Unbiased niche labeling maps immune-excluded niche in bone metastasis

SUMMARY

Metastatic cancer cell fate is shaped by the local microenvironment niches. To unbiasedly define the cellular and molecular features of metastatic niches, we developed Sortase A–Based Microenvironment Niche Tagging (SAMENT), which selectively labels cells encountered by cancer cells during metastasis. Applying SAMENT across multiple cancer models and target organs revealed shared niche features, including macrophage enrichment and T cell depletion, alongside marked organ-specific phenotype heterogeneity in niche macrophages. In bone, metastatic niches are enriched for macrophages expressing estrogen receptor alpha (ERα) with active ERα signaling. Conditional deletion of Esr1 in macrophages significantly impaired bone colonization by enabling T cell infiltration. ERα⁺ macrophages were also identified in human bone metastases across multiple cancer types. Together, these findings define a distinct ERα⁺ macrophage niche and establish macrophage ERα signaling as a key driver of T cell exclusion during metastatic colonization.

Graphical Abstract

INTRODUCTION

Throughout tumor progression and evolution, cancer cells are constantly interacting with the host, and this interaction profoundly impacts their cellular fates, metastatic behaviors and therapeutic responses. Numerous previous studies have elucidated the roles of specific microenvironment niches (i.e., cells that are immediately adjacent to cancer cells) in the progression of metastasis. For instance, dormant disseminated tumor cells (DTCs) in the bone marrow tend to remain adjacent to the perivascular niche^1–3.Endothelial cells have been shown to induce mesenchymal-to-epithelial transition of DTCs⁴. Bone remodeling may activate DTCs and attract them to the osteogenic niche comprised of osteoblasts and their precursors⁵. Osteogenic cells and cancer cells can form heterotypic adherens junctions (hAJs) and gap junctions (GJs), which in turn activate the mTOR and calcium signaling, respectively, to promote progression of micrometastases^6–8. Interaction with the osteogenic cells transiently reduces estrogen receptor alpha (ERα) expression and increases stem cell properties in the cancer cells⁹, which may in turn promote bone colonization, therapeutic resistance, and even further dissemination to other organs¹⁰. Taken together, the crosstalk between cancer cells and different BME niches appears to be of pivotal importance. However, this crosstalk is usually difficult to investigate because of its dynamic nature¹¹ – many interactions may be transient and occur simultaneously between cancer cells and multiple other cell types. Thus, an unbiased approach is highly critical to capture all niche cells at specific stages of tumor progression.

A previous metastasis niche-labeling method was based on expression of secreted lipid-permeable mCherry (sLP-mCherry)^12,13. sLP-mCherry autonomously and continuously diffuses from cancer cells to neighboring cells, allowing these labeled cells to be distinguished from others. This approach is easy to apply, although the timing and range of diffusion are difficult to control and may lead to unintended labeling.

Sortase A (SrtA) is a bacterial enzyme linking proteins to cell walls. It catalyzes cleavage of the threonine of the LPXTG motif, followed by fusion of this cleaved peptide to another protein with (Glycine)_n at the opposite end¹⁴. Recently, SrtA has been adopted to study cell-cell interactions mediated by tight receptor-ligand interactions¹⁵. In these studies, one cell type was engineered to express SrtA-conjugated ligands. Upon provision of tag-LPXTG substrates, SrtA could transfer tag-LPXTG to the cognate receptors expressed on the other cell type. This strategy is powerful but is limited to only studying one ligand-receptor pair at a time. More recently, a more generalized approach was developed to record immune cell interactions mediated by any ligand-receptor pairs, which was named uLIPSTIC (universal labeling immune partnerships by SorTagging intercellular contacts)¹⁶, which evolved from the original LIPSTIC technology^17,18.When expressed in metastatic cells, the uLIPSTIC system may be used to label metastatic niche cells that stably interact with cancer cells through ligand-receptor binding or cell-cell adhesion. However, whether it can also capture more transient and dynamic interactions remains to be tested.

In this study, we report a system named SrtA-based MicroEnvironment Niche Tagging (SAMENT). By combining SrtA and synthetic ligand-receptor binding, we aim to label any cells that are physically encountered by cancer cells. Compared to sLP-mCherry, labeling by SAMENT is dependent on simultaneous induction of SrtA and provision of substrates, and is amenable to downstream single-cell omics and spatial transcriptomic analyses. Compared to uLIPSTIC, SAMENT is independent of stable ligand-receptor binding or cell-cell junction, and is therefore more capable of recording transient and weak cell-cell interactions. The robustness and specificity of SAMENT enable us to conduct cellular and molecular profiling of the tumor immune microenvironment (TIME) landscape of different cancer models, in different organs and temporal stages, and under different treatments – thereby establishing a comprehensive metastatic niche atlas.

ER signaling is a major therapeutic target and biomarker of ER⁺ breast cancers¹⁹. Anti-ER agents, including selective estrogen receptor modulators (SERMs), aromatase inhibitors (AIs), and selective estrogen receptor degraders (SERDs), have been used in treating ER⁺ tumors in preventive, neoadjuvant, adjuvant and metastatic settings^20,21. However, the roles of ER in non-cancer cells have just begun to be elucidated²². By applying SAMENT for unbiased immune profiling in the tumor microenvironment, we observed an enrichment of ERα⁺ macrophages across multiple cancer models among tumor-contacting immune cellular components and then performed a macrophage-specific Esr1 knockout to investigate its functional contribution to bone colonization. Interestingly, ERα expression and signaling were also observed in human bone metastases from multiple cancer types. This discovery exemplifies the utility of SAMENT and raises an interesting possibility to target macrophage-derived ERα signaling in treating bone metastasis of multiple cancer types.

RESULTS

Design and validation of SAMENT

In cancer cells, a transgene consisting of fused the N-terminal mono-glycine recognition sortase A variant (mgSrtA)²³ and a green fluorescent protein (GFP) nanobody (GFPn)²⁴ is expressed under the control of a doxycycline-inducible promoter. MgSrtA is a variant of SrtA that recognizes a single glycine at protein termini such that a substantial proportion of cell surface proteins can serve as targets for unbiased labeling. Glycosylphosphatidylinositol-anchored GFP (GFP-GPI) mice were used as hosts. These mice constitutively express a membrane-anchored, outward-facing surface GFP (surGFP) on all cell types²⁵, which can bind GFPn, thereby allowing unbiased tagging of any cells that come to the vicinity of cancer cells. When cancer cells are provided with doxycycline to induce mgSrtA-GFPn and with biotin-LPETGS (sortase substrate), a five–amino acid SrtA recognition motif in which biotin serves as a detectable tag, the surGFP-GFPn binding stabilizes cell-cell interactions and allows mgSrtA to robustly label these adjacent cells for further analysis (Figure 1A).

Figure 1. SAMENT approach enables specific labeling of metastatic niches.

(A) Schematic of SAMENT validation: doxycycline (Dox)-induced tdTomato–mgSrtA–GFPn (SAMENT) in sender cells binds surGFP on receiver cells and transfers biotin-LPETGS (500 μM) during a 2h co-culture.

(B-D) Flow cytometry of biotin⁺ surGFP⁺ receivers under indicated Dox/substrate conditions (n=4); labeling required both SAMENT induction and biotin-LPETGS.

(E-G) Ex vivo co-culture of LLC1-SAMENT cells with bone marrow (BM) from GFP-GPI or WT mice (1:50, 2 h) ± Dox (1 μg/mL) (n=3). SAMENT expression alone minimally labeled WT BM cells.

(H) Bone metastasis in GFP-GPI mice after intra-iliac artery (IIA) injection of LLC1-SAMENT cells ± Dox; biotin, white; SAMENT, red.

(I-J) Biotin⁺ CD45⁺ BM frequency (n=16/group) and correlation with SAMENT⁺ tumor cells in bone metastases (n=21).

(K) SAMENT labeling in bone micro-metastases.

(L) Composition of biotin⁺ versus biotin⁻ BM cells from mice with IIA-induced bone metastases (LLC1 n=12; B16F10 n=7; PyMT-N n=8; 4T1 n=7).

SAMENT: Sortase A-Based Microenvironment Niche Tagging; mgSrtA: mono-glycine recognition sortase A variant; GFPn: green fluorescent protein (GFP) nanobody; surGFP: surface GFP; biotin-LPETGS: biotin-labeled LPETGS (a five–amino acid SrtA recognition motif with biotin serving as a tag to monitor labeling); LLC1: Lewis Lung Carcinoma cells; MFI: median fluorescence intensity; GFP-GPI: glycosylphosphatidylinositol-anchored GFP; WT: wildtype; Neu: Neutrophils; Mono: Monocytes; Macs: Macrophages; MSC: Mesenchymal stem cell; OPC: Osteoprogenitor cell; OB: Osteoblast cell; EC: Endothelial cell.

Data are mean ± SEM; N = biological replicates. Scale bars (H, K), 100 μm. P values: one-way ANOVA + least significant difference (LSD) (B, D), two-way ANOVA + LSD (G), unpaired t-test (I), Pearson correlation (J), paired t-test (L).

See also Figure S1

We first tested the SAMENT system in vitro. Lewis Lung Carcinoma cells (LLC1) were engineered to express tdTomato/mgSrtA-GFPn or surGFP to generate red sender cells or green receiver cells, respectively. The sender and receiver cells were then co-cultured. We observed that cell labeling between sender and receiver cells occurred if and only if the biotin-LPETGS peptides and doxycycline were simultaneously provided (Figures 1B–1D). Intracellular GFP (intraGFP) expression in receiver cells did not result in labeling (Figures S1A–S1C). Within a wide range, the fraction of labeled receiver cells (over all receiver cells) was linearly proportional to incubation time (0–4 hours) (Figures S1D and S1E), concentration of biotin-LPETGS peptides (0.01–0.5 mM) (Figures S1F and S1G), and doxycycline (0.01–1 μg/ml) (Figure S1H). When the substrate was washed off, the biotin signal remained stable for about 4 hours and then decayed to the background level within 24 hours, with a half-life of about 16 hours (Figures S1I and S1J).

We then tested bone marrow (BM)-derived cells as receiver cells. These cells were extracted from GFP-GPI or control mice and co-cultured with sender cancer cells. As expected, labeling of host BM cells occurred only when 1) mgSrtA was induced by doxycycline, and 2) biotin-LPETGS substrate was provided (Figures 1E–1G). Without GFP-GPI, we observed a baseline level of labeling (~2–3%), which is presumably mediated by strong ligand-receptor binding or cell-cell junctions. However, the labeling is dramatically increased to 17–18% by GFP-GPI expression (Figure 1G), suggesting that the synthetic GFPn-GFP-GPI interaction captures a wider range of adjacent cell that are not necessarily tightly binding cancer cells. Taken together, these results support that SAMENT functions as designed in vitro and exhibits a strong regulatability, sensitivity and specificity.

We then performed in vivo validation of SAMENT. Given our previous knowledge of bone metastatic niches, we decided to use bone as a primary model. Cancer cells expressing inducible mgSrtA-GFPn were delivered to the bone through intra-iliac artery (IIA) injection²⁶. Upon establishment of bone lesions, we examined biotin tagging in animals with or without provision of doxycycline to induce mgSrtA-GFPn. It was evident that labeling robustly occurred when the SAMENT system was induced by doxycycline (Figure 1H). Flow cytometry analyses showed that induction of SAMENT resulted in 6–8 fold higher biotin labeling efficiency compared to background level (Figure 1I), which is consistent with the in vitro result (Figure 1G). Moreover, the frequencies of biotin⁺ cells were proportional to tumor burden (Figure 1J). Importantly, this labeling was also effective in microscopic lesions (Figure 1K). These data support the effectiveness of SAMENT in vivo.

