biomarkers for prioritization or deprioritization of individual, patient-specific specific drugs.
<p>Supplementary method, Supplementary figures and supplementary figure legends: Supplementary Figure 1. Viral VBIM integration site in MAML3 gene and equal sensitivity to cisplatin of parental and SD3.23 cells. Supplementary Figure 2. Overexpression of MAML3, but not MAML1 and MAML2, mediates resistance to retinoic acid in multiple neuroblastoma cell lines. Supplementary Figure 3. Induction of neuronal markers upon RA treatment is decreased in MAML3 overexpressing cells. Supplementary Figure 4. MAML3 binds to RXR. Supplementary Figure 5. IGF2 is upregulated in MAML3 overexpressing cells and results in hyperproliferation, but not RA resistance.</p>
Abstract Background: Gene expression-based subtyping is widely accepted as a relevant source of disease stratification. Despite the widespread use, its translational and clinical utility is hampered by discrepant results, likely related to differences in data processing and algorithms applied to diverse patient cohorts, sample preparation methods, and gene expression platforms. In the absence of a clear methodological gold standard to perform such analyses, a more general framework that integrates and compares multiple strategies is needed to define common disease patterns in a principled, unbiased manner. Methods: We formed a consortium of 6 independent experts groups - each with a previously published CRC classifier, ranging from 3 to 6 subtypes - to understand similarities and differences of their subtyping systems. Sage Bionetworks functioned as neutral party to aggregate public and proprietary data (Synapse platform) and perform meta-analysis. Each group applied its CRC subtyping signature to the collection of data sets with gene expression (n = 4,151, predominantly stage II and III). Using the resulting subtype labels, we developed a network-based model and applied a Markov cluster algorithm to detect robust network substructures that would indicate recurring subtype patterns and therefore a consensus subtyping system. Correlative analyses using clinico-pathological, genomic and epigenomic features was performed to robustly characterize the identified subtypes. Results: This analytical framework revealed significant interconnectivity between the six independent classification systems, leading to the identification of four biologically distinct consensus molecular subtypes (CMS) enriched for key pathway traits: CMS1 (MSI Immune), hypermutated, microsatellite unstable, with strong immune activation; CMS2 (Canonical), epithelial, chromosomally unstable, with marked WNT and MYC signaling activation; CMS3 (Metabolic), epithelial, with evident metabolic dysregulation; and CMS4 (Mesenchymal), prominent TGFβ activation, angiogenesis, stromal invasion. Patients diagnosed with MSI Immune tumors had worse survival after relapse and those with mesenchymal tumors had increased risk of metastasis and worse overall survival. Discussion: We describe a novel methodological paradigm for deriving benchmarks of disease subtyping. Our work represents the first example of a community of experts identifying and advocating for a single reproducible model for cancer subtyping, effectively unifying previous classifiers. In the CRC domain, the uniformity afforded by this new classification system and its application to a large data set revealed important subtype-specific biological associations that were previously unnoticed or marginally significant, supporting a new taxonomy of the disease. Citation Format: Justin Guinney, Rodrigo Dienstmann, Xin Wang, Aurelien de Reynies, Andreas Schlicker, Charlotte Soneson, Laetitia Marisa, Paul Roepman, Gift Nyamundanda, Paolo Angelino, Brian Bot, Jeffrey S. Morris, Iris Simon, Sarah Gerster, Evelyn Fessler, Felipe de Sousa e Melo, Edoardo Missiaglia, Hena Ramay, David Barras, Krisztian Homicsko, Dipen Maru, Ganiraju Manyam, Bradley Broom, Valerie Boige, Ted Laderas, Ramon Salazar, Joe W. Gray, Josep Tabernero, Rene Bernards, Stephen Friend, Pierre Laurent-Puig, Jan P. Medema, Anguraj Sadanandam, Lodewyk Wessels, Mauro Delorenzi, Scott Kopetz, Louis Vermeulen, Sabine Tejpar. Consensus molecular subtyping through a community of experts advances unsupervised gene expression-based disease classification and facilitates clinical translation. [abstract]. In: Proceedings of the 106th Annual Meeting of the American Association for Cancer Research; 2015 Apr 18-22; Philadelphia, PA. Philadelphia (PA): AACR; Cancer Res 2015;75(15 Suppl):Abstract nr 603. doi:10.1158/1538-7445.AM2015-603
