Deep learning-based computer vision methods have recently made remarkable breakthroughs in the analysis and classification of cancer pathology images. However, there has been relatively little investigation of the utility of deep neural networks to synthesize medical images. In this study, we evaluated the efficacy of generative adversarial networks (GANs) to synthesize high resolution pathology images of ten histological types of cancer, including five cancer types from The Cancer Genome Atlas (TCGA) and the five major histological subtypes of ovarian carcinoma. The quality of these images was assessed using a comprehensive survey of board-certified pathologists (n = 9) and pathology trainees (n = 6). Our results show that the real and synthetic images are classified by histotype with comparable accuracies, and the synthetic images are visually indistinguishable from real images. Furthermore, we trained deep convolutional neural networks (CNNs) to diagnose the different cancer types and determined that the synthetic images perform as well as additional real images when used to supplement a small training set. These findings have important applications in proficiency testing of medical practitioners and quality assurance in clinical laboratories. Furthermore, training of computer-aided diagnostic systems can benefit from synthetic images where labeled datasets are limited (e.g., rare cancers). We have created a publicly available website where clinicians and researchers can attempt questions from the image survey at http://gan.aimlab.ca/. This article is protected by copyright. All rights reserved.
The somatic missense point mutation c.402C>G (p.C134W) in the FOXL2 transcription factor is pathognomonic for adult-type granulosa cell tumors (AGCT) and a diagnostic marker for this tumor type. However, the molecular consequences of this mutation and its contribution to the mechanisms of AGCT pathogenesis remain unclear. To explore these mechanisms, we engineered V5-FOXL2WT- and V5-FOXL2C134W-inducible isogenic cell lines and performed ChIP-seq and transcriptome profiling. FOXL2C134W associated with the majority of the FOXL2 WT DNA elements as well as a large collection of unique elements genome-wide. This model enabled confirmation of altered DNA binding specificity for FOXL2C134W and identification of unique targets of FOXL2C134W including SLC35F2, whose expression increased sensitivity to YM155. Our results suggest FOXL2C134W drives AGCT by altering the binding affinity of FOXL2-containing complexes to engage an oncogenic transcriptional program.
Alignment-free classification tools have enabled high-throughput processing of sequencing data in many bioinformatics analysis pipelines primarily due to their computational efficiency. Originally k-mer based, such tools often lack sensitivity when faced with sequencing errors and polymorphisms. In response, some tools have been augmented with spaced seeds, which are capable of tolerating mismatches. However, spaced seeds have seen little practical use in classification because they bring increased computational and memory costs compared to methods that use k-mers. These limitations have also caused the design and length of practical spaced seeds to be constrained, since storing spaced seeds can be costly. To address these challenges, we have designed a probabilistic data structure called a multiindex Bloom Filter (miBF), which can store multiple spaced seed sequences with a low memory cost that remains static regardless of seed length or seed design. We formalize how to minimize the false-positive rate of miBFs when classifying sequences from multiple targets or references. Available within BioBloom Tools, we illustrate the utility of miBF in two use cases: read-binning for targeted assembly, and taxonomic read assignment. In our benchmarks, an analysis pipeline based on miBF shows higher sensitivity and specificity for read-binning than sequence alignment-based methods, also executing in less time. Similarly, for taxonomic classification, miBF enables higher sensitivity than a conventional spaced seed-based approach, while using half the memory and an order of magnitude less computational time.
Immunoglobulin (Ig) gene rearrangements and oncogenic translocations are routinely assessed during the characterization of B cell neoplasms and stratification of patients with distinct clinical and biological features, with the assessment done using Sanger sequencing, targeted next-generation sequencing, or fluorescence in situ hybridization (FISH). Currently, a complete Ig characterization cannot be extracted from whole-genome sequencing (WGS) data due to the inherent complexity of the Ig loci. Here, we introduce IgCaller, an algorithm designed to fully characterize Ig gene rearrangements and oncogenic translocations from short-read WGS data. Using a cohort of 404 patients comprising different subtypes of B cell neoplasms, we demonstrate that IgCaller identifies both heavy and light chain rearrangements to provide additional information on their functionality, somatic mutational status, class switch recombination, and oncogenic Ig translocations. Our data thus support IgCaller to be a reliable alternative to Sanger sequencing and FISH for studying the genetic properties of the Ig loci.
Structural variants (SVs) may be an underestimated cause of hereditary cancer syndromes given the current limitations of short-read next-generation sequencing. Here we investigated the utility of long-read sequencing in resolving germline SVs in cancer susceptibility genes detected through short-read genome sequencing.
The most aggressive B cell lymphomas frequently manifest extranodal distribution and carry somatic mutations in the poorly characterized gene TBL1XR1. Here, we show that TBL1XR1 mutations skew the humoral immune response toward generating abnormal immature memory B cells (MB), while impairing plasma cell differentiation. At the molecular level, TBL1XR1 mutants co-opt SMRT/HDAC3 repressor complexes toward binding the MB cell transcription factor (TF) BACH2 at the expense of the germinal center (GC) TF BCL6, leading to pre-memory transcriptional reprogramming and cell-fate bias. Upon antigen recall, TBL1XR1 mutant MB cells fail to differentiate into plasma cells and instead preferentially reenter new GC reactions, providing evidence for a cyclic reentry lymphomagenesis mechanism. Ultimately, TBL1XR1 alterations lead to a striking extranodal immunoblastic lymphoma phenotype that mimics the human disease. Both human and murine lymphomas feature expanded MB-like cell populations, consistent with a MB-cell origin and delineating an unforeseen pathway for malignant transformation of the immune system.
