Metastatic Tumor-in-a-Dish, a Novel Multicellular Organoid to Study Lung Colonization and Predict Therapeutic Response

Postcancer survival has improved dramatically in the last few decades, in part, due to early diagnoses, improvement in imaging techniques, radiotherapy, and chemotherapy regimens (1). Therapeutic discovery has been slow, especially for metastatic cancers begging the question of whether we are using preclinical models that closely represent the primary tumor to test oncology drugs. A significant hindrance to the discovery of new cancer therapies is that current screening methods are poorly predictive of the clinical efficacy and safety of drug candidates.

The applicability of two-dimensional (2D) cultures, which along with murine models has been the mainstay of basic cancer biology research, has become increasingly called into question (2). Despite rapid advances in the ability to screen large libraries of compounds, there has not been a corresponding increase in the number of approved drugs that had been expected from such large screens using 2D cultures followed by three-dimensional (3D) spheroid culture and animal models to select clinical candidates (3). An advantage of the 3D system is the cell-to-cell communication between cancer cells affecting how cells react to therapeutic agents (4). However, a significant disadvantage is the lack of complete tumor microenvironment including the stroma and vasculature. This reflects the inadequacy of current culture systems that does not correlate clinically to predict chemoresistance or propensity for metastasis. Hence, developing an in vitro model to better represent a more physiologically relevant system, one that uses healthy cells as a matrix, to model the development, colonization, and response of metastatic tumors to treatment is essential. While modeling “mini” healthy organs has been fast paced and prolific, efforts to model various solid tumors and their microenvironment have been lacking. Further improvement in organoids will allow advance understanding of hypoxia, metastasis, treatment effects, and drug resistance at the cellular level (5). We have now combined the progress made in modeling healthy organelles in organoid culture, and introduced tumor cells, creating an organoid system that recapitulates the tumor microenvironment with surrounding healthy tissue.

Using a combination of commercially available cell lines and primary patient-derived cells, we created a lung organoid in vitro, which we have named “primitive lung-in-a-dish” (PLiD). We then grew commercially available colon or ovarian cancer cell lines in PLiD and named it “metastatic tumor-in-a-dish” (mTiD). This model now represents the metastatic potential of tumor cells in vitro. Using this model, we tested the therapeutic efficacy of several standard-of-care chemotherapeutic agents. Furthermore, we used primary patient-derived colon or ovarian tumor cells and corroborated the patient's response to therapy with that of the mTiD cultures in vitro. Thus, we have now developed a novel platform with which to screen drug responses, not only for novel therapeutics, but also to optimize a precision approach where a patient's tumor can be quickly tested in an mTiD system for effective chemotherapeutics.

Human normal lung epithelial cells (NL20, ATCC) were cultured using airway epithelial cell basal medium containing bronchial epithelial growth kit (ATCC). Human normal lung fibroblasts (MrC5, ATCC) were grown in fibroblast basal media containing 10% FBS. Human umbilical vein endothelial cells (HUVEC) and human normal lymphatic endothelial cells (HLEC) were grown in endothelial basal medium (EBM, Lonza) containing endothelial growth factor (EGM-2 Bullet Kit, Lonza). The cancer cell lines were obtained from ATCC and cultured in appropriate media. A2780 and cisplatin-resistant A2780/C30 were gifted by Dr. Andrew Godwin (University of Kansas Cancer Center, Kansas City, KS). The established cell lines were validated by short tandem repeat analyses. For 2D culture, cells were trypsinized and plated in the adherent cell culture–treated plate. Single-cell organoids were grown using DMEM containing rhEGF (20ng/mL), rhFGF (20ng/mL), 1× B27, and heparin calcium salt (0.4%) from Thermo Fisher Scientific) in ultralow attachment plates (Corning). All the compounds were obtained from Tocris Bioscience or Sigma-Aldrich. Bevacizumab was obtained from Bio-Rad.

Peripheral blood mononuclear cells (PBMC) were isolated from fresh blood using Ficoll-Paque PLUS (Thermo Fisher Scientific). Briefly, blood was diluted 1:1 ratio in PBS, overlay on top of Ficoll–Paque, and centrifuged at 400 × g for 30 minutes at room temperature. The PBMCs were collected from the Ficoll–Plasma interface, washed twice with PBS at 300 ×g at 4°C, and then labeled with PKH26 dye as described previously (6). Fluorescence Z-stack images were captured with a Nikon Eclipse Ti Epi-fluorescent microscope connected with a Nikon intensilight C-HGFIE (Nikon) with ProScan H31 controller (Prior).

Deidentified tissue explants were collected with informed written consent from patients under the auspices of the University of Kansas Medical Center Biospecimen Repository Core Facility (Kansas City, KS). All protocols for collection of sample were in accordance with recognized ethical guidelines (Declaration of Helsinki, Belmont Report, and U.S. Common Rule) by the Human Subject Committee at the University of Kansas Medical Center (Kansas City, KS). Patient characteristics are listed in Supplementary Table S1. Briefly, tumor and ascitic fluid samples from 4 patients were used in this study. Primary colon cancer cells were derived from a female (hCol1) and a male (hCol2) patient. The patient was administered a combination of avastin/bevacizumab, 5-fluorouracil (5-FU), leucovorin, irinotecan capecitabine, and oxaliplatin. Tissue from two patients with ovarian cancer was acquired in the study. The hOVR1 cultures were derived from ascites fluid from a patient with clear cell adenocarcinoma of the ovary. The patient was treated with taxol and cisplatin. Unfortunately, the patient had progressive disease and the chemotherapeutic agent was changed to doxorubicin. The hOVR2 cancer cells were cultured from ascites fluid from a patient with adenocarcinoma of the ovary. The patient was treated with taxol and carboplatin, but recurred and was treated with the same regimen.

