Download PDF

Hannah A. Strobel, PhD | Staff Scientist | Advanced Solutions Life Sciences Victoria Marsh, PhD | Director, Custom Organoid Services | Molecular Devices

Maria R. Oyaga, PhD | Scientist | Core Life Analytics Angeline Lim, PhD | Applications Scientist | Molecular Devices

Introduction

Patient-derived organoids (PDOs), such as those from tumors, are being used as predictive models of patient responses to drugs and therapies. To optimize these complex and valuable 3D tissue models for drug screening, we leveraged the BioAssembly™ platform. This state-of-the-art, agile automation solution integrated the BioAssemblyBot® (BAB), the BioStorageBot™ modular incubator, and the BioApps™ intuitive workflow control software with confocal imaging in an automated workflow. Following the automated preparation, feeding, drug treatment, and image acquisitions, image sets were analyzed to generate high-content assessments of PDO responses to the drugs.

The BAB platform improves accuracy, reproducibility and efficiency of dispensing PDOs into multiwell plates due to highly tunable operation parameters, such as pipetting speed. This minimizes organoid fragmentation and non-uniform plating. Here, we performed an automated PDO drug screen involving automated imaging and High Content Analysis (HCA) with the BAB platform. The workflow (Figure 1) involved assessing the response of colorectal cancer PDOs from 2 different patients (Molecular Devices assay-ready organoids) to the drugs trametinib and adavosertib (4 different doses each). Up to 76 image features (StratoMineR™; Core Life Analytics) were extracted from the image sets (IN Carta®; Molecular Devices) generated during the automated workflows. These were evaluated for drug- and dose-specific responses. As expected, the automated PDO assay with HCA was able to discriminate differences between PDOs treated with different drugs.

Materials and methods

Colorectal cancer (CRC) PDOs from two different donors were provided by Molecular Devices. PDOs were suspended in 80% Matrigel® and dispensed into a 96-well plate using the BAB and pipette hand. 5µl were dispensed into the center of each well, such that a dome was formed. After gelling, BAB added culture medium to each well and moved the plate to the modular BioStorageBot incubator. On day 2 of culture, BAB changed culture media and added a serial dilution series of trametinib (MEK1/2 inhibitor) or adavosertib (inhibitor of the tyrosine kinase WEE1) to the wells. On day 7, cultures were imaged using an integrated confocal imager following a Hoechst staining protocol. Offline, image analysis was performed on the maximum projection images using IN Carta image analysis software. Multiparametric analyses was performed using StratoMineR (Core Life Analytics). After imaging, a CellTiter-Glo® 3D assay was performed on plates to measure Adenosine triphosphate (ATP) production. All steps of the PDO assay, including the automated imaging protocol, were controlled by a user-defined, PDO BioApp software module.

Figure 1. Workflow for automated PDO drug screen.

Results and discussion

PDOs in Matrigel can be challenging to properly dispense as it must be done precisely in the center of the well to form a dome. The dome is necessary to prevent Matrigel from forming a meniscus at the well edge, which complicates imaging. It takes considerable practice for a scientist to achieve a high success rate of dome formation. In this experiment, the use of the BAB improved the dome formation rate by 10% and dome dispensing time was reduced by 50% compared to an experienced scientist. This demonstrates that BAB can improve speed and consequently throughput, without compromising dome formation (Figure 2).

Figure 2. CRC PDOs seeded in Matrigel domes in a 96 well plate (A). PDOs after 7 days of culture (B) phase contrast image, and (C) calcein AM.

After 2 days, serial dilutions of the trametinib (MEK1/2 inhibitor) or adavosertib (WEE1 kinase inhibitor) were added to wells. The PDOs were incubated for a further 5 days after which an end-point viability assay was carried out. CellTiter-Glo 3D, which measures ATP production as an indirect measure of cell viability, showed clear differences between the highest and lowest drug concentrations, and untreated controls (Figure 3). Despite this, there was no discernable change in organoid diameter in response to the inhibitors (Figure 4).

Figure 3. ATP production of PDOs, as measured by CellTiter-Glo 3D, performed at the end of the assay.

Figure 4. Area of PDOs treated with different concentrations of trametinib or adavosertib, compared to untreated controls. Plates were seeded with PDOs from one of two different donors. Bars are mean ± SD.

Features such as area, form factor, intensity, and texture were extracted from the data set using deep learning-based segmentation in IN Carta and then analyzed using StratoMineR’s cloud-based advanced analytics. While a variety of analytical outcomes are possible with these solutions, we focused on principal component analysis (PCA) and Euclidian distance analyses as baseline readouts for this automated workflow. Principal component analysis (PCA) of up to 76 features identified differential responses by the PDOs to the inhibitors (Figure 5). From the PCA, it appears that PDOs from Donor 1 had a limited response to the two inhibitors (Figure 5). In contrast, PDOs from Donor 2 exhibited clear differences in inhibitor responses in a dose-dependent manner (Figure 5). An evaluation of the hit rate via phenotypic distance mapping confirms that Donor 1 PDOs were relatively unresponsive to the 2 inhibitors (hit rate of 31.5%, Figure 6), while both inhibitors had a significant impact on Donor 2 PDOs (hit rate of 100%, Figure 6).

Figure 5. Heat maps and principal component analysis (PCA). For Donor 1 (PDO 1), 2 components are shown, which show limited clustering within or between drug concentrations. Donor 2, however, had clear clustering in components 1, 2, and 3, within each drug compared to controls.

Figure 6. Box plots showing the Euclidian distance from the median of the control (untreated) for PDO 1 and PDO 2. Everything above the red dotted line has a p value <0.05.

Here, we demonstrate a fully integrated automation platform for screening the effects of therapeutic compounds on patient-derived tumor organoids. In this case, the BioAssemblyBot performed operational tasks in coordination with a modular incubator and high-content imager. Importantly, PDO images were acquired as part of the automation protocol. PDO plates were transferred to the imager and a pre-determined imaging protocol was executed as directed by the PDO BioApps software module.

As expected, the automated PDO assay identified patient-specific differences in PDO responses to selected signaling inhibitors via high-content image analyses. Whilst a more expanded study is required to establish measurements of consistency, accuracy, and precision, this automated workflow produced results comparable to the work performed by a skilled scientist in approximately half the time and with improved efficiencies (e.g. in dome formation). The combination of automated PDO formatting (domes), culturing, treating, and imaging with high-content analysis creates a powerful system for high-throughput drug screening that is highly adaptable yet easy to implement.

Download PDF