Download PDF

Angeline Lim, PhD | Senior Applications Scientist | Molecular Devices Zhisong Tong, PhD | Scientist | Molecular Devices

Introduction

Many oncology drugs fail at the later stages of the drug development pipeline and in clinical trials, despite promising data for their efficacy in vitro¹. This high failure rate is partly attributed to the lack of predictive models used to screen drug candidates in the early stages of drug discovery. As such, there is a need to develop and utilize more representative models that are amendable for efficient compound testing to discover new therapeutic targets.

3D cell models, specifically patient-derived organoids (PDOs), offer a promising solution to this problem. Cells grown in 3D can better mimic cell-cell interactions and the tissue microenvironment, including cancer stem cell niches. Studies show that patients and their derived organoids respond similarly to drugs, suggesting the therapeutic value of using PDOs to improve therapeutic outcomes². However, challenges such as assay reproducibility, scalability, and cost have limited the use of PDOs in mainstream drug discovery pipelines.

To address some of the challenges associated with the use of PDOs in large scale applications, a semi-automated bioreactor was developed for the controlled production of standardized PDOs at scale. Organoids are cultured in a regulated environment that ensures constant delivery of nutrients and growth factors to the culture while preventing the accumulation of toxins, which can lead to cell death. This culture approach enables the large-scale production of assay-ready organoids that are uniform in size and have high viability.

To demonstrate the utility of these PDOs for downstream drug discovery applications, we developed an automated workflow demonstrating the use of patient-derived colorectal cancer (CRC) organoids for a compound screen (Figure 1). Assay-ready CRC organoids were established in culture, maintained, and screened in an automation-enabled workcell consisting of an incubator, high-content imager, liquid handler, and robotic plate handler. Liquid handling methods were developed in microtiter plates (96 or 384 wells) for seeding of CRC organoids in an extracellular matrix (ECM) (Matrigel) to establish the culture. For maintenance, the liquid handler was programmed to carry out media changes with routine imaging of the plate to monitor organoid growth (pre- and post-treatment). CRC PDOs were treated with selected anti-cancer compounds at various concentrations. For the analysis of organoid growth, a deep learning-based image segmentation model was developed to automate image analysis of the label-free organoids. Using this approach, we tracked the effects of the compounds on colorectal organoid size, morphology, texture, and additional phenotypic readouts. A viability assay was carried out using live/dead cell dyes and the PDOs were imaged in 3D on a high-content confocal imager. Out of the tested panel of known anti-cancer drugs, we found that PDOs treated with romidepsin and trametinib showed the most significant reduction in size. The reduction in organoid size is also accompanied by an increase in dead cell count for the romidepsin-treated tissues, suggesting cytotoxicity. Interestingly, while a decrease in organoid size was observed for trametinib-treated organoids, there was little impact on viability. Overall, our results show the potential for the utility of PDOs from other tissues in both precision medicine and high throughput drug discovery applications using automation with high-content imaging.

Materials and methods

Cell culture

Colorectal cancer organoids (Line ISO38, Cellesce) were handled according to manufacturer’s instructions. Briefly, organoids were thawed quickly at 37°C, gently resuspended, washed in media, and pelleted. Organoids in the pellet were resuspended in Matrigel (Corning) and then seeded in a 384 well plate (50% final concentration of Matrigel), at 200 organoids per well either manually or with the Hamilton STAR liquid handler. For automated seeding, the organoid Matrigel suspension was prepared in three columns of a 96well compound plate. The Hamilton STAR liquid handler was used to seed 10µl of suspension in each well of a 384 well plate using the multichannel probe. Columns 1–12 were seeded with automation, columns 13–24 were seeded manually. Organoids were incubated with media containing ROCK inhibitor for 48 hours to improve recovery. Organoids were then treated with selected compounds for 6 days at varying concentrations (approximately half-log dilutions or 4-fold) and in quadruplicates. Compounds (highest concentration used): 5-Fluorouracil (5FU) (100µM), cisplatin (20µM), doxorubicin (60µM), romidepsin (10µM), trametinib (20µM).

Figure 1. Workflow for using assay-ready colorectal cancer organoids

Image acquisition and analysis

The effects of compound treatment were monitored over time using the ImageXpress® Micro Confocal System. CRC organoids were imaged using 4X Plan Apo objective (NA 0.2) with z-stacks enabled. For viability assay, organoids were incubated with Hoechst, Calcein AM, and ethidium homodimer (Thermo Fisher) for 2hrs at 37°C. Images were acquired on the ImageXpress® Confocal HT.ai High-Content Imaging System using the 10X Plan Apo (NA 0.45) objective with Z-stacking.

