Biomimetic bone niche reconstructs proliferation-inhibited and therapy-resistant bone-metastatic prostate cancer

Abstract

Prostate cancer bone metastases often harbor a rare subset of tumor cells with suppressed proliferation, contributing to therapy resistance and disease relapse. However, the lack of physiologically relevant in vitro models has hindered mechanistic and therapeutic advances. Here, we engineered a 3D bone-like microenvironment by integrating calcium phosphate scaffolds, decellularized extracellular matrix (dECM), mesenchymal stem cells (MSCs), and osteoblasts. This biomimetic niche induced a proliferation-inhibited state in prostate cancer cells, closely mirroring transcriptomic signatures identified from patient-derived single-cell RNA sequencing datasets. Tumor cells in this niche also displayed enzalutamide resistance, accompanied by metabolic reprogramming and activation of pro-survival signaling. This platform provides a clinically relevant tool for modeling bone metastatic prostate cancer and accelerating the development of therapies targeting resistant tumor states.

Graphical abstract

Highlights

• •A novel 3D bioprinted biomimetic bone microenvironment (BME) was engineered, recapitulating the proliferation-inhibited tumor phenotype in clinical prostate cancer bone metastases.
• •This platform integrates a CPC scaffold mimicking trabecular bone and dECM hydrogel, showing robust biocompatibility and osteoinductive properties.
• •The biomimetic BME suppressed prostate cancer cell proliferation in vitro and in vivo, inducing G0/G1 arrest and reducing tumor growth.
• •Transcriptomic profiling of tumor cells in the BME revealed downregulated cell cycle/proliferative pathways and altered metabolism, matching patient scRNA-seq data from bone metastatic prostate cancer.
• •The 3D-printed BME model conferred enzalutamide resistance to prostate cancer cells, highlighting microenvironment-driven metabolic reprogramming and pro-survival pathways for therapeutic evasion.

1. Introduction

Bone metastases are a principal cause of prostate cancer-related mortality, driven by intricate molecular interactions and formidable challenges in treatment [1]. The fate of disseminated tumor cells (DTCs) within the bone microenvironment (BME) is highly site-dependent, governed by both spatial specificity and complex signaling networks [2,3]. In regions of active bone remodeling—accounting for less than 20 % of the endosteal surface—growth-promoting factors drive rapid tumor proliferation. In contrast, over 80 % of the endosteal niche consists of quiescent zones, where DTCs enter a state of proliferation arrest, a process stringently regulated by homeostatic bone signaling pathways [[4], [5], [6]].

The BME itself is a remarkably heterogeneous, vascularized, and innervated connective tissue that provides a specialized niche for DTCs. This dynamic environment integrates a range of cellular populations, including osteoblasts, osteoclasts, endothelial cells, immune cells, and hematopoietic progenitors, all of which coordinate the fate decisions of tumor cells. Intrinsic physicochemical factors, such as hypoxia and acidic pH, further influence neoplastic behavior. Among these regulators, NG2⁺/Nestin⁺ MSCs enforce tumor dormancy through TGF-β2 paracrine signaling, while osteoblast-derived factors, such as Wnt5a, BMP7, and GAS6, provide additional suppressive signals that curb tumor proliferation [[7], [8], [9], [10]]. However, disruptions in the BME can trigger the reactivation of dormant tumor cells, leading to drug resistance and metastatic recurrence. Despite this understanding, the molecular mechanisms governing the transition from proliferation arrest to reactivation remain poorly defined [11].

Clinically, only about 0.02 % of DTCs evolve into detectable metastases, while the majority fail to adapt to the new microenvironment, are eliminated by immune surveillance, or enter a state of proliferative arrest. Preclinical studies highlight the inefficiency of metastatic progression: within two weeks of intravenous inoculation, only 30 % of tumor cells persist in organs such as the lung, liver, and bone, with only 0.001 %–0.02 % forming macroscopically visible metastases [12]. Despite the widespread elimination of non-metastatic cells, a significant subset enters a non-proliferative state, evading conventional diagnostic methods. Recent single-cell transcriptomic profiling has identified a distinct population of proliferation-inhibited tumor cells within prostate cancer bone metastases, characterized by downregulated cell cycle pathways, though lacking full dormancy hallmarks [11,13,14]. Yet, replicating this phenotype in experimental models remains challenging, due to an incomplete understanding of the underlying regulatory mechanisms [15]. This knowledge gap hinders both the biological investigation of tumor cell behavior and the development of therapeutic strategies aimed at preventing metastatic relapse.

Building upon our prior discovery that osteoblasts play a critical role in suppressing prostate cancer cell proliferation, we hypothesize that the broader BME actively maintains this proliferation-inhibited state. However, existing experimental models remain limited in their ability to recapitulate the structural complexity [[16], [17], [18]], and biochemical diversity of the bone niche [19,20]. To overcome this limitation, we have developed a sophisticated in vitro platform that biomimetically reconstructs key features of the BME, enabling in-depth studies of tumor cell fate regulation and providing a valuable tool for therapeutic testing.

Leveraging our expertise in creating 3D heterogeneous tumor models—including spatially patterned breast cancer and hepatocellular carcinoma systems that capture microenvironmental gradients—we now introduce a precision-engineered bone niche analog, fabricated using advanced 3D bioprinting techniques [21,22]. This model features spatial control over architectural motifs and cellular distribution, incorporating MSCs, osteoblasts, and prostate cancer cells from bone metastases into a calcium-enriched, soft matrix designed to replicate the mechanical properties and trabecular topology of bone marrow. Strikingly, our model is the first to faithfully reproduce the proliferation-inhibited tumor cell phenotype identified through single-cell transcriptomics in prostate cancer bone metastases. This system provides a controlled platform for investigating the mechanisms that regulate tumor cell quiescence, resistance to therapy, and metastatic reactivation. Through this approach, we offer new insights and opportunities for therapeutic development targeting metastatic relapse.

2. Results

2.1. Single-cell data reveal a subset of proliferation-arrested cells in bone metastatic prostate cancer

We analyzed previously published single-cell RNA sequencing (scRNA-seq) data (GSE143791) from 9 patients with prostate cancer spinal metastases [23], including solid tumor lesions and bone marrow samples from involved and distal vertebrae, along with primary tumor samples from 11 patients undergoing prostatectomy (GSE176031) (Fig. 1a–e) [24]. Unbiased clustering identified diverse cell types within the bone metastatic niche, including tumor cells, immune cells, endothelial cells, osteoblasts, osteoclasts, and MSCs (Fig. 1b). Based on prostate cancer markers (KLK2/3/4, AR), we annotated 885 tumor cells and stratified them by anatomical origin (Supplementary Fig. 1a). Tumor cells from distal sites showed distinct transcriptional profiles compared to those from tumor and involved sites (Fig. 1c). The signaling pathways in distal tumor cells, as revealed by pathway enrichment analysis and illustrated in Fig. 1d, which show suppression of cell cycle and replication pathways such as the G2/M checkpoint and mitotic spindle [25], and enrichment of DNA repair signaling, indicate a proliferation-inhibited state [26,27].

Fig. 1.

Single-cell data revealed a subset of proliferation-arrested cells in bone metastatic prostate cancer. a) Classification of tumor cells into three subpopulations based on anatomical origin: distal site, involved site, metastatic tumor site. b) UMAP projection of single-cell transcriptomes from spinal metastatic tissues of 9 prostate cancer patients, annotated by cell type. c) Heatmap displaying DEGs across the three tumor cell subpopulations. d) Heatmap of differentially enriched Hallmark pathways (GSEA) among the tumor subpopulations. e) Schematic overview of the analysis pipeline for primary tumor cells from 11 patients who underwent radical prostatectomy. f) UMAP projection of single-cell transcriptomes from radical prostatectomy specimens, annotated by identified cell types. g) UMAP plot of epithelial cells from radical prostatectomy tissues showing 17 clusters (0–16); clusters 1 and 11 identified as malignant via inferCNV analysis. h) Schematic summarizing the integrated analysis of primary and metastatic tumor cell datasets. i) Volcano plot of DEGs between primary and distal tumor cells. j) Comparison of Hallmark pathway enrichment related to cell cycle and proliferation between primary and distal tumor cells (GSEA). k) Expression levels of representative genes associated with mitosis and prostate cancer proliferation in primary versus distal tumor cells. l) UMAP projection of integrated single-cell data from distal tumor cells, MSCs, and osteoblasts. m) Volcano plot of signature genes distinguishing distal tumor cells, MSCs, and osteoblasts. n) Ligand-receptor interaction network among distal tumor cells, MSCs, and osteoblasts, identified using the iTALK algorithm. o) Top 20 ligand-receptor interactions between MSCs/osteoblasts and tumor cells predicted by iTALK.

