Application of deep learning models on whole slide images uncover new histological markers related to high-risk malignant pleural mesothelioma

e13580

Background: Malignant Pleural mesothelioma (MPM) is a highly aggressive cancer of the pleural surface and represents 80-90% of mesothelioma diagnosis. MPM is broadly subclassified into three histological subgroups (epithelioid, sarcomatoid, biphasic), however tissue heterogeneity has resulted in diagnostic challenges and suboptimal patient care. There are currently no specific histological markers of high/low-risk MPM patients, which is critical in predicting patient prognosis. Methods: Owkin developed MesoNet, a deep learning model that predicts overall survival (OS) of MPM patients from whole slide images (WSI) and trained on the French MESOBANK and TCGA cohorts (Courtiol et al, 2019). In this study, we sought to validate MesoNet’s performance on an independent cohort of 127 WSI stained with haematoxylin/eosin from MPM patients collected at the University of Pittsburgh as part of the National Mesothelioma Virtual Bank (funding by U24OH009077). Patient demographics, survival data, expertly curated pathology annotations were also collected. Results: Our analyses showed that MesoNet predicted OS as risk score based on WSI, which validated high-risk MPM patients exhibited poorer OS, as compared to low-risk patients. Analyses on histological subtypes revealed sarcomatoid and biphasic patients were overrepresented in high-risk groups, as compared to epithelioid patients, which correlates with observed OS data. Notably, histological features associated with high-risk patients revealed tumor pleomorphism and anaplastic nuclear features, whereas low-risk tiles appear to be enriched in tumor infiltrating lymphocytes (TILs) with accompanying stromal proliferation and dense fibrosis. Conclusions: Collectively, our studies validate MesoNet performance on an independent cohort and identify new features related to MPM risk groups, which may inform future treatment stratification and personalization of immunotherapies.

Macrophage migration inhibitory factor is overproduced through EGR1 in TET2low resting monocytes

Somatic mutation in TET2 gene is one of the most common clonal genetic events detected in age-related clonal hematopoiesis as well as in chronic myelomonocytic leukemia (CMML). In addition to being a pre-malignant state, TET2 mutated clones are associated with an increased risk of death from cardiovascular disease, which could involve cytokine/chemokine overproduction by monocytic cells. Here, we show in mice and in human cells that, in the absence of any inflammatory challenge, TET2 downregulation promotes the production of MIF (macrophage migration inhibitory factor), a pivotal mediator of atherosclerotic lesion formation. In healthy monocytes, TET2 is recruited to MIF promoter and interacts with the transcription factor EGR1 and histone deacetylases. Disruption of these interactions as a consequence of TET2-decreased expression favors EGR1-driven transcription of MIF gene and its secretion. MIF favors monocytic differentiation of myeloid progenitors. These results designate MIF as a chronically overproduced chemokine and a potential therapeutic target in patients with clonal TET2 downregulation in myeloid cells.

To improve our understanding of the pathological role of TET2 mutations, Pronier, Imanci et al. use mice and human cells to show that TET2 downregulation promotes the production of macrophage migration inhibitory factor (MIF). In addition they show that whilst TET2 is recruited to the MIF promoter in healthy monocytes, decreased TET2 expression results in chronic overproduction of MIF - suggesting that MIF signaling could therefore constitute a potential therapeutic target for conditions associated with TET2 mutations.

Integrating deep learning CT-scan model, biological and clinical variables to predict severity of COVID-19 patients

The SARS-COV-2 pandemic has put pressure on intensive care units, so that identifying predictors of disease severity is a priority. We collect 58 clinical and biological variables, and chest CT scan data, from 1003 coronavirus-infected patients from two French hospitals. We train a deep learning model based on CT scans to predict severity. We then construct the multimodal AI-severity score that includes 5 clinical and biological variables (age, sex, oxygenation, urea, platelet) in addition to the deep learning model. We show that neural network analysis of CT-scans brings unique prognosis information, although it is correlated with other markers of severity (oxygenation, LDH, and CRP) explaining the measurable but limited 0.03 increase of AUC obtained when adding CT-scan information to clinical variables. Here, we show that when comparing AI-severity with 11 existing severity scores, we find significantly improved prognosis performance; AI-severity can therefore rapidly become a reference scoring approach.

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.