For human colon cancer tissue, tissue was washed with sterile PBS and minced into small pieces. Minced tissue pieces were incubated with collagenase (type IV, Sigma-Aldrich), for 10–20 minutes. The resultant cell suspension was passed through a fine fire-polished glass pipette to generate a single-cell suspension. Cells were subjected to centrifugation at 3,000 rpm for 2 minutes. The cell pellet was suspended in specialized media [spMedia, DMEM containing 20% FBS, hydrocortisone (1 μg/mL), rhFGF (5 ng/mL), rhEGF (5 ng/mL), VEGF (5 ng/mL), R3-IGF-1 (15 ng/mL), ascorbic acid (50 μg/mL), 2% FBS, and 1% Gentamicin, Amphotericin-B]. Human ovarian cancer cells from ascitic fluid were cultured in CTC Enrichment cocktail containing anti-CD36 antibody (STEMCELL Technologies) according to the manufacturer's instructions. Briefly, ascitic fluid was incubated with antibody cocktail (RosetteSep CTC Enrichment Cocktail, STEMCELL Technologies) in the ratio of 50 μL/mL for 20 minutes at room temperature. The suspension was then diluted (1:1) with PBS containing 2% FBS and cells were separated using density gradient medium, Ficoll-Paque (GE Healthcare Life Sciences). Cells were washed twice in PBS containing 5% FBS at 300 × g for 10 minutes and cultured in spMedia.

Cell lines and patient-derived cells were grown in a humidified incubator at 37°C with 5% CO₂. GFP-overexpressing plasmids were generated using the pLVX Tet-On Advanced Plasmid System (ClonTech Laboratories). Lentiviral particles were generated using Lenti-X cells transfected with the pGIPZ set of packaging plasmids generously donated by Roy Jensen (University of Kansas Medical Center, Kansas City, KS). HCT116, SW480, and DLD1 cells were cotranduced with pLVX-Tet-On and pLVX-Tight plasmids followed by selection with 1 mg/mL G418 (Mediatech) and 2 μg/mL puromycin (Life Technologies) for 7 days. Cells were consistently maintained in 500 μg/mL G418 and 1 μg/mL puromycin following selection, unless otherwise noted.

For PLiD cultures, lung epithelial NL20 cells, lung fibroblast MRC5, HUVECs, and HLECs were mixed in equal proportions (5,000 cells/mL) and cultured in mTiD-spMedia-1 [DMEM containing hydrocortisone (0.5 μg/mL), rhFGF (3 ng/mL), rhEGF (3 ng/mL), VEGF (2 ng/mL), R3-IGF-1 (5 ng/mL), ascorbic acid (10 μg/mL), Extract P (0.1%), plasma protein fraction (0.125%), l-Alanyl-l-Glutamine (2.4 mmol/L), rhEGF (5 ng/mL), rhInsulin (2 μg/mL), 2% FBS, and 1% penicillin-streptomycin]. Single or admixed cultures were plated in ultralow attachment plates. For mTiD cultures, the cancer cells were mixed in equal proportions to lung epithelial NL20 cells, lung fibroblast MRC5, HUVECs, and HLECs. After 3–5 days in cultures, PLiD and mTiD organoids were treated with drug and imaged after 48–72 hours using light microscope (Jenco International) or immunofluorescence (Nikon Eclipse Ti Epi-fluorescent microscope connected with Nikon intensilight C-HGFIE with ProScan H31 controller). IC₅₀ values for the response to therapeutic agents were determined following flow cytometry analyses of the cultures.

Tumor tissue and patient-derived primary tumor cells were submitted to the Genomics Core facility at the Kansas University Medical Center (Kansas City, KS). Genomics Core facility performed exome interrogation using the TruSeq Exome Library Prep Kit (Illumina FC-150-1001). Following sequence data collection on the Illumina Compute server (iCompute), the raw base call files (.bcl) are converted to fastq files and demultiplexed into individual libraries. Data were uploaded to a secure FTP site and sequencing data were analyzed by Bioinformatic core facility at Kansas University Medical Center (Kansas City, KS).

For transmission electron microscopy (TEM), tissue or organoid samples were fixed and then dehydrated in a graded series of ethanol for 10 minutes each. Samples were then placed into propylene oxide for 20 minutes followed by a solution of 1:1 propylene oxide:Embed 812 resin (Electron Microscopy Sciences) overnight. Samples were then placed into fresh Embed 812 medium formulation resin mixture for 1 hour. Samples were transferred to molds with fresh resin and cured in a 65°C oven overnight. Sections were cut and imaged on a JEOL. JEM-1400 TEM.

For scanning electron microscope, organoids were fixed and dehydrated in graded series of ethanol for 5 minutes/step. After being coated with gold, the organoids were viewed in a Hitachi S-2700 scanning electron microscope (Hitachi).

Exosomes were isolated from HCT116 cell culture medium using differential centrifugation method (7). Briefly, HCT116 were cultured in DMEM containing 10% exosome-depleted FBS (Thermo Fisher Scientific). The culture medium was collected and centrifuged at 3,000 × g for 10 minutes to remove all the floating cells and cell debris. Supernatant was collected and centrifuged at 10,000 × g for 30 minutes at 4°C (Sorvall High Speed Centrifuge, Thermo Fisher Scientific). Exosome-containing supernatant was subsequently centrifuged for 60 minutes at 100,000 × g at 4°C (Sorvall Ultracentrifuge, Thermo Fisher Scientific). The isolated exosomes were washed twice by resuspending in 5 mL PBS and recentrifuged at 100,000 × g for 60 minutes. Then the pellet was resuspended in 200 μL of PBS and protein was measured using Bio-Rad Protein Assay Kit.

Hexosaminidase assay was used to study the effects of various drugs including 5-FU, oxaliplatin, irinotecan, SN38, cisplatin, and carboplatin on proliferation of cancer cells. Cell proliferation was assessed as percent proliferation rate = [(A/B) × 100], where A and B are the absorbance of treated and control cells, respectively. Half maximal inhibitory concentration (IC₅₀) was obtained by 50% of cell death after 72 hours of treatment by using GraphPad PRISM software (GraphPad Software).

3D spheroid formation assay was performed as described previously (8). Briefly, cells were plated in ultralow attachment plates (Corning) at a density of 5,000 cells/mL in DMEM supplemented with rhEGF, rhFGF, and 2% B27 (Life Technologies) and 1% antibiotic–antimycotic (Mediatech) at 37°C in a humidified atmosphere containing 5% CO₂. For determining therapeutic effects, the spheroids were treated with 1×, 5×, 10×, and 25× IC₅₀ of 2D cultures and imaged after 3–5 days.

