Bioengineered Implantable Scaffolds as a Tool to Study Stromal-Derived Factors in Metastatic Cancer Models

Despite significant advances in targeted therapies against cancer, metastatic disease remains incurable (1). Paget's “seed and soil” hypothesis suggests that a receptive premetastatic microenvironment must evolve in order for tumor cells to engraft and proliferate at secondary sites (2, 3). Tumor cell–derived factors that enhance or direct metastasis to specific organs have been described (4–6). In contrast, proteins secreted by stromal cells within a metastatic niche are less readily identified, in part because of the difficulty in isolating and characterizing such cells even in mouse models. Such studies are particularly challenging in prostate and breast cancer, which metastasize to bone, a site that is not easily accessible for sampling.

A number of stromal-derived survival factors have been proposed on the basis of in vitro coculture experiments, although most remain to be validated in vivo (7, 8). No robust mouse models exist to test for metastasis-enhancing stromal-derived factors on a scale that would allow identification of novel pathways and testing of potential therapeutic suppressors of critical tumor/stromal interactions. Bioengineered scaffolds have been previously employed for the study of primary tumors (9) and, for bone metastasis in particular, engineered bone marrow–like structures have recently been described in the context of hematopoietic reconstitution (10–12), but their application to generating and studying blood-borne metastasis have not been extensively explored. Here, we present a bioengineered bone marrow–modeling scaffold, which can be implanted subcutaneously, monitored in situ through live imaging, and either serially transplanted in vivo or resected for detailed cellular and molecular analysis. Hematogenous seeding of the scaffold by orthotopically and systemically introduced tumor cells recapitulates the initiation of metastasis and allows molecular characterization of mouse-derived metastasis-associated stromal cells. As a functional validation of this method, we identify IL1β as a stromal-secreted cytokine that enhances the initiation of metastasis in two different cancer models, and whose suppression can be achieved through systemic administration of a receptor antagonist. Our data demonstrate the efficacy of a metastasis-capturing device in uncovering stromal signals and evaluating the effect of their modulation in vivo, thus enabling broad applications in the study of cancer metastasis.

All chemicals and supplies were purchased from Sigma Aldrich or Fisher Scientific unless otherwise stated. In vivo studies were performed in accordance with an animal protocol approved by the Massachusetts General Hospital (MGH) Subcommittee on Research Animal Care.

Scaffold design and bone marrow stromal cell seeding density were based on our previous studies that optimized these parameters (11). Specifically, we used a polyacrylamide hydrogel composed of 30% (w/w) acrylamide (monomer) and 5% (w/w) bis-acrylamide (crosslinker). Mechanical property of hydrogel scaffold measured by dynamic storage modulus was 18.3 ± 6.8 kPa. Cavity and junction diameters were about 250 and 65 μm, respectively. Half a million human bone marrow stromal cells were seeded per scaffold. Pore dimension and porosity of scaffolds are comparable with the marrow tissue formed within trabecular bones, which consist of 300 to 900 μm cavities. Mechanical stiffness of hydrogel scaffolds is approximately 50 times higher than reported central marrow stiffness (13). Although decreasing the polymer content of the hydrogel matrix could further reduce mechanical stiffness, it cannot support open porous three-dimensional (3D) structure during cell seeding and after subdermal implantation, which would in turn cause poor tissue development. Technical details about scaffold fabrication and 3D culture of primary human bone marrow stromal cells are provided in Supplementary Materials and Methods.

PC-3 cells (ATCC) were cultured in F-12K Medium (ATCC), while DU-145 (ATCC) and MDA-231 #1833 (a kind gift by J. Massagué, Memorial Sloan Kettering Cancer Center, New York, NY) were grown in DMEM (Gibco/Life Technologies), both supplemented with 10% FBS and 1% penicillin/streptomycin (Gibco/Life Technologies). All cell lines were obtained in 2011 and immediately expanded and frozen at early passages. When thawed for experimental use, they were never passaged for more than 4 months in culture. ATCC-derived cell lines were authenticated by the cell bank via short tandem repeat profiling. MDA-231 #1833 cell lines were validated in our laboratory for KRAS and BRAF mutation by Sanger sequencing. All cells were tested for mycoplasma contamination giving negative results. Stable Luc-GFP cell lines were generated using high titer lentivirus (Lenti Luc-GFP) as previously described (14).

