A simple high-throughput approach to identify actionable drug responses in patient-derived tumor organoids

There is increasing interest in developing 3D tumor organoid models for drug development and personalized medicine applications. While tumor organoids are in principle amenable to high-throughput drug screenings, progress has been hampered by technical constraints and extensive manipulations required by current methodologies. Here, we introduce a miniaturized, fully automatable, flexible high-throughput method using a simplified geometry to rapidly establish 3D organoids from cell lines and primary tissue and robustly assay drug responses. Introduction Cancer therapy is rapidly progressing toward individualized regimens not based on the organ of origin, but rather on the molecular characteristics of tumors. Next generation sequencing (NGS) is typically regarded as the key to access this potentially actionable molecular information1,2. However, recent studies showed how only a small number of cancers can be singled out and targeted with this approach, in part because very few gene alteration-drug pairs are unequivocally established3–7. Thus, functional precision therapy approaches where the primary tumor tissue is directly exposed to drugs to determine which may be efficacious have the potential to boost personalized medicine efforts and influence clinical decisions4,8. Establishing patient-derived xenografts (PDX) is a costly and time consuming option that only allows to screen very few potential drugs. Conversely, ex vivo 3D tumor spheroids or organoids derived from primary cancers can be easily established and potentially scaled to screen hundreds to thousands of different conditions. 3D cancer models have been consistently shown to faithfully recapitulate features of the tumor of origin in terms of cell differentiation, heterogeneity, histoarchitecture and clinical drug response4,9–16. Various methods to set up tumor spheroids or organoids have been proposed, including using low-attachment U-bottom plates, feeding layers or various biological and artificial matrices10,13,14,17–23. Methods using low-attachment U-bottom plates ideally only carry one organoid per well, have limited automation and final assay capabilities19–21. In addition, not all cells are capable of forming organized 3D structures with this method. Approaches that include a bio-matrix, such as Matrigel, have the potential to offer a scalable alternative in which cancer cells thrive10,15,24,25. However, most approaches so far rely on thick volumes of matrix which is not cost-effective, potentially hard for drugs to efficiently penetrate and difficult to dissolve fully at the end of the experiment24. In other applications, organoids are first formed and then transferred to different plates for drug treatment or final readout which can result in the tumor spheres sticking to plastic or breaking15,25. In addition, some assays require to disrupt the organoids to single cell suspensions at the end of the experiment17,23. All of these manipulations introduce significant variability limiting applicability in screening efforts13. To overcome these limitations, we introduce a facile assay system to screen 3D tumor organoids that takes advantage of a specific geometry. Our miniaturized ring methodology does not require functionalized plates. Organoids are assayed in the same plate where they are seeded, with no need for sample transfer at any stage or dissociation of the pre-formed tumor organoids to single cell suspensions. Here, we show that the mini-ring approach is simple, robust, requires few cells and can be easily automated for high-throughput applications. Using this method, we were able to rapidly identify clinically actionable drug sensitivities for several primary ovarian cancer and high-grade serous tumors. A simple high-throughput approach to identify actionable drug responses in patient-derived tumor organoids Nhan Phan1,‡, Jenny J. Hong1, Bobby Tofig2, Matthew Mapua1, Jin Huang3, Sanaz Memarzadeh3-7, Robert Damoiseaux3,8 and Alice Soragni1,7* 1Division of Hematology-Oncology, Department of Medicine, David Geffen School of Medicine, University of California Los Angeles, CA, 90095 2Molecular Screening Shared Resource, California NanoSystems Institute, University of California Los Angeles, CA 90095 3Department of Obstetrics and Gynecology, David Geffen School of Medicine, University of California Los Angeles, CA, 90095 4Eli and Edythe Broad Center of Regenerative Medicine and Stem Cell Research, University of California Los Angeles, CA 90095 5The VA Greater Los Angeles Health Care System, Los Angeles, CA, 90073 6Department of Biological Chemistry, University of California Los Angeles, CA 90095 7Molecular Biology Institute, University of California Los Angeles, CA 90095 8Department of Molecular and Medicinal Pharmacology, David Geffen School of Medicine, University of California Los Angeles, CA, 90095 ‡Present address: Laboratory of Stem Cell Research and Application, University of Science, Vietnam National University, HCM City, Vietnam *Correspondence to AS (alices@mednet.ucla.edu)


Introduction
