Pleiotropic Mechanisms Drive Endocrine Resistance in the Three-Dimensional Bone Microenvironment

Systemic endocrine therapies, which encompass HR blockade with small-molecule antagonists (e.g., antiestrogens and antiandrogens for ER and PR, respectively), can prolong disease-free and overall survival for patients with hormone receptor–positive (HR⁺) breast cancer/prostate cancer. However, some initially sensitive HR⁺ tumors acquire hormonal therapy (HT) resistance and others exhibit de novo resistance (reviewed in refs. 1–3).

Potential HT resistance mechanisms in breast cancer include the deregulation of ER coactivators/corepressors, modulation of ER activity by growth factor receptor pathways, and deregulation of cell-cycle signaling molecules (1, 4, 5). Castration-resistant prostate cancer (CRPC) can involve multiple nonmutually exclusive mechanisms including AR overexpression though amplification of the AR gene (6) and/or its enhancer region (7), suppression of miRNAs downregulating AR expression (8, 9), AR splice variants (10), and increased expression or activation of AR-binding partners (11, 12). Another common mechanism of HT resistance involves mutations in ER or AR genes (13, 14). Second-generation HR antagonists (e.g., fulvestrant and enzalutamide) can overcome HT resistance mediated by incomplete blockade of hormone signaling (15, 16). However, acquisition of hormone independence remains a common mechanism of HT resistance. Alternative cancer cell growth signals, including amplification/overexpression of HER2 and acquisition of activating mutations in the PI3K/Akt signaling pathway or PTEN loss, are frequently observed in HT-resistant breast cancer/prostate cancer (17–20). These preclinical and clinical observations raise the possibility that loss of HR expression or attenuation of its activity may serve as alternate means of escaping HR-targeted therapies, plausibly through concomitant activation of compensatory signaling pathways.

As new targeted therapies are being developed against HT-resistant breast cancer and prostate cancer tumors (21–23), identifying response predictors is essential for their rational use. However, putative biomarkers in primary tumors may not account for potential HT resistance mechanisms operating in metastatic tumors. Furthermore, preclinical modeling of HT resistance mechanisms has principally involved in vitro cultures of cancer cells on 2D plastic surfaces: these conventional systems do not necessarily recapitulate the biological properties of HR⁺ cancer cells growing in 3D anchorage-independent conditions and their evasion of anoikis, a key property acquired during malignant transformation (24, 25). Distinctly from 2D cultures, 3D cultures and coculture systems, which involve nonmalignant cells of the metastatic milieu, can better recapitulate the cancer architecture and the heterotypic cellular interactions within the physical space of the tumor microenvironment (26), which can putatively affect the response of cancer cells to HT.

Bone is the most common metastatic site for HR⁺ breast cancer and prostate cancer. To assess how the cellular, structural and paracrine elements of the bone metastatic milieu can simultaneously interact with the tumor cells and affect their sensitivity to HT, we developed 3D cocultures of breast cancer/prostate cancer spheroids and bone marrow stromal cells (BMSC). We observed that HR-targeting agents have modest effect against cancer cells in the extracellular matrix (ECM) niche, but specifically abrogate the anchorage-independent growth of HR⁺ breast cancer/prostate cancer cells, a hallmark of tumorigenesis. Importantly, BMSCs restore the ability of HR⁺ tumor cells to evade anoikis in the presence of HT through distinct and context-dependent mechanisms. Our results highlight that tumor–stroma interactions in the metastatic microenvironment are a potential source of HT resistance in patients with breast cancer/prostate cancer.

Cancer cell lines (MCF7, T47D, ZR75-1, and LNCaP) and immortalized nonmalignant cells lines HS5, HS27A, HOBIT, hFOB, THLE3, and SVGP12 were originally obtained from ATCC. Lenti-X-293T cells were purchased from Clontech. The MCF7 cells stably expressing doxycycline-inducible mutant ER (Y537S) and the respective WT control were established by the laboratory of Dr. Myles Brown, as described previously (27). The cells were used for up to 15 passages after thawing. The identity of the cell lines was confirmed by short tandem repeat analysis. The cells were last tested for the presence of mycoplasma in May 2019, using the MycoAlert Mycoplasma Detection Kit (Lonza). The culture conditions and reagents used in this study are further described in the Supplementary Materials.

The breast cancer and prostate cancer cells were cultured in 3D conditions as described previously (28). Briefly, cells were suspended in 80% Matrigel or in rat tail collagen (1 mg/mL; prepared according to the manufacturer's instructions), and liquid gels were casted in either 12-well (for histologic and molecular analyses) or multi-well plates (for cell viability assays; described below). Gels were incubated for 1 hour at 37°C to allow for solidification, supplemented with medium, and used for further downstream applications (see other sections of Materials and Methods). To extract cells from 3-D organoids, gels were first disrupted mechanically in cold PBS; next, spheroids were isolated by centrifugation and incubated in trypsin for 20 to 30 minutes to generate single-cell solution. When applicable, separation of tumor and stromal cell subpopulations was performed using mesenchymal cells depletion microbeads (cat. # 130-050-601; Miltenyi Biotec) and magnetic columns, according to the manufacturer's instructions.

