Monitoring metastatic events in distal tissues is challenged by their sporadic occurrence in obscure and inaccessible locations within these vital organs. A synthetic biomaterial scaffold can function as a synthetic metastatic niche to reveal the nature of these distal sites. These implanted scaffolds promote tissue ingrowth, which upon cancer initiation is transformed into a metastatic niche that captures aggressive circulating tumor cells. We hypothesized that immune cell phenotypes at synthetic niches reflect the immunosuppressive conditioning within a host that contributes to metastatic cell recruitment and can identify disease progression and response to therapy. We analyzed the expression of 632 immune-centric genes in tissue biopsied from implants at weekly intervals following inoculation. Specific immune populations within implants were then analyzed by single-cell RNA-seq. Dynamic gene expression profiles in innate cells, such as myeloid-derived suppressor cells, macrophages, and dendritic cells, suggest the development of an immunosuppressive microenvironment. These dynamics in immune phenotypes at implants was analogous to that in the diseased lung and had distinct dynamics compared with blood leukocytes. Following a therapeutic excision of the primary tumor, longitudinal tracking of immune phenotypes at the implant in individual mice showed an initial response to therapy, which over time differentiated recurrence versus survival. Collectively, the microenvironment at the synthetic niche acts as a sentinel by reflecting both progression and regression of disease.

Significance:

Immune dynamics at biomaterial implants, functioning as a synthetic metastatic niche, provides unique information that correlates with disease progression.

All too often, the detection of cancer occurs after some tissue such as the lung or bone begin to fail, which typically represents an advanced stage of disease that is challenging to treat. Systems for the early detection of cancer would provide the opportunity to intervene at earlier stages and prior to tissue dysfunction. The risk for distant recurrence is being determined by gene and protein expression in primary tumor samples that define molecular signatures of patient tumors. Clinically, these molecular signatures are reduced to predictive scores, which help guide decisions about systemic therapy for patients with cancer (1). Clinical examples include Oncotype DX (2), Mammaprint (3), PAM50 ROR Score (4, 5), Breast Cancer Index (6), and EndoPredict (7). Yet, the information obtained from the primary tumor is focused on measurements taken prior to the initiation of first-line treatment of the primary tumor, which limits this approach to pretreatment predictions of disease progression. Liquid biopsies show promise, yet face challenges associated with low cell numbers, as well as phenotypic relevance, as approximately 0.01% of circulating tumor cells will become metastatic foci and circulating immune cells are distinct from those in distal tissue (8–10).

We and others have previously reported that biomaterial implants function as a synthetic metastatic niche, with recruitment of early metastatic cells that represent aggressive populations observed within the lung (11–19). The development of the metastatic niche is driven by circulating acellular material from the primary tumor that alters the immune system of the host, including discrete sites in distal organs (20–23). These sites precede and support colonization by metastatic tumor cells (Supplementary Fig. S1). The identification of the metastatic niche was enabled through mouse models of metastasis, which recapitulated the conditioning of distal sites and validated the long standing “seed and soil” hypothesis (24). The metastatic niche develops, in part, from an influx of bone marrow–derived cells, such as hematopoietic progenitor cells and myeloid-derived suppressor cells (MDSC), that differentiate within the distal sites and create a permissive metastatic site in target organs (20, 21). Monitoring the natural metastatic niche could alter the course of therapeutic interventions, yet the potential is hindered because of metastatic niche formation in obscure and inaccessible locations within vital organs, such as the lung and the liver. A synthetic metastatic niche provides a technical solution by predefining sites in readily accessible locations.

We tested the hypothesis that analysis of immune dynamics at biomaterial implants, functioning as a synthetic metastatic niche, would provide unique information that correlates with disease progression. The tissue within the implant integrates with the host and is subsequently modified by disease onset and progression (25, 26). Following implantation, a porous biomaterial scaffold supports cell infiltration, vascularization, and the persistent recruitment of immune cells due to the localized foreign body response to the material (27). The initiation of cancer metastasis is associated with systemic alteration of immune responses, which we hypothesize also alters the composition of the foreign body response (28). In the context of cancer progression, the immune response at an implanted microporous polymer scaffold would acquire immunosuppressive trademarks, thus mimicking the properties of disease-targeted organs. We propose to analyze the immune cell composition and phenotype to draw correlations with disease progression, while sparing vital organs or other sites from invasive sampling. For cancer metastasis, this scaffold could advance disease detection capabilities by providing a predefined site that indicates an altered immune environment and disease progression in vital organs (20). Identifying early metastatic events has the potential to inform a new generation of therapeutic approaches that target the earliest stages of metastatic disease.

An expanded Materials and Methods section with more detail is provided in the Supplementary Data.

Microporous PCL scaffold fabrication

To form microporous scaffolds, poly(ϵ-Caprolactone) (PCL) microspheres were mixed with salt porogen (NaCl, 250–425 μm) at 1:30 (w/w) ratio, pressed into 5-mm wide 2-mm thick PCL/NaCl disc (Fig. 1A), then heated to 135°C prior to NaCl dissolution. The resulting microporous PCL scaffolds were disinfected and stored at −80°C until time of implantation.

Figure 1.

Microporous polymer scaffolds support host tissue integration when implanted, which is then progressively modified as a function of disease burden and presents unique dynamics compared with blood. A, PCL scaffolds and experimental timeline for scaffold implantation, tumor inoculation, implant biopsies, and high-throughput gene expression analysis. B and C, Representative macroscopic and SEM images of PCL scaffold illustrating microporous architecture. Scale bar, 500 μm and interconnected 250–425 μm pores. D, DAPI staining illustrates dense and uniform cellularity throughout the implant. Scale bar, 500 μm. Highlighted area (dashed box) indicates region of higher magnification (E). Scale bar, 100 μm. F, Progressive gene expression changes in tissue biopsied from implants during metastatic disease course. BALB/c mice were implanted with microporous polymer implants (day −14), then inoculated with syngeneic 4T1 tumor cells at day 0. Implants were biopsied at days 7, 14, and 21 and tissue was analyzed with a high-throughput qRT-PCR platform (OA, 632 target and 16 reference genes). The heatmap illustrates unsupervised hierarchal clustering of the 10-gene panel, with expression levels for each gene depicted as standardized data (n = 14 per condition). Cohort size at days 7 (n = 3), 14 (n = 8), and 21 (n = 3) represent different mice as biological replicates. G, BALB/c leukocytes were isolated from blood of tumor-bearing mice, gene expression profiled, and compared against tumor-free and implant-free control mice. Many of the genes in the panel showed significant differences in the same directionality as those in the implant. H, The heatmap shows differences in dynamics between the implant to blood gene expression ratio and the goodness of fit for this relationship between days 7 and 21. Fit for the regression analyses was determined by normalized root mean squared error, where negative values indicate a worse fit and a value of 1 indicates a perfect fit. S100a9, S100a8, Pglyrp1, and Ltf had the worst fits when comparing implant with blood. Bmp15 is not shown as it was not measurable in blood. Box plots and indications of significance for scaffold and blood gene expression are shown in Supplementary Fig. S6 and Supplementary Data File S4.

Figure 1.

Microporous polymer scaffolds support host tissue integration when implanted, which is then progressively modified as a function of disease burden and presents unique dynamics compared with blood. A, PCL scaffolds and experimental timeline for scaffold implantation, tumor inoculation, implant biopsies, and high-throughput gene expression analysis. B and C, Representative macroscopic and SEM images of PCL scaffold illustrating microporous architecture. Scale bar, 500 μm and interconnected 250–425 μm pores. D, DAPI staining illustrates dense and uniform cellularity throughout the implant. Scale bar, 500 μm. Highlighted area (dashed box) indicates region of higher magnification (E). Scale bar, 100 μm. F, Progressive gene expression changes in tissue biopsied from implants during metastatic disease course. BALB/c mice were implanted with microporous polymer implants (day −14), then inoculated with syngeneic 4T1 tumor cells at day 0. Implants were biopsied at days 7, 14, and 21 and tissue was analyzed with a high-throughput qRT-PCR platform (OA, 632 target and 16 reference genes). The heatmap illustrates unsupervised hierarchal clustering of the 10-gene panel, with expression levels for each gene depicted as standardized data (n = 14 per condition). Cohort size at days 7 (n = 3), 14 (n = 8), and 21 (n = 3) represent different mice as biological replicates. G, BALB/c leukocytes were isolated from blood of tumor-bearing mice, gene expression profiled, and compared against tumor-free and implant-free control mice. Many of the genes in the panel showed significant differences in the same directionality as those in the implant. H, The heatmap shows differences in dynamics between the implant to blood gene expression ratio and the goodness of fit for this relationship between days 7 and 21. Fit for the regression analyses was determined by normalized root mean squared error, where negative values indicate a worse fit and a value of 1 indicates a perfect fit. S100a9, S100a8, Pglyrp1, and Ltf had the worst fits when comparing implant with blood. Bmp15 is not shown as it was not measurable in blood. Box plots and indications of significance for scaffold and blood gene expression are shown in Supplementary Fig. S6 and Supplementary Data File S4.

