Quantification of Hepatocellular Mitoses in a Toxicological Study in Rats Using a Convolutional Neural Network.

Convolutional neural networks (CNNs) have been recognized as valuable tools for rapid quantitative analysis of morphological changes in toxicologic histopathology. We have assessed the performance of CNN-based (Halo-AI) mitotic figure detection in hepatocytes in comparison with detection by pathologists. In addition, we compared with Ki-67 and 5-bromodesoxyuridin (BrdU) immunohistochemistry labeling indices (LIs) obtained by image analysis. Tissues were from an exploratory toxicity study with a glycogen synthase kinase-3 (GSK-3) inhibitor. Our investigations revealed that (1) the CNN achieved similarly accurate but faster results than pathologists, (2) results of mitotic figure detection were comparable to Ki-67 and BrdU LIs, and (3) data from different methods were only moderately correlated. The latter is likely related to differences in the cell cycle component captured by each method. This highlights the importance of considering the differences of the available methods upon selection. Also, the pharmacology of our test item acting as a GSK-3 inhibitor potentially reduced the correlation. We conclude that hepatocyte cell proliferation assessment by CNNs can have several advantages when compared with the current gold standard: it relieves the pathologist of tedious routine tasks and contributes to standardization of results; the CNN algorithm can be shared and iteratively improved; it can be performed on routine histological slides; it does not require an additional animal experiment and in this way can contribute to animal welfare according to the 3R principles.

Deep Learning-Based Detection of Endothelial Tip Cells in the Oxygen-Induced Retinopathy Model.

Proliferative retinopathies, such as diabetic retinopathy and retinopathy of prematurity, are leading causes of vision impairment. A common feature is a loss of retinal capillary vessels resulting in hypoxia and neuronal damage. The oxygen-induced retinopathy model is widely used to study revascularization of an ischemic area in the mouse retina. The presence of endothelial tip cells indicates vascular recovery; however, their quantification relies on manual counting in microscopy images of retinal flat mount preparations. Recent advances in deep neural networks (DNNs) allow the automation of such tasks. We demonstrate a workflow for detection of tip cells in retinal images using the DNN-based Single Shot Detector (SSD). The SSD was designed for detection of objects in natural images. We adapt the SSD architecture and training procedure to the tip cell detection task and retrain the DNN using labeled tip cells in images of fluorescently stained retina flat mounts. Transferring knowledge from the pretrained DNN and extensive data augmentation reduced the amount of required labeled data. Our system shows a performance comparable to the human level, while providing highly consistent results. Therefore, such a system can automate counting of tip cells, a readout frequently used in retinopathy research, thereby reducing routine work for biomedical experts.

The lytic activity of VSV-GP treatment dominates the therapeutic effects in a syngeneic model of lung cancer.

BACKGROUND: Oncolytic virotherapy is thought to result in direct virus-induced lytic tumour killing and simultaneous activation of innate and tumour-specific adaptive immune responses. Using a chimeric vesicular stomatitis virus variant VSV-GP, we addressed the direct oncolytic effects and the role of anti-tumour immune induction in the syngeneic mouse lung cancer model LLC1. METHODS: To study a tumour system with limited antiviral effects, we generated interferon receptor-deficient cells (LLC1-IFNAR1-/-). Therapeutic efficacy of VSV-GP was assessed in vivo in syngeneic C57BL/6 and athymic nude mice bearing subcutaneous tumours. VSV-GP treatment effects were analysed using bioluminescent imaging (BLI), immunohistochemistry, ELISpot, flow cytometry, multiplex ELISA and Nanostring® assays. RESULTS: Interferon insensitivity correlated with VSV-GP replication and therapeutic outcome. BLI revealed tumour-to-tumour spread of viral progeny in bilateral tumours. Histological and gene expression analysis confirmed widespread and rapid infection and cell killing within the tumour with activation of innate and adaptive immune-response markers. However, treatment outcome was increased in the absence of CD8+ T cells and surviving mice showed little protection from tumour re-challenge, indicating limited therapeutic contribution by the activated immune system. CONCLUSION: These studies present a case for a predominantly lytic treatment effect of VSV-GP in a syngeneic mouse lung cancer model.

Introducing a fluorescence-based standard to quantify protein partitioning into membranes.