Cell lines MCF-7, T47D, ZR-75-1, LNCaP, and prostate cancer patient-derived cells MSK-PCa1, MSK-PCa2, and MSK-PCa3 were stably transfected with lentiviral vectors expressing the luciferase gene, allowing the use of the emitted bioluminescence to longitudinally and nondisruptively measure cancer cell viability in the presence of luciferase-negative nonmalignant cells (compartment-specific bioluminescence imaging, CS-BLI; ref. 29). For 2D culture cell viability assays, the cells were seeded in 384-well plates (500–2,000 cells/well), in supplemented media (50 μL–100 μL/well) and incubated for 24 hours prior to addition of compound/s at indicated concentrations. For 3D spheroid cell viability assays, cells were suspended in Matrigel, collagen, or mixture of the two matrices, at a density of 10⁵ cells/ml, and plated in 384-well plates. After 24 hours, antiestrogens were added in the medium at the indicated final concentrations. At each indicated time point, either luciferin (for the luciferase-positive cells; concentration per the manufacturer's instructions) or CTG reagent (per the manufacturer's instructions) was added to each well, and the plates were read using a microplate reader (BioTek Synergy 2). Cell viability assays were performed in quadruplicates and error bars represent SEM.

The rate of cell apoptosis in 2D and 3D cultures was estimated by luminescence measurement, using the RealTime-Glo Annexin V Apoptosis assay (Promega) following the manufacturer's instructions. Briefly, the cells were incubated with two Annexin V fusion proteins (Annexin V-LgBiT and Annexin V-SmBiT), which contain complementary subunits of NanoBiT Luciferase, and a time-released luciferase. The cell surface exposed phosphatidylserine in apoptotic cells, bringing the Annexin V-LgBiT and Annexin V-SmBiT luciferase subunits into complementing proximity, which is reflected by the strength of bioluminescence signal emitted in culture (quadruplicate measurements).

Cells suspended in Matrigel were casted in 12-well plates (1.5 mL/well) and gels were let to solidify for 1 hour at 37°C. After gel solidification, the 3D cultures were supplemented with medium and incubated for 7 to 8 days to allow for organoid formation and antiestrogen/antiandrogen treatment as indicated. Gels were then scooped out of the culture vessel and fixed in 10% formalin. Subsequently, gels were processed for paraffin embedding and IHC (slide sections), as described previously (28).

A CRISPR/Cas9–based gene editing screen of MCF7 cells in the presence versus absence of 4-hydroxytamoxifen (4-OHT) was performed similarly to prior studies (further details in the Supplementary Material; ref. 30).

Two-photon excited fluorescence imaging of 3D spheroids was performed as described previously (further details in the Supplementary Material; ref. 31).

The in vivo experiments were conducted in accordance with the guidelines of the DFCI Institutional Animal Care and Use Committee. MCF7 cells alone or mixed with HS5 cells were suspended in medium containing 50% Matrigel and injected subcutaneously in the flank of NU(NCr)- Foxn1nu mice implanted with estradiol pellets (5 × 10⁶ cells of each type per injection). When the average tumor diameter reached approximately 1 cm, mice were separated into treatment versus control groups as indicated and treated with either tamoxifen (3 mg/kg in corn oil, biweekly subcutaneous injections) or vehicle. Tumor growth and response to treatment was quantified by in vivo bioluminescence measurements. Statistical analyses (GraphPad Prism 8.2.0) compared all time-matched mean bioluminescence values between treatment versus vehicle cohorts via two-way ANOVA with correction for multiple comparisons (Sidak test, default option of the statistical package).

The cells in 2D cultures were collected by scraping in cold cell recovery solution, pelleted by centrifugation and stored at −80°C. The 3D gels were cultured for 8 to 10 days after seeding to allow for spheroid formation, followed by mechanical disruption using cold PBS and centrifugation to collect the cells. The breast cancer cells were separated from BMSCs as described above (breast cancer cells from monocultures were also subjected to the same protocol, as controls), counted, and stored at −80°C. The RNA was isolated using Qiagen RNA Isolation Kit (experimental triplicates). The RNA isolated from collagen gels was analyzed by hybridization in Affymetrix Prime View 3′ microarrays at the Molecular Biology Core Facility (Dana-Farber Cancer Institute, Boston, MA) and further evaluated for differential gene expression analysis for microarray data of each cell line in 2D and 3D organoid models using the bioconductor LIMMA packageA. The samples isolated from Matrigel cultures were supplemented with ERCC RNA spike-in mix to adjust for the cell number and analyzed by RNA sequencing using Illumina NextSeq 500 Next Gen at the Molecular Biology Core Facility (Dana-Farber Cancer Institute), followed by differential gene expression analysis for the RNA sequencing (RNA-seq) data using the bioconductor DEseq2 package. The data have been deposited in the Gene Expression Omnibus database (GEO accession number GSE152312).

