Sage Journals: Discover world-class research

Abstract

The widespread use of two-dimensional (2D) monolayer cultures for high-throughput screening (HTS) to identify targets in drug discovery has led to attrition in the number of drug targets being validated. Solid tumors are complex, aberrantly growing microenvironments that harness structural components from stroma, nutrients fed through vasculature, and immunosuppressive factors. Increasing evidence of stromally-derived signaling broadens the complexity of our understanding of the tumor microenvironment while stressing the importance of developing better models that reflect these interactions. Three-dimensional (3D) models may be more sensitive to certain gene-silencing events than 2D models because of their components of hypoxia, nutrient gradients, and increased dependence on cell-cell interactions and therefore are more representative of in vivo interactions. Colorectal cancer (CRC) and breast cancer (BC) models composed of epithelial cells only, deemed single-cell-type tumor spheroids (SCTS) and multi-cell-type tumor spheroids (MCTS), containing fibroblasts were developed for RNAi HTS in 384-well microplates with flat-bottom wells for 2D screening and round-bottom, ultra-low-attachment wells for 3D screening. We describe the development of a high-throughput assay platform that can assess physiologically relevant phenotypic differences between screening 2D versus 3D SCTS, 3D SCTS, and MCTS in the context of different cancer subtypes. This assay platform represents a paradigm shift in how we approach drug discovery that can reduce the attrition rate of drugs that enter the clinic.

Keywords

three-dimensional culture spheroid high-throughput screen tumor microenvironment RNAi

Introduction

RNA interference (RNAi) is a biological process involving the targeting of specific mRNA transcripts for degradation resulting in gene loss of function and has become a powerful tool in high-throughput screening (HTS) for the discovery of novel drug targets and the dissection of molecular mechanisms of action.^1–3 Although HTS tools and methods enable the collection of data on the genomewide scale, screens are limited by the same technical aspects of current in vitro assay systems in low throughput that have proven to be unable to provide robust, high-priority targets that can withstand extensive validation.⁴

Cell-based proliferation studies of cancer in both high and low throughput are typically conducted using cells of a single type. This is problematic because the conventional solid tumor has been shown to be influenced by local and systemic factors produced by noncancerous cells.⁵ Emerging evidence supports the critical role of cancer-associated stromal cells in promoting tumor initiation, differentiation, angiogenesis, invasion, metastasis, and immunosuppression.^6,7 Conventional two-dimensional (2D) cell culture methods involve growing cells on a flat, plastic surface, but little is known about how key aspects of in vivo biology, such as nutrient and oxygen gradients, govern gene expression patterns in three-dimensional (3D) orientation.^8,9 This limits our understanding of intra- and extracellular signaling influenced by 3D cell-cell interactions as well as our ability to identify clinically relevant molecular targets. Therefore, we hypothesized that the fate of cells is intimately related to contributions from surrounding cells and their 3D configuration.

We designed an RNAi screen that would explore this hypothesis in the context of breast and colorectal cancer. The goal of this screen is threefold: to elucidate subtle differences in 2D versus 3D tumor growth driven by siRNA-mediated gene silencing, to identify specific factors that affect the viability of breast and colorectal cancer, and to determine the influence of fibroblasts on the proliferation of cancer cells.

Materials and Methods

Cell Lines and Reagents

DLD-1 (ATCC) cells were grown in RPMI medium 1640 (Gibco, Waltham, MA), whereas MDA-MB-231-GFP (Cell Biolabs, Inc., San Diego, CA) and RFP-expressing human neonatal dermal fibroblasts (HNDF; Angio-Proteomie, Boston, MA) cells were grown in DMEM (Gibco), all according to vender protocols. All media were supplemented with 10% fetal bovine serum (FBS), 100 U penicillin, and 100 µg streptomycin, components purchased from ThermoFisher Scientific (Waltham, MA).