Abstract 2105: HE2RNA: A deep learning model for transcriptomic learning from digital pathology

Today, pathology imaging is one of the most common and inexpensive diagnostic/prognostic tools used in oncology, while more sophisticated methods such as next generation sequencing (NGS) remain relatively expensive and not routinely used in a clinical setting. Deep convolutional neural networks (CNNs) have emerged as an important image analysis technology enhancing the workflow of pathologists and improving the prediction of patient prognosis and response to treatment​​. Recently, a few attempts have been made to predict molecular features from tissue imaging using CNNs. While these preliminary results are encouraging, there have been no systematic attempts to link Whole Slide Images (WSIs) to transcriptomic profiles. In this study, we developed a cutting-edge deep learning model named HE2RNA, specifically customized for the direct prediction of gene expression from H&E-stained WSIs without need for annotation from pathologists. Our model was trained and tested on 8,725 patients from 28 different cancer types available at The Cancer Genome Atlas (TCGA).

HE2RNA accurately predicted the expression of six gene signatures related to well known cancer hallmarks (angiogenesis, hypoxia, DNA repair, cell cycle and immunity) and performed particularly well for signalling pathways involved in immune cell activation. This indicates that suitably designed deep learning models can recognize subtle structures in tissue imaging and relate them to molecular portraits.

Moreover, HE2RNA is designed to generate a spatial representation (virtual map) of any well-predicted gene expression overlaying the H&E slide. Such a virtual map was validated on a double-stained H&E/CD3 slide obtained from an independent hepatocellular carcinoma sample. This spatialization could be useful in augmenting the pathologists' workflow by providing a virtual multiplexed staining for each H&E slide while overcoming the technical issues associated with immunohistochemistry.

Various important prognostic factors, such as microsatellite instability (MSI), are derived from molecular features. Microsatellite instability refers to the hypermutability of short repetitive genomic sequences caused by impaired DNA mismatch repair. These mutations frequently observed in gastric and colorectal cancer are associated with better response to immunotherapy. We show that the transcriptomic representation learned by our model can be used to improve the performance of MSI status prediction for small datasets of WSI. This type of setting is common since large databases of matched RNA-Seq profiles and WSI are widely available, while databases of matched MSI status and WSI are more scarce. In the future, such technologies could therefore facilitate universal screening of molecular biomarkers and improved identification of patients that could benefit from new therapeutic strategies.

Citation Format: Elodie Pronier, Benoît Schmauch, Alberto Romagnoni, Charlie Saillard, Pascale Maillé, Julien Calderaro, Meriem Sefta, Sylvain Toldo, Mikhail Zaslavskiy, Thomas Clozel, Matahi Moarii, Pierre Courtiol, Gilles Wainrib. HE2RNA: A deep learning model for transcriptomic learning from digital pathology [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 2105.

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.

RNA-sequencing of tumour tissue can provide important diagnostic and prognostic information but this is costly and not routinely performed in all clinical settings. Here, the authors show that whole slide histology slides—part of routine care—can be used to predict RNA-sequencing data and thus reduce the need for additional analyses.

Unfolding the Role of Calreticulin in Myeloproliferative Neoplasm Pathogenesis

In 2013, two seminal studies identified gain-of-function mutations in the Calreticulin (CALR) gene in a subset of JAK2/MPL-negative myeloproliferative neoplasm (MPN) patients. CALR is an endoplasmic reticulum (ER) chaperone protein that normally binds misfolded proteins in the ER and prevents their export to the Golgi and had never previously been reported mutated in cancer or to be associated with hematologic disorders. Further investigation determined that mutated CALR is able to achieve oncogenic transformation primarily through constitutive activation of the MPL-JAK-STAT signaling axis. Here we review our current understanding of the role of CALR mutations in MPN pathogenesis and how these insights can lead to innovative therapeutics approaches.

Targeting the CALR interactome in myeloproliferative neoplasms

Mutations in the ER chaperone calreticulin (CALR) are common in myeloproliferative neoplasm (MPN) patients, activate the thrombopoietin receptor (MPL), and mediate constitutive JAK/STAT signaling. The mechanisms by which CALR mutations cause myeloid transformation are incompletely defined. We used mass spectrometry proteomics to identify CALR-mutant interacting proteins. Mutant CALR caused mislocalization of binding partners and increased recruitment of FLI1, ERP57, and CALR to the MPL promoter to enhance transcription. Consistent with a critical role for CALR-mediated JAK/STAT activation, we confirmed the efficacy of JAK2 inhibition on CALR-mutant cells in vitro and in vivo. Due to the altered interactome induced by CALR mutations, we hypothesized that CALR-mutant MPNs may be vulnerable to disruption of aberrant CALR protein complexes. A synthetic peptide designed to competitively inhibit the carboxy terminal of CALR specifically abrogated MPL/JAK/STAT signaling in cell lines and primary samples and improved the efficacy of JAK kinase inhibitors. These findings reveal what to our knowledge is a novel potential therapeutic approach for patients with CALR-mutant MPN.