IN Carta® Image Analysis Software (version 1.17) was used to analyze images acquired during monitoring. SINAP, a deep learning-based module, was used to create a model for organoid segmentation and analysis was carried out using the 2D projection images. The total organoid area on day 6 post-treatment was normalized over the total organoid area pre-treatment to determine the amount of growth over time. For live/dead analysis, the intensity of Calcein AM dye (FITC channel) was divided by the intensity of ethidium homodimer dye per organoid.

Figure 2. Layout of the automation work cell is illustrated in (A). The instruments are controlled by an integrated software that allows for set up of cell culture workflows. The curve arrows shown an example of the process to monitor cells in culture where plates are moved from the incubator to the ImageXpress Confocal HT.ai for imaging in brightfield and then back to the incubator. B) Assay setup in a 384well plate with four technical replicates for each condition. Compounds are added with the highest concentration in Row B (Border wells, columns 12 and 13 are controls). C) Organoids were seeded with the liquid handler (left half) or manually (right half). Shown here is a heat map representing the number of organoids per well. Organoids seeded with the liquid handler show a more homogenous distribution between wells while manual seeding show a gradient of organoid count from high to low (starting column 13).

Results

Automation of organoid seeding

Compared to 2D cell culture, 3D organoid culture has a more complex workflow that is performed by highly trained individuals. One of the challenges in organoid culture lies in the use of ECM, which requires special handling. ECM, such as Matrigel, is particularly sensitive to temperature – its liquid properties changes with increasing temperature and therefore needs to be handled carefully to prevent premature polymerization. To automate organoid seeding in Matrigel, the liquid properties were optimized for the STAR liquid handler (Figure 2). To ensure minimum fluctuations in temperature, the source plate containing the Matrigel organoid suspension was placed on a cold plate for the duration of the seeding process. The 384W destination plate and pipette tips were also pre-chilled before they were loaded on the deck. The counts of organoids in each well are shown as a heatmap (Figure 2). While we observe some variability in the number of organoids between wells, the manual seeding approach shows an obvious skew in organoid distribution – more organoids were seeded in the initial dispense steps than in the later step. This suggest that the liquid handling protocol can mix and dispense more consistently than with manual pipetting.

To monitor organoids over time, plates containing organoids were incubated in an automated incubator and a monitoring protocol was set up to routinely transport the plates to the ImageXpress Confocal HT.ai. for live imaging (Figure 2A).

Deep learning-based segmentation for robust analysis

Automated image analysis is an integral part of an automation-enabled platform. The ability to monitor cells and organoids in real time and to extract meaningful information is dependent on robust image analysis of label-free transmitted-light images. Due to the meniscus effect and small seeding volume, organoids tend to settle at the corners of the well. This leads to shading artifacts in images acquired in brightfield, which makes it challenging to segment and analyze the organoids. In addition, the occasional presence of bubbles, either in the Matrigel or culture media, distorts the light path such that not all organoids are analyzed.

Conventional image analysis is dependent on the implementation of a series of image processing steps leading up to object segmentation. Each of these steps requires parameter input (such as threshold intensity) in order to detect the object of interest. The highly variable nature of brightfield images makes it challenging to effectively segment objects based on a fixed set of parameters. Users will frequently need to make changes to the image analysis steps before obtaining satisfactory object segmentation. To circumvent these challenges, a deep-learning approach was used to identify the organoids for analysis using IN Carta image analysis software (Figure 3A). A model was trained based on a total of \~40 annotated images in the training set (\~1.4% of the dataset). The resulting model is able to segment organoids independent of size and even those at the well edges and those beneath air bubbles (Figure 3B).

Figure 3. A) Overview of the SINAP workflow in IN Carta software used to generate a model for organoid segmentation. B) Examples of whole-well images of organoids and their respective segmentation masks (cyan). Inset in the figure is shown with adjusted brightness and contrast in order to bring out the organoids that are obscured by artifacts such as bubbles or edge effects. Note that the model can segment these “hard to see” organoids.

Monitoring compound effects on CRC growth

Selected anti-cancer compounds were added to the organoids 48hrs after seeding and their effects on CRC growth over the next 6 days were monitored (Figure 4). Untreated organoids showed an increase in size over time in culture. In contrast, organoids treated with doxorubicin, romidepsin and trametinib showed decreased or arrested growth.