Cancer therapy is rapidly progressing toward individualized regimens not based on the organ of origin, but rather on the molecular characteristics of tumors. Next generation sequencing (NGS) is typically regarded as the key to access this potentially actionable molecular information 1,2 . However, recent studies showed how only a small number of cancers can be singled out and targeted with this approach, in part because very few gene alteration-drug pairs are unequivocally established 3-7 . Thus, functional precision therapy approaches where the primary tumor tissue is directly exposed to drugs to determine which may be efficacious have the potential to boost personalized medicine efforts and influence clinical decisions 4,8 . Establishing patient-derived xenografts (PDX) is a costly and time consuming option that only allows to screen very few potential drugs. Conversely, ex vivo 3D tumor spheroids or organoids derived from primary cancers can be easily established and potentially scaled to screen hundreds to thousands of different conditions. 3D cancer models have been consistently shown to faithfully recapitulate features of the tumor of origin in terms of cell differentiation, heterogeneity, histoarchi-tecture and clinical drug response 4,9-16 . Various methods to set up tumor spheroids or organoids have been proposed, including using low-attachment U-bottom plates, feeding layers or various biological and artificial matrices 10, 13,14,[17][18][19][20][21][22][23] . Methods using low-attachment U-bottom plates ideally only carry one organoid per well, have limited automation and final assay capabilities [19][20][21] . In addition, not all cells are capable of forming organized 3D structures with this method. Approaches that include a bio-matrix, such as Matrigel, have the potential to offer a scalable alternative in which cancer cells thrive 10, 15,24,25 . However, most approaches so far rely on thick volumes of matrix which is not cost-effective, potentially hard for drugs to efficiently penetrate and difficult to dissolve fully at the end of the experiment 24 . In other applications, organoids are first formed and then transferred to different plates for drug treatment or final readout which can result in the tumor spheres sticking to plastic or breaking 15,25 . In addition, some assays require to disrupt the organoids to single cell suspensions at the end of the experiment 17,23 . All of these manipulations introduce significant variability limiting applicability in screening efforts 13 . To overcome these limitations, we introduce a facile assay system to screen 3D tumor organoids that takes advantage of a specific geometry. Our miniaturized ring methodology does not require functionalized plates. Organoids are assayed in the same plate where they are seeded, with no need for sample transfer at any stage or dissociation of the pre-formed tumor organoids to single cell suspensions. Here, we show that the mini-ring approach is simple, robust, requires few cells and can be easily automated for high-throughput applications. Using this method, we were able to rapidly identify clinically actionable drug sensitivities for several primary ovarian cancer and high-grade serous tumors.

Mini-ring setup and assays optimization
In order to rapidly screen organoids, we first established a miniaturized system that allows to setup hundreds of wells and perform assays with minimal manipulation. We adapted the geometry used to plate tumor cells in Matrigel to generate mini-rings around the rim of the wells. This is attained by plating single cell suspensions obtained from a cell line or a surgical specimen pre-mixed with cold Matrigel (3:4 ratio) in a ring shape around the rim in 96 well plates (Fig. 1a). The combination of small volume plated (10 µl) and surface tension holds the cells in place until the Matrigel solidifies upon incubation at 37°C and prevents 2D growth at the center of the wells. This configuration allows for media addition and removal so that changes of conditions or treatment addition to be easily performed by pipetting directly in the center of the well, preventing any disruption of the gel. Cancer cell lines grown in mini-ring format give rise to organized tumor organoids that recapitulate features of the original histology (Fig  1b and S1; Table S1). Similarly, we can routinely establish patient-derived tumor organoids (PDTOs) using the same geometry. Primary patient samples grow and maintain the heterogeneity of the original tumor as expected (Fig. 1b). Next, we optimized treatment protocols and readouts for the mini-ring approach. Our standardized paradigm includes: seeding cells on day 0, establishing organoids for 2-3 days followed by two consecutive daily drug treatments, each performed by complete medium change (Fig. 1c). As an example, small scale screenings were performed using three drugs at five different concentrations in triplicates, ReACp5, Staurosporine and Doxorubicin (Fig. 1d-g). We optimized different readouts in order to adapt the method to a specific research question or instrument availability. After seeding cells in standard white plates, we performed a lumines-cence-based ATP assay to obtain a metabolic readout of cell status, calculate EC 50 and identify cell-specific sensitivities (Fig. 1, S2 and S3). Results show how the Matrigel in the mini-ring setup is thin enough to allow penetration not only of small molecules but also of higher molecular weight biologics such as peptides 17 . We performed two consecutive treatments which allows the drugs to not only penetrate the gel but also to reach organoids that may be bulky 17 . However, the assay is flexible and can be easily adapted to single treatments followed by longer incubations, multiple consecutive recurring treatments, multi-drug combinations or other screening strategies (Fig. S3). We also implemented assays to quantify drug response by measuring cell viability after staining of live organoids with specific dyes followed by imaging. A calcein-release assay coupled to propidium iodide (PI) staining as well as a caspase 3/7 cleavage assay can be readily performed after seeding the cells in standard black plates ( Fig. 1e-g and S4). Tumor organoids are stained with the reagents after dispase release. After a 30-45 minute incubation, organoids can be imaged, followed by segmentation and quantification of the pictures ( Fig. 1e-g and S4). All the assays described here are performed by measuring cell status in within the same well in which spheroids are seeded. Although the various assays we introduce are testing different aspects of cell viability and measure distinct biological events, results were mostly concordant across the methods for the three drugs tested (Fig. 1, S4).

