3D-Hepatocellular Carcinoma (3D-HCC) Models of Diffused and Aggregated Spheroids using a 96-Pillar/Well Plate

Background: Hepatocellular carcinoma (HCC) is the sixth most common cancer and the third most frequent cause of cancer-related mortality worldwide. The genetic and physiological complexity of HCC is a major barrier to the study of tumorigenesis and identication of therapeutic targets. The three-dimensional (3D) culture of cancer cells within an extracellular matrix (ECM) provides an in vitro tumor model that best recapitulates in vivo tumor pathophysiology. Methods: In the current study, we cultured cells and made spheroids by aggregating cells or diffusing cells in Matrigel attached on the tip of the 96-pillar plate. According to the initial cell position in Matrigel, two 3D-HCC cancer in-vitro models (diffused spheroid model and aggregated spheroid model) were established. These two models were applied to drug sensitivity assays to identify drug sensitivity changes. The protein expression and cytokine activation related to the drug resistance and maintenance of the physiological properties of HCC cells were analyzed in both models. This 3D culture method not only maintained the phenotype of the tumor but allowed easy high-throughput screening (HTS) due to its 96-array plate conguration. Results: The Aggregated Spheroid Model (ASM) showed higher expression of cancer markers associated with proliferation, tight junctions formation and epithelial cell identity in HCC cells, as well as cytokine factors associated with immune cell recruitment/activation, ECM regulation, cancer interaction, and angiogenesis regulation. Conclusions: Overall, the proposed ASM better recapitulates the tumor microenvironment, demonstrating the involvement of the ECM/tight junctions in the cell-to-cell signaling processes. This provides more instructive data for in vitro screening of tumor cells.


Background
Hepatocellular carcinoma (HCC) is the sixth most common cancer and the third most frequent cause of cancer-related mortality worldwide [1][2][3]. The genetic and physiological complexity of HCC is a major barrier to the study of tumorigenesis and identi cation of therapeutic targets. Sorafenib is a drug that is the standard of care for patients in advanced stage HCC, and Regorafenib is recommended as a secondline therapy in patients who are nonresponsive but tolerant to Sorafenib [4][5][6]. Sorafenib [7][8] and Regorafenib [9][10][11] are small, novel oral multikinase inhibitors that were developed to target different tyrosine kinases involved in the angiogenic, tumor growth-promoting, and tumor microenvironment signaling pathways, including the Ras/Raf/MEK/ERK pathway, vascular endothelial growth factor receptors (VEGFR1, 2, and 3), platelet-derived growth factor receptor beta (PDGFR-β), and broblast growth factor receptor 1 (FGFR1).
However, HCC patients typically have poor tolerance of systemic chemotherapy due to underlying chronically damaged tissue and liver dysfunction that contains considerable in ammation, brosis, and cirrhosis. Therefore, to nd the most responsive and effective drug for HCC patients, many preclinical studies and in vitro drug assays are conducted using patient derived tumor cells grown under twodimensional (2D) monolayer cell culture conditions. The conventional 2D cell culture system is well established and generates highly reproducible data. Due to these opportunities, high-throughput screening (HTS) methods have been widely used for scienti c experimentation. However, particularly in the eld of drug discovery, its limitations have been increasingly recognized. The conventional 2D cell culture-based HTS platform does not re ect the actual patient's tumor microenvironment such as physiological cell-to-cell and cell-to-extracellular matrix (ECM) interactions [12]. Also, many cells lose some of their actual phenotypic and genomic properties when grown using in vitro methods with 2D cell culture. As a result of these limitations, experimental results using 2D cell culture may provide ambiguous and nonpredictive data for in vivo and clinical responses [13][14][15].
