Background Hepatotoxicity is one of the leading causes for withdrawal of drugs from the market or limitations in their use1. The design of cell-based assays to identify toxic effects early in discovery should reduce late-stage attrition. One potential indicator of hepatotoxicity is phospholipidosis, an increased uptake of phospholipid in multi-lamellar vesicles. Here we present data from cultured primary hepatocytes treated with phospholipidosis-inducing compounds, followed by analysis with the Definiens Enterprise Image IntelligenceTM Suite. This method not only identifies global changes but can also reveal subtle differences in cellular physiology, potentially leading to more informed decisions when taking compounds forward. Methods Rat primary hepatocytes were cultured in 96 well plates, treated with control and test compounds and labelled with synthetic phospholipid coupled to a fluorescent marker. Cellular DNA was labelled with Hoechst 33342 (Fig. 1). Images from the plates were captured with a GE INCell Analyzer 3000. The images were analysed using Definiens Enterprise Image IntelligenceTM platform. This uses Definiens Cognition Network Technology2 to build a topological hierarchical network of image objects (fig 2A). Modules derived from the High Content Analysis application plug-in Definiens Cellenger® were used to identify cells, nuclei and phospholpipidotic vesicles, while customised modules to address other tasks were written in Definiens Developer and plugged into the solution (fig. 2B). Results 1. Cellular Segmentation Due to heterogeneity in the phospholipid (red) channel across treatments, information in the Hoechst channel was used to generate both nucleus and cell boundary information. This was accomplished despite very low signals at cell boundaries (fig. 3). 2. Live/Dead Cellular Classification Four distinct cellular phenotypes were observed, based on cellular and nuclear morphology. Healthy cells possessed large, round nuclei, while dead cells either showed large, granular nuclei, small, bright nuclei, or no nuclei (enucleated). These phenotypes were reflected in the class hierarchy, allowing extraction of information at different levels of detail (fig. 4). 3. Standard readout: drug toxicity Cells were classified as alive or dead based on the classification criteria above. Toxicity-response curves of a representative plate’s data are shown below. One compound (C1) exhibited significant toxicity at higher doses of drug, from which the LD50 could be calculated (fig. 5). 4. Standard readout: Phospholipidosis The mean integrated intensity of phospholipid dye is one of the standard readouts to measure phospholipidosis. Only vesicles in viable cells were considered for this response. For the compounds examined, C1 exhibited a strong phospholipidotic response, C2, C3 and C6 exhibited weak responses, while C4 and C5 were negative. The maximal phospholipidotic response occurs at concentrations around 7 times lower than the toxic dose of drug (see compound 1, figure 6). 5. Subtle effects: vesicle location and clustering Examination of many of the images, it would appear that there is a very strong bias for phospholipidotic vesicles to align themselves with membranes bordering other adjacent cells, as opposed to “naked”cell borders, bordering background. We examined whether this was the case by sub-classifying vesicles according to their locations relative to the two different categories of membrane. Analysis of total numbers of vesicles demonstrate over 80% are classified as being aligned to membranes associated with other cells; however, when normalised to the length of cellular or background membrane, the response was far less marked (fig 7). 6. Probing toxicity: nuclear phenotype We then examined in more detail the mechanism of death due to toxic compounds. The overall proportion (and number) of enucleated cells remains relatively constant across the wells, demonstrating that this phenotype appears to be due to natural attrition, independent of treatment. However, at high doses, compound 1 causes a shift in nuclear phenotype away from the small/bright irregular (typical apoptotic) phenotype to the large, granular phenotype. This effect was not seen in cells treated with 80µM amiodarone, the cytotoxic control (fig. 8), indicating that different mechanisms of death are associated with different compounds. 