Bioengineered human bone-mimetic niche compositions regulate breast cancer cell quiescence and therapy response

bone metastasis , bone marrow niche, tumor dormancy, perivascular niche, 3D 26 bioreactor culture 27 This PDF file includes: 28 Main Text 29 Figures 1 to 4 30 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.18.695089doi: bioRxiv preprint 2

1 Disseminated tumor cells (DTCs) in the bone are widely regarded as the cellular seeds of late 2 metastatic relapse in estrogen receptor-positive (ER+) breast cancer (BC). However, how distinct 3 bone compartments influence BC cell quiescence and endocrine therapy response remains 4 unclear. Mechanistic insight has been hindered by limited access to human bone samples and by 5 the lack of relevant human models that permit controlled manipulation of stromal compartments. 6 Here, we developed a fully human, 3D bone-mimetic system for modular, controllable assembly of 7 engineered osteoblastic (eON), vascularized (eVN), and vascularized osteoblastic (eVON) niche 8 compositions. These niches were generated by perfusion culture of human bone marrow-derived 9 mesenchymal stromal cells (hBM-MSCs) and/or human adipose tissue-derived stromal vascular 10 fraction (hAT-SVF) cells within porous ceramic scaffolds. The resulting tissue microenvironments 11 were then used as a substrate for the culture of an ER+ BC cell line, expressing a mutant reporter 12 of p27 to monitor the quiescent status . We found that the eON enhanced BC cell proliferation, 13 whereas the eVN was enriched in quiescent BC cells positive for NR2F1, a dormancy-associated 14 transcription factor, and located near perivascular elements. Treatment with the selective ER 15 degrader fulvestrant reduced BC cell numbers in vascularized niches (eVN and eVON) but not in 16 eON, despite comparable receptor degradation. In summary, we developed a modular human 17 platform for dissecting niche -specific regulation of BC quiescence, proliferation, and endocrine 18 therapy response. The system can be further used to investigate perivascular niche-dependent 19 mechanisms of BC cell dormancy and to guide the development of therapeutic strategies 20 preventing recurrence in ER+ BC patients. 21 22 Main Text 23

24 Bone is the predominant site of metastatic relapse in estrogen receptor -positive (ER+) breast 25 cancer (BC) [1-4]. Disseminated tumor cells (DTCs) are frequently detected in bone marrow (BM) 26 aspirates of patients lacking clinically detectable metastases. The presence of DTCs in the bone 27 compartments correlates with poor prognosis, highlighting their potential as seeds for future 28 metastases [5-9]. Understanding the contribution of bone niches to DTC outgrowth is therefore 29 critical for developing strategies to prevent metastatic relapse. 30 Within the bone, DTCs localize to specialized microenvironments, primarily the osteogenic 31 and perivascular niches [10, 11] . The osteogenic niche comprises osteoblast - and osteoclast-32 lineage cells lining the bone surface [12]. The osteoclast-driven “vicious cycle” of advanced stage 33 of bone metastasis, marked by tumor-induced osteolysis and release of growth factors that promote 34 further tumor growth, is well -characterized [2, 9, 13 -16]. Once the osteolytic lesions emerge, 35 treatment is largely palliative [17-21]. In contrast, the contribution of osteoblasts during early 36 metastatic stages is understudied [12, 22]. Growing evidence suggests that osteoblasts promote 37 early colonization, therapeutic resistance, and proliferation of ER+ BC cells [23-25]. Targeting these 38 early cellular interactions, before the onset of irreversible osteolysis, may represent a more effective 39 therapeutic window. 40 The other specialized bone compartment to which DTCs mainly traffic is the perivascular 41 niche, which is often anatomically adjacent to or overlapping with the osteogenic niche [26]. While 42 the osteogenic niche is primarily associated with promoting DTC proliferation during early 43 metastatic progression, prior work has shown that the perivascular niche can sustain DTC 44 dormancy [27]. Using intravital imaging on bone marrow (BM) of a BC xenograft model, DTCs were 45 found predominantly localized near perisinusoidal vessels. Analyses of patient BM samples also 46 revealed that non-proliferative, Ki67-negative (Ki67-) DTCs are more frequently found adjacent to 47 sinusoidal vessels than endosteal surface [26]. Thrombospondin 1 (TSP1) expression was strongly 48 associated with BC cell dormancy in an in vitro organotypic model composed of endothelial and 49 stromal cells, mimicking the perivascular niche [27]. Despite their distinct features, the osteogenic 50 and perivascular niches are interconnected through shared mesenchymal progenitors that give rise 51 to both endothelial and osteoblastic lineages. Indeed, specialized type H capillaries, enriched in 52 CD31 and endomucin (encoded by Emcn in mice), typically reside near the endosteum and 53 contribute to both angiogenesis and osteogenesis via factors such as Noggin [10, 28-30]. These 54 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.18.695089doi: bioRxiv preprint 3 developmental and anatomical overlaps underscore the importance of not only dissecting the 55 individual effects of osteogenic and perivascular microenvironments on DTC outgrowth, but also of 56 modeling transitional zones, where osteogenic and vascular cues converge. 57 Although recent in vivo studies have advanced understanding of how discrete bone 58 compartments influence DTC outgrowth, they also underscore the need for fully human ex vivo 59 systems to dissect and compare niche -specific contributions to metastatic progression. A central 60

is the scarcity of solitary DTCs in mouse bone, which hampers spatial and functional 61 interrogation of discrete bone niches during early metastatic stages. Additionally, murine models 62 often fail to recapitulate spontaneous tumor cell dissemination and are further constrained by the 63 rarity of ER+ mammary tumor cell lines compatible with immunocompetent hosts [31-33]. 64 Complementing these challenges, existing in vitro and ex vivo models often lack the modularity to 65 independently engineer bone-mimetic niches and the dynamic perfusion required for physiological 66 delivery of oxygen, nutrients, and drugs [27, 34-40]. Together, these limitations restrict the ability 67 to resolve niche-specific effects on BC cell dormancy and reactivation, particularly in ER+ disease, 68 where the underlying mechanisms remain poorly understood. 69 We have previously developed a human, bioreactor -based 3D BM niche model using 70 hydroxyapatite scaffolds seeded with human BM-derived MSCs (hBM -MSCs), and/or adipose 71 tissue-derived stromal vascular fraction (hAT -SVF) cells [41-43]. These engineered constructs 72 mimic the mineralized architecture of trabecular bone with or without vascularization . They also 73 reproduce key features of mineralized and perivascular bone regions [42, 44, 45]. 74 In this study, we investigated the influence of distinct engineered bone-mimetic niches on 75 BC cell quiescence by adapting our previously established 3D perfusion-based culture systems. 76 This platform enables modular engineering of osteogenic, vascular, and vascularized osteogenic 77 bone niches under controlled conditions. We used MCF7-tdTomato/mVenus-p27K- cells to track 78 BC cells in G0 cell cycle arrest by a mutant p27 reporter [46, 47] . To validate niche -specific 79 dormancy, we further evaluated the expression of nuclear receptor subfamily 2, group F member 80 1 (NR2F1), an orphan nuclear receptor used as a functional dormancy marker in metastatic mouse 81 models and a prognostic biomarker of dormant DTCs in the BM of BC patients [48-50]. Indeed, our 82 side-by-side comparative approach revealed that nuclear NR2F1+ MCF7 cells were enriched in 83 eVN, underscoring its utility as a platform to study cell-extrinsic regulation of BC dormancy in bone. 84 85