Stable Cell Line Generation

Puromycin kill curves were established for DLD-1 cells with a final concentration of 2 µg/mL. Cignal Lenti-positive control green fluorescent protein (GFP) particles were transduced into DLD-1 cells (multiplicity of infection = 10). Cignal Lenti reporters are ready-to-transduce lentiviral particles constitutively expressing GFP, driven by cytomegalovirus promoter. Transduction was performed in Corning 6-well microplates; clonal isolation was performed by single-cell dilution into Corning 384-well microplates. Fluorescence and proliferation were monitored daily for 3 wk using the Invitrogen EVOS FL Auto Imaging System. Clonal populations with high-viability bright fluorescence expression were isolated and tested for transfectability. The chosen clone for screening was designated “DLD-1-GFP.”

Primary Screen

RNAi screening was conducted using the Ambion Silencer Select Human Kinome siRNA Library Version 4, which consists of three unique, nonoverlapping, nonpooled siRNAs per 710 human kinase genes. siRNA reagents (2 µL; 400 nM) were stamped into microplates (described in detail below) using a Velocity11 VPrep liquid-handling system (Agilent, Santa Clara, CA) integrated into a BioCel robotic platform (Agilent) in columns 1 to 22, leaving columns 23 and 24 empty for negative (Ambion Silencer Select Negative Control #2) and positive (Qiagen Allstars Hs Cell Death) controls, respectively. The median value of each plate’s negative control column was used to normalize sample wells, whereas the positive control was used to assess transfection efficiency and assay performance.

Two-dimensional screening was conducted in Greiner Bio-One 384-well black, clear, flat-bottom microplates, whereas 3D screening was conducted in Corning 384-well black, clear, round-bottom ultra-low-attachment spheroid microplates. Invitrogen lipofectamine RNAiMAX transfection reagent (2D: DLD-1-GFP, 0.15 µL; HNDF, 0.20 µL; multi-cell-type tumor spheroids [MCTS], 0.17 µL; 3D: DLD-1-GFP, 0.05 µL; HNDF, 0.08 µL; MCTS, 0.07 µL) was added in 20 µL serum-free, antibiotic-free media to plate wells using a Thermo Scientific Matrix WellMate and Microplate Stacker. Plates were incubated for 45 min at room temperature to allow for the sufficient formation of siRNA-to-lipid complexes. Cells were added to plate wells in 20 µL media containing 20% FBS and without antibiotics (2D: DLD-1-GFP, HNDF, 2000 cells/well; MCTS, 2000 cells/well, 3:2 ratio; 3D: DLD-1-GFP, HNDF, 500 cells/well; MCTS, 500 cells/well, 3:2 ratio). The final 40 µL per well contained 20 nM siRNA in media with 10% FBS. Cells were cultured for 72 h at 37 °C in 5% CO₂ before endpoint assays.

Viability Assay

Cell viability was assessed using the CellTiter-Glo Luminescent Cell Viability Assay. CellTiter-Glo reagent was added directly to microplates (2D: 20 µL, 60 µL final volume; 3D: 30 µL containing 10% trypsin, 70 µL final volume) and incubated at room temperature for 20 min for 2D culture. For 3D culture, plates were incubated on a shaker at room temperature for 5 min and then set still for 20 min. Luminescence signal was detected using the EnVision Multilabel Plate Reader. The EnVision Multilabel Plate Reader uses an ultrasensitive aperture and detector system to collect emitted luminescence from the sample in 384-well plate format. Plates were read using the following parameters: 0.1 mm distance between plate and detector, 0.3 s measurement time.

Image Acquisition

Plates were imaged with the IN Cell Analyzer 6000 Cell Imaging System (GE Healthcare Life Sciences). The IN Cell Analyzer 6000 uses a scientific CMOS 5.5 megapixel camera and a proprietary epifluorescence-based line scanning confocal system for image acquisition in confocal, open aperture, and transmitted light modes. Image acquisition was performed with a 10× Nikon lens objective.

