Carboplatin response in preclinical models for ovarian cancer: comparison of 2D monolayers, spheroids, ex vivo tumors and in vivo models

Epithelial ovarian cancer (EOC) is the most lethal gynecological cancer. Among the key challenges in developing effective therapeutics is the poor translation of preclinical models used in the drug discovery pipeline. This leaves drug attrition rates and costs at an unacceptably high level. Previous work has highlighted the discrepancies in therapeutic response between current in vitro and in vivo models. To address this, we conducted a comparison study to differentiate the carboplatin chemotherapy response across four different model systems including 2D monolayers, 3D spheroids, 3D ex vivo tumors and mouse xenograft models. We used six previously characterized EOC cell lines of varying chemosensitivity and performed viability assays for each model. In vivo results from the mouse model correlated with 2D response in 3/6 cell lines while they correlated with 3D spheroids and the ex vivo model in 4/6 and 5/5 cell lines, respectively. Our results emphasize the variability in therapeutic response across models and demonstrate that the carboplatin response in EOC cell lines cultured in a 3D ex vivo model correlates best with the in vivo response. These results highlight a more feasible, reliable, and cost-effective preclinical model with the highest translational potential for drug screening and prediction studies in EOC.

Xenograft mouse model. All animal procedures were performed in accordance with the guidelines for the Care and Use of Laboratory Animals of the CRCHUM as well as the recommendations in the ARRIVE guidelines. This study was approved by the Comité institutionnel de protection des animaux (Animal Ethics Committee, protocol number C18028AMMs). NOD.Cg-Rag1 tm1Mom Il2rg tm1Wjl /SzJ immunodeficient female mice (007,799, The Jackson Laboratory-JAX, Bar-harbor, Maine, USA) 20,21 were used to establish xenograft tumors with cell lines. A 200 µL suspension of 1 × 10 6 cells in 100 µL cold Dulbecco's PBS (311-425-CL, Wisent) with 100 µL of Matrigel® Matrix (CACB356237, Corning Inc., NY, USA) was injected subcutaneously in the flank of each mouse for the TOV112D, TOV21G and OV90 cells, while 5 × 10 6 cells were injected for the OV1946, OV4453 and OV4485 cells. Eight mice were used for the control (vehicle) group and for each of the three carboplatin treatment groups per cell line (see section Carboplatin treatment). Treatment was initiated once tumor size was 200 mm 3 as drug effects can vary if below this value 22 . Mice were between the ages of 11-24 weeks at the start of treatment and given dietary supplementation, DietGel® Recovery and DietGel® Boost (Clear H 2 O, Portland, USA), twice weekly. Tumors were measured with calipers 2-3 times weekly. To alleviate the known negative side effects of carboplatin treatment, anti-nausea medications (1 mg/kg of maropitant and 0.8 mg/kg of ondansetron) were given one hour before the chemotherapy dose and at 24-and 48-h following treatment. Mice were sacrificed at the end of treatment period or if ethical limits were attained through an intraperitoneal injection of euthanyl (pentobarbital sodium) at a dose of 400 mg/kg (concentration of 240 mg/ml). Tumors were collected, measured and were formalin-fixed and paraffin-embedded (FFPE). FFPE tumor blocks were cut into 4 µm sections for histological hematoxylin & eosin (H&E) staining. 3D spheroid formation. Rapid, compact and uniform homogenous formation of EOC spheroids was achieved by using 96-well concave-bottom, ultra-low attachment (ULA) microplates (4515/4520, Corning) 8,11,13,16,23 . For all cell lines, 2,000-2,500 cells in 100 µL of complete OSE medium were seeded in each well. Plates were centrifuged at 1,000 rpm for 5 min at room temperature. Spheroids were allowed to form over 48 h in their respective incubation conditions (see Cell lines), generating spheroids of approximately 500 µm in diameter. Spheroids were treated with three carboplatin concentrations (based on optimized IC 50 dose ranges; see Carboplatin treatment). For each cell line, 20 spheroids were seeded for each carboplatin concentration as well as control groups. Two replicates per condition, containing 10 spheroids for each replicate, were analyzed for each cell line by flow cytometry. In parallel, 10 untreated spheroids were transferred into microfluidic devices at 48 and 96 h for fixation.