Close modal

Animals and scaffold implantation

Scaffolds were implanted in the subcutaneous space of female mice. All animal studies were performed in accordance with institutional guidelines and protocols approved by the University of Michigan Institutional Animal Care and Use Committee. Female mice of the BALB/c and C57BL/6 strains were purchased from Jackson Laboratories at an age of 8 weeks. For implantation, mice were anesthetized under isoflurane and administered the analgesic Carprofen prior to and 24 hours after surgery. Scaffolds were then surgically inserted into dorsal subcutaneous space.

Metastatic tumor cell lines and animal inoculation

Orthotopic inoculation of tumor cells in the 4th right mammary fat pad was performed 2 weeks after scaffold implantation. Our previous data shows that tumor cells colonize scaffolds if they are implanted prior to or following tumor inoculation (11, 12). In the studies here, implantation of the scaffolds prior to tumor inoculations allows our data to start from a tumor-free baseline. This approach reflects the translatable strategy of implanting scaffolds following effective treatment of a primary tumor to provide a tumor-free baseline for monitoring distant recurrence. 4T1-luc2-tdTomato cells were obtained from Perkin Elmer, which had been authenticated previously via short tandem repeat (STR) and comparison with ATCC STR database. E0771 GFP+ cells were a kind gift from the Laboratory of Gary and Kathryn Luker at the University of Michigan Center for Molecular Imaging. 4T1 and E0771 are syngeneic breast cancer cell lines for the BALB/c and C57BL/6 strains, respectively. E0771 lung metastases were serially inoculated to develop a unique metastatic variant (Lu.2; Supplementary Fig. S2A–S2G), similar to reports on 4T1 phenotype. The Lu.2 E0771 variant we developed was not analyzed by STR.

Biopsies of tissue from scaffold, blood, and lung tissue

Biopsies of tissue from microporous scaffolds, leukocytes from blood, and lung tissue were isolated to study gene expression changes due to disease progression. Microporous scaffolds that had been implanted for either 7, 14, or 21 days following inoculation of primary tumors (as indicated in figures) were surgically biopsied (Supplementary Fig. S3A) to eliminate caveats associated with biopsies of regenerated tissue in the implant, yet this aspect would need to be analyzed for translation. For core-needle biopsy (CNB) of the implant, a disposable CNB tool (Bard Mission Disposable Core Biopsy Instrument) was inserted through the skin and used to isolate a portion of the implant and infiltrating tissue (Supplementary Fig. S3B and S3C).

Surgical resection of primary tumor and longitudinal tracking

To test the responsiveness of the implant microenvironment to therapeutic treatment, mammary gland removal was performed to excise the 4T1 primary tumor from BALB/c mice. In this model, the primary tumor was resected 14 days after inoculation to align with data acquired during disease progression. Following primary tumor excision, a scaffold was biopsied weekly to monitor gene expression dynamics longitudinally in individual mice. Following tumor resection and mammary gland removal, animal health was monitored daily for activity and responsiveness, including posture, mobility, body weight, grooming behavior, and respiratory conditions. Animals were euthanized if found in a moribund condition or when a primary tumor regrowth was positively identified.

RNA isolation, purity, integrity, and cDNA synthesis

Total RNA was isolated from tissue biopsied from implants, lung tissue, and leukocytes from blood. Samples were immersed in TRIzol, homogenized with a rotor stator homogenizer, centrifuged to remove particulate, then processed in a silica-based matrix spin column with DNase I treatment to isolate total RNA. Concentration, purity, and RNA integrity number analysis were used to validate the samples prior to generation of first strand cDNA through reverse transcription.

OpenArray high-throughput quantitative reverse-transcription PCR

Tissue biopsied from implants was analyzed for gene expression with OpenArray (OA), a high-throughput quantitative reverse-transcription PCR (qRT-PCR) platform that analyzes 648 genes per sample in parallel. cDNA from tumor-free and tumor-bearing mice at days 7, 14, and 21 were analyzed, with OA qRT-PCR run and sample quality control being performed by the Affymetrix Group at the University of Michigan DNA Sequencing Core.

Analysis of gene expression and selection of genes of interest

Gene expressions from OA were screened to identify genes of interest during disease course. qRT-PCR quality control resulted in 559 genes for full analysis selection of 5 reference genes Gapdh, Tbp, Ywhaz, Hmbs, and Ubc. ΔCq values were calculated against the average of these reference genes. For computational analysis ΔCq values are log2 transformed and centered on the healthy, tumor-free control time-matched median.

Ten-gene panel qRT-PCR analysis in single-tube format

Experiments for signature validation, analysis of blood and lung, and postexcision monitoring was performed by qRT-PCR analysis in 384-well and 96-well plate formats using matched TaqMan probes from the OA platform (see Supplementary Data 1 for probe details), with data quality control and ΔCq measures performed in alignment with OA data.

Single-cell RNA-seq of immune cells in tissue biopsied from implants

Scaffolds from BALB/c mice inoculated with 4T1 tumor cells for 14 days, were removed and digested to obtain a single-cell suspension. The cells were prepared for single-cell sequencing using Drop-Seq. mRNA was collected from each cell and subsequently converted to cDNA before being taken through tagmentation and PCR. Samples were pooled and sequenced on the Illumina HiSeq-4000. The final reads were mapped to the mm10 mouse reference genome to quantify gene expression and a read-count matrix was assembled. Single-cell gene expression was conducted using Seurat, an open source program in R developed by the Satija lab at the New York Genome Center.

Adoptive transfer and implant trafficking analysis of Ly6G+ and Gr-1+Ly6G immune populations

Splenocytes were isolated from 4T1 tumor-bearing BALB/c mice. G-MDSCs (Ly6G+) and M-MDSCs (Gr1+Ly6G) were taken from splenocytes by magnetic bead separation using the Myeloid-Derived Suppressor Cell Isolation Kit (Miltenyi Biotec). G-MDSCs and M-MDSCs were labeled with lipophilic membrane dyes then injected intravenously into mice, and analyzed for trafficking by flow cytometry (Supplementary Fig. S4A–S4F) of scaffolds at 36 hours after transfer.

Gene expression dimensionality reduction and classification

Genes of interest were identified on the basis of fold change, level of significance, and expression stability (continuous trajectory for the expression of a gene). Clustering of samples was performed with the Matlab clustergram function on standardized ΔCq data. Unsupervised dimensionality reduction was performed using singular value decomposition and a supervised machine learning algorithm, bootstrap aggregated (bagged) decision tree ensemble (i.e., random forest), was used to construct a predictive model based on the 10-gene panel.

Statistical analysis

Error bars and the exact n for each test are detailed in the legends along with the methods for multiple comparisons corrections. All n throughout the article indicate biological replicates from different mice. Exact values for significant and nonsignificant P values, F values, and degrees of freedom are available in the Supplementary Data S4. All statistical indicators within the article were computed using packages and syntax in the Statistical Package for the Social Sciences (SPSS, International Business Machines Co.).

Breast cancer patient microarray Kaplan–Meier correlation

Genes of interest from the implant-derived data were queried in human patient microarray data for correlations between high or low gene expression and patient overall and recurrence-free survival. Microarray data from Gene Expression Omnibus (GEO) was analyzed using Kaplan–Meier Plotter (KMPlot, kmplot.com/analysis/).

Protocol, material, data, and code availability

Additional protocol specifics and the Lu.2 cell line are available from authors. qRT-PCR and scRNA-seq can be accessed from GEO (GSE138993) and ArrayExpress (E-MTAB-8503) repositories, which comply with MIAME and MINSEQE, respectively. Code used for signature development is available in the Supplementary Data Files. SPSS syntax used for statistical analysis is available alongside the statistical outputs in Supplementary Data. All data and materials are available in public repositories or available from the authors as outlined in the Materials and Methods sections on protocol, data, and code availability.