The affinity of peripheral membrane proteins for a lipid bilayer can be described using the partition coefficient (KP). Although several methods to determine KP are known, all possess limitations. To address some of these issues, we developed both: a versatile method based on single molecule detection and fluorescence imaging for determining KP, and a simple measurement standard employing hexahistidine-tagged enhanced green fluorescent protein (eGFP-His6) and free standing membranes of giant unilamellar vesicles (GUVs) functionalized with NTA(Ni) lipids as binding sites. To ensure intrinsic control, our method features two measurement modes. In the single molecule mode, fluorescence correlation spectroscopy (FCS) is applied to quantify free and membrane associated protein concentrations at equilibrium and calculate KP. In the imaging mode, confocal fluorescence images of GUVs are recorded and analyzed with semi-automated software to extract protein mean concentrations used to derive KP. Both modes were compared by determining the affinity of our standard, resulting in equivalent KP values. As observed in other systems, eGFP-His6 affinity for membranes containing increasing amounts of NTA(Ni) lipids rises in a stronger-than-linear fashion. We compared our dual approach with a FCS-based assay that uses large unilamellar vesicles (LUVs), which however fails to capture the stronger-than-linear trend for our NTA(Ni)-His6 standard. Hence, we determined the KP of the MARCKS effector domain with our FCS approach on GUVs, whose results are consistent with previously published data using LUVs. We finally provide a practical manual on how to measure KP and understand it in terms of molecules per lipid surface.

Learning image representations for anomaly detection: Application to discovery of histological alterations in drug development.

We present a system for anomaly detection in histopathological images. In histology, normal samples are usually abundant, whereas anomalous (pathological) cases are scarce or not available. Under such settings, one-class classifiers trained on healthy data can detect out-of-distribution anomalous samples. Such approaches combined with pre-trained Convolutional Neural Network (CNN) representations of images were previously employed for anomaly detection (AD). However, pre-trained off-the-shelf CNN representations may not be sensitive to abnormal conditions in tissues, while natural variations of healthy tissue may result in distant representations. To adapt representations to relevant details in healthy tissue we propose training a CNN on an auxiliary task that discriminates healthy tissue of different species, organs, and staining reagents. Almost no additional labeling workload is required, since healthy samples come automatically with aforementioned labels. During training we enforce compact image representations with a center-loss term, which further improves representations for AD. The proposed system outperforms established AD methods on a published dataset of liver anomalies. Moreover, it provided comparable results to conventional methods specifically tailored for quantification of liver anomalies. We show that our approach can be used for toxicity assessment of candidate drugs at early development stages and thereby may reduce expensive late-stage drug attrition.

Monocyte-derived alveolar macrophages are key drivers of smoke-induced lung inflammation and tissue remodeling.

Smoking is a leading risk factor of chronic obstructive pulmonary disease (COPD), that is characterized by chronic lung inflammation, tissue remodeling and emphysema. Although inflammation is critical to COPD pathogenesis, the cellular and molecular basis underlying smoking-induced lung inflammation and pathology remains unclear. Using murine smoke models and single-cell RNA-sequencing, we show that smoking establishes a self-amplifying inflammatory loop characterized by an influx of molecularly heterogeneous neutrophil subsets and excessive recruitment of monocyte-derived alveolar macrophages (MoAM). In contrast to tissue-resident AM, MoAM are absent in homeostasis and characterized by a pro-inflammatory gene signature. Moreover, MoAM represent 46% of AM in emphysematous mice and express markers causally linked to emphysema. We also demonstrate the presence of pro-inflammatory and tissue remodeling associated MoAM orthologs in humans that are significantly increased in emphysematous COPD patients. Inhibition of the IRAK4 kinase depletes a rare inflammatory neutrophil subset, diminishes MoAM recruitment, and alleviates inflammation in the lung of cigarette smoke-exposed mice. This study extends our understanding of the molecular signaling circuits and cellular dynamics in smoking-induced lung inflammation and pathology, highlights the functional consequence of monocyte and neutrophil recruitment, identifies MoAM as key drivers of the inflammatory process, and supports their contribution to pathological tissue remodeling.

Lateral membrane diffusion modulated by a minimal actin cortex.

