An Optimized High-Content Imaging Workflow for 3D Spheroid Cell Models

InSphero and Yokogawa Life Sciences discuss best practices for model selection, sample processing, image acquisition, and analysis

By virtue of their more physiologically relevant architecture and cell-cell interactions, 3D cell culture models are proving to be a better alternative to 2D cell culture models for predicting the efficacy and safety of drugs in human patients.

3D microtissues, resembling human tissue in both form and function, have been shown to be amenable to parallelized screening approaches. However, most 3D screens rely on global endpoints such as viability, size, or integrated fluorescent signal, and therefore do not examine drug action at the level of the individual cell populations or provide spatial information. Alternative higher resolution methods, such as classical histology and flow cytometry, are generally low throughput and therefore are incompatible with screening approaches.

Pairing 3D models with translational endpoints

High-content imaging (HCI) is a high-resolution platform originally developed for automated imaging and analysis of 2D cell–based assays. Newer confocal HCI instruments, designed with 3D cell models in mind, have the potential to extract high-resolution data, akin to clinical histology.

To prepare 3D models for HCI, a whole-mount fix and stain procedure is utilized which eliminates the need to prepare physical sections of the tissue. Further, 3D whole-mount samples can be fixed, stained, and imaged in situ (that is, in the original culture plate), reducing the risk of compromising or losing the sample during transfer. If one has the ability to generate optical sections many cell layers deep, one may use this ability to obtain important spatial information, such as drug effects within nutrient-starved or hypoxic zones, as well as the penetration depth of biologics or infiltrating immune cells.

As with implementation of any new technology platform, there are challenges to be considered. In the case of 3D imaging, the breadth of these challenges is not always fully appreciated. In this tutorial, we examine best practices for each step including the model selection, sample processing, image acquisition, and analysis.

Balancing complexity with scalability

As cell model complexity increases, as is the case when transitioning from 2D to multicellular 3D models, scalability inevitably decreases. Two major contributors are variability introduced by the 3D model and limitations of the 3D analytical method.

To maximize scalability, a 3D model destined for HCI should be only as complex as necessary, or “fit for purpose.” 3D models that vary considerably in terms of size and cellular composition can lead to reproducibility issues, thereby limiting assay robustness. Standardizing the 3D model production process and maintaining large inventories of prequalified cell lots can be instrumental in reducing model variation.

Large 3D models (>250 μm in Z-height) may also have limited compatibility with HCI approaches. Light is attenuated as it passes through dense 3D models, until it reaches a depth where the emitted light is no longer sufficient to produce a high-contrast image. Therefore, size limits must be considered and balanced with downstream analysis requirements. For example, imaging depth may be relevant depending on the biological question of interest.

Understanding 3D challenges

Nonuniform fluorescent labeling is a common issue for 3D models and often results from uneven diffusion of the fluorescent probe into thick 3D samples. Fix-and-stain protocols originally developed for 2D cell culture may require substantial adaptation for 3D models. For example, permeabilization and staining steps for 3D models may need to be carried out over a more extended time frame and at higher temperatures and require addition of harsher detergents and solvents to sufficiently “open up” the tissues. In addition, selecting the optimal probe concentration requires careful titration to avoid a binding-site barrier effect or nonspecific background.

Significant light scattering, resulting from refractive index (RI) mismatches, occurs as the light passes through 3D samples (Figure 1). Although 3D models are mostly water (RI = 1.33), they also contain regions of concentrated proteins (RI > 1.44), lipids, and organelles (RI > 1.45), and in some cases acellular matrices and scaffolds. Light scattering may limit effective image analysis to only 25–50 μm for uncleared spheroids, and 100–175 μm for cleared spheroids. The application of 3D clearing methods (BABB, ScaleS, etc.) can greatly ameliorate light scattering issues by homogenizing the refractive index of 3D tissue, rendering it transparent.

Figure 1. The effect of light scattering on object segmentation. (A) A cartoon of light scattering caused by refractive index mismatches within uncleared and cleared spheroids. (B) Images of Z sections acquired at a depth of approximately 50 μm from both uncleared and cleared pancreatic islet microtissues. Spheroid imaging was performed in a 25-μm ultrathin FEP-bottomed 384-well plate using a 20 × 0.75 numerical aperture dry objective on a confocal high-content imaging instrument equipped with spinning disk plus microlens technology, a laser light source, and a scientific CMOS camera. The corresponding nuclear segmentation overlays and X–Z centroid maps illustrate the significantly reduced depth at which nuclei are accurately segmented in uncleared spheroids.