Organoids were fixed, paraffin embedded, and sectioned. The sections were deparaffinized and stained using mouse anti-E-cadherin (610181, BD Biosciences), mouse anti-vimentin (sc-6260, Santa Cruz Biotechnology), rabbit anti-CD31 (PA5-16301, Thermo Fisher Scientific), or rabbit anti-LYVE1 (ab14917, Abcam). Secondary antibodies donkey anti-mouse/rabbit tagged with FITC (Jackson ImmunoResearch) or Alexa 488 (Thermo Fisher Scientific). For hypoxia detection, organoids were incubated with 100 μmol/L hypoxia probe-1 (pimonidazole hydrochloride) for 2 hours at 37°C, as per the company protocol (Hypoxyprobe-1 Kit, Hypoxyprobe) and fixed, stained using anti-pimonidazole mouse antibody followed by secondary antibody anti-mouse Dylight 549. Images were captured using Nikon Eclipse Ti microscope using NIS element software.

Western blotting was carried out as described previously (8). Briefly, mTiD/PLiD lysates were prepared and quantified using Bio-Rad Protein Assay Kit (Thermo Fisher Scientific). Cell lysates were subjected to SDS-PAGE and blotted onto polyvinylidene difluoride membrane (EMD Millipore), blocked, incubated with primary antibodies [mouse anti-E-cadherin (610181, BD Biosciences), mouse anti-vimentin (sc-6260, Santa Cruz Biotechnology), rabbit anti-CD31 (PA5-16301, Thermo Fisher Scientific), rabbit anti-LYVE1 (ab14917, Abcam), mouse anti-surfactant protein (SP-B, ab3282, or SP-D, ab1778 Abcam), rabbit anti-carcinoembryonic antigen (CEA; ab135781, Abcam), a mouse anti-MUC16 (CA125, ab10029, Abcam), or mouse anti-GAPDH (sc-365062, Santa Cruz Biotechnology)], and then incubated with appropriate sheep anti-mouse secondary antibodies (NA911V, GE Healthcare Life Sciences) or donkey anti-rabbit (NA934V, GE Healthcare Life Sciences) tagged with horseradish peroxidase. Proteins of interest were detected using Enhanced Chemiluminescence System (RPN2106, GE Healthcare Life Sciences or 34095, Thermo Fisher Scientific) and imaged on ChemiDoc XRS with Image lab software (Bio-Rad).

Cells were dissociated from mTiD culture by trypsin/EDTA for 15–30 minutes at 37°C and centrifuged at 2,000 rpm for 2 minutes. The pellet was washed twice with PBS containing 2% FBS. Percentage of GFP- or mCherry-positive cells were determined using the Accuri C6 flow cytometer and analyzed by using CFlow Plus software (BD Biosciences). For IC₅₀ calculations, a plot between the drug concentration versus 50% reduction in GFP-positive cells were calculated using GraphPad Prism software.

The levels of carcinoembryonic antigen (CEA) and cancer antigen 125, CA125 (MUC 16) were detected using a chemiluminescence kit (ABNOVA Corporation) according to the manufacturer's instructions. Briefly, mTiD supernatant was centrifuged at 2,000 × g for 10 minutes to remove debris. Supernatants (50 μL) were incubated for 2 hours at 37°C in antibody-coated wells. Chemiluminescence was detected in a microplate reader (BioTek Synergy Neo) using the Gen5 software. Percentage of chemiluminescence in test wells was calculated relative to control.

Student t tests were used to calculate statistical significance between treatment groups, and data are reported as mean ± SEM. Significance between test groups was determined by P < 0.05. Statistical calculations were performed using Microsoft Office Software (version 2013).

Following regional lymph nodes, the lung represents the most frequent site of metastasis for solid tumors (9, 10). An in vitro model that mirrors the complex microenvironment and architecture of the human lung would enable an accurate assessment of the biology of tumor metastasis to the lung and also facilitate the development of therapeutic agents that specifically abrogate lung metastasis. Furthermore, an in vitro lung model would be useful for high-throughput screening of effective therapeutic agents. Toward this, we systematically developed culture conditions to grow normal human lung epithelial NL20 cells with other major cell types in the lung including fibroblasts (MrC5), lymphatic (HLEC) and blood vessel (HUVEC) endothelial cells. While individual cell types aggregated to form small spherical clusters, the cells when put together formed a multicell-type robust organoid that we named PLiD (Fig. 1A and B; Supplementary Fig. S1A and S1B). To demonstrate the persistence of each individual cell type in the multicellular PLiD organoids, we performed immunofluorescence for molecular markers. Epithelial cells make up the majority of PLiD as indicated by the epithelial marker E-cadherin (Fig. 1C). Fibroblasts were also present and dispersed throughout the organoid as indicated by the vimentin staining. It is of interest that a higher concentration of vimentin is seen at the periphery of the organoid, which is consistent with stromal presence in the region for matrix deposition and increased stability of the organoid. Blood endothelial cells were seen at the core and edge of the organoids as marked by CD31 staining. LYVE1 staining in the periphery of the organoids also demonstrated the presence of lymphatic endothelial cells (Fig. 1C). To confirm that the different cells were present in the organoid, we performed Western blot analyses, which showed markers for epithelial cells (E-cadherin), fibroblasts (vimentin), HUVECs (CD31), and lymphatic endothelial cells (LYVE1; Fig. 1D).

Scanning electron microscopic analyses of the PLiD reveals an uneven surface architecture with irregular bulging segments, which are typical of surface epithelial cells in the lung (Fig. 1E; refs. 11, 12). TEM micrographs of ultrathin organoid sections at 800× magnification show a number of projections including surface blebs and short microvilli, which are protruding into the luminal spaces (Fig. 1F). TEM images also show unusual cytoplasmic features with the presence of numerous multilamellar bodies with parallel membranes (Fig. 1G). Pulmonary lamellar bodies store surfactant proteins that are processed in the rough endoplasmic reticulum and the Golgi apparatus (13). Some of the lamellar bodies contain small vesicles. These organelles are implicated in the secretion of surfactant proteins that spread over the alveolar surfaces providing an extracellular coating. Pulmonary surfactant proteins secreted by epithelial cells consist of phospholipids, glycosaminoglycans, and proteins that together reduce the surface tension allowing the exchange of gases and material during inspiration (14).

