A robust and simple patient-derived multicellular spheroid model enables liver cancer precision drug screening (111 characters)

intra-heterogeneity therapeutic targets of bevacizumab and of sorafenib. Tumorspheres from two HCC patients were treated for 6 days. Cell viability was measured. Results are expressed as percentage of viable cells to control and shown as mean ± s.d. A minimum of triplicates were performed for each condition.


Abstract. (164 words)
Background & Purpose. Hepatocellular carcinoma (HCC) is the second leading and fastest rising cause of cancer death worldwide. HCC is often diagnosed when it is not curable. Thus, there are many challenges associated with the treatment of HCC. Until now there is no simple and robust system for reliable pre-treatment screening of approved drugs.
Methods. We developed a simple and robust patient-derived multicellular cell clustering model from 18 liver resections. We applied this technology on 9 HCC liver tissues and assessed FDA-approved drug responses.
Results. The success rate of our model is 100% independently from the type of liver underlying disease. The model shows differential responses to HCC targeted-and immune-based treatments with a high reproducibility. Our work highlights that preserving the integrity of the primary liver architecture permits to perpetuate the physiopathology.
Conclusion. This model system will help to understand the biological role of the tumor microenvironment and to accelerate the clinical research program for personalized liver cancer treatment based on functional screening. Graphical abstract. 4

Introduction.
Liver cancer is the fifth most common cancer and the second most frequent cause of cancer-related death worldwide (1). Current therapeutic options are limited by a response rate of less than 30% and often exhibit severe adverse effects that lead to the discontinuation of the treatment (2). The absence of accurate predictive models makes impossible the right decision-making process in which a patient is enrolled in a treatment protocol (3). Several strategies were proposed to explore ways to fast-track drug screening. However, extended culture of cell lines can lead to secondary genomic changes that modify their response/behavior to anticancer drugs (4). Organoids must be grown by stem cells through a complex induction process that hampers the success rate of organoid cultures (5). Their growth relies on rigid extracellular matrices that create biochemical forces on cells reducing drug penetration (6). Organoids are epithelial cultures that lack tumor stroma, thus they do not allow drug testing that target host-tumor interactions (7). The establishment of patient-derived xenograft (PDX) models is time consuming and not all tumor tissues are suitable (8). Moreover, PDX mice lacked the immune system and a murine-specific tumor evolution may occur (9).
We established a heterogeneous and functional patient-derived model for drug screening in individual patients. We show that the model is successful for studying drug treatment responses including cancer cell targeted-and immune-based therapies. We report a high intra-individual reproducibility of the drug response.

Materials and methods.
Human subjects and liver tissue collection. 5 Resected tissues are from the Centre Hospitalier Universitaire (CHU) of Strasbourg.
They were stored in HypoThermosol FRS Preservation Solution (Sigma-Aldrich) and kept at +4°C prior processing or cryopreserved at -80°C in CryoStor® cell cryopreservation media (Sigma-Aldrich). The protocols were approved by the Ethics Committee of the Strasbourg University Hospitals (DC-2016-2616). Written and informed consent was obtained from all patients.

Generation of patient-derived spheroids.
Resected liver tissues (5x5mm to 8x8mm size) were chopped in small fragments and digested with a 0.05% collagenase-EGTA solution followed by a gentle mechanical disruption and size filtering with a 70µM strainer, to generate aggregates of cells.
Aggregates are assembled into spheroids Corning® 96-well Black/Clear Bottom Low For cryopreserved liver tissues, resected liver tissues were thawed in a +37°C water bath prior processing as described above.

Cell viability assay.
Cell viability was performed using CellTiter3D Glo (Promega). Results are expressed as RLUs mean ± s.d. or as percentage to control mean ± s.d. A minimum of three replicates were performed for each culture or treatment condition.