For in vitro treatments, recombinant human IL1β (PeproTech) and IL1Ra (Anakinra) were resuspended as 100 μg/mL stocks in DMSO and used at a concentration of 100 ng/mL.

For in vivo experiments, IL1Ra (Anakinra) stock solution (100 mg/0.67 mL) was freshly diluted in PBS at the moment of use and administered daily by intraperitoneal injection at a concentration of 25 mg/kg.

Invasiveness and proliferation of PC-3 cells were assessed according to standard procedures with slight modifications. Details are provided in Supplementary Materials and Methods.

Six- to eight-week-old male NOD/SCID/IL2γ^null (NSG) mice (Jackson Laboratory) were used for xenograft studies. The dorsal side of mice was shaved under isofluorane anesthesia. Bone marrow stromal cells (BMSC)-coated scaffolds were washed with PBS (×3 times). After wiping the skin with alcohol swab, an incision (∼4 mm) was made to create a subcutaneous pocket, a scaffold was carefully inserted using forceps and the incision was closed with surgical staples (Roboz). Four scaffolds were implanted in each mouse. On the last day of the study, mice were sacrified and implanted scaffolds were carefully removed. For serial transplant, explanted scaffolds were subdermally reimplanted as described above.

Four BMSC-coated scaffolds were subcutaneously implanted in the back of each mouse as previously described. Four weeks later, orthotopic prostate tumors were generated by injecting 1 × 10⁶ PC-3 or DU-145 cells into the dorsal prostatic gland of NSG mice in a volume of 20 μL (1:1 ratio of PBS and Matrigel Basement Membrane Matrix, BD Biosciences). Similarly, experimental breast cancer metastases were generated by tail vein injection of 1 × 10⁶ MDA-231 #1833 cells in the same strain of immunocompromised mice. Animals were imaged weekly 5 minutes after intraperitoneal injection of 150 μL RediJect d-luciferin Ultra Bioluminescent Substrate (Caliper Life Sciences) using an IVIS Lumina II platform (Caliper Life Sciences) with constant settings throughout the entire experiment. Four or eight weeks later (breast and prostate cancer model, respectively), mice were sacrificed and scaffolds were explanted for either culture, histologic analysis, or serial transplantation into syngeneic tumor-free mice. Bone and visceral metastases were confirmed on dissected organs immediately after necropsy based on bioluminescence imaging (BLI) and GFP fluorescence. Bioluminescence in recipient mice was monitored weekly for up to eight more weeks as described above. For drug treatment, PC-3 tumor-bearing mice were randomized on the basis of primary tumor size and treated daily by intraperitoneal injection of 25 mg/kg IL1Ra or PBS vehicle (4 mice per group) for 2 weeks starting 6 weeks after establishment of the primary tumor, followed by scaffold transplantation as above. The investigator performing live imaging of transplanted mice (experiment readout) was blinded to the group allocation.

Before scaffold transplant, PC-3 tumor-bearing mice were anesthetized to perform cardiocentesis through the intercostal muscle of the left chest and approximately 1 mL of blood was drawn into a syringe primed with 100 μL of PBS with 10 mmol/L EDTA pH 7.4 (Gibco). Circulating tumor cells (CTC) were isolated by depleting the normal blood cell component using a microfluidic platform (^negCTC-iChip) as previously described (15). Cells were collected, spun down onto a microscope slide, and fixed with 4% paraformaldehyde. Samples were then permeabilized with 1% NP40 in PBS, blocked with 2% goat serum, and immunostained with chicken anti-GFP primary antibody (Abcam, ab13970) followed by donkey anti-chicken Dylight 488 secondary antibody (Jackson ImmunoResearch, 703-486-155). Slides were imaged under 10× magnification using the BioView Ltd automated imaging system, and hand validated to enumerate CTCs.

