Introduction

Metastases are the direct cause of more than 90% of cancer-related deaths1. Current cancer treatments usually involve surgical removal of primary tumors followed by adjuvant chemotherapies or endocrine therapies. In many cases, tumors relapse in distant organs years or even decades after primary tumor resection, and often, these relapses are accompanied by resistance to the original therapies2. Bone metastases are one of the major causes of morbidity in 65-75% of patients with advanced breast cancer. Albeit less aggressive than other visceral metastases, two-thirds of bone metastases are followed by metastasis to other organs and ultimately lead to patient mortality3,4,5.

Breast cancer bone metastasis is characterized by severe pain, impaired mobility, pathologic fractures, numbness, paralysis, anemia, and hypercalcemia3,6. In its advanced stage, breast cancer bone metastasis drive a vicious cycle involving cancer cells, osteoblasts and osteoclasts, which together establish osteolytic lesions7,8,9. Specifically, cancer cells secrete factors, such as parathyroid hormone-related protein (PTHrP), which act on osteoblasts to modulate the expression of genes including receptor activator of nuclear factor kappa-Β ligand (RANKL) and osteoprotegerin (OPG). The alterations of these factors in turn boost osteoclast formation and maturation, thereby accelerating bone resorption and leading to widespread bone destruction. Many growth factors deposited in the bone matrix are then released, and reciprocally stimulate tumor growth7,8,9.

In addition to the observation of osteolytic morphology in advanced bone metastases of breast cancer, our previous study revealed an intermediate phase before the establishment of overt bone lesions, namely the “pre-osteolytic stage”10. We found that in some breast cancers, especially for ER+ luminal-like, the microscopic lesions were heavily osteoblast-dependent before they developed osteolytic lesions. Such complexity is associated with continuous dynamics during the metastasis progression, which leads to the hypothesis that some intrinsic traits of cancer cells may drive varying metastatic behaviors and alterations of the microenvironment in different breast cancer subtypes. By applying the newly-developed model systems that enable us to develop bone metastases with breast cancer cells of different subtypes in mice11, we compared bone lesions formed by aggressive ER− tumors with indolent ER+ tumors. Our results suggested that, among the models and samples investigated in our study, ER− breast cancer tends to form more osteolytic and hypoxic bone lesions compared to ER+ breast cancer.

Results

ER+ and ER− breast cancer cells respond differently to osteogenic niche and cells

In our previous studies, we investigated the early colonization of breast cancer cells in bone10,11,12. Our lab developed intra-iliac artery (IIA) injection to selectively deliver cancer cells into hind limbs via the external iliac artery10,13. Typically applied intraosseous breast cancer cell delivery does not replicate the vascular route of cancer spreading to bone, whereas intracardiac or tail vein delivery exhibits high pulmonary metastasis and animal death before bone metastases develop. However, our IIA approach enables delivery of breast cancer cells to the ipsilateral femur and tibia and facilitates efficient colonization and metastasis formation in bone10,13. With this approach, we successfully modeled the early stage of bone metastasis, which is a major advantage over the conventional techniques. We discovered that micro-metastases of both ER+ MCF-7 and ER− MDA-MB-231 breast cancer cells tend to localize in an “osteogenic niche” composed of ALP+ osteoblasts and their precursors (e.g., mesenchymal stem cells, MSC), which is consistent with our previous publications10. However, these microscopic lesions exhibited different interactions with cells expressing alkaline phosphatase (ALP+). Specifically, GFP+ MCF-7 cells forming micro-metastases recruited more ALP+ cells to the surrounding regions, regardless of exogenous estradiol (E2) supplementation. In contrast, while GFP+ MDA-MB-231 cells occasionally made contact with ALP+ cells, they did not recruit additional osteoblasts beyond those already lining bone surfaces. Furthermore, some clustered MDA-MB-231 cells were located distantly from the bone-lining osteoblasts and failed to establish direct cell-cell contact with ALP+ cells, in contrast to MCF-7 cells (Fig. 1A).

Fig. 1: Breast cancer cells responded differently to osteogenic niche and cells.
figure 1

A Representative imaging showing osteogenic niches in tumor free bones or tumor-bearing bones with the residency of ER+ and ER− micro-metastases. Cancer cells are tagged by GFP (green); ALP (red) marks the osteogenic cells. Scale bar: 25 µm. For experiments in need of estradiol supply, sustained-released estradiol tubes (7 mg/mouse) were prepared and transplanted to mice subcutaneously before cancer cell injection. B Bar graphs showing the quantification of cancer cells mixed with various ratios of MSCs in 2D co-cultures. Cancer cells and ER status in each experiment are indicated below the bars. Bioluminescence signal intensity is used for the quantification, which is measured on Day 7 under each condition and normalized to that of untreated cancer cells (n = 3 biologically independent samples). Error bars SD. P values are calculated by one-way ANOVA test. C Quantification of MSC-conferred growth advantages in 3D cultures for 7 days. The data was then normalized to the values of control group for each well (n = 3 biologically independent samples). Error bar: SD. D Heatmap of qPCR results showing the expression of osteogenesis-related genes in bone-resident cells of mouse limbs carrying size-matched small, medium, and large pairs of MCF-7 or MDA-MB-231 bone lesions in both estradiol-provided and estradiol-free conditions (n = 3 animals). The values are log2-transformed. P-values are calculated by two-tailed paired t-test. BM: bone metastases. For experiments in need of estradiol supply, sustained-released estradiol tubes (7 mg/mouse) were prepared and transplanted to mice subcutaneously before cancer cell injection.

Full size image

In line with the divergent interplay with the osteogenic niche in vivo, we observed distinct growth dependency on osteogenic cells in different subtypes of breast cancer cells in vitro. When breast cancer cells were mixed with mesenchymal stem cells (MSC, the progenitor of osteogenic lineage) in 2D cocultures, 4 of 4 ER+ cell lines exhibited a growth dependence to MSCs, whereas the growth of all the ER− cells (5 of 5) were suppressed by MSCs (Fig. 1B). The MSC-conferred growth advantage to ER+ cells was further confirmed in 3D cocultures (Fig. 1C).

To determine whether ER+ and ER− cancer cells affect osteogenic cells differently, we performed transcriptional profiling of bone stromal cells derived from bone tissues carrying MCF-7 or MDA-MB-231lesions. Given that MDA-MB-231 proliferates faster than MCF-7 cells, we carefully quantified the tumor burdens and selected three pairs of different size-matched bone tissues (small, medium, and large lesions representing different endpoints) for RNA-seq (Supplementary Fig. 1A-B, see more methodological details in Methods). Interestingly, different subtypes of cancer cells appeared to impact osteogenic niche cells in distinct ways: compared to MDA-MB-231 cells, the MSC/osteoblasts-favored MCF-7 induced more intensive osteogenesis in vivo, which is reflected by a global enrichment of osteogenesis genes (except for Col1a1) in bone samples carrying MCF-7 lesions (Supplementary Fig. 2A).