Micro-dissected tissue (MDT) production from cell line xenograft tumors. The micro-dissection procedure was adapted from previously published work 14,24  Fixation of MDTs and spheroids within microfluidic devices. MDTs were fixed with 10% formalin (F6050, Produits Chimiques A.C.P. Chemicals Inc, Saint-Leonard, Qc, Canada) after carboplatin treatment and recovery periods, including respective controls. Untreated spheroids were similarly fixed after 48 and 96 h of formation. All specimens were further processed through the previously published paraffin-embedding lithography procedure to create micro-dissected tissue microarray (MDTMA) blocks 14 , which were cut into 4 µm sections for histological H&E staining. Specimen size shrink after this processing technique as previously reported 14 .   [25][26][27] . A 24-h recovery was chosen based on published in vitro studies demonstrating the effect of chemotherapy only after its removal 27,28 and to mimic the physiologic metabolism of the drug. The optimal concentration range was determined for each cell line for minimal growth inhibition and for an effect well below the 50% threshold.
3D ex vivo tumor model. MDTs obtained from untreated xenograft tumors of our cell lines were treated with 6-7 different carboplatin concentrations based on the IC 50 values from monolayers and 3D spheroids. Two carboplatin regimens were tested for MDTs: a 10-h treatment induction followed by a 14-h recovery, and a 16-h induction followed by a 24-h recovery.
Clonogenic survival assay. The IC 50 values for carboplatin for OV4453, OV4485, TOV112D and OV90 were previously determined by clonogenic survival assay 13,16,19 . Carboplatin sensitivity for the OV1946 and TOV21G cell lines was determined in this study using the same clonogenic assay 16 . Briefly, cells were seeded in a 6-well plate at a volume of 1 mL/well and at a density that allowed the formation of individual colonies (1,000 or 1,500 cells/well for TOV21G or OV1946, respectively). Cells were allowed to adhere for 16 h in 5% CO 2 at 37 °C. Then an additional 1 mL of medium containing carboplatin (final concentrations 0-100 μM) was added in each well and cells were incubated for 24 h. After this period, medium was completely removed and replaced with fresh OSE complete medium. When colonies became visible at 2X magnification, plates were fixed with cold methanol and stained with a solution of 0.5% blue methylene (Sigma-Aldrich Inc., St. Louis, MO) in 50% methanol. Colonies were counted under a stereomicroscope and reported as percent of control. IC 50 values were determined using Graph Pad Prism 6 (GraphPad Software Inc., San Diego, CA). Each individual experiment was performed in duplicate and repeated three times. Normalized live and dead cell rates were plotted using GraphPad Prism 6 (GraphPad Software Inc.) to generate dose-response inhibition curves with respective IC 50 values. Each experimental analysis was performed in duplicate and repeated three times.  29 . Xenografts were generated from EOC cell lines and treated following the protocol depicted in Fig. 2A. Tumor volumes were recorded throughout carboplatin treatment (Fig. 2B). Chemosensitivity of each cell line was based on inhibition of in vivo tumor growth. OV1946 and OV4453 were categorized as sensitive, demonstrating tumor volumes that were significantly lower than the controls at time of sacrifice for all three carboplatin doses (highly responsive). OV90 and OV4485 showed intermediate responses with a significant decrease in tumor volumes at the two highest doses but no response to the lower dose (partially responsive). TOV21G and TOV112D were resistant as they showed no statistical difference at even the highest dose (unresponsive). IF with Ki-67 was quantified from collected xenografts after carboplatin treatment (Fig. 2C) and showed that the response was dose-and cell-line dependent. Results were largely concordant with the tumor volume measurements, confirming chemosensitivity classification. However, treatment response in the xenograft model varied significantly from the 2D culture ranking (Supplementary Table S1); a positive correlation of 2D sensitivity with the in vivo response was found in 3/6 EOC cell lines (Table 1).

Carboplatin response of 3D EOC spheroids improves the correlation with the in vivo response compared to 2D cultures. All six EOC cell lines formed 3D spheroids in ULA plates. OV90 and OV1946
formed compact spheroids (Fig. 3A), whereas TOV112D, TOV21G, OV4485 and OV4453 formed dense aggregates (Fig. 3B). To demonstrate that cells in the spheroids remained proliferative throughout the experiment, we performed IHC staining to evaluate the level of apoptotic (CC3) and proliferative cells (Ki-67) in spheroids at 48 h (time of spheroid formation) and at 96 h (end of experiment) in the untreated controls ( Supplementary  Fig. S1). Cells in the spheroids stained strongly for Ki-67 at both time-points with low expression of CC3, demonstrating that they remained proliferative throughout the treatment course.