Metastatic disease alters gene expression at implants as a function of disease severity

We performed a high-throughput gene expression analysis of immune pathways to characterize the dynamics at a biomaterial implant that recruits metastatic cells, throughout the process of disease progression. We and others have shown that metastatic tumor cells colonize microporous implants, yet are vastly outnumbered by other cell types including stromal, immune, and epithelial cells (11–14, 17). Microporous implants (5-mm diameter, 2-mm thickness, interconnected 250–425 μm pores) composed of PCL (Fig. 1A and B), were inserted into dorsal subcutaneous space of BALB/c mice to allow for tissue ingrowth. Microporous PCL scaffolds (Fig. 1C) facilitate the longitudinal studies herein by provide greater implant stability compared with more rapidly degradable polymers like poly(lactide-co-glycolide) (12). This microporous architecture facilitated cell colonization throughout the thickness of the implant (Fig. 1D and E). After 2 weeks, mice were orthotopically inoculated (day 0) with triple-negative 4T1 tumor cells. At weekly intervals (days 7, 14, and 21), tissue from implants was biopsied and changes in the gene expression were screened via a high-throughput qRT-PCR platform, OA (Fig. 1A). This screening assessed the expression of 632 target and 16 reference genes per sample in parallel (Supplementary Fig. S5A). Gene expression from the biopsied implants of tumor-bearing mice was compared against time-matched tumor-free control mice that had implants yet were not inoculated with 4T1 cells. Altered expression was observed for 113 genes following tumor inoculation (Supplementary Data File S3). To focus our studies, a panel of 10 genes was identified on the basis of fold change (Supplementary Fig. S5B), level of significance (FDR corrected), and expression stability. The 10 target genes (Fig. 1F; Supplementary Fig. S6A) were normalized against an average expression of 5 reference genes. As a function of disease progression (Supplementary Fig. S6B), expression increased for 7 of the 10 genes within the implant, including: S100 Calcium Binding Protein A8 (S100a8), S100 Calcium Binding Protein A9 (S100a9), Peptidoglycan Recognition Protein 1 (Pglyrp1), Lactotransferrin (Ltf), Cathelicidin Antimicrobial Peptide (Camp), Elastase 2 (Ela2), Chitinase (Chi3l3). Expression decreased for 3 of the 10 genes within the implant, including: Bone Morphogenetic Protein 15 (Bmp15), C-C Motif Chemokine Ligand 22 (Ccl22), and C-C Motif Chemokine Receptor 7 (Ccr7). Unsupervised hierarchical clustering of the expression of these genes in the implant-derived tissue produced a separation between tumor-bearing and tumor-free control cohorts. Over time, the gene expression in the tissue infiltrate of tumor-bearing mice progressively separated from tumor-free controls with day 7 expression being most similar to tumor-free mice. No structured or time-based organization was observed in the gene expression from the tumor-free controls, presumably due to the microenvironment stability in healthy mice. Clustering also identified distinct gene groups (i.e., S100a8/S100a9/Pglyrp1) that displayed similar expression patterns.

The 10-gene panel was then analyzed in blood leukocytes isolated from blood at multiple time points following tumor inoculation to identify correlations or unique dynamics in synthetic niche. Liquid biopsies represent a relatively accessible source to analyze the intermediary stages of metastasis that occur between the primary tumor and distant metastatic sites. Most genes (except Bmp15) analyzed in the implant-derived tissue were expressed in the blood (Fig. 1G; Supplementary Fig. S6C), yet their expression dynamics differed from those of the implant. This difference was particularly evident for the genes S100a9, S100a8, Pglyrp1, and Ltf. At the implant, expression of these genes progressively increased over time, which aligned with the progression of disease. In the blood, expression of these genes was stable and lacked a correlation with disease progression, illustrated by a goodness-of-fit analysis in which S100a9, S100a8, Pglyrp1, and Ltf had the poorest alignment between the implant and blood (Fig. 1H). In contrast, gene expression for Ela2 and Camp progressively increased in both the implant and blood. Collectively, the data acquired from blood does not directly correlate with the synthetic niche, indicating that synthetic niche provides distinct information that has the potential to complement liquid biopsy.

Disease-driven differences at the implant have unique dynamics and are reflective of distal organs

The 10-gene panel was analyzed in diseased lung tissue to determine whether the synthetic niche correlates with native metastatic sites in distal organs. The lung is normally the primary metastatic site for 4T1 cells, and we thus analyzed lung biopsies from tumor-bearing and tumor-free control mice for comparison with synthetic niche. The gene expression patterns for the 10-gene panel in the lung at day 21 following tumor inoculation (Fig. 2A; Supplementary Fig. S6D) was highly aligned with that observed in the implant. Notable is that all 7 genes upregulated in tissue biopsied from synthetic niche are clustered in the lung data. Similarly, the 3 genes downregulated in the tissue biopsied from implants are clustered in the lung data. These measurements of tissue taken from the lung were consistent with data from other studies regarding metastatic niche development and the phenotype of MDSCs in distal tissues during metastasis (23). Collectively, the gene expression patterns of the synthetic metastatic niche reflect the patterns observed in native metastatic niche.

Figure 2.

Tissue biopsied from implants is indicative of diseased lung. A, BALB/c mice orthotopically inoculated with 4T1 tumor cells at day 0 had lung tissue biopsies taken and qRT-PCR analyzed at day 21 from tumor-bearing and tumor-free control mice. Heatmap of gene expressions from lung tissue of tumor-free control and tumor-bearing mice. Organization was based on unsupervised clustering of samples and genes. B, Tissue biopsied from implants in C57BL/6 was analyzed for gene expression in tumor-free and tumor-bearing mice that were inoculated with a lung-tropic, syngeneic metastatic cell line. C57BL/6 mice were implanted with microporous PCL scaffolds (day −14) and then orthotopically inoculated with a metastatic derivative (developed through serial inoculations of explanted lung metastases) of the E0771 syngeneic line (day 0). At day 14, scaffolds were biopsied from tumor-bearing and tumor-free control mice. Heatmap of gene expressions for a 10-gene panel normalized to reference genes, centered on the healthy tumor-free control median, and standardized. Organization is based on unsupervised clustering of samples and genes. Box plots and indications of significance for lung tissue and the C57BL/6-E0771Lu.2 model are shown in Supplementary Fig. S6 and Supplementary Data File S4.

Figure 2.

Tissue biopsied from implants is indicative of diseased lung. A, BALB/c mice orthotopically inoculated with 4T1 tumor cells at day 0 had lung tissue biopsies taken and qRT-PCR analyzed at day 21 from tumor-bearing and tumor-free control mice. Heatmap of gene expressions from lung tissue of tumor-free control and tumor-bearing mice. Organization was based on unsupervised clustering of samples and genes. B, Tissue biopsied from implants in C57BL/6 was analyzed for gene expression in tumor-free and tumor-bearing mice that were inoculated with a lung-tropic, syngeneic metastatic cell line. C57BL/6 mice were implanted with microporous PCL scaffolds (day −14) and then orthotopically inoculated with a metastatic derivative (developed through serial inoculations of explanted lung metastases) of the E0771 syngeneic line (day 0). At day 14, scaffolds were biopsied from tumor-bearing and tumor-free control mice. Heatmap of gene expressions for a 10-gene panel normalized to reference genes, centered on the healthy tumor-free control median, and standardized. Organization is based on unsupervised clustering of samples and genes. Box plots and indications of significance for lung tissue and the C57BL/6-E0771Lu.2 model are shown in Supplementary Fig. S6 and Supplementary Data File S4.

Close modal

We developed a lung-tropic tumor cell line for C57BL/6 as an additional immunocompetent model of lung-biased metastasis to determine whether the expression panel may reflect lung metastasis. To match the aggressive and lung-tropic metastasis of the BALB/c-4T1 model, we developed an aggressive variant of the E0771 cell line through serial inoculation of lung metastases. This aggressive and lung-tropic E0771 variant, denoted as Lu.2, was inoculated in C57BL/6 mice implanted with the polymer scaffolds. This experiment also used an orthotopic inoculation of tumor cells into the mammary fat pad, which maintained the experimental design of a primary tumor conditioning distal sites and spontaneous metastasis. The C57BL/6-Lu.2 had similar gene expression changes and clustering to the lung tissue from the BALB/c-4T1 model (Figs. 1C and 2A), with 80% (8/10) of altered genes progressing in a similar direction (Fig. 2B; Supplementary Fig. S6E). The two genes that demonstrated a difference in expression (Camp and Chi3l3) may be due to the slower progression of disease that we observed in the B6-Lu.2 model, yet additional studies would be necessary to fully understand the impact of mouse strain and cell line metastatic potential on distal conditioning. Our data illustrates that the synthetic niche consistently mimics and serves as a sentinel for metastatic conditioning of lungs.

Select immune populations dictate alterations in the implant microenvironment during metastasis

We next investigated the gene expression of innate and adaptive immune cells (Fig. 3A) within implants during cancer metastasis to determine the cell types driving metastatic conditioning of the synthetic niche. Single-cell RNA-sequencing (scRNA-seq) was employed to investigate cell-specific gene expression. Our data supports previous studies that MDSCs are a driving force of immunosuppressive S100a8/9 (23). MDSCs had increased expression of S100a8, S100a9, Pglyrp1, and Ltf relative to the other cell types (Fig. 3B). Specifically, Pglyrp1 was 1.8-fold higher in MDSCs than other neutrophils from implants of tumor-bearing mice. The scRNA-seq data are consistent with our previous data, showing an influx of MDSCs to the implant, and current literature on MDSC phenotype (29). Interestingly, when the gene expression was normalized to the number of cells in each population, a phenotypic shift was observed. S100a8 was significantly upregulated in dendritic cells (DC) of tumor-bearing mice and S100a8 and S100a9 were significantly upregulated in macrophages of tumor-bearing mice, relative to their tumor-free counterparts (Fig. 3C; Supplementary Fig. S7A and S7B). This phenotypic shift in antigen-presenting cells, which reside as sentinels in tissues, suggests that multiple branches of the myeloid lineage are influenced by primary tumor conditioning of distal sites (30). In our data, diseased neutrophils cluster closely with MDSCs as they are also a source of S100a8, S100a9, Pglyrp1, and Ltf. However, individual neutrophils did not significantly change their expression of these genes, yet overall increases from neutrophils at the implants were observed because of their increased cell numbers following the onset of metastasis (Fig. 3C). The scRNA-seq data indicates that the genes most associated with MDSC phenotype (Fig. 3B) were also those genes that poorly aligned with blood leukocyte gene expression (Fig. 1H). Perhaps this is supported by previous studies showing that MDSCs can activate within metastatic niche distal sites (21), similar to our scRNA-seq data showing that a phenotypic shift occurs in antigen-presenting cells within the synthetic niche (Fig. 3C). This data on phenotypic shift of myeloid lineage cells in tissue biopsied from implants underlies the function of synthetic niche.