In both our and others’ studies, estradiol (E2) pellets are often used to promote MCF-7 growth in vivo10,14,15,16. In contrast, to our knowledge, E2 is rarely used in in vivo studies of MDA-MB-231 cells17,18,19,20. Our in vivo experiments have inherited these settings; however, the discrepancy in E2 supply presents a challenge to interpret our results, as estrogen itself is known to affect osteoblasts by stimulating their proliferation and differentiation, meanwhile inhibiting their apoptosis21. To determine whether the altered osteogenesis was due to the direct effect of E2 on the general bone environment, we started with performing RNA sequencing on bone tissues collected from tumor-free mice with and without E2 supplementation. Using CIBERSORT algorism22, we analyzed and compared the relative cell compositions across the major bone cell types, including 4 key subpopulations of the osteogenic lineage: MSC, osteoprogenitor, immature osteoblast, and osteoblast. Although a modest change in the MSC-to-osteoprogenitor ratio was observed, E2 supplementation did not significantly alter the overall composition of osteogenic cells, particularly the enrichment of osteoblasts (Supplementary Fig. 1C). These results suggest that the distinct osteoblast activities are primarily driven by the differential behaviors of cancer cells, rather than E2 supplementation.

To further determine the direct impact of cancer cells on osteogenesis in the tumor microenvironment, we reevaluated the comparative analysis between MCF-7 and MDA-MB-231 lesions under identical settings, both with and without E2 supply. Notably, our IIA injection approach allowed MCF-7 cells to develop bone lesions even in E2-free mice, although the tumor progression was slower compared to those with E2 supply12 (and Supplementary Fig. 1D). Again, three pairs of size-matched bone tissues (small, medium and large lesions) were collected to represent different endpoints. Using real-time PCR (RT-PCR) with mouse-specific primers, we assessed the transcript levels of selected osteogenic marker genes. The observation of relative enrichment of osteogenesis markers in MCF-7-bearing bones, in both E2-supplemented and E2-free scenarios (Fig. 1D), validated our RNA-seq results. Taken together, these results indicate that the bone lesions developed by ER− MDA-MB-231 breast cancer are less osteoblastic than those formed by ER+ MCF-7 tumors.

ER− breast cancer induces more active osteolysis than ER+ breast cancer

Despite the “osteoblastic-like” phenotype in early bone colonization of some breast tumors (especially ER+ tumors), most breast cancer bone metastases ultimately become osteolytic and are driven by a robust interaction between cancer cells and bone-resorbing osteoclasts. To compare the development of osteolysis in ER+ and ER− tumors side-by-side, we performed immunofluorescence (IF) staining for cathepsin K (CTSK) and tartrate-resistant acid phosphatase (TRAP) staining to observe osteoclasts activity in different metastatic stages. CTSK is a marker indicative of differentiated osteoclasts, which is applicable across all stages of bone metastases at the cellular level; TRAP staining presents a method to gauge significant osteoclast activity on a broader scale, which is often employed to visualize the osteolysis phenotype evident in advanced bone metastases. Microscopic lesions (i.e., micrometastases) of ER + MCF-7 tumors did not exhibit evident osteoclast activity (Fig. 2A). By contrast, size-matched micrometastases of ER− MDA-MB-231 tumors were marked by CTSK staining, suggesting the recruitment of differentiated osteoclasts even at a relative early stage of bone colonization. As the metastases progressed to late stage (i.e., macrometastases), the bone lesions of MDA-MB-231 displayed strong TRAP staining and excessive bone loss on day 21, representing fully developed osteolytic tumors. By contrast, the MCF-7 remains pre-osteolytic for an additional 14 days even after the tumor burdens reached a comparable size to the MDA-MB-231 derived lesions (Fig. 2B).

Fig. 2: ER+ and ER− breast cancer cells responded differently to osteoclast niches.
figure 2

A Representative imaging showing osteoclast niches in tumor-free bone or with the residency of ER+ and ER− micro-metastases. Cancer cells are tagged by GFP (green); CTSK (red) marks the osteoclasts. Scale bar: 25 µm. B TRAP staining of tumor-free bone, MCF-7 (day 35 after tumor inoculation), and MDA-MB-231 (day 21 after tumor inoculation) osteolytic bone lesions after IIA injection. C Bar graphs showing the quantification of cancer cells mixed with various ratios of U937s in 2D co-cultures. Cancer cells and ER status in each experiment are indicated below the bars. Bioluminescence signal intensity is used for the quantification, which is measured on Day 7 under each condition and normalized to that of untreated cancer cells (n = 3 biologically independent samples). Error bars: SD. The values are log2-transformed. P-values are calculated by one-way ANOVA test. D Heatmap of qPCR results showing the expression of osteolysis-related genes in bone-resident cells of mouse limbs carrying size-matched small, medium, and large pairs of MCF-7 or MDA-MB-231 bone lesions in both estradiol-provided and estradiol-free conditions (n = 3 animals). For experiments in need of estradiol supply, sustained-released estradiol tubes (7 mg/mouse) were prepared and transplanted to mice subcutaneously before cancer cell injection. P-values are calculated by two-tailed paired t-test. BM: bone metastases. E Transcriptional levels (Z-score) of osteolysis-related genes in ER+ (n = 6 patients) and ER− (n = 10 patients) bone metastatic specimens of breast cancer metastases clinical dataset GSE-14020. Error bar: S.D. P-values are calculated by two-way ANOVA test. Error bars: SD. Centerline: median; box limits: upper and lower quartiles; whiskers: 1.5x interquartile range.

Full size image

In 2D cocultures, the precursor cell line of macrophage and osteoclast, namely U937, promoted growth of more ER− breast cancer (4 of 5, 80%) than ER+ (2 of 4, 50%) cell lines (Fig. 2C). Conversely, when we examined stromal RNA profiles from in vivo bone metastases formed by breast tumors, we found increased expression of numerous osteoclast-differentiation markers23, including RANKL (Tnfsf11), RANK (Tnfrsf11a), c-fms (Csf1r), integrin β3 (Itgb3), cathepsin K (Ctsk) and c-src (Src) in bone carrying ER− MDA-MB-231 tumors, compared to ER+ MCF-7 bone lesions (Supplementary Fig. 2B). To account for the effect of E2 supply, we reassessed the transcription of selected osteoclast marker genes between MCF-7 and MDA-MB-231 lesions under identical conditions, both with and without E2 supply. The RT-PCR results confirmed the trends observed in the original RNA-seq data. (Fig. 2D). Similarly, in clinical dataset GSE1402024, which comprises the microarray data of over 60 breast cancer metastasis samples including 16 bone metastasis specimens with ER status identified, the transcriptional levels of osteoclast markers were globally more enriched in ER− bone lesions compared to ER+ bone metastases (Fig. 2E). Taken together, this data indicates that the bone metastases formed by ER− breast cancer are more osteolytic than ER+ breast cancer.