86 Engineered osteoblastic niche promotes MCF7 cell proliferation 87 To assess the effects of an osteogenic niche on BC cell quiescence and proliferation, hBM-MSCs 88 were seeded on hydroxyapatite scaffolds and expanded for one week, followed by three weeks of 89 osteogenic differentiation to engineer an osteoblastic niche (eON) [41, 43, 44] . MCF7 -90 tdTomato/mVenus-p27K- cells were then seeded and cultured for two weeks under perfusion in the 91 eON or in the niche-free scaffold (NFS) as control (Fig. 1A) [46, 47]. The total number of MCF7-92 tdTomato/mVenus-p27K- cells increased after two weeks of culture in the eON compared to NFS 93 (Fig. 1B). Immunofluorescence (IF) staining for tdTomato and mVenus further revealed a reduced 94 fraction of quiescent (mVenus+) MCF7 cells in eON (Fig. 1C). Consistently, Ki67 staining showed 95 a higher proportion of proliferating MCF7 cells in eON compared to the control, highlighting that 96 eON enhances MCF7 cell proliferation (Fig. 1D). The eON was previously validated by matrix 97 deposition of collagen type I alpha 1 (COL1A1) and osteocalcin (OCN) [44, 45]. 98 COL1A1 and OCN staining confirmed osteogenic differentiation in eON, whereas NFS lacked 99 detectable levels of these markers (SI Appendix, Fig. S1). These experiments validated the 100 feasibility of culturing MCF7 cells in an engineered niche as a prerequisite to evaluate how bone 101 niche compositions influence BC cell cycle states. 102 eON, eVN, and eVON model distinct osteoblastic and vascular features of bone 103 Although prior studies have suggested that the perivascular niche of the bone may contribute to 104 BC dormancy, the specific influence of vascular and perivascular cues on BC cell cycle state, 105 particularly in the presence of osteogenic elements, has not yet been addressed [26, 27, 51]. To 106 address this, we engineered two additional environments: a vascularized niche (eVN) and a 107 vascularized osteogenic niche (eVON). eVN was generated by seeding hAT -SVF cells into 108 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.18.695089doi: bioRxiv preprint 4 hydroxyapatite scaffolds and culturing them for two weeks in medium supplemented with fibroblast 109 growth factor 2 (FGF2). For eVON, hAT-SVF cells were seeded into a pre -osteogenic niche and 110 cultured in osteogenic medium supplemented with FGF2 for two weeks [42]. Perfusion bioreactors 111 ensured uniform distribution of MCF7-tdTomato/mVenus-p27K- cells upon subsequent seeding. 112 MCF7 cells were then introduced into each niche configuration and co-cultured for an additional 113 two weeks (Fig. 2A). 114 To determine whether the engineered niches recapitulate key features of bone 115 microenvironments, we assessed osteogenic matrix deposition (i.e., COL1A1 and OCN ) and 116 endothelial (i.e., CD31) network formation. The osteogenic compartment of the three niches was 117 characterized by IF staining and quantified using an optimized image analysis pipeline (SI Appendix 118 Fig. S2A, S2B). The highest levels of COL1A1 deposition were observed in eVON (57.8 ± 12.4%), 119 followed by eON (35.0 ± 13.3%), and lowest in eVN (6.9 ± 1.9%) (Fig. 2B, 2C), likely due to the 120 higher stromal cell content in eVON. OCN deposition was low in eVN (2.6 ± 0.8%), but higher in 121 both eON (14.1 ± 6.4%) and eVON (16.0 ± 9.2%) (Fig. 2D, 2E). These results confirm that eON and 122 eVON retain osteogenic matrix features compared to eVN. 123 Staining for CD31 revealed the self-organized, branched endothelial networks in both eVN 124 and eVON after two weeks of co -culture with MCF7 -tdTomato/mVenus-p27K- cells (Fig. 2F). 125 Notably, these networks formed and persisted without exogeneous angiogenic factor 126 supplementation, indicating stable endothelial organization. eON lacked CD31+ endothelial 127 structures, consistent with the avascular nature of mineralized osteoblastic niches (Fig. 2F). Using 128 the pericyte marker neuron-glial antigen 2 -positive (NG2+; also known as chondroitin sulfate 129 proteoglycan 4, CSPG4) , we measured proportion of perivascular cells associated with CD31+ 130 structures using an optimized image analysis pipeline ( SI Appendix Fig. S2C). CD31+ network 131 density was comparable between eVN and eVON (Fig. 2G). However, NG2+ cells more frequently 132 colocalized with CD31+ networks in eVN, reflecting a more developed perivascular architecture 133 (Fig. 2F, 2H). These findings indicate that endothelial network -forming capacity is maintained in 134 both niches, while perivascular organization is less extensive in eVON. 135 Nuclear NR2F1+ MCF7 cells are enriched in eVN 136 Next, we assessed the effect of distinct cellular compositions provided by eON, eVN, and eVON 137 on MCF7-tdTomato/mVenus-p27K- cell quiescence and proliferation. To this end, we first quantified 138 the number of MCF7 cells retrieved from each niche after two weeks of co-culture. Flow cytometry 139 analysis revealed the highest number of MCF7 cells in eVON, followed by eON, with eVN yielding 140 the lowest cell numbers. These findings suggest that combined osteoblastic and vascular 141 components correlate with enhanced MCF7 cell count (Fig. 3A). 142 IF staining showed that mVenus -p27K- and Ki67 expression were mutually exclusive, 143 confirming the specificity of the quiescence reporter ( mVenus-p27K-) (Fig. 3B). Both quiescent 144 (tdTomato+/mVenus+/Ki67-) and proliferating (tdTomato+/mVenus -/Ki67+) cancer cells were 145 detected across all niches. Quiescent cancer cells were also found as solitary cells or as clusters 146 with proliferative cells in all niches (Fig. 3B, SI Appendix, Fig. S3). Quiescent and proliferating 147 cancer cells were quantified using an optimized image analysis pipeline ( SI Appendix, Fig. S4). 148 eVN had the highest proportion of quiescent MCF7 cells relative to eON and eVON (Fig. 3C). 149 Coherently, it had the lowest fraction Ki67+ MCF7 cells (Fig. 3D). 