Flow cytometry was used to evaluate the proportion of viable cells in spheroids after carboplatin treatment (Fig. 3C). IC 50 values were generated using dose-response inhibition analyses (Fig. 3D). In all cell lines, the 3D spheroid IC 50 values were significantly higher than that seen in 2D models (Supplementary Table S1). However, the fold change in carboplatin sensitivity between 3 and 2D models varied significantly, depending on the cell line (Supplementary Table S2). The change from 2 to 3D models increased the IC 50 value by 280-fold for TOV21G, but only sevenfold for OV90, highlighting cell line-dependent changes. Cut-off for resistance to carboplatin treatment was based on response to the physiologic conversion of carboplatin bioavailability (269.4 µM, rounded to 250 µM) in patients (carboplatin dosing of an AUC of 5 corresponding to an average concentration of 300 mg/ m 2 , body surface area 1.6 m 2 , blood volume 4.8 L) 30 . Therefore, cell lines with IC 50 values higher than 250 µM were considered resistant. On the other hand, response to doses below 100 µM were considered sensitive based on previous reports of carboplatin treatment of 3D ovarian cancer models 31,32 . Response between the two cut-offs were considered intermediate. Using these criteria, OV1946 was categorized as sensitive, OV90 as intermediate, and TOV21G, TOV112D, OV4453 and OV4485 as resistant. These results show a positive correlation with the in vivo response for 4/6 EOC cell lines ( Table 1).

The 3D ex vivo tumor model demonstrates a reliable correlation with the in vivo carboplatin response.
Previous studies have shown that MDTs can assess the response to chemotherapeutic drugs in cell line xenograft models 14,24 . Here, we sought to compare the carboplatin sensitivity profiles of cell line xenograft tumor-derived MDTs (Supplementary Fig. S2A) to our ovarian cancer model systems including 3D spheroids and in vivo xenografts using IF analysis (Fig. 4A)     www.nature.com/scientificreports/ (16)(17)(18)(19)(20)(21)(22)(23)(24) (Fig. 4B-C, Supplementary Figure S2B). Both treatment regimens gave similar cell fate responses (proliferation, apoptosis) as well as similar IC 50 (Supplementary Fig. S2C and S2D, and Fig. 4C). Based on these results, we used the 16-24 treatment regimen to perform the remaining experiments. To compare the chemosensitivity of cell line-based MDTs to 3D spheroids and the in vivo model, we quantified the proliferation capacity of cells after carboplatin treatments (Fig. 4B) and determined IC 50 values (Fig. 4C). According to our criteria for 3D spheroids, OV1946 and OV4453 were categorized as sensitive (IC 50 < 100 µM), TOV21G and TOV112D were resistant (IC 50 > 250 µM), and OV4485 was intermediate. These findings were in complete agreement with the in vivo chemosensitivity results (5/5 EOC cell lines) and provided the best correlation compared to the other two in vitro models (Table 1).

Discussion
This study highlights the importance of preclinical model selection for drug sensitivity analysis and understanding the variation that exists between experimental models. As most early-phase clinical trial designs rely heavily on preclinical data, it is important to consider these variations when performing drug screening or therapeutic response prediction studies, especially in the era of personalized medicine.