Diffusion of lipids and proteins within the cell membrane is essential for numerous membrane-dependent processes including signaling and molecular interactions. It is assumed that the membrane-associated cytoskeleton modulates lateral diffusion. Here, we use a minimal actin cortex to directly study proposed effects of an actin meshwork on the diffusion in a well-defined system. The lateral diffusion of a lipid and a protein probe at varying densities of membrane-bound actin was characterized by fluorescence correlation spectroscopy (FCS). A clear correlation of actin density and reduction in mobility was observed for both the lipid and the protein probe. At high actin densities, the effect on the protein probe was ∼3.5-fold stronger compared to the lipid. Moreover, addition of myosin filaments, which contract the actin mesh, allowed switching between fast and slow diffusion in the minimal system. Spot variation FCS was in accordance with a model of fast microscopic diffusion and slower macroscopic diffusion. Complementing Monte Carlo simulations support the analysis of the experimental FCS data. Our results suggest a stronger interaction of the actin mesh with the larger protein probe compared to the lipid. This might point toward a mechanism where cortical actin controls membrane diffusion in a strong size-dependent manner.

Quantifying lipid diffusion by fluorescence correlation spectroscopy: a critical treatise.

Fluorescence correlation spectroscopy (FCS) measurements are widely used for determination of diffusion coefficients of lipids and proteins in biological membranes. In recent years, several variants of FCS have been introduced. However, a comprehensive comparison of these methods on identical systems has so far been lacking. In addition, there exist no consistent values of already determined diffusion coefficients for well-known or widely used membrane systems. This study aims to contribute to a better comparability of FCS experiments on membranes by determining the absolute diffusion coefficient of the fluorescent lipid analog 1,1'-dioctadecyl-3,3,3',3'-tetramethylindodicarbocyanine (DiD) in giant unilamellar vesicles (GUVs) made of dioleoylphosphatidylcholine (DOPC), which can in future studies be used as a reference value. For this purpose, five FCS variants, employing different calibration methods, were compared. Potential error sources for each particular FCS method and strategies to avoid them are discussed. The obtained absolute diffusion coefficients for DiD in DOPC were in good agreement for all investigated FCS variants. An average diffusion coefficient of D = 10.0 ± 0.4 µm(2) s(-1) at 23.5 ± 1.5 °C was obtained. The independent confirmation with different methods indicates that this value can be safely used for calibration purposes. Moreover, the comparability of the methods also in the case of slow diffusion was verified by measuring diffusion coefficients of DiD in GUVs consisting of DOPC and cholesterol.

Deep learning-based quantification of NAFLD/NASH progression in human liver biopsies.

Non-alcoholic fatty liver disease (NAFLD) affects about 24% of the world's population. Progression of early stages of NAFLD can lead to the more advanced form non-alcoholic steatohepatitis (NASH), and ultimately to cirrhosis or liver cancer. The current gold standard for diagnosis and assessment of NAFLD/NASH is liver biopsy followed by microscopic analysis by a pathologist. The Kleiner score is frequently used for a semi-quantitative assessment of disease progression. In this scoring system the features of active injury (steatosis, inflammation, and ballooning) and a separated fibrosis score are quantified. The procedure is time consuming for pathologists, scores have limited resolution and are subject to variation. We developed an automated deep learning method that provides full reproducibility and higher resolution. The system was established with 296 human liver biopsies and tested on 171 human liver biopsies with pathologist ground truth scores. The method is inspired by the way pathologist's analyze liver biopsies. First, the biopsies are analyzed microscopically for the relevant histopathological features. Subsequently, histopathological features are aggregated to a per-biopsy score. Scores are in the identical numeric range as the pathologist's ballooning, inflammation, steatosis, and fibrosis scores, but on a continuous scale. Resulting scores followed a pathologist's ground truth (quadratic weighted Cohen's κ on the test set: for steatosis 0.66, for inflammation 0.24, for ballooning 0.43, for fibrosis 0.62, and for the NAFLD activity score (NAS) 0.52. Mean absolute errors on a test set: for steatosis 0.29, for inflammation 0.53, for ballooning 0.61, for fibrosis 0.78, and for the NAS 0.77).

Keratocyte lamellipodial protrusion is characterized by a concave force-velocity relation.

We report on the characterization of actin driven lamellipodial protrusion forces and velocities in keratocytes. A vertically mounted glass fiber acted as a flexible barrier positioned in front of migrating keratocytes with parallel phase contrast microscopy. A laser beam was coupled into the fiber and allowed detecting the position of the fiber by a segmented photodiode. Calibration of the fiber was carried out with the thermal oscillation method. Deflection and force signals were measured during lamellipodial protrusion. Velocity was constant during initial contact whereas loading force increased until finally the cell was stalled at higher forces. Stall forces were on the order of 2.9 ± 0.6 nN, which corresponds to a stall pressure of 2.7 ± 1.6 nN/µm(2). Assuming a density of actin filaments of 240 filaments per µm, we can estimate a stall force per actin filament of 1.7 ± 0.8 pN. To check for adaption of the cell against an external force, we let the cell push toward the glass fiber several times. On the timescale of the experiment (∼1 min), however, the cell did not adapt to previous loading events.