As mentioned above, lamellar bodies produced by the epithelial cells in the lung store multiple surfactant proteins, of which SP-B and SP-C are essential for lowering the surface tension at the air/liquid interface within the alveoli of the lung (15). As a surrogate for lung function, we determined surfactant proteins' expression. Western blotting of PLiD lysates demonstrated the presence of SP-B and SP-D, which were at higher levels in the PLiD organoids than 2D cultures of NL20 cells, further demonstrating that the PLiD organoid is functionally similar to the human lung (Fig. 1H).

Previous studies have demonstrated that PBMCs can infiltrate 3D cultures of human non–small cell lung cancer (NSCLC) cell lines, grown alone or with a human fibroblast cell line (16). To determine whether the PLiD cultures are amenable to PBMC infiltration, we labeled PBMCs with PKH26 dye and then added them to PLiD cultures. Immunofluorescence image, taken after 3 days, shows that the labeled cells are within the lung organoid (Supplementary Fig. S1C, top). To determine the depths to which the cells have penetrated, we took images at 50-μm intervals to show penetration of PBMCs through the entirety of the PLiD spheroids (Supplementary Fig. S1C, bottom).

Currently, no models are available to efficiently study lung colonization of metastatic cells. The only available method to study lung colonization is an animal model, where cancer cells are injected into the tail vein and subsequent growth in the lungs is measured. Although more reliable, animal experiments are expensive and do not take into consideration the complexity of multiple human cell types interacting in a concerted manner with the tumor (17). To circumvent this critical issue, we developed a mTiD model that recreates the tumor microenvironment and allows for medium- to high-throughput screening of therapeutic agents (Supplementary Fig. S1D). We seeded colon carcinoma cells along with the PLiD cells in nonadherent dishes to generate mTiD organoids. Histologic examination of the mTiD organoids looks similar to that of colon cancer metastasis to the lung (Fig. 2A). Focal central necrosis is a common feature of tumors with blood supply that is primarily on the periphery of the tumor (18). We observed central necrosis in the mTiD cultures from two colon cancer cell lines due to a hypoxic core as assessed by hypoxia tracer hypoxyprobe-Red549 (Fig. 2B). To further confirm that the endothelial cells are present in the mTiD, we generated the tumor organoids using fluorescently labeled endothelial cells, along with three colon carcinoma cell lines. An even distribution of GFP-labeled colon carcinoma cells was observed throughout the organoid (Fig. 2C). In discrete regions interspersed between the carcinoma cells were vessel-like tubule structures of mCherry (red) labeled endothelial cells. Organoids were dissociated and single-cell suspensions were assessed by flow cytometry to determine the percentage of carcinoma and endothelial cells in the organoids. We found that close to 50% of the organoid consisted of carcinoma cells and 9%–11% of endothelial cells (Fig. 2D). Western blotting analyses revealed that all the cell types seeded were represented within the organoids and expressed the individual markers, namely E-cadherin for epithelial cells, vimentin for fibroblasts, CD31 and LYVE1 for blood and lymph endothelial cells, respectively, and SP-B and SP-D for lung epithelial cells (Fig. 2E).

We next determined whether the cancer cells can colonize PLiD, for which we first allowed PLiD to develop, and subsequently added GFP-labeled HCT116 cancer cells. We obtained images at various timepoints after addition of the GFP-labeled cells in PLiD culture. At 12 hours, we observed one cell to be attached to the PLiD, which formed a small colony after 3 days and grew to be a larger GFP-positive HCT116 colony after 5 days (Fig. 3A). Because angiogenesis is important for tumor growth, we next determined the effect of the cancer cell colonization in the lung organoid on blood vessel formation. IHC staining of sections from the PLiD shows a single CD31-positive blood vessel–like structure in the center of PLiD and some positive staining in periphery. However, in the mTiD, there were multiple branches of CD31-positive cells, suggesting that the cancer cells enhanced the angiogenic process in the mTiD model organoid (Fig. 3B).

Exosomes are small membrane-bound vesicles that carry various cargo including nucleic acids such as DNA, mRNA, miRNA, and proteins. Exosomes are involved in intercellular signaling and epithelial-mesenchymal transition (EMT), resulting in cancer cells presenting a more aggressive phenotype (19–21). Recent studies have demonstrated that exosomes isolated from highly metastatic murine breast cancer cells distributed predominantly to the lung, and enhanced breast cancer cell metastases (22). To determine whether cancer cell colonization of PLiD is affected by exosomes, we pretreated PLiD with exosomes secreted by HCT116 colon cancer cells grown in 2D culture. Subsequently, we added HCT116 cells to the PLiD and allowed the cells to colonize for 3 days. We observed significantly bigger and higher number of colonization following exosome treatment, suggesting that the exosome sufficiently modulate the lung to enhance the colonization process (Fig. 3C).

Current in vitro screening methods are poorly predictive of the clinical efficacy of drug candidates. Also, as mentioned above, animal experiments are expensive and not appropriate to study human cancer cell response to therapeutics. It is also well known that the stromal cells modulate the response of tumor cells to therapeutic agents (23). We therefore determined whether the mTiD model would be efficient for studying therapeutic response of cancer cells. First, we tested the effect of the anti-VEGF antibody bevacizumab because VEGF plays an important role in tumor growth and metastasis (24, 25). Moreover, targeting VEGF was considered to be good therapeutic option for patients with metastatic colorectal cancer, especially in combination with current therapeutic agents (26, 27). We treated the mTiD model containing GFP-labeled colon cancer cells DLD1 and SW480 with increasing concentrations of bevacizumab. There was a dose-dependent decrease in GFP-labeled cells in mTiD upon treatment (Fig. 4). This suggests that the model is amenable to testing biological agents, which may include immune therapy agents.

Because patients are administered various combinations of agents including 5-FU, oxaliplatin, irinotecan, and leucovorin, we tested the antitumor efficacy of these compounds in the mTiD model. We observed significant inhibition in growth of the three colon cancer cell lines HCT116, SW480, and DLD1 when treated with either irinotecan or SN38, an active metabolite of irinotecan (Fig. 5A; Supplementary Table S2A).