Albumin and CYP3A4 activity.
Albumin release was measured by ELISA (Bethyl Laboratories) and CYP3A4 activity was determined using CYP3A4 Activity Assay Kit (Abcam).
Triglycerides content was analyzed using Triglyceride Assay Kit (Abcam). Total collagen was quantified using Total Collagen Assay Kit (Abcam).

Actin cytoskeleton and focal adhesion staining.
Patient-derived spheroids were transferred onto glass chamber slide. Actin cytoskeleton and FAK were visualized using the Actin Cytoskeleton & Focal Adhesion Staining Kit (Sigma-Aldrich). The kit consists of TRITC-conjugated phalloidin, anti-Vinculin, and DAPI for the immunofluorescence staining of actin filaments in the cytoskeleton as well as the nuclei.
Immunofluorescence. 7 Patient-derived spheroids were transferred onto glass chamber slide. The procedure for immunofluorescence used is a standard protocol for cell cultures described elsewhere. Briefly, cells were fixed with formaldehyde, permeabilized with Triton-100, blocked with BSA, and incubated for overnight with E-cadherin, CD14, or αSMA specific antibodies (Cell Signaling). Cells were visualized by confocal microscopy.

Results.
A simple and robust protocol to establish a patient-derived multicellular spheroids. A workflow was setup to generate patient-derived spheroids using diseased liver tissues (Fig. 1a). All liver tissues obtained from patients enrolled in this study and their clinical characteristics are reported in Table S1. We generate aggregates of primary cells in which the native architecture of cell complexes is retained (Fig. 1a). We observed dense and well-defined shape spheroids 48h after seeding (Fig. 1b). Immunofluorescence staining revealed positive cells for the epithelial marker, E-cadherin, for CD14, and for αSMA suggesting that patient-derived spheroids retain hepatocytes, Kupffer cells, and hepatic stellate cells, respectively (Fig. 1c, upper and middle pictures). Analysis of the mechanosignaling components showed the presence of F-actin and of the focal adhesion kinase (FAK), key proteins to conduct mechanosignals from the EM into specific intracellular signaling responses modulating cell behavior and function (10,11) (Fig. 1c, lower picture). Cell viability is high and remains constant up to 8 days (Fig. 1d and 1e). Liver-specific functions, albumin 8 release and CYP activity, are maintained (Fig. 1f). We detected the release of MIP1a/CCl3 (Fig. 1g), the production of hepatic triglycerides (Fig. 1h), and the deposition of total collagen (Fig. 1i) suggesting functional macrophages (12), endothelial cells (13), and hepatic stellate cells (14,15). The establishment success rate of our models is 100% independently from the type of liver disease.

HCC tumorspheres show heterogeneous responses to FDA-approved drugs.
We assessed the response of HCC tumorspheres to a panel of targeted-and immunebased compounds (Fig. 2a and 2b). Tumorspheres generated from nine HCC patients, showed a heterogeneous response to FDA-approved molecules suggesting inter-and intra-individual tumor characteristics. Drugs (lenvatinib, sorafenib, and regorafenib) that target common signaling pathway (VEGFR) induced similar reduction of cancer cells viability (346, S19, S22, 422). Cabozantinib, a Met tyrosine kinase inhibitor, decreased tumor cells viability in six patients out of nine (S19, S22, 381, 394, 404, 422). Tumorspheres from three patients (S22, S18, 422) were sensitive to immunecheck point inhibitors, nivolumab (PD1) and atezolizumab (PDL1). One patient out of nine was resistant to all therapies (S12). To reinforce our model value, we compared the efficacy of bevacizumab, a recombinant humanized monoclonal antibody that binds to soluble VEGFA, preventing receptor binding and inhibiting cell function (16), to sorafenib, a multikinase inhibitor, that targets VEGFR (17). As expected, compounds that target common signaling pathway, although with a different mode of action, mediate similar response in HCC tumorspheres (422 is sensitive while 404 is resistant to both bevacizumab and sorafenib) (Fig. 2b)  High reproducibility of the model enables robust functional analyses following cryopreservation. We investigated the reproducibility of the model by performing a drug efficacy test on tumorspheres generated from fresh and cryopreserved liver tissues, both from the same patient (Fig. 3). Tumorspheres generated from HCC fresh liver tissue (346) that are sensitive to sorafenib, remained sensitive after a cryopreservation period. As control of absence of response of normal primary liver cells to sorafenib (18), we generated spheroids from fresh or from cryopreserved healthy liver tissue (349). We found that those spheroids were insensitive to sorafenib in both conditions. We confirmed that the model is reproducible even from cryopreserved HCC tissues showing similar response/unresponse to sorafenib in two independent experiments with a time interval of 4 months (S12, S19).