206 Background: In prostate cancer, homologous recombination deficiency (HRD) is associated with poor prognosis, and sensitivity to DNA damaging agents and DNA damage repair (DDR) inhibitors. As new classes of DDR inhibitors become available, identifying patients with HRD will be critical for treatment selection. Here, we present machine learning (ML)-based models trained to predict HRD status directly from hematoxylin and eosin (H&E) whole slide images (WSI). Methods: ML models were trained to predict and segment cells and tissue regions within the tumor microenvironment (TME) using annotated (N=91,021 annotations) WSI of H&E-stained resections from the cancer genome atlas prostate adenocarcinoma (TCGA PRAD) dataset (N=401) and needle core biopsies from a proprietary dataset (N=1,000). Quantified Human Interpretable Features (HIFs) that describe the TME composition were extracted. Three models were trained to predict HRD status using 373 WSI with known HRD score (TCGA PRAD; train N=259, validation N=76, and test N=38). Two models used input from the TME model: An HIF multivariate logistic regression model, and a graph neural network (GNN) where predictions are based on the complex spatial relationships within the TME. An end-to-end (E2E) multiple instance learning model predicted directly from the WSI. Two cutoffs for HRD were defined using Gaussian Mixture Models, resulting in 99 WSI (train N=72, validation N=18, and test N=9) positive for the Genomic Instability (>16 events) cutoff, and 58 WSI (train N=44, validation N=10, test N=4) positive for the Genomic Instability (>22 events) cutoff. An independent validation set of 45 biopsies and 16 resections from a biobank of metastatic castration resistant prostate cancer with HRD status determined by whole-exome sequencing was compared to ML model H&E-based HRD prediction. Results: In the TCGA test set of resection samples, all three models moderately or strongly predicted HRD status, with the HIF model showing the best performance (AUROC 0.87, sensitivity 0.88, specificity 0.62). The same HIF model performed equally well (AUROC 0.85, Sensitivity 0.93, specificity 0.67) in the resection samples from the independent validation set. However, the model performance went down (AUROC 0.69, sensitivity 0.91, specificity: 0.3) when both resection and needle biopsy samples were included, highlighting the importance of a representative training set to achieve robust performance in a real world setting. Further model training and validation with a more diverse dataset is required to accurately assess the performance of the model on needle biopsies. Conclusions: ML models trained on resection prostate cancer samples performed well in predicting HRD status when applied to the same sample type, demonstrating the potential of ML models to predict genomic biomarkers status in surgical specimens for treatment decision.
<p>Supplementary Figures S1-S5 and Table S1. Fig. S1: BRCA2 and p53 deletion in KB2P tumors. Fig. S2: aCGH clustering of KB2P tumors. Fig. S3: Acquired resistant carcinomas retain their epithelial phenotype. Fig. S4: SAM analysis of KB2P tumors. Fig. S5: EMT score and ABCB1 expression in metaplastic breast cancers. Table S1: EMT signature genes in each data set.</p>
3530 Background: Unbiased genome-wide analyses of gene expression patterns have been successfully used for molecular classification of breast cancer into subtypes that have clear relevance for prognosis and treatment. A similar classification is still missing for colorectal cancer (CRC). Methods: Using full genome expression data of 188 stage I-IV CRC patients, an unsupervised clustering revealed three major subtypes (A-, B-, C-type). A molecular subtype classification was developed and validated in 543 stage II and III patients. The subtypes were analyzed for correlation to clinical information, mutations in the kinome, known molecular markers status and chemotherapy response. In addition, subtypes were determined on 173 samples from The Cancer Genome Atlas (TCGA) colon dataset with Agilent genome expression data. Results: C-type patients have the worst outcome, a mesenchymal phenotype, and show no benefit from adjuvant chemotherapy treatment. Patients having A- or B-type tumors have a better clinical outcome, a more proliferative and epithelial phenotype and benefit from adjuvant chemotherapy. A- and C-type groups are enriched for tumors having oncogenic BRAF mutations and a deficient DNA mismatch repair system. B-type tumors showed a low overall kinome mutation frequency (1.6%), while both A- and C-type patients harbor a higher mutation frequency (respectively 4.2 and 6.2%), in agreement with their mismatch repair deficiency. Finally, CRC subtyping was confirmed in the colon TCGA dataset with 26 samples classified as A-type, 110 as B-type and 37 as C-type. In agreement with the different aggressiveness of the subtypes, A-type tumors were less prevalent in stage IV while C-type were less prevalent in stage I CRC. Conclusions: The heterogeneity of the intrinsic subtypes is largely based on three biological hallmarks of the tumor: an epithelial-to-mesenchymal transition, deficiency in mismatch repair genes that result in a high mutation frequency associated with MSI, and cellular proliferation. These subtypes are clinically relevant, as they differ in their underlying biology and might require different treatment strategies.