There is a need for suitable experimental models to address these limitations. One model that may overcome these issues is the 3D culture model which was devised to better mimic the in vivo behavior of tumor cells under more adaptable conditions. In contrast to the 2D culture system, a recent study reported that 3D cultured spheroids composed of patient-derived tumor cells could maintain tumor characteristics in vitro for an extended period [16]. Because of the structural and physical differences in dimensionality, 3D cell culture methods have an advantage that leads to differences in the physiological activity of cells.
These biological and physical aspects of 3D cell cultures affect not only the expression of the cell surface receptors but also induce the activation of signal transduction from the outside to the inside of cells. Finally, the characteristics of 3D cell culture in uence gene expression and cellular responses. It has been reported that 3D cultured cell responses are more similar to in vivo testing results compared to those from 2D cell culture [17][18].
To address these issues our research group has developed Pillar/well plates for the 3D cell culture on the surface of the pillar as previously reported [19][20][21][22][23][24]. Since alginate is easier to handle, our previous 3D cell culture studies were conducted by adopting alginate as an extracellular matrix. Commercially available alginates are extracted from harvested brown seaweeds. Although alginates have been used as biomaterials in biological and pharmaceutical elds, its use in biological experiments is controversial as alginate is not an animal-derived material [25][26].
In this regard, Matrigel is widely used and adapted for the 3D cell culture ECM. Matrigel is a soluble extract of basement membrane proteins derived from the Engelbreth-Holm-Swarm (EHS) tumor that forms a 3D gel at 37 ℃. It is known to enhance biological features re ecting their actual cancer cells malignancy and is related to cell proliferation, differentiation, morphology, gene and protein expression, and cellular responses to external stimuli. Matrigel is typically used in monolayer or thin gel conformations, mainly for short-term analysis such as invasion assays, ranging from few hours to 4 days [27][28][29][30]. However, spheroid formation in Matrigel differs depending on the initial cell position. Generally, cells are diffused in Matrigel and form multi-spheroids that are connected. This is the diffused spheroid model (DSM). On other hand, when cells were aggregated in Matrigel, big spheroids are formed. Cells are aggregated on the curvature of the Matrigel spot before gelling with Matrigel, which leads to formation of large spheroids. This is the Aggregated Spheroid Model (ASM).
In this study, our research group adapted Matrigel as the 3D cell culture ECM and optimized gelation methods of the cell-Matrigel mixtures to implement two types of 3D cell culture models. (Diffused Spheroid Model "DSM", Aggregated Spheroid Model "ASM"). Based on two 3D cell culture models, we conducted drug sensitivity assays in order to select e cacious anti-cancer drugs. We analyzed the proteins and cytokine activations related to drug resistance and maintenance of the physiological properties of HCC cells. Through these experiments, we found that the ASM could provide better in vivolike data.