To further characterize this state, we integrated scRNA-seq data from primary tumors and identified 9 epithelial subclusters, including 2502 tumor cells confirmed via inferCNV (Fig. 1f and g). Comparative analysis between distal bone metastases and primary tumor cells revealed downregulation of proliferative pathways and mitosis-associated genes in distal cells, supporting a bone-specific proliferation arrest (Fig. 1h, i, j, k).

Given previous reports implicating osteoblasts and MSCs in modulating tumor cell behavior [[7], [8], [9], [10]], we examined their potential interactions with distal tumor cells (Fig. 1l and m). The collagen-integrin and LAMA2/LAMB2-RPSA signaling pathways, which were identified through ligand-receptor analysis using iTALK (Fig. 1n and o), may contribute to tumor adhesion, EMT, and drug resistance through ECM remodeling and MAPK/ERK activation [[28], [29], [30], [31]]. It is noteworthy that we found that distal cells do not exhibit the characteristics of dormancy in prostate cancer bone metastasis [13], but rather are in a state of inhibited proliferation (Supplementary Fig. 1b).

While these findings offer mechanistic clues, the proliferation-inhibited tumor population remains rare, challenging to isolate, and poorly understood, highlighting the need for experimental models to enable functional validation and deeper mechanistic investigation.

2.2. Fabrication of a biomimetic 3D bone microenvironment

To replicate the complex architecture and biomechanical properties of native bone, we established a composite 3D BME using extrusion-based bioprinting. This system integrates a calcium phosphate cement (CPC) scaffold mimicking trabecular bone with a dECM hydrogel recapitulating the marrow niche (Fig. 2a).

Fig. 2.

Fabrication of a Biomimetic 3D Bone Microenvironment. a) Schematic illustration of the biomimetic BME model. b) Representative photographs of dECM hydrogels before and after gelation. c) Proteomic profiling of lyophilized dECM hydrogels. d) Heatmap showing effective Young's modulus of dECM hydrogels (n = 3). e) Thermal gelation curve of dECM during a temperature ramp from 4 °C to 40 °C. f) Viscosity and shear stress profiles of dECM hydrogels across shear rates ranging from 10⁻¹ to 10² s⁻¹. g) Degradation profile of dECM hydrogels over 20 days. h) Swelling behavior of dECM hydrogels. i) SEM images of lyophilized dECM hydrogels at magnifications of 50 × , 100 × , 200 × , and 500 × (scale bars: 500 μm, 250 μm, 125 μm, and 50 μm). j) Schematic of the custom-designed 3D bioprinter. k) Schematic representation of the CPC scaffold. l) Image of a 3D-printed CPC scaffold (15 mm × 15 mm). m) Image of a 3D-printed trabecular bone-like CPC structure. n) Chemical structure of CPC. o) Temperature-dependent rheological profile of CPC showing storage and loss modulus changes from 4 °C to 40 °C. p) Viscosity and shear stress profiles of CPC across shear rates from 10⁻¹ to 10² s⁻¹. q) Compressive stress-strain curves of CPC material. r) SEM images of lyophilized CPC scaffolds at magnifications of 27 × , 100 × , 1000 × , and 5000 × (scale bars: 400 μm, 200 μm, 20 μm, and 2 μm).

The dECM hydrogel, derived through an optimized decellularization protocol, exhibited robust thermosensitive gelation and was primarily composed of type I collagen, as confirmed by mass spectrometry [32,33] (Fig. 2b and c). Proteomic analysis also revealed the retention of key ECM constituents, including fibrin, hyaluronic acid, and proteoglycans, supporting its bioactivity. Mechanical characterization showed a low effective Young's modulus (∼100 Pa Fig. 2d), suitable for modeling soft marrow tissue [34]. Previously, we measured the Young's modulus of the bone marrow using a nanoindenter and found it to range between 500 and 1200 Pa (Supplementary Fig. 5). Rheological assays demonstrated rapid sol-gel transition at 32 °C (Fig. 2e), pronounced shear-thinning behavior at 10 °C (viscosity decreasing to ∼0.1 Pa s under shear; Fig. 2f), and a strong water absorption profile (Fig. 2h). Degradation kinetics indicated that approximately 50 % of the hydrogel mass remained after 20 days (Fig. 2g), aligning with the requirements for mid-term in vitro studies. SEM imaging revealed a porous 3D fibrous network with collagen-like structures that provide a supportive matrix for cell adhesion and migration (Fig. 2i).

For the bone-mimetic scaffold, a bioactive calcium phosphate ink was formulated and printed using a suspension-assisted strategy to achieve high structural fidelity and recapitulate the anisotropic architecture of cancellous bone [35] (Fig. 2j, k, l, m). The incorporation of hydroxyl-functionalized polymers facilitated hydrogen bonding with calcium phosphate, enhancing viscosity, print resolution, and post-printing mechanical strength through improved particle cohesion during hydration and curing [36,37] (Fig. 2n). For the CPC scaffold, mechanical testing revealed favorable properties tailored for bioprinting and bone mimicry. In temperature-dependent rheological profiling (Fig. 2o), the storage modulus (G′) consistently exceeded the loss modulus (G″) across a ramp from 10 °C to 40 °C (e.g., G' ≈ 10⁴–10⁵ Pa vs. G" ≈ 10³–10⁴ Pa at lower temperatures), confirming a dominant elastic, solid-like response (G'/G" > 1). A sharp transition near 30 °C drove a marked increase in both moduli (from ∼10⁴ Pa to >10⁶ Pa for G′), attributable to enhanced crosslinking and hydration during curing, which bolsters the scaffold's integrity for stable 3D constructs. Complementing this, shear-dependent viscosity measurements (Fig. 2p) demonstrated pronounced shear-thinning, with viscosity dropping from ∼10⁴ Pa s at low shear rates (0.1–1 s⁻¹) to ∼10 Pa s at higher rates (10–100 s⁻¹), embodying non-Newtonian flow (power-law index n < 1) that enables smooth extrusion during printing and rapid viscosity recovery for post-deposition shape fidelity. Finally, compressive stress-strain curves (Fig. 2q) yielded an effective Young's modulus of ∼20–30 MPa in the initial linear elastic regime (0–2 % strain), with robust stress-bearing capacity evidenced by a peak stress of ∼0.8 MPa at ∼5 % strain, followed by minor yielding (to ∼0.7 MPa) indicative of plastic deformation or microcracking, and subsequent strain hardening up to ∼1.0 MPa at 10 % strain. These properties of the CPC scaffold effectively emulate the toughness of trabecular bone; in contrast to the softer dECM hydrogel (∼100 Pa), they establish a hierarchically structured biomimetic niche.

The choice of porcine skin-derived decellularized extracellular matrix (dECM) in our bone microenvironment model was strategically designed to replicate native bone matrix biology through a composite approach. While tissue-specific dECM is ideal, our system leverages dECM's abundant type I collagen (Fig. 2c) as the organic scaffold for cell adhesion, integrated with a 3D-printed calcium phosphate cement (CPC) scaffold mimicking mineralized trabecular structure for mechanical and topographical cues. This enables precise delivery of biochemical and biophysical signals, while porcine sourcing offers scalability, ethical advantages, and regulatory ease, supported by high conservation of ECM proteins between porcine and human tissues.

2.3. Biocompatibility and osteoinductive properties of the engineered BME

To evaluate the biocompatibility of the biomimetic BME, we independently encapsulated osteoblasts (MC3T3), bone-derived MSCs, and prostate cancer cells (C4-2B) within the dECM hydrogel (Fig. 3a). Fluorescence imaging revealed high viability and well-spread cell morphology across all three cell types, confirming the dECM's cytocompatibility and structural support (Fig. 3b). To mimic the native trabecular architecture, MC3T3 cells were seeded onto the printed CPC scaffold, where they selectively adhered to the scaffold surface, replicating the in vivo spatial distribution of osteoblasts on trabecular bone (Fig. 3c). Confocal microscopy imaging indicated robust osteoblast proliferation within the scaffold (Fig. 3d). When MSCs were introduced into the co-culture system, both cell types exhibited favorable growth within the composite scaffold-hydrogel construct (Fig. 3d), validating the overall biocompatibility and stability of the model.

Fig. 3.

Biocompatibility and osteoinductive properties of the engineered BME. a) Schematic and representative fluorescence images showing individual culture of MC3T3 (mCherry), MSC (GFP), and C4-2B (BFP) cells in dECM hydrogels. Representative x-y plane projections at days 1, 3, 7, 14, and 21. b) Cell proliferation of MC3T3, MSC, and C4-2B in dECM hydrogels over 21 days (n = 3). c) Schematic and fluorescence images showing MC3T3 attachment to the CPC scaffold and subsequent co-culture with MSCs. d) Proliferation dynamics of MC3T3 and MSCs in the hydrogel-based BME model over 14 days (n = 3), quantified as fluorescence intensity normalized to Day 1.e) ALP staining at day 7. f) ARS at day 14. g) Expression of osteogenic genes at day 7. h) Images of the BME model 7 days after implantation and upon retrieval at day 28. i, j, k) Histological and immunohistochemistry analysis of retrieved tissues: H&E staining (i), Masson trichrome staining (j), and immunostaining of osteogenic markers (k). Statistical analysis was performed using one-way ANOVA (n = 3). Significance is indicated as: ns (not significant), *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001, respectively.