150 Given the higher percentage of quiescent MCF7-tdTomato/mVenus-p27K- cells in eVN, we 151 asked whether this phenotype was associated with a dormancy -linked transcriptional program. 152 Prior studies have shown that microenvironmental cues in the bone, including cytokine-mediated 153 activation of p38 mitogen-activated protein kinase (MAPK) signaling, can induce dormancy through 154 upregulation of specific transcriptional regulators such as NR2F1 [48-50, 52]. Notably, we found 155 nuclear colocalization of mVenus-p27K- and NR2F1 expression (SI Appendix, Fig. S5) and more 156 frequently within the eVN compared to eON and eVON (Fig. 3E, F), suggesting that eVN promotes 157 dormancy phenotype. Furthermore, the higher abundance of NG2+/CD31+ endothelial network in 158 eVN correlated with increased frequencies of nuclear NR2F1+ and quiescent MCF7 cells. 159 Fulvestrant-induced MCF7 cell number reduction is confined to eVN and eVON 160 To test whether distinct niche compositions influence therapeutic response, we treated engineered 161 co-cultures with fulvestrant for one week. Qualitative assessment of ER α immunohistochemistry 162 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.18.695089doi: bioRxiv preprint 5 (IHC) revealed a pronounced loss in nuclear ERα staining in MCF7 cells across all three conditions 163 (Fig. 4A). Consistent with this, the percentage of n uclear ERα+ MCF7 -tdTomato/mVenus-p27K- 164 cells w as markedly reduced across all three niches, confirming target engagement (Fig. 4B). 165 Notably, MCF7 cell numbers significantly decreased in eVN and eVON upon treatment, but 166 remained unchanged in eON (Fig. 4C), indicating that fulvestrant-induced reduction in cell number 167 was confined to the niches incorporating endothelial networks. 168 169

170 In this study, we developed a fully human, modular 3D bone-mimetic model to investigate cancer-171 stroma interactions during BC bone metastatic progression. This platform allows the stepwise 172

of microenvironmental complexity and supports direct comparison of different cellular 173 components in a modular fashion under controlled conditions. Importantly, our model allows cancer 174 cells to be seeded after bone -mimetic niche formation thanks to perfusion flow, thereby better 175 mimicking the temporal sequence of spontaneous cancer cell dissemination. This design enables 176 spatial analysis of niche -specific effects on cancer cell cycle states, including quiescence and 177 proliferation, in a physiologically relevant and experimentally tractable context. 178 Our findings that the eON promotes MCF7-tdTomato/mVenus-p27K- cell proliferation are 179 consistent with previous in vivo studies [24] showing that osteoblast-rich microenvironments 180 enhance early colonization and proliferation of BC cells in the bone [23, 24]. This underscores the 181 often-overlooked contribution of osteoblasts to the early, non -osteolytic stages of metastatic 182 progression, and highlights the importance of modeling these niches independently of the 183 osteoclast-driven vicious cycle. 184 The eVN and eVON enabled us to investigate the influence of endothelial and perivascular 185 elements on cancer cell quiescence in the absence or presence of an osteoblastic niche. eVN 186 yielded the lowest total number of MCF7 cells yet contained a higher fraction of quiescent cells 187 than eON and eVO N. In contrast, eVON supported the highest MCF7 cell numbers , while the 188 quiescent fraction was reduced relative to eVN. This pattern suggests that osteogenic cues may 189 modulate the impact of vascular elements on BC cell quiescence, potentially by interfering with 190 perivascular organization [10, 25, 28]. 191 Prior studies have shown that specialized capillaries in the BM, such as type H vessels 192 coordinate osteogenesis through endothelial Notch signaling [28, 29]. Disruption of these vascular-193 osteogenic interactions may alter the microenvironmental signaling, potentially affecting DTC 194 outgrowth. Importantly, the NG2+/CD31+ networks resemble the dormancy-supportive 195 vasculature, where DTCs adjacent to stable microvessels had elevated p27 expression and 196 remained quiescent, in contrast to reactivation near sprouting vessels [27]. Perivascular cells 197 expressing NG2 were more frequently associated with CD31+ endothelial networks in eVN than in 198 eVON, suggesting a more stable engineered vasculature in the absence of osteogenic 199 differentiation. Whether this contributes to the higher proportion of quiescent cells in eVN compared 200 to eVON remains to be addressed. 201 In line with the enrichment of the quiescent cell fraction in eVN, we also observed a 202 significantly higher proportion of nuclear NR2F1+ cells in this niche compared to eON and eVON. 203 BM-derived soluble factors, including TGF -β2 and BMP7, particularly enriched in perivascular 204 regions, can activate the p38 MAPK pathway and upregulate dormancy-associated genes such as 205 NR2F1 [48, 49, 52 -54]. This association suggests that structural features of the eVN niche, 206 including enriched NG2+/CD31+ networks, may potentially contribute to the activation of 207 dormancy-related signaling programs. Together, these results validate the pathophysiological 208 relevance of the eVN and underscore its utility for modeling dormancy -permissive 209 microenvironments. 210 A striking observation was the presence of quiescent cells both as solitary cells and within 211 clusters that also contained proliferative cells. This spatial distribution mirrors features of both 212 cellular dormancy and tumor mass dormancy [55]. While tumor mass dormancy is defined by a 213 balance between proliferation and apoptosis that limits net growth, we did not assess apoptosis in 214 this study. Thus, we cannot determine whether a dormancy equilibrium exists in the observed 215 clusters. Future studies incorporating apoptotic markers (e.g., cleaved PARP, Annexin V/PI , or 216 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.18.695089doi: bioRxiv preprint 6 TUNEL) and longitudinal live -cell imaging will be necessary to distinguish between cellular 217 dormancy and tumor mass dormancy. 218 Beyond cell cycle regulation, our system also demonstrates utility for therapeutic testing. 219 Fulvestrant treatment reduced BC cell numbers in eVN and eVON, compared to eON, despite the 220 confirmed nuclear ERα degradation . One potential explanation for the limited response in eON 221 could be fulvestrant-induced changes in ECM, as previous studies have shown that collagen-rich 222 microenvironments can reduce the efficacy of ER -targeted therapies [56, 57] . These data 223 underscore the potential of our platform in dissecting microenvironment -dependent treatment 224 responses and in refining specific therapeutic strategies. 225 There are limitations specifically regarding the further validation of the dormancy model. 