Methods
Preparation of the 96 pillar/well plates The 96-pillar/well plate was manufactured by plastic injection molding and is a robust and exible platform for mammalian cell cultures, enzymatic reactions, viral infections, and compound screenings (Fig. 1). The 96-pillar plate is made of polystyrene (PS) and contains 96-pillars (with a 2 mm pillar diameter and 9 mm pillar-to-pillar distance). The 96-pillar plate was coated with poly-L-lysine (PLL) solution to support Matrigel attachment on the hydrophilic pillar surface. The 96-well plate has 96 complementary wells (with a 7 mm well diameter and 9 mm well-to-well distance) and it can be combined (or "stamped") with a 96-pillar plate to conduct the 3D cell culture and the drug e cacy test. Plastic molding was performed with an injection molder (Sodic Plustech Inc., IL).

Experimental Procedure of High-Throughput Drug Screening
For high-throughput screening and implementation of 3D cell culture models, 1.5 µL volume of mixtures containing approximately 4000 cells and 50% Matrigel were automatically dispensed onto a 96-pillar plate surface using an ASFA™ Spotter ST (Medical & Bio Device, South Korea). The ASFA™ Spotter ST uses a solenoid valve (The Lee Company, USA) to dispense 1.5 µL droplets of the cell-Matrigel mixture on the 96-pillar plate surface.
After dispensing the cells, two types of 3D cell culture models were implemented (DSM and ASM), as shown in (Fig. 2-a). To implement a DSM model, the icing and gelation steps continued with the pillar surface facing upward (Figs. 2-b and -c). To implement the ASM model, the icing and gelation steps continued with the pillar surface facing down (Figs. 2-d and -e). The 96-pillar plate was combined with the pre-warming culture medium and subjected to an additional stabilization for 10 minutes (Fig. 2-f). The 96-pillar plate containing HCC cells in Matrigel was sandwiched (or "stamped") with the 96-well plate for 3D cell culture ( Fig. 2-g). Each 96 well plate was comprised of 96 wells, each 7 mm in diameter, which contained 200uL of cell culture medium. After the cell-Matrigel mixture was stabilized using each gelation method, 3D cells were observed with an optical microscope and appeared to be evenly spread over the pillar surface in the DSM model. In the ASM model, it was con rmed that cells were aggregated in the center of the pillar surface, as shown in (Fig. 2-k).
The cells needed to be stabilized in the Matrigel spots before being treated with drugs. After stabilizing cells, cell viability was checked by 3d live cell staining using Calcein AM dye (Invitrogen, Carlsbad, CA). Majority of cells were alive based on the level of green uorescence as shown in (Fig. 2-j). To analyze the e cacy of all six drugs, we designed a dose-response curve (DRC). The 96-well plate was divided into 7 regions. Each region was composed of a 3 × 7 well array corresponding to six different drug doses (including one control) and their triplicates. After treatment with the six different drugs for a period of six days, drug response was analyzed through viability quanti cation of 3d live cell staining images. As shown in (Fig. 2-j), the area value of 3D live cells differed depending on the drug response.
Experimental Procedure for Image Scanning and Data analysis An automatic optical uorescence scanner (ASFA™ Scanner HE, Medical & Bio Device, South Korea) was used to measure green uorescence intensities (excitation/emission, 494/517 nm to lasers) on the 96pillar surface using an 8-bit code among the RGB codes (0-255), as shown in ( Cell culture Human hepatocellular carcinoma cell lines Hep3B and HepG2 were purchased from the Korean Cell Line Bank (Seoul, South Korea). Both Hep3B and HepG2 were cultured in DMEM medium (Gibco, Grand Island, NY) with 1 µg/mL of puromycin; both media were supplemented with 100 µg/mL of streptomycin, 100 units/mL of penicillin, 250 ng/mL of amphotericin B, and 10% fetal bovine serum. Cell lines were maintained at 37°C in a 5% CO2-humidi ed atmosphere and passaged every four days. Typically, we use the Hep3B and HepG2 cell lines under 20 passages after thawing the frozen cell stock. Under 20 passages, we observed that the Hep3B and HepG2 cell lines easily formed 3D cells in 50% Matrigel on the chip platform.

3D-Cell Viability Assay
Hep3B and HepG2 cells were seeded in a 96-pillar tray at a density of 4000 cells per pillar in triplicate for each treatment. One day after seeding, cells were treated with drugs in a two-fold and seven-dose point serial dilution from 100 µM to 3.12 µM. After 6 days of incubation at 37°C in a 5%-CO2 humidi ed incubator, cell viability was determined using an adenosine triphosphate (ATP) monitoring system based on re y luciferase (CellTiter-Glo® 3D Cell Viability Assay, Promega, Madison, WI) and 3D-live cell staining (Calcein AM, Invitrogen, Carlsbad, CA), according to manufacturer's protocol. Brie y, the assay mixture was prepared in an ATP monitoring system by adding 20 µL of CellTiter-Glo® 3D reagent into 60 µL of medium per well. Cells were lysed with an assay mixture by shaking for 5 min, followed by incubation for 25 min at room temperature. Viable cells were estimated using the SpectraMax iD3 Reader (Molecular Devices LLC, San Jose, CA). In the 3D live-cell staining, the staining solution was prepared by adding 1 µL of Calcein AM into 7 mL of DMEM medium. Cells were incubated with staining solution for 1 hour at 37°C in a 5% CO2-humidi ed atmosphere. Live cell images were acquired using an automatic uorescence microscope scanner (

Cytokine Assay
The expression levels of cytokines and chemokines were analyzed using a Human Cytokine Antibody Array (C5) (RayBiotech, GA). According to the manufacturer's instructions, antibody-embedded membranes were incubated with 1 ml of conditioned media (CM) at 4°C overnight, followed by incubation with HRP-streptavidin at room temperature. Proteins were then visualized using a Chemidoc chemiluminescence imaging system (Bio-rad, Hercules, CA) and chemiluminescent substrate reagent. The signal intensities were quanti ed using Image lab software (Bio-rad, Hercules, CA).