Bone metastases of ER− breast cancer are characterized by intensive hypoxic stress

The discrepancy between ER+ and ER− breast cancers, as well as their distinct predilection on different bone cells, raise the question of how different subtypes of breast cancer initiate osteolysis and evolve to overt lesions. To illuminate the associated mechanisms, we further analyzed the RNAseq data of in vivo bone lesions formed by MCF-7 and MDA-MB-231. The reads, comprising a mixture of human (cancer cells) and mouse (stromal/niche cells) transcripts, were separated using the Xenome in silico sorting approach25, which allowed us to conduct transcriptomic profiling on xenograft cancer cells and host bone cells respectively (Fig. 3A). Given the limited sample size, we performed single-sample Gene Set Enrichment Analysis (ssGSEA) on stromal contents (i.e., mouse sequence) to identify the distinct ortholog hallmark gene sets in bone milieu favoring the progression of ER− tumor cells. In comparison to MCF-7 lesions, the bone environment of MDA-MB-231 cells up-regulated genes regulating hypoxia (Fig. 3B) including the increased expression of Hif1a, the major regulator in response to low oxygen stress (Fig. 3C). Compared to MCF-7, MDA-MB-231 also induced upregulation of the most top-ranked genes derived from a highly prognostic hypoxia metagene26 in bone cells in the tumor microenvironment, including “Vegfa”, “Slc2a1”, “Pgam1”, “Eno1”, “Ldha”, “Tpi1”, “P4ha1”, “Cdkn3”, ““Tubb6” (Hypoxia signature = Σ 12 top-rank hypoxia genes + HIF1a), indicating an enhanced hypoxic stress in the bone milieus (Fig. 3C and Supplementary Fig. 2C). On the tumor-intrinsic side, we found the expression of HIF1A and hypoxia metagene were also more enriched within cancer cells of MDA-MB-231-formed bone lesions (Fig. 3D and Supplementary Fig. 2D). Intriguingly, the difference in hypoxic response, as indicated by hypoxia signature, appears to be more pronounced in tumor cells than in the bone stromal cells.

Fig. 3: Hypoxia signaling is enriched in bone metastasis induced by ER− breast cancer cells.
figure 3

A The data processing pipeline after RNAseq. Xenome was used to separate the sequences of cancer cells (human) and bone cells (mouse). B The orthology-mapped hallmark gene sets upregulated in MDA-MB-231 bone lesions, compared to MCF-7 lesions, and based on ssGSEA analysis (n = 3 animals for each group). C Hif1a expression and hypoxia-associated gene signature are enriched in stromal content of MDA-MB-231 skeletal lesions. Sustained-released estradiol tubes (7 mg/mouse) were only transplanted to mice with MCF-7 cells (n = 3 animals/group). D HIF1A expression and hypoxia-associated gene signature are enriched in tumor tissues of MDA-MB-231 skeletal lesions (n = 3 animals for each group). Sustained-released estradiol tubes (7 mg/mouse) were only transplanted to mice with MCF-7 cells. E Heatmap of qPCR results showing the expression of representative hypoxia-related genes in bone-resident cells of mouse limbs carrying size-matched small, medium, and large pairs of MCF-7 or MDA-MB-231 bone lesions in both estradiol-provided and estradiol-free conditions (n = 3 animals). For experiments in need of estradiol supply, sustained-released estradiol tubes (7 mg/mouse) were prepared and transplanted to mice subcutaneously before cancer cell injection. The values are log2-transformed. P-values are calculated by two-tailed paired t-test. BM: bone metastases. F Heatmap of qPCR results showing the expression of representative hypoxia-related genes in cancer cells derived from size-matched small, medium, and large pairs of MCF-7 or MDA-MB-231 bone lesions in both estradiol-provided and estradiol-free conditions (n = 3 animals). For experiments in need of estradiol supply, sustained-released estradiol tubes (7 mg/mouse) were prepared and transplanted to mice subcutaneously before cancer cell injection. The values are log2-transformed. P-values are calculated by two-tailed paired t-test. G Evaluation of HIF1A expression in breast cancer bone metastases in clinical dataset GSE-14020. Brain/Liver/Lung/Ovary metastases: n = 19/5/18/7 patients for each group; bone metastasis: n = 16 patients. H HIF1A and hypoxia signature expression are especially increased in bone metastasis of ER− breast cancer. ER+ patient samples: n = 6; ER− patient samples: n = 10. P-values by two-tailed t-test. Error bars: SD. Centerline: median; box limits: upper and lower quartiles; whiskers: 1.5x interquartile range.

Full size image

Considering the previous findings that ER signaling is implicated in the direct regulation of HIF1A and its downstream targets27, it is important to distinguish the differential effects of ER+ and ER− tumors on hypoxia from the alteration induced by E2. Using a strategy similar to that applied to study osteogenesis and osteolysis (Figs. 1D and 2D), we compared the transcription of representative hypoxia-related genes (HIF1A/Hif1a, VEGFA/Vegfa, SLC2A1/Slc2a1 and LDHA/Ldha) between MCF-7 and MDA-MB-231 lesions under identical conditions, both with and without E2 supply. Importantly, we employed both human and murine-specific primers to assess the transcriptional levels of selected genes in both tumor (human) and bone (mouse) cells. The differences in most tested gene transcripts between MCF-7 and MDA-MB-231 bone metastases generally indicate an enhanced hypoxia response in both stromal and cancer cells (Fig. 3E, F). However, the elevation of hypoxic-associated genes in MDA-MB-231-bearing bones without E2 supplementation is less pronounced compared to those with E2, probably due to the high baseline activation of HIF1A in bone cells, particularly in osteoclasts, under estrogen deficiency conditions28,29. Overall, these results further support the notion that bone metastases of ER− breast cancer exhibit elevated activity in response to hypoxia.

Next, we validated our findings in the clinical dataset GSE14020. We found increased HIF1A expression in bone compared to other metastatic sites (Fig. 3G). Of note, the bone specimens of ER− breast cancer displayed higher levels of HIF1A gene and hypoxia signature than those of ER+ breast cancer, suggesting a correlation between skeletal hypoxia level and subtypes of breast tumor (Fig. 3H). Taken together, this data suggests that hypoxia signaling is more upregulated in both cancer cells and bone cells in the bone metastases of ER− breast cancer.

Hypoxia is associated with the osteoclast differentiation and osteolytic activity

Given that the osteolytic activity and hypoxia level changed simultaneously in bone metastases formed by distinct subtypes, we asked if a one-on-one correlation can be established between hypoxia and osteolysis in bone tumors. The association between hypoxia and the production of osteolytic cytokines was supported by the in vivo dataset GSE110451, which compared the bone lesions of MCF-7 with orthotopic tumors established by the same cells. The bone lesions in this dataset showed a positive correlation between the HIF1A level and hypoxia signature as well as the collective expression of osteolytic inducers including “PTHLH (PTHrP)”, “CXCL8 (IL-8)”, “IL11”, “IL6”, “VEGFA”, “MMP1”, “SPP1(OPN)”, “CCN2 (CTGF)” and “CXCL12”30,31,32. Of note, no such correlation was seen in orthotopic tumors, indicating that the positive correlation between osteolysis and HIF1A signaling is more specific to bone metastases (Fig. 4A). Next, we computed a composite score (Osteolysis signature = Σ6 osteoclast-differentiation genes) to reflect the osteoclast activity; and correlated it with the hypoxia signature score. Interestingly, in our own experiments, we observed a strong correlation between the hypoxia signature and osteolysis signature in the stromal cells of the bone lesion of both MCF-7 and MDA-MB-231, which was more intensified in the latter lesions (Fig. 4B).