226 Although our data support the presence of NR2F1-associated dormancy in eVN, several important 227 criteria remain unmet for full dormancy validation [58]. Most critically, dormancy is defined not only 228 by G0 cell cycle arrest, but also by its reversibility [59]. Although we observed enrichment of 229 mVenus-p27K-+ and Ki67- proportion of MCF7 cells, additional assays are needed to demonstrate 230 the capacity of these cells to re-enter the cell cycle upon niche alteration or exogenous stimulation. 231 Furthermore, the possibility that some quiescent cells are senescent rather than dormant 232 populations must be considered, and should be addressed through assessment of senescence 233 markers such as senescence associated β-galactosidase (SA-β-Gal) or p16INK4a [55, 60]. 234 Another critical gap is the lack of functional perturbation studies to establish causality 235 between niche components and dormancy induction. Although our data suggest that stable 236 perivascular architecture in eVN is associated with NR2F1+ quiescence, functional validation such 237 as selective depletion of endothelial or perivascular cell populations, or genetic manipulation of 238 dormancy-related signaling pathways will be essential to define causality. 239 Despite these limitations, our system offers a modular platform that recapitulates key 240 phenotypic features of the human bone metastatic microenvironment. It enables direct comparison 241 of distinct niche types and allows for phenotypic interrogation of cancer cell states in a 242 pathophysiologically relevant context. Given its human origin and architectural fidelity, this platform 243 holds promise for in-depth studies on the cellular and molecular determinants of tumor dormancy 244 and for drug testing applications. In particular, it could be used to assess the efficacy of agents 245 targeting dormant cells or preventing metastatic outgrowth in niche -specific contexts, offering a 246 valuable preclinical model for therapeutic development. 247 248

249 hBM-MSC isolation and culture 250 Human bone marrow derived mesenchymal stromal cells (hBM-MSCs) were isolated as previously 251 described [41, 44, 45]. Cells were cultured in complete medium (CM) composed of a α-minimum 252 essential medium (αMEM) (Gibco; cat# 22571 -020), supplemented with 10% fetal bovine serum 253 (FBS, Invitrogen; cat# 548-62-9), 1% HEPES (1M, Gibco; cat# 15630 -056), 1% sodium pyruvate 254 (100 mM), 1% GlutaMAX (100X, Gibco; cat# 35050 -061), and 1% Penicillin –Streptomycin (PS, 255 Gibco; cat# 15140 -122). Nucleated cells were plated at a density of 5 x 103 cells/cm2 and 256 maintained at 37°C in a water-jacketed incubator with 5% CO2. CM was supplemented with 5 ng/ml 257 of fibroblast growth factor -2 (FGF -2). Medium was changed twice weekly. hBM -MSCs were 258 selected on adherence and proliferation after one week. 259 hAT-SVF cell isolation and culture 260 Adipose tissue was obtained from three healthy, female donors via liposuction or excision at the 261 University Hospital Basel. Isolation of human adipose tissue derived stromal vascular fraction (hAT-262 SVF) cells was conducted by enzymatic digestion with collagenase type II (Worthington; cat# 263 LS004176), and followed by several centrifugation and purification steps as described [42, 43, 45]. 264 Cell counting was performed using Acridine Orange/Propidium Iodide Stain (Logos biosystems; 265 cat# F23001) at 1:100 dilution in the cell suspension using Luna -FX7™ Automated Cell Counter 266 (Logos biosystems). 267 Ethics statement 268 Human BM aspirates and adipose tissue were collected with informed consent from healthy donors 269 at the University Hospital Basel. The study was approved by the local ethics committee 270 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.18.695089doi: bioRxiv preprint 7 (Ethikkommission Nordwest- und Zentralschweiz, ref. 78/07) and conducted in accordance with EU 271 ethical guidelines. 272 Cancer cell culture 273 MCF7 cells (ATCC) were cultured in Dulbecco’s Modified Eagle’s Medium (DMEM) high glucose 274 (Sigma; cat# D6429), supplemented with 10% FBS, 1% PS, and 1µg/mL insulin at 37°C with 5% 275 CO2. Cell line identity was confirmed by short tandem repeat (STR) sequencing, and routinely 276 tested for mycoplasma contamination. 277 Generation of lentivirus and transduced MCF7 cells 278 Lentiviral particles were generated by co -transfecting HEK293T cells with 2µg each of VSVG 279 envelope plasmid, 2 µg each of third-generation packaging plasmids, and 4µg plasmid DN (pFU-280 Luc2-tdTomato or pCDH-EF1-mVenus-p27K−), using FuGENE HD (Promega) at a 3:1 (reagent:μg 281 DNA) ratio. Viral supernatants were collected at 48- and 72-hours post-transfection, pooled, filtered, 282 and concentrated using Lenti-X Concentrator (Takara). MCF7 cells were sequentially transduced; 283 first with pFU-Luc2-tdTomato lentivirus and Fluorescence-activated Cell Sorting (FACS)-sorted for 284 tdTomato+ expression (BD Aria), followed by transduction with pCDH -EF1-mVenus-p27K− and 285 puromycin selection 0.75 μg/mL. 286 Generation of engineered niches 287 eON was generated by using hBM-MSC as previously described [41, 44]. Briefly, 0.75x106 hBM-288 MSCs were seeded into hydroxyapatite scaffolds (Engipore®, Finceramica-Faenza; 4 mm x 8 mm) 289 embedded in perfusion bioreactors. Cells were perfused at 3 mL/min superficial velocity for 24 290 hours (seeding phase), followed by 0.3 mL/min (culture phase). Constructs were cultured in 291 proliferative medium (PM) consisting of CM supplemented with 100 nM dexamethasone (Sigma; 292 cat# D4902), 0.1 mM ascorbic acid -2-phosphate (Sigma; cat# A92902) and 5 ng/mL FGF -2, 293 followed by three weeks in osteogenic medium (OM) consisting in CM supplemented with 10 nM 294 dexamethasone, 10 mM β-glycerophosphate, and 0.1 mM ascorbic acid-2-phosphate. The medium 295 was changed twice weekly. 296 eVN was generated by seeding 1x10 5 hAT-SVF cells into the hydroxyapatite scaffolds 297 embedded in perfusion bioreactors. Cells were exposed to 3 mL/min superficial velocity over 24 298 hours. Superficial velocity was reduced to 0.3 mL/min after 24 hours of cell seeding phase. Cells 299 were then cultured for two weeks in CM supplemented with 5 ng/mL FGF2. The culture medium 300 was changed twice per week. 301 eVON was generated by seeding hAT-SVF cells at a ratio of 1:1 into pre-engineered eON 302 after one week of osteogenic differentiation [42, 45]. After 24 hours of cell seeding phase, the 303 superficial velocity was reduced to 0.3 mL/min for perfusion co-culture for additional two weeks in 304 OM supplemented with 5 ng/mL FGF2. The culture medium was changed twice per week. 305 Co-culture with MCF7 cells and fulvestrant treatment 306 MCF7-tdTomato/mVenus-p27K- cells were seeded at 1:10 ratio onto eON, eVN, and eVON 307 constructs under the perfusion bioreactor at 3 mL/min for 24 hours. Cells were co-cultured with the 308 respective niches in CM at 0.3 mL/mmin for two weeks. Medium was refreshed twice weekly. 