Gene set enrichment analysis (GSEA) analysis was performed using the GSEA software version 3.0 (gsea-3.0.jar) based on the preranked option. For each differential gene expression analysis, we used the per gene fold change for gene ranking. For the analysis of microarray datasets, if a gene was represented by multiple probes, we used the fold change of the probe with the smallest nominal P value. For visualizations, we used the GSEA enrichment scores (ES) and the gene set size normalized enrichment scores (NES). For genes with transcripts upregulated (log₂FC > 2) in MCF7 spheroids cocultured in Matrigel with BMSCs (vs. monoculture) and with sgRNA significantly (P < 0.05) depleted in CRISPR-Cas9 screen in the presence of 4-OHT, we queried the C6 Oncogenic Signatures (representing 189 gene sets deregulated in cancer) of the Molecular Signatures Database (MSigDB https://www.gsea-msigdb.org/gsea/msigdb/annotate.jsp; last accessed on February 12, 2020).

To assess the translational relevance of our 3D coculture models, we used the transcriptional profiles of MCF7 spheroids co-cultured with BMSCs in Matrigel or collagen gels (vs. respective monocultures). We initially established a consensus molecular signature comprised of genes that were commonly upregulated (log₂FC > 1) or commonly downregulated (log₂FC < −1) in both Matrigel and collagen cocultures (53 and 27 genes, respectively). These differentially expressed genes were initially used to stratify the n = 1,445 ER⁺ patients in the METABRIC dataset (last accessed on May 28, 2019) and examine potential differences in clinical outcome between the patients with high versus low BMSC-induced MCF7 molecular signature. Next, the correlation of MCF7 spheroid gene profiling data with breast cancer patient outcome was also confirmed using the online meta-analysis tool KM plotter (which aggregates data from several clinical trials; http://kmplot.com/analysis/; ref. 32) for patients with ER⁺ breast cancer. To assess the effects of genes of interest in KM plotter (last accessed on March 2, 2020), we used the mean expression of the upregulated or the downregulated genes in cocultures to analyze data from n = 355 patients with ER⁺ cancer that received systemic endocrine therapy; quality control and other tool options were used at default setting; and compared the relapse-free survival of patients of the upper tertile versus lower two tertiles of expression of the respective gene signatures. For the METABRIC data, we compared the disease-specific survival of patients in the upper tertile versus lower two tertiles of expression of the respective gene signatures using log-rank (Mantel–Cox) test (GraphPad Prism 8.2.0). Similar results were obtained in KM plotter and METABRIC databases by analyses of additional stroma-induced signatures from the Matrigel or collagen cocultures, and/or with use of additional cut-off points to stratify patients with low versus high expression of these respective signatures.

To assess how HR signaling affects anchorage-independent cancer cell growth, we cultured ER⁺ breast cancer cells in ECM, where tumor cells grow as 3D spheroids by evading anoikis, a form of apoptosis in matrix-detached epithelial cells that enables lumen formation (24, 33). Each spheroid was formed by one initial cell, with the outer cellular layer attached to the ECM and the centrally localized cells growing in matrix-detached conditions. Exposure of MCF7 3D spheroids to antiestrogens prevented anchorage-independent growth, leading to lumen clearance and acinar differentiation (Fig. 1A). Antiestrogens also induced apoptosis in 3D (but not 2D) cultures (Fig. 1B), corroborating the morphologic observations. Antiestrogen treatment had only modest effect on the peripheral ECM-attached breast cancer cells, suggesting that cell adhesion–mediated signaling attenuates the effect of antiestrogens in breast cancer cells. After antiestrogen washout, the peripheral cells resumed lumen filling, indicating that the malignant potential is retained in the ECM niche during treatment (Fig. 1A). The survival of anchorage-independent cancer cells depends on suppression of redox stress by oncogenic signals (33). To measure the redox stress within breast cancer spheroids we used two-photon excited fluorescence metabolic imaging (31). Antiestrogens significantly increased redox stress in the interior of 3D spheroids within 24 hours (Fig. 1C and D), suggesting that ER contributes in redox homeostasis during the acquisition of anchorage independence by cancer cells. Indeed, N-acetyl cysteine (NAC), a ROS-neutralizing antioxidant that can prevent anoikis in the center of acini (33), could rescue the MCF7 cells treated with fulvestrant in 3D cultures, but had negligible effect in 2D cultures (Fig. 1E). These observations suggest that ER signaling is necessary for lumen filling by ER⁺ breast cancer cells, and that HT induces anoikis through selective cytotoxic effect on matrix-detached tumor cells.