Primary Screen Analysis and Hit Selection

The assay readout for each well was normalized by dividing by the median negative control readout for the plate and multiplying by 100, yielding a percentage of negative control value that corrects for plate-to-plate variance. A robust z-score was then calculated by subtracting the median screen percentage and dividing by the screen median absolute deviation. The difference between z-scores was then used to assess differences between each screen. The RSA algorithm¹⁰ was used to rank gene-level differences between the screens. All analysis was performed in R.¹¹ The gene-level aggregation of the data is the result of taking the median of a particular metric (percentage viability, z-score, z-diff, etc.) at the siRNA level, usually for three different siRNAs against the same gene.

Secondary Screen

The secondary screen was conducted using the Ambion Select Library. Genes with the highest median z-score from the primary screen were selected, and three independent siRNAs against these genes were cherry-picked and screened in the same assay. In a few cases, the siRNA sequence in the Ambion Select Library was the same as in the primary screening library, in which case it was eliminated from the data set as it would not be an independent data point. Data in the secondary screen was analyzed as it was in the primary, including negative control normalization and z-score calculation. Ranking the follow-up genes by the median z-score for all independent siRNAs tested against the gene (usually six) provided the basis for selecting genes for further validation.

Results

A 384-Well HTS-Compatible Spheroid Assay for RNAi Screening

The workflow for this screening campaign involved eight distinct culture conditions (Suppl. Fig. S1). The assay involved the generation of 3D spheroids with an approximate size of 500 µm in 384 ultra-low-attachment plates, generating one spheroid per well. The corresponding 2D-monolayer cultures were grown in 384-well flat-bottom plates. For this particular study, we studied four different 3D models: single-cell-type tumor spheroids (SCTS) representing (1) breast cancer, (2) colorectal cancer, (3) stromal cells, and (4) MCTS (DLD-1-GFP cells and fibroblasts). The four models as described above were also grown separately as 2D monolayers. Following transfection with the Kinome library and assay controls after 72 h, CellTiter-Glo was added to the same plates and read on the EnVison for viability as assessed by measurement of ATP. The results from the viability screen were then funneled through an informatics analysis to identify genes whose knockdown significantly and differentially reduced growth under conditions of either geometry (2D vs. 3D), biology (SCTS vs. MCTS), or cancer type (BC vs. CRC).

The first step of establishing a robust screening assay was to determine efficient transfection conditions for all eight parameters of this HTS screen ( Fig. 1A–H ). Optimal cell-seeding quantity for MDA-MB-231-GFP, DLD-1-GFP, HNDF, and MCTS was determined by observing cell density in 2D/3D 384-well plates over 72 h (not shown). The lipid-based transfection reagent RNAiMax was titrated in wells containing 20 nM siRNA of either a nontargeting control (negative) or a cell death control (positive) following vendor protocols (see the Materials and Methods section). After 72 h, CellTiter-Glo was added to determine viability. The lipid concentration that resulted in the greatest difference in viability between these two controls while minimizing lipid-associated toxicity was chosen to move forward with screening.

Figure 1.

Transfection optimization. (A–H) Lipid reagent and transfection efficiency optimization for MDA-MB-231-GFP, DLD1-GFP, HDNF, and MCTS in 2D (A–D) and 3D (E–F) are delineated.