Drug Penetration Assay
DOX was used to evaluate drug accumulation in spheroids because of its uorescent properties. After 1 d of culture, media was replaced with DOX containing media (300 µM). After 12 h of exposure, the 3D cultured cells were washed with PBS before imaging to remove background uorescence noise. Optical sections were acquired at 10 µm intervals and stacked into a Z-projection from which uorescence intensity was calculated.

Results
Quantitative analysis of the 3D cell-forming uniformity in each 3D cell model When analyzing the coe cient of variation (CV) value of the DSM model using the 96-pillar plate for 3D cell culture of this study, the CV value was 2.92 % after 1 day of culture, showing high dispensing uniformity and reproducibility as shown in (Fig. 3-a). Similarly, the ASM model had a CV value of 7.89 %, which was con rmed to be of suitable quality for high-throughput screening analysis for drug discovery as shown in (Fig. 3-b).
High-Throughput screening of six anti-cancer drugs responses in each 3D-cell model Two types of 3D cell models were implemented and then stabilized for 1 d. After the 1 d stabilization, the drugs were applied, and additional 3D cell culture was performed for 6 d.  Western blot analysis of epithelial cell markers and cell proliferation receptor protein expression As shown in (Fig. 6), we performed western blot analysis by recovering cells cultured in 2D and 3D cell models (DSM and ASM). P-Akt was overexpressed in the ASM model, where HepG2 and Hep3B had 3.39fold and 1.05-fold as compared to the 2D, respectively. These values indicated there was a higher p-Akt expression in the ASM model than in the corresponding 2D and DSM cell culture model conditions.
Similarly, the ASM model of HepG2 and Hep3B cells had higher levels of p-Erk expression in the ASM model, HepG2 and Hep3B had levels of 3.14-fold and 2.33-fold as compared to the 2D, respectively ( Fig. 6- 6-e and -f). β-actin, established as unchanged in each of the three cell culture models, was used as a loading control for immunoblot assay. The quantitatively analyzed densitometry graphs for these immunoblots are displayed in (Fig. 6).

High Content image analysis of epithelial cancer cell protein and drug penetration related proteins expression
We conducted immuno uorescence analysis for E-cadherin, bronectin and ZO-1. In the ASM model Ecadherin was overexpressed, where HepG2 and Hep3B had values of 2.66-fold and 1.86-fold as compared to the 2D, respectively ( Fig. 7-a). As shown in (Figs. 7-b and -c), the expression level of bronectin was measured as 1.56-fold and 2.13-fold as compared to the 2D in HepG2 and Hep3B, more than in the corresponding 2D and DSM cell culture models, respectively. Additionally, the expression level of ZO-1 was measured as 1.81-fold and 3.44-fold as compared to the 2D in HepG2 and Hep3B, more than the 2D and DSM cell culture models, respectively.

Cytokine secretion analysis related to cell proliferation and drug resistance
We recovered the cell culture medium in 2D cell culture and 3D cell models (DSM and ASM) and quantitatively analyzed the secreted cytokines. Among the secreted cytokines, those corresponding to the top 30% in the ASM model were selected. These cytokines had more than a 2-fold increase when compared to the 2D cell culture condition as shown in (Fig. 8-a). Cell-to-cell interactions in HepG2 in the ASM model stimulated cancer cell interaction, angiogenesis, ECM regulation, immune cell recruitment, and activation via paracrine signaling of Interleukin (IL)-8, IL-15, Transforming growth factor-β (TGF-β), VEGF-A, Chemokine ligand 2 (CCL2), CCL5, CCL17, Chemokine (C-X-C motif) ligand 7 (CXCL7), CXCL8, CXCL10, CXCL13, Insulin-like growth factor-binding protein-1 (IGFBP-1), IGFBP-2, osteopontin (OPN), angiogenin, tissue inhibitors of metalloproteinases-1 (TIMP-1), and TIMP-2, all of which were increased in the ASM model when compared to 2D and DSM model conditions.

