Predicting Survival After Hepatocellular Carcinoma Resection Using Deep Learning on Histological Slides.

BACKGROUND AND AIMS: Standardized and robust risk-stratification systems for patients with hepatocellular carcinoma (HCC) are required to improve therapeutic strategies and investigate the benefits of adjuvant systemic therapies after curative resection/ablation. APPROACH AND RESULTS: In this study, we used two deep-learning algorithms based on whole-slide digitized histological slides (whole-slide imaging; WSI) to build models for predicting survival of patients with HCC treated by surgical resection. Two independent series were investigated: a discovery set (Henri Mondor Hospital, n = 194) used to develop our algorithms and an independent validation set (The Cancer Genome Atlas [TCGA], n = 328). WSIs were first divided into small squares ("tiles"), and features were extracted with a pretrained convolutional neural network (preprocessing step). The first deep-learning-based algorithm ("SCHMOWDER") uses an attention mechanism on tumoral areas annotated by a pathologist whereas the second ("CHOWDER") does not require human expertise. In the discovery set, c-indices for survival prediction of SCHMOWDER and CHOWDER reached 0.78 and 0.75, respectively. Both models outperformed a composite score incorporating all baseline variables associated with survival. Prognostic value of the models was further validated in the TCGA data set, and, as observed in the discovery series, both models had a higher discriminatory power than a score combining all baseline variables associated with survival. Pathological review showed that the tumoral areas most predictive of poor survival were characterized by vascular spaces, the macrotrabecular architectural pattern, and a lack of immune infiltration. CONCLUSIONS: This study shows that artificial intelligence can help refine the prediction of HCC prognosis. It highlights the importance of pathologist/machine interactions for the construction of deep-learning algorithms that benefit from expert knowledge and allow a biological understanding of their output.

SF3B1-initiating mutations in MDS-RSs target lymphomyeloid hematopoietic stem cells.

Mutations in the RNA splicing gene SF3B1 are found in >80% of patients with myelodysplastic syndrome with ring sideroblasts (MDS-RS). We investigated the origin of SF3B1 mutations within the bone marrow hematopoietic stem and progenitor cell compartments in patients with MDS-RS. Screening for recurrently mutated genes in the mononuclear cell fraction revealed mutations in SF3B1 in 39 of 40 cases (97.5%), combined with TET2 and DNMT3A in 11 (28%) and 6 (15%) patients, respectively. All recurrent mutations identified in mononuclear cells could be tracked back to the phenotypically defined hematopoietic stem cell (HSC) compartment in all investigated patients and were also present in downstream myeloid and erythroid progenitor cells. While in agreement with previous studies, little or no evidence for clonal (SF3B1 mutation) involvement could be found in mature B cells, consistent involvement at the pro-B-cell progenitor stage was established, providing definitive evidence for SF3B1 mutations targeting lymphomyeloid HSCs and compatible with mutated SF3B1 negatively affecting lymphoid development. Assessment of stem cell function in vitro as well as in vivo established that only HSCs and not investigated progenitor populations could propagate the SF3B1 mutated clone. Upon transplantation into immune-deficient mice, SF3B1 mutated MDS-RS HSCs differentiated into characteristic ring sideroblasts, the hallmark of MDS-RS. Our findings provide evidence of a multipotent lymphomyeloid HSC origin of SF3B1 mutations in MDS-RS patients and provide a novel in vivo platform for mechanistically and therapeutically exploring SF3B1 mutated MDS-RS.

Integrative DNA methylation and gene expression analysis to assess the universality of the CpG island methylator phenotype.

BACKGROUND: The CpG island methylator phenotype (CIMP) was first characterized in colorectal cancer but since has been extensively studied in several other tumor types such as breast, bladder, lung, and gastric. CIMP is of clinical importance as it has been reported to be associated with prognosis or response to treatment. However, the identification of a universal molecular basis to define CIMP across tumors has remained elusive. RESULTS: We perform a genome-wide methylation analysis of over 2000 tumor samples from 5 cancer sites to assess the existence of a CIMP with common molecular basis across cancers. We then show that the CIMP phenotype is associated with specific gene expression variations. However, we do not find a common genetic signature in all tissues associated with CIMP. CONCLUSION: Our results suggest the existence of a universal epigenetic and transcriptomic signature that defines the CIMP across several tumor types but does not indicate the existence of a common genetic signature of CIMP.

Changes in correlation between promoter methylation and gene expression in cancer.