Explanted scaffolds recovered from tumor-bearing mice were cultured for up to 4 weeks in F-12K Medium (ATCC) or DMEM (Gibco/Life Technologies) + 10% FBS + 1% penicillin/streptomycin (Gibco/Life Technologies) and imaged weekly with an IVIS Lumina II platform (Caliper Life Sciences) after incubation in a 1:200 dilution of RediJect d-Luciferin Ultra Bioluminescent Substrate (Caliper Life Sciences) for 15 minutes at 37°C.

Seeded (met⁺) and unseeded (met⁻) scaffolds were minced with a scalpel and the cellular component was digested with collagenase IV in 1× HBSS (Gibco/Life Technologies) for 45 minutes at 37°C and filtered through a 70 μm cell strainer. After 1-minute incubation in ACK lysing buffer (Gibco/Life Technologies), cells were washed and resuspended in PBS–2% FBS solution. Cells were stained with anti-mouse CD45-Biotin (eBioscience, 13-0451-81, clone 30-F11) and anti-mouse CD31-Biotin (eBioscience, 13-0311-81, clone 390) followed by streptavidin–APC incubation (eBioscience, 17-4317-82). To avoid unwanted spillover from the very bright GFP signal into channels excited by the same laser and to reach maximum sorting stringency of pure stromal cells, we opted for a combination of CD45 and CD31 in the same channel (APC) using a different excitation laser. Cell sorting was performed on a four laser FACSAria II (BD Biosciences) based on endogenous GFP and antibody-linked APC fluorescence. 4′,6-Diamidino-2-phenylindole staining was used for dead cell exclusion. Collected GFP⁻/CD45⁻/CD31⁻ cells were lysed in TRIzol followed by standard RNA isolation. Equivalent amounts of RNA (50 ng) were reverse transcribed using RT² First Strand Kit and cDNA was preamplified with RT² PreAMP Primer Mix (SABiosciences/Qiagen) according to the manufacturer's instructions. Each sample was loaded on a Mouse Tumor Metastasis RT² Profiler PCR Array (SABiosciences/Qiagen) for real-time PCR-based quantitation of gene expression on an ABI/PRISM 7500 platform (Applied Biosystems/Life Technologies). Quality control, normalization against the arithmetic mean of four different housekeeping genes, data analysis, and generation of scatter plots were done with the RT² Profiler PCR Array Data Analysis v3.5 software (SABiosciences/Qiagen). Briefly, normalized expression of each gene between the two groups (met⁺ and met⁻) was plotted against one another. The central line in the plots in Fig. 4A indicates unchanged gene expression. Boundaries (fold regulation cutoff) were set at 5-fold change. Gene-wise ANOVA followed by t test between PC-3 met⁺, MDA-231 #1833 met⁺, and met⁻ qPCR samples was performed to identify differentially expressed genes. Correction for multiple hypotheses was carried out using the Benjamini–Hochberg procedure and genes with FDR < 0.1 were considered significant. For standard flow cytometry analysis (BD FACSCalibur), CD45⁻/CD31⁻ cells sorted from scaffolds 4 weeks after subdermal implant were stained with APC-anti-human HLA-ABC (BD Pharmingen, 555555) or biotin-anti-mouse CD44 (BD Pharmingen, 553132, clone IM7) followed by streptavidin-APC (eBioscience, 17-4317-82).

For immunohistochemistry, PC-3 cells and tissue sections (see below) were fixed with 4% paraformaldehyde, permeabilized with 0.1% Triton-X solution, and then blocked with 10% normal goat serum and 1% bovine serum albumin solution diluted in PBS. Samples were serially incubated with rat anti-mouse CD31 (BD Pharmingen, 55027), mouse anti-human cytokeratin 7 and 8 (BD Biosciences, 349205) or mouse anti-human Vimentin antibody (DAKO, 7165), followed by biotinylated goat anti-rat or goat anti-mouse IgG (BA-9400 and BA-9200, Vecta Lab) and ABC solution (Vecta Lab). Signal was visualized by applying DAB substrate (Vecta Lab).