We then extended SAMENT to multiple models and determined if we could replicate previous findings. We sorted biotin⁺ vs. biotin⁻ cells based on expression of surface markers using flow cytometry. The impact of doxycycline treatment was more dramatic on biotin⁺ cells than on biotin⁻ cells (Figure S1K), suggesting that the observed changes were due to induction of SAMENT rather than doxycycline itself. To deduce enrichment or depletion of different cell types in metastatic niches, the frequency of each cell type among biotin⁺ cells was normalized to that among biotin⁻ cells. In all four models, we observed consistent enrichment of endothelial cells and osteoblast/osteoprogenitor cells (Figure 1L), which is consistent with previous discoveries of the roles of perivascular niches^1,2,4, osteogenic niches^6–8,27, and skeletal stem cells²⁸. In addition, we also noticed an unbalanced distribution of monocytes, macrophages, and T cells between biotin⁺ and biotin⁻ populations. In particular, metastatic niches appeared to be devoid of T cells, which is expected according to the paradigm of immune editing. The simultaneous depletion of monocytes and enrichment of macrophages suggests accelerated monocyte-to-macrophage differentiation upon encountering cancer cells (Figure 1L).

Comparing SAMENT to other cell-labeling technologies

A few approaches have been established to track cell-cell interactions. Among these, we selected two representative ones to compare with SAMENT. A secreted monomeric Cherry red fluorescent protein (mCherry) containing a modified lipo-permeable Transactivator of Transcription (TATk) peptide (sLP-mCherry) has been used to label the metastatic niche via diffusion from cancer cells to nearby niche cells (Figure 2A). Universal labeling of immune partnerships by SorTagging intercellular contacts (uLIPSTIC) was designed for immune cell crosstalk. However, it can be readily repurposed for cancer-niche interactions (Figure 2A). We adapted and optimized both systems to compare with SAMENT. In addition, we also included a variant of SAMENT lacking the GFP nanobody (designated as SAMENT-no GFPn) such that labeling only occurs when cell-cell contact is stabilized by strong ligand-receptor interactions and cell-cell junctions. In theory, SAMENT-no GFPn is similar to uLIPSTIC (Figure 2A).

Figure 2. Benchmarking niche-labeling strategies in bone metastasis.

(A) Schematics of SAMENT, uLIPSTIC, and sLP-mCherry.

(B) Ex vivo co-culture labeling of CD45⁺ BM cells (left) and fraction of tumor cells expressing labeling components (right) (n=3/system). Controls: -Dox (SAMENT/uLIPSTIC) or BM only (sLP-mCherry).

(C) In vivo labeling efficiency (left) and tumor cell machinery expression (right) in bone metastases (SAMENT-no GFPn: control n=4, labeled n=8; SAMENT: control n=5, labeled n=5; uLIPSTIC: control n=5, labeled n=6; sLP-mCherry: control n=3, labeled n=7).

(D) Discrete lesions: density of labeled niche cells across 50-μm bands from the tumor edge.

(E) Diffuse lesions: distance from each labeled niche cell to its nearest tumor cell nucleus.

sLP-mCherry: secreted lipid-permeable mCherry; uLIPSTIC: Universal Labeling Immune Partnerships by SorTagging Intercellular Contacts; ROIs: regions of interest.

Scale bars, 100 μm. Data are mean ± SEM; N = biological replicates. P values: two-way ANOVA + LSD (B, C).

See also Figure S2

When induced to comparable levels in vitro, SAMENT exhibited the highest efficiency of cell labeling within a few hours of induction, followed by uLIPSTIC and SAMENT-no GFPn. The poor labeling efficiency of sLP-mCherry under this condition may be due to the slow diffusion rate of cell-permeable mCherry compared with sortase-mediated peptide transfer (Figure 2B). When applied to bone metastasis models in vivo, the sensitivity of SAMENT remained higher than that of uLIPSTIC. However, sLP-mCherry appeared to label a greater number of cells (Figure 2C). We then carefully examined metastatic lesions expressing uLIPSTIC, SAMENT and sLP-mCherry, and analyzed the frequency and spatial distribution of labeled cells. We used a customized spatial immunofluorescence quantification pipeline (deposited at Zenodo: https://zenodo.org/records/15374531) to computationally quantitate the results. The pipeline is robust across magnifications and tissues, enabling cross-batch, cross-tissue, and cross-image comparisons with minimal Fiji preprocessing. For quantification, we classified metastatic lesions as either (1) isolated lesions with well-defined tumor beds and tumor–adjacent tissue borders or (2) diffusive lesions with tumor cells intermingled with bone marrow. For isolated lesions, the frequency of biotin⁺ or mCherry⁺ cells was evaluated as a function of distance from the lesion borders (Figures 2D and S2A). For diffusive lesions, we evaluated the distance of labeled cells from the nearest cancer cells (Figures 2E and S2B). In both cases, sLP-mCherry appeared to be able to label cells over a much longer range than uLIPSTIC and SAMENT, as expected due to the paracrine nature of labeling by sLP-mCherry. Among the niche cells that are immediately adjacent to cancer cells, 60–70% were labeled by SAMENT and 30–40% were labeled by uLIPSTIC (Figure S2C). This is consistent with the different designs of the two systems, i.e., whereas uLIPSTIC requires strong ligand-receptor interaction to stabilize cell-cell contact, SAMENT obviates this need by leveraging GFPn-GFP binding. Therefore, SAMENT can label a wider range of cells that are geographically close, whereas uLIPSTIC labels a subset of cells with stronger associations to metastatic cells. Regarding controllability, SAMENT and uLIPSTIC both require doxycycline and biotin-LPETGS for efficient labeling, enabling precise temporal and spatial control. In contrast, sLP-mCherry labels autonomously without tight controllability, thereby progressively labeling surrounding niche cells as tumors develop (Figures S2D–S2F). Taken together, we conclude that the three systems may suit different yet complementary needs in metastasis research. The three systems also have limitations, which will be discussed later.

SAMENT generates a molecular and cellular atlas of metastatic niches.

The “seed-and-soil” hypothesis has been used to understand preferential metastatic colonization in different organs^29,30. The crosstalk between disseminated cancer cells and the immediate neighboring cells constitutes a significant part of “seed and soil” interactions. Having established SAMENT and validated it in bone metastasis, we went on to determine how the same cancer cells may encounter and adapt to various milieus when arriving at different organs. To this end, we ignored organ-specific parenchymal cells (e.g., osteoblasts in bone and hepatocytes in liver) and only focused on immune cells to allow meaningful inter-organ comparisons.

First, we validated that SAMENT worked as expected in the lung and liver (Figure 3A), in addition to bone. We then performed intracardiac (IC) injection of LLC1 cells to generate experimental metastases in multiple organs. Consistent with previous experiments (Figures S1K), Fluorescence-Activated Cell Sorting (FACS) analyses of biotin⁺ niche cells across bone, brain, lung, and liver metastases revealed stronger doxycycline effects on biotin⁺ than biotin⁻ cells, indicating that the changes resulted from SAMENT induction (Figures S3A). Similar to bone, metastases in other organs were also devoid of T cells but relatively enriched with macrophages (Figure 3B). However, there are also interesting differences noticed among organs. For instance, liver metastases exhibited less depletion of T cells and there appears to be a decrease as opposed to increase of neutrophils that are directly contacting cancer cells in lung metastases. These differences warrant further examination using additional models and lay the foundation for future research.

Figure 3. Macrophages are major, phenotypically diverse components of metastatic niches across organs.

(A) SAMENT labeling in lung and liver metastases after intracardiac (IC) injection of LLC1-SAMENT cells into GFP-GPI mice (biotin, white; tumor, red).

(B) Frequencies of biotin⁺ and biotin⁻ cells across metastatic organs (n=5; paired t test).

(C–E) scRNA-seq comparison of biotin⁻ versus biotin⁺ niche cells across tissues: PCA (C,D) and transcriptional distance (E; t test). Samples: bone n=10, brain n=4, liver n=4, lung n=4.

(F–I) UMAP of myeloid subclusters annotated by identity (F), biotin status (G), tissue origin (H), and subpopulation frequencies by tissue and biotin group (I).

Mo/Mφ: Monocyte / Macrophage; cDC: conventional Dendritic Cell; Pro-Mo: Pro-monocyte; Mo: Monocyte; NK: Natural Killer cells; Pre-Neu: Precursor neutrophil; pDC: plasmacytoid Dendritic Cell; Baso-Mast: Basophil/Mast cell; GMP: Granulocyte-Monocyte Progenitor; HSPCs: Hematopoietic Stem and Progenitor Cells; Nφ: Neutrophils; Endo: Endothelial cells.

Data are mean ± SEM. See also Figure S3

We next investigated transcriptomic variations of the same cell types at different metastatic sites. Single-cell RNA sequencing (scRNA-seq) data of major cell types (Figure S3B) were first divided into biotin⁻ and biotin⁺ (non-niche vs. niche) and then subjected to principal component analysis (PCA). In this analysis, the distance between any two points or the area of a polygon connecting multiple points indicates the degree of dissimilarity among the underlying data. Among the biotin⁻ cells, points representing the same cell type from different organs clustered tightly together, indicating high similarity across organs (Figure 3C). However, large inter-organ variations were observed among biotin⁺ cells for many cell types, especially for macrophages (Figure 3D). The phenotypic diversity of macrophages has been intensively studied before^31–36. Here, our data indicate that direct contact with cancer cells induces phenotypic heterogeneity of macrophages in different organs to the extent exceeding their normal diversity across tumor-free tissues. NK cells were ranked 2^nd in terms of their variations across different metastatic sites. An unbiased analysis of niche NK cells revealed multiple macrophage regulation pathways (Figure S3C), suggesting macrophages as a downstream effector of NK cells in metastatic niches.

We then asked how biotin⁺ and biotin⁻ cells vary from one another in different organs. To this end, we calculated the “distance” between biotin⁺ and biotin⁻ cells and compared the average distance of each cell type in four different organs (Figure 3E). Macrophages were again the top cell type in this comparison, indicating that their interaction with tumor cells causes the largest phenotypic alteration. In fact, among the four organs, macrophages in bone exhibit the largest difference between biotin⁺ and biotin⁻ cells, prompting further investigation elaborated in later sections. Other noteworthy tumor-induced changes include immature B cells in brain metastasis (Figure S3D) and neutrophil and monocyte progenitors in lung metastasis (Figures S3E and S3F). Further analyses revealed interesting cancer-induced transcriptomic shifts in these cell types in an organ-specific fashion. Among all cell types, granulocyte-monocyte precursors (GMPs) were included because of our special interest stemming from our previous research³⁷. Since GMPs are confined to the bone marrow, we could not analyze them in other organs. Nevertheless, biotin⁺ and biotin⁻ GMPs displayed a substantial transcriptomic difference (Figure S3G) that is consistent with our previous finding of tumor-entrained systemic reprogramming of hematopoietic stem and precursor cells including GMP³⁷.

Having observed the remarkable macrophage heterogeneity and plasticity upon direct contact with tumor cells, we performed detailed analyses on isolated scRNA-seq data of the monocyte lineage (Figures 3F–3H) and identified several clusters. Tabulation of frequencies of these clusters across different metastatic sites and biotin statuses led to interesting findings (Figure 3I). Compared to biotin⁺ cells, biotin⁻ cells enrich monocytes (Clusters C0, C4, and C9) similarly across all metastatic sites. On the other hand, the macrophage clusters (C1, C3, C5, C7, and C10) are the dominant populations in biotin⁺ cells but exhibited largely diverse distributions across different metastatic sites. Specifically, C1 and C3 abound in multiple sites, whereas C5, C7, and C10 are predominantly enriched in one specific organ. Taken together, the data suggest that C1 and C3 are monocyte-derived macrophages, whereas C5, C7, and C10 are tissue-specific macrophages recruited to metastases. Indeed, detailed examination of known markers of macrophages of different origins supports this hypothesis (Figures S3H and S3I).

Collectively, we have established an immune cell atlas of metastatic niches using SAMENT. We identified cell types that are preferentially contacted by cancer cells, and that are commonly or differentially enriched in metastases at different sites. Among all cell types, macrophages occur most frequently surrounding disseminated cancer cells and appear to be phenotypically reprogrammed upon interaction with metastases. Therefore, we will focus on the roles of macrophages in metastatic niches for the remainder of this study.