Bone metastases are frequent in HR⁺ breast cancer and prostate cancer. To model these lesions, we cocultured breast cancer cells and nonmalignant cells of the bone microenvironment in 3D conditions in the presence versus absence of antiestrogens. We adapted our previously developed CS-BLI approach (29), which involves a bioluminescence-based viability method to selectively quantify the cancer cell viability in the presence versus absence of nonmalignant cells from the bone milieu (e.g., osteoblasts and BMSCs). Some of these nonmalignant cells (e.g., HS5 BMSCs) mitigated the antiestrogen-mediated suppression of breast cancer growth in 3D cultures (Supplementary Fig. S1A). Notably, BMSCs abrogated the effect of antiestrogen treatment on inducing redox stress in anchorage-independent breast cancer cells (Fig. 1D) and acinar-like differentiation of cancer spheroids (Fig. 1F). CS-BLI assays confirmed that BMSCs attenuated the effect of antiestrogens on breast cancer spheroid growth in Matrigel (Fig. 2A), concordant with the morphologic observations about BMSC-induced rescue of ECM-detached cancer cells in the presence of antiestrogens (Fig. 1F). To account for any putative Matrigel-borne growth factors potentially affecting antiestrogen responses, we also confirmed these results in separate, collagen-based, 3D cocultures (Supplementary Fig. S1B). To further assess the translational relevance of this antiestrogen resistance model, we compared the antiestrogen responses of breast cancer 3D spheroids cocultured with primary BMSCs isolated from the bone metastatic lesion of a cancer patient versus from the bone marrow of a normal donor. Notably, the metastasis-derived, but not donor-derived, BMSCs induced antiestrogen resistance in breast cancer spheroids (Fig. 2B; Supplementary Fig. S1C), consistent with the notion that nonmalignant accessory cells within the 3D architecture of the bone metastatic environment can confer HT resistance to breast cancer cells. Furthermore, similar cocultures of LNCaP cells with BMSCs abrogated the inhibitory effect of antiandrogens on prostate cancer spheroid growth (Fig. 2C), supporting the generalizability of this microenvironmentally induced HT resistance phenotype in hormone-dependent cancers.

We examined whether HR mutations associated with HT resistance in in vitro models and in patients (13, 27) can supersede the microenvironmentally induced endocrine resistance phenotype in 3D cultures. We thus quantified the antiestrogen sensitivity of MCF7 3D spheroids harboring the ER mutation Y537S, in the presence versus absence of BMSCs. The Y537S ER mutation induced antiestrogen resistance in MCF7 3D spheroid monocultures, confirming previous results in 2D cultures (27). However, the antiestrogen response of Y537S MCF7 spheroids was even further attenuated in 3D cocultures with BMSCs (Fig. 2D; Supplementary Fig. S1D), suggesting that nonmalignant cells of the bone metastatic microenvironment can enhance HT resistance in breast cancer, even in the presence of ER mutations.

Coculturing breast cancer cells with nonmalignant “accessory” cells from other frequently colonized organs (e.g., brain and liver) also tended to attenuate the inhibitory effect of antiestrogens on breast cancer spheroid growth (Supplementary Fig. S1E and S1F), suggesting that microenvironmentally induced HT resistance mechanisms may operate in other metastatic sites beyond the bone, the focus of the current study.

To validate our findings in vivo, we implanted immunocompromised mice with MCF7 cells alone (monotypic xenografts) or together with BMSCs (heterotypic xenografts). Engraftment was higher for heterotypic xenografts (Supplementary Fig. S2A). Tamoxifen treatment delayed tumor growth in monotypic xenografts, but not their heterotypic counterparts, without affecting ER expression (Supplementary Fig. S2A–S2C), consistent with our in vitro findings about BMSC-induced antiestrogen resistance in breast cancer spheroids.

Coculture with BMSCs induced in breast cancer spheroids (compared with their monoculture counterparts) transcriptional changes that were highly concordant across cell lines (Fig. 3A). The transcriptomes of breast cancer spheroids in cocultures reflected decreased estrogenic signaling (Fig. 3B and C; Supplementary Fig S3A–S3C). Because attenuation of HR expression can be associated with hormone independence and HT resistance, we examined the ER gene expression in breast cancer 3D cultures. BMSCs induced downregulation (but not complete abrogation) of ER transcript levels in breast cancer spheroids (Fig. 3D). Next, we examined the ER protein expression in breast cancer 3D monocultures and cocultures (Fig. 3E and F; Supplementary Fig. S3D). IHC confirmed the antiestrogen-induced acinar differentiation in MCF7, T47D, and ZR75-1 3D spheroid monocultures. The ECM-attached breast cancer monolayer cells in all monoculture models retained ER protein expression during the antiestrogen-induced acinar differentiation. In addition, MCF7 and T47D spheroids in coculture with BMSCs were similar in size and number to the respective monocultures and retained ER protein expression (Fig. 3E; Supplementary Fig. S3D). Congruously, ER protein was expressed in vivo in heterotypic xenografts (Supplementary Fig. S2B) and has also been shown to remain expressed in the bone metastases of at least a proportion of patients with breast cancer (example shown in Supplementary Fig. S2C; ref. 34). In contrast to MCF7 and T47D cocultures, ER protein was lost in ZR75-1 spheroids cocultured with BMSCs (Fig. 3F). Furthermore, the overall number of ZR75-1 spheroids in cocultures was smaller and their overall growth was delayed (compared with the respective monoculture). Because ER transcriptional changes were similar across the three breast cancer coculture models (Fig. 3D), we examined whether protein degradation via the ubiquitin–proteasome system contributes to the distinct ER protein levels in 3D cocultures of ZR75-1 versus cocultures of MCF7 or T47D cells. Exposure of breast cancer 3D monocultures and cocultures to the proteasome inhibitor MG132 confirmed that BMSC-conditioned medium induced proteasome-depended degradation of ER in ZR75-1 (but not in MCF7) 3D cultures (Supplementary Fig. S3E).