Drug penetration assay for validation of drug resistance in the ASM model using an auto-uorescent drug
We investigated the penetration behavior of DOX in DSM and ASM models as shown in (Fig. 9). DOX, which is widely used in clinical practice and has its own orange-red uorescence (excitation 488 nm, emission 517-614 nm), was chosen as a low-molecular-weight antitumor agent. We treated the cells with DOX for 12 h at 300 µM dosage. Representative spheroid cross-sections observed by confocal microscopy are shown in (Fig. 9). The average penetrated areas where the drug uorescence signals are detectable and at the middle Z-position of 3D cancer spheroids were normalized by the total area of ROI.
The penetrated DOX uorescence was detected at 14.69 % and 72.2 % in the HepG2 DSM and ASM models respectively ( Fig. 9-a). In the Hep3B DSM and ASM models, the penetrated DOX uorescence was detected at 5.75 % and 86.37 % respectively as shown in (Fig. 9-b).

Discussion
In this study, we established 3D in vitro cancer spheroid models (DSM and ASM) that are advantageous for target discovery and drug screening because they recapitulate important characteristics of in vivo tumors and the tumor microenvironment. Based on our previously developed 3D cell culture platforms by means of a Pillar/well plates, we established a 3D-HCC cancer models that can well maintain the physiological characteristics of HCC cancer cells. A549-eGFP-Puro is a LV-eGFP-PGK-Puro (LV031)transduced polyclonal population derived from human lung carcinoma cell line A549. To achieve stable reporter expression in the polyclonal population, the cells were selected using puromycin. Since this cell line does not require an additional 3D live cell staining step, we veri ed the uniformity of cell dispensing in the DSM and ASM models through measurement of the area of A549_GFP cells expressing green uorescence in the live cell state. To determine a robust and reproducible assay, a means coe cient of variation (CV) value of less than 10% is used for a good cell-based assay [31]. When analyzing the CV values of two 3D cancer spheroid models, the CV values was under the 10 % respectively. Also, when observing the morphology of the 3D live cells cultured in each 3D cell model, the 3D cultured cells are attached to the surface of the 96-pillar plate in the DSM model. Conversely, in the ASM model, a dramatic morphological difference was observed in which 3D cultured cells formed as a single spheroid morphology in the ASM model. (white arrows mean 3D cultured live cells in each 3D cell models as shown in (Figs. 3-a and -b).
After establishing the reproducibility and uniformity of 3D cancer spheroid models for HTS analysis, we selected drugs that are commonly used for HCC patients in the clinics, particularly drugs targeting VEGFR, tyrosine-protein kinase Met (c-Met), mitogen-activated protein kinases/extracellular signal-regulated kinases (Ras/Raf/MEK/ERK), PDGFR-β, and those that cause DNA/RNA synthesis damage, etc. All drugs were in phase III or IV trials or were approved oncology drugs from public data provided by the US Food and Drug Administration (FDA). The six drugs selected (including three controls) were dispensed into 96well plates in a dose-dependent manner. Five of the six drugs were treated and diluted three times from 100 µM as the highest concentration in a dose-dependent manner. For DOX, if it was treated at the highest concentration of 100 µM, the drug e cacy would be too high and we would not be able to analyze the IC 50 value of the doxorubicin. Therefore, the highest concentration was set to 1.56 µM. After the drug treatment, we analyzed the 3D cell viability through the 3D Calcein AM live cell staining method, and by quantitatively measuring the area of the 3D live cells. The e cacy of each drug showed a striking difference between the two 3D cell models, which were quantitatively analyzed by IC 50 values as shown in (Fig. 4 and Table 1). We found that 5-FU targeting DNA/RNA synthesis inhibition drugs severely inhibited the HepG2 cell viability in 2D HTS and DSM model conditions. Conversely, the ASM model showed no response based on IC 50 values at 100 µM for the 5-FU drug treatment (Table 1). In addition, when Hep3B cells were treated with 5-FU, the ASM model showed approximately six times higher drug resistance compared to the DSM model. The current study demonstrated that HepG2 and Hep3B cells acquire substantial resistance to 5-FU when the cell-to-cell and cell-to-ECM interactions are re-established in a 3D ASM model (Fig. 5). Thus, in the conventional 2D cell culture model, low doses of 5-FU can induce massive cell death and the drug response results of the 2D cell culture model could be mismatched when compared to actual clinical treatment results. These additional ECM compositions may in uence drug response through altered local drug availability, by affecting the expression of drug targets, or by changing intrinsic cellular defense mechanisms such as increased repair upon DNA damage or evasion of apoptosis [32].