Identification of actionable drug responses in PDTOs
A rapid functional assay to determine drug sensitivities of primary specimens can offer actionable information to help tailoring therapy to individual cancer patients 3 . We tested suitability of our approach to rapidly and effectively identify drug susceptibilities of three primary ovarian cancer samples and one high-grade serous peritoneal cancer specimen obtained from the operating room (Table S1; Fig. 2 and 3). In all cases, ascites or tumor samples were processed after surgery (see Methods) and then plated as mini-rings as described above. In order to maximize the amount of information Results of kinase screening experiment. Three readouts were used for this assay: ATP quantification as measured by CellTiter-Glo 3D and organoid number or size quantification evaluated by brightfield imaging. Brightfield images were segmented and quantified using the Celigo S Imaging Cell Cytometer Software. Both organoid number as well as total area were evaluated for their ability to capture response to drugs. In this plot, each vertical line is one drug, all 252 tested are shown. Values are normalized to the respective vehicle controls for extracted from irreplaceable clinical samples, we investigated the possibility to concurrently perform multiple assays on the same plate. To do so, we first optimized the initial seeding cell number (5000 cells/well) to couple an ATP metabolic assay to 3D tumor count and total organoid area measurement. This seeding density yields a low-enough number of organoids to facilitate size distribution analysis but sufficient ATP signal to be within the dynamic range of the CaspaseGlo 3D assay. For each patient sample, we seeded six 96 well plates and tested 252 different kinase inhibitors at two different concentrations (120 nM and 1 µM). We used the same experimental paradigm optimized above. All steps (media change, drug treatment) were automatb c     ed and performed in less than 2 minutes/plate using a Beckman Coulter Biomek FX integrated into a Thermo Spinnaker robotic system. At the end of each experiment, PDTOs are first imaged in brightfield mode for organoid count/size distribution analysis followed by the ATP assay. The measurements yielded high quality data that converged on several hits, highlighting the feasibility of our approach to identify potential leads ( Fig. 2 and 3). Cells obtained from Patient #1 at the time of cytoreductive surgery 26 were chemo-naïve, and the heterogeneous nature of this clear cell/HGSC tumor was recapitulated in the PDTOs (Table 1 and Fig. 1b). The organoids were sensitive to 16/252 molecules tested and responded mostly to a variety of cyclin-dependent kinase (CDK) inhibitors with a stronger response to inhibitors hitting CDK1/2 in combination with CDK 4/6 or CDK 5/9 (Fig. 2b-d and S5a-b). Interestingly, CDK inhibitors have found limited applicability in ovarian cancer therapy so far 27 . Based on the profiles of the CDK inhibitors tested and on the response observed ( Fig. S5a-b), we selected four untested molecules to assay. We anticipated that Patient #1 would not respond to Palbociclib (targeting only CDK4/6) and THZ1 (CDK7) while expecting a response to JNJ-7706621 (CDK1/2/3/4/6) and AZD54338 (CDK1/2/9; Fig. S5a-b). However, we observed a strong response to THZ1 (Fig. 2e). Both THZ1 and BS-181 HCl specifically target CDK7. Nevertheless, Patient #1 PDTOs showed a strong response to the former but no response to the latter which could be attributed to the different activity of the two as recently observed in breast cancer 28 . We also attempted to establish patient-derived xenografts (PDX) in vivo by injecting Patient #1 cells injected in NSG mice (500K/mouse, 12 mice). Only three mice developed PDXs over the course of several months, with xenografts closely resembling the original tumors (Fig. 2f). Organoids established from one of Patient #1 PDXs qualitatively recapitulated the response to CDK inhibitors, confirming that our strategy can be successfully used to test patient samples that are recalcitrant to grow in vivo, reducing time and costs (Fig. 2e). Patient #2 was diagnosed with progressive, platinum-resistant HGSC and was heavily pretreated prior to sample procurement (Table S1). Patient #2 PDTOs showed a strong response to only 3/252 drugs tested ( Fig. 3a and S5c). Moderate responses (50-60% residual cell viability at 1 µM) were observed for EGFR inhibitors and we could detect high expression of EGFR at the plasma membrane of the tumor cells (Fig. 3a,  3d and S5f). Remarkably, Patient #2 PDTOs showed a very moderate response to our positive control, Staurosporine, a pan-kinase inhibitor with very broad activity 29 . The significant lack of response to multiple thera-pies observed for Patient #2 led us to hypothesize that there could be over-expression of efflux membrane proteins. Indeed, the PDTOs showed a high level of expression of ABCB1 (Fig. 3b). High expression of the ATP-dependent detox protein ABCB1 is frequently found in chemoresistant ovarian cancer cells and recurrent ovarian cancer patients' samples and has been correlated with poor prognosis 30,31 . Patient #3 presented with carcinosarcoma of the ovary, an extremely rare and aggressive ovarian tumor which has not been fully characterized at the molecular level yet 32 (Table 1, Fig. 3c, 3d and Fig. S5d) (Fig. 3f). A Phase I basket trial of this PI3K/ mTOR inhibitor highlighted moderate responses in unstratified patients 34 . Overall, patient with or without PI3K alterations have been shown to respond to PI3K inhibitors 34 . Our assay could supersede the lack of predictors of response to PI3K inhibitors, and identify responsive tumors from a functional standpoint.