Intermittent exposure to whole cigarette smoke alters the differentiation of primary small airway epithelial cells in the air-liquid interface culture.

Cigarette smoke (CS) is the leading risk factor to develop COPD. Therefore, the pathologic effects of whole CS on the differentiation of primary small airway epithelial cells (SAEC) were investigated, using cells from three healthy donors and three COPD patients, cultured under ALI (air-liquid interface) conditions. The analysis of the epithelial physiology demonstrated that CS impaired barrier formation and reduced cilia beat activity. Although, COPD-derived ALI cultures preserved some features known from COPD patients, CS-induced effects were similarly pronounced in ALI cultures from patients compared to healthy controls. RNA sequencing analyses revealed the deregulation of marker genes for basal and secretory cells upon CS exposure. The comparison between gene signatures obtained from the in vitro model (CS vs. air) with a published data set from human epithelial brushes (smoker vs. non-smoker) revealed a high degree of similarity between deregulated genes and pathways induced by CS. Taken together, whole cigarette smoke alters the differentiation of small airway basal cells in vitro. The established model showed a good translatability to the situation in vivo. Thus, the model can help to identify and test novel therapeutic approaches to restore the impaired epithelial repair mechanisms in COPD, which is still a high medical need.

Deep learning enables pathologist-like scoring of NASH models.

Non-alcoholic fatty liver disease (NAFLD) and the progressive form of non-alcoholic steatohepatitis (NASH) are diseases of major importance with a high unmet medical need. Efficacy studies on novel compounds to treat NAFLD/NASH using disease models are frequently evaluated using established histological feature scores on ballooning, inflammation, steatosis and fibrosis. These features are assessed by a trained pathologist using microscopy and assigned discrete scores. We demonstrate how to automate these scores with convolutional neural networks (CNNs). Whole slide images of stained liver sections are analyzed using two different scales with four CNNs, each specialized for one of four histopathological features. A continuous value is obtained to quantify the extent of each feature, which can be used directly to provide a high resolution readout. In addition, the continuous values can be mapped to obtain the established discrete pathologist-like scores. The automated deep learning-based scores show good agreement with the trainer - a human pathologist.

Deep neural network based histological scoring of lung fibrosis and inflammation in the mouse model system.

Preclinical studies of novel compounds rely on quantitative readouts from animal models. Frequently employed readouts from histopathological tissue scoring are time consuming, require highly specialized staff and are subject to inherent variability. Recent advances in deep convolutional neural networks (CNN) now allow automating such scoring tasks. Here, we demonstrate this for the case of the Ashcroft fibrosis score and a newly developed inflammation score to characterize fibrotic and inflammatory lung diseases. Sections of lung tissue from mice exhibiting a wide range of fibrotic and inflammatory states were stained with Masson trichrome. Whole slide scans using a 20x objective were acquired and cut into smaller tiles of 512x512 pixels. The tiles were subsequently classified by specialized CNNs, either an "Ashcroft fibrosis CNN" or an "inflammation CNN". For the Ashcroft fibrosis score the CNN was fine-tuned by using 14000 labelled tiles. For the inflammation score the CNN was trained with 3500 labelled tiles. After training, the Ashcroft fibrosis CNN achieved an accuracy of 79.5% and the inflammation CNN an accuracy of 80.0%. An error analysis revealed that misclassifications are almost exclusively with neighboring scores, which reflects the inherent ambiguity of parts of the data. The variability between two experts was found to be larger than the variability between the CNN classifications and the ground truth. The CNN generated Ashcroft score was in very good agreement with the score of a pathologist (r2 = 0.92). Our results demonstrate that costly and time consuming scoring tasks can be automated and standardized with deep learning. New scores such as the inflammation score can be easily developed with the approach presented here.

Reflection of successful anticancer drug development processes in the literature.

The development of cancer drugs is time-consuming and expensive. In particular, failures in late-stage clinical trials are a major cost driver for pharmaceutical companies. This puts a high demand on methods that provide insights into the success chances of new potential medicines. In this study, we systematically analyze publication patterns emerging along the drug discovery process of targeted cancer therapies, starting from basic research to drug approval - or failure. We find clear differences in the patterns of approved drugs compared with those that failed in Phase II/III. Feeding these features into a machine learning classifier allows us to predict the approval or failure of a targeted cancer drug significantly better than educated guessing. We believe that these findings could lead to novel measures for supporting decision making in drug development.