Fig. 4: Hypoxia correlates to osteolysis in breast cancer bone metastases.
figure 4

A Correlation between hypoxia response genes and osteolysis inducers with HIF1a expression in bone lesions or breast tumors in mouse models. R-values by Pearson test. Bone lesion samples: n = 7; orthotopic tumors: n = 7. B The hypoxia-signature level positively correlates with the osteoclast-signature level in niche cells of lesions. Blue dots: pre-osteolytic lesions, n = 3; orange dots: osteolytic lesions, n = 3. R and P-value by Pearson correlation test. TRAP staining quantification (C) and images (D) of monocytes being collected from bone tissue and treated by conditioned media from MCF-7 or MDA-MB-231 cells after hypoxia culture (n = 4/6/6 biological replicates for individual groups). Scale Bar: 25 µm. The levels of hypoxia-responding genes expressed by MCF-7 (E) and MDA-MB-231 (F) cells after cultured in hypoxia chamber for 24 h (n = 3 biologically independent samples). P values by two-tailed t-test.

Full size image

Based on the observed correlation between osteolytic activity and hypoxia, we hypothesized that hypoxia enhances osteolysis by inducing tumor production of osteolytic cytokines. Aligned with this hypothesis, we found that the bone-derived monocytes treated by conditioned media from MDA-MB-231 hypoxia culture displayed more osteoclast differentiation than those treated with MCF-7 conditioned media, indicating that ER− cells may exhibit a stronger activity of inducing osteogenesis in comparison to ER+ cells in bone metastases (Fig. 4C, D).

Previous studies demonstrated that under low oxygen condition, the expression of bone metastatic genes by cancer cells were associated with osteoclast differentiation and activity30,33. To further dissect this connection in ER+ and ER− cells, we cultured MCF-7 and MDA-MB-231 cancer cells in hypoxia conditions (2% O2). After 24 h, we observed upregulation of various bone metastatic genes, especially secreted factors34, in response to hypoxic stress that promoted osteolysis in both cancer cell types. Of note, the hypoxia-dependent production of osteolytic cytokines is likely more favorable in MDA-MB-231 compared to MCF-7 cells. In MCF-7, while the detected gene levels increase with fold changes from 1 to 2.3 (except for IL6), only the hypoxic enrichment of VEGFA and SLC2A1 showed significant difference (Fig. 4E). In contrast, in MDA-MB-231, the detected genes exhibited a more global elevation in the hypoxia condition, with fold changes from 2 to 7 as well as more significant statistics (Fig. 4F). To this end, our data suggests that in response to hypoxia, ER− cells likely secrete more osteolytic factors than ER+ cells.

The feasibility of suppressing bone metastases via HIF1A inhibitors is assessed in bone-in-culture array (BICA)

In our previous studies, we have established the bone-in-culture array (BICA) as a high-throughput and powerful platform to study bone metastasis progression in ex vivo cultures11,12. BICA provides a bone-like microenvironment and offers opportunities to study bone lesions of multiple types and subtypes of cancer. Advanced BICA lesions can successfully induce spontaneous osteoclastogenesis and bone resorption. These studies provided the justifications for using BICA as a reliable and efficient experimental platform to investigate osteolytic bone metastases.

As we observed the central importance of the hypoxia pathways in facilitating the osteolytic progression in breast cancer bone metastases, we decided to test if intervening in the HIF pathway may yield a growth suppression in the bone environment. Given the inherently hypoxic and low-perfused nature of bone tissue31,35, bone metastases may present a unique context in which the HIF1A pathway plays a critical role. To investigate this, we induced chemical hypoxia in BICA by adopting the well-established method of using cobalt chloride (CoCl2) to mimic hypoxia conditions36,37. This approach enabled us to explore whether HIF1A modulators could ultimately suppress the expansion of bone lesions under hypoxia condition.

Several HIF modulators have progressed to the clinical trial phase for cancer treatment38,39,40,41,42, but have not yet achieved the expected success, likely due to insufficient understanding of where and when hypoxia and the HIF pathway represent the true vulnerability in cancers. Even the intervention strategy remains controversial and often context dependent. While many studies advocate for inhibiting HIF activity, there are reports suggesting that activating HIF, rather than inhibiting it, can exert anti-tumor effects. For instance, Su et al. recently demonstrated that stabilizing and activating HIF1A using the HIF prolyl hydroxylase inhibitor FG-4592 suppresses glioblastoma growth in vivo by inducing ferroptosis43. Similarly, IOX2, another well-characterized HIF prolyl hydroxylases inhibitor, has been shown to suppress proliferation in breast cancer cell lines44. To determine the most effective strategy for treating bone metastases, we conducted a pilot assay on the BICA platform to test both HIF stabilizer and inhibitors. Our assay included FG-4592, IOX2, and 2-Methoxyestradiol (2-MeOE2), a natural HIF1A inhibitor currently under evaluation in phase 1 and 2 clinical trials for advanced solid tumors and multiple myeloma38,39,40. Among these compounds, we identified 2-MeOE2 as exerting potent inhibitory effects on bone lesions of both ER+ (MCF-7) and ER−(HCC-1937) tumors in BICA culture at a relatively low concentration (200 nM). Notably, its inhibitory effect was more pronounced in ER− HCC-1937 bone lesions compared to ER+ MCF-7 bone tumors (51% vs 32%) (Fig. 5A). In contrast, neither FG-4592 nor IOX2 achieved inhibitory effects. In fact, the application of FG-4592 and IOX2 even led to a 62% and 10% increase in MCF-7 growth in BICA, although this was not statistically significant due to intra-group variability. These results suggested that inhibition, rather than activation, is the preferred strategy for treating breast cancer bone metastases.

Fig. 5: The feasibility of reducing osteolytic metastases via HIF1 inhibitors in BICA.
figure 5

A Inhibition of HIF pathway by inhibitor 2-MeOE on MCF-7 and HCC-1937 BICA cultures at the CoCl2 concentration of 50 µM. All the compounds were applied at 200 nM. Treatments lasted for three weeks for BICA cultures. The acquired intensity data of BICA cultures was normalized over the day 1 post-seeding signal intensity of the same well. Inhibition rates were calculated based on the difference between control and treatment groups (see Methods for formula). p-values by Tukey’s multiple comparisons, n = 6 bone fragments for each group. B Comparison of the efficacy of 2-MeOE and CoCl2 between normoxic and hypoxic BICA cultures of SCP28. (n = 11 biologically independent samples as indicated in the dot plot). C Comparison of the efficacy of 2-MeOE and CoCl2 between normoxic and hypoxic 2D cultures of SCP28. (n = 12 biologically independent samples as indicated in the dot plot). CoCl2 and 2-MeOE were applied at 50 µM and 200 nM, respectively. Treatments lasted for three weeks for BICA and two weeks for 2D culture (as 2D signal reached saturation earlier than BICA). The acquired intensity data of both BICA and 2D cultures was normalized over the day 1 post-seeding signal intensity of the same well. Representative pictures of bioluminescence imaging are shown at the bottom of dot plots. p values by Bonferroni’s test. Norm: Normoxia; Hypo: Hypoxia. p-values by Bonferroni’s test. Norm: Normoxia; Hypo: Hypoxia. P values by one-way ANOVA tests with Bonferroni’s test.