Objective: Heat shock protein 47 (HSP47) is a collagen-specific molecular chaperone that facilitates collagen maturation. Its role in cancer remains largely unknown. In this study, we investigated the roles of HSP47 in colorectal cancer (CRC) and therapy resistance. Methods: Expression of HSP47 in CRC tissues was examined (1) in paired human CRC/adjacent normal tissues, using real time quantitative reverse transcription polymerase chain reaction (qRT-PCR), The Cancer Genome Atlas (TCGA) database, and 22 independent microarray databases (curated CRC). In vitro studies on several CRC cell lines (HCT116, RKO and CCL228) with modulated HSP47 expression were conducted to assess cell viability and apoptosis (TUNEL assay and caspase-3/-7) during exposure to chemotherapy. AKT signaling and co-immunoprecipitation studies were performed to examine HSP47 and PHLPP1 interaction. In vivo studies using tumor xenografts were conducted to assess the effects of HSP47 modulation on tumor growth and therapy response. Results: HSP47 was upregulated in CRC and was associated with poor prognosis in individuals with CRC. In vitro, HSP47 overexpression supported the survival of CRC cells, whereas its knockdown sensitized cells to 5-fluorouracil (5-FU). HSP47 promoted survival by inhibiting apoptosis, enhancing AKT phosphorylation, and decreasing expression of the AKT-specific phosphatase PHLPP1 when cells were exposed to chemotherapy. These effects were partly results of the interaction between HSP47 and PHLPP1, which decreased PHLPP1 stability and led to more persistent AKT activity. In vivo, HSP47 supported tumor growth despite 5-FU treatment. Conclusions: HSP47 supports the growth of CRC tumors and suppresses the efficacy of chemotherapy via modulation of AKT signaling.
Diffuse large B-cell lymphoma (DLBCL) patients are typically treated with immunochemotherapy containing rituximab (rituximab, cyclophosphamide, hydroxydaunorubicin-vincristine (Oncovin), and prednisone [R-CHOP]); however, prognosis is extremely poor if R-CHOP fails. To identify genetic mechanisms contributing to primary or acquired R-CHOP resistance, we performed target-panel sequencing of 135 relapsed/refractory DLBCLs (rrDLBCLs), primarily comprising circulating tumor DNA from patients on clinical trials. Comparison with a metacohort of 1670 diagnostic DLBCLs identified 6 genes significantly enriched for mutations upon relapse. TP53 and KMT2D were mutated in the majority of rrDLBCLs, and these mutations remained clonally persistent throughout treatment in paired diagnostic-relapse samples, suggesting a role in primary treatment resistance. Nonsense and missense mutations affecting MS4A1, which encodes CD20, are exceedingly rare in diagnostic samples but show recurrent patterns of clonal expansion following rituximab-based therapy. MS4A1 missense mutations within the transmembrane domains lead to loss of CD20 in vitro, and patient tumors harboring these mutations lacked CD20 protein expression. In a time series from a patient treated with multiple rounds of therapy, tumor heterogeneity and minor MS4A1-harboring subclones contributed to rapid disease recurrence, with MS4A1 mutations as founding events for these subclones. TP53 and KMT2D mutation status, in combination with other prognostic factors, may be used to identify high-risk patients prior to R-CHOP for posttreatment monitoring. Using liquid biopsies, we show the potential to identify tumors with loss of CD20 surface expression stemming from MS4A1 mutations. Implementation of noninvasive assays to detect such features of acquired treatment resistance may allow timely transition to more effective treatment regimens.
Hematopoietic clones with leukemogenic mutations arise in healthy people as they age, but progression to acute myeloid leukemia (AML) is rare. Recent evidence suggests that the microenvironment may play an important role in modulating human AML population dynamics. To investigate this concept further, we examined the combined and separate effects of an oncogene (c-MYC) and exposure to IL3, GM-CSF and SCF on the experimental genesis of a human AML in xenografted immunodeficient mice. Initial experiments showed that normal human CD34+ blood cells transduced with a lentiviral MYC vector and then transplanted into immunodeficient mice produced a hierarchically organized, rapidly fatal and serially transplantable blast population, phenotypically and transcriptionally similar to human AML cells, but only in mice producing IL3, GM-CSF and SCF transgenically, or in regular mice in which the cells were exposed to IL3 or GM-CSF delivered using a co-transduction strategy. In their absence, the MYC+ human cells produced a normal repertoire of lymphoid and myeloid progeny in transplanted mice for many months but, upon transfer to secondary mice producing the human cytokines, the MYC+ cells rapidly generated AML. Indistinguishable diseases were also obtained efficiently from both primitive (CD34+CD38-) and late (GMPs) cells. These findings underscore the critical role that these cytokines can play in activating a malignant state in normally differentiating human hematopoietic cells in which MYC expression has been deregulated. They also introduce a robust experimental model of human leukemogenesis to further elucidate key mechanisms involved and test strategies to suppress them.