The most striking observation was that HCT116 cells did not respond to 5-FU while DLD1 cells did not respond to oxaliplatin in the mTiD model, with IC₅₀s >25 μmol/L (Fig. 5A). However, in 2D single-cell–type cultures, 5-FU suppressed HCT116 cell growth with IC₅₀ of 13.2 ± 2.08 μmol/L, and oxaliplatin suppressed DLD1 cells with an IC₅₀ of 7.4 μmol/L ± 3.05 (Fig. 5B; Supplementary Table S2B). We also performed studies on single-cell–type 3D cultures of HCT116 (Fig. 5C), SW480 (Fig. 5D) and DLD1 (Fig. 5E). Interestingly, the 3D cultures were not susceptible at these doses. Much higher dose of compounds was required to observe some effect in the 3D single-cell–type cultures. In addition, irinotecan did not affect the cells in the 3D cultures, although its active metabolite SN38 showed significant efficacy. This suggests that the poor response to irinotecan in 3D single-cell–type cultures may be due to poor metabolism of the drug to its active component. These data further suggest that the mTiD model is the most appropriate model for testing drugs because it is a better representative of tumor physiology than the 3D single-cell–type spheroids.

To determine whether PLiD cultures has utility for assessing toxicity of therapeutics to the lung, we evaluated the dose response of PLiD cultures to increasing concentrations of 5-FU, cisplatin, carboplatin, oxaliplatin, irinotecan, and SN38. PLiD cultures demonstrated reduced spheroid sizes or dispersed spheroids only at very high concentrations of these compounds suggesting that the drugs are not directly affecting the cultures at pharmacologic doses (Supplementary Fig. S2).

To further assess the effectiveness of the mTiD platform in other tumor types, we also performed extensive testing of ovarian cancer cell lines. We grew A2780 human ovarian cancer cell line and its cisplatin-resistant variant C30 in 2D and mTiD cultures. Both cells were efficiently taken up in the mTiD cultures (Fig. 5F). We next determined the effect of 5-FU, cisplatin, oxaliplatin, and carboplatin in 2D and mTiD cultures. Specifically, we saw a striking difference in drug response in the mTiD cultures compared with 2D cultures. While the parent A2780 cells showed IC₅₀ values between 24.6 and 53.2 μmol/L, the C30 cells had IC₅₀ values of >100 μmol/L in mTiD cultures (Fig. 5G; Supplementary Table S2C). However, both the parent A2870 and its C30 derivative demonstrated lower IC₅₀ values in 2D cultures (Fig. 5H; Supplementary Table S2D). These data again confirm that the responses to drug treatment in the mTiD model, but not 2D or 3D cultures, replicates what is observed in humans.

Finally, we tested the ability of additional cancer cell types including glioblastoma (U251), medulloblastoma (DAOY), and breast cancer (MCF7) to form mTiD cultures (Supplementary Fig. S3A–S3C). In addition, we tested the response of the mTID cultures to 5-FU because it is a standard of therapy for these diseases. While glioblastoma cells in the mTiD are responsive to 5-FU, medulloblastoma and breast cancer cells in the mTiD cultures were resistant (Supplementary Fig. S3A–S3C). We also tested these cancer cells in the mTiD with honokiol, a natural compound found in the bark of the magnolia tree. All the cultures were sensitive to the compound. Finally, melanoma cells are known to frequently metastasize to the lung. Moreover, patients who are treated with vemurafenib, the first-line treatment for this indication, develop resistance. We tested the response of paired vemurafenib-sensitive and -resistant A2058 melanoma cells to honokiol. We observed that honokiol treatment effectively reduced growth of both vemurafenib-sensitive and -resistant cells (Supplementary Fig. S3D).

Developing potential strategies for preventing tumor metastasis has been somewhat difficult study, in part, due to the lack of adequate models of disease. Currently, the only methods to study would be using mouse models, but this model may not work for every cancer type because of lack of disease models for metastatic research. A second point to consider is whether the cancer cells that have disseminated from the primary tumor site are able to colonize and grow in the metastatic site. Given that the mTiD model is designed to study colonization and growth of disseminated cells, we performed two different experiments that would replicate a prevention setting. In the first experiment, we treated PLiD with chemotherapeutic agents (5-FU, irinotecan, and oxaliplatin) at IC₅₀ values for 24 hours and then added mCherry-labeled cancer cells (Supplementary Fig. S4A). We observed significantly lower numbers of cancer cells colonizing the PLiD in the treated samples when compared with controls, suggesting that pretreatment of PLiD can suppress tumor cell colonization. In a second experiment, we first treated mCherry-expressing cancer cells with the drugs at the IC₅₀ dose for 24 hours and then added these cells to the PLiD (Supplementary Fig. S4B). Again, there was significantly lower number of cancer cells colonizing the PLiD, suggesting that drug treatment suppresses colonization. Taken together, these data suggest that the mTiD system would be an effective method to determine mechanisms to prevent cancer cell colonization.

Currently, there are no standard chemotherapeutic agents for metastatic colon cancer. As mentioned above, patients are administered combinations of several agents. Also, there are no tests or biomarkers that can effectively predict the sensitivity of patients to the various chemotherapeutic agents. We used the mTiD platform to test the efficacy of various chemotherapeutic agents in two primary colon carcinoma lines (hCol1 and hCol2). The patient characteristics are listed in Supplementary Table S1. We used whole-genome exome sequencing on hCol2 primary tumor and matched cells in culture to determine the genomic variance. Available paraffin-embedded tissue samples from hCOL1 did not yield the necessary quality of RNA for sequencing analyses. Despite cellular heterogeneity, the primary tumor had 64.96% homology in the mutation profile compared with the matched colon cancer cells growing in culture, demonstrating that the primary cells are similar to the patient samples (Supplementary Fig. S5; Supplementary Table S3). Again, the GFP-labeled patient-derived primary cells were able to colonize the lung and form metastatic tumors (Fig. 6A). We then tested the antitumor efficacy of 5-FU, irinotecan, oxaliplatin, and SN38 in the primary patient-derived mTiD cultures. In these cultures, both hCol1 and hCol2 were resistant to oxaliplatin with IC₅₀s >25 μmol/L (Fig. 6B; Supplementary Table S4). In addition, the hCol1 mTiDs were resistant to 5-FU and irinotecan. The hCol2 mTiD was sensitive to irinotecan with an IC₅₀ of 0.3 ± 0.07 μmol/L. Both hCol1 and hCol2 were sensitive to SN38 with IC₅₀ values of 0.45 ± 0.19 μmol/L and 0.075 ± 0.01 μmol/L, respectively. In 2D cultures, both hCol1 and hCol2 were sensitive to all the agents with IC₅₀ values ranging from 0.05 ± 0.01 μmol/L – 9 ± 1.12 μmol/L (Supplementary Table S4; Supplementary Fig. S6A). Moreover, when cancer cells were grown as single-cell–type 3D spheroids, we needed much higher concentrations of drug than what was required for 2D or mTiD cultures, further suggesting that this method of growing and testing culture is not appropriate (Supplementary Fig. S6B).

