Showing papers by "Frederick Klauschen published in 2020"

Resolving challenges in deep learning-based analyses of histopathological images using explanation methods.

Abstract: Deep learning has recently gained popularity in digital pathology due to its high prediction quality. However, the medical domain requires explanation and insight for a better understanding beyond standard quantitative performance evaluation. Recently, many explanation methods have emerged. This work shows how heatmaps generated by these explanation methods allow to resolve common challenges encountered in deep learning-based digital histopathology analyses. We elaborate on biases which are typically inherent in histopathological image data. In the binary classification task of tumour tissue discrimination in publicly available haematoxylin-eosin-stained images of various tumour entities, we investigate three types of biases: (1) biases which affect the entire dataset, (2) biases which are by chance correlated with class labels and (3) sampling biases. While standard analyses focus on patch-level evaluation, we advocate pixel-wise heatmaps, which offer a more precise and versatile diagnostic instrument. This insight is shown to not only be helpful to detect but also to remove the effects of common hidden biases, which improves generalisation within and across datasets. For example, we could see a trend of improved area under the receiver operating characteristic (ROC) curve by 5% when reducing a labelling bias. Explanation techniques are thus demonstrated to be a helpful and highly relevant tool for the development and the deployment phases within the life cycle of real-world applications in digital pathology.

Deep Learning for the Classification of Small-Cell and Non-Small-Cell Lung Cancer

Abstract: Reliable entity subtyping is paramount for therapy stratification in lung cancer Morphological evaluation remains the basis for entity subtyping and directs the application of additional methods such as immunohistochemistry (IHC) The decision of whether to perform IHC for subtyping is subjective, and access to IHC is not available worldwide Thus, the application of additional methods to support morphological entity subtyping is desirable Therefore, the ability of convolutional neuronal networks (CNNs) to classify the most common lung cancer subtypes, pulmonary adenocarcinoma (ADC), pulmonary squamous cell carcinoma (SqCC), and small-cell lung cancer (SCLC), was evaluated A cohort of 80 ADC, 80 SqCC, 80 SCLC, and 30 skeletal muscle specimens was assembled; slides were scanned; tumor areas were annotated; image patches were extracted; and cases were randomly assigned to a training, validation or test set Multiple CNN architectures (VGG16, InceptionV3, and InceptionResNetV2) were trained and optimized to classify the four entities A quality control (QC) metric was established An optimized InceptionV3 CNN architecture yielded the highest classification accuracy and was used for the classification of the test set Image patch and patient-based CNN classification results were 95% and 100% in the test set after the application of strict QC Misclassified cases mainly included ADC and SqCC The QC metric identified cases that needed further IHC for definite entity subtyping The study highlights the potential and limitations of CNN image classification models for tumor differentiation

High-protein diet more effectively reduces hepatic fat than low-protein diet despite lower autophagy and FGF21 levels

Abstract: Background and aims Non-alcoholic fatty liver disease (NAFLD) is becoming increasingly prevalent and nutrition intervention remains the most important therapeutic approach for NAFLD. Our aim was to investigate whether low- (LP) or high-protein (HP) diets are more effective in reducing liver fat and reversing NAFLD and which mechanisms are involved. Methods 19 participants with morbid obesity undergoing bariatric surgery were randomized into two hypocaloric (1500-1600 kcal/day) diet groups, a low protein (10E% protein) and a high protein (30E% protein), for three weeks prior to surgery. Intrahepatic lipid levels (IHL) and serum fibroblast growth factor 21 (FGF21) were measured before and after the dietary intervention. Autophagy flux, histology, mitochondrial activity, and gene expression analyses were performed in liver samples collected during surgery. Results IHL levels decreased by 42.6% in the HP group, but were not significantly changed in the LP group despite similar weight loss. Hepatic autophagy flux and serum FGF21 increased by 66.7% and 42.2%, respectively, after 3 weeks in the LP group only. Expression levels of fat uptake and lipid biosynthesis genes were lower in the HP group compared with those in the LP group. RNA-seq analysis revealed lower activity of inflammatory pathways upon HP diet. Hepatic mitochondrial activity and expression of β-oxidation genes did not increase in the HP group. Conclusions HP diet more effectively reduces hepatic fat than LP diet despite of lower autophagy and FGF21. Our data suggest that liver fat reduction upon HP diets result primarily from suppression of fat uptake and lipid biosynthesis.