Effect of Gene Knockdown in 2D versus 3D SCTS Models

Having established a robust and reproducible platform to measure spheroid viability and morphology, we ran the Kinome library against both MDA-MB-231-GFP and the DLD-1-GFP cell lines grown as SCTS and 2D monolayers ( Fig. 2 ). A scatter plot of the median Z-scores for the screen run for both the breast cancer cell line MDA-MB-231-GFP and the colorectal cancer cell line DLD-1-GFP under 2D and 3D conditions is shown in Figure 2A . The breast cancer cell line shows less of a correlation between the 2D and 3D SCTS screen conditions as compared with the correlation between the 2D and 3D conditions for the colorectal cell line DLD-1-GFP. This is reflected in the number of genes showing differential activity between the 2D and 3D breast and colorectal cancer. Of the 704 genes screened, less than 2% of the library showed a differential effect between the 2D and 3D culture conditions. The genes that when knocked down caused a differential viability effect between 2D monolayer and 3D SCTS conditions are depicted in Figure 2 and Supplementary Table S1 for MDA-MB-231-GFP and Supplementary Table S2 for DLD-1. Among the genes showing a differential effect between the 2D and 3D conditions for breast cancer, knockdown of ALDH18A1, PLK1, and MAP3K8 results in the most significant decrease in cell viability in 2D as compared with 3D. This is supported by Z-scores of −6.94 and −2.19 for viability as a consequence of silencing ALDH18A1 under 2D and 3D conditions, respectively. The gene-level aggregation of the data is the result of taking the median of a particular metric (percentage viability, z-score, z-diff, etc.) at the siRNA level, usually for three different siRNAs against the same gene. In this case, the Z-score for the viability for every gene is calculated from the median values of the z-score for viability for each siRNA targeting that gene. The z-diff is the difference between the Z scores between the 2D and 3D conditions. For loss-of-function screening, any z-diff scores below a value of −2 are considered statistically significant. Similarly, for genes PLK1 and MAP3K8, the Z-scores and z-diff are −6.52, −2.55, −3.58 and −3.13, 0.57, and −3.70, respectively (Suppl. Table S1; Fig. 2B ). Furthermore, these Z-scores are reflective of corresponding decreases in viability ( Fig. 2C ). Loss of ALDH18A1 causes a 53% and 23% decrease in viability in 2D and 3D, respectively. Similarly, loss of PLK1 causes a 56% and 26% decrease in viability in 2D and 3D conditions ( Fig. 2C ; Suppl. Table S1). For MAP3K8, a 22% loss in viability occurred under 2D conditions but not 3D conditions. Figure 2D and E delineate the genes whose knockdown caused differential cell killing in 3D conditions as compared with 2D conditions. Knockdown of genes EPHA10, PHKG1, and SLK are associated with Z-scores of −2.82, −4.00, and −3.09 under 3D conditions while remaining unaffected under 2D conditions. This is reflected in both the bar graph in Figure 2D and the scatter plot in Figure 2E where these Z-scores are associated with a 29%, 39%, and 31% loss in viability for these genes in 3D as compared with 2D. We have highlighted the reproducibility of this effect in loss of viability between 2D and 3D conditions for one particular gene, ALDH18A1 ( Fig. 2D ). This scatter plot delineates the viability of MBA-MD-231-GFP under 2D and 3D conditions in both the primary and the secondary screen. The secondary screen was run with three different siRNA reagents from the primary screen targeting the same gene to validate the hit. Knockdown of ALDH18A1 in MDA-MB-231-GFP cells leads to a significant loss in viability in 2D as compared with 3D ( Fig. 2F ). In DLD1-GFP, knockdown of gene PRPS2 causes a 70% loss in viability in 2D as compared with a 58% loss in viability in 3D. Conversely, knockdown of PLK1 causes a 70% decrease in viability in 3D as compared with a 66% loss of viability in 2D (Suppl. Table S2).

Figure 2.

Effect of gene knockdown in single-cell-type tumor spheroids in two dimensions (2D) versus three dimensions (3D). (A) Correlation plots for the median Z-scores from the Kinome screen for MDA-MB-231 in 2D versus 3D and for DLD-1 in 2D versus 3D. (B) Bar graph showing the difference in the median Z-scores for viability between 2D and 3D conditions for MDA-MB-231 for genes selectively killing in 2D, and (C) scatter plot for the normalized viability (y-axis) for genes selected in (B) under both 2D and 3D conditions (x-axis). (D) Bar graph showing the difference in the median Z-scores for viability between 2D and 3D conditions for MDA-MB-231 for genes selectively killing in 3D, and (E) scatter plot for the normalized viability (y-axis) for genes selected in (D) under both 2D and 3D conditions (x-axis). (F) Scatter plot comparing MDA-MB-231-GFP in 3D against 2D, with a line at the “x=y” axis to aid in viewing trends where all siRNAs lead to greater or less survival under 2D conditions as compared with 3D conditions. Each dot represents a single siRNA reagent.