BACKGROUND: Methylation of high-density CpG regions known as CpG Islands (CGIs) has been widely described as a mechanism associated with gene expression regulation. Aberrant promoter methylation is considered a hallmark of cancer involved in silencing of tumor suppressor genes and activation of oncogenes. However, recent studies have also challenged the simple model of gene expression control by promoter methylation in cancer, and the precise mechanism of and role played by changes in DNA methylation in carcinogenesis remains elusive. RESULTS: Using a large dataset of 672 matched cancerous and healthy methylomes, gene expression, and copy number profiles accross 3 types of tissues from The Cancer Genome Atlas (TCGA), we perform a detailed meta-analysis to clarify the interplay between promoter methylation and gene expression in normal and cancer samples. On the one hand, we recover the existence of a CpG island methylator phenotype (CIMP) with prognostic value in a subset of breast, colon and lung cancer samples, where a common subset of promoter CGIs hypomethylated in normal samples become hypermethylated. However, this hypermethylation is not accompanied by a decrease in expression of the corresponding genes, which are already lowly expressed in the normal genes. On the other hand, we identify tissue-specific sets of genes, different between normal and cancer samples, whose inter-individual variation in expression is significantly correlated with the variation in methylation of the 3' flanking regions of the promoter CGIs. These subsets of genes are not the same in the different tissues, nor between normal and cancerous samples, but transcription factors are over-represented in all subsets. CONCLUSION: Our results suggest that epigenetic reprogramming in cancer does not contribute to cancer development via direct inhibition of gene expression through promoter hypermethylation. It may instead modify how the expression of a few specific genes, particularly transcription factors, are associated with DNA methylation variations in a tissue-dependent manner.

A deep learning model to predict RNA-Seq expression of tumours from whole slide images.

Deep learning methods for digital pathology analysis are an effective way to address multiple clinical questions, from diagnosis to prediction of treatment outcomes. These methods have also been used to predict gene mutations from pathology images, but no comprehensive evaluation of their potential for extracting molecular features from histology slides has yet been performed. We show that HE2RNA, a model based on the integration of multiple data modes, can be trained to systematically predict RNA-Seq profiles from whole-slide images alone, without expert annotation. Through its interpretable design, HE2RNA provides virtual spatialization of gene expression, as validated by CD3- and CD20-staining on an independent dataset. The transcriptomic representation learned by HE2RNA can also be transferred on other datasets, even of small size, to increase prediction performance for specific molecular phenotypes. We illustrate the use of this approach in clinical diagnosis purposes such as the identification of tumors with microsatellite instability.

Comprehensive Molecular and Pathologic Evaluation of Transitional Mesothelioma Assisted by Deep Learning Approach: A Multi-Institutional Study of the International Mesothelioma Panel from the MESOPATH Reference Center.

INTRODUCTION: Histologic subtypes of malignant pleural mesothelioma are a major prognostic indicator and decision denominator for all therapeutic strategies. In an ambiguous case, a rare transitional mesothelioma (TM) pattern may be diagnosed by pathologists either as epithelioid mesothelioma (EM), biphasic mesothelioma (BM), or sarcomatoid mesothelioma (SM). This study aimed to better characterize the TM subtype from a histological, immunohistochemical, and molecular standpoint. Deep learning of pathologic slides was applied to this cohort. METHODS: A random selection of 49 representative digitalized sections from surgical biopsies of TM was reviewed by 16 panelists. We evaluated BAP1 expression and CDKN2A (p16) homozygous deletion. We conducted a comprehensive, integrated, transcriptomic analysis. An unsupervised deep learning algorithm was trained to classify tumors. RESULTS: The 16 panelists recorded 784 diagnoses on the 49 cases. Even though a Kappa value of 0.42 is moderate, the presence of a TM component was diagnosed in 51%. In 49% of the histological evaluation, the reviewers classified the lesion as EM in 53%, SM in 33%, or BM in 14%. Median survival was 6.7 months. Loss of BAP1 observed in 44% was less frequent in TM than in EM and BM. p16 homozygous deletion was higher in TM (73%), followed by BM (63%) and SM (46%). RNA sequencing unsupervised clustering analysis revealed that TM grouped together and were closer to SM than to EM. Deep learning analysis achieved 94% accuracy for TM identification. CONCLUSION: These results revealed that the TM pattern should be classified as non-EM or at minimum as a subgroup of the SM type.

Deep learning-based classification of mesothelioma improves prediction of patient outcome.

Malignant mesothelioma (MM) is an aggressive cancer primarily diagnosed on the basis of histological criteria1. The 2015 World Health Organization classification subdivides mesothelioma tumors into three histological types: epithelioid, biphasic and sarcomatoid MM. MM is a highly complex and heterogeneous disease, rendering its diagnosis and histological typing difficult and leading to suboptimal patient care and decisions regarding treatment modalities2. Here we have developed a new approach-based on deep convolutional neural networks-called MesoNet to accurately predict the overall survival of mesothelioma patients from whole-slide digitized images, without any pathologist-provided locally annotated regions. We validated MesoNet on both an internal validation cohort from the French MESOBANK and an independent cohort from The Cancer Genome Atlas (TCGA). We also demonstrated that the model was more accurate in predicting patient survival than using current pathology practices. Furthermore, unlike classical black-box deep learning methods, MesoNet identified regions contributing to patient outcome prediction. Strikingly, we found that these regions are mainly located in the stroma and are histological features associated with inflammation, cellular diversity and vacuolization. These findings suggest that deep learning models can identify new features predictive of patient survival and potentially lead to new biomarker discoveries.