Full size image

Next, to assess the suppressive effect of 2-MeOE2 on osteolytic tumor growth in bone, we employed an osteotropic subpopulation of MDA-MB-231, namely SCP28, known for their rapid induction of osteoclastogenesis in BICA11. Notably, under CoCl2-induced chemical hypoxia, the BICA samples exhibited larger lesions compared to normoxia conditions, whereas 2-MeOE2 demonstrated significantly enhanced therapeutic efficacy (Fig. 5B). The inhibitory effect was bone-specific, as the same treatment did not show an inhibitory effect in 2D cultures (Fig. 5C). These results stand as a strong rationale for considering the use of HIF1 inhibitors in the treatments on breast cancer bone metastases, especially for ER− tumors.

Inhibition of hypoxia signaling delays outgrowth of bone metastases induced by ER−tumors in vivo

Next, we tested the efficacy of 2-MeOE2 on the progression of SCP28 in vivo. Again, we conducted bone-oriented IIA injection to enrich the cancer cells in the hind limbs of SCID mice, thereby avoiding metastasis to other organs (e.g., the lung metastases) and allowing us to focus on bone metastatic disease. As an aggressive osteotropic cell model, SCP28 quickly established overt bone lesions within 3 weeks, with the bioluminescence signal reaching saturation as the scientific endpoint for the treatment experiment. The administration of 60 mg/kg 2-MeOE2 twice a week significantly reduced the skeletal tumor burdens, as evidenced by both the bone colonization kinetics in intact mice and the endpoint evaluation of tumor burdens in the extracted bone tissues (Fig. 6A, B, Supplementary Fig. 3). As expected, 2-MeOE2 treatment hindered HIF signaling, evidenced by abolished nuclear staining of HIF1A in bone metastases specimens (Fig. 6C). 2-MeOE2 also reduced CTSK staining associated with bone metastases (Fig. 6D). Notably, although the 2-MeOE2 administration was initiated very early (24 h after the tumor inoculation), the treatment effect was not observed in the first two weeks. It only became evident when tumor progression reached the late stage, suggesting that 2-MeOE2 may be more effective during the late stage of bone metastases (Fig. 6A). Interestingly, 2-MeOE2 treatment appeared to be more effective at suppressing bone lesions in the metaphysis region (around the knee joint) compared to the diaphysis region (the shaft of long bone of the mouse tibia (Fig. 6B). The underlying mechanism of this intriguing phenomenon will be investigated in future studies.

Fig. 6: Pharmacological inhibition of HIF delays outgrowth of bone metastases in vivo.
figure 6

A Growth curves of IIA-injected SCP28 bone lesions in hind limbs. n = 7 animals for both groups. p-values by 2-way ANOVA for assessing treatment effects over time. Representative IVIS images (three out of seven mice in each group) are shown at the bottom of growth curves. Estradiol tubes were not transplanted. B Bioluminescence intensity of hindlimb bones extracted from vehicle and treatment groups described in (A). p-values by 2-sided t-test. Representative IVIS images are shown. Error bars: SD. n = 7 bones. C Representative IHC staining of bone-lesion samples derived from (A, B). Multiple bone fragments (n = 3) are shown. Scale bar, 100 μm. SCP28 cancer cells can be distinguished from bone marrow cells by their enlarged nuclei. High magnification pictures are shown to highlight nucleus staining of HIF1 in cancer cells. D Representative H&E and IHC staining of bone-lesion samples derived from (A, B). Multiple bone fragments (n = 3 bones) are shown. SCP28 cancer cells can be distinguished from bone marrow cells by their enlarged nuclei. High magnification pictures are shown to highlight CTSK staining around tumor area. Scale bar, 1 mm.

Full size image

Discussion

In the clinic, the propensity and progression of bone metastases can vary across different molecular subtypes of breast cancer. For example, the overall incidence of bone metastasis is higher in ER+ tumor patients (71%) than those with ER− tumors (47%)45. Yet, despite the lower rate of bone metastases presentation, ER− tumors often develop symptomatic bone relapse earlier than ER+ tumors, usually within 5 years of diagnosis45. Conversely, there is a typically observed latency up to 10 years before the appearance of overt bone recurrence of ER+ breast cancer24. These observations indicate that the phenotypes and the underlying mechanisms of bone metastasis may differ considerably between ER+ and ER− breast cancer. Accordingly, the varying bone colonization of ER+ and ER− tumors can provide different time ranges and opportunities to eradicate bone lesions and reduce the lytic destruction. In this paper, we initiate a pilot characterization on this divergency regarding osteolytic phenotypes between bone metastases induced by ER+ and ER− breast cancer cells.

It is widely recognized that different subtypes of breast cancer exhibit varying responses to treatment and diverse clinical outcomes46. Consequently, therapeutic approaches have been tailored to address specific patient populations, such as hormone therapy for ER/PR-positive cancer, HER2-targeting treatment for HER2-positive cancer, and immune checkpoint blockade therapy for triple-negative cancer46. However, the treatment options for skeletal related events (SRE) remain limited and lack specificity for the diverse biological background of breast cancer6. Our research endeavors aim to address this gap and offer insights into the development of precision/personalized treatment for bone metastases originating from different breast cancer subtypes. Our data indicates that in comparison to ER+ cells, ER− cells may engage in more robust interaction with osteoclast lineage cells, leading to the induction of more osteolytic morphology in their bone metastatic lesions. This observation is substantiated by both experimental and clinical evidence. Notably, Fig. 1A demonstrated that the osteoblast-recruiting ability of MDA-MB-231 cells is not only significantly lower than that of MCF-7, but also weaker than that of previously reported ER− murine breast cancer cells, 4T1 and 4TO710. The osteoblastic phenotype observed in 4T1 and 4TO7 lesions may be attributed to their preserved expression of E-cadherin, which is largely absent in MDA-MB-231 cells47. Our previous findings showed that E-cadherin facilitates the formation of E-N heterotypic adhesion junction between bone-colonized cells and osteoblasts10. Clinically, reduced E-cadherin expression is significantly associated with lack of estrogen receptor expression and basal-like phenotype in breast cancer48,49. The more pronounced loss of E-cadherin in human ER−cell lines compared to murine cells may help explain the greater reduction in osteoblastic activity observed in ER−xenograft models. Nevertheless, our data did not investigate the direct regulation of bone turnover in breast cancer bone metastases by ER signaling in cancer cells, which needs to be addressed by additional efforts.