Single-cell RNA sequencing reveals distinct tumor microenvironmental patterns in lung adenocarcinoma

Abstract: Recent developments in immuno-oncology demonstrate that not only cancer cells, but also features of the tumor microenvironment guide precision medicine. Still, the relationship between tumor and microenvironment remains poorly understood. To overcome this limitation and identify clinically relevant microenvironmental and cancer features, we applied single-cell RNA sequencing to lung adenocarcinomas. While the highly heterogeneous carcinoma cell transcriptomes reflected histological grade and activity of relevant oncogenic pathways, our analysis revealed two distinct microenvironmental patterns. We identified a prognostically unfavorable group of tumors with a microenvironment composed of cancer-associated myofibroblasts, exhausted CD8+ T cells, proinflammatory monocyte-derived macrophages and plasmacytoid dendritic cells (CEP2 pattern) and a prognostically favorable group characterized by myeloid dendritic cells, anti-inflammatory monocyte-derived macrophages, normal-like myofibroblasts, NK cells and conventional T cells (MAN2C pattern). Our results show that single-cell gene expression profiling allows to identify patient subgroups based on the tumor microenvironment beyond cancer cell-centric profiling.

TP53 Mutations Predict Sensitivity to Adjuvant Gemcitabine in Patients with Pancreatic Ductal Adenocarcinoma: Next-Generation Sequencing Results from the CONKO-001 Trial

Abstract: PURPOSE: We performed next-generation sequencing (NGS) in the CONKO-001 phase-3 trial to identify clinically relevant prognostic and predictive mutations and conducted a functional validation in TCGA sequencing data. PATIENTS AND METHODS: Patients of the CONKO-001 trial received curatively intended surgery for pancreatic adenocarcinoma (PDAC) followed by adjuvant chemotherapy with gemcitabine (Gem) or observation only (Obs). Tissue samples of 101 patients were evaluated by NGS of 37 genes. Cox proportional hazard models were applied for survival analysis. Additionally, functional genomic analyses were performed in an NGS and RNASeq dataset of 146 pancreatic tumors from The Cancer Genome Atlas (TCGA). RESULTS: The most common mutations in the CONKO-cohort were KRAS (75%), TP53 (60%), SMAD4 (10%), CDKNA2 (9%), as well as SWI/SNF (12%) complex alterations. In untreated patients, TP53 mutations were a negative prognostic factor for DFS (HR mut vs. WT 2.434, p=0.005). In Gem treated patients, TP53 mutations were a positive predictive factor for gemcitabine efficacy (TP53mut: HR for DFS Gem vs Obs: 0.235 (0.130 - 0.423; p

Novel methods for in vitro modeling of pancreatic cancer reveal important aspects for successful primary cell culture

Abstract: Pancreatic cancer remains a fatal disease. Experimental systems are needed for personalized treatment strategies, drug testing and to further understand tumor biology. Cell cultures can serve as an excellent preclinical platform, but their generation remains challenging. Tumor cells from surgically removed pancreatic ductal adenocarcinoma (PDAC) specimens were cultured under novel protocols. Cellular growth and composition were analyzed and culture conditions were continuously optimized. Characterization of cell cultures and primary tumors was performed via hematoxylin and eosin (HE) and immunofluorescence (IF) staining. Protocols for two- and three-dimensional PDAC primary cell cultures could successfully be established. Primary cell culture depended on dissociation techniques, growth factor supplementation and extracellular matrix components containing Matrigel being crucial for the transformation to three-dimensional PDAC organoids. The generated cultures showed to be highly resemblant to established PDAC primary cell cultures. HE and IF staining for cell culture and corresponding primary tumor characterization could successfully be performed. The work presented herein shows novel and effective methods to successfully establish primary PDAC cell cultures in a distinct time frame. Factors contributing to cell growth and differentiation could be identified with important implications for further primary cell culture protocols. The established protocols might serve as novel tools in personalized tumor therapy.