To assess osteoinductive potential, constructs were cultured in osteogenic differentiation medium containing β-glycerophosphate, ascorbic acid, and dexamethasone. Alkaline phosphatase (ALP) staining on day 7 showed widespread enzymatic activity (Fig. 3e), while Alizarin Red staining (ARS) on day 14 revealed extensive calcium nodule deposition (Fig. 3f). Quantitative RT-PCR analysis demonstrated significantly elevated expressions of key osteogenic markers, including Alpl, Ibsp, Bglap, Sp7, Runx2, and Spp1 compared to non-osteogenic controls (Fig. 3g, all p < 0.05).

To further evaluate in vivo osteogenic capacity, constructs were subcutaneously implanted into BALB/c-nu mice (Fig. 3h). After 4 weeks, histological analysis via H&E (Fig. 3i) and Masson's trichrome staining (Fig. 3j) revealed abundant extracellular matrix deposition and collagen maturation. Immunohistochemistry confirmed marked osterix (OSX) expression, supporting active osteogenic differentiation within the implant (Fig. 3k).

2.4. Recapitulating Bone Microenvironment-Mediated Tumor Suppression in Prostate Cancer

To investigate the influence of the bone BME on prostate cancer progression, we established a 3D in vitro co-culture system comprising prostate cancer cells (C4-2B), bone-derived MSCs, and osteoblasts. Tumor spheroids within the BME exhibited significantly attenuated volumetric expansion compared to monoculture controls (Fig. 4a), as further corroborated by quantitative fluorescence signal analysis (Fig. 4c). Fluorescence quantification also confirmed that MSCs and osteoblasts retained high proliferative activity within the co-culture system (Fig. 4b), indicating that the observed suppression of C4-2B growth was not due to generalized proliferation inhibition. To further exclude the possibility of nutrient competition as the cause of reduced tumor expansion, we replaced C4-2B cells with PC-3 cells, a more aggressive, bone-tropic prostate cancer line. Notably, PC-3 cells rapidly proliferated and became dominant in the co-culture within several days (Supplementary Fig. 2a), suggesting that the inhibitory effect observed in C4-2B cells is not due to resource limitation, but rather to specific regulatory interactions within the engineered microenvironment.

Fig. 4.

Recapitulating Bone Microenvironment-Mediated Tumor Suppression in Prostate Cancer. a) Schematic and fluorescence images showing C4-2B tumor spheroids cultured with either CPC scaffolds (control) or the BME model. b) Normalized proliferation of MC3T3, MSC, and C4-2B cells in the BME group (n = 3), calculated from fluorescence intensity and expressed as fold-change relative to Day 1. c) A quantitative comparison of the relative proliferation (normalized to Day 1) of C4-2B tumor spheroids between the control group and the BME group over time (Days 1, 7, and 14). d) Schematic of flow cytometry-based sorting and cell cycle analysis of C4-2B cells. e) Cell cycle distribution of C4-2B cells from both groups by flow cytometry. f, g, h) Heatmaps of DEGs in MSC (f), MC3T3 (g), and C4-2B (h) cells. i) Tumors retrieved from mice in control and BME groups on day 21. j, k, l) Quantification of body weight (j), tumor volume (k), and tumor weight (l) in mice from both groups (n = 5). m)In vivo bioluminescent imaging at days 0, 7, 14, and 21. n, o) Histological and molecular analysis of tumors: H&E staining (n) and Ki-67 immunohistochemistry (o). Statistical significance was assessed using one-way ANOVA (n = 3). Significance levels: ns (not significant), *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001, respectively.

Flow cytometry-based cell cycle analysis demonstrated that BME group significantly increased the proportion of cells in the G0/G1 phase (76.8 % versus 52.7 % in the control group) and concurrently led to a significant reduction in the percentage of tumor cells in the S phase (13 % in the BME group compared to 30.9 % in the control group) (Fig. 4d and e). These findings suggest that BME may induce cell cycle arrest at the G0/G1 phase, potentially by interfering with DNA replication processes, thereby inhibiting the progression of tumor cells into the S phase. Transcriptome sequencing of MSCs, osteoblasts, and prostate cancer cells identified significant differential gene expression profiles, suggesting that cell-cell and cell-matrix interactions in the BME influence DNA replication and cell cycle regulation, contributing to tumor proliferation suppression (Fig. 4f, g, h).

To validate the in vitro findings, a subcutaneous xenograft model was established in BALB/c-nu mice. The tumor volume and tumor weight in the BME group were significantly smaller than that in the control group (Fig. 4i–k, l), with no significant differences in body weight or weight changes observed between the two groups (Fig. 4j). Bioluminescence imaging further confirmed decreased tumor burden in the BME group (Fig. 4m). Histological evaluation of excised tumors demonstrated reduced mass (Fig. 4n) and significantly lower Ki-67 proliferation indices (Fig. 4o), consistent with suppressed tumor cell proliferation.

Collectively, this biomimetic model effectively replicates key features of the prostate cancer bone metastasis microenvironment, inhibiting C4-2B proliferation, offering a promising platform for mechanistic studies and drug screening.

2.5. Transcriptomic profiling of 3D-Printed model recapitulates prostate cancer proliferation inhibition phenotype observed in human bone metastasis

To investigate the molecular mechanisms underlying the effects of the 3D-printed biomimetic BME on prostate cancer cells, we performed transcriptomic sequencing of tumor cells cultured within the 3D-printed BME, comparing them with cells cultured in a 3D environment without bone-specific components. Principal component analysis (PCA) demonstrated distinct clustering of biological replicates between groups, confirming the consistency and reliability of the transcriptomic data (Fig. 5a).

Fig. 5.

Transcriptomic profiling of 3D-printed model recapitulates prostate cancer proliferation inhibition phenotype observed in human bone metastasis. a) Principal component analysis (PCA) of RNA-seq data from prostate cancer cells in BME versus control conditions. b) Volcano plot of DEGs between BME and control groups. c) GO enrichment of DEGs, highlighting altered biological processes in BME. d) GSEA showing downregulation of mitosis- and proliferation-related GO terms in BME. e) KEGG pathway enrichment analysis of DEGs between BME and control groups. f, g) Heatmaps of DEGs involved in steroid hormone biosynthesis (f) and focal adhesion (g) pathways from KEGG analysis. h) GSEA comparison of Hallmark pathways between BME and control tumor cells. i) GSEA of key regulatory pathways (P53, mitotic spindle, G2M checkpoint, DNA repair) between BME and control. j) Venn diagram showing shared enriched Hallmark pathways between BME vs. control and primary vs. distal tumor cells. k) Sankey diagram illustrating overlap in Hallmark pathway enrichment across comparisons. l, m) Heatmaps of significantly upregulated (l) and downregulated (m) genes from primary and distal tumor cells in BME and control groups. n) Single-sample GSEA (ssGSEA) analysis showing expression trends of top 100 DEGs (by p-value) from primary and distal tumor cells applied to BME and control samples. o) ssGSEA analysis showing expression trends of top 100 DEGs from BME and control tumor cells applied to primary and distal tumor datasets.

Differential gene expression analysis identified approximately 1000 significantly altered genes between the BME and control groups (p < 0.05) (Fig. 5b). Notably, genes associated with cell proliferation—including FN1, KIF14, NAMPT, PIK3C2A, and SLC7A11—were markedly downregulated in the BME group (p < 0.05). These genes regulate critical processes such as DNA replication, cell adhesion, and mitotic progression, suggesting that the BME imposes a suppressive effect on tumor proliferation (Supplementary Fig. 3c)

Gene Ontology (GO) enrichment analysis of the differentially expressed genes (DEGs) revealed significant pathway alterations across biological processes, molecular functions, and cellular components, with a predominant downregulation pattern (FDR <0.05) (Fig. 5c). Gene Set Enrichment Analysis (GSEA) further corroborated these findings, highlighting the downregulation of mitosis-related and proliferation-related pathways, including chromosome segregation (GO:0007059), microtubule binding (GO:0008017), spindle pole organization (GO:0090148), DNA helicase activity (GO:0003678), and centrosome cycle regulation (GO:0007098) (Fig. 5d). These results indicate that the BME exerts regulatory pressure on cell cycle progression and mitotic machinery, thereby inhibiting tumor proliferation.

Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis revealed significant alterations in multiple signaling pathways within the BME group (p < 0.05) (Fig. 5e). Of particular interest, the steroid hormone biosynthesis pathway was notably suppressed (Fig. 5f), which may contribute to decreased androgen production—a key factor in prostate cancer proliferation, therapy resistance, and metabolic regulation. Previous studies suggest that metastatic prostate cancer cells may downregulate androgen synthesis and cytochrome P450 activity to reduce both proliferation and drug metabolism, potentially facilitating therapeutic evasion [[38], [39], [40]]. Additionally, cell adhesion pathways were suppressed (Fig. 5g), likely reflecting BME-induced extracellular matrix remodeling and its impact on tumor cell adhesion, invasion, and growth [20,41].

Further pathway interrogation using the Hallmark gene set revealed significant negative enrichment of G2/M checkpoint and mitotic spindle pathways, consistent with cell cycle arrest. In contrast, the p53 pathway and DNA repair pathways were positively enriched, suggesting activation of tumor-suppressive responses within the BME (Fig. 5h and i). These transcriptomic signatures align closely with the inhibited proliferative state observed in bone-metastatic prostate cancer tissues.

To evaluate the clinical relevance of our model, we compared the transcriptional profiles of prostate cancer cells in the BME to distant metastasis-derived tumor cells from single-cell RNA-sequencing datasets. Remarkably, 9 out of the 13 Hallmark pathways altered in our model were also differentially regulated in patient-derived metastatic cells, particularly those associated with cell proliferation and oxygen metabolism (Fig. 5j and k).

Consistently, heatmap visualization of key upregulated and downregulated genes in bone metastasis single-cell data—compared to primary prostate cancer—showed strong concordance with gene expression patterns in our BME model (Fig. 5l and m). Bidirectional ssGSEA analyses further validated this overlap: the top 100 DEGs in patient metastases recapitulated the same trends in our bulk transcriptomic data (Fig. 5n), and vice versa (Fig. 5o). These results provide compelling evidence that our 3D-printed BME model faithfully reproduces the proliferation-inhibited phenotype characteristic of clinical bone metastases.

Therefore, our 3D-printed biomimetic BME model not only suppresses prostate cancer cell proliferation in vitro but also recapitulates the molecular hallmarks of clinical bone metastasis. This system offers a robust and clinically relevant platform for mechanistic studies and therapeutic screening targeting the bone metastatic niche.

2.6. Potential mechanisms underlying bone microenvironment-induced proliferation inhibition in prostate cancer cells

To elucidate underlying mechanisms, we employed multiply bioinformatics approaches including differentially expressed genes (DEGs) analysis, pathway enrichment analysis, metabolic flux prediction, and single-cell RNA-sequencing (scRNAseq). The comparison in transcriptional pattern between contrast of BME-cultured cells to control cells and contrast of distal tumor cells to primary tumor cells isolated from patient's bone marrow reveals remarkable consensus. These overlapping signatures converge on four main interconnected mechanisms: oxidative stress and immediate early response, MAPK and growth signal negative regulation, mitochondrial and metabolic reprogramming, and prostate differentiation and inhibition (Supplementary Fig. 7a–i). Collectively, these molecular features suggest suppression in cell cycle progression (e.g., G2/M checkpoint and mitotic spindle downregulation; Fig. 5h and i), metabolic shift (e.g., steroid hormone biosynthesis suppression; Fig. 5e and f), and onset of therapeutic resistance (Fig. 6k and l), provide mechanistic insights into BME-recapitulated proliferative inhibition and potential therapeutic opportunities.

Fig. 6.

Cell interactions and drug resistance in the 3D-printed BME. a) Schematic illustrating cellular interactions among prostate cancer cells, MSCs, and osteoblasts within the BME. b) Ligand-receptor interaction network among cancer cells, MSCs, and osteoblasts in the BME, identified using the iTALK algorithm. c) Top 20 ligand-receptor pairs between MSCs/osteoblasts and tumor cells in the BME (iTALK analysis). d, e) Comparison of the top 100 ligand-receptor pairs from bulk RNA-seq and single-cell RNA-seq data, highlighting overlapping interactions between MSCs/osteoblasts and tumor cells. f) Schematic and fluorescence images showing tumor spheroids cultured in BME or control (3D-only) conditions under 10 μM enzalutamide treatment; images captured on days 0, 3, 5, and 7. g) Following enzalutamide administration, a quantitative comparison of the relative proliferation (normalized to Day 1) of C4-2B tumor spheroids between the control and BME-treated groups was conducted over time (Days 1, 3, 5, and 7). h) Schematic of flow cytometry sorting and analysis of dissociated tumor cells from BME and control groups, stained for live/dead and apoptosis markers. i, j) Flow cytometry quantification of dead and apoptotic tumor cells in BME and control groups on days 1 and 7 post-enzalutamide treatment. k) KEGG enrichment analysis of DEGs showing altered metabolism-related pathways in BME compared to control. l) GSEA highlighting differences in MYC targets (V1) and fatty acid metabolism pathways between BME and control tumor cells. m) Kaplan-Meier curve showing association of BME-derived gene signatures (bulk RNA-seq) with reduced progression-free survival in TCGA prostate cancer cohort (p = 0.0576). n) Kaplan-Meier curve showing association of distal tumor gene signatures (single-cell RNA-seq) with reduced progression-free survival in TCGA dataset (p = 0.0595). A paired t-test was employed to ascertain significance: ns (not significant), *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001, respectively.

The BME elicits an adaptive oxidative stress response that quenches accumulative reactive oxygen species (ROS) for cell survival [42]. Among the common DEGs are the upregulated ROMO1, SH3BGRL3 and S100A10, and downregulated RAD50/SETX and USP8 (Supplementary Fig. 7a). ROMO1 acts as a ROS amplifier to promote growth inhibition. SH3BGRL3 is reported to enhance ROS-mediated anti-proliferative effects, and S100A10 is known to bolster stress resilience, while RAD50/SETX could impair DNA repair and USP8 can disrupt ROS scavenging. The collaborative regulation of these five genes presumably results in augmentation in ROS clearance and DNA repair [43,44], leading to improved cell survival under environmental stress at the cost of proliferative arrest.

This aligns with the enrichment in KEGG pathways including "Chemical carcinogenesis - ROS" and neurodegeneration-related clusters, linking ROS and protein misfolding to cancer inhibition (Supplementary Fig. 7e). The enrichment in Hallmark pathway "oxidative phosphorylation" suggest the excess ROS production is driven by active mitochondrial respiratory activity, while enriched METAFlux pathway "Xenobiotics metabolism" supports augmented detoxification for cell survival (Supplementary Fig. 7f–i). Integrated with mitosis downregulation and DNA repair enrichment, this ROS-dependent signaling axis highlights oxidative vulnerability as potential target for eliminating metastatic prostate cancer cells [45,46].

By dampening MAPK signaling, the BME restricts growth factor-driven proliferation, representing a critical niche-specific adaptation for growth inhibition [47]. The overlapping DEGs feature upregulation of ARPC1B (supporting cytoskeletal stabilizing to limit migration and proliferation) [48] and GNG5 (suppressing MAPK activation), alongside downregulation of TAOK1, TRIM33, and YES1 (attenuating ERK/MAPK, Src, and TGF-β signaling) [49,50] (Supplementary Fig. 7b). Downregulation of KEGG pathway "FoxO signaling" further fine-tunes stress-induced arrest (Supplementary Fig. 7e), while downregulation of Hallmark pathways "Kras signaling up" and "mitotic spindle" indicates suppressed RAS/MAPK-driven mitosis (Supplementary Fig. 7f–g). METAFlux analysis also reveals downregulation of "Ether lipid metabolism", implicating impaired lipid homeostasis (Supplementary Fig. 7h–i). Extending the ligand-receptor findings (e.g., counterbalanced collagen-ITGB1; Fig. 6b–e), these results collectively position MAPK as a gatekeeper of proliferation control, with potential implications of its inhibition for overcoming resistance in AR-dependent prostate cancer.

Mitochondrial reprogramming in the BME favors oxidative phosphorylation (OXPHOS) over glycolysis, thereby reducing anabolic demands while enhancing stress survival in proliferation-inhibited cells [51,52]. Among upregulated DEGs are electron transport chain components, such as COX6B1, COX8A, NDUFB1, ROMO1 and UQCR10. Overexpression of these genes is thought to enhance energetic efficiency and maintain low glycolytic flux in an proliferation-inhibited state, while downregulation of NAMPT and SLC25A36 limits NAD + salvage, glycolysis and nucleotide synthesis [53] (Supplementary Fig. 7d). In line with this, KEGG and Hallmark pathway "Oxidative phosphorylation" is enriched in BME group, demonstrating elevated bioenergetic level (Supplementary Fig. 7e–g), while enriched Hallmark pathway "MYC targets" may sustain tumorigenesis amid proliferative suppression (Supplementary Fig. 7f–g). METAFlux analysis revealed upregulated "Oxidative phosphorylation", downregulated "Pyruvate metabolism" and (up or down?) fatty acid metabolism in resistance, supporting the notion that tumor cells under growth arrest prefer OXPHOS over glycolysis (Fig. 6k and l; Supplementary Fig. 7h–i). This findings align with mechanism of enzalutamide resistance (Fig. 6g–j), suggesting OXPHOS inhibitors as adjuncts to target latent metastases [54].

