A deep learning image-based intrinsic molecular subtype classifier of breast tumors reveals tumor heterogeneity that may affect survival.

BACKGROUND: Breast cancer intrinsic molecular subtype (IMS) as classified by the expression-based PAM50 assay is considered a strong prognostic feature, even when controlled for by standard clinicopathological features such as age, grade, and nodal status, yet the molecular testing required to elucidate these subtypes is not routinely performed. Furthermore, when such bulk assays as RNA sequencing are performed, intratumoral heterogeneity that may affect prognosis and therapeutic decision-making can be missed. METHODS: As a more facile and readily available method for determining IMS in breast cancer, we developed a deep learning approach for approximating PAM50 intrinsic subtyping using only whole-slide images of H&E-stained breast biopsy tissue sections. This algorithm was trained on images from 443 tumors that had previously undergone PAM50 subtyping to classify small patches of the images into four major molecular subtypes-Basal-like, HER2-enriched, Luminal A, and Luminal B-as well as Basal vs. non-Basal. The algorithm was subsequently used for subtype classification of a held-out set of 222 tumors. RESULTS: This deep learning image-based classifier correctly subtyped the majority of samples in the held-out set of tumors. However, in many cases, significant heterogeneity was observed in assigned subtypes across patches from within a single whole-slide image. We performed further analysis of heterogeneity, focusing on contrasting Luminal A and Basal-like subtypes because classifications from our deep learning algorithm-similar to PAM50-are associated with significant differences in survival between these two subtypes. Patients with tumors classified as heterogeneous were found to have survival intermediate between Luminal A and Basal patients, as well as more varied levels of hormone receptor expression patterns. CONCLUSIONS: Here, we present a method for minimizing manual work required to identify cancer-rich patches among all multiscale patches in H&E-stained WSIs that can be generalized to any indication. These results suggest that advanced deep machine learning methods that use only routinely collected whole-slide images can approximate RNA-seq-based molecular tests such as PAM50 and, importantly, may increase detection of heterogeneous tumors that may require more detailed subtype analysis.

Integrative clustering reveals a novel split in the luminal A subtype of breast cancer with impact on outcome.

BACKGROUND: Breast cancer is a heterogeneous disease at the clinical and molecular level. In this study we integrate classifications extracted from five different molecular levels in order to identify integrated subtypes. METHODS: Tumor tissue from 425 patients with primary breast cancer from the Oslo2 study was cut and blended, and divided into fractions for DNA, RNA and protein isolation and metabolomics, allowing the acquisition of representative and comparable molecular data. Patients were stratified into groups based on their tumor characteristics from five different molecular levels, using various clustering methods. Finally, all previously identified and newly determined subgroups were combined in a multilevel classification using a "cluster-of-clusters" approach with consensus clustering. RESULTS: Based on DNA copy number data, tumors were categorized into three groups according to the complex arm aberration index. mRNA expression profiles divided tumors into five molecular subgroups according to PAM50 subtyping, and clustering based on microRNA expression revealed four subgroups. Reverse-phase protein array data divided tumors into five subgroups. Hierarchical clustering of tumor metabolic profiles revealed three clusters. Combining DNA copy number and mRNA expression classified tumors into seven clusters based on pathway activity levels, and tumors were classified into ten subtypes using integrative clustering. The final consensus clustering that incorporated all aforementioned subtypes revealed six major groups. Five corresponded well with the mRNA subtypes, while a sixth group resulted from a split of the luminal A subtype; these tumors belonged to distinct microRNA clusters. Gain-of-function studies using MCF-7 cells showed that microRNAs differentially expressed between the luminal A clusters were important for cancer cell survival. These microRNAs were used to validate the split in luminal A tumors in four independent breast cancer cohorts. In two cohorts the microRNAs divided tumors into subgroups with significantly different outcomes, and in another a trend was observed. CONCLUSIONS: The six integrated subtypes identified confirm the heterogeneity of breast cancer and show that finer subdivisions of subtypes are evident. Increasing knowledge of the heterogeneity of the luminal A subtype may add pivotal information to guide therapeutic choices, evidently bringing us closer to improved treatment for this largest subgroup of breast cancer.