Discussion.
We describe a simple model that retains the intrinsic tumor microenvironment allowing the analysis of targeted-and immune-based therapies on the same patient. Our models are differentially sensitive to drugs targeting distinct signaling components, suggesting that they are valuable tools for the discovery of drug mode of action and drug repositioning. The model offers a unique opportunity to assess easily combination therapies helping to identify synergistic sorafenib/bortezomib (19) and additive tivantinib/nivolumab or capmatinib/atezolizumab (20) molecules to accelerate HCC treatment options.
We have successfully generated spheroids from needle liver biopsies (data not shown). However, in this study, we only report data obtained from resected HCC liver tissues as these samples allow to produce a sufficient number of patient-derived spheroids to perform a simultaneous screen of targeted-and immune-based molecules on the same patient for a proof-of-concept study. We're aware that this approach does not allow a retrospective correlation analysis with clinical response because resected patients do not undergo systemic therapies. However, our data emphasize the potential prospective use of our model, that could be prepared from needle liver biopsies, to predict clinical responses enhancing success in finding adequate treatment options. Moreover, strategies to reconstruct the patient's own tumor as tumor organoids, have revealed 88% positive predictive and 100% negative predictive individual response to chemotherapy for patients with colorectal and gastroesophageal cancer (21), and 85% of tumor organoids that incorporate autologous immune cells, showed a treatment response that correlates to the clinical outcome (22). Thus the use of our model to predict clinical treatment response is, at a minimum, as good as using patient-derived organoids, encouraging the development of clinical research programs to accelerate personalized medicine.
HCC retains inter-and intra-tumor heterogeneity, and thus poses tremendous challenges for systemic treatments. In clinical practice, needle liver biopsies are used to clarify diagnosis, to predict prognosis, and to make treatment decisions. This sampling strategy may not fully reflect the intra-tumor heterogeneity and thus may not truly reproduce the clinical outcome. Nevertheless, it has been reported that organoids generated from multiple needle liver biopsies from the same location in a tumor, are mutually representative and are similar to originating tumors (5) suggesting that 11 patient-derived spheroids generated from needle liver biopsy are representative of the intra-tumor heterogeneity and thus are relevant to predict treatment responses.
We demonstrate that preserving the integrity of the primary tumor architecture enhances the success rate for the generation of HCC tumorspheres. This improvement may occur via the release of soluble factors or through native cell-to-cell contacts that regulate cell vitality (23). However, this action is often masked by the biological function Treating hepatic malignancies requires new anticancer molecules. However, clinical trials are costly, time consuming, and most compounds do not reach final criteria (25) because they were incorrectly selected in imperfect models that do not fully recapitulate the physiopathology. Our model offers a substantial add-on basis platform for a better preselection of drug candidates to reduce drastically the cost of drug development enhancing the number of molecules that will reach the end-goal.
In summary, we describe a simple and robust patient-derived model for HCC that retains the tumor heterogeneity and that is fully suitable for implementation into clinical research programs to accelerate personalized medicine. Moreover, the above model will be a valuable tool for the scientific community studying deeper the cellular network within the tumor microenvironment.