Conclusions
We devised and optimized a facile high-throughput approach to establish and screen tumor organoids. While we applied the mini-ring setup to drug screenings, the same methodology is suitable for studies aiming at characterizing organoids' biological and functional properties with medium to high throughput. Complete automation, scalability to 384 well plates, and flexibility to use different supports beside Matrigel can further extend the applicability of our mini-ring approach. Our methodology can be a robust tool to standardize functional precision medicine efforts 3 , given its ease of applicability to many different systems and drug screening protocols, as well as its limited cell requirement which allows testing of samples as obtained from biopsies/ surgical specimens without the need for expansion in vitro or in vivo which can lead to substantial divergence from the tumor of origin 35 . As demonstrated above, the method rapidly allowed us to pinpoint individual drug sensitivities and identify a tumor "fingerprint", with multiple inhibitors converging on a given pathway. Interestingly, many of the drugs identified in our screening do not have a specific, unequivocal biomarker or genomic signature predictive of response.  were dissociated to single cells and cryopreserved or plated right after processing. In short, fresh tumor specimens or ascites samples are obtained from consented patients (UCLA IRB 10-000727). Solid tumor specimens are minced, then enzymatically digested with collagenase IV (200 U/ml). The resulting cell suspension is filtered through a 40 μM cell strainer.
Chemicals: Doxorubicin hydrochloride was purchased from Sigma (#44583). Staurosporine was purchased from Cell Signaling Technology (#9953S). ReACp53 was synthesized by GL Biochem and prepared as described in Soragni et al, 2016.
3D organoids seeding/treatment procedure: Single-cell suspensions (2K-10K/well) were plated around the rim of the well of 96 well plates in a 3:4 mixture of PrEGM medium and Matrigel (BD Bioscience CB-40324). White plates (Corning #3610) were used for ATP assays while black ones (Corning #3603) were used for caspase or calcein assays. Plates are incubated at 37°C with 5% CO2 for 15 minutes to solidify the gel before addition of 100 µl of pre-warmed PrEGM to each well using an EpMotion (Eppendorf). Two days after seeding, medium is removed and replaced with fresh PrEGM containing the indicated drugs. The same procedure is repeated daily on two consecutive days. 24h after the last treatments, media is removed and wells are washed with 100 µl of pre-warmed PBS. To prepare for downstream experiments, organoids are then released from Matrigel by 40 minutes of incubation in 50 µl of 5mg/mL dispase (Life Technologies #17105-041). All steps are performed with the EpMotion for small scale experiments and medium is removed/added from the center of the wells. For the high-throughput kinase screening experiment, we utilized a Beckman Coulter Biomek FX system with 96 channel head integrated into a Thermo Spinnaker robotic system with Momentum scheduling software. In short, an intermediary dilution plate (Axygen P-96-450V-C-S) was filled with 100 µl/well of media and pre-warmed to 37°C. Using pre-sterilized p50 tips, 1 µl of drug is transferred from a library compound plate to the intermediary media plate and thoroughly mixed. Next, the robot gently removed 100 µl of media from the matrigel/cell plate. The liquid handler was set up to hit the dead center of each well with no contact to the Matrigel mini-ring. As a last step, the robot transferred 100 µl from the intermediary plate (media+drug) to the matrigel/cell plate. Media was easily dispensed without touching or disrupting the Matrigel mini-ring. The total process time outside of the CO2 incubator was less than 2 minutes allowing the temperature to be controlled throughout.     Kinase inhibitors to which the HGSC control patient-derived line S1 GODL responded to. (b) List of CDK inhibitors that induced cell death in >75%Patient #1's organoids. Targets and specificity of each is listed. The patient responded to CDK inhibitors hitting CDK1/2 in combination with CDK 4/6 or CDK 5/9. (c), (d) and (e) Representative images of post-treatment and post-dispase PDTOs (e) Expression of EGFR in S1 GODL, Patient #1 and Patient #2 3D tumors. Magnification: 40x.