We conducted a molecular study to verify differences in drug resistance and the maintenance of the original epithelial characteristics of the two HCC cancer cells according to the 3D cell models when compared to the conventional 2D cell culture method. p-Akt and p-Erk activities, which are known to be related to cell proliferation and important components of cell survival pathways [33], had the greatest increase in the ASM model compared to the 2D and DSM cell culture models (Fig. 6-a). HCC cell lines HepG2 and Hep3B were both classi ed as epithelial based on their expression of E-cadherin protein and mRNA [34]. However, the epithelial characteristics of HCC cells were not well maintained in the 2D cell culture and the DSM model. Conversely, in the ASM model, the expression of E-cadherin, an epithelial marker, was relatively high. ZO-1 and Occludin, both of which are tight junction markers, could be identi ed when the 3-dimensional structure was formed (Figs. 6-e and -f). Interestingly, Fibronectin, a high-molecular-weight glycoprotein of ECM, was overexpressed in the ASM model. It has been reported that Fibronectin was observed in drug-resistant tumor tissues [35]. Fibronectin production occurred in the actual tumor microenvironment and appeared to promote drug resistance [36,37]. Given these ndings, we con rmed that the ASM model implemented in this study has the advantage of maintaining the actual cell characteristics more e ciently than the conventional 2D cell culture method.
The current study demonstrated differences in drug response and protein expression through the HTS and western blot assay in each cell culture model. We also conducted immuno uorescence analysis for E-cadherin, an epithelial cancer cell marker, as well as bronectin, an ECM marker, and ZO-1 which is a tight junction marker. Since HCC cells have epithelial properties, it can be assumed that culture conditions in which epithelial markers are well expressed can maintain the actual tumor characteristics well. The epithelial characteristics of HCC cells were not well maintained in the 2D cell culture and the DSM model. However, in the ASM model E-cadherin was overexpressed as compared to the 2D. These levels were greater than those in the corresponding 2D and DSM cell culture model counterparts as shown in (Fig. 7a). Fibronectin plays a crucial role in growth, differentiation, adhesion, and migration [38]. Additionally, production of bronectin promotes drug resistance and protects from apoptosis induced by chemotherapeutic drugs when compared to those that were grown on plastic [39]. Loss of E-cadherin increases tight junction permeability and alters the localization of ZO-1 [40]. Therefore, the decrease in the expression of the tight junction marker ZO-1 allows for increased drug delivery into cancer cells. The expression levels of the ECM component and structure of tight junctions were measured by immunostaining for bronectin and ZO-1 to understand the differences of the drug response in each cell culture model as shown in (Figs. 7-b and -c). Fibronectin and ZO-1 were not well expressed in the 2D cell culture and DSM models. However, in the ASM model, these components which are related to the drug resistance were well maintained.
Also, secreted cytokines were analyzed to understand differences in drug resistance and phenotypic characteristics of the two HCC cancer cells according to the 2D and 3D cell models (Fig. 8). HepG2 and Hep3B cells cultured in ASM model stimulate cancer cell-to-cell interaction, angiogenesis, ECM regulation, immune cell recruitment, and activation via paracrine signaling, all of which were increased in the ASM model when compared to the 2D and DSM model conditions. These results indicated that the ASM model implemented in this study has the advantage of maintaining the actual HCC cell characteristics. We analyzed the different penetration behavior of DOX in DSM and ASM models. Therefore, the difference in drug permeability causes a difference in the drug response.

Conclusions
We tested two in vitro 3D cancer spheroid models (DSM and ASM) of HCC cells using 96-pillar/well plates. We cultured cells and created spheroids by either aggregating and diffusing cells in Matrigel attached on the tip of a 96-pillar plate. According to the initial cell position in Matrigel, two 3D-HCC cancer in vitro models (DSM and ASM) were established. Among them, the ASM showed high resistance to drugs that were related to the high expression of cancer proliferation, ECM/tight junction markers, and epithelial markers. Additionally, we found that ASM secreted more cytokines related to ECM regulation (TIMP02) and cancer interaction (Il-8, TGF beta 2). The ASM showed lower drug penetration compared to DSM. Overall, the proposed ASM better recapitulates the tumor microenvironment, demonstrating the involvement of the ECM/tight junctions in the cell-to-cell signaling processes. This provides more instructive data for in vitro screening of tumor cells.