Explanted scaffolds were embedded in optical cutting temperature compound and snap-frozen with dry ice-chilled 2-methylbutane. Frozen scaffold blocks were cut to 10 to 30 μm thickness and stored at −80°C until use. For hematoxylin/eosin, Giemsa, Trichrome, and Periodic Acid–Schiff stainings, frozen tissue sections were fixed with 10% buffered formalin solution and stained following the manufacturer's protocol (American Master Tech). For immunofluorescence stainings, frozen tissue sections were fixed with ice-cold acetone, blocked with 10% normal goat serum, and 1% bovine serum albumin diluted in PBS. Slides were incubated with rat anti-mouse CD31 (BD Pharmingen, 55027), mouse anti-human vimentin antibody (DAKO, 7165), or rabbit anti-human/mouse-IL1β (Abcam, ab9722), followed by goat anti-rat IgG Alexa Fluor 594 (A11007), goat anti-mouse IgG Alexa Fluor 488 (A11029), or goat anti-rabbit IgG Alexa Fluor 680 (A21109; Invitrogen/Life Technologies). Finally, VectaShield mounting medium with DAPI was applied and slides were imaged under confocal (Leica SP5) and fluorescence (Zeiss 200) microscopes.

For scanning electron microscope (SEM) imaging, in vitro cell-cultured and explanted scaffolds were fixed with 2% glutaradehyde, serially dehydrated with 20%, 50%, 70%, 90%, 95%, and 100% ethanol solution, and further dried using a lyophilizer overnight. A thin platinum/gold film was deposited on the samples using a sputter coating machine (208HR, Cressington) and then samples were imaged under FESEM Ultra55 (Zeiss).

We looked at the effect of grade/Gleason score on IL1β expression in previously published datasets for prostate (16) and breast (17) cancer. To determine the relationship between IL1β expression and grade/Gleason score, we collected log2 median centered intensity data and grade/Gleason categorical data from the Oncomine website (https://www.oncomine.org). With respect to the prostate cancer dataset, a t test was used to calculate the significance of the IL1β expression change between 12 Gleason score 6 samples and 15 Gleason score 9 samples. Similarly, we used a t test to calculate the significance of the IL1β expression change between 19 grade I tumor and 69 grade III breast tumor samples. Statistics were calculated using R (http://www.r-project.org).

A meta-analysis comparing IL1β expression between metastatic and primary tumor samples was performed using data from two prostate (18, 19) and one breast (20) published datasets. We selected metastatic samples with higher IL1β expression compared with the average IL1β level of the primary samples (or paired primary in the case of ref. 20) within each experiment. Using this criterion, we retained 8 unpaired (18), 11 unpaired (19), and 4 paired (20) sets of samples. The meta-analysis was performed using an in-house R script.

Unless otherwise stated, P values were calculated on the basis of a two-tailed Student t test. All sets of data met normal distribution. As an exception, a nonparametric Wilcoxon signed-rank test was applied to calculate the P value of the difference between the two cohorts of mice in the IL1Ra drug experiment, where values were not following a parametric model. Values of P < 0.05 were considered significant. All error bars represent SD of three or more replicates, as indicated in each figure. The sample size of each experiment is specified in figures and legends.

We adapted a bioengineered surrogate to mimic the 3D microenvironment of bone marrow (11, 21–24) to study blood-borne metastases in epithelial cancers. The marrow-like structure, which enhances engraftment by hematopoietic elements and leukemia cell lines, made it an attractive model to study prostate and breast cancer metastases, which have tropism for bone (25). Hydrogel scaffolds were synthesized with the microarchitecture of bone marrow sinusoids through a pattern of repeating honeycombs, matching mechanical properties of soft trabecular spongy bone (Fig. 1A and Supplementary Fig. S1A). These scaffolds were conjugated with type I collagen and they effectively supported the adhesion and proliferation of introduced primary human BMSCs. The BMSCs secreted growth factors, chemokines, and ECM that are concentrated within the hydrogel (Fig. 1B). Compared with noncellularized scaffolds, BMSC coating led to increased invasion by human cancer cells, as measured in a modified Boyden chamber assay (Fig. 1C and Supplementary Fig. S1B). The seeding and proliferation of luciferase- and GFP-tagged (Luc-GFP) cells was also significantly enhanced by BMSC-coating of scaffolds in a cell density–dependent manner (Fig. 1D and Supplementary Fig. S1C).