The UCSC Interaction Browser: multidimensional data views in pathway context.

High-throughput data sets such as genome-wide protein-protein interactions, protein-DNA interactions and gene expression data have been published for several model systems, especially for human cancer samples. The University of California, Santa Cruz (UCSC) Interaction Browser (http://sysbio.soe.ucsc.edu/nets) is an online tool for biologists to view high-throughput data sets simultaneously for the analysis of functional relationships between biological entities. Users can access several public interaction networks and functional genomics data sets through the portal as well as upload their own networks and data sets for analysis. Users can navigate through correlative relationships for focused sets of genes belonging to biological pathways using a standard web browser. Using a new visual modality called the CircleMap, multiple 'omics' data sets can be viewed simultaneously within the context of curated, predicted, directed and undirected regulatory interactions. The Interaction Browser provides an integrative viewing of biological networks based on the consensus of many observations about genes and their products, which may provide new insights about normal and disease processes not obvious from any isolated data set.

Integrated molecular profiles of invasive breast tumors and ductal carcinoma in situ (DCIS) reveal differential vascular and interleukin signaling.

We use an integrated approach to understand breast cancer heterogeneity by modeling mRNA, copy number alterations, microRNAs, and methylation in a pathway context utilizing the pathway recognition algorithm using data integration on genomic models (PARADIGM). We demonstrate that combining mRNA expression and DNA copy number classified the patients in groups that provide the best predictive value with respect to prognosis and identified key molecular and stromal signatures. A chronic inflammatory signature, which promotes the development and/or progression of various epithelial tumors, is uniformly present in all breast cancers. We further demonstrate that within the adaptive immune lineage, the strongest predictor of good outcome is the acquisition of a gene signature that favors a high T-helper 1 (Th1)/cytotoxic T-lymphocyte response at the expense of Th2-driven humoral immunity. Patients who have breast cancer with a basal HER2-negative molecular profile (PDGM2) are characterized by high expression of protumorigenic Th2/humoral-related genes (24-38%) and a low Th1/Th2 ratio. The luminal molecular subtypes are again differentiated by low or high FOXM1 and ERBB4 signaling. We show that the interleukin signaling profiles observed in invasive cancers are absent or weakly expressed in healthy tissue but already prominent in ductal carcinoma in situ, together with ECM and cell-cell adhesion regulating pathways. The most prominent difference between low and high mammographic density in healthy breast tissue by PARADIGM was that of STAT4 signaling. In conclusion, by means of a pathway-based modeling methodology (PARADIGM) integrating different layers of molecular data from whole-tumor samples, we demonstrate that we can stratify immune signatures that predict patient survival.

Learning subgroup-specific regulatory interactions and regulator independence with PARADIGM.

UNLABELLED: High-dimensional '-omics' profiling provides a detailed molecular view of individual cancers; however, understanding the mechanisms by which tumors evade cellular defenses requires deep knowledge of the underlying cellular pathways within each cancer sample. We extended the PARADIGM algorithm (Vaske et al., 2010, Bioinformatics, 26, i237-i245), a pathway analysis method for combining multiple '-omics' data types, to learn the strength and direction of 9139 gene and protein interactions curated from the literature. Using genomic and mRNA expression data from 1936 samples in The Cancer Genome Atlas (TCGA) cohort, we learned interactions that provided support for and relative strength of 7138 (78%) of the curated links. Gene set enrichment found that genes involved in the strongest interactions were significantly enriched for transcriptional regulation, apoptosis, cell cycle regulation and response to tumor cells. Within the TCGA breast cancer cohort, we assessed different interaction strengths between breast cancer subtypes, and found interactions associated with the MYC pathway and the ER alpha network to be among the most differential between basal and luminal A subtypes. PARADIGM with the Naive Bayesian assumption produced gene activity predictions that, when clustered, found groups of patients with better separation in survival than both the original version of PARADIGM and a version without the assumption. We found that this Naive Bayes assumption was valid for the vast majority of co-regulators, indicating that most co-regulators act independently on their shared target. AVAILABILITY: http://paradigm.five3genomics.com. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

The UCSC Cancer Genomics Browser: update 2011.