The BME fosters a differentiated, non-aggressive phenotype by blocking dedifferentiation signals, thereby limiting metastatic potential [55,56]. This process is reinforced by upregulation of APOD (an androgen-regulated proliferation suppressor that promotes differentiation), together with downregulation of SEMA3C (blocking pro-metastatic cues) [57] and TRIM33 (enhancing TGF-β-mediated arrest) (Supplementary Fig. 7c). Consistently, KEGG pathway "Adherens junction" and "Bile secretion" are downregulated, reflecting disruption of dedifferentiation-promoting interactions (Supplementary Fig. 7e), while METAFlux analysis indicates downregulation of "Steroid hormone biosynthesis", curtailing androgen-driven proliferation (Supplementary Fig. 7h–i). Integrated with evidence of AR signaling (Fig. 1k) and mitotic suppression (Fig. 1j), these findings imply differentiation as a barrier to proliferation, amenable to disruption by AR modulators in AR-positive prostate cancer.

In summary, these synergistic mechanisms—adaptation to oxidative stress, MAPK signaling suppression, mitochondrial reprogramming, and differentiation promotion — sustain proliferation inhibition within the BME, as corroborated by patient data. This not only affirms the fidelity of the model but also identifies exploitable vulnerabilities, including ROS or OXPHOS targeting, that may enable eradication of proliferation-inhibited cells and prevention of relapse.

2.7. Cell interactions and drug resistance in the 3D-Printed BME

To explore intercellular communication within the BME, we applied the iTALK algorithm to analyze ligand-receptor interactions among prostate cancer cells, MSCs, and osteoblasts (Fig. 6b). Our analysis revealed that MSCs and osteoblasts highly express collagen family proteins (COL1A2, COL5A1, COL6A1, COL6A2), which interact with integrin β1 (ITGB1) on tumor cells (Fig. 6c). This interaction activates integrin signaling, potentially promoting tumor cell adhesion, migration, extracellular matrix remodeling, and stemness. Moreover, collagen-ITGB1 engagement may confer resistance to therapy via FAK and Src signaling pathways [58]. Additional interactions of interest include HSP90B1-ERBB2, which may promote resistance to androgen receptor (AR) antagonists through PI3K/Akt signaling [59], and APP-SLC45A3, where APP secreted by MSCs and osteoblasts may induce neuroendocrine differentiation in tumor cells, further contributing to drug resistance [60].

We further compared ligand-receptor interactions in our model with those derived from single-cell transcriptomic datasets of bone metastatic lesions (Fig. 6d and e). Notably, interactions such as collagen-ITGB1 and APP-SLC45A3 were conserved across both datasets, supporting the biological relevance of our 3D-printed model and identifying key communication networks in prostate cancer bone metastasis.

To evaluate therapeutic resistance, we assessed the response of prostate cancer cells to enzalutamide (10 μM, 7 days) in the BME model versus a 3D mono-culture control (Fig. 6f, Supplementary Fig. 4a). Cell viability assays showed significantly higher survival in the BME model on days 3, 5, and 7 (p < 0.05), indicating increased drug resistance (Fig. 6g). Flow cytometry further confirmed reduced proportions of apoptotic and dead cells in the BME model compared to the control on days 1 and 7 (Fig. 6i and j).

Transcriptomic KEGG pathway analysis revealed altered metabolic profiles in tumor cells within the BME, many of which are implicated in drug resistance (Fig. 6k). Hallmark GSEA identified upregulation of MYC targets and fatty acid metabolism pathways (Fig. 6l), both of which have been associated with enzalutamide resistance. For example, Hwang et al. demonstrated that CREB5 enhances AR signaling and MYC expression, promoting drug resistance [61]. Similarly, Panja et al. reported that increased NME2-MYC activity predicts enzalutamide resistance [62]. Additionally, Xu et al. showed that ELOVL5-mediated elevation of polyunsaturated fatty acids (PUFAs) activates the AKT-mTOR axis, contributing to both drug resistance and neuroendocrine differentiation [63].

To further evaluate clinical relevance, we performed Kaplan-Meier analysis using the TCGA Prostate Adenocarcinoma (PRAD) cohort. Patients were stratified based on expression scores calculated from the top 100 upregulated genes identified in the BME model versus control. Patients in the high-expression group exhibited shorter progression-free survival (p = 0.0576) (Fig. 6m). A similar trend was observed using a gene set derived from the single-cell dataset of distant metastases (p = 0.0595) (Fig. 6n).

These results demonstrate that our 3D-printed BME model not only recapitulates the proliferation-inhibited state characteristic of prostate cancer bone metastases but also reflects critical features of drug resistance. This model offers a valuable platform for investigating the mechanisms underlying tumor-stroma interactions and therapeutic resistance in bone metastatic prostate cancer.

3. Discussion

This study addresses a critical and long-standing gap in bone metastasis research by establishing the first in vitro model that faithfully recapitulates the rare, proliferation-inhibited tumor cell phenotype observed in clinical prostate cancer bone metastases. While previous studies have employed co-culture systems—including 3D printing and organ-on-chip technologies—to model aspects of the bone metastatic niche [[64], [65], [66], [67]], these models have not focused on, nor successfully recapitulated, the proliferation-inhibited tumor phenotype seen in clinical prostate cancer bone metastases, and they often lacked in-depth biological characterization or direct alignment with patient-derived molecular features. In contrast, our 3D bioprinted platform overcomes these limitations through a highly biomimetic design. By integrating a trabecular bone-like architecture, a calcium-enriched matrix, marrow-mimicking stiffness, and key stromal cell types—such as MSCs and osteoblasts—we recreate the essential biomechanical and cellular cues that regulate tumor behavior within the bone niche.

Crucially, transcriptomic analysis confirms that tumor cells in this model exhibit a molecular signature strikingly similar to that of patient-derived proliferation-inhibited tumor cells, characterized by the downregulation of proliferative pathways [23]. This phenotype has proven elusive in traditional 2D cultures and most in vivo models, where human-specific microenvironmental interactions cannot be precisely controlled or observed [68]. Consequently, our system provides not only a tractable platform for the mechanistic dissection of metastatic latency and reactivation but also a powerful tool for investigating drug resistance and relapse—clinical phenomena that have long defied detailed modeling [69,70].

A central achievement of this study lies in the faithful recapitulation of the proliferation-inhibited tumor cell phenotype—previously identified through single-cell transcriptomics in clinical prostate cancer bone metastases—within our engineered 3D biomimetic BME. By integrating a calcium phosphate scaffold mimicking trabecular bone, a decellularized extracellular matrix hydrogel simulating marrow softness, and key stromal cells like MSCs and osteoblasts, our model induces a state in prostate cancer cells characterized by suppressed proliferation, cell cycle arrest at G0/G1, and transcriptomic signatures mirroring patient-derived data, including downregulation of mitotic pathways (e.g., G2/M checkpoint and spindle organization) and upregulation of survival mechanisms like DNA repair and p53 signaling [71]. This alignment was rigorously validated through bulk RNA sequencing, GO/KEGG/GSEA analyses, and flow cytometry, demonstrating not only phenotypic but also molecular concordance with distal metastatic cells from scRNA-seq datasets. Such recapitulation addresses a longstanding gap in preclinical models, where traditional 2D cultures or xenografts often fail to capture the subtle, niche-dependent regulation of tumor quiescence, leading to overestimation of proliferative behaviors and limited translational relevance [72,73].

Importantly, the proliferation-inhibited phenotype in our BME is not a nonspecific artifact of the culture environment but a targeted response reflective of cell-intrinsic and niche-specific interactions. This specificity is underscored by the differential behaviors of two bone-tropic prostate cancer cell lines: AR-positive C4-2B cells exhibit marked growth suppression, G0/G1 arrest, and transcriptomic shifts toward quiescence, while the more aggressive, AR-negative PC-3 cells proliferate unchecked and dominate the co-culture [74,75]. Such divergence rules out generalized factors like nutrient limitation or cytotoxicity, instead pointing to AR-dependent vulnerabilities modulated by the bone niche. Literature supports this: C4-2B cells, derived from bone metastases and reliant on androgen signaling, show heightened sensitivity to stromal cues that dampen proliferation [76,77], whereas PC-3 cells, representing castration-resistant disease, resist similar inhibition due to AR independence and enhanced adaptability to bone matrix signals [[78], [79], [80]]. This cell-type selectivity mirrors clinical heterogeneity, where AR-positive tumors may enter inhibited states in bone, contributing to dormancy and resistance, while AR-negative variants drive rapid osteolytic progression [[81], [82], [83]]. By capturing these nuances, our model facilitates tailored investigations into subtype-specific therapies [74,84].