When subcutaneously implanted on the back of immunocompromised (NSG) mice, BMSC-coated scaffolds progressively lost their human stromal cell constituent, as shown by in vivo imaging analysis of tagged stromal cells, as well as by loss of staining of the retrieved devices for cells bearing human HLA markers (Supplementary Fig. S2A–S2C). Within 4 weeks from implantation, human BMSC cells were fully replaced by endogenous mouse stromal cells (Fig. 1E and Supplementary Fig. S2C and D). At this time point, the majority (89.2%) of CD45⁻/CD31⁻ mouse cells within the scaffold were scored positive for CD44 (Supplementary Fig. S2C), a marker enriched in mesenchymal cells, which plays an active role as a homing receptor and interacts with bone-specific ligands such as osteopontin (26–28). In addition, Masson trichrome staining revealed abundant presence of connective tissue (Fig. 1E), which makes this platform ideal to study microenvironmental changes at the stromal cell level.

We previously reported that the transient colonization by human BMSCs serves to initiate the recruitment of murine hematopoietic precursors to the device (11), which is then sustained by mouse cells. Although we have not tested the sublineage differentiation ability of hematopoietic cells following recruitment to implanted scaffolds, we previously demonstrated active homing of endogenous mouse Lin⁻/Sca1⁺/c-kit⁺ progenitors as well as engraftment capacity for human CD34 cells (11).

Taken together, while the model that we established does not precisely recapitulate the bone marrow environment, it mimics some of its main physical and cellular characteristics and provides an effective platform to enable capture and analysis of newly established metastases in a model organism.

We developed an orthotopic tumor model, in which Luc-GFP–tagged cancer cells can be used for in vivo tracking of both the primary tumor and the metastatic deposits derived from hematogenous seeding of the implanted scaffolds (Fig. 1F). For these experiments, we implanted four BMSC-functionalized scaffolds on the back of each recipient NSG mouse, allowing them to become coated with mouse-derived stromal cells and vasculature. Four weeks later, 1 × 10⁶ Luc-GFP PC-3 human prostate cancer cells were injected directly into the prostate, and mice were monitored using BLI as they demonstrated a progressive increase in primary tumor burden (Fig. 2A and Supplementary Fig. S3A). PC-3 cells were chosen for their proven ability to generate CTCs in the blood of mice bearing orthotopic xenografts in the prostate (data not shown). Animals were sacrificed at 2 months, at which time primary prostate tumors weighted approximately 400 mg. CTCs were readily detectable in blood samples (range: 13– 44 cells/mL/mouse) using a microfluidic capture platform (CTC-iChip; Fig. 2B; ref. 15). At the same time, ex vivo analysis of scaffolds using BLI revealed tumor cell–derived signal in 12 of 97 scaffolds (∼12%; average met⁺ scaffolds/mouse: 0.58; range: 0–4; Fig. 2C). In the 13 mice analyzed, the number of CTCs was correlated with the fraction of luciferase-positive scaffolds (R² = 0.97224), consistent with the presumed hematogenous seeding of the scaffolds (Fig. 2D). This finding highlights the ability of the scaffold to efficiently recruit tumor cells from circulation in vivo. PC-3 cells that had metastasized to the scaffolds were clearly identified by immunohistochemical staining using their characteristic markers, human vimentin and cytokeratin (Fig. 2E and Supplementary Fig. S4). Further culturing of individual explanted scaffolds in vitro for up to a month, yielded additional luciferase signal, with 14 of 40 (35%) becoming positive as rare tumor cells that had initially seeded the scaffolds in vivo were allowed to further proliferate in culture (Fig. 2F).

