3D Cell-Based Assays for Drug Screens: Challenges in Imaging, Image Analysis, and High-Content Analysis

Abstract

The introduction of more relevant cell models in early preclinical drug discovery, combined with high-content imaging and automated analysis, is expected to increase the quality of compounds progressing to preclinical stages in the drug development pipeline. In this review we discuss the current switch to more relevant 3D cell culture models and associated challenges for high-throughput screening and high-content analysis. We propose that overcoming these challenges will enable front-loading the drug discovery pipeline with better biology, extracting the most from that biology, and, in general, improving translation between in vitro and in vivo models. This is expected to reduce the proportion of compounds that fail in vivo testing due to a lack of efficacy or to toxicity.

Keywords: 3D culture; 3D 배양; high-content screening; image analysis; imaging; イメージアナリシス; イメージング; ハイコンテントスクリーニング; 三次元細胞培養; 三维细胞培养; 关键词 高内涵筛选; 图像分析; 成像; 대용량 스크리닝; 이미지 분석; 이미징.

Figure 1.

3D cell cultures provide a more physiologically relevant context for drug screening. (A) Prostate carcinoma (PC-3) cells cultured as 2D monolayer (top) show negligible morphological changes in response to growth factor (20 ng/mL hrEGF) stimulation but become invasive if embedded in 3D hydrogels (bottom) after growth factor stimulation. These invasive characteristics can be used to investigate the efficacy of inhibitors of receptor tyrosine kinases. Images in the top panel were obtained using a wide-field BD pathway 855 with a 10× objective, and images in the bottom panel were obtained using a Nikon Ti Eclipse confocal microscope with a 20× objective. (B) mIMCD3 cells transduced with a short-hairpin targeting Pkd1 form a monolayer in 2D culture (left panel, BD pathway 855 with 10× objective), but form cysts in 3D hydrogels, representing a more pathophysiologically relevant model of PKD (right panel, Nikon Ti Eclipse confocal microscope with 20× objective). F-actin (rhodamine-phalloidin), red; nuclei (Hoechst 33258), blue.

Figure 2.

Nomenclature of 3D cell-based assays.

Figure 3.

3D cultures of PDX material. PDX material from different tumors can be cultured in 3D hydrogels to form complex microtissues that can be used for compound screening in a preclinically relevant context. Actin cytoskeleton visualized with rhodamine-phalloidin. PDX tumor material provided by Charles River Labs (Freiburg, Germany). Annotations refer to tumor type and PDX model number; BX = bladder; MAX = mammary; GX = gastric; PAX = pancreatic; LX = lung. 3D cultures and images generated by OcellO B.V.

Figure 4.

Maximum-intensity projections can cause loss of phenotypic information in 3D cultures. (A) Schematic representation of 2D maximum-intensity projections modified from Booij et al. (2016). Structures embedded in hydrogels are captured in xy and z directions using automated microscopy, and in-focus regions from all sections are projected into a 2D reconstruction. (B) 2D projections from 3D structures can cause loss of important phenotypic characteristics. These images display human kidney cyst-derived organoids, cultured in Matrigel and stained for F-actin (rhodamine-phalloidin, red) and nuclei (Hoechst 33258, blue) and imaged on a Nikon Ti Eclipse confocal microscope. Example 3D reconstructions from image sections obtained after imaging with a Nikon Ti Eclipse confocal microscope are included as Supplemental Video 1. Maximum-intensity projection performed with ImageJ software prevents lumen and cell shape detection.

Figure 5.

Exploiting multiparametric data to discriminate responses. (A) Metformin, rapamycin, roscovitine, and sorafenib inhibit forskolin-stimulated PKD cyst swelling, with roscovitine and sorafenib inducing the most potent response based on evaluation of individual parameters. Analysis was performed using Ominer software (OcellO B.V.). All displayed phenotypic parameters are derived from the rhodamine-phalloidin (f-actin) staining of 3D-cultured cysts. (B) Left panel: Three principal components summarizing 84% of variance in the data show a desirable phenotypic change (green arrow) in which 5 mM metformin (blue) and 10 nM rapamycin (green) revert a 2.5 µM forskolin-stimulated phenotype (swollen cyst, empty circles) to one indistinguishable from an unstimulated (solvent) phenotype (solid black circles). Right panel: 31.6 µM roscovitine and 10 µM sorafenib induce an aberrant phenotype (orange arrow); data points represent single wells. Figures adapted from Booij et al. (2017). (C) Two principal components from B showing multiple inhibitors targeting cyclin-dependent kinases (CDK), mammalian target of rapamycin (mTOR), phosphatidylinositol-4,5-bisphosphate 3-kinase (PI3K), human epidermal growth factor receptor 2 (HER2), and polo-like kinase (PLK1). Contour plots represent density estimations to emphasize the locations of forskolin-stimulated (empty circles) and unstimulated controls (solid black circles), respectively. Green arrow represents desirable compound efficacy, from forskolin-stimulated control (empty circles, swollen cysts) to unstimulated control (solid black circles, small cysts), and the orange arrow represents aberrant phenotypes that are observed after treatment with PLK1 inhibitors or high-dose CDK inhibitors, indicative of cytotoxicity.