Despite its vascularity, bone constitutes a notably hypoxic environment31,35. Previous studies have highlighted the role of hypoxia in promoting osteolysis by stimulating the secretion of osteolytic inducers from osteogenic cells and bone marrow stromal cells35,50,51. Our findings add a new dimension to this understanding by revealing that cancer cells also respond to hypoxia stimulation, ultimately influencing the maturation of osteoclasts. Notably, ER− cells, when compared to ER+ cells, exhibit accentuated activation of HIF signaling under hypoxia stress, which plays a pivotal role in the induction of bone metastatic genes, particularly the osteolysis regulators. This aligns with previous studies demonstrating that HIF1A knockdown in MDA-MB-231 cells reduced bone colonization and increased survival in nude mice33. In addition to the genes being tested in our paper, previous studies also indicate that hypoxia could induce the expression of lysyl oxidase (LOX)50,51,52. While our study did not observe differentiated LOX expression between the ER+ and ER− cells (data not shown), potentially due to the artificial hypoxia models employed in vitro, we could not dismiss the possibility of a more pronounced involvement of LOX in ER− cells contributing to osteolysis in physiological conditions or in intact mouse models. Additionally, HIF signaling may interact with other bone metastases-related molecular mechanisms to intensify the tumor progression and osteolysis. For example, previous studies indicate that tumor-derived Jagged1 promotes osteolytic bone metastasis of osteotropic breast cancer cells by engaging Notch signaling in osteoblasts and osteoclasts, which can be inhibited by the anti-Jagged 1 antibody treatment17,19. Given the evidence of the crosstalk between Notch and HIF signaling53,54, it is intriguing to explore whether a simultaneous blockade of these two pathways could achieve a synergistic effect in treating bone metastases.

Presently, the standard of care for treating breast-cancer-related SREs primarily is to target bone turnover, focusing on arresting pronounced bone resorption and pain relief through agents like bisphosphonates and denosumab6. Our data proposes that targeting the hypoxia pathway, especially with HIF inhibitors like 2-MeOE2, may achieve a dual therapeutic effect by suppressing tumor growth as well as preventing osteolysis simultaneously. Unlike bisphosphonates that might be administered to the patient cohort of all breast cancer subtypes as adjuvant therapy, our findings suggest that HIF inhibitors may be more favorable for targeting ER− bone metastases. This represents a crucial stride toward a more precise and personalized treatment for bone metastases.

While our in vivo treatment demonstrated therapeutic effect, two potential limitations warrant further consideration. First, the effectiveness of 2-MeOE2 was tested only in a preventive setting (i.e., treatment initiated before overt lesions were formed) rather than in treating established bone metastases. Encouragingly, comparison of growth kinetics suggested a preferential effect against SCP28 progression in late stages (Fig. 6A). Therefore, there is a compelling need to investigate the therapeutic effect of 2-MeOE2 on established bone metastases in future studies. Second, 2-MeOE2 is known for its off-target effects, particularly as a high-affinity agonist of the G protein-coupled estrogen receptor (GPER) which may influence the progression of triple-negative breast cancer55. Some studies have shown that disrupting GPER function can reduce cell proliferation in standard 2D cultures56,57. However, in our 2D cultures treated with 2-MeOE2, this reduction was not observed. Instead, the inhibitory effect of 2-MeOE2 seems specific to the hypoxic bone environment. Nevertheless, the possibility that 2-MeOE2 exerts its effect through GPER cannot be entirely ruled out. Future research should address this caveat by evaluating the impact of 2-MeOE2 on GPER knockout cells in vivo.

Overall, our study revealed a significant correlation between hypoxia and osteogenesis in breast cancer bone metastases, which represents a promising target for bone metastases therapies. Our findings, particularly those from MDA-MB-231 cells, align with prior research on the role of hypoxia and HIF1A in promoting breast cancer bone metastases33,35,58,59. Particularly, Hiraga et al. used genetic approaches, by overexpressing constitutively active and dominant negative HIF1A in MDA-MB-231, to demonstrate that hypoxia and HIF1A suppress osteogenesis and promote osteolysis in bone59. Building upon this, we expanded our investigation scope to include more breast cancer models and clinical data, indicating that the hypoxia-associated osteolysis may be more pronounced in ER− breast cancer bone metastases compared to those induced by ER+ tumors. Dunn et al. pioneered the application of 2-MeOE2 to in vivo bone metastases models, showing that the treatment decreased osteolytic lesion areas following intracardiac inoculation of parental MDA-MB-231 cells33. However, the intracardiac injection of parental MDA-MB-231 also induced metastases in other visceral organs including lung, adrenal glands, liver, and brain34,60, which posed challenges in evaluating the specific impact of bone metastases to the entire animal. In this study, we assessed the inhibitory effect of 2-MeOE2 in a more bone-tropic MDA-MB-231 subpopulation and conducted bone-oriented IIA injection. These approaches provide more specific justification for the application of anti-hypoxia therapies in treating more advanced osteolytic bone metastases. Recently, the FDA approved the use of belzutifan as an HIF2α inhibitor for treating renal cell carcinoma, central nervous system hemangioblastomas, and pancreatic neuroendocrine tumors61. Further efforts are needed to assess whether FDA-approved or clinically-trialed HIF inhibitors could reduce tumor lesions in immune-competent mice; and whether their efficacy can be enhanced through combination with other anti-cancer regimens (e.g., anti-Jagged 1 antibodies). This will mark a promising avenue for advancing the field of precision medicine in bone metastasis therapies.

Methods

Animals

Immune-deficient athymic nude mice were purchased from Envigo for RNAseq and qPCR in Figs. 1–4. NOD SCID (NSG) mice were purchased from Roswell Park Comparative Oncology Shared Resource for Figs. 5–6. The 6-week-old female mice were used for all experiments. All animal work was done in accordance with a protocol approved by the Institutional Animal Care and Use Committee of Roswell Park Comprehensive Cancer Center. Mice were euthanized when the bioluminescence signals of metastatic tumors saturate at default setting of the IVIS imager or two months after injection, whichever comes first. During this time, if the mice exhibited any sign of distress or pain, they were euthanized before the terminal time point. None of the experiments exceeded these limits. We have complied with all relevant ethical regulations for animal use.

Cell lines

MCF-7, MDA-MB-231, HCC-1937, ZR-75-1, T47D, MDA-361, BT20, MDA-436, and BT549 were purchased from American Type Culture Collection (ATCC). Dr. Xiang Zhang (Baylor College of Medicine) generously provided SCP28 cells. The immortalized human MSC cell line was a generous gift from Dr. Max Wicha’s laboratory at University of Michigan. Human U937 cells were purchased from ATCC. For cell passage and maintenance, cells were cultured in DMEM media supplemented with 10% fetal bovine serum (FBS). As the parallel control for BICA assay, 500 cells were seeded in DMEM/F12 media supplemented with 2% FBS in 96-well plates. MCF7, MDA-MB-231, and their derivative cells were authenticated by the Cytogenetics and Cell Authentication Core at MD Anderson Cancer Center by STR profiling. Cell lines were subjected to testing for mycoplasma contamination once every three months.