Macrophages in the metastatic niche express estrogen receptor alpha (ERα)

Having obtained a general atlas of metastatic niche cells across different organs, we turned our attention back to bone metastasis for more in-depth discoveries. The inducibility of SAMENT allowed us to examine different temporal stages of bone colonization. Comparison between an early time point (10 days after IIA injection) and a relatively late time point (21 days after IIA injection) revealed evolution of metastatic niche. The noticeable changes with time include recruitment of neutrophils, reduction of monocytes, further increase of macrophages, and exacerbated depletion of T cells (Figures 4A and S4A). The prominent changes in macrophages prompted us to validate these findings by multiplexed immunofluorescence staining. Indeed, macrophages (F4/80⁺) were enriched in the metastatic niche (Figure 4B).

Figure 4. Macrophages within the bone metastasis niche exhibit enhanced ERα signaling.

(A) Flow cytometry quantification of major immune/stromal subsets in biotin⁺ versus biotin⁻ BM at early (day 10; n=11, osteogenic n=7) and late (3 weeks; n=12) stages of bone metastasis following LLC1-SAMENT IIA injection (paired t test).

(B) Immunofluorescence of biotin-labeled F4/80⁺ macrophages (red) near LLC1-SAMENT cells (green) in bone metastases. Scale bar, 100 μm.

(C) Heatmap of DEGs between biotin⁺ and biotin⁻ macrophages.

(D) GSVA of pathway enrichment across organs; estrogen-related gene sets highlighted (Benjamini–Hochberg-adjusted P).

(E) Human bone metastasis sections showing ERα (white) in CD68⁺ macrophages (red) within CK8/CK19⁺ tumors (green). Scale bar, 50 μm.

(F) ESR1 expression in published human bone-metastasis scRNA-seq (healthy n=5; breast cancer n=12; lung cancer n=8; kidney cancer n=14).

DEGs: differentially expressed genes; GSVA: Gene Set Variation Analysis; MsigDB: Molecular Signatures Database; CK8: Cytokeratin8; CK19: Cytokeratin19.

See also Figure S4

Differentially expressed genes and pathways were identified by unbiased comparison between biotin⁺ and biotin⁻ macrophages (Figures 4C and 4D). We noticed a number of up-regulated pathways are related to ER signaling (Figures 4D and S4B). Interestingly, the activation of ER signaling appeared to be highly specific to biotin⁺ macrophages, because 2 out of 5 top differentially expressed genes that distinguish biotin⁺ macrophages from other myeloid cells including biotin⁻ macrophages are classic ER targets^38–41(Figure S4C). The enrichment of ER was further validated at the levels of RNA, protein, and target genes (Figures S4D–S4F). In addition, we adoptively transferred macrophages expressing an estrogen response elements (EREs) reporter to animals carrying bone metastases and observed stronger ER reporter activity (Figures S4G–S4I) and higher ERα expression (Figures S4J–S4L) in macrophages located within metastatic niches compared with those in non-niche areas. ER signaling is a major driver of ER⁺ breast cancer. It also plays an important role in many other cell types, including macrophages, T cells, osteoblasts, and osteoclasts. The immunoregulatory functions of ER signaling may result from evolutionary needs for tolerating the fetus during pregnancy. The immunosuppressive impact of ERα in macrophages was recently reported in melanoma^42–44. The pleiotropy of ERα makes it an interesting target both for the validation of SAMENT methodology and for the discovery of emerging biology in metastasis.

We next asked if ERα expression in macrophages also occurs in human bone metastases. In a recent study, we collected and analyzed 42 clinical samples of human bone metastases colonized from a variety of primary cancer sites⁴⁵. By immunofluorescence staining, we observed robust macrophage ERα expression in 10 metastases, including male patients (Figures 4E and S4M). For comparison, we also examined a primary colorectal cancer and did not observe ERα expression in associated macrophages (last column of Figure S4M). Enhanced expression of ESR1 in macrophages was further validated by single-cell RNA sequencing in human bone metastases from breast cancer (BC), lung cancer (LC), and kidney cancer (KC), compared to healthy bone marrow (Figure 4F).

ERα expression in the niche macrophages is activated by fatty acid (FA)-PPARγ signaling and confers immunosuppressive phenotype

We then set out to characterize the ERα⁺ niche macrophages. By correlating estrogen response signatures with other HALLMARK pathway signatures, we found that estrogen signaling in biotin⁺ macrophages was significantly associated with lipid metabolism–related pathways (Figure S5A). Given the central role of PPARγ (Peroxisome proliferator-activated receptor gamma) in regulating lipid metabolism⁴⁶, we next examined the correlation between estrogen response and PPAR pathways, and found a strong positive association specifically in biotin⁺ macrophages (Figure 5A). Treating macrophages with PPARγ agonist, rosiglitazone, upregulated Esr1 RNA expression (Figure 5B). PPARγ is a hallmark of M2 macrophages. Indeed, besides PPARγ, the ER responsive signature also exhibited strong correlation with IL4 and IL10 responsive genes exclusively in biotin⁺ macrophages (Figure 5C). Beyond their correlation with the ER signature, the overall pathway activities of PPARγ, IL4, and IL10 were markedly higher in biotin⁺ than in biotin⁻ macrophages (Figures 5A and 5C). Interestingly, conditionally knockout Esr1 using LysM-cre (Figure S5B) resulted in significantly increased interferon pathways and decreased IL6/STAT3 pathway (Figure 5D), indicative of a conversion from M2-like, anti-inflammation macrophages to M1-like, pro-inflammatory macrophages. Consistent with this notion, we also observed SPP1 and CXCL9 among the top down-regulated and up-regulated genes upon conditional Esr1 KO (Figure 5E). CXCL9/SPP1 ratio has been recently suggested to be a key parameter to distinguish between pro-tumor and anti-tumor macrophages⁴⁷. Furthermore, conditional KO of Esr1 in macrophages also led to stimulatory changes to the T cell compartment (Figure S5C). Together, these data strongly indicate that ERα⁺ niche macrophages are pro-tumor and M2-like, and the ER pathway is an important driver of this phenotype.

Figure 5. Paracrine fatty acids from cancer cells induce ERα upregulation in macrophages via PPARγ.

(A) Single-cell correlation between Hallmark Estrogen Response and KEGG PPAR signaling scores in biotin⁺ versus biotin⁻ macrophages (5,000 cells shown; Benjamini–Hochberg-adjusted P).

(B) Esr1 RNA in BMDMs treated with rosiglitazone.

(C) Correlation of estrogen-response scores with macrophage responses to rosiglitazone+IL-4 or IL-10.

(D-E) scRNA-seq of TAMs from Esr1^LysM versus Esr1^fl/fl tumors: GSEA (D) and DEGs (E). n=12/group.

(F) ERE-luciferase reporter activity in BMDMs after co-culture with LLC1 cells. Scale bar, 50 μm.

(G-J) ERα protein and Esr1 RNA in BMDMs treated with LLC1 CM or control (G), BSA-palmitate or BSA (H), LLC1 CM with FASN (I) or RAB27a knockdown (J).

KEGG: Kyoto Encyclopedia of Genes and Genomes; PPARγ: Peroxisome proliferator-activated receptors gamma; GSEA: Gene Set Enrichment Analysis; TAMs: tumor-associated macrophages; BMDM: Bone Marrow Derived Macrophages; ERE: Estrogen Response Element; CM: conditioned medium; FASN: Fatty acid synthase.

Data are mean ± SEM. Adjusted P values (A, C, E) used Benjamini–Hochberg correction. Significance: repeated-measures one-way ANOVA + LSD (B, H–J); unpaired two-tailed t-test (G).

See also Figure S5

To identify the mechanism triggering ERα expression in macrophages, we performed in vitro co-culture experiments using cancer cells and bone marrow-differentiated macrophages expressing luciferase/GFP reporters driven by EREs. Compared to the monoculture control, both luciferase activities and GFP expression were significantly elevated in co-cultures of 5/7 models (Figures 5F and S5D), suggesting up-regulation of ER transcriptional activity upon interaction with cancer cells. ERα expression was increased in macrophages when conditioned medium (CM) from cancer cells is provided (Figure 5G), indicating cancer-derived soluble factors responsible for ERα activation. Having found the strong correlation between PPARγ and ER pathways, we hypothesized that fatty acid signaling may be an upstream regulator of ERα in macrophages. Addition of palmitate to culture medium induces ERα expression in macrophages (Figure 5H). Knocking down fatty acid synthesis in cancer cells diminished the CM-induced ERα expression (Figure 5I). We asked how cancer-derived fatty acids may diffuse to act on adjacent macrophages. Among several possible mechanisms, our preliminary data supported extracellular vesicles (EVs) as vehicles of FA transmission because knockdown of Rab27a in cancer cells, a key mediator of EV formation, reduced CM’s effect on ERα expression in macrophages (Figure 5J). The same results were also reproduced using RAW264.7 cells as the macrophage model (Figures S5E–S5M). Taken together, our data indicates that ERα expression in macrophages is driven by cancer cell-derived fatty acids through paracrine interaction mediated by EVs.

ERα in monocytes and macrophages does not regulate bone formation or remodeling.

Osteoclasts drive the osteolytic vicious cycle, which is a major mechanism underlying advanced bone metastases of multiple cancer types^48–51. In our scRNA-seq dataset, we could robustly identify a subset of cells (C5) overexpressing characteristic genes of osteoclasts (Figures S5N and S5O)⁵². Although fully matured osteoclasts are typically too large to be captured by scRNA-seq, their precursors and some newly formed osteoclasts are smaller and more amenable for the technology^53,54. Monocytes and macrophages are precursors of osteoclasts. However, the osteoclast differentiation is determined by the microenvironment and intrinsic properties of precursors, and only occurs to a subset of monocytes and macrophages^43,44. It is important to ask if ERα⁺ macrophages are indeed osteoclasts or their precursors. ERα activity in C5 is heterogeneous, indicating that there are indeed ERα^high osteoclasts (Figure S5P). We then performed pseudotime analyses to deduce precursors of C5 and identified C9 (monocytes) (Figure 3I) as the likely osteoclast precursors in both biotin⁺ and biotin⁻ samples (Figure S5Q). We assessed ERα expression in osteoclasts by multi-color immunofluorescence staining and tartrate resistant acid phosphatase (TRAP) staining. Among the TRAP⁺RANK⁺/CTSK⁺ osteoclasts, we observed higher ERα expression in mononuclear, early osteoclasts as compared to multi-nuclear mature osteoclasts (Figure S5R). Functionally, conditional knockout of Esr1 did not affect osteoclast frequency and activation (Figures S5S and S5T). Therefore, we conclude that ERα signaling diminishes as osteoclasts fully differentiate.

Since a subset of new osteoclasts and osteoclast precursors exhibit ERα activity, we asked if blocking ERα signaling may affect bone formation or remodeling. We examined Esr1^LysM mice with LysM-cre driven Esr1 knockout in monocytes, mature macrophages and granulocytes⁵⁵, which include osteoclast precursors and differentiated osteoclasts. The knockout was validated by immunofluorescence staining (Figure S5U) and Western blots (Figure S5B). Loss of ERα in these cells did not generate discernible effects on bone formation (Figure S5V), nor did it affect trabecular or cortical bones (Figures S5W–S5Z). This is consistent with the fact that ERα upregulation in macrophages is induced by close contact with cancer cells but remains at a relatively low level in normal bones. Thus, perturbation of macrophage-derived ERα does not appear to alter bone homeostasis.

Conditional knockout of Esr1 in macrophages hinders metastatic colonization in bone.

To examine if macrophage-derived ERα plays a role in bone metastasis, we performed IIA injection to introduce experimental bone metastases of LLC1, E0771, 4T1 and PyMT-M cells in syngeneic hosts with or without LysM-cre driven Esr1 knockout. Since LLC1 is a lung cancer cell line and the known functions of ERα are gender-specific, we also investigated bone colonization of LLC1 in male vs. female hosts (Figure 6A). In all experiments, bone colonization in Esr1^LysM animals was significantly retarded. Interestingly, for the LLC1 model, the impact of conditional Esr1 knockout was even more striking in male animals (Figure 6A). This is consistent with patient data in Figure 4E, which clearly showed a similar level of ERα in bone metastases of male patients. Importantly, the decrease in metastatic burden was not only evident in the hind limbs subjected to IIA injection (Figure 6B), but in most models also noticeable in contralateral limbs (non-injected) (Figure 6C) and lungs (Figure 6D). These latter metastatic lesions resulted from further dissemination of the initial bone metastases as shown in our previous studies¹⁰.