Effect of Gene Knockdown in Different SCTS Models

Having established the effect of geometry on viability, we then investigated the effect of the different cancer types on viability. Figure 3A shows the correlation plots comparing the median Z-scores for breast and colorectal cancer with each other in both 2D and 3D SCTS conditions. Within the 2D conditions, the correlation between breast and colorectal cancer is 0.27, which allows for a reasonable overlap in genes affecting viability. However, a comparison between breast and colorectal cancer for the 3D SCTS condition reveals an R-score of 0.12, which indicates very little overlap between the genes affecting viability. A comparison of the genes differentially regulated in 2D between breast and colorectal cancer implicates ALDH18A, WEE1, and PRPS2. Although knockdown of ALDH8A1 and WEE1 shows a 46% and 33% decrease in viability in MDA-MB-231-GFP cells under 2D conditions, knockdown of PRPS2 shows a 71% decrease in viability in DLD1-GFP cells as compared with virtually no change in the MDA-MB-231-GFP cells ( Fig. 3B ; Suppl. Table S3). Figures 3C , D show the median Z-scores for viability and the percentage viability for differentially regulated genes under 3D conditions for both cell lines. Knockdown of PLK1, AURKB, and PRPS2 shows significant decrease in viability under 3D conditions for DLD1-GFP while not affecting the MDA-MB-231-GFP cell line (Suppl. Table S4). Conversely, knockdown of PHKG1, PRKDC, EPHA10, RPS6KC1, and ALDHA81 significantly affects viability in MDA-MB-231-GFP cells as compared with DLD1-GFP ( Fig 3C ; Suppl. Table S4). Figure 3E shows the scatter plots for the validation of the hits PLK1, AURKB, and PRPS2 from the primary screen. All six siRNA reagents from both the primary and the validation screen show a loss in viability after knockdown of PLK1, AURKB, and PRPS2 in DLD1-GFP 3D cultures as compared with MDA-MB-231 GFP cells. Similarly, knockdown of PRPS2 and PLK1 caused a significant loss in viability in DLD1-GFP as compared with MDA-MB-231-GFP ( Fig. 3F ).

Figure 3.

Effect of gene knockdown in single-cell-type tumor spheroids in breast cancer versus colorectal cancer and two dimensions (2D) versus three dimensions (3D). (A) Correlation plots for the median Z-scores from the Kinome screen for MDA-MB-231 versus DLD-1 in 2D and 3D. (B) Scatter plot showing the difference in viability (y-axis) plotted for genes that showed a difference between breast and colorectal cancer under 2D conditions. (C) Bar graph showing the difference in the median Z-scores for viability between breast cancer and colorectal cancer under 3D conditions. (D) Scatter plot for the normalized viability (y-axis) for genes selected in (C) under 3D conditions for both breast and colorectal cancer (x-axis). (E) Scatter plot for PLK1, AURKB, and PRPS2 comparing MDA-MB-231 GFP in 3D against DLD1-GFP in 3D, with a line at the “x=y” axis to aid in viewing trends where all siRNAs lead to less survival under DLD1-GFP conditions as compared with MDA-MB-231-GFP in 3D (E) and in 2D (F). Each dot represents a single siRNA reagent.

Effect of the Stroma-Differential Sensitivities to Gene Knockdown between Multi- and Single-Cell-Type Tumor Models