Diffusion coefficients and dissociation constants of enhanced green fluorescent protein binding to free standing membranes.

Recently, a new and versatile assay to determine the partitioning coefficient [Formula: see text] as a measure for the affinity of peripheral membrane proteins for lipid bilayers was presented in the research article entitled, "Introducing a fluorescence-based standard to quantify protein partitioning into membranes" [1]. Here, the well-characterized binding of hexahistidine-tag (His6) to NTA(Ni) was utilized. Complementarily, this data article reports the average diffusion coefficient [Formula: see text] of His6-tagged enhanced green fluorescent protein (eGFP-His6) and the fluorescent lipid analog ATTO-647N-DOPE in giant unilamellar vesicles (GUVs) containing different amounts of NTA(Ni) lipids. In addition, dissociation constants [Formula: see text] of the NTA(Ni)/eGFP-His6 system are reported. Further, a conversion between [Formula: see text] and [Formula: see text] is provided.

Mapping nanomechanical properties of freshly grown, native, interlamellar organic sheets on flat pearl nacre.

We imaged surfaces of freshly grown flat pearl nacre (Haliotis tuberculata) in different stages of growth in seawater using an atomic force microscope (AFM). Characteristic mineral phases of nacre, such as aragonitic stacks of coins, as well as the associated organic sheets, could be detected. Apart from imaging, the acquisition of force volumes on freshly grown organic surface areas on flat pearl nacre was conducted with the AFM. The evaluation of the force volumes with the Hertz-Sneddon model resulted in Young's moduli in the MPa range. The presented values are considerably smaller than values previously determined from macroscopic tensile tests. This might reflect the anisotropy of the organic nacre layers.

Myosin motors fragment and compact membrane-bound actin filaments.

Cell cortex remodeling during cell division is a result of myofilament-driven contractility of the cortical membrane-bound actin meshwork. Little is known about the interaction between individual myofilaments and membrane-bound actin filaments. Here we reconstituted a minimal actin cortex to directly visualize the action of individual myofilaments on membrane-bound actin filaments using TIRF microscopy. We show that synthetic myofilaments fragment and compact membrane-bound actin while processively moving along actin filaments. We propose a mechanism by which tension builds up between the ends of myofilaments, resulting in compressive stress exerted to single actin filaments, causing their buckling and breakage. Modeling of this mechanism revealed that sufficient force (∼20 pN) can be generated by single myofilaments to buckle and break actin filaments. This mechanism of filament fragmentation and compaction may contribute to actin turnover and cortex reorganization during cytokinesis.DOI:http://dx.doi.org/10.7554/eLife.00116.001.

The design of MACs (minimal actin cortices).

The actin cell cortex in eukaryotic cells is a key player in controlling and maintaining the shape of cells, and in driving major shape changes such as in cytokinesis. It is thereby constantly being remodeled. Cell shape changes require forces acting on membranes that are generated by the interplay of membrane coupled actin filaments and assemblies of myosin motors. Little is known about how their interaction regulates actin cell cortex remodeling and cell shape changes. Because of the vital importance of actin, myosin motors and the cell membrane, selective in vivo experiments and manipulations are often difficult to perform or not feasible. Thus, the intelligent design of minimal in vitro systems for actin-myosin-membrane interactions could pave a way for investigating actin cell cortex mechanics in a detailed and quantitative manner. Here, we present and discuss the design of several bottom-up in vitro systems accomplishing the coupling of actin filaments to artificial membranes, where key parameters such as actin densities and membrane properties can be varied in a controlled manner. Insights gained from these in vitro systems may help to uncover fundamental principles of how exactly actin-myosin-membrane interactions govern actin cortex remodeling and membrane properties for cell shape changes.

Gastropod nacre: structure, properties and growth--biological, chemical and physical basics.

The biogenic polymer/mineral composite nacre is a non-brittle biological ceramic, which self-organizes in aqueous environment and under ambient conditions. It is therefore an important model for new sustainable materials. Its highly controlled structural organization of mineral and organic components at all scales down to the nano- and molecular scales is guided by organic molecules. These molecules then get incorporated into the material to be responsible for properties like fracture mechanics, beauty and corrosion resistance. We report here on structure, properties and growth of columnar (gastropod) nacre with emphasis on the genus Haliotis in contrast to sheet nacre of many bivalves.