The PWIPZ system was used to introduce firefly luciferase fused with GFP or RFP into breast cancer cells via lentivirus transduction. The Fluc-GFP fusion gene was cloned into expression vector pwpt (Addgene #12255). The pwpt-Fluc-GFP vector was transfected into 293 T cells with pMD2.G (Addgene #12259) and psPAX2 (Addgene #12260) to package lentivirus. Successfully labeled cells were isolated by FACS sorting of GFP or RFP-positive cells. The MSC and U937 are unlabeled so that the growth of cancer cells can be determined by bioluminescence imaging in both 2D and 3D co-culture.

IIA injection

Bone Metastases lesions induced by MCF-7 and MDA-MB-231 were both developed by intra-iliac injections13. Briefly, mice were anesthetized by isoflurane and then restrained on a far infrared warming pad (Kent Scientific). A 1.5 cm incision was made between the 4th and 5th nipples in the lower right abdomen. Blunt dissection was performed to separate muscles and expose the common iliac artery. 5 × 105 MCF7, 2.5 × 105 MDA-MB-231, or 2.5 × 105 SCP28 cancer cells were suspended in 0.1 ml PBS were injected via 31 G needles. The wound was then closed by a 7 mm EZ clip. If needed, estradiol pellet was implanted under the back neck.

Bioluminescence signaling was checked by injection of 100 µl of 15 mg/ml filtered luciferase substrate D-luciferin (LUCNA-1G, Goldbio) via the intra-orbital sinus. Animals were imaged weekly using IVIS Lumina II (Advanced Molecular Vision).

In vivo pharmacological treatment

2-MeOE2 (A4188, APExBio) was dissolved in 10% DMSO + 40% PEG300 + 5% Tween-80 + 45% Saline and applied by oral gavage twice per week at the dosage of 60 mg/kg. For experiments in need of estradiol supply, sustained-released estradiol tubes (7 mg/mouse) were prepared and transplanted to mice subcutaneously before cancer cell injection according to a published protocol62.

BICA

Bone-In-Culture pieces were prepared from the proximal epiphysis and metaphysis of the tibias from mice aging 5–6 weeks in age. The bone was crushed using sterilized bone pliers, trimmed with micro dissecting scissors, and then transferred to a low-attachment 96-well plate, containing 100 µL of DMEM/F12 with 2% FBS per well, using micro dissecting forceps. Bone fragments from multiple animals were randomized and mixed into each group. The cells were then seeded into the wells at a concentration of 1000 per well. Cells were incubated at 37 °C for 24–48 h to allow for sufficient homing and bone colonization. The cancer cells that have successfully attached should establish strong adhesion with the bone matrix and bone niche cells, enabling their growth within the bone fragment throughout the entire experiment period. Afterward, wells that contained bone fragments were washed and rinsed with PBS to remove unattached cancer cells. The medium was refreshed 2–3 times a week.

For 2D cultures done in parallel to the BICA cultures, cells were seeded into a 96-well plate at a concentration of 500 per well, using DMEM/F12 with 2% FBS medium.

IVIS Spectrum imaging was performed by using luciferase-tagged cells and measuring bioluminescence intensities using the IVIS Lumina II (PerkinElmer, Advanced Molecular Vision). Before imaging, sterilized firefly luciferase substrate D-Luciferin (GoldBio LUCK-1G) was added to cell cultures at a final concentration of 150 µg/mL. Residual luciferin was removed by rinsing with PBS and replacing the medium after imaging occurred. Data was analyzed using Living Image Software (PerkinElmer, v4.7.3). The acquired intensity data of both BICA and 2D cultures was normalized over the day 1 post-seeding bioluminescence signal intensity (BioLI) of the same well. Inhibitory rate was calculated by the formula: (1-BioLItreatment/Mean value of BioLIvehicle) × 100%.

IF and IHC staining

Bone pieces were collected over 3 weeks after tumor cell inoculation. Sample preparation was assisted by Experimental Tumor Model Resource (ETM) of Roswell Park Comprehensive Cancer Center. Staining was performed using antibodies against ALP (Abcam, ab108337, 1:500), CTSK (Abcam, ab19027, 1:1000), GFP (Invitrogen, GF28R, 1:500), and HIF1 (CST, #3716, 1:300). Antibody validation information is presented on the vendor’s website.

2D co-culture

Cancer cells were seeded at a concentration of 200 per well with or without various numbers of MSC or U937 cells in 96 well plates containing DMEM media with 2% FBS. Medium was refreshed 2–3 times each week. Samples were incubated and traced by bioluminescence imaging for 7 days. Because MSC and U937 were unlabeled, the bioluminescence intensity only indicated the growth of cancer cells. The acquired intensity data were normalized over the 0 MSC or 0 U937 group.

3D co-culture

In 24-well low-attachment plates, cancer cells were seeded into the wells at a concentration of 2 × 104 per well, with MSC cells at a concentration of 2 × 104 per well for 4–7 days. Because MSC was unlabeled, the bioluminescence intensity only indicated the growth of cancer cells. The acquired intensity data of co-cultured cells was normalized over the mono-culture groups.

RNAseq

Bone Metastases lesions induced by MCF-7 and MDA-MB-231 were both developed by intra-iliac injections, with technical details available in “Method—intra-iliac artery injection”. Following the experimental settings of previous studies, estradiol (E2) pellets were implanted subcutaneously for mice inoculated with MCF-7 tumor but not for those inoculated with MDA-MB-231.

RNA-seq was conducted by collecting size-matched in vivo bone lesions (small, medium, and large lesion-pairs for MDA-MB-231 and MCF-7). The quantification of bone lesions was performed by IVIS imaging at the epiphysis and metaphysis of extracted femur and tibia at the endpoint. Since the efficiency of luciferase labeling varied in the two cells, same bioluminescence intensities do not indicate the same sizes of tumor burdens. Therefore, the actual lesion sizes were determined according to a standard curve that was generated using a diluted gradient series of cell amounts ranging from 12,500 to 400,000. Three pairs of bone metastatic lesions were selected to ensure that the sizes of tumor burdens are matched between the two cell models (Supplementary Fig. 1A, B).

Using IVIS ex vivo imaging as reference, bone regions with high bioluminescence intensity were selectively cut with dissecting scissors, grinded by bone pliers, and homogenized by Precellys 24 Homogenizer (Bertin Instrument) with 2.8 mm stainless steel beads in 2 mL reinforced tubes (Bertin Instruments, MK28R). Total RNA of the cancer/bone mixture was extracted using the Direct-zol RNA Miniprep Kit (Zymo Research, R2051).