Tumor stroma plays a very important role in therapeutic resistance, cell proliferation, and drug response. To better mimic the tumor microenvironment, we screened MCTS composed of DLD1-GFP cells with fibroblasts. Figure 4A shows three plots that investigate the correlation between the median Z-scores of MCTS in 2D and 3D, fibroblasts alone in 2D and 3D, and colorectal SCTS and MCTS. None of these correlation plots show a strong correlation between viability responses, with R values of 0.13, 0.11, and 0.14, respectively. Figure 4B delineates the differential response between the MCTS and fibroblasts alone under 2D and 3D conditions. In both the primary and validation screens, knockdown of PRPS2 caused a severe decrease in viability in DLD1-GFP SCTS and MCTS. However, knockdown of PRPS2 did not affect the viability of fibroblasts alone ( Fig. 4B ). The addition of fibroblasts to MCTS does not confer a protective advantage to the DLD1-GFP cells in either 2D or 3D in terms of viability (Suppl. Table S5). For example, knockdown of genes IP6K3, TNK2, KIAA1804, and CIB4 causes a decrease in viability in 2D multi-cell-type monolayers while resulting in a substantial increase in viability in 3D MCTS. The difference in viability in 3D between the MCTS and the DLD-1 SCTS is at the crux of explaining the effect of the stroma on tumor cell proliferation. Down-regulation of genes MST1R, EGFR, ATM, CDK18, TGFBR2, PRKACA, DAPK1, CSK, KHK, NTRK2, FRK, and TEC cause a significant decrease in viability in 3D MCTS as compared with the 3D DLD1-GFP SCTS ( Fig. 4C ; Suppl. Table S6). Knockdown of CKB, DGKH, JAK1, LCK, MAP3K14, NTRK2, NTRK3, PGK1, PTK7, TRIB3, and VRK2 resulted in a loss in viability in fibroblasts grown in 3D cultures as compared with 2D monolayers ( Fig. 4D ; Suppl. Table S7).

Figure 4.

Multi-cell-type tumor spheroids (MCTS) versus single-cell-type tumor spheroids in two dimensions (2D) versus three dimensions (3D). (A) Correlation plots for the median Z-scores from the Kinome screen for MCTS 2D versus 3D, fibroblasts 2D versus 3D, and MCTS 3D versus colorectal 3D. (B) Scatter plot for gene PRPS2 comparing MCTS 3D against fibroblasts in 3D, with a line at the “x=y” axis to aid in viewing trends where all siRNAs lead to less survival under MCTS conditions as compared with fibroblast conditions. Each dot represents a single siRNA reagent.

Discussion

The tumor microenvironment plays an important role in the processes of tumor growth, metastasis, and drug resistance, yet most oncology drug discovery programs in HTS rely on cell proliferation assays with cancer cells cultured in a 2D monolayer on plastic.^5,12–14 Although 2D assay systems have been useful for discovering compounds that target genetic drivers of cancers, they are devoid of influence from the heterogeneous cell population associated with the tumor microenvironment. Two-dimensional cultures also lack the complex architecture of the tumor microenvironment in vivo. The intricate network of distinct populations of stromal and immune cells in epithelial tumors communicate and contribute to the physiology and biochemistry of the tumor, including the responses to chemotherapeutic treatments. Accordingly, in vitro 3D organotypic models designed to recapitulate the natural tumor composition and environment are expected to provide a biologically relevant assay platform that will fill a critical gap between the 2D cell proliferation assays and animal models. Here, we proposed to build an HTS compatible, multi-cell- and single-cell-type 3D organotypic model of solid tumors and investigate the genes necessary for differential survival between the 2D and 3D models so as to be able to understand the mechanisms of survival specific to each growth mode and target them appropriately. Furthermore, in the context of understanding how viability was differentially regulated between cells growing in 2D and 3D models, we were interested in knowing which genes were unique to a particular tumor type versus the other. Ultimately, from the context of building and understanding complex, physiologically relevant systems, we were also interested in understanding how the stromal cells in a tumor in vivo affect the proliferation of cancer cells.