7. Responses of mono-vs multinucleated cells It is not clear whether mono- or multinucleated hepatocytes behave differently with drug treatments. Below we examine the average integrated vesicle intensity for the phospholipidotic control and test compounds at the concentration exhibiting maximal phospholipidotic response. For rigorous examination individual intensities are normalised the area of the individual cells. Multinucleated cells are on average larger than mononucleated cells, the spot intensity values were normalised to the area of each individual cell. While there appeared to be a small difference between single and multinucleated cells in response to compound C1, this difference was not statistically significant (fig. 9). Summary ● Automated, object-based image analysis was used to investigate phospholipidosis. ● Accurate and robust cellular classification was achieved despite low signal: background ratios. ● Straightforward readouts gave basic information on phospholipidotic and toxic cellular responses. ● Hierarchical segmentation and classification of cellular constituents allows extensive, multivariate analysis of subtle biological effects. ● These can be used to obtain deeper insights into assay systems and make more informed decisions when progressing candidate compounds. References 1. Fung M; et al (2001). Drug Information Journal 35, 293–317 2. Baatz M, et al. (2006). Cytometry Analysis (2006) 69, 652-658 AUTOMATED MULTIVARIATE ANALYSIS OF PHOSPHOLIPIDOSIS IN PRIMARY HEPATOCYTES USING DEFINIENS CELLENGER® AND DEFINIENS DEVELOPER Edward Ainscow, James Pilling, Nick Brown, Elaine Sullivan, Mike Sullivan. Advanced Science and Technology Laboratory, AstraZeneca, Charnwood, Loughborough LE11 5RH, UK. Nick Arini, Bettina Linssen, Barbara Zenger-Landolt, René Korn, Thomas Harberichter, Andrzej Cholewinski, Colin Blackmore. Definiens AG, Trappentreustrasse 1, 80339 Muenchen, Germany www.definiens.com Deeper insights. Faster results. Better decisions. Figure 1: Phospholipidosis Assay. Primary rat hepatocytes labelled with synthetic phospholipid (red) and Hoechst (blue). A = negative control; B = toxic control; C = phospholipidosis. A B C Figure 1: Phospholipidosis Assay. Primary rat hepatocytes labelled with synthetic phospholipid (red) and Hoechst (blue). A = negative control; B = toxic control; C = phospholipidosis. A B C Figure 1: Phospholipidosis Assay. Primary rat hepatocytes labelled with synthetic phospholipid (red) and Hoechst (blue). A = negative control; B = toxic control; C = phospholipidosis. A B C A B C Figure 3: Cellular segmentation. The Hoechst channel (A = shown with gamma correction) was used for cellular segmentation, despite low contrast between cells (B). Segmentation, showing mononuclear and multi-nucleated hepatocytes is shown in (C). A C Figure 3: Cellular segmentation. The Hoechst channel (A = shown with gamma correction) was used for cellular segmentation, despite low contrast between cells (B). Segmentation, showing mononuclear and multi-nucleated hepatocytes is shown in (C). A C Figure 3: Cellular segmentation. The Hoechst channel (A = shown with gamma correction) was used for cellular segmentation, despite low contrast between cells (B). Segmentation, showing mononuclear and multi-nucleated hepatocytes is shown in (C). A C 0 200 400 600 800 1000 1200 0 50 100 150 200 Distance (pixels) Intensity (grey levels) A B C 0 10 20 30 40 50 60 70 80 90 100 1 10 100 1000 Relative Dose % DeadCells LD 50 =34 a.u. Figure 5: Drug Toxicity Curves. A = Heatmap, with positive, negative and phospholipidosis controls in columns 2 and 11, and dose responses (decreasing col. 3 to col. 10) for 6 compounds. B = Toxicity curves for 6 compounds (C1-C6). B A 0 10 20 30 40 50 60 70 80 90 100 1 10 100 1000 Relative Dose % DeadCells C1 C2 C3 C4 C5 C6 LD 50 =34 a.u. Figure 5: Drug Toxicity Curves. A = Heatmap, with positive, negative and phospholipidosis controls in columns 2 and 11, and dose responses (decreasing col. 3 to col. 10) for 6 compounds. B = Toxicity curves for 6 compounds (C1-C6). B A 0 5000 10000 15000 20000 25000 30000 1 10 100 1000 Relative Dose Mean Integrated Spot Intensity EC 50 C1 = 4.5 a.u. EC 50 C6 = 10.8 a.u Figure 6: Phospholipidosis dose responses. Dose responses of 6 compounds, with EC50 calculations for compounds 1 and 6. 