The first cDNA strands were prepared using the RevertAid First Strand cDNA Synthesis Kit (Thermo Fisher Scientific, K1622) with >200 ng total RNA input. The second cDNA strands were prepared using the NEBNext mRNA Second Strand Synthesis Module (NEB, E6111L). Libraries were generated using the Illumina Nextera XT DNA Library Prep Kit (illumine). Cluster generation was conducted with the Illumina Nextseq 500/550 High Output v2 Kit and then sequenced on the Illumina Nextseq 500 equipment with the assistance of Genomics Shared Resource in Roswell Park Comprehensive Cancer Center.

qPCR in supplement to RNAseq

Concerning the potential effects of E2 on bone morphology and tumor behavior, additional IIA-inoculated in vivo bone lesions were generated to compare MCF7- and MDA-MB-231 bone metastases in both E2-supplemented and E2-free scenarios. qPCR was performed in supplement to RNAseq to avoid redundant sequencing. MCF-7 bearing mice and MDA-MB-231 bearing mice were subject to both conditions with and without implantation of 7 mg sustained-released E2 tubes. Notably, IIA-induced MCF7 bone lesions can grow without E2 supply but at a slower rate (Supplementary Fig. 1C). Tumor sizes were normalized and matched as described in “Method—RNAseq”. Bone regions with high bioluminescence intensity were selectively cut with dissecting scissors, grinded by bone pliers, and homogenized by Precellys 24 Homogenizer (Bertin Instrument) with 2.8 mm stainless steel beads in 2 mL reinforced tubes (Bertin Instruments, MK28R). RNA was isolated from cells using the Direct-zol RNA Miniprep Kit (Zymo Research, R2051). Reverse transcription was performed using the BioRad iScript cDNA Synthesis Kit (Biorad, 1708891). Quantitative PCR was performed using the iTaq Universal SYBR Green Supermix (BioRad, 1725121) with the BioRad CFX96 Thermal Cycler. The qPCR primer sequences used in this study are as follows: Gapdh, 5′-CATCACTGCCACCCAGAAGACTG-3′, 5′-ATGCCAGTGAGCTTCCCGTTCAG-3′; Alpl, 5′-CCAGAAAGACACCTTGACTGTGG-3′, 5′- TCTTGTCCGTGTCGCTCACCAT -3′; Runx2, 5′- CCTGAACTCTGCACCAAGTCCT-3′, 5′- TCATCTGGCTCAGATAGGAGGG-3′, Cola1, 5′- CCTCAGGGTATTGCTGGACAAC-3′, 5′- CAGAAGGACCTTGTTTGCCAGG-3′ Sp7, 5′- GGCTTTTCTGCGGCAAGAGGTT-3′, 5′- CGCTGATGTTTGCTCAAGTGGTC-3′; Ibsp, 5′- AATGGAGACGGCGATAGTTCCG-3′,5′- GGAAAGTGTGGAGTTCTCTGCC-3′; Bgalp, 5′-GCAATAAGGTAGTGAACAGACTCC -3′, 5′- CCATAGATGCGTTTGTAGGCGG-3′; Csf1r, 5′- TGGATGCCTGTGAATGGCTCTG-3′, 5′- GTGGGTGTCATTCCAAACCTGC-3′; Tnfrsf11a, 5′- GGACAACGGAATCAGATGTGGTC-3′, 5′- CCACAGAGATGAAGAGGAGCAG-3′; Itgb3, 5′- GTGAGTGCGATGACTTCTCCTG-3′, 5′- CAGGTGTCAGTGCGTGTAGTAC-3′; Ctsk, 5′- AGCAGAACGGAGGCATTGACTC-3′, 5′- CCCTCTGCATTTAGCTGCCTTTG-3′; Tnfsf11, 5′- GTGAAGACACACTACCTGACTCC-3′, 5′- GCCACATCCAACCATGAGCCTT-3′; Ldha, 5′- ACGCAGACAAGGAGCAGTGGAA-3′, 5′-ATGCTCTCAGCCAAGTCTGCCA-3′; Vegfa, 5′-CTGCTGTAACGATGAAGCCCTG-3′, 5′-GCTGTAGGAAGCTCATCTCTCC-3′; Slc2a1, 5′-GCTTCTCCAACTGGACCTCAAAC-3′,5′- ACGAGGAGCACCGTGAAGATGA-3′; Hif1a, 5′-CCTGCACTGAATCAAGAGGTTGC-3′, 5′- CCATCAGAAGGACTTGCTGGCT-3′; GAPDH, 5′- GTCTCCTCTGACTTCAACAGCG-3′, 5′- ACCACCCTGTTGCTGTAGCCAA-3′; HIF1A, 5′-TATGAGCCAGAAGAACTTTTAGGC-3′, 5′-CACCTCTTTTGGCAAGCATCCTG-3′; VEGFA, 5′- AGGGCAGAATCATCACGAAGT-3′, 5′- AGGGTCTCGATTGGATGGCA -3′; SLC2A1, 5′- ATTGGCTCCGGTATCGTCAAC-3′, 5′- GCTCAGATAGGACATCCAGGGTA-3′; LDHA, 5′- GGATCTCCAACATGGCAGCCTT-3′, 5′- AGACGGCTTTCTCCCTCTTGCT-3′.

qPCR for 2D culture

Cells were cultured in normal condition or in hypoxia chamber for 24 hours. RNA was isolated from cells using the Direct-zol RNA Miniprep Kit (Zymo Research, R2051). Reverse transcription was performed using the BioRad iScript cDNA Synthesis Kit (Biorad, 1708891). Quantitative PCR was performed using the iTaq Universal SYBR Green Supermix (BioRad, 1725121) with the BioRad CFX96 Thermal Cycler. The qPCR primer sequences used in this study are as follows: GAPDH, 5′- GTCTCCTCTGACTTCAACAGCG-3′, 5′- ACCACCCTGTTGCTGTAGCCAA-3′; HIF1A, 5′-TATGAGCCAGAAGAACTTTTAGGC-3′, 5′-CACCTCTTTTGGCAAGCATCCTG-3′; SPP1, 5′-CTCCATTGACTCGAACGACTC-3′, 5′-CAGGTCTGCGAAACTTCTTAGAT-3′; IL6, 5′-ACTCACCTCTTCAGAACGAATTG-3′,5′-CCATCTTTGGAAGGTTCAGGTTG-3′; CTGF, 5′- AAAAGTGCATCCGTACTCCCA -3′, 5′- CCGTCGGTACATACTCCACAG -3′; PTHLH, 5′-AAGGTGGAGACGTACAAAGAGC-3′, 5′-CAGAGCGAGTTCGCCGTTT-3′;VEGFA, 5′- AGGGCAGAATCATCACGAAGT-3′, 5′- AGGGTCTCGATTGGATGGCA -3′; SLC2A1, 5′- ATTGGCTCCGGTATCGTCAAC-3′, 5′- GCTCAGATAGGACATCCAGGGTA-3′; MMP1, 5′-CTCTGGAGTAATGTCACACCTCT-3′, 5′- TGTTGGTCCACCTTTCATCTTC -3′.

Statistics and reproducibility

In vitro assays were performed with 3–5 biological replicates, each being an average of at least 3 technical replicates. For BICA assays, each bone piece is deemed as a biological replicate. Randomization process was performed by randomly assigning pre-treated bone fragments into separate groups after tumor inoculation. All results are presented in the form of mean ± SD unless otherwise specified. The sample size was determined on the basis of early experience and noted in the corresponding figures or figure legends. Student’s t-test was used for two-group comparison. For qPCR analysis in Figs. 1D, 2D, and 3E, F, data were log2-transformed and subject to paired t-test. In experiments consisting of more than two groups, one-way ANOVA tests were applied and performed with the assumption of equal variation. Two-way ANOVA test was applied for comparison of growth curves. No data is excluded. The investigator was not blinded to the group allocation during the experiment.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.