309 Fulvestrant (100 nM) was added onto the respective niches after a week of co -culture, and 310 refreshed twice weekly over a week. 311 Cell isolation from the engineered niches (eON, eVN, and eVON) 312 The constructs were washed with phosphate-buffered saline (PBS) (Gibco; cat# 20012-027), and 313 perfused at 3 mL/min superficial velocity with 0.3% collagenase type II (Worthington; cat# 314 LS004176) in PBS for an hour at 37°C. Supernatants were collected and filtered through a 100 μm 315 nylon mesh strainer (Corning; cat# 352360) and centrifuged at 500g for three minutes. Cells were 316 then resuspended in FACS buffer (2% FBS (Gibco; cat# 10500 -064), 2 mM 317 ethylenediaminetetraacetic acid (EDTA) (Sigma; cat# E7889), PBS). Additional digestion with 318 0.05% trypsin-EDTA (Gibco; cat# 25300-054) under perfusion at 3 mL/min was performed for six 319 minutes at 37°C. All fractions were pooled and, centrifuged at 500g for three minutes, and 320 resuspended in FACS buffer. 321 Flow cytometry 322 Phenotypic analysis of total MCF7 cells and mVenus-p27K-+ subpopulations was performed using 323 CytoFLEX flow cytometer (Beckman Coulter) and BD FACSAria cell sorter (BD Biosciences). Cell 324 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.18.695089doi: bioRxiv preprint 8 suspensions were collected from the niche -free scaffold and the respective engineered niches. 325 They were filtered through 30-µm mesh cell strainer tubes to obtain single cells. Debris exclusion 326 and doublet discrimination were achieved by gating based on forward and side-scatter profiles and 327 pulse width, respectively. Dead cells were identified and excluded using 4′,6 -diamidino-2-328 phenylindole (DAPI) staining (DAPI+ cells). Cancer cells were determined by the positive 329 expression of tdTomato. Single-stained and fluorescence minus one (FMO) controls were used to 330 set gating parameters. 331 Immunofluorescence staining 332 Medium was removed from the bioreactors, and the scaffolds with engineered niches were taken 333 out of the bioreactors using tweezers. Scaffolds were rinsed with PBS, and fixed in 4% 334 paraformaldehyde (PFA, Thermoscientific; cat# 28908) for 24 hours at 4°C. After the fixation, 335 scaffolds were rinsed in PBS, and cut in half using a scalpel. One half of the scaffolds were used 336 for whole -mount staining, and the other hal f were decalcified in 15% EDTA (0.5 M, pH: 7.2), 337 renewed every other day for a week at 37°C with agitation. They were then embedded in paraffin, 338 and sections were cut in 6µm thickness using Microtome (Thermo Scientific; cat# HM 340E). Tissue 339 sections were hydrated in Ultraclear™ (J. T. Baker; cat# 3905.500PE), and a graded alcohol series. 340 Slides were then subjected to heat -induced epitope retrieval (HIER) (pH: 6, Citrate Buffer, 1x, 341 Quartett, AR-001-0120) for 15 min at 96°C. Sections were permeabilized and blocked with 3% 342 bovine serum albumin (BSA) in PBST (0.2% Triton X -100 (Sigma; cat# 9002 -93-1)) for an hour. 343 Primary antibodies (SI Appendix, Table S1) were diluted in 0.5% BSA in PBST and incubated 344 overnight at 4°C. Secondary antibodies (SI Appendix, Table S2) diluted in 0.5% BSA in PBST was 345 applied for 30 minutes at room temperature. Slides were counterstained for a minute with DAPI 346 (BD Biosciences; cat# 564907) solution. Slides were then mounted with Fluoromount ™ aqueous 347 mounting medium (Sigma; cat# F4680), and images were obtained using Nikon Ti2 microscope 348 (NIS version 5.30.07) equipped with X-Light V3 confocal unit, photometrics Kinetix (29.4mm, back-349 illuminated sCMOS) camera and Apo Plan lambda 20x, NA0.75 objective. Images were stored and 350 figures were prepared in OMERO [61] and image analyses were conducted using QuPath software 351 (v0.5.1) [62]. 352 Whole-mount IF staining 353 Half of the fixed scaffolds were processed for whole -mount IF staining. Samples were 354 permeabilized and blocked in 3% BSA (Sigma; cat# A9647), 0.2% Triton X-100 (Sigma; cat# 9002-355 93-1) in PBS for 6 hours at room temperature, followed by 72 hours of primary antibody incubation 356 at 4°C with agitation. Antibodies were diluted in 0.5% BSA, 0.2% Triton X-100 in PBS and incubated 357 overnight at 4°C. The following human primary antibodies were used: Anti -CD31 (Abcam; cat# 358 ab9498), anti-NG2 (Abcam; cat# ab255811). After serial washes with PBST (0.2% Triton X-100 in 359 PBS) and PBS, samples were incubated with secondary antibodies (Invitrogen) overnight at 4°C, 360 washed again, and counterstained with DAPI. Samples were stored in PBS at 4°C until imaging. 361 Confocal imaging was performed on a Nikon X-Light V3 spinning disk microscope using 0.9 µm z-362 step across ~150 µm total depth (~165 optical sections per scaffold). Image stacks were converted 363 into maximum intensity projections using NIS-Elements software and further processed in OMERO 364 software. 365 Immunohistochemistry (IHC) staining of ERα 366 Formalin-fixed paraffin-embedded (FFPE) sections (6 µm) were stained for ERα (Thermo Scientific; 367 cat# MA5-14501) using the Ventana Discovery Ultra (RocheDiagnostics) automated slide strainer. 368 Briefly, tissue sections were deparaffinized and rehydrated, and were subjected to HIER, followed 369 by incubation with the primary antibody (manually applied; 1 hour at 37°C). After washing, sections 370 were incubated with the secondary antibody for 1 hour at 37°C. Detection was performed using the 371 Ventana DISCOVERY ChromoMAP 3,3’ -Diaminobenzidine (DAB) ( Ventana, cat# 760 -159) 372 detection kit. Slides were then counterstained with hematoxylin II, followed by a bluing reagent 373 (Ventana; cat# 790-2208, cat# 760 -2037). The sections were dehydrated, cleared and mounted 374 with permanent mounting medium. Slides were digitized using a Hamamatsu NanoZoomer S60 375 slide scanner equipped with a 40x objective (NA 0.95). Nuclear ER α+ MCF7 cell abundance 376 (DAB+) was quantified by manual evaluation. 377 378 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.18.695089doi: bioRxiv preprint 9 Quantification of CD31+ endothelial network and NG2+ cell abundance 379 Maximum intensity projection images were generated using NIS-Elements software, subjected to 380 ball-correction filtering to minimize signal to noise ratio. Files were then converted into 381 pyramidal.ome.tiff format using ImageJ [63] software with Kheops plugin [64], and uploaded into 382 QuPath. Full image annotation w as created for each field of view (FOV). A single ANN pixel 383 classifier to detect CD31+ (GFP channel) endothelial networks across each FOV was trained with 384 examples of positive and negative signal with gaussian filter and sigma 2.0. Once the CD31+ 385 endothelial network annotation was generated, a second pixel classifier was applied to detect 386 NG2+ (Cy5 channel) structures within or adjacent to the CD31+ networks. The ratio of 387 NG2+/CD31+ area per FOV was then calculated based on the workflow. The script containing the 388 workflow was then run for all images, and the ratio of CD31+ endothelial network per FOV, and 389 NG2+/CD31+ area per FOV was calculated based on the obtained measurements. 