0 5000 10000 15000 20000 25000 30000 1 10 100 1000 Relative Dose Mean Integrated Spot Intensity C1 C2 C3 C4 C5 C6 EC 50 C1 = 4.5 a.u. EC 50 C6 = 10.8 a.u Figure 6: Phospholipidosis dose responses. Dose responses of 6 compounds, with EC50 calculations for compounds 1 and 6. Membrane vesicles (red) Background vesicles (yellow) Dead cell vesicles (green) 0 20 40 60 80 100 120 C1 C2 C3 PLD Control % Vesicles Classified by Location (Normalised) 0 20 40 60 80 100 120 C1 C2 C3 PLD Control % Vesicles Classified by Location Treatment A B C Figure 7: Location of phospholipidotic vesicles. The majority of vesicles appear to be preferentially aligned with membranes adjacent to other cells (A, B). Normalisation to the length of cell-to-cell or cell-to-background membrane indicates that the effect exists, but is more subtle than visual inspection. Membrane vesicles (red) Background vesicles (yellow) Dead cell vesicles (green) 0 20 40 60 80 100 120 C1 C2 C3 PLD Control 0 20 40 60 80 100 120 C1 C2 C3 PLD Control % Vesicles Classified by Location (Normalised) 0 20 40 60 80 100 120 C1 C2 C3 PLD Control % Vesicles Classified by Location 0 20 40 60 80 100 120 C1 C2 C3 PLD Control % Vesicles Classified by Location Membrane vesicles Background vesicles Treatment A B C Figure 7: Location of phospholipidotic vesicles. The majority of vesicles appear to be preferentially aligned with membranes adjacent to other cells (A, B). Normalisation to the length of cell-to-cell or cell-to-background membrane indicates that the effect exists, but is more subtle than visual inspection. Figure 8: Dead cell phenotype. The profile of profile of dead nuclei shifts towards the large, granular phenotype with increasing doses of compound 1. This behaviour is different from treatment with cytotoxic controls, indicating different mechanisms of cell death occur at higher doses. 0 20 40 60 80 100 120 Neg. Control Toxic Control 1 2 4 8 16 32 64 128 Relative Dose % Dead Cells of Class Figure 8: Dead cell phenotype. The profile of profile of dead nuclei shifts towards the large, granular phenotype with increasing doses of compound 1. This behaviour is different from treatment with cytotoxic controls, indicating different mechanisms of cell death occur at higher doses. 0 20 40 60 80 100 120 Neg. Control Toxic Control 1 2 4 8 16 32 64 128 Relative Dose % Dead Cells of Class 0 20 40 60 80 100 120 Neg. Control Toxic Control 1 2 4 8 16 32 64 128 Relative Dose % Dead Cells of Class Granular Irregular Enucleated Figure 9: Comparison of mono-and multinucleated cells. Mean integrated spot intensity normalised to cell area. Error bars, ± sem. 0 5 10 15 20 25 30 35 40 45 C1 C2 C6 Controls Normalised Spot Intensity per Cell Single Nuc Multi Nuc Figure 9: Comparison of mono-and multinucleated cells. Mean integrated spot intensity normalised to cell area. Error bars, ± sem. Figure 2: Definiens Cognition Network Technology - Topology and Modules. The technology builds a network of image objects to describe elements in the image (A). Pre-defined and customised analysis modules were configured to perform the analysis task (B). Individual pixels Cell- background membrane vesicles Cell Background Nucleus Cytoplasm Nuclear substructure Cell-cell membrane vesicles Background Background Spot Level Nucleus Level Cell Level A B Figure 2: Definiens Cognition Network Technology - Topology and Modules. The technology builds a network of image objects to describe elements in the image (A). Pre-defined and customised analysis modules were configured to perform the analysis task (B). Individual pixels Cell- background membrane vesicles Cell Background Whole Image (Scene) Nucleus Cytoplasm Nuclear substructure Cell-cell membrane vesicles Background Background Spot Level Nucleus Level Cell Level A B Figure 4: Nucleus and Cellular Classification. Images (left) show different phenotypes recognised and classified during the analysis, while the class hierarchy (right) enables multivariate data reporting of different object classes. Alive Granular Irregular Enucleated Figure 4: Nucleus and Cellular Classification. Images (left) show different phenotypes recognised and classified during the analysis, while the class hierarchy (right) enables multivariate data reporting of different object classes. Alive Granular Irregular Enucleated Figure 4: Nucleus and Cellular Classification. Images (left) show different phenotypes recognised and classified during the analysis, while the class hierarchy (right) enables multivariate data reporting of different object classes. Alive Granular Irregular Enucleated SPL-0001-01-040407