When comparing the growth in 2D versus 3D architecture of breast cancer and colorectal cancer cells, there were some genes that when knocked down differentially regulated viability in 2D monolayer and 3D SCTS models. For example, knockdown of the genes ALDH18A1 (aldehyde dehydrogenase 18 family, member A1) and PLK1 (polo-like kinase 1) in breast cancer cells caused a significant decrease in viability in the 2D monolayer as compared with 3D SCTS ( Fig. 2B , C ). The aldehyde dehydrogenase family of genes is associated with cancer stemlike activity,^15,16 whereas PLK1 is a key regulator of cell division and is overexpressed in many types of cancer. Reduced viability as a consequence of knocking down these genes is not surprising, especially in the context of cells that proliferate faster in 2D. In 3D cultures, cells proliferate at a slower pace because of the gradients of hypoxia, oxygen, and nutrients; as a consequence, they are less affected by the knockdown of genes that are more directly involved with active proliferation. Similarly, in colorectal cancer, knockdown of PLK1 and PRPS2 causes a significant drop in viability in both 2D and 3D models of growth ( Fig. 3B–D ). However, although the effect of PLK1 knockdown in 2D monolayers is common to both breast and colorectal cancer, the effect of PRPS2 knockdown on viability is seen only in the colorectal cell line DLD1-GFP in both 2D and 3D SCTS. Furthermore, it is also one of the genes whose knockdown significantly decreases viability in 3D SCTS ( Fig. 3B ). In colorectal cancer, it is known that the purine nucleotide biosynthesis enzyme PRPS2 is selectively arginylated, unlike its close homologue PRPS1, and this arginylation of PRPS2 directly facilitates its biological activity.¹⁶ Moreover, selective arginylation of PRPS2 but not PRPS1 is regulated through a coding sequence–dependent mechanism that combines elements of mRNA secondary structure with lysine residues encoded near the N-terminus of PRPS1. Up-regulation of PRPS2 is a driving factor in colorectal cancer, and it is intuitive that down-regulation of PRPS2 would result in a decrease in viability in a colorectal cancer model.

In the context of 3D breast cancer SCTS, knockdown of PHKG1, PRKDC, EPHA10, NEK4, and IP6K1 causes significant decreases in viability in breast cancer 3D SCTS as compared with colorectal cancer 3D SCTS. IP6K1 belongs to a family of inositol phosphatase kinases that regulate cell-cell adhesion and thereby metastasis in breast cancer.¹⁷ Knockdown of this gene resulting in a significant loss of viability is conceivable given its role in cell-cell signaling. Furthermore, Ephrin receptor A10 (EphA10) is a relatively uncharacterized protein that is expressed in many breast cancers but not expressed in normal breast tissue.¹⁸ Knockdown of this gene also results in decreased viability in the 3D breast cancer SCTS model.

One of the most important aspects of building this screening platform was the ability to address the differences between SCTS and MCTS models. We found that the knockdown of ATM, PRKACA, KHK, FRK, and NTRK2 caused a significant decrease in viability in the 3D MCTS model as compared with the SCTS model. Most of these genes have been shown to be associated with increased differentiation, and in fact, FRK functions as an oncogene.¹⁹ Knockdown of these genes is purported to be associated with reduced viability, and this is indeed what we see in our 3D models. Surprisingly, these were not associated with an SCTS effect. Instead, it seemed like the growth of the fibroblasts and DLD-1 cancer cells together made them more susceptible to the knockdown of these genes. The effect of the stromal component on the sensitivity of the cancer cells to growth inhibition is an essential component of this screen.

The development of our 3D MCTS and SCTS RNAi screening platform is a step toward understanding the contribution of the tumor microenvironment to cancer cell viability. There is a dire need for chemotherapeutic combinations that will selectively attack and kill both the actively proliferating cancer cells and the slower-growing cancer stem cell population that is responsible for generating the drug-resistant cohort of cells that survive and repopulate the tumor. Our work aims to study and understand the role of tumor heterogeneity, clonal evolution, dormancy, and cell death and, in doing so, to bring forth novel mechanisms of action that will help uncover druggable targets. In the long term, we hope that our work will help alleviate the pain and suffering caused by these diseases by providing the means to better manage and combat them. We also hope that our HTS 3D model represents a paradigm shift in the way we view and define relevant biological models and that this screening regimen will be broadly applied to more than one disease type.

Footnotes

Acknowledgements

We would like to thank Carleen Klumpp-Thomas for help with the robotics implementation of the screens.

Supplementary material is available online with this article.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was funded by the National Institutes of Health Intramural Research Program to NCATS and by the NCGC Spark award established by Ajit Jadhav.

