Human HCT116 colorectal cancer cells (ATCC; Teddington, UK) were cultured in Dulbecco's modified Eagle's medium (DMEM) + Glutamax™ (Thermo Fisher Scientific, Inc., Waltham, USA) containing 10% fetal calf serum (Thermo Fisher Scientific, Inc.) with penicillin/streptomycin (Sigma-Aldrich, St. Louis, MO, USA) in a humidified atmosphere of 5% CO₂ at 37°C. A suspension of cells was inoculated in Biomimesys^® (Celenys, Rouen, France), commercially available 96-well plates containing a hyaluronic acid matrix. A total of 50,000 cells were seeded in 25 µl of culture media per well. Following incubation at 37°C, in 5% CO₂ for 30 min, 175 µl of culture media was added to each well. After 3 days, half of the volume of the culture medium was renewed. Six days following cell seeding, 100 µl of culture media per well was replaced by 100 µl of culture media containing a 2X topotecan concentration (Sigma-Aldrich, St. Louis, MO, USA). On each plate, 6 different concentrations were applied with 6 wells treated per concentration. Plates were fixed 72 h following treatment.

Hydrogel matrices were fixed for 15 min at room temperature with 4% paraformaldehyde in phosphate-buffered saline (PBS). Following 3 washes in PBS, the samples were permeabilized for 6 min in 0.5% Triton X-100 solution, washed 3 times in PBS and subsequently incubated with 1 µg/ml DAPI fluorescent nuclear stain for 10 min prior to PBS wash.

Image acquisition was performed using an inverted widefield fluorescence microscope, Axio Observer. Z1, with ApoTome fitted with an axiocam 506 mono camera (Zeiss GmbH, Jena, Germany; objective magnification, ×10 and numerical aperture 0.3). For each well, 9 tiles with 10% surface overlap were acquired. For each tile, a 400–500 µm-z stack with a 20 µm-z step was performed with 3 images corresponding to different ApoTome grid positions per z position. A total of 9 z-stacks were acquired per well. For each z-stack, ApoTome deconvolution was performed using ZEN software ver. 2012 (Zeiss GmbH).

For each z-stack, a maximum projection was performed and the 9 images obtained per well were stitched together. Image filtering, projections and the stitching process were executed by an ImageJ macro running on a Dell™ T7X series Workstation equipped with 192 GB RAM, a Xeon-phi 32 core processor and a 4 GB VRAM NVidia Graphics card. This level of power is required to process large batches of images (>60 megapixels per image). The stitched images were subsequently processed using a MATLAB^® script to remove noise and background, set saturation level, and perform Otsu's automatic threshold and morphological dilatation on stitched images (11). Combining the resulting binary mask with the original image made it possible to select only those pixels corresponding to clusters along with those in their close vicinity. This pre-segmented image was then resampled on 10 levels of grey, with the elimination of the first one leading to final segmentation. For the final step of the analysis process, i.e. quantification, MATLAB's built-in methods of clustering and quantification were used exclusively to compute the per cluster projected area and average intensity response.

Fig. 1 summarizes the workflow and the specific procedures reported, starting from the seeding step of cancer cells in the matrix and 3D sphere formation to the final results of the quantification analysis. As illustrated, 96-well plates containing a hydrogel made of cross-linked hyaluronic acid were seeded with HCT116 colon adenocarcinoma cells. It has been demonstrated that cancer cells proliferate actively in such a matrix (12), recapitulating optimal microenvironmental conditions. After 6 days of growth, a number of 3D cell clusters of various shapes and sizes were formed and detected under bright field microscopy (Fig. 1).

To individually characterize a large number of spheres per well, spheres were fixed inside the matrix and stained with DAPI to be detected by fluorescence microscopy. The aim of the present study was to acquire images of a large number of spheres from each of the 96 wells, therefore, multi-position 3D fluorescence microscopy was utilised. Structured illumination was used with an ApoTome fitted to a conventional wide-field fluorescence microscope. This technology is not based on pixel-by-pixel scanning, therefore it allows for a much more rapid z-stack acquisition of all in-depth information from each of the 96 wells. Individual acquisitions consist of a 400–500 µm z-stack, with each z separated by 20 µm (Fig. 1). A movie (available for download at https://mycore.core-cloud.net/public.php?service=files&t=bcd82c01cd45e93c3c55ce5e175cd7cb) presents a z-stack acquisition of a field of view prior to and following ApoTome deconvolution. These data indicate that following ApoTome deconvolution, nuclei of individual spheres are visible and it is possible to identify each sphere. These findings are confirmed with the maximal projection of the z-stack (data not shown), which makes it possible to qualitatively assess proliferation and count the number of individual cell clusters.