Figure 6. Conditional knockout of Esr1 in macrophages significantly retarded tumor colonization in bone.

(A) Longitudinal tumor growth after IIA injection of indicated cell lines in Esr1^LysM versus control mice (n indicated); individual mice (dotted) and group mean (bold).

(B-D) Ex vivo bioluminescence imaging (BLI) of injected (B) and contralateral (C) hindlimbs, and lungs (D) from mice in (A).

(E) Schematic of spontaneous bone metastasis assay.

(F-H) Subcutaneous tumor weight, spontaneous metastasis burden by BLI, and incidence in hindlimb bones (n=25/group).

Data are mean ± SEM; N = biological replicates. P values: two-way ANOVA + LSD (A), Mann–Whitney (B–D, F–G), Chi-square (H).

See also Figure S6

LLC1 cells spontaneously metastasize to bone, which allows us to examine the role of ERα⁺ macrophage in a complete metastatic cascade (Figure 6E). Subcutaneous source tumors grew at a similar rate in Esr1^LysM and control Esr1^fl/fl animals (Figure 6F) and were removed after reaching 1 cm³ (Figure 6E). 18 days after tumor resection, we harvested hind limb bones and lungs to assess metastases to these organs. A 5- to 10-fold difference was observed for both hind limbs (Figures 6G and 6H). Lung metastases trended toward a reduction but did not reach statistical significance (Figure 6G).

To determine if the role of ERα is specific to macrophages, we also conditionally knocked out ERα by crossing Esr1^fl/fl with a number of other cre- lines that target major cell types in bone including Cx3cr1-cre, Osterix-cre, Col1a1-cre, S100A8-cre, and Itgax-cre in both experimental metastasis and spontaneous metastasis assays (Figures S6). Among these, Esr1^Cx3cr1 animals exhibited significant decrease of bone colonization (Figure S6A), albeit to a lesser degree compared to Esr1^LysM (Figures 6A–6C). CX3CR1⁺ marks a subset of macrophages⁵⁶, and this result supports the role of ERα in macrophages. On the other hand, S100A8 is predominantly expressed in granulocytes⁵⁷, and Itgax (a.k.a. CD11c) is a dendritic cell marker. The lack of effect in these models distinguish macrophages from other major myeloid cell populations. Taken together, our results strongly support the hypothesis that ERα in macrophages plays an important role in bone colonization.

Conditional knockout of Esr1 in macrophages leads to increased T cell infiltration and activation in metastatic niches.

We next set out to determine the cellular mechanism underlying the observed decrease of bone colonization upon Esr1 knockout in macrophages. Major immune cell types in metastasis-associated bone microenvironment were quantitated using flow cytometry and compared between Esr1^LysM and Esr1^fl/fl hosts and uncovered a significant increase of biotin⁺ CD8⁺ T cells (Figure S7A), suggesting an elevation of direct contact between cancer cells and CD8⁺ T cells. However, the overall frequency of T cells did not appear to change by the Esr1 knockout. This is contrasted by anti-PD1 treatment, which significantly increased overall CD8⁺ T cells (Figure S7B). Consistent with the SAMENT output, multiplexed immunofluorescence staining revealed dramatic changes in the spatial distribution of T cells. In wild type hosts, T cells appear to be sequestered by ERα⁺ macrophages at the edge of metastatic lesions (Figures 7A and 7B). When ERα is depleted in macrophages, metastasis-infiltrating, Granzyme B⁺ T cells significantly increased (Figures 7C–7D). We then used a computational pipeline to quantitate the distribution of T cells and macrophages. The algorithm identified the border between metastatic lesions and adjacent cells. The distance of macrophages and T cells to the boarder was then profiled. In wild type hosts, macrophages were concentrated surrounding the boarder, T cells appeared to be sequestered by macrophages and kept outside of the tumor bed. In Esr1^LysM hosts, this pattern of spatial distribution was disrupted, and T cell infiltration was markedly increased (Figures 7E and 7F). Thus, conditional knockout of Esr1 in macrophages appears to impact T cells differently as compared to anti-PD1 treatment.

Figure 7. ERα signaling in macrophages restricts CD8⁺ T cells tumor infiltration.

(A-B) Immunofluorescence of CD8⁺ T cells (white), F4/80⁺ macrophages (red), and LLC1 cells (green) in Esr1^LysM vs. Esr1^fl/fl bone metastases (A) and quantification of tumor-infiltrating CD8⁺ T cells (B, n=11–12).

(C-D) Granzyme B⁺ CD8⁺ T cells (C) and quantification (D, n=10–12).

(E-F) Spatial analysis of CD8⁺ T-cell distribution relative to tumor boundaries and heterotypic cell clustering.

GrB: Granzyme B.

Data are mean ± SEM. Scale bars (A, C, E), 100 μm. P values: unpaired two-tailed t-test (B, D).

See also Figure S7

We next asked if fulvestrant, a SERD clinically used to treat metastatic breast cancer, could achieve the same effect as Esr1 knockdown. Indeed, both SAMENT (Figure S7C) and immunofluorescence staining (Figure S7D) confirmed that fulvestrant, but not anti-PD1, significantly increases T cell infiltration. Interestingly, we observed decreased tumor burden in Esr1^LysM hosts but not in anti-PD1 treated animals (Figure S7E), suggesting that T cell infiltration is more important in controlling metastatic colonization. However, combining anti-PD1 and Esr1 knockout did not achieve additive or synergistic effects (Figure S7E), suggesting more complicated interactions between ERα signaling in macrophages and the PD1 pathway in T cells. The underlying mechanism remains unclear and beyond the scope of this study, although it may be caused by increased monocytes and macrophages that selectively occur to animals subjected to the combined treatment (Figure S7F). Further studies are needed to identify other combination strategies to leverage the increased T cell infiltration when macrophage ERα is inhibited.

DISCUSSION

We established SAMENT to unbiasedly profile microenvironment niche cells that are immediately adjacent to metastatic seeds. Compared to previous approaches such as uLIPSTIC and sLP-mCherry, SAMENT is unique in that it most sensitively labels niche cells that are geographically adjacent to metastatic cells independent of strong ligand-receptor binding or cell-cell junctions. Therefore, SAMENT is suitable for studies that are seeking to identify niche cells interacting with cancer cells through a wider range of mechanisms. Like uLIPSTIC, SAMENT builds in precise control of labeling. Efficient labeling occurs within a few hours in vivo upon and only upon simultaneous administration of doxycycline and biotin-LPETG. Nonetheless, SAMENT also has its limitations compared with other approaches. The synthetic binding between GFP-GPI and GFPn helps capture transient and dynamic cell-cell interactions but also interrupts the natural progression of tumors. For this reason, the induction of SAMENT needs to be conducted immediately before sample collection and cannot track cell-cell interaction longitudinally. This limitation is less stringent for uLIPSTIC and sLP-mCherry, which can theoretically record cancer-niche crosstalk over time, depending on the turnover rates of biotin–LPETGS–conjugated G5–Thy1.1 or mCherry (43 hour)¹². Therefore, the three approaches compared in this study are complementary and may be used for different purposes.

In general, the niche-labeling approaches, including SAMENT, are complementary to state-of-the-art spatial transcriptomics (ST) platforms such as Nanostring CosMx and 10x Genomics Xenium. Compared with these financially and computationally intensive technologies, niche-labeling approaches meet distinct research needs. First, they allow the isolation of live niche cells, which can be used for functional studies. Second, niche-labeling enables unbiased sampling of all niche cells from the entire tumor-bearing organs, rather than relying on thin sections or a few selected small regions of interest as in ST. Third, biotin⁺ or mCherry⁺ niche cells are more amenable to further downstream analyses including FACS and multi-omics profiling (e.g., simultaneous RNA and ATAC sequencing), which provides more in-depth information about metastatic niches. Owing to these unique advantages, niche-labeling approaches can be combined with ST to advance our understanding of metastasis-niche interactions.

Macrophages play important roles in tumor progression including metastases at different sites and development of therapeutic resistance^58–65. Aside from being precursors of osteoclasts, the specific functions of macrophages in bone metastasis have only recently been studied in breast and prostate cancers^31,32. Herein, our study elucidated a distinct subset of macrophages expressing ERα. Interestingly, ERα expression is induced by cell-cell communication between cancer cells and macrophages, and conditional knockout of Esr1 appears to alleviate the exclusion of T cells from infiltrating metastatic lesions. Mechanistically, our data suggest that ERα signaling biases macrophages toward an SPP1^high/CXCL9^low state, which has been associated with reduced T cell recruitment in the tumor microenvironment^47,66,67. This provides a potential molecular explanation for the observed immune exclusion and highlights how ERα activity in macrophages may shape the metastatic niche.

Notably, the more pronounced decrease in bone metastases observed in Esr1^LysM mice relative to Esr1^Cx3cr1 mice suggests that ERα functions mainly within macrophages, with monocyte-specific deletion insufficient to recapitulate the full phenotype. The immunosuppressive effects of the ERα pathway in macrophages have been investigated before⁴². However, the spatial relationships among ERα⁺ macrophages, cancer cells and T cells, as revealed by SAMENT and immunofluorescence staining, have never been discovered. This highlights the relevance of spatial information in our understanding of tumor-host interactions. Toward this end, SAMENT provides an objective measurement of geographic, immediate proximity.

While our data support in situ induction of ERα in macrophages, we cannot definitively exclude the possibility that pre-existing ERα⁺ macrophages are recruited to the metastatic niche. Similarly, NK cell-mediated modulation may contribute to macrophage phenotypic heterogeneity, as suggested by Figure S3C. These alternative mechanisms underscore that the observed ERα⁺ macrophage population likely arises from a combination of in situ induction and potential recruitment or NK cell editing. Future lineage-tracing studies will be needed to clarify the relative contributions of these processes.

Despite its prominent female-specific roles, ERα signaling is also involved in multiple physiological processes in males⁶⁸. This has been largely neglected in cancer research. In this study, we were surprised to find that conditional knockout of Esr1 generated an even stronger effect on bone metastases in male animals. Consistently, we also observed robust ERα expression in macrophages associated with bone metastases in male patients. These observations raised an interesting possibility of applying well established anti-ER endocrine therapies to male patients with bone metastases of multiple cancer types. The downstream effectors of ERα signaling in macrophages will need to be identified in future research.

The importance of ERα signaling in macrophages may be questioned in breast cancer, as endocrine therapies have been shown to have no benefit for ER- patients⁶⁹. However, the previous studies have never focused exclusively on bone metastases, which are relatively infrequent and often co-occurring with metastases to other organs in ER- diseases. Furthermore, as shown in the final set of experiments, inhibition of ERα in macrophages may not be effective by itself but could synergize with immunotherapies because it facilitates T cell infiltration into metastatic lesions. Although we did not observe synergy between ERα knockout in macrophage and anti-PD1 treatment, it is still worth exploring the combinatory effects with other immunotherapies. Therefore, our findings may warrant future clinical trials on combined endocrine and immunotherapies on patients with bone metastases, and this combination may be extended to other cancer types and to patients of both genders.

Limitations of the Study

SAMENT effectively labels niche cells because the tight interaction between GFP-GPI and GFPn helps stabilize cell-cell interactions. However, this also poses a limitation as the same interaction also hinders cancer cells from migrating. As such, we can only induce GFPn immediately before we terminate the experiment and examine a snapshot of metastatic niches. Complementary approaches are under development to track niche cells longitudinally even after they depart from metastatic cells.