The UCSC Cancer Genomics Browser (https://genome-cancer.ucsc.edu) comprises a suite of web-based tools to integrate, visualize and analyze cancer genomics and clinical data. The browser displays whole-genome views of genome-wide experimental measurements for multiple samples alongside their associated clinical information. Multiple data sets can be viewed simultaneously as coordinated 'heatmap tracks' to compare across studies or different data modalities. Users can order, filter, aggregate, classify and display data interactively based on any given feature set including clinical features, annotated biological pathways and user-contributed collections of genes. Integrated standard statistical tools provide dynamic quantitative analysis within all available data sets. The browser hosts a growing body of publicly available cancer genomics data from a variety of cancer types, including data generated from the Cancer Genome Atlas project. Multiple consortiums use the browser on confidential prepublication data enabled by private installations. Many new features have been added, including the hgMicroscope tumor image viewer, hgSignature for real-time genomic signature evaluation on any browser track, and 'PARADIGM' pathway tracks to display integrative pathway activities. The browser is integrated with the UCSC Genome Browser; thus inheriting and integrating the Genome Browser's rich set of human biology and genetics data that enhances the interpretability of the cancer genomics data.

Inference of patient-specific pathway activities from multi-dimensional cancer genomics data using PARADIGM.

MOTIVATION: High-throughput data is providing a comprehensive view of the molecular changes in cancer tissues. New technologies allow for the simultaneous genome-wide assay of the state of genome copy number variation, gene expression, DNA methylation and epigenetics of tumor samples and cancer cell lines. Analyses of current data sets find that genetic alterations between patients can differ but often involve common pathways. It is therefore critical to identify relevant pathways involved in cancer progression and detect how they are altered in different patients. RESULTS: We present a novel method for inferring patient-specific genetic activities incorporating curated pathway interactions among genes. A gene is modeled by a factor graph as a set of interconnected variables encoding the expression and known activity of a gene and its products, allowing the incorporation of many types of omic data as evidence. The method predicts the degree to which a pathway's activities (e.g. internal gene states, interactions or high-level 'outputs') are altered in the patient using probabilistic inference. Compared with a competing pathway activity inference approach called SPIA, our method identifies altered activities in cancer-related pathways with fewer false-positives in both a glioblastoma multiform (GBM) and a breast cancer dataset. PARADIGM identified consistent pathway-level activities for subsets of the GBM patients that are overlooked when genes are considered in isolation. Further, grouping GBM patients based on their significant pathway perturbations divides them into clinically-relevant subgroups having significantly different survival outcomes. These findings suggest that therapeutics might be chosen that target genes at critical points in the commonly perturbed pathway(s) of a group of patients. AVAILABILITY: Source code available at http://sbenz.github.com/Paradigm,. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Identification and validation of expressed HLA-binding breast cancer neoepitopes for potential use in individualized cancer therapy.

BACKGROUND: Therapeutic regimens designed to augment the immunological response of a patient with breast cancer (BC) to tumor tissue are critically informed by tumor mutational burden and the antigenicity of expressed neoepitopes. Herein we describe a neoepitope and cognate neoepitope-reactive T-cell identification and validation program that supports the development of next-generation immunotherapies. METHODS: Using GPS Cancer, NantOmics research, and The Cancer Genome Atlas databases, we developed a novel bioinformatic-based approach which assesses mutational load, neoepitope expression, human leukocyte antigen (HLA)-binding prediction, and in vitro confirmation of T-cell recognition to preferentially identify targetable neoepitopes. This program was validated by application to a BC cell line and confirmed using tumor biopsies from two patients with BC enrolled in the Tumor-Infiltrating Lymphocytes and Genomics (TILGen) study. RESULTS: The antigenicity and HLA-A2 restriction of the BC cell line predicted neoepitopes were determined by reactivity of T cells from HLA-A2-expressing healthy donors. For the TILGen subjects, tumor-infiltrating lymphocytes (TILs) recognized the predicted neoepitopes both as peptides and on retroviral expression in HLA-matched Epstein-Barr virus-lymphoblastoid cell line and BC cell line MCF-7 cells; PCR clonotyping revealed the presence of T cells in the periphery with T-cell receptors for the predicted neoepitopes. These high-avidity immune responses were polyclonal, mutation-specific and restricted to either HLA class I or II. Interestingly, we observed the persistence and expansion of polyclonal T-cell responses following neoadjuvant chemotherapy. CONCLUSIONS: We demonstrate our neoepitope prediction program allows for the successful identification of neoepitopes targeted by TILs in patients with BC, providing a means to identify tumor-specific immunogenic targets for individualized treatment, including vaccines or adoptively transferred cellular therapies.