A total of 9 z-stack acquisitions were performed for each well to avoid restricting the analysis to a single field of view and to provide global quantitative information (Figs. 1 and 2A). The nine 400 µm z-stack acquisitions with ApoTome covered half of the well area and made it possible to identify each sphere in the volume. To generate the global image of each micro-well (Fig. 2A), the z-stacks of the 9 overlapping tiles of each well were stitched together using a previously described procedure (13). This procedure was applied to wells treated with increasing doses of the DNA topoisomerase inhibitor, topotecan. A total of 6 concentrations were used and 6 wells were treated for each concentration. The stitched images obtained for each well are presented in the mosaic image (Fig. 2B), which represents a virtual multiwell plate reconstructing each well using the aforementioned procedure and including fluorescence information for spheres throughout the matrix in the well's central area (~13 mm²). Cluster size decreases upon treatment with higher concentrations of topotecan, thereby making it possible to qualitatively evaluate the effect of this drug on matrix-embedded multispheres. This representation provides a global view of the experiment, offering an assessment of the effect of a given drug.

Stitched images from ApoTome acquisition provided all of the cancer cell cluster information available in each well. However, its analysis and interpretation require further processing (Figs. 1 and 3). To this end, a methodology was developed based on image analysis tools and was subsequently applied to images of the mosaic (Fig. 2B). This provided morphometric and multiparametric quantitative data, including cluster size distribution.

The objective was to avoid any manual threshold for image segmentation and to use a single tool to identify large clusters, as well as individual cells. The first step was to filter stitched images and perform a pre-segmentation pass to eliminate background and focus on the upper class of Otsu's segmentation method (11). Extracted areas were then expanded, creating a very coarse mask (binary image, Fig. 3) to identify areas of interest containing cell clusters. Following the combination of this binary mask with the original stitched image, the pixels corresponding to the background were set to 0, leaving only pixels corresponding to clusters and their near vicinity. The goal here was to avoid removing any of the few multi-sized clusters. To account for the variability of nuclei fluorescence intensities that disrupt traditional segmentation algorithms and to more accurately distinguish clusters from background noise, an intensity scale was used, resampling in 10 levels. Finally, an additional segmentation and refinement was performed from the 9 stitched image tiles to obtain the boundary of each cluster (pink lines on the final image, Fig. 3) and to extract morphometric features.

The quantitative analysis methodology was applied to the stitched images of the wells following topotecan treatment at different concentrations. Fig. 4A illustrates the result of the final segmentation performed on cancer tumor spheroids grown in matrix in a 96-well format and treated with increasing concentrations of the topoisomerase inhibitor topotecan. The left panel presents an untreated control well, where a large number of 3D clusters of various sizes were detected and segmented. The middle and right panel show wells treated, respectively, with 75 nM and 10 µM topotecan. As evidenced, the segmentation process detects fewer and mostly smaller clusters in a topotecan concentration-dependent manner.

Fig. 4B illustrates the distribution of the size (median, quartile and extreme values) of the detected clusters in these three conditions. The data show that the median decreases in cells treated with ≥250 nM topotecan, with a clear decrease of the interquartile range from 1 µM, thus highlighting a marked decrease of cluster size in response to an increasing dose of topotecan. Cluster size distribution in the control illustrates the great heterogeneity normally observed in such an assay. This heterogeneity is largely lost upon treatment with topotecan and only very small clusters or single cells are detected.

Additional parameters may be extracted from this experiment, including the evolution of the distribution by class of cluster size (Fig. 4C). These data quantitatively reflect the loss of large clusters and the accumulation of aggregates of a very small number of cells, in cells treated with ≥250 nM.

In conclusion, these results illustrate the power of the approach and capability of this imaging and processing pipeline to detect and accurately quantify the effect of a compound, such as a cytotoxic agent, on a heterogeneous cell population grown in 3D in a hyaluronic acid matrix. Powerful methods for reconstructing and live 3D in vivo analysis of the process of cancer cell aggregation and growth have already been documented, however, they rely on the use of sophisticated microscopy and/or image processing tools (14,15). The approach reported in the present study aims to provide a simple, robust pipeline to do so based on technologies that are readily available to many investigators.