References

Martineau

H. M.

Pyrah

I. T.

Review of the Application of RNA Interference Technology in the Pharmaceutical Industry. Toxicol. Pathol. 2007, 35, 327–336.

Iorns

Lord

C. J.

Turner

. Utilizing RNA Interference to Enhance Cancer Drug Discovery. Nat. Rev. Drug Discov. 2007, 6, 556–568.

Gao

Yang

Jiang

., Applications of RNA Interference High-Throughput Screening Technology in Cancer Biology and Virology. Protein Cell 2014, 5, 805–815.

Prinz

Schlange

Asadullah

Believe It or Not: How Much Can We Rely on Published Data on Potential Drug Targets?

Nat. Rev. Drug Discov. 2011, 10, 712–712.

Olumi

A. F.

Grossfeld

G. D.

Hayward

S. W.

. Carcinoma-Associated Fibroblasts Direct Tumor Progression of Initiated Human Prostatic Epithelium. Cancer Res. 1999, 59, 5002.

Le Bitoux

M.-A.

Stamenkovic

. Tumor-Host Interactions: The Role of Inflammation. Histochem. Cell Biol. 2008, 130, 1079.

Liotta

L. A.

Kohn

E. C.

The Microenvironment of the Tumour-Host Interface. Nature 2001, 411, 375–379.

Lee

G. Y.

Kenny

P. A.

Lee

E. H.

. Three-Dimensional Culture Models of Normal and Malignant Breast Epithelial Cells. Nat. Methods 2007, 4, 359–365.

Weaver

V. M.

Fischer

A. H.

Peterson

O. W.

. The Importance of the Microenvironment in Breast Cancer Progression: Recapitulation of Mammary Tumorigenesis Using a Unique Human Mammary Epithelial Cell Model and a Three-Dimensional Culture Assay. Biochem. Cell Biol. 1996, 74, 833–851.

10.

Konig

Chiang

C. Y.

B. P.

. A Probability-Based Approach for the Analysis of Large-Scale RNAi Screens. Nat. Methods 2007, 4, 847–849.

11.

Team

T. R. C.

R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing; 2015.

12.

Garnett

M. J.

Edelman

E. J.

Heidorn

S. J.

. Systematic Identification of Genomic Markers of Drug Sensitivity in Cancer Cells. Nature 2012, 483, 570–575.

13.

Liotta

L. A.

Kohn

E. C.

The Microenvironment of the Tumour-Host Interface. Nature 2001, 411, 375–379.

14.

Bissell

M. J.

Radisky

D. C.

Rizki

. The Organizing Principle: Microenvironmental Influences in the Normal and Malignant Breast. Differentiation 2002, 70, 537–546.

15.

Qiu

. The Expression of Aldehyde Dehydrogenase Family in Breast Cancer. J. Breast Cancer 2014, 17, 54–60.

16.

Zhang

Patel

D. M.

Colavita

. Arginylation Regulates Purine Nucleotide Biosynthesis by Enhancing the Activity of Phosphoribosyl Pyrophosphate Synthase. Nature Commun. 2015, 6, 7517.

17.

Rao

. Inositol Pyrophosphates Promote Tumor Growth and Metastasis by Antagonizing Liver Kinase B1. Proc. Natl. Acad. Sci. U.S.A. 2015, 112, 1773–1778.

18.

Nagano

Maeda

Kanasaki

. Ephrin Receptor A10 Is a Promising Drug Target Potentially Useful for Breast Cancers Including Triple Negative Breast Cancers. J. Control. Release 2014, 189, 72–79.

19.

Simon

E. P.

Freije

C. A.

Farber

B. A.

. Transcriptomic Characterization of Fibrolamellar Hepatocellular Carcinoma. Proc. Natl. Acad. Sci. U.S.A. 2015, 112, E5916–E5925.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

1.93 MB

RNAi High-Throughput Screening of Single- and Multi-Cell-Type Tumor Spheroids: A Comprehensive Analysis in Two and Three Dimensions