Interpretable Deep Neural Network to Predict Estrogen Receptor Status from Haematoxylin-Eosin Images

Abstract: The eligibility for hormone therapy to treat breast cancer largely depends on the tumor’s estrogen receptor (ER) status Recent studies show that the ER status correlates with morphological features found in Haematoxylin-Eosin (HE) slides Thus, HE analysis might be sufficient for patients for whom the classifier confidently predicts the ER status and thereby obviate the need for additional examination, such as immunohistochemical (IHC) staining Several prior works are limited by either the use of engineered features, multi-stage models that use features unspecific to HE images or a lack of explainability To address these limitations, this work proposes an end-to-end neural network ensemble that shows state-of-the-art performance We demonstrate that the approach also translates to the prediction of the cancer grade Moreover, subsets can be selected from the test data for which the model can detect a positive ER status with a precision of 94% while classifying 13% of the patients To compensate for the reduced interpretability of the model that comes along with end-to-end training, this work applies Layer-wise Relevance Propagation (LRP) to determine the relevant parts of the images a posteriori, commonly visualized as a heatmap overlayed with the input image We found that nuclear and stromal morphology and lymphocyte infiltration play an important role in the classification of the ER status This demonstrates that interpretable machine learning can be a vital tool for validating and generating hypotheses about morphological biomarkers

Distinct immune evasion in APOBEC-enriched, HPV-negative HNSCC.

Abstract: Immune checkpoint inhibition leads to response in some patients with head and neck squamous cell carcinoma (HNSCC). Robust biomarkers are lacking to date. We analyzed viral status, gene expression signatures, mutational load and mutational signatures in whole exome and RNA-sequencing data of the HNSCC TCGA dataset (n = 496) and a validation set (DKTK MASTER cohort, n = 10). Public single-cell gene expression data from 17 HPV-negative HNSCC were separately reanalyzed. APOBEC3-associated TCW motif mutations but not total single nucleotide variant burden were significantly associated with inflammation. This association was restricted to HPV-negative HNSCC samples. An APOBEC-enriched, HPV-negative subgroup was identified, that showed higher T-cell inflammation and immune checkpoint expression, as well as expression of APOBEC3 genes. Mutations in immune-evasion pathways were also enriched in these tumors. Analysis of single-cell sequencing data identified expression of APOBEC3B and 3C genes in malignant cells. We identified an APOBEC-enriched subgroup of HPV-negative HNSCC with a distinct immunogenic phenotype, potentially mediating response to immunotherapy.

Human Lungs Show Limited Permissiveness for SARS-CoV-2 Due to Scarce ACE2 Levels But Strong Virus-Induced Immune Activation in Alveolar Macrophages

Abstract: SARS-CoV-2 utilizes the ACE2 transmembrane peptidase as essential cellular entry receptor. Several studies have suggested abundant ACE2 expression in the human lung, inferring strong permissiveness to SARS-CoV-2 infection with resultant alveolar damage and lung injury. Against this expectation, we provide evidence that ACE2 expression must be considered scarce, thereby limiting SARS-CoV-2 propagation in the human alveolus. Instead, spectral imaging of ex vivo infected human lungs and COVID-19 autopsy samples depicted that alveolar macrophages were frequently positive for SARS-CoV-2, indicating viral phagocytosis. Single-cell transcriptomics of SARS-CoV-2 infected human lung tissue further revealed strong inflammatory and anti-viral activation responses in macrophages and monocytes, comparable to those induced by MERS-CoV, but with virus-specific gene expression profiles. Collectively, our findings indicate that severe lung injury in COVID-19 likely results from an overwhelming immune activation rather than direct viral damage of the alveolar compartment. Funding: ACH, LES, SH were supported by Berlin University Alliance GC2 Global Health (Corona Virus Pre-Exploration Project). ACH, SH, TW and CD were supported by BMBF (RAPID) and ACH, SH by BMBF (alvBarriereCOVID-19). KH, LB, SL, SH, CD, TW, ACH were funded by BMBF (NFN-COVID 19, Organo-Strat). KH, NS, LES, MW, SH, ADG, CD, TW and ACH were supported by DFG (SFB-TR 84). ACH was supported by BIH, Charite 3R, and Charite-Zeiss MultiDim. KH was supported by BMBF (Camo-COVID-19). MW, NS and SH was supported by BMBF (PROVID). MW and NS was supported by BIH and BMBF (SYMPATH, CAPSyS, NAPKON). BO and DB were funded through the BIH Clinical Single Cell Bioinformatics Pipeline. LB was supported by the BMBF (CoIMMUNE), the DFG (KFO 342) and the IZKF of the Medical Faculty of the WWU. Conflict of Interest: The authors declare no competing interests. Ethical Approval: The study was approved by the ethics committee at the Charite clinic (projects EA2/079/13) and Arztekammer Westfalen-Lippe and of the Westfalischen Wilhelms-Universitat (AZ: 2016-265-f-S). Written informed consent was obtained from all patients.