Figure 3.

Select innate immune populations drive phenotypic changes within the synthetic niche. At day 14, immune populations within the scaffolds of tumor-free control (TFC) and tumor-bearing (TB) mice were profiled by scRNA-seq, with a focus on the 10-gene panel derived from analysis of the whole microenvironment. A, T-distributed Stochastic Neighbor Embedding (tSNE) plot of innate and adaptive immune cells analyzed in implants during disease progression. B, Heatmap and unsupervised hierarchal clustering of aggregated gene expression of the total immune cell population in the tumor-bearing models. This data represents the expression level of a gene in a specific cell type as compared with other cell types (e.g., cumulative MDSCs express the highest amount of S100a8 and S100a9 compared with all other cell types). C, Violin plots of single-cell gene expression from tumor-free control and tumor-bearing mice including B cells, DCs, macrophages, monocytes, neutrophils, natural killer (NK) cells, and T cells. Each dot represents the gene expression (x-axis) of a single cell within a cell type and health state (tumor-bearing or tumor-free control; y-axis). Additional scRNA-seq violin plots for genes with little expression are shown in Supplementary Fig. S7B. Bmp15 is absent as it was undetected using scRNA-seq. D and E, Immune cells adoptively transferred from tumor-bearing mice readily infiltrate implants of tumor-free control and tumor-bearing mice. Gr1+Ly6G and Ly6G+ immune cells were isolated from spleens of tumor-bearing mice through magnetic-activated cell sorting, then labeled with DiO or DiD, respectively. Labeled immune cells were adoptively transferred into tumor-free control or tumor-bearing mice implanted with scaffolds. For tumor-bearing mice, tumors were inoculated 7 days prior to adoptive transfer, where approximately 30 tumor cells were expected in implants. An injection of PBS was used as an unlabeled control to establish the analytical background (12). Adoptively transferred DiO-labeled Gr1+Ly6G cells (C) and DiD-labeled Ly6G+ cells (D) readily trafficked into implants in tumor-free control and tumor-bearing mice. One-way ANOVA and post hoc Šidák multiple comparisons analysis with significance indicated as *, P < 0.01; ***, P < 0.0001; n.s., not significant. See Supplementary Fig. S4A–S4F for gating scheme and axes numerical increments.

Figure 3.

Select innate immune populations drive phenotypic changes within the synthetic niche. At day 14, immune populations within the scaffolds of tumor-free control (TFC) and tumor-bearing (TB) mice were profiled by scRNA-seq, with a focus on the 10-gene panel derived from analysis of the whole microenvironment. A, T-distributed Stochastic Neighbor Embedding (tSNE) plot of innate and adaptive immune cells analyzed in implants during disease progression. B, Heatmap and unsupervised hierarchal clustering of aggregated gene expression of the total immune cell population in the tumor-bearing models. This data represents the expression level of a gene in a specific cell type as compared with other cell types (e.g., cumulative MDSCs express the highest amount of S100a8 and S100a9 compared with all other cell types). C, Violin plots of single-cell gene expression from tumor-free control and tumor-bearing mice including B cells, DCs, macrophages, monocytes, neutrophils, natural killer (NK) cells, and T cells. Each dot represents the gene expression (x-axis) of a single cell within a cell type and health state (tumor-bearing or tumor-free control; y-axis). Additional scRNA-seq violin plots for genes with little expression are shown in Supplementary Fig. S7B. Bmp15 is absent as it was undetected using scRNA-seq. D and E, Immune cells adoptively transferred from tumor-bearing mice readily infiltrate implants of tumor-free control and tumor-bearing mice. Gr1+Ly6G and Ly6G+ immune cells were isolated from spleens of tumor-bearing mice through magnetic-activated cell sorting, then labeled with DiO or DiD, respectively. Labeled immune cells were adoptively transferred into tumor-free control or tumor-bearing mice implanted with scaffolds. For tumor-bearing mice, tumors were inoculated 7 days prior to adoptive transfer, where approximately 30 tumor cells were expected in implants. An injection of PBS was used as an unlabeled control to establish the analytical background (12). Adoptively transferred DiO-labeled Gr1+Ly6G cells (C) and DiD-labeled Ly6G+ cells (D) readily trafficked into implants in tumor-free control and tumor-bearing mice. One-way ANOVA and post hoc Šidák multiple comparisons analysis with significance indicated as *, P < 0.01; ***, P < 0.0001; n.s., not significant. See Supplementary Fig. S4A–S4F for gating scheme and axes numerical increments.

Close modal

The trafficking of innate immune cells to implants was investigated for its dependence on primary tumor conditioning of the implant. We adoptively transferred purified Gr1+Ly6G and Ly6G+ cells (isolated from tumor-bearing mice) into tumor-free control and tumor-bearing mice that had implanted scaffolds. Gr1+Ly6G and Ly6G+ cells were labeled prior to transfer with DiO and DiD lipophilic dyes, respectively. Scaffolds were analyzed for labeled immune cells at 36 hours after the transfer. Both labeled immune cell populations were present in implants of the tumor-free control and tumor-bearing mice at an equal order of magnitude. Compared with PBS controls, transferred Gr1+Ly6G cells were increased by 8.4-fold and 7.9-fold in tumor-free control and tumor-bearing mice, respectively (Fig. 3D). Transferred Ly6G+ cells were increased by 31-fold and 55-fold in tumor-free control and tumor-bearing mice, respectively (Fig. 3E). These results suggest that the foreign body response to the material largely determines immune recruitment to the scaffold, as supported by previous studies (17, 18), and that the preexisting condition of the implant does not selectively control immune cell trafficking at the implant. Thus, the similarities in trafficking and phenotypic shifts in the scRNA-seq data illustrate that the synthetic niche provides both a predefined site for disease-conditioned immune cells to aggregate, then further differentiate into metastasis-supportive populations.

Gene expression signatures monitor disease progression

We next developed a scoring system from our 10-gene panel that represented the progression of metastatic conditioning in tumor-bearing mice. We applied computational approaches to derive single-metric scores based on gene expression signatures from the tissue of biopsied implants, and thereby develop a predictive model for the probability that a mouse was either tumor bearing or a tumor-free control (Fig. 4). The computational reduction of gene expression signatures from primary tumor tissues to single-metric scores and predictions has guided clinical management of patients with breast cancer (31). Two computational techniques were investigated in parallel (Supplementary Fig. S8) as a spectrum of regression and modeling approaches provides a more robust interpretation of data and biological networks (32). First, data dimensionality was reduced via an unsupervised matrix factorization (singular value decomposition, SVD; ref. 33), which linearly transformed the data into eigengenes and eigenarrays that indicate gene significance and sample grouping, respectively. The SVD separation of the samples in three-dimensional subspace (Fig. 4A) established a scoring metric by setting a reference point at the centroid of the tumor-free controls and calculating Euclidean distance to each sample data point (Fig. 4B; x-axis of 4D), scores are scaled between 0 and 1. Second, supervised machine learning created a predictive model via a bootstrap aggregated (bagged) decision tree ensemble (i.e., random forest; ref. 34), which used a forest of decision trees based on a random selection of a gene from the 10-gene panel to predict a mouse status as tumor free or tumor bearing. For each sample, a prediction score (Fig. 4C; y-axis of 4D) was determined via leave-one-out analysis. In all models, the scores in tumor-bearing mice increased over time and corresponded with disease progression, while the scores in tumor-free control mice remained consistent (Fig. 4D). Thus, the metastatic conditioning of the implant was formulated into a trained model for monitor disease progression.

Figure 4.