To assess how the bone microenvironment may affect the HR expression in prostate cancer, we also examined LNCaP spheroids in 3D monocultures versus cocultures with BMSCs. IHC indicated that coculture with BMSCs induced loss of AR protein expression in LNCaP cells (Supplementary Fig. S3F) and tended to decrease the number and size of the formed spheroids, a pattern similar to ZR75-1 cocultures. Interestingly, the AR-negative LNCaP spheroids formed in coculture with BMSCs also expressed higher levels of neuroendocrine markers (Supplementary Fig. S3G).

Prior secretome profiling of various stromal cell types, including immortalized HS5 BMSCs (reanalyzed in Supplementary Fig. S4A; ref. 35) identified a multitude of cytokines, including high levels of IL6, a ligand known to activate several intracellular signaling cascades including JAK/STAT, PI3K, and MAPK/ERK. We observed strong enrichment for IL6/JAK/STAT3 pathway target genes in the examined breast cancer spheroids cocultured with BMSCs (Fig. 4A). We hypothesized that BMSC-secreted IL6 induces activation of alternative pathways and estrogen-independent growth in breast cancer spheroids. Indeed, exposing ZR75-1 3D monocultures to recombinant IL6 induced proteasome-depended degradation of ER (Supplementary Fig. S3E) and slower ZR75-1 spheroid growth (similar to cocultures with BMSCs) and attenuation of antiestrogen response (Supplementary Fig. S4B), phenocopying the respective coculture of ZR75-1 with BMSCs. Concordantly, antibodies against IL6 itself or IL6 receptor, or the JAK/STAT pathway inhibitor ruxolitinib, restored ER expression (Fig. 4B) and attenuated the BMSC-induced antiestrogen resistance (Fig. 4C) in ZR75-1 cocultures with BMSCs. Similarly, blocking IL6/JAK/STAT signaling also restored antiandrogen sensitivity in the LNCaP 3D cocultures with BMSCs (Supplementary Fig. S4C), suggesting that similar IL6-dependent microenvironmental signaling mechanism may operate in the bone metastatic microenvironment to induce androgen independence in prostate cancer.

In contrast to the results with ZR75-1, recombinant IL6 had only limited effect in MCF7 and T47D 3D spheroid growth (Supplementary Fig. S4B). Inhibition of IL6/JAK/STAT signaling pathway did not have substantial impact on antiestrogen sensitivity in MCF7 and T47D spheroid cocultures with BMSCs (Fig. 4D; Supplementary Fig. S4D). Therefore, BMSC-induced HT resistance in MCF7/T47D spheroids may be driven by distinct, IL6-independent, paracrine signals.

To assess the translational relevance of these putative IL6-independent microenvironmental models, we examined transcriptional signatures derived from MCF7 spheroids cultured in the presence of BMSCs (see Materials and Methods). In the METABRIC study, patients with ER⁺ breast cancer with shorter disease-specific survival had in their pretreatment tumors low levels of transcripts downregulated in MCF7 spheroids in coculture (Fig. 5A, left) or, conversely, high levels of transcripts upregulated in MCF7 spheroids in coculture (Fig. 5A, right). This association with clinical outcome was also confirmed in additional publicly available molecular profiling data, aggregated from several clinical trials, of ER⁺ breast cancer tumors (Fig. 5B; ref. 32), but not patients with ER-negative breast cancer (Supplementary Fig. S5). Concordantly, genes upregulated in MCF7 spheroids cocultured with BMSCs were significantly enriched for genes upregulated in tamoxifen-resistant breast cancer patient–derived xenografts (Fig. 5C; ref. 36).

To identify functional mediators of antiestrogen-resistance in 3D cocultures with BMSCs, we conducted a genome-scale CRISPR-Cas9 screen of MCF7 cells in 2D monocultures in the presence versus absence of 4-OHT and compared the patterns of sgRNA changes for genes differentially expressed in MCF7 spheroids cocultured with BMSCs (Fig. 5D). Genes upregulated in cocultures with BMSCs tended to have their respective sgRNAs depleted in the CRISPR/Cas9 screen in the presence of 4-OHT (Wilcoxon signed rank rest for depletion rank P < 0.0001), including genes previously associated with tamoxifen resistance in MCF7 cells (e.g., SOX9; ref. 37). These functional data support the notion that BMSCs induce in breast cancer spheroids genes, which, individually or in concert, contribute to decreased antiestrogen sensitivity. Gene set enrichment analysis for the collection of genes that were upregulated in the 3D cocultures and were also associated with 4-OHT resistance in the genome-scale CRISPR/Cas9 screen, indicated an overall transcriptional signature consistent with stimulation of MAPK and PI3K/Akt pathways or upstream growth factor receptors in MCF7 cells (Fig. 5E and F).