A factor graph nested effects model to identify networks from genetic perturbations.

Complex phenotypes such as the transformation of a normal population of cells into cancerous tissue result from a series of molecular triggers gone awry. We describe a method that searches for a genetic network consistent with expression changes observed under the knock-down of a set of genes that share a common role in the cell, such as a disease phenotype. The method extends the Nested Effects Model of Markowetz et al. (2005) by using a probabilistic factor graph to search for a network representing interactions among these silenced genes. The method also expands the network by attaching new genes at specific downstream points, providing candidates for subsequent perturbations to further characterize the pathway. We investigated an extension provided by the factor graph approach in which the model distinguishes between inhibitory and stimulatory interactions. We found that the extension yielded significant improvements in recovering the structure of simulated and Saccharomyces cerevisae networks. We applied the approach to discover a signaling network among genes involved in a human colon cancer cell invasiveness pathway. The method predicts several genes with new roles in the invasiveness process. We knocked down two genes identified by our approach and found that both knock-downs produce loss of invasive potential in a colon cancer cell line. Nested effects models may be a powerful tool for inferring regulatory connections and genes that operate in normal and disease-related processes.

Transcriptomic silencing as a potential mechanism of treatment resistance.

Next-generation sequencing (NGS) has not revealed all the mechanisms underlying resistance to genomically matched drugs. Here, we performed in 1417 tumors whole-exome tumor (somatic)/normal (germline) NGS and whole-transcriptome sequencing, the latter focusing on a clinically oriented 50-gene panel in order to examine transcriptomic silencing of putative driver alterations. In this large-scale study, approximately 13% of the somatic single nucleotide variants (SNVs) were unexpectedly not expressed as RNA; 23% of patients had ≥1 nonexpressed SNV. SNV-bearing genes consistently transcribed were TP53, PIK3CA, and KRAS; those with lower transcription rates were ALK, CSF1R, ERBB4, FLT3, GNAS, HNF1A, KDR, PDGFRA, RET, and SMO. We also determined the frequency of tumor mutations being germline, rather than somatic, in these and an additional 462 tumors with tumor/normal exomes; 33.8% of germline SNVs within the gene panel were rare (not found after filtering through variant information domains) and at risk of being falsely reported as somatic. Both the frequency of silenced variant transcription and the risk of falsely identifying germline mutations as somatic/tumor related are important phenomena. Therefore, transcriptomics is a critical adjunct to genomics when interrogating patient tumors for actionable alterations, because, without expression of the target aberrations, there will likely be therapeutic resistance.

A20 and ABIN1 Suppression of a Keratinocyte Inflammatory Program with a Shared Single-Cell Expression Signature in Diverse Human Rashes.

Genetic variation in the NF-κB inhibitors, ABIN1 and A20, increase risk for psoriasis. While critical for hematopoietic immune cell function, these genes are believed to additionally inhibit psoriasis by dampening inflammatory signaling in keratinocytes. We dissected ABIN1 and A20's regulatory role in human keratinocyte inflammation using an RNA sequencing-based comparative genomic approach. Here we show subsets of the IL-17 and tumor necrosis factor-α signaling pathways are robustly restricted by A20 overexpression. In contrast, ABIN1 overexpression inhibits these genes more modestly for IL-17, and weakly for tumor necrosis factor-α. Our genome-scale analysis also indicates that inflammatory program suppression appears to be the major transcriptional influence of A20/ABIN1 overexpression, without obvious influence on keratinocyte viability genes. Our findings thus enable dissection of the differing anti-inflammatory mechanisms of two distinct psoriasis modifiers, which may be directly exploited for therapeutic purposes. Importantly, we report that IL-17-induced targets of A20 show similar aberrant epidermal layer-specific transcriptional upregulation in keratinocytes from diseases as diverse as psoriasis, atopic dermatitis, and erythrokeratodermia variabilis, suggesting a contributory role for epidermal inflammation in a broad spectrum of rashes.