390 Quantification of COL1A1, OCN deposition on IF staining images 391 IF staining images were converted into pyramidal.ome.tiff format using ImageJ software for 392 compatibility with QuPath (v.0.5.1). Tissue boundaries were detected by the autofluorescence 393 signal captured in the GFP channel (SI Appendix, Fig S2B) using a thresholder with sigma 5 and 394 threshold 150. For COL1A1 quantification, a pixel classifier was created using a manual 395 thresholding method on the Cy5 channel, where COL1A1 was visualized. Classifier resolution was 396 set to full (0.56µm/px), with a Gaussian pre -filter applied and smoothing set to 0. The intensity 397 threshold for COL1A1 and OCN positivity was empirically defined based on the signal from 398 samples. Pixels above this threshold were classified as COL1A1+ ECM or OCN+ ECM, 399 respectively. All classifiers were stored and batch-applied across full-section datasets. 400 Quantification of tdTomato, mVenus, Ki67, and NR2F1 expression 401 IF staining images were first converted into pyramidal.ome.tiff format using ImageJ software, then 402 uploaded into QuPath. Full image annotations were created to define FOVs. Nuclei were 403 segmented using Watershed Cell Detection based on DAPI staining. Next, annotation for MCF7 404 cells (tdTomato+) were generated by creating a single measurement classifier based on the mean 405 tdTomato signal intensity in the cytoplasm of MCF7 cells in the Cy3 channel. Thresholds were 406 empirically defined through visual inspection. Following td Tomato+ cell annotation, cancer cells 407 were segmented based on their nuclear expression of mVenus (GFP channel), Ki67 (Cy5 channel) 408 and NR2F1 (Cy5 channel). Single measurement object classifier was created for each marker by 409 applying channel-specific filters, and manually thresholding mean nuclear signal intensity to classify 410 the positive cells. Cells co -expressing tdTomato with each nuclear marker were classified using 411 composite object classifiers e.g., tdTomato+/Ki67+. Scripts for each marker panel were exported 412 from QuPath and applied across all images. Manual curation was performed to remove staining 413 artifacts and exclude false positives. 414 Statistics 415 Data are presented as means ± standard deviation of the mean, and were analyzed by using 416 GraphPad Prism software (v10.1.1). Unless otherwise stated/indicated, multiple (pairwise) 417 comparisons were performed using Tukey’s test. Statistical significance was determined by one -418 way ANOVA followed by Tukey's multiple comparison test. Statistically significant differences were 419 defined as: *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. 420 421 Acknowledgments 422 We thank past and present members of the Martin and Bentires-Alj laboratories for feedback and 423 discussions. We also thank the Department of Biomedicine (DBM) Flow Cytometry Core Facility, 424 specifically, Morgane Hilpert, Mihaela Barbu-Stevanovic, Jelena Markovic Djuric, Stella Stefanova 425 for assistance with FACS. We are grateful to the DBM Microscopy Core Facility, particularly Loïc 426 Sauteur, for assistance with imaging and image processing. We thank the DBM Histology Core 427 Facility, particularly Diego Calabrese for performing the immunohistochemical staining. We thank 428 T. Oki and T. Kitamura for providing the pMXs-IRES-puro/mVenus-p27K- vector, and A. Bottos and 429 N. E. Hynes for the pFU -Luc2-tdTomato vector. We also thank for past and present members of 430 the Scherberich and Barbero laboratories for their helpful feedback and discussions. We are 431 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.18.695089doi: bioRxiv preprint 10 grateful to Noemi Torriero, Gangyu Zhang, and Benedetta Guagnini for their technical assistance, 432 regular scientific exchange, and support. This work was supported by the European Commission 433 under the Horizon Europea Marie Skłodowska-Curie Actions (MSCA) program (Grant No. 860715; 434 SINERGIA), by the Freiwillige Akademische Gesellschaft Basel (FAG Basel), and by the Stiftung 435 zur Förderung von chirurgischer Forschung und Spitalmanagement. 436 437

438 1. Wang, H., et al., Bone Tropism in Cancer Metastases. Cold Spring Harb Perspect Med, 439 2020. 10(10). 440 2. Suva, L.J., et al., Bone metastasis: mechanisms and therapeutic opportunities. Nat Rev 441 Endocrinol, 2011. 7(4): p. 208-18. 442 3. Nuckhir, M., et al., State of the Art Modelling of the Breast Cancer Metastatic 443 Microenvironment: Where Are We? J Mammary Gland Biol Neoplasia, 2024. 29(1): p. 14. 444 4. Quayle, L., P.D. Ottewell, and I. Holen, Bone Metastasis: Molecular Mechanisms Implicated 445 in Tumour Cell Dormancy in Breast and Prostate Cancer. Curr Cancer Drug Targets, 2015. 446 15(6): p. 469-80. 447 5. Braun, S., et al., A pooled analysis of bone marrow micrometastasis in breast cancer. N 448 Engl J Med, 2005. 353(8): p. 793-802. 449 6. Pantel, K., C. Alix-Panabieres, and S. Riethdorf, Cancer micrometastases. Nat Rev Clin 450 Oncol, 2009. 6(6): p. 339-51. 451 7. Pantel, K., R.H. Brakenhoff, and B. Brandt, Detection, clinical relevance and specific 452 biological properties of disseminating tumour cells. Nat Rev Cancer, 2008. 8(5): p. 329-40. 453 8. Pantel, K. and C. Alix -Panabieres, Bone marrow as a reservoir for disseminated tumor 454 cells: a special source for liquid biopsy in cancer patients. Bonekey Rep, 2014. 3: p. 584. 455 9. Coleman, R.E., et al., Bone metastases. Nat Rev Dis Primers, 2020. 6(1): p. 83. 456 10. Muscarella, A.M., et al., Exploiting bone niches: progression of disseminated tumor cells to 457 metastasis. J Clin Invest, 2021. 131(6). 458 11. Azevedo, A.S., et al., Metastasis of circulating tumor cells: favorable soil or suitable 459 biomechanics, or both? Cell Adh Migr, 2015. 9(5): p. 345-56. 460 12. Haider, M.T., D.J. Smit, and H. Taipaleenmaki, The Endosteal Niche in Breast Cancer Bone 461 Metastasis. Front Oncol, 2020. 10: p. 335. 462 13. Mundy, G.R., Mechanisms of bone metastasis. Cancer, 1997. 80(8 Suppl): p. 1546-56. 463 14. Clines, G.A. and T.A. Guise, Molecular mechanisms and treatment of bone metastasis. 464 Expert Rev Mol Med, 2008. 10: p. e7. 465 15. Haider, M.T. and H. Taipaleenmaki, Targeting the Metastatic Bone Microenvironment by 466 MicroRNAs. Front Endocrinol (Lausanne), 2018. 9: p. 202. 467 16. Mundy, G.R., Metastasis to bone: causes, consequences and therapeutic opportunities. 468 Nat Rev Cancer, 2002. 2(8): p. 584-93. 469 17. Clezardin, P., Mechanisms of action of bisphosphonates in oncology: a scientific concept 470 evolving from antiresorptive to anticancer activities. Bonekey Rep, 2013. 2: p. 267. 471 18. Clezardin, P. and A. Teti, Bone metastasis: pathogenesis and therapeutic implications. Clin 472 Exp Metastasis, 2007. 24(8): p. 599-608. 473 19. Coleman, R., Bone targeted treatments in cancer - The story so far. J Bone Oncol, 2016. 474 5(3): p. 90-92. 