The mainstay of preclinical studies remains cell line-based and the in vivo response from animal models is often used as the gold standard in preclinical testing of novel therapies/combinations. To our knowledge, only one study 33 using bladder carcinoma cells reported that the 3D spheroids model reflected better the chemoresponse found in their mouse xenograft model, and higher drug resistance was seen with the 3D model compared to 2D cultures. In our study, we compared four translational model systems, including 2D monolayers, 3D spheroids, ex vivo MDTs and in vivo xenografts. Our data suggest better concordance in carboplatin sensitivity between our 3D ex vivo model (MDTs) and in vivo responses. Interestingly, we observed some notable differences between 2D culture and in vivo carboplatin responses. In the case of TOV21G, both its clear cell histology and microsatellite instability 17,34,35 supports the in vivo response of a platinum-resistant cell line. However, 2D culture experiments have consistently shown this cell line as carboplatin sensitive (Supplementary Table S1 and 29,36,37 ). For OV90 and OV1946, an increased sensitivity to carboplatin is seen in mice. This may be due to their histological highgrade serous subtype of which the majority of patients respond to first-line platinum treatment 3 . Indeed, none of our high-grade serous cell lines showed in vivo resistance to carboplatin, but showed sensitivity in the 2D or 3D spheroid models. Given that each model has unique features, their relative response to cytotoxic therapy may vary. Immortalized EOC monolayer cultures offer little cell-cell interaction and consist uniquely of a sub-clonal population of epithelial cancer cells. While spheroids also consist of mostly epithelial cancer cells, they offer a 3D structure with inherent cell layers, cell-cell interactions and chemical/nutrient gradients 25,38 . Increasing in model complexity is our ex vivo tumor model of MDTs that not only offers a 3D structure, but also includes mouse-infiltrating stromal cells which may impact the tumor response to a therapeutic agent. In vivo models further increase model complexity by incorporating important elements such as drug metabolism, influence of endogenous hormones and mammalian physiology 39 . In general, our study suggests that the relative carboplatin response of our 3D models was in line with in vivo results. However, two cell lines, OV4453 and OV4485, did not have concordant results as 3D spheroids and demonstrated higher carboplatin resistance in our spheroid model. We suspect that this may be related to their low oxygen culture conditions (7%), which was specific for only these two cell lines. Hirst et al. 40 showed that an increase in hypoxia-regulated genes and markers of stemness were present in the core of 3D spheroids but not in monolayered cells and that this induced chemoresistance and phenotypic changes. In addition, these cell lines formed spheroid aggregates that were not compact and had larger spheroid diameters, which has been shown to influence drug resistance 38,41 .
Importantly, our 3D ex vivo model provided a complete concordant correlation with in vivo responses. Ex vivo models are attractive for fundamental and translational research as they can predict patient response to drugs in a clinically relevant timeframe. Important advantages of this model include minimal waste of tissue and culture/ drug reagents 24 , control of fluids and constant supply of nutrients 42 , long-term viability 14,43,44 and the maintenance of MDTs and their TME 14,24 without need of growth supplements 45,46 . This model further allows testing multiple cycles of cytotoxic therapies as well as studying the effects of cytostatic drugs that require longer incubation periods. This underscores the need to incorporate these models into the drug development pipeline to better evaluate the potential efficacy of new drugs or combinations prior to entering in expensive clinical trial settings.
Although this study has some limitations, such as use of a single drug, choice of flow cytometry for 3D spheroid analyses and limited utility of ex vivo tumors with low epithelial count, the reproducible comparison between model systems while using the same cell lines clearly shows the relevance of using various preclinical models to better characterize response to novel therapies. We are aware that cell lines are often devoid of many elements of the natural TME such as stromal and immune cells, which have been shown to influence response, and may not fully represent the primary tumor heterogeneity. However, the use of cell lines currently remains common practice in most preclinical models. Furthermore, a mouse model may not entirely reflect the human drug response. Hence, ex vivo models derived directly from patient samples would eliminate this bias in the drug development pipeline. Alternatively, similar analyses in comparing these preclinical models could be applied to other drugs in ovarian cancer treatment, such as paclitaxel, poly (ADP-ribose) polymerase (PARP) inhibitors and other drugs currently in preclinical studies such as HIF, WEE1 and TGFß inhibitors. Unfortunately, 3D analysis methods have traditionally relied on 2D culture methods or confocal microscopy analyses, which have their limitations. Thus, it would be interesting to include novel techniques such as light sheet microscopy and tissue clearing as well as IF staining of MDTMAs to allow optimal analysis of tissue without disturbing its natural environment.
Platinum resistance remains an important obstacle in EOC with dismal survival and limited options at advanced stages of disease progression. With the overall high attrition rate of oncologic treatments, more Scientific Reports | (2021) 11:18183 | https://doi.org/10.1038/s41598-021-97434-w www.nature.com/scientificreports/ cost-effective predictive cancer models that accurately reflect patient response are needed. With this study, we clearly demonstrate a heterogeneity in therapeutic responses of EOC cell lines when cultured in different systems, which underscores the need to consider multiple factors when selecting a preclinical model for drug discovery and screening studies. This may avoid rejecting potentially effective drugs while eliminating ineffective drugs at the preclinical stage. This could also help reduce the rate of failed clinical trials in which patients experience drug toxicities with minimal efficacy, particularly for rare cancers 23 , which are more difficult to accrue for clinical trials. In the era of personalized medicine, future applications would be to optimize treatment selection based on the individual tumor and patient characteristics rather than a 'one treatment fits all' approach. Thus, validation and feasibility studies of newer and more complex models are needed to enhance the current standards.

Data availability
The data generated or analyzed during this study are available from the corresponding author upon request.