Cooperative Epigenetic Remodeling by TET2 Loss and NRAS Mutation Drives Myeloid Transformation and MEK Inhibitor Sensitivity

Mutations in epigenetic modifiers and signaling factors often co-occur in myeloid malignancies, including TET2 and NRAS mutations. Concurrent Tet2 loss and Nras^G12D expression in hematopoietic cells induced myeloid transformation, with a fully penetrant, lethal chronic myelomonocytic leukemia (CMML), which was serially transplantable. Tet2 loss and Nras mutation cooperatively led to decrease in negative regulators of mitogen-activated protein kinase (MAPK) activation, including Spry2, thereby causing synergistic activation of MAPK signaling by epigenetic silencing. Tet2/Nras double-mutant leukemia showed preferential sensitivity to MAPK kinase (MEK) inhibition in both mouse model and patient samples. These data provide insights into how epigenetic and signaling mutations cooperate in myeloid transformation and provide a rationale for mechanism-based therapy in CMML patients with these high-risk genetic lesions.

Genetic hierarchy and temporal variegation in the clonal history of acute myeloid leukaemia

In acute myeloid leukaemia (AML) initiating pre-leukaemic lesions can be identified through three major hallmarks: their early occurrence in the clone, their persistence at relapse and their ability to initiate multilineage haematopoietic repopulation and leukaemia in vivo. Here we analyse the clonal composition of a series of AML through these characteristics. We find that not only DNMT3A mutations, but also TET2, ASXL1 mutations, core-binding factor and MLL translocations, as well as del(20q) mostly fulfil these criteria. When not eradicated by AML treatments, pre-leukaemic cells with these lesions can re-initiate the leukaemic process at various stages until relapse, with a time-dependent increase in clonal variegation. Based on the nature, order and association of lesions, we delineate recurrent genetic hierarchies of AML. Our data indicate that first lesions, variegation and treatment selection pressure govern the expansion and adaptive behaviour of the malignant clone, shaping AML in a time-dependent manner.

Pre-leukaemic clones, together with the propensity to cause disease in mice, are characterized by appearing early in myeloid leukaemia and being found at relapse. Here, the authors identify clones in human samples and find that they are characterized by hierarchically organized genetic lesions, which can be used to track evolution of the disease.

IDH1/2 mutations and BCL-2 dependence: an unexpected Chink in AML's armour

There is a pressing need to develop novel, mechanism-based therapeutic approaches that can be used to improve therapies for genetically defined tumor subtypes. Chan and colleagues have demonstrated recently that BCL-2 inhibitors can target IDH1/2 mutant cancers through a mutant-specific dependency in metabolic regulation.

Identification of small molecule activators of BMP signaling

Bone Morphogenetic Proteins (BMPs) are morphogens that play a major role in regulating development and homeostasis. Although BMPs are used for the treatment of bone and kidney disorders, their clinical use is limited due to the supra-physiological doses required for therapeutic efficacy causing severe side effects. Because recombinant BMPs are expensive to produce, small molecule activators of BMP signaling would be a cost-effective alternative with the added benefit of being potentially more easily deliverable. Here, we report our efforts to identify small molecule activators of BMP signaling. We have developed a cell-based assay to monitor BMP signaling by stably transfecting a BMP-responsive human cervical carcinoma cell line (C33A) with a reporter construct in which the expression of luciferase is driven by a multimerized BMP-responsive element from the Id1 promoter. A BMP-responsive clone C33A-2D2 was used to screen a bioactive library containing ∼5,600 small molecules. We identified four small molecules of the family of flavonoids all of which induced luciferase activity in a dose-dependent manner and ventralized zebrafish embryos. Two of the identified compounds induced Smad1, 5 phosphorylation (P-Smad), Id1 and Id2 expression in a dose-dependent manner demonstrating that our assays identified small molecule activators of BMP signaling.

Role of TET2 mutations in myeloproliferative neoplasms

Recently, 5-hydroxymethylcytosine (5-hmC), the 6th base of DNA, was discovered as the product of the hydroxylation of 5-methylcytosine (5-mC) by the ten-eleven translocation (TET) oncogene family members. One of them, TET oncogene family member 2 (TET2), is mutated in a variety of myeloid malignancies, including in 15% of myeloproliferative neoplasms (MPNs). Recent studies tried to go further into the biological and epigenetic function of TET2 protein and 5-hmC marks in the pathogenesis of myeloid malignancies. Although its precise function remains partially unknown, TET2 appears to be an important regulator of hematopoietic stem cell biology. In both mouse and human cells, its inactivation leads to a dramatic deregulation of hematopoiesis that ultimately triggers blood malignancies. Understanding this leukemogenic process will provide tools to develop new epigenetic therapies against blood cancers.