Analysis of the genomic landscape of multiple myeloma highlights novel prognostic markers and disease subgroups.

In multiple myeloma, next-generation sequencing (NGS) has expanded our knowledge of genomic lesions, and highlighted a dynamic and heterogeneous composition of the tumor. Here we used NGS to characterize the genomic landscape of 418 multiple myeloma cases at diagnosis and correlate this with prognosis and classification. Translocations and copy number abnormalities (CNAs) had a preponderant contribution over gene mutations in defining the genotype and prognosis of each case. Known and novel independent prognostic markers were identified in our cohort of proteasome inhibitor and immunomodulatory drug-treated patients with long follow-up, including events with context-specific prognostic value, such as deletions of the PRDM1 gene. Taking advantage of the comprehensive genomic annotation of each case, we used innovative statistical approaches to identify potential novel myeloma subgroups. We observed clusters of patients stratified based on the overall number of mutations and number/type of CNAs, with distinct effects on survival, suggesting that extended genotype of multiple myeloma at diagnosis may lead to improved disease classification and prognostication.

Classification and risk assessment in AML: integrating cytogenetics and molecular profiling.

In recent years, the composite molecular architecture in acute myeloid leukemia (AML) has been mapped out. We now have a clearer understanding of the key genetic determinants, the major genetic interactions, and the broad order in which these mutations occur. The next impending challenge is to discern how these recent genomic discoveries define disease biology as well as how to use molecular markers to deliver patient-tailored clinical decision support.

A Stromal Immune Module Correlated with the Response to Neoadjuvant Chemotherapy, Prognosis and Lymphocyte Infiltration in HER2-Positive Breast Carcinoma Is Inversely Correlated with Hormonal Pathways.

INTRODUCTION: HER2-positive breast cancer (BC) is a heterogeneous group of aggressive breast cancers, the prognosis of which has greatly improved since the introduction of treatments targeting HER2. However, these tumors may display intrinsic or acquired resistance to treatment, and classifiers of HER2-positive tumors are required to improve the prediction of prognosis and to develop novel therapeutic interventions. METHODS: We analyzed 2893 primary human breast cancer samples from 21 publicly available datasets and developed a six-metagene signature on a training set of 448 HER2-positive BC. We then used external public datasets to assess the ability of these metagenes to predict the response to chemotherapy (Ignatiadis dataset), and prognosis (METABRIC dataset). RESULTS: We identified a six-metagene signature (138 genes) containing metagenes enriched in different gene ontologies. The gene clusters were named as follows: Immunity, Tumor suppressors/proliferation, Interferon, Signal transduction, Hormone/survival and Matrix clusters. In all datasets, the Immunity metagene was less strongly expressed in ER-positive than in ER-negative tumors, and was inversely correlated with the Hormonal/survival metagene. Within the signature, multivariate analyses showed that strong expression of the "Immunity" metagene was associated with higher pCR rates after NAC (OR = 3.71[1.28-11.91], p = 0.019) than weak expression, and with a better prognosis in HER2-positive/ER-negative breast cancers (HR = 0.58 [0.36-0.94], p = 0.026). Immunity metagene expression was associated with the presence of tumor-infiltrating lymphocytes (TILs). CONCLUSION: The identification of a predictive and prognostic immune module in HER2-positive BC confirms the need for clinical testing for immune checkpoint modulators and vaccines for this specific subtype. The inverse correlation between Immunity and hormone pathways opens research perspectives and deserves further investigation.

Epigenomic alterations in breast carcinoma from primary tumor to locoregional recurrences.

INTRODUCTION: Epigenetic modifications such as aberrant DNA methylation has long been associated with tumorogenesis. Little is known, however, about how these modifications appear in cancer progression. Comparing the methylome of breast carcinomas and locoregional evolutions could shed light on this process. METHODS: The methylome profiles of 48 primary breast carcinomas (PT) and their matched axillary metastases (PT/AM pairs, 20 cases), local recurrences (PT/LR pairs, 17 cases) or contralateral breast carcinomas (PT/CL pairs, 11 cases) were analyzed. Univariate and multivariate analyzes were performed to determine differentially methylated probes (DMPs), and a similarity score was defined to compare methylation profiles. Correlation with copy-number based score was calculated and metastatic-free survival was compared between methods. RESULTS: 49 DMPs were found for the PT/AM set, but none for the others (FDR < 5%). Hierarchical clustering clustered 75% of the PT/AM, 47% of the PT/LR, and none of the PT/CL pairs together. A methylation-based score (MS) was defined as a clonality measure. The PT/AM set contained a high proportion of clonal pairs while PT/LR pairs were evenly split between high and low MS score, suggesting two groups: true recurrences (TR) and new primary tumors (NP). CL were classified as new tumors. MS score was significantly correlated with copy-number based scores. There was no significant difference between the metastatic-free survival of groups of patients based on different classifications. CONCLUSION: Epigenomic alterations are well suited to study clonality and track cancer progression. Methylation-based classification of TR and NP performed as well as clinical and copy-number based methods suggesting that these phenomenons are tightly linked.