To validate the sensitivity of the mTiD cultures to chemotherapeutic agents, we assessed the levels of CEA in the cell lysates and supernatants of primary hCol mTiD cultures. The hCol2 mTiD cells were sensitive to 5-FU, irinotecan, and SN38 and demonstrated a corresponding reduction in CEA levels in both the cell lysates and supernatant (Fig. 6C and D). Furthermore, hCol2 mTiD cells demonstrated oxaliplatin resistance, and a corresponding lack of change in CEA levels in both the cell lysate and supernatant. To further validate these findings, we used the chemiluminescent immunoassay to measure the levels of CEA in cell supernatants. Corroborating our cell viability data, hCol1 cells that were sensitive to only SN38, showed a corresponding 50% reduction in CEA levels with SN38 treatment but a 35%–40% reduction in CEA levels on treatment with 5-FU, irinotecan, and oxaliplatin, respectively (Fig. 6E). Similarly, hCol2 cells showed >60% reduction in CEA levels on 5-FU, irinotecan, and SN38 treatment but only 20% reduction in CEA levels when treated with oxaliplatin.

To further confirm that the mTiD model recapitulated the patient's response to therapy, we interrogated serum from the patient whose tumor was used to derive the hCol2 line. The hCol1 patient was not treated with chemotherapy because of severe toxicity. The hCol2 patient was treated with a combination of avastin, FOLFIRI (irinotecan, 5-FU, and leucovorin), capecitabine, and oxaliplatin. Serum levels of CEA went down from 100 to 3 ng/mL postchemotherapy and stayed low even 200 days postsurgery (Fig. 6F). This finding is consistent with the data from the mTiD platform, which effectively demonstrated the sensitivity of the patient to FOLFIRI as determined by the cell survival and CEA levels. Thus, mTiD may be an effective platform to predict the sensitivity of patients undergoing colon cancer chemotherapy.

To further assess the effectiveness of the mTiD platform in a second tumor type, we tested ovarian cancer cell lines and ovarian cancer cells from ascitic fluid for the propensity to colonize the lung. For this, we isolated ovarian cancer cells from ascites fluid of two patients with ovarian cancer (hOVR1 and hOVR2). The patient characteristics are listed in Supplementary Table S1. As above with colon cancers, we performed whole-genome exome sequencing on hOVR1 ascites fluid and matched cells in culture. Despite cellular heterogeneity, analyses of mutation profiles showed 98.41% homology between DNA isolated from the ascites fluid and the matched cancer cells growing in culture (Supplementary Fig. S5; Supplementary Table S3). We first tested two primary ovarian carcinoma cells for growth in PLiD. Our data demonstrate that GFP-labeled primary metastatic ovarian cancer cells grow well in the mTiD model with distribution throughout the organoid (Fig. 7A). Furthermore, flow cytometry analyses revealed that organoids contained 25%–27% GFP-labeled tumor cells. We next checked the sensitivity of 2D- and mTiD-hOVR1 and hOVR2 cultures to 5-FU, irinotecan, oxaliplatin, cisplatin, and carboplatin. In 2D cultures, hOVR1 and hOVR2 cells were sensitive to 5-FU, oxaliplatin, irinotecan, and cisplatin with IC₅₀ values ranging from 1 to 18 μmol/L (Supplementary Fig. S7A; Supplementary Table S5). Again, similar to that seen with colon cancer cells above, when cells were grown as single-cell–type 3D spheroids, the levels of drug required for cytotoxicity was much higher than what was observed in 2D cultures, ranging from 15 to 212.5 μmol/L (Supplementary Fig. S7B and S7C). In mTiD cultures, hOVR1 was sensitive to 5-FU, irinotecan, oxaliplatin, and cisplatin with IC₅₀ values of 10, 4, 4.25, and 10.25 μmol/L, respectively (Fig. 7B; Supplementary Table S5). In mTiD cultures, hOVR2 was sensitive to 5-FU and irinotecan at IC₅₀ values of 10 and 4 μmol/L, respectively. However, hOVR2 cells were resistant to oxaliplatin (IC₅₀ > 25 μmol/L) and cisplatin (IC₅₀ >25 μmol/L).

The current choice of first-line chemotherapy in patients with ovarian cancer is not based on rational design. The ability to predict response of patients to specific chemotherapeutics would facilitate the rapid and effective treatment of patients with cancer. Generally, the patients receive cisplatin or carboplatin, and they generally appear to respond to the drug. However, a majority of the cases have reemergence of the tumor, and these are now resistant to the platin drugs. In our studies, hOVR1 is from a patient that had never been treated with platin drugs (carboplatin naïve), while hOVR2 is from a patient that is resistant to carboplatin. To confirm that our OVR mTiD cultures replicate this, we tested them for sensitivity to carboplatin. Our data demonstrate that hOVR1 was sensitive to carboplatin at an IC₅₀ value of 30 μmol/L while hOVR2 was resistant with an IC₅₀ value of >25 μmol/L (Fig. 7C). We assessed for secreted CA125 levels from the two OVR mTiD cultures. The data demonstrate that carboplatin treatment of the naïve patient-derived hOVR1 culture had a corresponding 80% reduction in secreted CA125 levels when treated with various concentrations of carboplatin (Fig. 7D). The hOVR2 cultures on the other hand, demonstrated resistance to carboplatin in the mTiD cultures with a reduction in CA125 levels by only 30% at the highest carboplatin dose of 25 μmol/L. To determine the clinical relevance of these in vitro findings, we obtained the data from the clinical laboratory on CA125 levels in patient's serum beginning on the day of surgery (day 0) for up to 300 days. The hOVR1 patient was treated with taxol, cisplatin, and doxorubicin and demonstrated a reduction in CA125 levels with levels below 35 U/mL (Fig. 7E). The hOVR2 patient, in contrast, demonstrated an initial drop in CA125 levels postsurgery that were reduced 100 days postsurgery. The patient was administered a combination of taxol and carboplatin as first-line therapy. CA125 levels went up in the patient serum following relapse of the tumor, which was corroborated by the hOVR2-mTiD in vitro data (Fig. 7F).