Light scattering can also result from microplates that are not optimized for imaging. The ideal 3D imaging plate has a flat, thin, and transparent bottom with high planarity to minimize spherical aberrations; a low skirt height to accommodate high numerical aperture objectives; black walls to prevent interwell crosstalk; and a well geometry that minimizes sample loss during processing.

3D image acquisition

The keys to acquiring volumetric images are selecting sampling parameters that enable accurate measurements and understanding the resolution limits of the system (Figure 2). Sampling 3D cell models typically involves counting individual cells and measuring the shape, volume, and intensity of those cells. Optical sectioning is achieved using a confocal microscope, typically powered by laser sources.

Figure 2. High-throughput, high-content screening in a heterotypic 3D tumor model. (A) Maximum intensity projections from a 384-well plate of a tumor co-culture model comprised of NCI-N87–GFP (green fluorescent protein) and NIH-3T3–RFP (red fluorescent protein). (B) Channel 1, channel 2, and merged images from a representative Z section, plus segmentation of whole tumor (yellow) and stroma (pink) regions of interest. (C) Bar graphs showing results from a 3D volumetric analysis of the tumor and stroma total volumes.

There are two primary types of confocal systems used for 3D high-content cell measurements. Point scanners build each image a single pixel at a time, whereas spinning disk modules can acquire an entire field of view simultaneously.

The faster acquisition of the spinning disk platform reduces sample degradation caused by photobleaching. Consequently, living samples can remain viable over days or even weeks with proper environmental control.

Fast acquisition is a key factor for high-content analysis involving thousands of wells. Live-cell imaging by inclusion of environmental control can reduce the total number of plates and experiment cost. Tracking the time course of individual cells may produce more relevant data, compared with samples which have been fixed at different time points. Spinning disk systems are optimized to block out-of-focus light through a combination of pinhole size and distance between pinholes. To increase light collection, more advanced systems utilize a microlens on the pinhole disk which improves the optical sectioning and signal-to-noise ratio and has a major impact on the segmentation and analysis of cells in microtissues.

3D data analysis considerations

3D image analysis enables the creation of 3D models which more accurately represent the true shape, location, and relationships between cells. In addition to intensity and morphological measures, the analysis software may use local intensity variation, known as texture, to further describe features of cells or subcellular components. The increased information allows for a more physiologically relevant measure of phenotypical responses from the sampled tissues.

A high signal-to-noise ratio can enable automated and consistent segmentation of 3D images and is increasingly important for larger 3D data sets. Object identification protocols should intelligently vary settings to compensate for lower signals deeper inside the samples.

For automated measurements of 3D microtissues, correct lateral (X,Y) and especially axial (Z) sampling needs to be considered. Axial undersampling can result in incorrect segmentation across adjacent optical planes. For accurate cell counts, at least two image planes per countable object should be acquired.

To fully exploit the potential of 3D microtissues combined with 3D HCI, we encourage researchers to regularly review and adapt their experimental designs (Figure 3).

Figure 3. A checklist of considerations for high-resolution, high-content imaging of complex 3D models.

In conclusion, extracting translational data from 3D models can enable better drug selection by evaluating efficacy, selectivity, and safety of drug candidates in an in vitro correlate of human tissue. The added throughput of HCI assays enables the application of a 3D platform during the early stages of drug discovery prior to chemotype selection and molecular pharmacology decisions. If successful, this approach could one day reduce the use of preclinical animal models in drug discovery.

Judith Wardwell-Swanson is a senior scientist at InSphero, Arvonn Tully is a life sciences senior software applications specialist at Yokogawa Life Science , Özlem Yavas, PhD, is an engineer of organ-on-chip technologies at InSphero, Mahomi Suzuki is a senior application specialist at Yokogawa Life Science, and Olivier Frey, PhD, is vice president of technologies and platforms at InSphero. Websites: InSphero and Yokogawa Life Science.