Our integrative transcriptomic analyses identified four interconnected mechanisms driving the proliferation-inhibited state in prostate cancer cells within the biomimetic BME: oxidative stress adaptation and immediate early response, negative regulation of MAPK and growth signaling, mitochondrial and metabolic reprogramming, and promotion of prostate differentiation and inhibition. These mechanisms align closely with established literature on proliferation arrest and bone metastasis [85]. For instance, the adaptive oxidative stress response in our model mirrors ROS-mediated proliferation inhibition in bone-metastatic prostate cancer [86]. Similarly, adipocyte-induced oxidative and ER stress in the bone niche sustains proliferation-arrested tumor cells [87], while broader evidence shows oxidative stress inhibiting distant metastasis in prostate and melanoma models, highlighting ROS as a conserved regulator in hypoxic environments [88,89]. This transitions into MAPK signaling attenuation, which echoes p38-MAPK activation's role in niche-driven proliferation arrest [90,91]; studies show elevated p38-MAPK/ERK ratios from factors like TGF-β2 maintaining inhibited states [92,93], with osteoblast mediators such as GDF10 and BMP7 enforcing similar inhibition via p38 pathways [94]. In parallel, mitochondrial reprogramming toward OXPHOS aligns with metabolic shifts in aggressive prostate tumors [95,96], where elevated OXPHOS supports survival during proliferation arrest [97,98], and inhibitors targeting this pathway reduce outgrowth in bone niches [99]. Complementing this, promotion of differentiation recapitulates niche signals blocking dedifferentiation, as osteoblast factors like BMP7 induce senescence [7], and NG2+/Nestin + MSCs enforce proliferation inhibition through similar mechanisms [7]. Collectively, these alignments affirm our platform's utility for probing proliferation-arrested states in metastasis and identify conserved targets to eliminate persistent cells and prevent relapse [85,100].

Beyond proliferation inhibition, our findings reveal that the bone-mimetic microenvironment also promotes therapeutic resistance through metabolic and signaling reprogramming. Transcriptome analysis shows suppression of G2M checkpoint, mitotic spindle, and DNA repair pathways—hallmarks of a quiescent-like state [[25], [26], [27]]. Notably, key regulators such as KIF14, FN1, and PIK3C2A are downregulated, suggesting impaired DNA replication and altered cytoskeletal organization. In parallel, upregulation of MYC signaling, enhanced fatty acid metabolism, and reduced androgen biosynthesis were observed, aligning with known mechanisms of enzalutamide resistance [[61], [62], [63]]. These metabolic adaptations suggest a shift toward survival-promoting states in the absence of proliferation, reflective of drug-tolerant persister populations.

Although our BME represents a significant technological and biological advancement, it remains a simplified model of the highly complex in vivo ecosystem. This model deliberately focuses on the core structural and stromal components known to primarily regulate tumor cell fate in bone tissue. However, it has not yet incorporated other critical cellular elements, including immune cells (such as macrophages, T cells, and myeloid-derived suppressor cells) and endothelial cells that form the vascular system [23,101]. These components are known to engage in intricate interactions with tumor cells, influencing proliferation inhibition, immune evasion, angiogenesis, and bone remodeling processes [102]. Nevertheless, these limitations underscore the platform's flexibility and substantial potential for future expansion. The modular nature of 3D bioprinting technology enables the systematic introduction of additional cell types and structural features [103]. Additionally, human tissue-derived dECM, such as that from the placenta, offers significant advantages over porcine skin-derived dECM, including greater biocompatibility, reduced immunogenicity, and improved biomimicry. In future studies, we plan to incorporate placental dECM as the hydrogel carrier to further optimize and refine the model system. Future iterations could incorporate perfusable vascular networks to study nutrient exchange and drug delivery [103], or integrate patient-derived immune cells to explore mechanisms of immune surveillance and evasion within the bone niche [104,105]. Furthermore, this model serves as a powerful and clinically relevant tool for therapeutic strategy development, capable of authentically recapitulating the characteristics of "drug-tolerant persister cells," making it an ideal platform for high-throughput screening of novel therapeutics aimed at specifically eradicating such highly resilient cell populations [106,107].

4. Conclusion

In conclusion, this study introduces an advanced 3D bioprinted biomimetic bone microenvironment model that, for the first time, faithfully recapitulates—at both transcriptomic and phenotypic levels—the hallmark features of the proliferation-inhibited state observed in clinical prostate cancer bone metastases. By integrating key structural, mechanical, and cellular components of the bone niche, this platform successfully induces prostate cancer cells into a state of growth arrest and therapy resistance, validated at the molecular level using patient-derived data. Our findings elucidate the complex integrative mechanisms by which the BME orchestrates this state transition through cascades of stress responses, signal attenuation, and metabolic reprogramming. This platform not only deepens our understanding of the intricate interplay between the metastatic microenvironment, tumor proliferation inhibition, and drug resistance but also provides a transformative research tool for developing novel therapeutic strategies aimed at eradicating residual lesions that drive metastatic relapse.

5. Methods

5.1. Single-cell RNA sequencing analysis

Data sources and preprocessing: Single-cell RNA sequencing (scRNA-seq) data were obtained from public datasets GSE143791 [23] and GSE176031 [24]. GSE143791 includes samples from nine patients with spinal metastatic prostate cancer, covering three anatomical compartments: solid metastatic tissue (Tumor), bone marrow adjacent to the tumor (Involved), and distal bone marrow within the surgical field but anatomically distant from the tumor site (Distal). GSE176031 comprises primary prostate tumor samples from 11 patients who underwent radical prostatectomy.

Quality control and integration: Cells were filtered using the following criteria: 200-50,000 unique molecular identifiers (UMIs), 100-6000 detected features, and the proportion of mitochondrial genes <15 %. Genes expressed in fewer than three cells were excluded. Batch correction and data integration were performed using the Harmony package (v1.2.1). Dimensionality reduction was achieved with Uniform Manifold Approximation and Projection (UMAP), followed by unsupervised clustering using the Louvain algorithm. DEGs were identified using the findallmarks function, which employs Wilcoxon rank-sum testing with a significance threshold of p < 0.05.

Cell type annotation and tumor cell identification:Tumor cells in bone metastasis samples were annotated based on the expression of prostate cancer-specific markers (KLK2, KLK3, KLK4, AR), resulting in the identification of 885 tumor cells. These were further classified by anatomical site (Tumor, Involved, Distal) to derive a bone metastasis-specific gene signature. For primary tumor samples, scRNA-seq data were processed similarly, and 2502 tumor cells were identified using inferCNV (v1.23.0), a tool implemented in Python 3.8 that infers copy number variations to confirm malignancy (https://github.com/broadinstitute/inferCNV).

Pathway enrichment analysis:Pathway analysis was performed using clusterProfiler (v4.4.4) and gene sets from the Molecular Signatures Database (MSigDB) (https://www.gsea-msigdb.org/gsea/msigdb). Key pathways related to proliferation, such as the G2/M checkpoint, mitotic spindle, and DNA repair, were compared between distal tumor cells and primary tumor cells to assess microenvironmental effects on cell cycling.

Ligand-receptor interaction analysis:Intercellular communication was analyzed using the iTALK algorithm (https://github.com/Coolgenome/iTALK), focusing on interactions among tumor cells, MSCs, and osteoblasts. Notable ligand-receptor pairs included collagen-integrin signaling (e.g., COL1A1, COL1A2, COL3A1, COL4A1, COL6A2 with ITGB1) and LAMA2/LAMB2-RPSA. The analysis leveraged iTALK's curated database of 2648 ligand-receptor pairs to construct intercellular interaction networks.

5.2. Metabolic flux analysis using METAFlux

Bulk RNA-seq data were processed to obtain transcripts per million (TPM) values after quality control, alignment, and normalization. Metabolic pathway activities were inferred using the METAFlux R package (https://github.com/KChen-lab/METAFlux), which applies flux balance analysis (FBA) on the Human1 genome-scale metabolic model (13,082 reactions, 8378 metabolites). Gene expression was mapped to reactions via Gene-Protein-Reaction rules to compute Metabolic Reaction Activity Scores (MRAS): for enzyme complexes (AND logic), the minimum subunit expression; for isoenzymes (OR logic), the sum. MRAS normalized across samples set flux bounds (upper: MRAS; lower: 0 or -MRAS for reversible reactions). Non-enzyme reactions had fixed bounds at 1. Differential activities between groups were assessed via Wilcoxon tests, with visualizations including heatmaps and enrichment plots.

5.3. Bulk RNA sequencing analysis

Sample preparation and sequencing: To investigate the influence of the 3D-printed biomimetic BME, bulk RNA sequencing was performed on prostate cancer cells (C4-2B), MSCs, and osteoblasts cultured within the BME and compared to cells cultured separately in 3D. RNA was sequenced on the MGI DNBSEQ-T7 platform (Kangce Technology Co., Ltd.) using paired-end sequencing to ensure high coverage and accuracy.