While primary neoplasms generally have a favorable prognosis, metastatic spread is the major cause of cancer-related mortality, for which there are few therapeutic options. A major hurdle in the development of antimetastatic therapies is the dearth in effective preclinical models to study and to develop targeted therapies that prevent or treat metastatic disease. To date, 2D assays were the main stay for drug screening, in part, because they can be easily observed and growth measured. However, it has become increasingly apparent that this method lacks the in vivo characteristics seen in a tumor. Rigid surfaces used in 2D culture enhance proliferation but likely inhibit differentiation due to lack of cell–cell interactions (28, 29). In addition, these rigid surfaces result in flattened cell growth and changes in the cytoskeleton, which then changes nuclear shape, which has been shown to cause differences in gene expression and protein synthesis (30). Finally, previous studies corroborate our data demonstrating that tumor cell lines in 2D cultures may be more susceptible to therapeutic agents when compared with cells grown as 3D spheroids (31–33). However, the single cancer cell–type 3D spheroids do not replicate the in vivo setting as they only employ cancer cells and do not have stromal cells. Moreover, 3D cultures of just the cancer cells require significantly higher levels of drug to see effects, and this is not representative of what happens in the in vivo setting. Another pitfall of current 3D tumor models is that they are created with commercially available cell cultures that no longer reflect the pathology and chemoresistance patterns that are seen in vivo. It is also becoming increasingly apparent that the tumor microenvironment plays a significant role in cancer progression and response to therapy (34, 35). Although tumor cells alter the microenvironment to support their growth and metastasis (36–38), little is known about the tumor-associated stromal cells. The major hurdle in carrying out these studies has been the acquisition and establishment of primary stromal cells from patient samples.

In recent years, there has been a surge in technologies wherein cells can be cultured in ways that maintain their natural morphology and structure. It is for this reason that 3D cell culturing such as tumor organoids were developed, resulting in enhanced physiologic relevance of experiments performed in vitro. Because there is more cell-to-cell contact, and in return more intercellular signaling, cells develop more complex structures. This cell culture environment could improve the predictive accuracy of the drug discovery process and aid in the understanding of tissue morphogenesis (39, 40). Thus, while 2D culture is reproducible, it is a highly artificial and less physiologic environment, as some in vivo characteristics and traits are lost or compromised. In contrast, 3D cell culture is more physiologic with these traits better preserved (1).

The choice of the appropriate organoid model and culturing conditions is a crucial decision in the design of experiments with 3D cultures, as each has advantages and disadvantages. Currently, 3D tumor models can essentially be divided into scaffold-based and scaffold-free models (41). Scaffold-based 3D models are numerous and varied and can incorporate both natural and synthetic materials including collagen, fibrin, hyaluronic acid, silk, and alginate (42–45). These materials are easier to reliably reproduce but are not always physiologically relevant. Studies have also tried using cancer fibroblast–associated matrices, which may be more accurate in resembling the tumor microenvironment as compared with artificial scaffolds; however, these too do not recapitulate the tumor microenvironment in a metastatic setting (46). Another issue is that even with fibroblast-derived matrices, the cultures do not represent the metastatic tumor in the lung (47).

In general, scaffold-free systems may better represent human lung metastatic tumor, wherein the cells form their own extracellular matrix. However, in most cases, only single-cell–type 3D spheroids have been studied. Nevertheless, these methods have provided us invaluable insights into cancer stem cell biology and differentiation as well as modeling tumor hypoxia, given that the innermost cells have limited oxygen exposure (48). Tumor organoids are well established in basic cancer research and experimental therapeutics, and have shown resistance to chemotherapy-induced apoptosis when compared with identical cells in monolayers (3). But it is clear that the versatility of organoids offers an advantage over other models to further understand cancer biology.

In our study, we have developed a new in vitro lung organoid, PLiD, which contains all the normal lung cells including epithelial cells, fibroblast, and endothelial cells. The data establishes expression of E-cadherin (epithelial cells), vimentin (fibroblast), CD31 (Heme-Endothelial cells), and LYVE1 (lymphatic endothelial cells) in PLiD using immunocytochemistry, indicating the presence of multiple viable cell types in the organoid. Furthermore, this PLiD model demonstrates significant expression of surfactant proteins (B and D), an indication of functional lung tissue. Expression of surfactant protein was seen only in PLiD model, and not in cells grown individually. Our studies further confirm that PLiD cultures may be an excellent platform to study the final process of metastasis namely colonization. Current in vitro methods to study colonization use artificial matrices or bioengineered 3D platforms (49, 50). However, while many such matrices may be useful to study cancer cell growth and migration (51), they have not been amenable to study angiogenesis. In the PLiD system, we have observed angiogenesis after colonization by the cancer cells. We planned to test both HUVECs and HLMVECs in our model. Initially, we tried both types of cells, but the HMVECs failed to thrive in the PLiD model, which is a limitation. However, HUVEC cells worked really well in the model. Given that these cells are robust lines to study growth and tube-forming ability, and they thrived in the model, we carried out all studies with this line. In addition to developing studies to understand the role of angiogenesis in cancer cell colonization, the system may be useful in learning about the mechanisms by which cancer cells induce the angiogenic process. In addition, given that pretreatment of PLiD with cancer cell–derived exosomes enhanced cancer cell colonization of the lung, this further underscores the role of this model to study the effect of circulating factors and those in the microenvironment cancer cell colonization.