475 20. Rucci, N., et al., Inhibition of protein kinase c-Src reduces the incidence of breast cancer 476 metastases and increases survival in mice: implications for therapy. J Pharmacol Exp Ther, 477 2006. 318(1): p. 161-72. 478 21. Le Gall, C., et al., A cathepsin K inhibitor reduces breast cancer induced osteolysis and 479 skeletal tumor burden. Cancer Res, 2007. 67(20): p. 9894-902. 480 22. Zhang, W., et al., Bone Metastasis: Find Your Niche and Fit in. Trends Cancer, 2019. 5(2): 481 p. 95-110. 482 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.18.695089doi: bioRxiv preprint 11 23. Wang, H., et al., The Osteogenic Niche Is a Calcium Reservoir of Bone Micrometastases 483 and Confers Unexpected Therapeutic Vulnerability. Cancer Cell, 2018. 34(5): p. 823-839 484 e7. 485 24. Wang, H., et al., The osteogenic niche promotes early -stage bone colonization of 486 disseminated breast cancer cells. Cancer Cell, 2015. 27(2): p. 193-210. 487 25. Zhang, W., et al., Bone Metastasis Initiation Is Coupled with Bone Remodeling through 488 Osteogenic Differentiation of NG2+ Cells. Cancer Discov, 2023. 13(2): p. 474-495. 489 26. Price, T.T., et al., Dormant breast cancer micrometastases reside in specific bone marrow 490 niches that regulate their transit to and from bone. Sci. Trabsl. Med, 2016. 8(340): p. 491 340ra73. 492 27. Ghajar, C.M., et al., The perivascular niche regulates breast tumour dormancy. Nat Cell 493 Biol, 2013. 15(7): p. 807-17. 494 28. Kusumbe, A.P., S.K. Ramasamy, and R.H. Adams, Coupling of angiogenesis and 495 osteogenesis by a specific vessel subtype in bone. Nature, 2014. 507(7492): p. 323-328. 496 29. Ramasamy, S.K., et al., Endothelial Notch activity promotes angiogenesis and 497 osteogenesis in bone. Nature, 2014. 507(7492): p. 376-380. 498 30. Crisan, M., et al., A perivascular origin for mesenchymal stem cells in multiple human 499 organs. Cell Stem Cell, 2008. 3(3): p. 301-13. 500 31. Sflomos, G., et al., A Preclinical Model for ERalpha-Positive Breast Cancer Points to the 501 Epithelial Microenvironment as Determinant of Luminal Phenotype and Hormone 502 Response. Cancer Cell, 2016. 29(3): p. 407-422. 503 32. Ozdemir, B.C., G. Sflomos, and C. Brisken, The challenges of modeling hormone receptor-504 positive breast cancer in mice. Endocr Relat Cancer, 2018. 25(5): p. R319-R330. 505 33. Sowder, M.E. and R.W. Johnson, Enrichment and detection of bone disseminated tumor 506 cells in models of low tumor burden. Sci Rep, 2018. 8(1): p. 14299. 507 34. Baldassarri, I., et al., An engineered model of metastatic colonization of human bone 508 marrow reveals breast cancer cell remodeling of the hematopoietic niche. Proc Natl Acad 509 Sci U S A, 2024. 121(42): p. e2405257121. 510 35. Taubenberger, A.V., et al., Delineating breast cancer cell interactions with engineered bone 511 microenvironments. J Bone Miner Res, 2013. 28(6): p. 1399-411. 512 36. Taubenberger, A.V., In vitro microenvironments to study breast cancer bone colonisation. 513 Adv Drug Deliv Rev, 2014. 79-80: p. 135-44. 514 37. Cox, R.F., et al., Osteomimicry of mammary adenocarcinoma cells in vitro; increased 515 expression of bone matrix proteins and proliferation within a 3D collagen environment. 516 PLoS One, 2012. 7(7): p. e41679. 517 38. Pathi, S.P., et al., A novel 3 -D mineralized tumor model to study breast cancer bone 518 metastasis. PLoS One, 2010. 5(1): p. e8849. 519 39. Sosnoski, D.M., et al., Dormancy and growth of metastatic breast cancer cells in a bone -520 like microenvironment. Clin Exp Metastasis, 2015. 32(4): p. 335-44. 521 40. Wang, H., et al., Bone-in-culture array as a platform to model early-stage bone metastases 522 and discover anti-metastasis therapies. Nat Commun, 2017. 8: p. 15045. 523 41. Sadr, N., et al., Enhancing the biological performance of synthetic polymeric materials by 524 decoration with engineered, decellularized extracellular matrix. Biomaterials, 2012. 33(20): 525 p. 5085-93. 526 42. Born, G., et al., Engineering of fully humanized and vascularized 3D bone marrow niches 527 sustaining undifferentiated human cord blood hematopoietic stem and progenitor cells. J 528 Tissue Eng, 2021. 12: p. 20417314211044855. 529 43. Papadimitropoulos, A., et al., A 3D in vitro bone organ model using human progenitor cells. 530 Eur Cell Mater, 2011. 21: p. 445-58; discussion 458. 531 44. Bourgine, P.E., et al., In vitro biomimetic engineering of a human hematopoietic niche with 532 functional properties. Proc Natl Acad Sci U S A, 2018. 115(25): p. E5688-E5695. 533 45. Garcia-Garcia, A., et al., Culturing patient-derived malignant hematopoietic stem cells in 534 engineered and fully humanized 3D niches. Proc Natl Acad Sci U S A, 2021. 118(40). 535 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.18.695089doi: bioRxiv preprint 12 46. Oki, T., et al., A novel cell -cycle-indicator, mVenus-p27K-, identifies quiescent cells and 536 visualizes G0-G1 transition. Sci Rep, 2014. 4: p. 4012. 537 47. Correia, A.L., et al., Hepatic stellate cells suppress NK cell -sustained breast cancer 538 dormancy. Nature, 2021. 594(7864): p. 566-571. 539 48. Bragado, P., et al., TGF-beta2 dictates disseminated tumour cell fate in target organs 540 through TGF-beta-RIII and p38alpha/beta signalling. Nat Cell Biol, 2013. 15(11): p. 1351-541 61. 542 49. Sosa, M.S., et al., NR2F1 controls tumour cell dormancy via SOX9- and RARbeta-driven 543 quiescence programmes. Nat Commun, 2015. 6: p. 6170. 544 50. Borgen, E., et al., NR2F1 stratifies dormant disseminated tumor cells in breast cancer 545 patients. Breast Cancer Res, 2018. 20(1): p. 120. 546 51. Carlson, P., et al., Targeting the perivascular niche sensitizes disseminated tumour cells to 547 chemotherapy. Nat Cell Biol, 2019. 21(2): p. 238-250. 548 52. Sosa, M.S., et al., ERK1/2 and p38alpha/beta signaling in tumor cell quiescence: 549 opportunities to control dormant residual disease. Clin Cancer Res, 2011. 17(18): p. 5850-550 7. 551 53. Kim, R.S., et al., Dormancy signatures and metastasis in estrogen receptor positive and 552 negative breast cancer. PLoS One, 2012. 7(4): p. e35569. 553 54. Nobre, A.R., et al., Bone marrow NG2(+)/Nestin(+) mesenchymal stem cells drive DTC 554 dormancy via TGFbeta2. Nat Cancer, 2021. 2(3): p. 327-339. 555 55. Kabak, E.C., et al., Microenvironmental Regulation of Dormancy in Breast Cancer 556 Metastasis: "An Ally that Changes Allegiances". Adv Exp Med Biol, 2025. 1464: p. 373 -557 395. 558 56. Reyes-Ramos, A.M., et al., Collagen I Fibrous Substrates Modulate the Proliferation and 559 Secretome of Estrogen Receptor -Positive Breast Tumor Cells in a Hormone -Restricted 560 Microenvironment. ACS Biomater Sci Eng, 2021. 7(6): p. 2430-2443. 561 57. Heydari, S., et al., The association between tumor -stromal collagen features and the 562 clinical outcomes of patients with breast cancer: a systematic review. Breast Cancer Res, 563 2025. 27(1): p. 69. 564 58. Montagner, M. and E. Sahai, In vitro Models of Breast Cancer Metastatic Dormancy. Front 565 Cell Dev Biol, 2020. 