Figure 4. Effects of compounds of organoid growth over time A) Organoids were treated with 5FU, cisplatin, doxorubicin, romidepsin and trametinib at various concentrations and their effect on the CRC organoids were monitored over 6 days. Shown here are examples of CRC organoids treated at the highest compound concentrations. Scale = 200µm. B) Boxplot of organoid area in response to compounds at the indicated concentrations (in µM). Each column shows a compound at three different concentrations (lowest, middle and highest), time (days) is represented on the x-axis. Compounds are added on day two. Compared to the untreated organoids, romidepsin and trametinib treated organoids showed minimal growth. C) To quantify organoid growth over time, the total area of organoids on day 7 was normalized to the total area of pre-treated organoids. Control organoids show an average of \~30% growth. Growth inhibition was significant in doxorubicin, romidepsin, and trematinib treated organoids (based on 2-way ANOVA test)

Assessment of organoid viability

To determine if the compounds affected cell viability, the organoids were stained with Calcein AM (live cell marker), Ethidium homodimer (dead cell marker) and Hoechst (nuclei dye). Organoids were imaged and analysis was carried out to quantify organoid viability (Figure 5).

Romidepsin-treated organoids show a significant increase in cell death at most of the tested concentrations compared to controls. Some cytotoxic effects were seen with doxorubincin-treated organoids at higher compound concentrations. Interestingly, although significant cytostatic effects were seen with trametinib-treated organoids, the increase in cell death was only observed in two of the tested concentrations. This suggests that trametinib has mostly cytostatic effects on CRC organoids while romidepsin and doxorubicin have both cytostatic and cytotoxic effects on these organoids. The results here are consistent with other studies that demonstrated the anti-tumor activities of romidepsin, trametinib and doxorubicin. Romidepsin is a selective inhibitor of Class I histone deacetylase (HDAC) with cytotoxicity observed in colorectal cancer cells⁵. Trametinib is an inhibitor of MEK1/MEK2 activation and kinase activity. Tumor xenograft studies using nude mice showed significant growth delay (tumor volume) when dosed with trametinib^3,4. Doxorubicin has shown effectiveness against a wide range of tumors in clinical studies⁶. Overall, we have demonstrated here a proof-of-concept study on the use of CRC organoids for screening. The implementation of automation in this workflow, including the use of a robust deep-learning based image analysis method, will enable scalability and higher throughput in organoid-based screening studies.

Figure 5. A) Viability assay was carried out on CRC organoids six days post treatment. Shown here are example images of organoids from the various treatment groups (treated with the highest concentrations). B) An example image with overlay of the segmentation mask shown as outlines. C) Viability was quantified by calculating the average ratio of Calcein AM to ethidium homodimer intensity per organoid. Bar graph here shows the data normalized to the controls (y-axis) vs concentration (x-axis) grouped by compound. Two-way ANOVA was performed on unnormalized data, Bonferroni post-hoc test to compare treatments to DMSO.

Conclusions

• Our results support the use of PDOs for high-throughput assays such as compound screening. The availability of standardized, assay-ready PDOs provides significant time savings for screens using 3D models
• We show that an AI-based approach can be successfully used to generate robust segmentation for the analysis of label-free biological modes such as 3D organoids. The data obtained from monitoring gives mechanistic insights into compound effects on cancer organoids and supports studies on drug efficacy.

References

1. Hay M, Thomas DW, Craighead JL, Economides C, Rosenthal J. Clinical development success rates for investigational drugs. Nat Biotechnol. 2014;32(1):40–51.
2. Vlachogiannis G, Hedayat S, Vatsiou A, et al. Patient-derived organoids model treatment response of metastatic gastrointestinal cancers. Science. 2018;359(6378):920–926.
3. Gilmartin AG, Bleam MR, Groy A, et al. GSK1120212 (JTP-74057) is an inhibitor of MEK activity and activation with favorable pharmacokinetic properties for sustained in vivo pathway inhibition [published correction appears in Clin Cancer Res. 2012 Apr 15;18(8):2413. Clin Cancer Res. 2011;17(5):989–1000.
4. Ueda H, Manda T, Matsumoto S, et al. FR901228, a novel antitumor bicyclic depsipeptide produced by Chromobacterium violaceum No. 968. III. Antitumor activities on experimental tumors in mice. J Antibiot (Tokyo). 1994;47(3):315–323.
5. Saijo K, Katoh T, Shimodaira H, Oda A, Takahashi O, Ishioka C. Romidepsin (FK228) and its analogs directly inhibit phosphatidylinositol 3-kinase activity and potently induce apoptosis as histone deacetylase/phosphatidylinositol 3-kinase dual inhibitors. Cancer Sci. 2012;103(11):1994–2001.
6. Minotti G, Menna P, Salvatorelli E, Cairo G, Gianni L. Anthracyclines: molecular advances and pharmacologic developments in antitumor activity and cardiotoxicity. Pharmacol Rev. 2004;56(2):185–229.

Download PDF