STAR METHODS

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Mouse Model

The in vivo procedures and use of animal models were conducted following protocol AN-5734 approved by the Baylor College of Medicine Institutional Animal Care and Use Committee. The mouse strains utilized, including C57BL/6J (RRID:IMSR_JAX:000664), BALB/cJ (RRID: IMSR_JAX:000651), LysMcre (RRID:IMSR_JAX:004781), NG2-CreERTM (RRID:IMSR_JAX:008538), Osx1-GFP::Cre (RRID:IMSR_JAX:006361), S100A8-Cre-ires/GFP (RRID:IMSR_JAX:021614), Col1a1-CreERTM (RRID:IMSR_JAX:016241), Cd11c-Cre (RRID:IMSR_JAX:008068), Cx3cr1-CreERTM (RRID:IMSR_JAX:020940) and Rosa26^uLIPSTIC(RRID:IMSR_JAX:038221), were obtained from The Jackson Laboratory. Mice genotyping was performed according to the JAX genotyping protocol. The CAG::GFP-GPI strain (C57BL/6J, RRID:IMSR_JAX:011106) was generously provided by Dr. Anna-Katerina Hadjantonakis²⁵ (Memorial Sloan-Kettering Cancer Center), and Esr1^fl/+ mice(RRID:IMSR_JAX:032173) were provided by Dr. Kenneth Korach⁸⁹ (National Institute of Environmental Health Sciences). Detailed information on the primers used for genotyping is available in the key resource table. Conditional Esr1 knockout strains were generated through crosses between male Cre+ transgenic mice and female Esr1^fl/fl mice over two generations. Esr1^LysM–GFP-GPI was generated by mating Esr1^LysM with GFP-GPI mice. Additionally, B6.LysMcre, B6. Esr1^fl/fl, and B6. CAG::GFP-GPI mice underwent over 10 generations of backcrossing to BALB/cJ in our lab, followed by further breeding to establish the BALB/cJ- Esr1^LysM strain.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Ghost Dye™ UV 450	Cytek® Biosciences	Cat#13-0868
CD16 / CD32 (Fc Shield)	Cytek® Biosciences	Cat# 40-0161; RRID:AB_2621443
CD45-VF450	Cytek® Biosciences	Cat#75-0451; RRID:AB_2621947
CD11b-APC/Cy7	Cytek® Biosciences	Cat#25-0112; RRID:AB_2621625
Ly6G-Percp/Cy5.5	Cytek® Biosciences	Cat#65-1276; RRID:AB_2621899
Ly6C-PE/Cy7	BioLegend	Cat#128018; RRID:AB_1732082
F4/80-BV510	BioLegend	Cat#123135; RRID:AB_2562622
B220-APC/Cy7	Cytek® Biosciences	Cat#25-0452; RRID:AB_2801487
CD3e-PerCP/Cy5.5	Cytek® Biosciences	Cat#65-0031; RRID:AB_2621872
CD4-PE/Cy7	Cytek® Biosciences	Cat#60-0041; RRID:AB_2621828
CD8a-BV711	BD Biosciences	Cat#752634; RRID:AB_2917619
PD-1-BV605	BioLegend	Cat#135220; RRID:AB_2562616
CD45-BV605	BioLegend	Cat#103140; RRID:AB_2650656
Ter119-BV605	BioLegend	Cat#116239; RRID:AB_2562447
CD31-APC/Cy7	BioLegend	Cat#102440; RRID:AB_2832288
Sca-1-Percp/Cy5.5	eBioscience	Cat#45-5981-82; RRID:AB_914372
CD51-BV421	BD Biosciences	Cat#740062; RRID:AB_2739827
CD140α-PE/Cy7	BioLegend	Cat#135912; RRID:AB_2715973
Flag-BV421	BioLegend	Cat#637322; RRID:AB_2750061
Streptavidin-APC	Cytek® Biosciences	Cat#20-4317
InVivoMAb anti-mouse PD-1 (CD279)	BioXcell	Cat#BE0146 RRID:AB_10949053
InVivoMAb Rat IgG2a	BioXcell	Cat# BE0089 RRID: AB_1107769
Mouse anti-Estrogen Receptor beta 1:100	Santa Cruz Biotechnology	Cat#sc-390243; RRID: AB_2728765
Mouse anti-alpha Tubulin 1:10000	Santa Cruz Biotechnology	Cat#sc-23948; RRID: AB_628410
Mouse anti-GAPDH 1:10000	Santa Cruz Biotechnology	Cat# sc-32233; RRID:AB_627679
rabbit anti-mRFP, 1:500	Rockland	Cat#600-401-379; RRID: AB_2209751
chicken anti-GFP, in 1:500	Novus Biologicals	Cat#NB100-1614; RRID:AB_10001164
rabbit anti-Granzyme B, 1:200	Novus Biologicals	Cat#NB100-684; RRID:AB_10001094
rat anti-mouse CD8a, 1:200	Thermo Fisher Scientific	Cat#14-0195-82; RRID:AB_2637159
Mouse anti-ERa(6F11) 1:100	Thermo Fisher Scientific	Cat#MA1-80216; RRID: AB_930763
Rabbit anti-ERα(MC-20) 1:500	Santa Cruz Biotechnology	Cat# sc-542; RRID:AB_631470
Rabbit anti-ERa(D8H8) 1:1000	Cell Signaling technology	Cat# 8644; RRID:AB_2617128
Goat anti-RANK 1:100	Novus Biologicals	Cat# AF692, RRID:AB_2205357
Rabbit anti- Cathepsin K 1:100	Proteintech	Cat# 11239-1-AP, RRID:AB_2245581
Rabbit anti-Rab27A (D7Z9Q)	Cell Signaling technology	Cat# 69295, RRID:AB_2799759
Rabbit anti-FASN	Proteintech	Cat# 10624-2-AP, RRID:AB_2100801
rabbit anti-mouse F4/80, 1:200	Cell Signaling technology	Cat#70076; RRID:AB_2799771
rat anti-mouse F4/80, 1:200	BIO-RAD	Cat#MCA497; RRID:AB_872005
rat anti-mouse Ly6G, 1:200	BioLegend	Cat#127602; RRID:AB_1089180
Mouse anti-Biotin APC, 1:100	Miltenyi Biotec	Cat#130-113-288; RRID:AB_2726071
Mouse anti-Biotin PE, 1:100	Miltenyi Biotec	Cat#130-113-291; RRID:AB_2726073
rabbit anti-human CD68 1:200	proteintech	Cat#25747-1-AP; RRID: AB_2721140
rat anti-human Cytokeratin 8, 1:100	DSHB	Cat#TROMA-I; RRID: AB_531826
rat anti-human Cytokeratin 19, 1:100	DSHB	Cat#TROMA-III; RRID: AB_2133570
rat anti-Flag tag, 1:200	BioLegend	Cat#637301; RRID:AB_1134266
Donkey anti-rat DyLight™ 405	Jackson ImmunoResearch,	Cat#712-476-153; RRID: AB_2632568
Donkey anti-rabbit Alexa Fluor 488 1:500	Jackson ImmunoResearch,	Cat#711-546-152; RRID:AB_2340619
Donkey anti-rat Alexa Fluor 488 1:500	Jackson ImmunoResearch,	Cat#712-545-153; RRID: AB_2340684
Donkey anti-chicken Alexa Fluor 488 1:500	Jackson ImmunoResearch	Cat#703-545-155; RRID: AB_2340375
Donkey anti-mouse Alexa Fluor 488 1:500	Jackson ImmunoResearch	Cat#715-545-151; RRID: AB_2341099
Donkey anti-mouse Alexa Fluor 555 1:500	Invitrogen™	Cat#A-31570; RRID: AB_2536180
Donkey anti-rat Alexa Fluor 555 1:500	Abcam	Cat#ab150154; RRID:AB_2813834
Donkey anti-rabbit Alexa Fluor 555 1:500	Invitrogen™	Cat#A-31572; RRID: AB_162543
Donkey anti-mouse Alexa Fluor 647 1:500	Jackson ImmunoResearch	Cat#715-605-151; RRID: AB_2340863
Donkey anti-rabbit Alexa Fluor 647 1:500	Jackson ImmunoResearch	Cat#711-605-152; RRID: AB_2492288
Donkey anti-rat Alexa Fluor 647 1:500	Jackson ImmunoResearch	Cat#712-605-153; RRID: AB_2340694
Donkey anti-chicken Alexa Fluor 647 1:500	Jackson ImmunoResearch	Cat#703-606-155; RRID:AB_2340380
Bacterial and virus strains
One Shot™ ccdB Survival™ 2 T1R Competent Cells	Invitrogen™	Cat#A10460
One Shot™ Stbl3™ Chemically Competent E. coli	Invitrogen™	Cat#C737303
Biological samples
Human bone metastasis samples	This paper	N/A
Chemicals, peptides, and recombinant proteins
Recombinant Murine M-CSF	Peprotech	Cat#315-02
Fulvestrant	ApexBio	Cat#A1428
Tamoxifen	Sigma-Aldrich	Cat#T5648
Calcein	Sigma-Aldrich	Cat#C0875
Alizarin Red S	Thermo Scientific Chemicals	Cat#400480250
Biotin-Ahx-LPETGS C-Terminal: Amidation	LifeTein	N/A
Doxycycline hyclate	Sigma-Aldrich	Cat#D9891
Dispase II	Sigma-Aldrich	Cat#D4693
Collagenase, Type I	Thermo Fisher Scientific	Cat#17100017
Collagenase, Type II	Thermo Fisher Scientific	Cat# 17101015
Dnase 1（Deoxyribonuclease I ）	Sigma-Aldrich	Cat# DN25
Hoechst 33342	Thermo Fisher Scientific	Cat#62249
D-Luciferin	Gold Biotechnology	Cat#LUCK-5G
Rosiglitazone	MedChemExpress	Cat#HY-17386
BSA-Palmitate Saturated Fatty Acid Complex	Cayman Chemical	Cat# 29558
BSA Control for Fatty Acid Complexes	Cayman Chemical	Cat# 29556
Critical commercial assays
Matrigel® Growth Factor Reduced (GFR) Basement Membrane Matrix	CORNING	Cat#356231
HIFI HOTSTART READY MIX	Roche Diagnostics	Cat#7958927001
Gateway™ BP Clonase™ II Enzyme mix	Invitrogen™	Cat#11789020
Gateway™ LR Clonase™ II Enzyme mix	Invitrogen™	Cat#11791100
NEBuilder® HiFi DNA Assembly Cloning Kit	New England Biolabs	Cat#E5520S
Fetal Bovine Serum,USA origin, Charcoal Stripped, sterile-filtered	Sigma-Aldrich	Cat#F6765
Gibco™ DMEM, high glucose, HEPES, no phenol red	Thermo Fisher Scientific	Cat#21063029
Standard Macrophage Depletion Kit	Encapsula NanoSciences	Cat#CLD-8901
X-tremeGENE HP DNA Transfection Reagent	Sigma-Aldrich	Cat#6366244001
3’ CellPlex Kit	10x Genomics	Cat#1000261
Chromium Single Cell 3’v3.1 kit	10x Genomics	N/A
Acid Phosphatase, Leukocyte (TRAP) Kit	Sigma-Aldrich	Cat# 387A-1KT
TotalSeq™-C0301 anti-mouse Hashtag 1 Antibody	BioLegend	Cat#155861; RRID: AB_2800693
TotalSeq™-C0302 anti-mouse Hashtag 2 Antibody	BioLegend	Cat#155863; RRID: AB_2800694
TotalSeq™-C0303 anti-mouse Hashtag 3 Antibody	BioLegend	Cat#155865; RRID: AB_2800695
TotalSeq™-C0304 anti-mouse Hashtag 4 Antibody	BioLegend	Cat#155867; RRID: AB_2800696
TotalSeq™-C0305 anti-mouse Hashtag 5 Antibody	BioLegend	Cat#155869; RRID: AB_2800697
TotalSeq™-C0306 anti-mouse Hashtag 6 Antibody	BioLegend	Cat#155871; RRID: AB_2819910
TotalSeq™-C0307 anti-mouse Hashtag 7 Antibody	BioLegend	Cat#155873; RRID: AB_2819911
TotalSeq™-C0308 anti-mouse Hashtag 8 Antibody	BioLegend	Cat#155875; RRID: AB_2819912
TotalSeq™-C0309 anti-mouse Hashtag 9 Antibody	BioLegend	Cat#155877; RRID: AB_2819913
TotalSeq™-C0310 anti-mouse Hashtag 10 Antibody	BioLegend	Cat#155879; RRID: AB_2819914
TotalSeq™-C0311 anti-mouse Hashtag 11 Antibody	BioLegend	Cat#155881; RRID: AB_2924482
TotalSeq™-C0312 anti-mouse Hashtag 12 Antibody	BioLegend	Cat#155883; RRID: AB_2924483
Deposited data
Raw and cellranger processed scRNA seq data	This paper	GSE263383; GSE296725
scRNA-seq Seurat and Scanpy objects	This paper	Zenodo: 10.5281/zenodo.15374531
Customized spatial analysis results based on fluorescence images	This paper	Zenodo: 10.5281/zenodo.15374531
Experimental models: Cell lines
PyMT-N (Murine TNBC)	Kim et al.70	N/A
PyMT-M (Murine TNBC)	Kim et al.70	N/A
E0771 (Murine TNBC)	CH3 Biosystems	Cat#94A001
LLC (Murine lung cancer)	Gift of S.I. Abrams at Roswell Park Cancer Institute	N/A
4T1 (Murine TNBC)	Michigan Cancer Foundation	N/A
67NR (Murine TNBC)	Gift from Dr. Fred Miller	N/A
2208L (Murine TNBC)	Gift of Dr. Jeffrey Rosen	N/A
4TO7 (Murine TNBC)	Barbara Ann Karmanos Cancer Center	N/A
B16F10 (Murine melanoma cancer)	ATCC	Cat#CRL-6475; RRID:CVCL_0159
Experimental models: Organisms/strains
Mouse: C57BL/6J	Jackson Laboratory	RRID:IMSR_JAX:000664
Mouse: BALB/cJ	Jackson Laboratory	RRID:IMSR_JAX:000651
Mouse: LysMcre, B6.129P2-Lyz2tm1(cre)Ifo/J	Jackson Laboratory	RRID:IMSR_JAX:004781
Mouse: NG2-CreERTM, B6.Cg-Tg(Cspg4-cre/Esr1*)BAkik/J	Jackson Laboratory	RRID:IMSR_JAX:008538
Mouse: Osx1-GFP::Cre, B6.Cg-Tg(Sp7-tTA,tetO-EGFP/cre)1Amc/J	Jackson Laboratory	RRID:IMSR_JAX:006361
Mouse: MRP8-Cre-ires/GFP, B6.Cg-Tg(S100A8-cre,-EGFP)1Ilw/J	Jackson Laboratory	RRID:IMSR_JAX:021614
Mouse: Col1a1-CreERTM, B6.Cg-Tg(Col1a1-cre/ERT2)1Crm/J	Jackson Laboratory	RRID:IMSR_JAX:016241
Mouse: Cd11c-Cre, B6.Cg-Tg(Itgax-cre)1-1Reiz/J	Jackson Laboratory	RRID:IMSR_JAX:008068
Mouse: Cx3cr1-CreERTM, B6.129P2(C)-Cx3cr1tm2.1(cre/ERT2)Jung/J	Jackson Laboratory	RRID:IMSR_JAX:020940
Mouse: CAG-GPIGFP, Tg(CAG-GFP*)1Hadj/J	Gift of Dr. Anna-Katerina Hadjantonakis	RRID:IMSR_JAX:011106
Mouse: Esr1fl/+, B6(Cg)-Esr1tm4.1Ksk/J	Gift of Dr. Kenneth Korach	RRID:IMSR_JAX:032173
Mouse: Rosa26uLIPSTIC	Jackson Laboratory	RRID:IMSR_JAX:038221
Oligonucleotides
Genotyping primer for CAG-GPIGFP. IMR 872: 5′-AAGTTCATCTGCACCACCG-3′; IMR 873: 5′-TGCTCAGGTAGTGGTTGTCG-3′;	This paper	N/A
Genotyping primer for Esr1fl/+. 5′-GACTCGCTACTGTGCCGTGTGC-3′ 5′-CTTCCCTGGCATTACCACTTCTCC T-3′	This paper	N/A
mGapdh-F: 5’-AGGTCGGTGTGAACGGATTTG-3’ mGapdh-R: 5’-TGTAGACCATGTAGTTGAGGTCA-3’	This paper	N/A
mEsr1-F: 5’-TGTGTCCAGCTACAAACCAATG-3’ mEsr1-R: 5’-CATCATGCCCACTTCGTAACA-3’	This paper	N/A
mWisp2-F: 5’-CGCTGTGATGACGGTGGTTT-3’ mWisp2-R: 5’-CCTGGCACCTGTATTCTCCTG-3’	This paper	N/A
mCtsd-F: 5’-GCTTCCGGTCTTTGACAACCT-3’ mCtsd-R: 5’-CACCAAGCATTAGTTCTCCTCC-3’	This paper	N/A
mHpgds-F: 5’-CAC TAG TTT CCT GGC TAG GGT-3’ mHpgds-R: 5’-TGT CAC AGC TCC TTT CCT TGT-3’	This paper	N/A
mRab27a-F: 5’-TCG GAT GGA GAT TAC GAT TAC CT-3’ mRab27a-R: 5’-TTT TCC CTG AAA TCA ATG CCC A-3’	This paper	N/A
mFASN-F: 5’- AGG TGG TGA TAG CCG GTA TGT −3’ mFASN-R: 5’- TGG GTA ATC CAT AGA GCC CAG −3’	This paper	N/A
FASN shRNA-1, target Sequence: GCTGGTCGTTTCTCCATTAAA	Millipore Sigma	Cat# TRCN0000075703
FASN shRNA-2, target Sequence: CGTCTATACCACTGCTTACTA	Millipore Sigma	Cat# TRCN0000075706
FASN shRNA-3, target Sequence: GCTGCGGAAACTTCAGGAAAT	Millipore Sigma	Cat# TRCN0000075707
Rab27a shRNA-1, target Sequence: GCCAGTTTAAGAGAAGTGTTT	Millipore Sigma	Cat# TRCN0000100575
Rab27a shRNA-2, target Sequence: GCTTCTGTTCGACCTGACAAA	Millipore Sigma	Cat# TRCN0000100577
Recombinant DNA
pDisplay-mgSrtA	Addgene	Cat#125792; RRID:Addgene_125792
pMD2.G	Addgene	Cat#12259; RRID:Addgene_12259
psPAX2	Addgene	Cat#12260; RRID:Addgene_12260
pHR_EGFPligand	Addgene	Cat#79129; RRID:Addgene_79129
pGreenFire 2.0 Estrogen Response Element Reporter Lentivector	System Biosciences	Cat#TR455PA-P
pinducer20	A gift from Dr. Thomas Westbrook’s lab at Baylor College of Medicine	Cat#44012; RRID:Addgene_44012
pMP71-tdTomato-Flag-mgSrtA-lag16_2 nanobody-PDGFR	This paper	N/A
pinducer20-tdTomato-Flag-mgSrtA-lag16_2 nanobody-PDGFR(SAMENT plasmid)	This paper	N/A
pinducer20-tdTomato-Flag-mgSrtA-PDGFR (SAMENT(no GFPn) plasmid)	This paper	N/A
pinducer20-Flag-SrtA-PDGFR (uLIPSTIC plasmid)	This paper	N/A
pcPPT-mPGK-attR-sLPmCherry-WPRE (sLP-mCherry plasmid)	CancerTools	Cat# 155083
SrtA-PDGFR Tm	Addgene	Cat# 121168; RRID:Addgene_121168
Software and algorithms
10x Genomics Software Cell Ranger v7.1.0	Zheng, et al.71	https://kb.10xgenomics.com/hc/en-us
Seurat v5.0.3	Hao et al.72	https://satijalab.org/seurat/
Scanpy	Wolf et al.73	https://scanpy-tutorials.readthedocs.io/en/latest/index.html
scVelo	Bergen et al.74	https://scvelo.readthedocs.io/en/stable/
velocyto	La Manno et al.75	https://velocyto.org/velocyto.py/tutorial/index.html
Dynamo	Qiu et al.76	https://dynamo-release.readthedocs.io/en/latest/
SVM-based cell type annotation	Abdelaal et al.77, Alquicira-Hernandez et al.78	https://powellgenomicslab.github.io/scPred/articles/introduction.html
Training dataset used in SVM-based cell type annotation	Baccin et al.79, Baryawno et al.80, Tikhonova et al.81	GSE122464 GSE122465 GSE134098 GSE128423 GSE108885 GSE108891
Pseudobulk analysis pipeline	Piper et al.82	https://github.com/hbctraining/scRNA-seq_online
ClusterProfiler GSEA and gene enrichment analysis	Yu et al.83	https://guangchuangyu.github.io/software/clusterProfiler/
GSVA analysis	Hanzelmann et al.84	https://alexslemonade.github.io/refinebio-examples/03-rnaseq/pathway-analysis_rnaseq_03_gsva.html
ImageJ	National Institute of Health	https://imagej.nih.gov/ij/
Graphpad Prism 10.1.2	GraphPad Software, Inc.	https://www.graphpad.com/scientificsoftware/prism/
FlowJo, v10.0	BD	https://www.flowjo.com/
Living Image v4.5.5	PerkinElmer	https://www.perkinelmer.com/
ZEN 3.4 (blue edition)	ZEISS	https://www.zeiss.com/microscopy/us/products/microscope-software.html
NRecon, v1.6.9.8	Bruker-MicroCT	https://www.microphotonics.com/micro-ct-systems/nrecon-reconstruction-software/
DataViewer, v1.5.6.2	Bruker-MicroCT	https://blue-scientific.com/bruker-micro-ct/micro-ct-software/
CT Analyzer, v1.15.4.0	Bruker-MicroCT	https://blue-scientific.com/bruker-micro-ct/micro-ct-software/
Ctvox, v3.0.0	Bruker-MicroCT	https://blue-scientific.com/bruker-micro-ct/micro-ct-software/
Scripts for scRNA-seq and customized spatial analysis	This paper	GitHub: https://github.com/xzhanglab/SAMENT-scrnaseq; Zenodo: 10.5281/zenodo.15374531
Other
Target Retrieval Solution, pH 9 (10X)	Agilent Technologies	Cat#S236784-2
Pierce BCA Protein Assay Kit	Thermo Fisher Scientific	Cat#23225
NuPAGE™ Novex™ 4-12% Bis-Tris Protein Gels	Thermo Fisher Scientific	Cat#NP0336BOX
iBlot® Transfer Stack, nitrocellulose	Thermo Fisher Scientific	Cat#IB301002
THINCERT CELL CULTURE INSERT	Greiner Bio-One	Cat#665640