Gene expression signatures reduced to scoring metrics and diagnostic predictions. A and B, Gene expression from analysis of implant-derived tissue was reduced to an unsupervised scoring metric through SVD, which was converted to a score by calculating Euclidean distance from the tumor-free control centroid to each sample. Scores were scaled between 0 and 1. C, In parallel, gene expression data were used to train a bootstrap aggregated (bagged) decision tree ensemble with leave-one-out cross-validation to predict the likelihood that a mouse was tumor bearing. D, Plot of the SVD score versus the bagged tree prediction. Dashed and solid line ellipses indicate the 99.9% confidence intervals for tumor-free control (TFC; n = 14) and tumor-bearing (TB; n = 14) cohorts, respectively. Filled ellipses indicate average and SEM for bagged tree and SVD data at days 7 (n = 3), 14 (n = 8), and 21 (n = 3). A two-way MANOVA showed a significant interaction effect between condition and time [Pillai trace = 0.568, F(4,44) = 4.361, P = 0.005]. Post hoc univariate ANOVA showed significant differences within the diseased cohort over time [indicated by #, df = (2,22), P < 0.001]. Simple effects analysis showed significant differences between tumor-bearing and time-matched tumor-free controls (indicated by *, Šidák adjusted P < 0.05; see Supplementary Data File S4 for exact F and P values).

Figure 4.

Gene expression signatures reduced to scoring metrics and diagnostic predictions. A and B, Gene expression from analysis of implant-derived tissue was reduced to an unsupervised scoring metric through SVD, which was converted to a score by calculating Euclidean distance from the tumor-free control centroid to each sample. Scores were scaled between 0 and 1. C, In parallel, gene expression data were used to train a bootstrap aggregated (bagged) decision tree ensemble with leave-one-out cross-validation to predict the likelihood that a mouse was tumor bearing. D, Plot of the SVD score versus the bagged tree prediction. Dashed and solid line ellipses indicate the 99.9% confidence intervals for tumor-free control (TFC; n = 14) and tumor-bearing (TB; n = 14) cohorts, respectively. Filled ellipses indicate average and SEM for bagged tree and SVD data at days 7 (n = 3), 14 (n = 8), and 21 (n = 3). A two-way MANOVA showed a significant interaction effect between condition and time [Pillai trace = 0.568, F(4,44) = 4.361, P = 0.005]. Post hoc univariate ANOVA showed significant differences within the diseased cohort over time [indicated by #, df = (2,22), P < 0.001]. Simple effects analysis showed significant differences between tumor-bearing and time-matched tumor-free controls (indicated by *, Šidák adjusted P < 0.05; see Supplementary Data File S4 for exact F and P values).

Close modal

Monitor response to surgical excision and early identification of responders

We subsequently investigated the regression of disease following therapy to test and validate the disease-monitoring capabilities of the synthetic niche. Gene expression of synthetic niche was profiled following excision of the tumor-bearing mammary gland. This procedure (analogous to a mastectomy in humans) was chosen as the intervention based on the role of surgery as a first-line treatment for breast cancer. We have previously reported that surgery alone in mice receiving scaffolds resulted in 40% long-term survival, while the remaining mice recurred (12). Following synthetic niche profiling, samples were then classified on the basis of the model developed in Fig. 4. Note, this study serves as validation because the trained model is based on progression of disease burden, while the test analysis is based on disease regression. Fourteen days following 4T1 tumor inoculation in BALB/c mice, tissue from the implant was biopsied (day 0 in Fig. 5A) and then immediately followed by an excision of the tumor-bearing mammary gland. Tumor-free control mice had a similar excision of the mammary gland. Implant biopsies were performed at weekly intervals following mammary gland excision (days 7, 14, and 21 post-excision in Fig. 5). Gene expression from the biopsies was quantified, and scores were derived from tumor-free control or tumor-bearing mice. Scores for each tumor-bearing animal showed a redirection toward the tumor-free control baseline 7 days after surgery for all mice (Fig. 5B). This difference was most evident at day 7 for the SVD (Fig. 5C) and at day 21 for the bagged tree prediction (Fig. 5D), with this difference originating from the distinct weighting of specific genes in the two computational approaches. We noted increased variability of scores at later time points with the SVD, which prompted an investigation of these metrics as a function of outcome.

Figure 5.

Primary tumor excision redirects signature and score trajectories, which predicts therapeutic efficacy. A, Mice with an array of implants and a primary tumor in their 4th mammary fat pad (n = 15 mice) had an implant surgically biopsied, then immediately had their primary tumor resected along with the surrounding fat pad (day 0, tumor burden equivalent to day 14 in Figs. 1 and 3). Tumor-free control (TFC) mice (n = 7) received sham inoculation with PBS and a mammary gland excision. All mice where then monitored by weekly (days 7, 14, and 21) biopsy of tissue from the implant, then analyzed by qRT-PCR and reduction to a predictive score. B–D, Signature analysis using the SVD and bagged tree algorithm that had been trained on the disease-progression group (Fig. 4) was used to classify each sample. Both SVD and bagged tree exhibited downward trends in the tumor-bearing cohort (n = 15) following therapeutic tumor resection. Next, the analysis was stratified on the basis of classification as therapy resistant (n = 9) or responsive (n = 6), as determined by survival monitoring following excision of the primary tumor. E–G, The predictive scores indicated a significant divergence between the resistant and responsive mice. Filled ellipses indicate the average and SEM for the bagged tree and SVD data at days 0, 7, 14, and 21. The cohort size for each group was decreased by one at day 21 due recurrence and animal censorship. For longitudinal data, statistics were performed via a linear mixed model. Post hoc simple effects analysis indicates significant differences (P < 0.05) between tumor-free control (*) and tumor bearing (TB; collective, not stratified by outcome), tumor bearing and tumor bearing day 0 (preexcision; #), tumor-free control and resistant (†), tumor-free control and responsive (‡), and resistant and responsive following significant (&; P < 0.05) or trending (P < 0.1) interactions in a two-way ANOVA (see Supplementary Data File S4 for exact F and P values).

Figure 5.

Primary tumor excision redirects signature and score trajectories, which predicts therapeutic efficacy. A, Mice with an array of implants and a primary tumor in their 4th mammary fat pad (n = 15 mice) had an implant surgically biopsied, then immediately had their primary tumor resected along with the surrounding fat pad (day 0, tumor burden equivalent to day 14 in Figs. 1 and 3). Tumor-free control (TFC) mice (n = 7) received sham inoculation with PBS and a mammary gland excision. All mice where then monitored by weekly (days 7, 14, and 21) biopsy of tissue from the implant, then analyzed by qRT-PCR and reduction to a predictive score. B–D, Signature analysis using the SVD and bagged tree algorithm that had been trained on the disease-progression group (Fig. 4) was used to classify each sample. Both SVD and bagged tree exhibited downward trends in the tumor-bearing cohort (n = 15) following therapeutic tumor resection. Next, the analysis was stratified on the basis of classification as therapy resistant (n = 9) or responsive (n = 6), as determined by survival monitoring following excision of the primary tumor. E–G, The predictive scores indicated a significant divergence between the resistant and responsive mice. Filled ellipses indicate the average and SEM for the bagged tree and SVD data at days 0, 7, 14, and 21. The cohort size for each group was decreased by one at day 21 due recurrence and animal censorship. For longitudinal data, statistics were performed via a linear mixed model. Post hoc simple effects analysis indicates significant differences (P < 0.05) between tumor-free control (*) and tumor bearing (TB; collective, not stratified by outcome), tumor bearing and tumor bearing day 0 (preexcision; #), tumor-free control and resistant (†), tumor-free control and responsive (‡), and resistant and responsive following significant (&; P < 0.05) or trending (P < 0.1) interactions in a two-way ANOVA (see Supplementary Data File S4 for exact F and P values).

Close modal

Scores were stratified on the basis of survival or recurrence outcomes (Fig. 5E–G). For mice that survived long-term post-excision, the regression toward the tumor-free control baseline continued over the subsequent 2 weeks (days 14 and 21 post-surgery), with no signs of recurrence through day 48. However, for the mice that developed disease recurrence, the scores remained significantly elevated above the tumor-free control baseline and significantly deviated from the scores of mice that survived long term. In the predictions obtained from the bagged decision tree, the average scores for mice that survived consistently indicated a lower degree of disease conditioning. The redirection in score and prediction trajectories of Fig. 5 is attributable to specific changes in gene expression. All genes from the panel with increased expression during tumor progression (S100a8, S100a9, Pglyrp1, Camp, Ltf, Chi3l3, and Ela2) regressed toward the tumor-free control baseline following tumor-bearing mammary gland excision (Fig. 6A). Interestingly, the three genes strongly associated with MDSC cell phenotype in our scRNA-seq data (S100a8, S100a9, and Pglyrp1; Fig. 3B), also demonstrated a pronounced separation between mice that remained disease free and those animals that developed a recurrence (Fig. 6B–E; Supplementary Fig. S9A–S9K). An interesting connection between data in Figs. 1F, 1H, 4B, and 6A–E is that S100a8, S100a9, Pglyrp1, and Ltf cluster together during disease progression, have the weakest correlation with blood, are upregulated in MDSCs, and are the earliest and most consistent predictors of survival in the therapeutic model. Results from the signature analysis, reduction to single-metric scores, and the gene expression components demonstrate that tissue from biopsied implants are a platform for monitoring context-specific pathological states.