Given the complex secretome of BMSCs (Supplementary Fig. S4A) and the redundancy of BMSC-induced upregulation of genes and signaling pathways associated with reduced sensitivity to antiestrogens in cancer cells (Fig. 5D–F), we postulated that the antiestrogen resistance in 3D cocultures could be mediated by pleiotropic effects of BMSCs on the MCF7 spheroids. The results of our genome-scale CRISPR/Cas9 screen of MCF7 cells in the presence of antiestrogens suggest that single gene-by-gene approaches may not be powerful enough to reveal the therapeutic vulnerabilities in the setting of a redundant network. To identify therapeutically targetable pathways driving MCF7 spheroid anchorage-independent growth in coculture with BMSCs, we turned to the alternative approach of functional phenotypic screening and examined the sensitivity of breast cancer spheroids in 3D monocultures and cocultures to a set of tool chemical compounds including 416 kinase inhibitors targeting collectively approximately 300 kinases (Fig. 5G; Supplementary Fig. S6A). We identified increased sensitivity of MCF7 + BMSCs 3D cocultures (vs. the respective monocultures) to several inhibitors of ERK and PI3K/MTOR signaling pathway components (Fig. 5G; Supplementary Fig. S6B), but not to other drug classes (Supplementary Fig. S6C). Overall, these observations are concordant with the molecular data in the 3D cocultures of MCF7 with BMSCs, indicating that acquisition of hormone independence by anchorage-independent breast cancer cells in cocultures is mediated, at least in part, through activation of ERK/PI3K pathways, a finding concordant with previous studies in cell-autonomous models (1).

To further explore the translational relevance of our in vitro 3D models of microenvironmentally-induced HT resistance, we studied tumor organoids established from bone metastases of three patients with antiandrogen-refractory prostate cancer, namely MSK-PCa1, MSK-PCa2, and MSK-PCa3, characterized in prior studies (38). We observed in 3D monocultures that MSK-PCa1 organoids were AR-negative and refractory to antiandrogens. MSK-PCa3 organoids were AR-positive, but insensitive to antiandrogen treatment, suggesting a cell-autonomous antiandrogen resistance model attributable to putative activation of alternative signaling pathways. Interestingly, MSK-PCa2 organoids expressed AR and were sensitive to antiandrogen treatment, indicating that the clinical refractoriness may be due to microenvironmental mechanisms. To test this hypothesis, we compared the characteristics of MSK-PCa2 cells in 3D monoculture versus coculture with BMSCs. In cocultures, the MSK-PCa2 organoids acquired distinct morphology marked by diffuse cells invading the ECM instead of spheroids. Congruously, BMSCs attenuated the sensitivity of MSK-PCa2 cells to enzalutamide (Fig. 6). Similarly to studies with ZR75-1 and LNCaP spheroids described above, exposing cocultures to ruxolitinib or neutralizing antibodies against IL6 and IL6R restored antiandrogen sensitivity in MSK-PCa2 organoids (Fig. 6). Overall, these observations support the translational relevance of the 3D coculture models as a system to simulate preclinically the context of HT resistance of breast cancer and prostate cancer bone metastatic lesions.

The preclinical efforts to dissect breast cancer/prostate cancer HT resistance have so far focused mostly on cell-autonomous mechanisms modeled in conventional 2D cultures (1, 3). Clinical studies have also reflected this preclinical emphasis, by focusing mostly on characterizing the molecular features of tumor cells (typically from the primary tumor), which eventually are associated with development of more rapid endocrine resistance and metastasis. Moreover, studies to identify genomic drivers of metastatic breast cancer (39) or prostate cancer (40) do not address whether cell nonautonomous signals can also operate as parallel or alternate resistance mechanisms. To study the potential contribution of cell nonautonomous mechanisms to endocrine resistance, we examined the responses of breast and prostate cancer cells to HT in 3D culture conditions (where clonal expansion of each initial cell forms a distinct spheroid) in the presence versus absence of nonmalignant accessory cells of the microenvironment of key metastatic sites. We reasoned that it is biologically plausible for BMSCs to play such a role in the bone milieu, a main site for metastases of HR⁺ breast cancer and prostate cancer, because extensive studies in hematologic neoplasias have documented that BMSCs can confer to them cell nonautonomous resistance against diverse pharmacologic and immune-based therapies (29, 41, 42). Although our study focuses on modeling the interactions between tumor cells and BMSCs in the 3D microenvironment, other cell populations in the bone marrow (e.g., osteoblasts, immune cells, etc.) may also function as “accessories” to the tumor cells in distinct contexts. Thus, microenvironment-mediated endocrine resistance may be determined by the aggregate effects of these different nonmalignant cell populations. Adapting our 3D coculture systems to include these additional tumor-adjacent cell types will improve the biological relevance of these microenvironmental models for breast cancer and prostate cancer metastases.

Here, we observed that cell nonautonomous mechanisms emerging within the local metastatic microenvironment can indeed profoundly affect the response of HR⁺ tumor cells to HR-targeting agents. We identified two cellular phenotypes that occur within the 3D tumor architecture and are distinct with regard to tumor cell sensitivity to HT. Hormone antagonists had limited effect on the viability of breast cancer/prostate cancer cells residing in the ECM-attached niche. In contrast, antiestrogens/antiandrogens abrogated the ability of breast cancer/prostate cancer cells to grow in anchorage-independent conditions and induced acinar-like differentiation in 3D cultures. Notably, cocultures with BMSCs rescued the anchorage independence of breast cancer/prostate cancer cells despite their exposure to HT.