Differential expression and pathway analysis: Gene expression was quantified using raw count data. DEGs between BME-cultured cells and 3D monocultures were identified via the Wilcoxon rank-sum test (p < 0.05), with data normalized using the DESeq2 R package. Visualizations, including heatmaps and volcano plots, were generated using the ggplot2 package in R. GO and KEGG pathway enrichment analyses were conducted using clusterProfiler (FDR <0.05). Additionally, GSEA was performed using MSigDB-derived gene sets to identify enrichment trends in proliferation- and mitosis-related pathways.

Hallmark pathway analysis: To assess the regulatory effects of the BME on tumor proliferation and metabolism, Hallmark GSEA was conducted. Key proliferative pathways (e.g., G2/M checkpoint, mitotic spindle) and anti-proliferative pathways (e.g., p53 signaling, DNA repair) were examined.

Ligand-receptor interaction analysis: Further intercellular interaction analysis was performed using iTALK on bulk RNA-seq data, focusing on tumor cell, MSC, and osteoblast crosstalk. Key interactions included COL1A2, COL5A1, COL6A1/2 with ITGB1, and HSP90B1-ERBB2 and APP-SLC45A3 ligand-receptor pairs. Interaction networks were visualized using RStudio and Adobe Illustrator.

5.4. Materials preparation

Preparation of porcine skin-derived dECM: Porcine dermal extracellular matrix was prepared using a multistep decellularization protocol. Fresh porcine skin obtained from a local abattoir was initially decontaminated by alternating rinses with 75 % ethanol. Following surface sterilization, the dermal layer was enzymatically digested using 0.25 % Trypsin-EDTA. Residual agents were removed via multiple rinses in distilled water. The tissue was then immersed in a decellularization solution comprising 1 % Triton X-100 (detergent), 0.25 % EDTA (chelating agent), and 0.69 % Tris buffer (pH stabilizer), and agitated at room temperature (20–25 °C) for 12–16 h. The decellularized tissue was solubilized in 1 % (w/v) pepsin at pH 2.0, followed by dialysis for 7 days to remove residual cellular material and reagents. The resulting solution was lyophilized and reconstituted in hydrochloric acid (adjusted to pH 3.5) to yield a 15 mg/mL dECM solution for downstream applications.

Proteomic Analysis of dECM by Mass Spectrometry:“To characterize the protein composition of the porcine skin-derived dECM, lyophilized hydrogel samples were analyzed by liquid chromatography-tandem mass spectrometry (LC-MS/MS). Briefly, 50 μg oflyophilized dECM powder was solubilized and denatured in a buffer containing 1 % SDS. Disulfide bonds were reduced with dithiothreitol (DTT) and subsequently alkylated with iodoacetamide (IAA). The proteins were then subjected to in-solution digestion using sequencing-grade trypsin overnight at 37 °C. The resulting peptide mixture was desalted using a C18 ZipTip. LC-MS/MS analysis was performed on a Q-Exactive HF mass spectrometer (Thermo Fisher Scientific) coupled with an Easy-nLC 1200 system. Peptides were separated on a C18 analytical column with a linear gradient. The mass spectrometer was operated in data-dependent acquisition (DDA) mode. Raw data files were processed using MaxQuant software (version 1.6.17.0) and searched against the Sus scrofa (porcine) UniProt database. The relative abundance of identified proteins was estimated using label-free quantification (LFQ) intensities.

Rheological and mechanical characterization of dECM hydrogel: Rheological measurements were performed using a rotational rheometer (Physica MCR301, Anton Paar, Austria). A steady shear sweep (0.1–100 s⁻¹) was conducted at 15 °C to evaluate viscosity, followed by dynamic frequency sweep tests to assess the storage and loss moduli of the crosslinked hydrogel. Mechanical stiffness was evaluated using a Piuma Nanoindenter (Optics11 Life, Netherlands) with a 0.26 N/m cantilever and 25.5 μm tip radius. Tests were conducted in PBS, and Young's modulus was mapped over a 500 × 500 μm² area at 100 μm intervals. The mechanical stiffness testing of porcine bone marrow was performed as described above.Data were analyzed using OriginLab 13.0 and reported as mean ± standard deviation.

Swelling and degradation analysis: Swelling behavior was assessed gravimetrically. Lyophilized hydrogels were weighed (m₀) and incubated in PBS at 37 °C. At specific time points, samples were blotted and weighed (m_t). The swelling ratio (%) was calculated as:

Swelling ratio = (m_t − m₀) / m₀ × 100 %

Water uptake capacity was defined by the equilibrium swelling ratio at 24 h.

For degradation assessment, hydrogels were incubated in an enzymatic solution (1.5 μg/mL lysozyme and 0.03 μg/mL collagenase in PBS, pH 7.4) at 37 °C for 14 days. The solution was refreshed every 48 h. At selected time points (days 2, 4, 6, 8, 10, 12, 14), samples were washed, lyophilized, and weighed (m_t). Degradation ratio (%) was calculated as:

Degradation ratio = (m₀ − m_t) / m₀ × 100 %

Composite bone cement fabrication: The composite bone cement consisted of 60 % CPC, 20 % barium sulfate, and 20 % pregelatinized corn starch. CPC powder was prepared by mixing α-tricalcium phosphate and dicalcium phosphate dihydrate in a 9:1 mass ratio, followed by 24-h ball milling in anhydrous ethanol. The suspension was oven-dried at 60 °C, ground, and sieved through a 60-mesh screen. Barium sulfate powder was similarly processed and sieved through a 200-mesh screen.

3D bioprinting of bone cement structures:A multi-nozzle extrusion-based 3D printer (SunP Biomaker 2, SunP Biotech, China) was used under sterile conditions. Bioink was prepared by mixing 5 % w/v CPC with sterile water and loaded into a syringe fitted with a 25 G needle. During printing, the cement was maintained at 15 °C. A 3-layer grid structure (10 mm × 10 mm, 0.9 mm line spacing, 0.3 mm layer height) was printed at 2.8 mm/s with an extrusion speed of 0.8 mm/s. Printed constructs were incubated at 37 °C for solidification.

Scanning electron microscopy (SEM):Microstructural analysis of dECM hydrogel and CPC was performed using a ZEISS Sigma 300 SEM (Germany). Prior to imaging, samples were freeze-dried and sputter-coated with platinum to enhance conductivity. Specimens were mounted and imaged at various magnifications to examine morphological features.

5.5. In vitro experiments

Cell lines and culture: Primary mouse bone marrow mesenchymal stem cells (mBMSCs) were obtained from Procell Bioscience Inc. (China). MC3T3-E1-S14 pre-osteoblasts were purchased from AoRuiCell (China). Also, C4-2B prostate cancer cells were acquired from the American Type Culture Collection (ATCC; USA). All these cells were cultured in MEMα complete medium (GIBCO; C12571500BT; USA) supplemented with 10 % fetal bovine serum (FBS) (Vazyme; F103; China) and 1 % penicillin - streptomycin. The culture conditions were a humidified atmosphere at 37 °C with 5 % CO₂, and the medium was changed every 48 h. When the cells reached approximately 80 % confluence, they were passaged using 0.25 % trypsin-EDTA (GIBCO; 25200072; USA). For hydrogel co-culture experiments, sterilized hydrogel constructs were incubated in confocal dishes under standard culture conditions, with daily medium replacement. Osteogenic differentiation of mBMSCs was induced using osteoinductive medium supplemented with 50 μM ascorbic acid, 10 nM dexamethasone, and 10 mM β-glycerophosphate.

Lentiviral transduction: Stable expression of fluorescent and luminescent proteins was achieved via lentiviral transduction (HBLV-EGFP-PURO, HBLV-mCherry-PURO, HBLV-LUC-PURO, HBLV-EBFP-PURO; OBiO, China). Cells were seeded in 6-well plates and allowed to adhere overnight. The culture medium was then replaced with lentivirus-containing medium at the appropriate multiplicity of infection (MOI): 10 for MC3T3, 30 for MSCs, and 20 for C4-2B, supplemented with 6 μg/mL polybrene (Hanbio, China). Fluorescent signal expression was assessed after 72 h. Cells were selected with 2 μg/mL puromycin (ST551, Beyotime, China) to establish stable lines: mCherry-labeled MC3T3, EGFP-labeled MSCs, and EBFP-labeled C4-2B cells.

Preparation of cell-laden hydrogels: Cell-laden hydrogels were prepared by gently mixing MSCs, MC3T3, and C4-2B cells with a bioink composed of 10 × DMEM, 2.5 % (w/v) sodium alginate, dECM, and 10 mM CaCl₂. A tri-cellular construct was generated by spatially controlling the distribution of each cell type within the hydrogel.