As control, we tested the efficiency of tumor cell seeding in vivo by other cancer cell types (Supplementary Fig. S3B and S3C). Poorly metastatic DU-145 prostate cancer cells (29) failed to engraft in the scaffold (0/10 scaffolds), despite generating sizable primary tumors in the prostate. In contrast, the highly aggressive MDA-231 #1833 breast cancer cell subline, which has been selected for enhanced tropism to bone (4), generated luciferase-positive signal from 13 of 24 scaffolds (∼54%), which all turned positive after one month in culture ex vivo (Supplementary Fig. S5).

The rapid expansion of primary prostate tumors in mice following orthotopic inoculation necessitated removal of the scaffolds before they are fully mature, as noted by the fact that some explanted devices turn luciferase-positive following ex vivo culture. To enable more complete monitoring of metastatic microenvironments as they evolve in vivo, we established a transplantation assay, whereby scaffolds are transplanted from tumor-bearing mice into naïve recipient mice (Fig. 3A). The specific biophysical characteristics of the scaffold allow successful transplants by minimizing time and manipulation of the premetastatic microenvironment formed in vivo, and recipient mice can then be imaged for an additional 2 months, without interference from a large primary tumor (Fig. 3B and C). Indeed, 26 of 68 scaffolds (∼38%) were scored as met⁺ using this in vivo imaging end point, matching the frequency obtained after ex vivo culture (Fig. 2F). Tumor cell colonization of the devices was consistent with observation of visceral metastases in primary tumor-bearing mice. Of note, with this serial transplantation assay, none of the transplanted scaffolds from mice harboring DU-145 primary tumors (negative control) and all of those from mice with MDA-231 #1833 tumors (positive control) were scored positive by in vivo imaging at 2 months (Supplementary Fig. S6A–S6C). Serial transplantation of scaffolds allowed us to capture within each device the full evolution of metastasis, from initial extravasation and seeding of a single CTC to colonization and ECM degradation (Fig. 3D). Thus, integration of transplantable scaffolds with live imaging in orthotopic mouse tumor models may enable detailed characterization of the formation and progression of the metastatic niche in vivo.

Coexistence of tumor-infiltrated (met⁺) and tumor-free (met⁻) scaffolds within the same animal prompted us to explore differences in their microenvironment at the molecular level. Live cells were recovered from paired met⁺ and met⁻ scaffolds, derived from three independent tumor-bearing mice. Distinct cell populations from matched met⁺ and met⁻ scaffolds were stained for various cell surface markers and quantified by flow cytometry (Fig. 3E and F and Supplementary Fig. S6D and S6E). Consistent with a prior study showing direct competition between prostate cancer cells and hematopoietic stem cells in the bone marrow niche (30), we found that the hematopoietic/endothelial compartment (GFP⁻/mCD45⁺/mCD31⁺) was reduced in the presence of metastatic prostate cancer cells (GFP⁺; Fig. 3F). This effect was even more marked for the highly invasive MDA-231 #1833 breast cancer cells (Supplementary Fig. S6E). In contrast, the relative number of stromal cells was not affected by tumor burden.

To further characterize the stromal compartment of tumor-infiltrated scaffolds, we compared RNA expression profiles from FACS-sorted stromal cells derived from either met⁻ or met⁺ scaffolds from the same tumor-bearing mouse, interrogating a predesigned qPCR array including 85 metastasis-related genes (Supplementary Table S1). Among the most commonly upregulated genes from met⁺ stromal cells (>5 fold change), we identified the cytokine IL1β (Fig. 4A), matrix metalloproteinases (Mmp-13, Mmp-10, and Mmp-9) that are known contributors to the premetastatic niche, as well as the angiogenesis-related tyrosine kinase Flt-4, the prometastatic gene Etv-4 and the Mcam adhesion molecule. Among these, IL1β was the most differentially induced gene in the PC-3 prostate cancer model (∼3,600-fold; P < 0.005; FDR < 0.02) and it was also upregulated (∼21 fold; P < 0.02; FDR < 0.05) in the MDA-231 breast cancer model (Fig. 4A). Immunohistochemical staining of met⁺ scaffolds confirmed that IL1β expression was almost exclusively derived from stromal cells, rather than cancer cells (Fig. 4B).