Normal epithelial cells require attachment to the basement membrane for survival (24, 28, 33). The activation of oncogenic signals enables cancer cells to evade anoikis and survive outside their normal ECM niches, a key step toward transformation of glandular epithelial cells (25). For instance, HER2 overexpression in normal mammary acini rescues the metabolic defects caused by matrix detachment through stabilization of EGFR/PI3K activation, resulting in lumen filling (33). Our results indicate that in hormone-dependent cancers, HR is directly involved in the evasion of anoikis, while HR blockade compromises redox homeostasis of ECM-detached HR⁺ cells and their ability for anchorage-independent growth (e.g., Fig. 1A–E). The ability of ECM-attached HR⁺ cancer cells to survive and grow in the presence of HR-targeting agents indicates that cell adhesion signals are also important in HT resistance, suggesting that alternative signaling axes (e.g., integrin/FAK/SRC kinase) could be targeted to overcome HT resistance. Maintenance of lumen clearance and acinar morphology in breast cancer spheroids was dependent on continuation of antiestrogen treatment. After antiestrogen washout, the peripheral monolayer cells repopulated the lumen (Fig. 1A), indicating that the ECM niche can preserve a reservoir of latent HR⁺ cancer cells that can reinitiate tumor growth if HT is interrupted. These results are congruent with the clinical observation that prolonging adjuvant HT treatment beyond 5 years has survival benefit for patients with hormone-dependent breast cancer (43).

The suppression of anchorage-independent growth of breast cancer/prostate cancer cells by HT was dramatically attenuated in 3D cocultures with BMSCs. Interestingly, although the global transcriptional signatures induced in breast cancer spheroids by BMSCs (e.g., suppression of estrogenic signals and activation of IL6/JAK/STAT3 pathway) were similar for the different breast cancer cell lines examined (e.g., Fig. 3A–C), the mechanisms of antiestrogen resistance were distinct, namely IL6-mediated versus IL6-independent. High serum IL6 concentrations are associated with poor prognosis in patients with metastatic ER⁺ breast cancer (44, 45). In an in vivo breast cancer model of acquired adaptive HT resistance, ER blockade led to increased IL6 cytokine levels and ER-independent, IL6/Notch3-driven, cancer stem cell growth via a feed-forward loop (46). Distinctly from the findings of this latter study, IL6-mediated antiestrogen resistance was not associated with concomitant IL6-stimulated cell growth in our microenvironmental HT resistance model of ZR75-1 3D spheroids (Supplementary Fig. S4B), indicating that other putative changes occurring during prolonged antiestrogen exposure in cell-autonomous models of acquired HT resistance may exert additional effects on the function of IL6/Notch signaling, which are not pertinent to our 3D coculture models of de novo resistance. Concordantly, in both these IL6-dependent models of acquired (46) and de novo (in our study) HT resistance, blocking IL6 signaling using IL6-targeting antibodies (and also confirmed with JAK inhibitors, in our study) restored HR expression and HT sensitivity.

ERK and PI3K/Akt signaling cascades are crucial for survival of transformed cancer cells during displacement from the normal ECM niche and anchorage-independent growth (33). The PI3K/Akt and ERK pathways can be indirectly activated by HR (1, 3), which could explain the dependence of breast cancer/prostate cancer cells on HR signaling for anoikis evasion in our 3D monocultures. These growth factor signaling cascades seem to be further activated through putative paracrine factors in cocultures with BMSCs (e.g., Fig. 5E). We hypothesize that BMSC-induced paracrine activation of ERK/PI3K/Akt axes in ECM-detached MCF7 cells bypasses their dependence on ER signaling for anchorage-independent growth, therefore providing an alternative mechanism of HT resistance in the 3D metastatic microenvironment. Interestingly, although MCF7 spheroids have functional IL6 receptor and BMSC-secreted IL6 can potentially activate the ERK and PI3K/Akt pathways in cancer cells, HT resistance in this coculture setting was not abrogated by IL6 inhibition, indicating either dependency on alternate BMSC-secreted cytokine/s or redundancy of paracrine extracellular signals in this context. Conversely, although IL6/JAK/STAT3 pathway was upregulated in MCF7 spheroid cocultures with BMSCs, antiestrogen resistance was not associated with ER expression loss (distinct from ZR75-1 cocultures) and was not reversed by JAK inhibition, indicating that alternate IL6-independent mechanisms confer endocrine resistance to MCF7 spheroids.

Although extensive preclinical studies have used a multitude of informative, but mostly cell-autonomous, models to identify actionable mechanisms of HT resistance in breast cancer and prostate cancer, a comprehensive panel of clinical biomarkers is needed for the efficient deployment of therapeutic agents targeting these mechanisms in patients. Our study shows that the use of metastatic biopsies to identify and/or validate biomarkers related to tumor–stroma interactions in the metastatic lesion can help to distinguish between the various forms of microenvironmentally induced HT resistance (which may operate in the absence of, or concurrently with, relevant genetic lesions) and to refine the treatment strategy. For instance, HT-refractory metastatic tumors that retain expression of HR and have increased ERK/PI3K/Akt activity through paracrine microenvironmental signal could be sensitive to kinase inhibitors targeting EGFR/HER2 and/or PI3K/Akt/MTOR. Conversely, loss of HR expression along with high local IL6 levels could indicate IL6-induced hormone independence; in this case, the use of agents targeting IL6 or JAK should perhaps be administered concomitantly with HT, because targeting IL6/JAK/STAT alone may not be effective and could also fuel HR-dependent tumor growth.