Figure 6.

Gene expressions corresponding to the post-therapeutic signature analysis shows a cumulative regression toward a tumor-free state following tumor resection and a subsequent bifurcation between several genes as a function of therapeutic outcome. A, For the 7 genes that showed an increased expression in the metastatic onset model (Fig. 2), there was an organized regression in expression 1 week after therapy. Tumor-free control (TFC), n = 7; tumor-bearing (TB) mice, n = 15. B–E, Radar plots illustrate the bifurcation of specific genes as a function of therapeutic outcome, as determined by survival monitoring following excision of the primary tumor. While the decrease in expression continued for several genes, other genes like S100a8, S100a9, Pglyrp1, and Ltf showed a rebound and increase in the resistant cohort after the initial decrease. tumor-free control, n = 7; resistant, n = 9; and responsive, n = 6. Detailed statistical plots and exact F and P values are in Supplementary Fig. S9B–S9K and Supplementary Data File S4.

Figure 6.

Gene expressions corresponding to the post-therapeutic signature analysis shows a cumulative regression toward a tumor-free state following tumor resection and a subsequent bifurcation between several genes as a function of therapeutic outcome. A, For the 7 genes that showed an increased expression in the metastatic onset model (Fig. 2), there was an organized regression in expression 1 week after therapy. Tumor-free control (TFC), n = 7; tumor-bearing (TB) mice, n = 15. B–E, Radar plots illustrate the bifurcation of specific genes as a function of therapeutic outcome, as determined by survival monitoring following excision of the primary tumor. While the decrease in expression continued for several genes, other genes like S100a8, S100a9, Pglyrp1, and Ltf showed a rebound and increase in the resistant cohort after the initial decrease. tumor-free control, n = 7; resistant, n = 9; and responsive, n = 6. Detailed statistical plots and exact F and P values are in Supplementary Fig. S9B–S9K and Supplementary Data File S4.

Close modal

The synthetic metastatic niche is based on engineering a tissue within a porous scaffold that is inserted subcutaneously and is thus readily accessible for analysis, allowing for longitudinal monitoring without risk to a solid organ. Following treatment with surgery, chemo- and/or radiotherapy, scaffolds would be implanted into patients with aggressive cancer variants, who are at high risk for disease recurrence. Insertion and biopsy of a synthetic niche using a trocar would be comparable with procedures using a CNB. The microporous scaffold supports cell infiltration from the host, which becomes vascularized within days (35). In addition to the fibroblasts and endothelial cells, the biomaterial scaffold elicits a foreign body response that recruits immune cells from the vasculature. Immune cell infiltration is highly dynamic during the acute phase (1–2 weeks after implantation) and is followed by a chronic phase that is relatively stable. However, as the health status of the host changes, the immune cells and acellular material in the circulation may be altered, which alters recruitment to the scaffold and transforms the immune composition at the synthetic niche from that of a healthy individual to reflect the diseased state. This relationship between the foreign body response and disease state has emerged recently, with observations in the context of diabetes and neoplasia (25, 26). In the work focused on neoplasia, Oliva and colleagues demonstrated that tissue adhesives perform differently in healthy, inflamed, and neoplastic colon tissue, which were then used to inform the synthesis of disease-specific adhesives (25). Our data builds on these observations by demonstrating that the immune pathways in tissue localized at biomaterial implants may be exploited as a cancer diagnostic platform.

Changes in the gene expression of tissue biopsied from implants of tumor-bearing mice suggest the adoption of an immunosuppressive and hospitable environment for tumor cells, which is reflective of metastatic niche biology and primary tumor conditioning of distant vital organs. The increases in S100a8/9 are hallmarks of premetastatic and metastatic niche formation in metastasis-targeted organs (e.g., lungs) and MDSC immunosuppressive functionality (23). Accordingly, S100a8/9 expression leads to T-cell inhibitory/cytotoxic byproducts (e.g., reactive oxygen species) and tumor cell proliferation (36, 37). Others have shown that tumor cells, specifically 4T1 cells, lack production of S100a8/9 (38), thus our results reflect changes in immune cell dynamics and not the presence of metastatic tumor cells. An interesting discovery was that S100a8/9 was also upregulated in macrophage and DC populations, supporting that multiple myeloid cell lineages are involved establishing the metastatic niche. Unlike the increase in S100a8/9, the increase in Pglyrp1 has not yet been associated with the metastatic niche, although Pglyrp1 expression correlation with S100a8/9 was previously associated with bone marrow–derived granulocytic MDSCs and Th17 to regulatory T (Treg) cell transdifferentiation (39, 40). Increases in Ela2, Camp, and Ltf had the greatest overall magnitude changes within the implant tissue at day 21. Ltf and Camp are upregulated in granulocytic MDSCs as compared with neutrophils, and increased levels of Ela2 and Camp have been identified as drivers of metastasis, with Ela2 associated with poor clinical outcomes and endocrine treatment failure (29, 41). Chi3l3 expression progressively increased in the tissue infiltrate, and this finding is supported by recent reports that chitinase-knockout mice had decreased metastasis (42). A decrease in Ccr7 would limit the chemoattraction of T cells, reflecting the decrease in CD8+ and CD4+ T cells in tissue at days 14 and 21, respectively (12, 43). Ccl22 expression can be upregulated in metastatic target organs and attracts Tregs in oncogenesis, thus the decrease in Ccl22 expression during disease progression was novel and not anticipated (44). Data accessed from a gene expression repository suggests that expression of S100a8, S100a9, Pglyrp1, and Camp in the primary tumors of patients with breast cancer is prognostic, which points toward some conserved biological mechanisms (Supplementary Fig. S10A and S10B), yet these primary tumor measurements are limited to a single predictive measure and do not facilitate the monitoring of oncogenic changes over time. Collectively, data obtained from implants in the BALB/c-4T1 metastatic model indicate that the scaffold is conditioned by the primary tumor, leading to increased expression of genes associated with an immunosuppressive microenvironment.

Analysis of the scaffold provides data relevant to the elusive metastatic niche that is distinct from the information that obtained from a liquid biopsy. Information obtained from the metastatic niche could be used to generate tools that enable comprehensive analysis of the metastatic cascade to permit early detection. Primary tumor and sentinel lymph node data are attainable from tissue biopsies but are limited to providing only early stage disease findings. Liquid biopsies have utility, yet may reflect an intermediate or potentially clinically noncontributory step in the disease process. Cell migratory events, particularly events involving immune cells leaving the vascular and entering distal tissues, are associated with phenotypic changes. As the formation of the metastatic niche is stochastic in distal organs, the ability to predetermine the location of the metastatic niche enables the analysis of the microenvironments that support the survival of disseminated tumor cells. Importantly, for the 4T1 metastatic breast cancer model, this data showed that the conditioning of the implant microenvironment was similar to the conditioning of the lung. Interestingly, the gene panel had several common responses in two distinct models of breast cancer metastasis to the lung. Thus, these findings point toward the development of an environment that can induce a common lung-tropic cellular phenotype in recruited cancer cells (45). We do not anticipate that this implant replicates the biology of the lung, yet is capturing common aspects of metastatic conditioning. These results highlight the spectrum of immune states that sites distal from a primary tumor exhibit following therapy and during cancer progression. Through the targeting of early immunologic events using immune modulatory agents, these dynamics could potentially be exploited (46). In addition, these treatments could be used in conjunction with anticancer therapies that focus on the cellular weaknesses gleaned from the analysis of tumor cells captured within the scaffold. The expectation that combinatorial approaches would be necessary to address different cancer subtypes will help to realize the promise of personalized medicine.

Analysis of the scaffold following surgical excision of the primary tumor allowed for monitoring of response to this therapy, and stratified the tumor-bearing treatment group based on survival outcomes. Predictive scores were derived from gene expression signatures to track disease status and enable longitudinal monitoring of individual mice before and after surgical treatment. One of the most interesting observations was the sharp regression in tumor-promoting gene expression in the implants following cancer excision of the tumor-bearing mammary fat pad. On the basis of scRNA-seq data, this rapid change in gene expression likely results from changing immune cell populations or altered phenotypes within the implant. This dynamic nature of immune cells at the implant underpins the concept of the synthetic niche, which could inform the initiation of novel therapeutic interventions. In the last year, novel therapies have become available for the treatment of hormone insensitive and HER2 receptor–negative breast cancer. Olaparib, a PARP inhibitor, has shown efficacy in extending progression-free survival for patients with metastatic, BRCA mutation–positive, and triple-negative breast cancer (47). In addition, an anti-PD-L1 antibody, atezolizumab, was also recently found to be efficacious in extending survival for patients with PD-L1–expressing metastatic triple-negative breast cancer (48). Accelerating the initiation time of these promising treatments, through early-detection diagnostic platforms, could improve disease-specific outcomes by allowing therapeutics to function when the burden of metastatic disease and heterogeneity of cancer mutations are relatively low. Furthermore, the niche can be analyzed for the presence of immunosuppressive macrophages and DCs, which presents unique diagnostic avenues and therapeutic targets for tumors that are driven by these myeloid subsets and complements efforts to characterize and overcome MDSC immunosuppression.