Another utility for PLiD/mTiD is in drug development and assessing toxicity. While all our current studies are performed with cells in traditional 2D and more recent 3D multicellular spheroid systems that may include fibroblasts, none of them replicate the organ in the body. Hence, after the preliminary cell culture work, one has to rely on studies in animals such as mice. There are two problems with this approach: the expenditure involved in maintaining the animals and a second, more ethical dilemma of having to use animals for such discovery studies because of issues such as distress. Moreover, animal tissues do not always replicate what is seen in the humans. With the development of the PLiD/mTiD system, we now have an effective platform to assess the ability of primary tumor and established cell lines to colonize the lung and further assess response to chemotherapy. Moreover, for those in the drug development field, this system would be an excellent intermediate between current 2D and 3D cultures and animal studies. The system can also be used to determine the role of immune cells in therapeutic responses. Finally, not only can the new compounds and biological agents be tested against various mTiD models (developed with different cancer cells), but also on PLiD to determine issues related to toxicity. Adverse drug reactions are a common form of injury and the lungs are a frequent target. However, current toxicity testing involves cell lines in traditional 2D cultures or in a chip (52). However, none of these replicate the organ. Hence, given the lung architecture seen in PLiD, it may be an excellent model to study drug-related toxicity. In this regard, all our studies presented in this article with the various drugs also included testing the compounds in PLiD to ensure that the dose of drug used to kill the cancer cells in mTiD was not also affecting the lung cells in PLiD. Finally, the possibility exists that the model can be used to study other lung-related diseases such as cystic fibrosis and infections.

The aim of precision medicine is to identify patients with tumors that are most likely to respond to specific chemotherapeutic agents. We and others have demonstrated that the response of primary tumor cells to chemotherapy varies in 2D versus 3D cultures (53). However, no one has been able to correlate this to patient response to the chemotherapeutic agent. One of the strengths of our study is that we have successfully cultured primary patient-derived cells ex vivo in the mTiD system, and the responses to the clinically relevant therapeutics observed in the mTiD organoids closely resembled that seen in the patients. However, while this is the first presentation of the model, we did not have access to metastatic lung tumors to compare with the mTiD model, because patients with metastatic disease are not surgically treated. Nevertheless, further studies are warranted to demonstrate that PLiD/mTiD model represents the in vivo patient tumor and its microenvironment in terms of genetic characteristics, which is a future direction of our studies. A matched primary analysis, such as global gene expression patterns or genomic mutations, would be valuable for comparing individual profiles but would inherently select for a subgroup of cancer cells and may not provide sufficient information about modeling intratumoral heterogeneity.

Our data suggest that the response of mTiD cultures from primary tumors may correlate with that of patients to chemotherapy. One of the limitations of our study is the sample size, and hence may not account for differences in therapeutic response of individual patient tumors. Hence, further studies with a large cohort of patient samples are required to confirm our findings. Another factor relates to intratumoral heterogeneity, something that has not been tested at this time in this mTiD model. It should be noted that not all tumors have intratumoral heterogeneity (54). Nevertheless, future studies are required with tumors known to have intratumoral heterogeneity and compare therapeutic response in the mTiD model to single-cell–type 3D cultures.

Patient-derived xenografts are being employed for drug screening (55). However, implanted tumors in mice need to be expanded into multiple mice prior to drug screening for sensitivity. This process is time consuming, cumbersome, and costly. Moreover, the xenografts have significant infiltration of mouse cells further complicating understanding of the results. On the other hand, the mTiD model effectively overcomes these limitations in that, the cells are in 3D in the context of the human lung microenvironment, thereby allowing for the screening of drugs that would effectively target lung metastatic disease. Thus, the mTiD platform can be used as a tool to inform decision pertaining to the selection of appropriate chemotherapeutic agents for individual patients depending on their drug sensitivity profile.

G. Sitta Sittampalam is a consultant/advisory board member for Indiana Bioscience Research Institute. No potential conflicts of interest were disclosed by the other authors.

Conception and design: P. Ramamoorthy, S.M. Thomas, K.M. Chastain, A. Dhar, S. Ramalingam, S.J. Weir, R.A. Jensen, G.S. Sittampalam, S. Anant

Development of methodology: P. Ramamoorthy, S.M. Thomas, G. Kaushik, D. Subramaniam, O. Tawfik

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): P. Ramamoorthy, K.M. Chastain, O. Tawfik

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): P. Ramamoorthy, S.M. Thomas, K.M. Chastain, O. Tawfik, W. Sun, S. Ramalingam, S. Gunewardena, S. Anant

Writing, review, and/or revision of the manuscript: P. Ramamoorthy, S.M. Thomas, D. Subramaniam, K.M. Chastain, A. Dhar, O. Tawfik, A. Kasi, W. Sun, S. Umar, J.M. Mammen, R.A. Jensen, G.S. Sittampalam, S. Anant

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): P. Ramamoorthy, S.M. Thomas, K.M. Chastain

Study supervision: S.M. Thomas, A. Dhar, S. Anant

Other (participated in intellectual discussion of the study): A. Kasi

Other (synthesis and structural characterization of compounds used for assay): S.B. Padhye

This work was supported by NIH grants CA182872 (S.M. Thomas), CA182872, CA135559, and CA109269 (S. Anant); the U.S. Department of Defense Breast Cancer Research Program (W81XWH-18-1-0031); the Kansas Bioscience Authority; the Thomas P. O'Sullivan Foundation; and grants from the Braden's Hope Foundation and Midwest Cancer Alliance. The authors would also like to acknowledge the support of the National Cancer Institute Cancer Center Support Grant P30CA168524, which supports the Biospecimen Repository Shared Resource that provided the deidentified patient samples. The authors thank the University of Kansas Medical Center–Genomics Core for generating the sequence datasets. The Genomics Core is supported by the University of Kansas School of Medicine, the Kansas Intellectual and Developmental Disability Research Center (NIH U54 HD090216), and the Molecular Regulation of Cell Development and Differentiation–COBRE (5P30GM122731-02).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.