Human Bone Metastasis Samples

The collection of human bone metastasis samples adhered to the Declaration of Helsinki guidelines and received approval from the Institutional Review Boards at Baylor College of Medicine (H-49396), The University of Texas MD Anderson Cancer Center (PA15–0225), and the University of Texas Medical Branch (H-46675). Written informed consent was obtained from all patients undergoing orthopedic surgery, authorizing the research use of their samples.

Cell lines

Murine triple-negative breast cancer (TNBC) lines, including PyMT-N (B6), PyMT-M (B6), 4T1 (BALB/c), 4TO7 (BALB/c), 67NR (BALB/c), and 2208L (BALB/c), along with the Mouse Lewis lung carcinoma cell line: LLC (B6) and Mouse melanoma cell line: B16F10 (B6), were cultured in DMEM/high glucose medium (Cytiva, Cat#SH30022.01) supplemented with 10% FBS (Thermo Fisher Scientific, Cat#A5256701) and antibiotics(fisher scientific, Cat#MT30004CI). Additionally, 67NR (BALB/c) cells received NEAA supplementation. The TNBC line E0771 (B6) was cultured in RPMI-1640 medium (Cytiva, Cat#SH30027.01) supplemented with 10% FBS, 1% HEPES (Cytiva, Cat#SH30237.01), and antibiotics.