The application of tissue engineering for cancer diagnostics represents a novel approach to engineer a system that reports on early metastatic sites. Our data indicate the ability to capture disease progression or regression, suggesting that the foreign body response has the potential to capture both immune activation and suppression, which may enable their application to a range of immunologic disorders. Collectively, our data indicate that the tissue within implants is dynamic, with a context-dependent profile that reflects disease course and response to tumor excision. Broadly, the concept of synthesizing tissues that are difficult to assess, such as the lung or central nervous system tissue, may enable diagnostics that have broad application for monitoring immunologic diseases.

R.S. Oakes has ownership interest in patent applications (U.S. Application No. 62/569,460, U.S. Application No. 62/571,702, and International Application No. PCT/US18/54859). R.M. Hartfield is a medical writer (paid consultant) at Biodesix, Inc. L.D. Shea is a paid consultant at Cour Pharmaceutical and has ownership interest (including patents) in University of Michigan. No potential conflicts of interest were disclosed by the other authors.

Conception and design: R.S. Oakes, G.G. Bushnell, P. Kandagatla, M.S. Hall, J.S. Jeruss, L.D. Shea

Development of methodology: R.S. Oakes, S.M. Orbach, P. LaFaire, J.S. Jeruss, L.D. Shea

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): R.S. Oakes, S.M. Orbach, P. Kandagatla, Y. Zhang, A.H. Morris, M.S. Hall, P. LaFaire, R.M. Hartfield, M.D. Brooks, J.S. Jeruss

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): R.S. Oakes, G.G. Bushnell, S.M. Orbach, P. Kandagatla, Y. Zhang, A.H. Morris, J.T. Decker, M.S. Wicha, J.S. Jeruss, L.D. Shea

Writing, review, and/or revision of the manuscript: R.S. Oakes, G.G. Bushnell, S.M. Orbach, P. Kandagatla, Y. Zhang, A.H. Morris, M.S. Hall, R.M. Hartfield, M.S. Wicha, J.S. Jeruss, L.D. Shea

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): R.S. Oakes

Study supervision: R.S. Oakes, J.S. Jeruss, L.D. Shea

Authors thank Jeff Riskin and Andrew Ao, MS for their technical assistance during implant fabrication and surgical procedures, Ravi Raghani for assistance with tumor growth monitoring, Angelica R. Galvan for her technical assistance and expertise in imaging the scaffold microstructure with SEM, Christopher Krebs, PhD, and the University of Michigan DNA Sequencing Core for assistance in OpenArray and scRNA-seq processing and analysis, Indika Rajapakse, PhD, for suggesting the use of SVD, the Luker Laboratory at the University of Michigan for donation of the E0771 GFP+ parental line, the NIH for their support through 5R01CA173745-06. G.G. Bushnell is a recipient of the NSF Graduate Research Fellowship and NIH NRSA F31CA224982-01. P. Kandagatla is a recipient of the NIH Institutional Research Training Grant T32CA009672. A.H. Morris is funded by a Michigan Life Sciences Fellowship and a Michigan Precision Health Scholars grant. P. LaFaire is a recipient of the NIH Short-Term Institutional Research Training Grant T35HL007690.

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.