Construction of the biomimetic BME: To construct the in vitro BME, MC3T3 cells were pre-seeded onto 3D-printed CPC scaffolds for 3 days to promote adhesion. Meanwhile, C4-2B tumor microspheres (2000–10,000 cells per microsphere) were prepared. These microspheres and MSCs were incorporated into the dECM hydrogel matrix, and the pre-seeded CPC scaffolds were embedded within this composite hydrogel system.

Flow cytometry analysis: To recover cells from the hydrogel-based bone BME, constructs were treated with 50 mg/mL sodium citrate (S1804, Sigma-Aldrich) for 25 min to dissolve alginate, followed by 1 mg/mL collagenase (C0130, Sigma-Aldrich) digestion for 30 min with intermittent pipetting. Subsequently, 0.25 % trypsin was applied to detach cells from CPC scaffolds. Cell suspensions were stained using Cell Cycle Assay Kit Plus (C598381, Aladdin, China), Live/Dead Fixable Aqua (L34957, Invitrogen, USA), FITC Plus Anti-Human Ki-67 Rabbit Recombinant Antibody(FITC-98143-100tests, Proteintech, China) and Annexin V-FITC/PI Apoptosis Kit (A211-02, Vazyme, China). After centrifugation (1100 rpm, 5 min), cell pellets were resuspended in PBS, filtered through a 40 μm cell strainer, and adjusted to 5 × 10⁶ cells/mL in PBS with 2 % FBS. Analyses were performed using a BD LSRFortessa flow cytometer.

Confocal microscopy and image analysis: Samples were imaged at days 3, 7, 14, and 21 using a Zeiss LSM 510 Meta confocal microscope (Zeiss, Germany). Imaging settings were as follows: enhanced green fluorescent protein (GFP) (Ex/Em: 488/515 nm), monomeric cherry (mCherry) (Ex/Em: 570/602 nm), and enhanced blue fluorescent protein (BFP) (Ex/Em: 380/440 nm). Quantitative analysis of fluorescent signal intensity was performed using ImageJ software (v1.53, NIH). To evaluate longitudinal cellular proliferation dynamics, the integrated fluorescent area for each distinct cell population at specified experimental time points (Day 3, Day 7, Day 14) was normalized relative to the corresponding baseline measurement obtained at Day 1 for that specific cell type. Normalized proliferation ratios were calculated using the formula: Relative Proliferation = (Fluorescent Area at Day X/Fluorescent Area at Day 1). This normalization approach accounts for initial seeding density variations while enabling quantitative comparison of proliferation kinetics across different cell types and experimental conditions.

ALP staining and ARS: Osteogenic activity was assessed via ALP and ARS staining. For ALP staining, MSC-laden hydrogels and scaffolds were cultured for 7 days, fixed in 4 % paraformaldehyde for 1 h after PBS washes, and stained using the ALP kit (P0321S, Beyotime, China). Samples were visualized using an inverted light microscope (Leica). For ARS staining, MSC-laden constructs were cultured for 14 days. After fixation with 4 % paraformaldehyde for 15 min, samples were washed and incubated with 1 % Alizarin Red S solution (pH 4.2) for 30 min. Excess stain was removed via repeated washing, and mineral deposition was visualized using an inverted light microscope (Leica).

RNA Isolation and quantitative Real-Time PCR: Total RNA was extracted from hydrogel/scaffold constructs using Trizol reagent (4478359, Invitrogen, USA). Constructs were first treated with 500 μL sodium citrate (50 mg/mL, 10 min) followed by 2 mg/mL collagenase digestion for 25 min at 37 °C to dissociate cells. After centrifugation, RNA was isolated from cell pellets. cDNA synthesis and quantitative PCR were performed using SYBR Green Master Mix. Relative gene expression was normalized to Gapdh using the 2^−ΔΔCt method.

5.6. Primer sequences

Gapdh-F: 5′-AGGTCGGTGTGAACGGATTTG-3′

Gapdh-R: 5′-GGGGTCGTTGATGGCAACA-3′

Bglap-F: 5′-GCAATAAGGTAGTGAACAGACTCC-3′

Bglap: 5′-CCATAGATGCGTTTGTAGGCGG-3′

Spp1-F: 5′-CACCATTCGGATGAGTCTGA-3′

Spp1-R: 5′-CCTCAGTCCATAAGCCAAGC-3′

Ibsp-F: 5′-ATGGAGACGGCGATAGTTCC-3′

Ibsp-R: 5′-CTAGCTGTTACACCCGAGAGT-3′

Runx2-F: 5′-GACTGTGGTTACCGTCATGGC-3′

Runx2-R: 5′-ACTTGGTTTTTCATAACAGCGGA-3′

Alpl-F: 5′-CCAACTCTTTTGTGCCAGAGA-3′

Alpl-R: 5′-GGCTACATTGGTGTTGAGCTTTT-3′

Sp7-F: 5′-GGAAAGGAGGCACAAAGAAGC-3′

Sp7-R: 5′-CCCCTTAGGCACTAGGAGC-3′

5.7. In vivo experiments

Animal ethics and housing: All animal experiments were approved by the Institutional Review Board and Animal Care Committee of Guangzhou Linfutuopu Testing Co. Ltd (Approval No. LFTOP-IACUC-2025-0029) and conducted at the First Affiliated Hospital of Sun Yat-sen University. Four-week-old male nude mice (average weight 18 ± 1.2 g) were used for all experiments. Mice were housed six per cage in a temperature-controlled environment (20 ± 1.0 °C, 40–70 % humidity) with free access to food and water.

Experimental grouping and surgical procedure: A total of ten mice were randomly divided into two groups:

Control group (n = 5)

Biomimetic BME group (n = 5)

On the day of surgery, mice were anesthetized using isoflurane (3–5 % for induction, 1.5 % for maintenance in air). Following skin disinfection, a 1.5-cm incision was made under the axilla. Hydrogel-encapsulated scaffolds were then implanted subcutaneously.

Bioluminescence imaging and sample collection: Bioluminescence imaging (BLI) was conducted on days 0, 7, 14, and 21 to monitor cell activity at the implantation site. At 21 days post-surgery, all mice were sacrificed. The implanted scaffolds and surrounding tissues were collected for further analysis, including measurement of sample volume and weight.

Histological processing and staining: Collected tissues were decalcified in 10 % EDTA for four weeks. Following decalcification, samples were dehydrated through a graded ethanol series and embedded in paraffin. Sections of 5 μm thickness were prepared for histological evaluation. Hematoxylin and Eosin (H&E) staining was used to assess general tissue morphology. Masson's trichrome staining was performed to evaluate collagen deposition. Immunohistochemistry staining was performed to visualize spatial expression of the osteogenic marker OSX. All staining protocols followed the manufacturer's instructions (Servicebio, Hubei, China).

5.8. Statistical analysis

All statistical analyses were conducted using R software (version 4.3.0.1). DEGs were identified using the Wilcoxon rank-sum test, with statistical significance defined as p < 0.05. For pathway enrichment analysis, a false discovery rate (FDR) threshold of <0.05 was applied to determine significance. Data visualization, including UMAP projections, DEGs heatmaps, volcano plots, and GO/KEGG enrichment graphs, was performed using the ggplot2, Seurat, and clusterProfiler packages. These tools ensured robust data interpretation and high-quality graphical representation of results.

CRediT authorship contribution statement

Xin Chen: Writing – review & editing, Writing – original draft, Investigation, Formal analysis, Data curation, Conceptualization. Yujiao Peng: Writing – review & editing, Writing – original draft, Software, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Ying Zhao: Writing – review & editing, Writing – original draft, Formal analysis, Data curation, Conceptualization. Huiling Liu: Methodology. Qijun Lin: Data curation. Xihong Fu: Data curation. Lianheng Chen: Formal analysis. Zhongte Peng: Formal analysis. Jianfeng Huang: Formal analysis. Yu Luo: Methodology. Xuenong Zou: Resources. Lei Yang: Resources. Xinsheng Peng: Writing – review & editing, Writing – original draft, Resources, Project administration, Funding acquisition. Chun Liu: Writing – review & editing, Writing – original draft, Funding acquisition.

Ethics approval and consent to participate

All animal experiments were conducted in strict accordance with ethical guidelines and protocols. The study received full ethical approval from the Institutional Review Board and Animal Care Committee of Guangzhou Linfutuopu Testing Co. Ltd. (Approval Number: LFTOP-IACUC-2025-0029). All efforts were made to minimize animal suffering and to reduce the number of animals used. As this study involved only animal experiments, obtaining consent to participate from human subjects was not applicable.

Declaration of competing interest

Lei Yang is an editorial board member for Bioactive Materials and was not involved in the editorial review or the decision to publish this article. All authors declare that there are no competing interests.

Appendix A. Supplementary data

The following is the Supplementary data to this article.