METHOD DETAILS

Plasmids, Virus Production, and Infection

The SAMENT construct was generated by assembling tdTomato, an N-terminal signal peptide, Flag tag, mgSrtA, Lag16_2 GFP nanobody²⁴ and the PDGFRβ transmembrane domain using NEBuilder DNA Assembly. Three point mutations (K105E, Q108E, and L200F) were introduced into the original mgSrtA, rendering it calcium-dependent and enhancing substrate binding. The resulting SAMENT or tdTomato-SrtA-PDGFRβ sequence was inserted into the pInducer20 vector⁹⁰ for inducible expression, and all pInducer20 constructs were produced by Gateway cloning. Details of all constructs are listed in Table S1.

To produce lenti- or retrovirus, plasmids were transfected into HEK293T cells together with packaging plasmids using X-tremeGENE HP DNA Transfection Reagent. After 48 h, the viral supernatant was collected and filtered through a 0.45 μm filter. Target cells were then infected in the presence of 10 μg/ml polybrene. Flag⁺ and/or tdTomato⁺ cells were subsequently sorted to establish stable cell lines.

SAMENT in vitro labeling

SAMENT- and surGFP-expressing cells were cultured individually in DMEM medium. Inducible SAMENT cells were pretreated with 1 μg/ml Doxycycline for 12 hours before collection. Following detachment with Trypsin, the cells were washed and resuspended in 100 μL PBS (without EDTA). Subsequently, these two cell populations were mixed in a 1.5-ml microcentrifuge tube, and biotin-LPETGS was added to a final concentration of 500 μM or as specified. After co-incubation at room temperature, cells were washed three times with PBS to remove unbound biotin. The cell mixture was then incubated with Streptavidin-APC and Flag-VF450 on ice for 15 min, followed by PBS washes prior to FACS analysis.

SAMENT in vivo labeling

Mice receiving IIA or IC injections were administered doxycycline (2 mg per mouse, intraperitoneally) once daily for two consecutive days prior to euthanasia. Subsequently, biotin-LPETGS (100 μL of 20 mM solution in PBS; LifeTein) was administered intraperitoneally six times at 20-minute intervals. Organs were collected 20 minutes after the final injection.

Induction of Cre-Mediated Recombination

If not stated otherwise, female mice were used for all in vivo experiments, with age-matched Cre littermates as controls. To activate Cre-ERT2, 5-week-old female mice received daily injections of 1 mg tamoxifen (Sigma-Aldrich, T5648) dissolved in a solution of 5% ethanol and 95% corn oil (Sigma-Aldrich, C8267). This regimen was administered for 10 consecutive days. In the case of Osx1-GFP::Cre mice, expression of the EGFP-Cre fusion protein was suppressed by adding doxycycline to water bottles at a final concentration of 0.2 mg/ml, which was changed weekly until 2 weeks before the experiment.

Micro-CT Analysis of Bone Samples

The femur bones were initially prepared by removing the skin and then fixed in 70% ethanol before scanning with a Bruker SkyScan 1272 scanner (Bruker-MicroCT). The X-ray energy was set at 50 kV and 200 μA, with images captured at a pixel size of 6.6 μm. A rotation angle of 0.1° was employed for a complete 360° rotation during scanning. Trabecular bone analysis focused on a 2.0 mm distal metaphysis region, starting 0.7 mm from the proximal end of the distal femoral growth plate, with a threshold of 75–255 permille. For cortical bone analysis, a 0.5 mm region of femoral cortical bone, starting 3.7 mm from the proximal end of the distal femoral growth plate, was assessed using a threshold range of 125–255 permille. Cross-sectional images were generated using NRecon (Bruker-MicroCT, v1.6.9.8) and DataViewer (Bruker-MicroCT, v1.5.6.2), while bone volumes were analyzed via CT Analyzer (Bruker-MicroCT, v1.15.4.0). Reconstruction of three-dimensional bone images was performed using CTvox (Bruker-MicroCT, v3.0.0).

Bone Formation Rate Measurement by Calcein-alizarin red S labeling

Ten-week-old mice were intraperitoneally injected with Calcein (Sigma-Aldrich, C0875) at 20 mg/kg on day 0 and with Alizarin Red S (Thermo Scientific Chemicals, 400480250) at 40 mg/kg on day 4. Upon euthanasia on day 7, hindlimb bones were collected, fixed overnight, and snap-frozen in OCT. Non-decalcified bone sections were obtained using the CryoJane tape-transfer system on a Leica CM3050S Cryostat equipped with Low-Profile Disposable Blades DB80LX (Leica Biosystems, 14035843496). These sections were then mounted with Prolong Gold Antifade Mountant with DAPI (Invitrogen, P36935). Imaging of the distance between Calcein and Alizarin Red S labeling was conducted using an LSM880 confocal microscope, followed by analysis using ImageJ.

Intra-Iliac Artery (IIA) injection and Intra-Cardiac (IC) injection

Intra-iliac artery injections were performed following established protocols²⁶. After anesthetizing the animals and sterilizing the surgical site, a small incision (7–8 mm) was made between the fourth and fifth nipples to access the iliac vessels. Cancer cells, suspended in 100μL of PBS, were then slowly injected into the iliac artery using a 31-G insulin syringe (Becton Dickinson, 328418). Gentle pressure was applied with cotton tips to manage any bleeding before closing the wound. For intra-cardiac injection, cancer cells suspended in 100μL of PBS were directly administered into the left ventricle of anesthetized animals using a 26-gauge syringe (Becton Dickinson, 309625). The needle was carefully inserted into the heart, followed by gradual withdrawal, and pressure was applied to minimize bleeding.

Implantation of Tumors in the Mammary Fat Pad or Subcutaneously and Surgical Tumor Removal

The LLC1 cells were suspended in PBS and combined with growth factor-reduced matrigel (Corning, Cat#356231) at a 1:1 ratio. An equivalent volume of growth factor–reduced matrigel was mixed with the cells and administered via subcutaneous injection into the skin near the left hindlimb. Tumor removal surgery was conducted approximately 3 weeks after implantation to completely excise the primary tumors. Most animals survived for about 2–3 weeks after the removal of primary tumors and were subsequently dissected and examined between days 37 and 40 for lung and bone metastasis.

Administration of Fulvestrant and Immune Checkpoint Blockade (ICB) treatment

For Fulvestrant treatment, the selective estrogen receptor degrader (SERD) fulvestrant (ApexBio, #A1428) was dissolved in a solution consisting of 5% DMSO and 95% corn oil (Sigma-Aldrich, C8267). Mice were subcutaneously administered with a dose of 250 mg/kg per mouse starting from day 3 after IIA injection. This treatment was given once a week for a total duration of 3 weeks. For ICB treatment, mice received intraperitoneal injections of either anti-PD1 (200 ug per mouse, BioXcell,BE0146) or anti-IgG2a (200 ug per mouse,BioXcell,BE0089) every three days, commencing from day 1 post IIA injection. This dosing regimen was repeated for a total of six doses.

Bioluminescence Imaging and Tissue Collection

For in vivo bioluminescence imaging, mice in the IIA model were scanned using mode E of the IVIS Lumina II system (PerkinElmer) at predetermined intervals. Prior to imaging, the fur on the posterior aspect of the injected hindlimb was carefully removed to enhance the penetration of bioluminescence. Subsequently, animals received a retro-orbital injection of 100 μL of 15 mg/mL D-luciferin (Gold Biotechnology, LUCK-5G).

For ex vivo bioluminescence imaging, mice from either the IIA model or the spontaneous metastasis model were used. Dissected tissues were promptly scanned in D mode, with exposure time adjusted to avoid signal saturation. To quantify the total bioluminescence intensity of specific tissues, a standardized region of interest (ROI) was consistently applied to all animals or dissected tissues, and the total flux (photons/second) was quantified.

The dissected bones or soft tissues were immediately snap-frozen or placed in 10% neutral buffered formalin overnight. After fixation, the bone tissues were washed in PBS to remove residual formalin, followed by decalcification in a 14% pH 7.4 EDTA solution for 7 days. Subsequently, the tissues were cryopreserved in a 30% sucrose-PBS solution and then embedded in OCT (Tissue-Tek).

Single-cell Resuspension Preparation

The bone marrow cells were extracted from the tibia and femur bones using a 26-G syringe containing FACS buffer (PBS with 2% FBS, antibiotics, and 2mM EDTA). The remaining bones were then crushed thoroughly with a mortar and pestle. The fragmented bones underwent digestion in DMEM containing 1 mg/mL collagenase I (Thermo Fisher Scientific,17100017), 1 mg/mL collagenase II (Thermo Fisher Scientific,17101015), 3 mg/mL Dispase II (Sigma-Aldrich,D4693), 1 mg/mL BSA, and 0.1 mg/mL DNase I (Sigma-Aldrich, DN25) at 37°C for 1 hour on a shaker. After digestion, the bones were washed with FACS buffer. The cells released from the enzyme digestion and FACS washing were combined with bone marrow cells, filtered through a 70 mm strainer, lysed in red blood cell (RBC) lysis buffer (Cytek^® Biosciences,TNB-4300-L100) and centrifuged for subsequent analysis.

For soft tissues, they were also crushed thoroughly with a mortar and pestle, followed by digestion in the same DMEM solution. After digestion, the tissues were washed with FACS buffer, filtered through a 70 mm strainer, lysed in red blood cell (RBC) lysis buffer (Cytek^® Biosciences,TNB-4300-L100), and centrifuged for subsequent analysis.

Flow Cytometry

For the pooled bone marrow cells, staining was initiated with Ghost Dye UV450 (Cytek^® Biosciences, 13–0868) in 1 mL PBS for 30 minutes. Following centrifugation, the cells were resuspended in 1 mL FACS buffer. Subsequently, samples were divided for staining with various panels and pre-treated with anti-CD16/32 antibody (Cytek^® Biosciences, 40–0161) for 10 minutes on ice to block non-specific binding. This was followed by incubation with fluorescent dye–conjugated primary antibodies on ice for 15 minutes. The antibodies used in this study included: Immune cell panel, CD45-VF450 (Cytek^® Biosciences,75–0451), CD11b-APC/Cy7 (Cytek^® Biosciences, 25–0112), Ly6G-Percp/Cy5.5 (Cytek^® Biosciences, 65–1276), Ly6C-PE/Cy7 (BioLegend,128018), F4/80-BV510 (BioLegend,123135), B220- APC/Cy7 (Cytek^® Biosciences, 25–0452), CD3e-Percp/Cy5.5 (Cytek^® Biosciences,65–0031), CD4-PE/Cy7 (Cytek^® Biosciences,60–0041), CD8a-BV711 (BD Biosciences,752634), PD-1-BV605 (BioLegend,135219), Biotin-APC (Miltenyi Biotec,130–113–288) or Biotin-PE(Miltenyi Biotec,130–113–291); Stromal cell panel, CD45-BV605 (BioLegend,103140), Ter119-BV605 (BioLegend,116239), CD31-APC/Cy7 (BioLegend,102440), Sca-1-Percp/Cy5.5 (eBioscience,45–5981–82), CD51-BV421 (BD Biosciences, 740062), CD140α-PE/Cy7 (BioLegend, 135912), Biotin-APC (Miltenyi Biotec,130–113–288) or Biotin-PE(Miltenyi Biotec,130–113–291)

Flow cytometry analysis was conducted using a BD LSRFortessa flow cytometer and data were further analyzed with FlowJo software. The gating strategy for each population was as follows: Monocytes: DAPI−CD45+CD11b+Ly6C+Ly6G−; Neutrophils: DAPI−CD45+CD11b+Ly6C^mid-lowLy6G+; Macrophages: DAPI−CD45+CD11b+Ly6C−Ly6G−F4/80+; B cells: DAPI−CD45+B220+; T cells: DAPI−CD45+CD3e+; CD4 T cells: DAPI−CD45+CD3e+CD4+; CD8a T cells: DAPI−CD45+CD3e+CD8a+; Endothelial cells: CD45- Ter119-CD31+; Osteoprogenitors: CD45-Ter119-CD31-CD51+CD140a+; Mature osteoblasts: CD45-Ter119-CD31-CD51+Sca1-; MSCs: CD45 Ter119-CD31- CD140a+;