The expression of IL1β by human cancers is well documented, although the source of its production may be variable. The cytokine may be derived from malignant cells themselves, from stromal cells in response to factors secreted by cancer cells (31), or as part of the inflammatory response that accompanies tumor growth (32). IL1β overexpression itself is associated with increased MMP expression, invasiveness, angiogenesis, and tumor-mediated immune suppression (33–35). To confirm the physiologic significance of IL1β expression as it relates to metastatic propensity, we tested a well annotated prostate cancer dataset (16). Expression of IL1β by primary prostate tumors was positively correlated with their Gleason score, a well-defined histologic marker of invasiveness and metastatic recurrence risk (Fig. 4C). A similar correlation between histologic grade and IL1β expression was evident in our analysis of a large breast cancer dataset (Fig. 4C; ref. 17). Furthermore, a meta-analysis of multiple available datasets (18–20) showed significant upregulation of IL1β in both prostate and breast cancer metastases, compared with primary tumors (Fig. 4D). All available datasets are based on gene expression profiling of bulk primary or metastatic tumors, thus containing a mixed population of cancer cells and stromal cells. The elevation in IL1β expression may therefore be even greater in a pure stromal cell population. These clinical correlations are consistent with previous reports of IL1 expression in several invasive tumor types (33, 36, 37).

To test the reliability of our new scaffold-based method to identify relevant metastatic stromal components and verify the functional significance of stromal cell IL1β expression, we treated prostate tumor-bearing mice carrying subcutaneous scaffolds with either an FDA-approved IL1 receptor antagonist (IL1Ra) or vehicle control (4 mice and 16 scaffolds per test group). Mice were treated daily for 2 weeks, beginning 6 weeks after orthotopic inoculation of prostate cancer cells. At the time of IL1Ra treatment, primary tumors were readily detectable by BLI in vivo imaging, but the scaffolds were not yet scored positive. After 2 weeks of treatment, scaffolds were transplanted into naïve recipient mice, which were monitored for the development of metastases without further treatment for an additional 2 months (Fig. 4E). IL1Ra-treated mice showed a 50% reduction in the number of met⁺ scaffolds: 3 of 16 (18.7%) met⁺ scaffolds in treated mice versus 6 of 16 (37.5%) scaffolds in controls (P < 0.005). The positive scaffolds derived from IL1Ra-treated mice also had a dramatic reduction in luciferase signal intensity (Fig. 4F). As tumor-bearing mice were treated only for a limited time before transplantation of scaffolds into naïve mice, it is likely that IL1β contributes to the early recruitment of CTCs to the scaffold rather than supporting tumor cell proliferation after engraftment. Consistent with this model, IL1Ra had no effect on PC-3 cell proliferation in vitro (Supplementary Fig. S7A). PC-3 met⁺ scaffolds derived from untreated mice also showed no growth advantage or growth suppression, following ex vivo culture in the presence of either recombinant IL1β or IL1Ra (Supplementary Fig. S7B).

The importance of the microenvironment in metastatic progression is well established, but our understanding of the precise contribution of specific factors and cell populations, which have been extensively studied at the primary tumor level (38, 39), is limited at secondary sites. Given the technical hurdles incidental to recovering and studying the stromal compartment of the metastatic niche in vivo, we aimed to develop an experimental method to test whether a bioengineered scaffold could be used to dissect the cellular components of a metastatic lesion. As increasing numbers of therapeutic agents are developed to target the tumor-supporting stroma within bone metastases (40), there is a need for controllable models that mimic physiologic tumor/stromal interactions in vivo, extending beyond standard available in vitro coculture experimental conditions (7, 8). Metastases to distant organs are infrequent in mouse tumor models (41), and these are rarely accessible for detailed molecular analysis. The biocompatible scaffold described here circumvents these technical limitations based on its high efficiency engraftment by blood-borne CTCs, as well as its susceptibility to serial in vivo passaging and monitoring and ex vivo cellular and molecular analysis. This versatile platform to capture viable metastasis-competent CTCs and analyze their interactions with host stromal cells builds on previous approaches (42, 43), enabling study of the evolution of the metastatic niche at different stages of tumor progression. For instance, we show that explanting tumor-seeded scaffolds at different time points allows characterization of changes within the stromal compartment that may allow the establishment of a metastatic focus.