The metastatic microenvironment is complex in structure and heterogeneous in cellular and paracrine elements. The recapitulation of these microenvironmental features in vitro, using 3D coculture systems, projects a complex landscape of various, possibly coexisting, forms of HT resistance in breast cancer and prostate cancer. Further development of these preclinical models of metastasis, ideally using patient-derived tumor and stromal cells, could help the identification of therapeutic targets to overcome HT resistance and of biomarkers to stratify patients likely to benefit from these therapies.

E. Dhimolea reports grants from Breast Cancer Alliance (Young Investigator Award), Claudia Adams Barr Program for Innovative Cancer Research Award, Hellenic Women's Club, Terri Brodeur Breast Cancer Foundation, Avon Foundation Breast Cancer Research Program, and Elsa U. Pardee Foundation during the conduct of the study. N. Mitsiades reports grants from NCI Grant (U54-CA233223) during the conduct of the study. Y. Chen reports personal fees from Oric Pharmaceuticals outside the submitted work. R. Jeselsohn reports grants from Lilly and Pfizer and personal fees from Carrick and Luminex outside the submitted work. M. Brown reports grants from Novartis and personal fees from Kronos Bio, H3 Biomedicine, and GV20 Therapeutics outside the submitted work. I. Georgakoudi reports a patent for System and Method for Assessing Cellular Metabolic Activity pending. C.S. Mitsiades reports grants from Department of Defense (grant W81XWH-15-1-0012), Breast Cancer Alliance Young Investigator Award, Claudia Adams Barr Program for Innovative Cancer Research, Hellenic Women's Club, Terri Brodeur Breast Cancer Foundation, Avon Foundation Breast Cancer Research Program, Leukemia and Lymphoma Society Scholar Award, Ludwig Center at Harvard, and Elsa U. Pardee Foundation during the conduct of the study. C.S. Mitsiades also reports relationships with Takeda (employment of a family member), Fate Therapeutics (consultant/honoraria), Ionis Pharmaceuticals (consultant/honoraria), and FIMECS (consultant/honoraria), grants from Sanofi (research funding), Arch Oncology (research funding), EMD Serono/Merck KGA (research funding), Abbvie (research funding), Karyopharm (research funding), Janssen/Johnson & Johnson (research funding), TEVA (research funding), and Novartis (research funding) outside the submitted work; and has been an inventor/coinventor over the last 15 years in several patent applications pertinent to the broader theme of resistance of different types of cancers to various therapeutics, including applications that have involved studies of tumor cells cocultured with nonmalignant cells of their microenvironment in conventional 2D culture or 3D systems. No disclosures were reported by the other authors.

E. Dhimolea: Conceptualization, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. R. de Matos Simoes: Resources, software, formal analysis, visualization. D. Kansara: Methodology, writing-review and editing. X. Weng: Methodology. S. Sharma: Methodology. P. Awate: Methodology. Z. Liu: Formal analysis, visualization, methodology, writing-review and editing. D. Gao: Resources. N. Mitsiades: Writing-review and editing. J.H. Schwab: Resources. Y. Chen: Resources. R. Jeselsohn: Resources, writing-review and editing. A.C. Culhane: Resources, software, formal analysis, methodology. M. Brown: Resources. I. Georgakoudi: Resources, methodology, analysis of two-photon microscopy data. C.S. Mitsiades: Conceptualization, resources, data curation, supervision, funding acquisition, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing.

We thank Zach Herbert and the members of DFCI Molecular Biology Core Facilities for the transcriptional analyses. This study was supported by the Breast Cancer Alliance Young Investigator Award (to E. Dhimolea and C.S. Mitsiades), Claudia Adams Barr Program for Innovative Cancer Research (to E. Dhimolea and C.S. Mitsiades), Hellenic Women's Club (to E. Dhimolea and C.S. Mitsiades), Terri Brodeur Breast Cancer Foundation Grant (to E. Dhimolea), Avon Foundation Breast Cancer Research Program (to E. Dhimolea and C.S. Mitsiades), Elsa U. Pardee Foundation Grant (to E. Dhimolea and C.S. Mitsiades), Department of Defense grants W81XWH-15-1-0013 (to A.C. Culhane) and W81XWH-15-1-0012 (to C.S. Mitsiades), National Cancer Institute Grant U54-CA233223 (to N. Mitsiades), Leukemia and Lymphoma Society Scholar Award (to C.S. Mitsiades), Ludwig Center at Harvard (to C.S. Mitsiades), and National Cancer Institute Grant U54-CA233223 (to N. Mitsiades). Raw CRISPR and RNA-seq sequencing data was processed on the Orchestra High Performance Computer Cluster at Harvard Medical School (grant NCRR 1S10RR028832-01, http://rc.hms.harvard.edu).

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.