8: p. 37. 566 59. Phan, T.G. and P.I. Croucher, The dormant cancer cell life cycle. Nat Rev Cancer, 2020. 567 20(7): p. 398-411. 568 60. Truskowski, K., S.R. Amend, and K.J. Pienta, Dormant cancer cells: programmed 569 quiescence, senescence, or both? Cancer Metastasis Rev, 2023. 42(1): p. 37-47. 570 61. Allan, C., et al., OMERO: flexible, model-driven data management for experimental biology. 571 Nat Methods, 2012. 9(3): p. 245-53. 572 62. Bankhead, P., et al., QuPath: Open source software for digital pathology image analysis. 573 Sci Rep, 2017. 7(1): p. 16878. 574 63. Schindelin, J., et al., Fiji: an open -source platform for biological -image analysis. Nat 575 Methods, 2012. 9(7): p. 676-82. 576 64. BIOP. ijp-kheops: Fiji plugin for image quantification 2025 05.09.2025]; Available from: 577 https://github.com/BIOP/ijp-kheops. 578 579 580 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.18.695089doi: bioRxiv preprint 13 Figures and Tables 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 Figure 1. Engineered osteoblastic niche (eON) promotes MCF7 cell proliferation. (A) 606 Experimental scheme to generate eON co -cultured with MCF7-tdTomato/mVenus-p27K⁻ for two 607 weeks in perfusion bioreactors. (B) Flow cytometry quantification of MCF7 cells after two weeks of 608 culture in niche-free scaffold (NFS) or eON. (C, D) Percentage of (C) quiescent (mVenus-p27K⁻) 609 MCF7 cells, and (D) proliferating (Ki67+) MCF7 cells relative to total tdTomato+ MCF7 cells per 610 field of view (FOV) in NFS or eON. Data presented as mean ± SD, n=3. Statistical significance was 611 determined by two-tailed, unpaired, parametric t test. *p<0.05, **p<0.01. (C, D) 5 FOV analyzed 612 per sample. 613 614 NFSeON 0 1×105 2×105 3×105 4×105 5×105 Total number of MCF7 cells ✱✱ B C D NFSeON 0 20 40 60 80 mVenus-p27K+cells(%ofMCF7cells)mVenus-p27K-+ cells (% of tdTomato+ MCF7 cells) ✱ NFSeON 0 20 40 60 Ki67+ MCF7 cells (% of MCF7 cells per FOV) Ki67+MCF7cells (% of MCF7 cells) ✱✱ A Quiescent MCF7MCF7Luc2 tdT omatoUbi EF1mVenus p27K- eON Osteogenic differentiationExpansion 1w 3w hBM-MSCs Ceramic scaffold 2w 4w NFS .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.18.695089doi: bioRxiv preprint 14 615 Figure 2. eON, eVN and eVON recapitulate osteoblastic, vascularized, and vascularized 616 osteoblastic bone microenvironments, respectively . (A) Experimental scheme illustrating 617 generation of eVN and eVON co-cultured with MCF7-tdTomato/mVenus-p27K⁻ cells for two weeks. 618 (B) Representative IF staining and (C) quantification of COL1A1 deposition (percentage per area) 619 across the niches. (D) Representative IF staining and (E) quantification of OCN deposition 620 (percentage per area) across the niches. Magnification: 20x; scale bar: 50 µm. (F) Representative 621 whole-mount immunofluorescence (IF) staining of CD31 (magenta) and NG2 (cyan) in eON, eVN, 622 and eVON. Magnification: 20x; scale bar: 50 µm. (G, H) Quantification of (G) CD31+ endothelial 623 network density (percentage per FOV) and (H) NG2+ perivascular cell density on CD31+ 624 endothelial network (H) (percentage per FOV) in eON, eVN, and eVON. (C, E, G, H) Data presented 625 as mean ± SD, n=3 -5 (two independent experiments). (G, H) 5 FOV analyzed per sample. 626 Statistical significance was determined by one -way ANOVA followed by Tukey's multiple 627 comparison test. *p<0.05, **p<0.01. Multiple comparisons were performed exclusively between 628 eVN and eVON for vascular analysis. 629 630 NG2CD31 DAPI MERGE eVON eVN eON F eONeVNeVON 0 5 10 15 20 NG2+/CD31+ endothelial network (% per FOV) ✱✱ eONeVNeVON 0 5 10 15 CD31+ endothelial network (% per FOV) HG COL1A1 DAPI MERGE eVNeVON eON B DAPIOCN MERGE eONeVNeVON DC eONeVNeVON 0 20 40 60 80 COL1A1 deposition(% per area) ✱ ✱✱ E eONeVNeVON 0 10 20 30 40 OCN deposition(% per area) ✱ ✱ eVON 2w1w 1w hBM-MSCs Differentiation DifferentiationExpansion 2w2w eVN hAT-SVF cells Ceramic scaffold Vasculogenic induction Vasculogenic induction MCF7 cells 2w A .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.18.695089doi: bioRxiv preprint 15 631 Figure 3. eVN enriches for quiescent, non -proliferative MCF7 cells with elevated nuclear 632 NR2F1. (A) Flow cytometry quantification of MCF7-tdTomato/mVenus-p27K- cells after two weeks 633 of co -culture with eON, eVN and eVON. (B) Representative IF staining showing quiescent 634 (mVenus-p27K⁻+; green) MCF7 cells appearing as solitary and clustered with proliferating (Ki67+; 635 cyan) MCF7 cells (tdTomato+; red) in eON, eVN and eVON after two weeks of co -culture. (C, D) 636 Percentage of (C) quiescent (mVenus-p27K-+) MCF7 cells, and (D) proliferating (Ki67+) MCF7 637 cells relative to total MCF7 cells per FOV. (E) Representative IF staining of NR2F1 (cyan) and for 638 tdTomato (red) after two weeks of co -culture in eON, eVN, and eVON. Magnification: 20x; scale 639 bar: 50 µm. (F) Percentage of nuclear NR2F1+ MCF7 cells relative to total MCF7 cells per FOV in 640 the respective niches. (A, C, D, F) Data presented as mean ± SD, n=3 (three independent 641 experiments; each data point represents the mean of three technical replicates per donor). 642 Statistical significance was determined by one-way ANOVA with Tukey’s multiple comparison test. 643 *p<0.05, **p<0.01, ***p<0.001. (C, D, F) 5 FOV analyzed per sample. 644 645 tdTomato DAPImVenus-p27K⁻ Ki67 MERGE eONeVNeVON B DAPItdTomato MERGENR2F1 eONeVNeVON E eONeVNeVON 0 20 40 60 80 Nuclear NR2F1+ MCF7 cells (% of MCF7 cells per FOV) ✱✱ ✱✱ F eONeVNeVON0102030Ki67+ MCF7 cells(% of MCF7 cells per FOV) ✱✱✱✱ D eONeVNeVON 0 20 40 60 80 mVenus-p27K-+ MCF7 cells (% of MCF7 cells per FOV) ✱ ✱C eONeVNeVON 0 5×105 1×106 2×106 Total number of MCF7 cells ✱ ✱✱ ✱✱✱A .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.18.695089doi: bioRxiv preprint 16 646 Figure 4. Fulvestrant reduces MCF7 cell numbers only in vascularized niches, despite 647 comparable ERα degradation across all conditions. (A) Representative IHC staining images of 648 ERα and (B) percentage of nuclear ERα+ MCF7 -tdTomato/mVenus-p27K- cells relative to total 649 MCF7 cells per FOV upon one week of treatment in eON, eVN and eVON. (C) Flow cytometry 650 quantification of MCF7 cells after one week of treatment in eON, eVN and eVON. Data presented 651 as mean ± SD, n=3-4 (two independent experiments). Magnification: 40x; scale bar: 50 µm. (A, B) 652 Statistical significance was determined by two-way ANOVA with Tukey’s multiple comparison test. 653 *p<0.05, ****p<0.0001. 654 655 FulvestrantVehicle eONeVNeVON A C eONeVNeVON02×1054×1056×1058×1051×106Total number of MCF7 cells ✱✱ 20406080Nuclear ERα+MCF7 cells(% of MCF7 cells per FOV) VehicleFulvestrant✱✱✱✱✱✱✱✱✱✱✱✱ B eVNeONeVON020406080Nuclear ERα+MCF7 cells(% of MCF7 cells per FOV) VehicleFulvestrant✱✱✱✱✱✱✱✱✱✱✱✱ eVNeONeVON020406080Nuclear ERα+MCF7 cells(% of MCF7 cells per FOV) VehicleFulvestrant✱✱✱✱✱✱✱✱✱✱✱✱ .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.18.695089doi: bioRxiv preprint