In providing a proof-of-concept of the ability of our platform to uncover relevant stromal signals involved in CTC recruitment and initiation of metastatic lesions, we confirmed the pivotal role for the IL1β cytokine in promoting tumor cell engraftment. In our experiments, IL1β expression was detected within stromal cells of tumor-bearing but not tumor-free scaffolds, and its role in enhancing metastatic seeding of the bioengineered scaffold by prostate PC-3 cancer cells was supported by the effectiveness of a receptor antagonist administered systemically. Our observations are consistent with a recent report of increased bone metastases following intracardiac injection of IL1β-transduced prostate cancer cells in the mouse (37). In these experiments, knockdown of endogenous IL1β within injected tumor cells had no effect in reducing the frequency of metastatic lesions, supporting our observation that stromal cell expression may be the more physiologic source of this cytokine, rather than tumor cells. Of note, additional stromal-derived factors were evident in our analysis of scaffolds bearing either prostate or breast cancer colonization, some of which overlap, suggesting that our platform may have broad utility in characterizing tumor/stromal interactions across multiple tumor types. Among the differentially expressed genes identified in metastatic stroma, some, such as MMPs, chemokines (Ccl7) and chemokine receptors (Cxcr2, Cxcr4) are likely involved in the formation of a receptive microenvironment, whereas others, such as tyrosine kinase receptors (Flt-4) and transcription factors (Etv-4), possibly favor the active growth of a secondary tumor mass. Both the engraftment and the growth-promoting aspects, although difficult to distinguish, contribute to the homing and generation of bone metastasis and can be interrogated using the transplantable scaffold described in this study. As exemplified by the consequences of suppressing IL1β in the PC-3 cell model, identification of key factors implicated in tumor/stromal crosstalk may lead to potential therapeutic intervention.

Taken together, this integrated approach, involving implantation of bioengineered devices into mouse tumor models followed by cellular and molecular characterization of colonizing cell populations, offers a powerful tool to study the metastatic microenvironment, dissect molecular changes within different components of the metastatic niche, and model tumor/stromal interactions in vivo.

B. Parekkadan has ownership interest (including patents) and is a consultant/advisory board member for Sentien Biotechnologies, Inc. No potential conflicts of interest were disclosed.

Conception and design: F. Bersani, J. Lee, M. Toner, D.A. Haber, B. Parekkadan

Development of methodology: F. Bersani, J. Lee, M. Yu, M. Toner, B. Parekkadan

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): F. Bersani, J. Lee, M. Yu

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): F. Bersani, J. Lee, M. Yu, R. Morris, S. Ramaswamy, M. Toner, B. Parekkadan

Writing, review, and/or revision of the manuscript: F. Bersani, J. Lee, M. Yu, M. Toner, D.A. Haber, B. Parekkadan

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Toner, B. Parekkadan

Study supervision: D.A. Haber, B. Parekkadan

Other (carried out experimental procedures): R. Desai

The authors thank S.M. Rothenberg for help in developing the mouse models, R. Murray for technical support in BMSC culture and in vitro characterization, and R. Mylvaganam from the MGH Flow Cytometry Core and M. Waring from the Ragon Institute for cell sorting. The authors also thank R. Taulli and all laboratory members for helpful discussions.

This work was supported in part by NIH grants R01EB012521 (B. Parekkadan), K01DK087770 (B. Parekkadan), K99CA163671 (J. Lee), R01CA129933 (D.A. Haber and F. Bersani), the Shriners Hospitals for Children (J. Lee and B. Parekkadan), the National Foundation for Cancer Research (D.A. Haber), and the Howard Hughes Medical Institute (D.A. Haber).

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.