Cell of origin and mutation pattern define three clinically distinct classes of sebaceous carcinoma.

Sebaceous carcinomas (SeC) are cutaneous malignancies that, in rare cases, metastasize and prove fatal. Here we report whole-exome sequencing on 32 SeC, revealing distinct mutational classes that explain both cancer ontogeny and clinical course. A UV-damage signature predominates in 10/32 samples, while nine show microsatellite instability (MSI) profiles. UV-damage SeC exhibited poorly differentiated, infiltrative histopathology compared to MSI signature SeC (p = 0.003), features previously associated with dissemination. Moreover, UV-damage SeC transcriptomes and anatomic distribution closely resemble those of cutaneous squamous cell carcinomas (SCC), implicating sun-exposed keratinocytes as a cell of origin. Like SCC, this UV-damage subclass harbors a high somatic mutation burden with >50 mutations per Mb, predicting immunotherapeutic response. In contrast, ocular SeC acquires far fewer mutations without a dominant signature, but show frequent truncations in the ZNF750 epidermal differentiation regulator. Our data exemplify how different mutational processes convergently drive histopathologically related but clinically distinct cancers.

Transcriptional Programming of Normal and Inflamed Human Epidermis at Single-Cell Resolution.

Perturbations in the transcriptional programs specifying epidermal differentiation cause diverse skin pathologies ranging from impaired barrier function to inflammatory skin disease. However, the global scope and organization of this complex cellular program remain undefined. Here we report single-cell RNA sequencing profiles of 92,889 human epidermal cells from 9 normal and 3 inflamed skin samples. Transcriptomics-derived keratinocyte subpopulations reflect classic epidermal strata but also sharply compartmentalize epithelial functions such as cell-cell communication, inflammation, and WNT pathway modulation. In keratinocytes, ∼12% of assessed transcript expression varies in coordinate patterns, revealing undescribed gene expression programs governing epidermal homeostasis. We also identify molecular fingerprints of inflammatory skin states, including S100 activation in the interfollicular epidermis of normal scalp, enrichment of a CD1C+CD301A+ myeloid dendritic cell population in psoriatic epidermis, and IL1ßhiCCL3hiCD14+ monocyte-derived macrophages enriched in foreskin. This compendium of RNA profiles provides a critical step toward elucidating epidermal diseases of development, differentiation, and inflammation.

Voltage-gated Na+ channel SCN5A is a key regulator of a gene transcriptional network that controls colon cancer invasion.

Voltage-gated Na(+) channels (VGSC) have been implicated in the metastatic potential of human breast, prostate, and lung cancer cells. Specifically, the SCN5A gene encoding the VGSC isotype Na(v)1.5 has been defined as a key driver of human cancer cell invasion. In this study, we examined the expression and function of VGSCs in a panel of colon cancer cell lines by electrophysiologic recordings. Na(+) channel activity and invasive potential were inhibited pharmacologically by tetrodotoxin or genetically by small interfering RNAs (siRNA) specifically targeting SCN5A. Clinical relevance was established by immunohistochemistry of patient biopsies, with strong Na(v)1.5 protein staining found in colon cancer specimens but little to no staining in matched-paired normal colon tissues. We explored the mechanism of VGSC-mediated invasive potential on the basis of reported links between VGSC activity and gene expression in excitable cells. Probabilistic modeling of loss-of-function screens and microarray data established an unequivocal role of VGSC SCN5A as a high level regulator of a colon cancer invasion network, involving genes that encompass Wnt signaling, cell migration, ectoderm development, response to biotic stimulus, steroid metabolic process, and cell cycle control. siRNA-mediated knockdown of predicted downstream network components caused a loss of invasive behavior, demonstrating network connectivity and its function in driving colon cancer invasion.