1.
Harris
LN
,
Ismaila
N
,
McShane
LM
,
Andre
F
,
Collyar
DE
,
Gonzalez-Angulo
AM
, et al
Use of biomarkers to guide decisions on adjuvant systemic therapy for women with early-stage invasive breast cancer: American Society of Clinical Oncology Clinical Practice Guideline
.
J Clin Oncol
2016
;
34
:
1134
50
.
2.
Paik
S
,
Shak
S
,
Tang
G
,
Kim
C
,
Baker
J
,
Cronin
M
, et al
A multigene assay to predict recurrence of tamoxifen-treated, node-negative breast cancer
.
N Engl J Med
2004
;
351
:
2817
26
.
3.
van 't Veer
LJ
,
Dai
H
,
van de Vijver
MJ
,
He
YD
,
Hart
AA
,
Mao
M
, et al
Gene expression profiling predicts clinical outcome of breast cancer
.
Nature
2002
;
415
:
530
6
.
4.
Ciriello
G
,
Gatza
ML
,
Beck
AH
,
Wilkerson
MD
,
Rhie
SK
,
Pastore
A
, et al
Comprehensive molecular portraits of invasive lobular breast cancer
.
Cell
2015
;
163
:
506
19
.
5.
Parker
JS
,
Mullins
M
,
Cheang
MC
,
Leung
S
,
Voduc
D
,
Vickery
T
, et al
Supervised risk predictor of breast cancer based on intrinsic subtypes
.
J Clin Oncol
2009
;
27
:
1160
7
.
6.
Ma
X-J
,
Wang
Z
,
Ryan
PD
,
Isakoff
SJ
,
Barmettler
A
,
Fuller
A
, et al
A two-gene expression ratio predicts clinical outcome in breast cancer patients treated with tamoxifen
.
Cancer Cell
2004
;
5
:
607
16
.
7.
Dubsky
P
,
Brase
JC
,
Jakesz
R
,
Rudas
M
,
Singer
CF
,
Greil
R
, et al
The EndoPredict score provides prognostic information on late distant metastases in ER+/HER2− breast cancer patients
.
Br J Cancer
2013
;
109
:
2959
64
.
8.
Minn
AJ
,
Kang
Y
,
Serganova
I
,
Gupta
GP
,
Giri
DD
,
Doubrovin
M
, et al
Distinct organ-specific metastatic potential of individual breast cancer cells and primary tumors
.
J Clin Invest
2005
;
115
:
44
55
.
9.
Luzzi
KJ
,
MacDonald
IC
,
Schmidt
EE
,
Kerkvliet
N
,
Morris
VL
,
Chambers
AF
, et al
Multistep nature of metastatic inefficiency: dormancy of solitary cells after successful extravasation and limited survival of early micrometastases
.
Am J Pathol
1998
;
153
:
865
73
.
10.
Zhu
Z
,
Qiu
S
,
Shao
K
,
Hou
Y
. 
Progress and challenges of sequencing and analyzing circulating tumor cells
.
Cell Biol Toxicol
2018
;
34
:
405
15
.
11.
Azarin
SM
,
Yi
J
,
Gower
RM
,
Aguado
BA
,
Sullivan
ME
,
Goodman
AG
, et al
In vivo capture and label-free detection of early metastatic cells
.
Nat Commun
2015
;
6
:
8094
.
12.
Rao
SS
,
Bushnell
GG
,
Azarin
SM
,
Spicer
G
,
Aguado
BA
,
Stoehr
JR
, et al
Enhanced survival with implantable scaffolds that capture metastatic breast cancer cells in vivo
.
Cancer Res
2016
;
76
:
5209
18
.
13.
Aguado
BA
,
Caffe
JR
,
Nanavati
D
,
Rao
SS
,
Bushnell
GG
,
Azarin
SM
, et al
Extracellular matrix mediators of metastatic cell colonization characterized using scaffold mimics of the pre-metastatic niche
.
Acta Biomater
2016
;
33
:
13
24
.
14.
Bersani
F
,
Lee
J
,
Yu
M
,
Morris
R
,
Desai
R
,
Ramaswamy
S
, et al
Bioengineered implantable scaffolds as a tool to study stromal-derived factors in metastatic cancer models
.
Cancer Res
2014
;
74
:
7229
38
.
15.
Aguado
BA
,
Hartfield
RM
,
Bushnell
GG
,
Decker
JT
,
Azarin
SM
,
Nanavati
D
, et al
Biomaterial scaffolds as pre-metastatic niche mimics systemically alter the primary tumor and tumor microenvironment
.
Adv Healthc Mater
2018
;
7
:
e1700903
.
16.
Lee
J
,
Li
M
,
Milwid
J
,
Dunham
J
,
Vinegoni
C
,
Gorbatov
R
, et al
Implantable microenvironments to attract hematopoietic stem/cancer cells
.
Proc Natl Acad Sci U S A
2012
;
109
:
19638
43
.
17.
Bushnell
GG
,
Hardas
TP
,
Hartfield
RM
,
Zhang
Y
,
Oakes
RS
,
Ronquist
S
, et al
Biomaterial scaffolds recruit an aggressive population of metastatic tumor cells in vivo
.
Cancer Res
2019
;
79
:
2042
53
.
18.
Bushnell
GG
,
Rao
SS
,
Hartfield
RM
,
Zhang
Y
,
Oakes
RS
,
Jeruss
JS
, et al
Microporous scaffolds loaded with immunomodulatory lentivirus to study the contribution of immune cell populations to tumor cell recruitment in vivo
.
Biotechnol Bioeng
2019
Sep 23
[Epub ahead of print]
.
19.
Bushnell
GG
,
Hong
X
,
Hartfield
RM
,
Zhang
Y
,
Oakes
RS
,
Rao
SS
, et al
High frequency spectral ultrasound imaging to detect metastasis in implanted biomaterial scaffolds
.
Ann Biomed Eng
2019
Sep 23
[Epub ahead of print]
.
20.
Kaplan
RN
,
Riba
RD
,
Zacharoulis
S
,
Bramley
AH
,
Vincent
L
,
Costa
C
, et al
VEGFR1-positive haematopoietic bone marrow progenitors initiate the pre-metastatic niche
.
Nature
2005
;
438
:
820
7
.
21.
Giles
AJ
,
Reid
CM
,
Evans
JD
,
Murgai
M
,
Vicioso
Y
,
Highfill
SL
, et al
Activation of hematopoietic stem/progenitor cells promotes immunosuppression within the pre-metastatic niche
.
Cancer Res
2016
;
76
:
1335
47
.
22.
Hiratsuka
S
,
Watanabe
A
,
Aburatani
H
,
Maru
Y
. 
Tumour-mediated upregulation of chemoattractants and recruitment of myeloid cells predetermines lung metastasis
.
Nat Cell Biol
2006
;
8
:
1369
75
.
23.
Sinha
P
,
Okoro
C
,
Foell
D
,
Freeze
HH
,
Ostrand-Rosenberg
S
,
Srikrishna
G
. 
Proinflammatory S100 proteins regulate the accumulation of myeloid-derived suppressor cells
.
J Immunol
2008
;
181
:
4666
75
.
24.
Kaplan
RN
,
Psaila
B
,
Lyden
D
. 
Bone marrow cells in the ‘pre-metastatic niche': within bone and beyond
.
Cancer Metastasis Rev
2006
;
25
:
521
9
.
25.
Oliva
N
,
Carcole
M
,
Beckerman
M
,
Seliktar
S
,
Hayward
A
,
Stanley
J
, et al
Regulation of dendrimer/dextran material performance by altered tissue microenvironment in inflammation and neoplasia
.
Sci Transl Med
2015
;
7
:
272ra11
.
26.
Socarras
TO
,
Vasconcelos
AC
,
Campos
PP
,
Pereira
NB
,
Souza
JP
,
Andrade
SP
. 
Foreign body response to subcutaneous implants in diabetic rats
.
PLoS One
2014
;
9
:
e110945
.
27.
Vegas
AJ
,
Veiseh
O
,
Doloff
JC
,
Ma
M
,
Tam
HH
,
Bratlie
K
, et al
Combinatorial hydrogel library enables identification of materials that mitigate the foreign body response in primates
.
Nat Biotechnol
2016
;
34
:
345
52
.
28.
Kveler
K
,
Starosvetsky
E
,
Ziv-Kenet
A
,
Kalugny
Y
,
Gorelik
Y
,
Shalev-Malul
G
, et al
Immune-centric network of cytokines and cells in disease context identified by computational mining of PubMed
.
Nat Biotechnol
2018
;
36
:
651
9
.
29.
Youn
JI
,
Collazo
M
,
Shalova
IN
,
Biswas
SK
,
Gabrilovich
DI
. 
Characterization of the nature of granulocytic myeloid-derived suppressor cells in tumor-bearing mice
.
J Leukoc Biol
2012
;
91
:
167
81
.
30.
Ehrchen
JM
,
Sunderkotter
C
,
Foell
D
,
Vogl
T
,
Roth
J
. 
The endogenous Toll-like receptor 4 agonist S100A8/S100A9 (calprotectin) as innate amplifier of infection, autoimmunity, and cancer
.
J Leukoc Biol
2009
;
86
:
557
66
.
31.
Kwa
M
,
Makris
A
,
Esteva
FJ
. 
Clinical utility of gene-expression signatures in early stage breast cancer
.
Nat Rev Clin Oncol
2017
;
14
:
595
610
.
32.
Janes
KA
,
Lauffenburger
DA
. 
A biological approach to computational models of proteomic networks
.
Curr Opin Chem Biol
2006
;
10
:
73
80
.
33.
Alter
O
,
Brown
PO
,
Botstein
D
. 
Singular value decomposition for genome-wide expression data processing and modeling
.
Proc Natl Acad Sci U S A
2000
;
97
:
10101
6
.
34.
Breiman
L
. 
Random forests
.
Mach Learn
2001
;
45
:
5
32
.
35.
Jang
JH
,
Rives
CB
,
Shea
LD
. 
Plasmid delivery in vivo from porous tissue-engineering scaffolds: transgene expression and cellular transfection
.
Mol Ther
2005
;
12
:
475
83
.
36.
Cheng
P
,
Corzo
CA
,
Luetteke
N
,
Yu
B
,
Nagaraj
S
,
Bui
MM
, et al
Inhibition of dendritic cell differentiation and accumulation of myeloid-derived suppressor cells in cancer is regulated by S100A9 protein
.
J Exp Med
2008
;
205
:
2235
49
.
37.
Hiratsuka
S
,
Watanabe
A
,
Sakurai
Y
,
Akashi-Takamura
S
,
Ishibashi
S
,
Miyake
K
, et al
The S100A8-serum amyloid A3-TLR4 paracrine cascade establishes a pre-metastatic phase
.
Nat Cell Biol
2008
;
10
:
1349
55
.
38.
Becker
A
,
Grosse Hokamp
N
,
Zenker
S
,
Flores-Borja
F
,
Barzcyk
K
,
Varga
G
, et al
Optical in vivo imaging of the alarmin S100A9 in tumor lesions allows for estimation of the individual malignant potential by evaluation of tumor-host cell interaction
.
J Nucl Med
2015
;
56
:
450
6
.
39.
Ouzounova
M
,
Lee
E
,
Piranlioglu
R
,
El Andaloussi
A
,
Kolhe
R
,
Demirci
MF
, et al
Monocytic and granulocytic myeloid derived suppressor cells differentially regulate spatiotemporal tumour plasticity during metastatic cascade
.
Nat Commun
2017
;
8
:
14979
.
40.
Downs-Canner
S
,
Berkey
S
,
Delgoffe
GM
,
Edwards
RP
,
Curiel
T
,
Odunsi
K
, et al
Suppressive IL-17A+Foxp3+ and ex-Th17 IL-17AnegFoxp3+ Treg cells are a source of tumour-associated Treg cells
.
Nat Commun
2017
;
8
:
14649
.
41.
Weber
G
,
Chamorro
CI
,
Granath
F
,
Liljegren
A
,
Zreika
S
,
Saidak
Z
, et al
Human antimicrobial protein hCAP18/LL-37 promotes a metastatic phenotype in breast cancer
.
Breast Cancer Res
2009
;
11
:
R6
.
42.
Kim
DH
,
Park
HJ
,
Lim
S
,
Koo
JH
,
Lee
HG
,
Choi
JO
, et al
Regulation of chitinase-3-like-1 in T cell elicits Th1 and cytotoxic responses to inhibit lung metastasis
.
Nat Commun
2018
;
9
:
503
.
43.
Muller
A
,
Homey
B
,
Soto
H
,
Ge
N
,
Catron
D
,
Buchanan
ME
, et al
Involvement of chemokine receptors in breast cancer metastasis
.
Nature
2001
;
410
:
50
6
.
44.
Curiel
TJ
,
Coukos
G
,
Zou
L
,
Alvarez
X
,
Cheng
P
,
Mottram
P
, et al
Specific recruitment of regulatory T cells in ovarian carcinoma fosters immune privilege and predicts reduced survival
.
Nat Med
2004
;
10
:
942
9
.
45.
Minn
AJ
,
Gupta
GP
,
Siegel
PM
,
Bos
PD
,
Shu
W
,
Giri
DD
, et al
Genes that mediate breast cancer metastasis to lung
.
Nature
2005
;
436
:
518
24
.
46.
Oakes
RS
,
Froimchuk
E
,
Jewell
CM
. 
Engineering biomaterials to direct innate immunity
.
Adv Ther
2019
;
2
. .
47.
Robson
M
,
Im
SA
,
Senkus
E
,
Xu
B
,
Domchek
SM
,
Masuda
N
, et al
Olaparib for metastatic breast cancer in patients with a germline BRCA mutation
.
N Engl J Med
2017
;
377
:
523
33
.
48.
Schmid
P
,
Adams
S
,
Rugo
HS
,
Schneeweiss
A
,
Barrios
CH
,
Iwata
H
, et al
IMpassion130: results from a global, randomised, double-blind, phase III study of atezolizumab (atezo) + nab-paclitaxel (nab-P) vs. placebo + nab-P in treatment-naive, locally advanced or metastatic triple-negative breast cancer (mTNBC)
.
Ann Oncol
2018
;
29
:
viii707–viii708
.

Supplementary data