Single-cell RNA Library Construction and Sequencing

Mice received SAMENT+ LLC1 cell injections either via intracardiac or intra-iliac routes. Following this, the previously described SAMENT in vivo labeling procedure was conducted before euthanization. Afterwards, tumor tissues from mice, encompassing lungs, livers, brains, and hindlimbs, were collected, and single-cell suspensions were prepared as previously outlined. Prior to sorting, debris removal was undertaken to minimize excessive cell debris contamination. Subsequently, cells were stained with Ghost Dye UV450 and biotin flow antibody. The tissue cells were then sorted into biotin⁺ and biotin⁻ populations using Arial. To optimize cost-effectiveness, we implemented a cell hashing strategy during sample processing and sequencing. Two distinct hashing methods were employed for two separate batches of single-cell experiments: CellPlex by 10X Genomics and TotalSeq C by BioLegend. Both approaches followed the Chromium Next GEM Single Cell 3ʹ v3.1 (10x Genomics): Cell Multiplexing protocol (CG000383) and were carried out in collaboration with the Single Cell Genomics Core at BCM. The resulting libraries, containing barcoded single-cell transcriptomes, were subjected to sequencing to a depth of 600 million reads using the Novoseq 6000 system (Illumina) at the Genomic and RNA Profiling Core (GARP) and Novogene. Data processing was performed using the CASAVA 1.8.1 pipeline (Illumina). Sequence reads were converted into FASTQ files, cell multiplexing oligo (CMO), antibody barcoding matrices, and unique molecular identifier (UMI) read counts using the CellRanger software from 10X Genomics.

Immunofluorescence Staining and quantification

The tissues, either paraffin-embedded or OCT-embedded, underwent sectioning with assistance from the Breast Center Pathology Core at Baylor College of Medicine. Paraffin-embedded slides were baked at 55°C for 2 hours, dewaxed, and rehydrated following standard protocols. Antigen retrieval was performed using Dako Retrieval Solution pH 9.0 (Agilent Technologies, S2367) in a pressure cooker at 115°C for 10 minutes. For frozen sections, slides were thawed at room temperature for 10 minutes and rinsed with PBS. All slides received treatment with 0.1M NH4Cl in PBS for 10 minutes to minimize autofluorescence and were subsequently blocked in 10% donkey serum in 0.4% Triton X-100 PBS for 1 hour at room temperature. In cases where mouse primary antibodies were used on mouse tissues, M.O.M. Blocking Reagent (Vector Laboratories, MKB-2213–1) was employed for additional blocking. The slides were then incubated with primary antibodies overnight at 4°C in 0.2% Triton X-100 PBS. Following three washes with PBS, slides were incubated with secondary antibodies for 1 hour at room temperature. Hoechst 33342(Thermo Fisher Scientific, 62249) was used for nuclear staining. After washing, slides were mounted with ProLongTM Gold antifade mountant(Invitrogen, P36934). Imaging was conducted using a LSM780 or LSM880 (Zeiss) confocal microscope, Cytation 5 Epi-fluorescence microscope, and Zeiss Axioscan.Z1 scanner. Image analysis was performed using ZEN (Zeiss) or ImageJ. Antibodies used for immunofluorescent staining are listed in key resources table.

TRAP⁺ osteoclasts were stained using the Acid Phosphatase, Leukocyte (TRAP) Kit according to the manufacturer’s instructions. CTSK⁺ osteoclasts were stained on paraffin-embedded sections following the above method. Images were acquired using a Zeiss Axioscan.Z1 scanner and exported as TIFF files via ZEN software. Bright-field TRAP images were processed in ImageJ using the Colour Deconvolution2 plugin and converted to 8-bit format. For CTSK fluorescence images, channels were separated, and the red channel (CTSK⁺ cells) was selected. Positive signals were identified using the “Threshold” function, refined with the “Mask,” “Erode,” and “Despeckle” tools to separate overlapping objects, and quantified using the “Analyze Particles” feature to obtain positive cell counts.

In vitro macrophage differentiation and co-culture assay

Murine hindlimb bone marrow cells were isolated following the protocol outlined in a previous study⁹¹. Subsequently, the cells were cultured in DMEM medium for 5 hours to eliminate stromal cells. After this incubation period, unattached bone marrow cells were collected and reseeded into a new plate containing DMEM culture medium. These cells were then infected with the ERE-GFP-luciferase virus for 2 days. Following viral transduction, the bone marrow cells were differentiated into macrophages by the addition of 25 ng/ml recombinant murine M-CSF (Peprotech,315–02) for 3 days. Estrogen-stripped DMEM, consisting of phenol red-free DMEM (Thermo Fisher Scientific,21063029), 10% charcoal-stripped FBS (Sigma-Aldrich,F6765) and antibiotics, was used to culture for an additional 4 days, with the medium being changed every 2 days. Subsequently, 600,000 ERE-macrophages were cultured with or without 25,000 various types of cancer cells in a 12-well plate for 2–3 days. In the non-contact co--culture assay, we initially seeded 600,000 ERE-macrophages in the lower wells of a 12-well plate and incubated them for 30 minutes to promote cell attachment. Following this, 25,000 LLC1 cancer cells suspended in 200 μl of medium were added to the upper culture inserts (0.4-μm pore size, Greiner Bio-One,665640). The luminescence signal of the ERE-reporter was detected using an IVIS Lumina II (PerkinElmer) at the indicated time points, while the fluorescence signal of the ERE-reporter was acquired using an LSM880 confocal microscope.

In vivo reconstitution of ERE-macrophages

Mice were injected with LLC1 cells via intra-iliac artery (IIA) injection to induce bone metastatic tumors for a duration of 2 weeks. Following this, in vivo macrophage depletion was achieved by administering 200 μl clodronate-containing liposomes (Encapsula NanoSciences,CLD-8901) via retro-orbital injection for two consecutive days. Two days post-clodronate-liposome treatment, approximately 1,500,000 bone marrow-derived ERE reporter-macrophages, as described above, were delivered into the mice via retro-orbital injection to reconstitute the macrophage-depleted mouse model. Four days thereafter, hindlimb bones were harvested for sectioning, and immunofluorescence staining was performed on the bone section slides. Imaging data were collected using an LSM880 confocal microscope.

Conditioned medium (CM) collection

Cancer cells were cultured in DMEM supplemented with 10% FBS until reaching 80–90% confluence. After washing 2–3 times with PBS, cells were incubated for 24 h in phenol red– and FBS-free DMEM to reduce background interference. The conditioned medium (CM) was collected, centrifuged at 500 × g for 5 min to remove debris, filtered through a 0.22 μm membrane, aliquoted, and stored at –80 °C. Before use, charcoal-stripped FBS (matching the serum concentration of the control medium) was added back to the CM.

mRNA Extraction and qRT-PCR

RNA extraction was performed using the Direct-zol RNA Miniprep Kit (Zymo Research, R2052) according to the manufacturer’s instructions. Subsequently, cDNA was synthesized from total RNA using the RevertAid First Strand cDNA Synthesis Kit (Thermo Fisher Scientific, K1622). For real-time PCR, PowerUp^™ SYBR^™ Green Master Mix (Thermo Fisher Scientific, A25741) was utilized on a CFX96 Real-Time PCR system (Biorad). The expression levels of GAPDH mRNA were employed as the internal control. All of the primer sequences provided in key resource table.

Protein Extraction and Western Blotting

For western blotting, bone marrow-derived macrophages were detached from the culture dish and lysed using RIPA buffer (10mM Tris-HCl, pH 8.0, 1mM EDTA, 0.5mM EGTA, 1% Triton X-100, 0.1% Sodium Deoxycholate, 0.1% SDS, 140mM NaCl). Protein concentration was determined by BCA assay (Pierce BCA Protein Assay Kit, Thermo Fisher Scientific, 23225). 25ng total protein were loaded onto Invitrogen NuPage^® Novex^® Gel (Thermo Fisher Scientific, NP0336BOX) for electrophoresis. Subsequently, proteins were transferred from the gel to the nitrocellulose membrane (Thermo Fisher Scientific, IB301002) using the iBlot^™ Dry Blotting device (Invitrogen). Following a 1-hour blocking step with 5% milk, the membranes were incubated with anti- ERα(D8H8), anti-ERβ, anti-GAPDH and anti-α Tubulin antibodies overnight at 4°C with primary antibodies. The next day, membranes were probed with secondary HRP antibodies(Goat anti-Mouse IgG, Invitrogen, 31430)(Goat anti-Rabbit IgG, Invitrogen, 31460) and scanned using the Amersham Imager 600 system.

Data Preprocessing, Filtering, and Batch Correction

We utilized the cellranger pipeline for processing the multiplexed raw data. In each dataset and for every patient, we generated multiplexing and count assays based on CellRanger outputs, with all subsequent steps performed using Seurat packages (versions 4.0 and 5.0)⁹². Specifically, sequencing files from TotalSeqC were demultiplexed through Seurat pipeline based on README table uploaded together with data.

Initially, within the Seurat framework, we conducted hashtag demultiplexing. To ensure data quality, we implemented a two-step doublet filtering process. Initially, we filtered cells based on demultiplexing results and hashtag assignments, retaining only the singlets (cells with a single barcode label). Subsequently, cells with fewer than 200 detected genes or displaying a deviation exceeding two-fold from the median gene count were excluded. Moreover, cells with more than 10% mitochondrial genes were systematically filtered out to minimize mitochondrial gene contamination.

We proceeded to normalize each individual dataset using the default parameters of the NormalizeData function. After normalization, the FindVariableFeatures function was employed with its default settings to identify variable genes within each normalized dataset. For data integration, we utilized the SelectIntegrationFeatures function, selecting the top 2,000 variable genes.

To identify anchors across all datasets, we compared different methods, including “rpca,” “cca,” and “harmony.” The “rpca” method was chosen due to its superior speed and efficiency⁹³. These anchors were subsequently employed to integrate matrices using the IntegrateData function, specifying the parameter dims as 1:50.

Unsupervised Clustering and Cell Type Annotation

Within the integrated assay, we conducted scaling and PCA analysis. To visualize the data, we employed Uniform Manifold Approximation and Projection (UMAP) based on the top 50 principal components. To explore transcriptomic diversity, we utilized the FindClusters function with a range of resolutions from 0.2 to 5. We opted to label cell types using high-resolution clustering results to enhance annotation accuracy. We utilized machine learning support vector machines (SVM)^77–78. We trained the SVM model using a well-annotated single-cell dataset integrated from various publicly available sources (GSE122464, GSE122465, GSE134098⁷⁹, GSE128423⁸⁰, GSE108885, GSE108891, GSE118436⁸¹, which collectively comprised over 500,000 cells from mice, covering more than 20 cell types and states. After predicting cell types in our dataset using the trained references, we conducted manual validation by assessing the expression of classical cell markers⁹⁴.

Pseudo-bulk Transcriptomic Comparisons and Pathway Analysis

We employed pseudo-bulk analysis to determine differentially expressed genes (DEGs) with the goal of assessing cell type-specific expression disparities at the population level. The process encompassed several key steps: Firstly, we selected cells of interest corresponding to the desired cell type(s) as a prerequisite for the subsequent DE analysis. Subsequently, raw counts were obtained through quality control (QC) filtering applied to the designated cells earmarked for DE analysis. These counts, along with associated metadata, were integrated at the sample level. During the DE analysis, we used DESeq2 packages⁹⁵ in R and the results were cross-validated through limma voom method. Following the identification of DEGs, we conducted Gene Set Enrichment Analysis (GSEA) using the ClusterProfiler R package⁸³. The pathway dataset employed for this analysis was sourced from MSigDB (https://www.gsea-msigdb.org/gsea/msigdb/). To assess pathway activity across different cell types and biotin statuses, we utilized the AUCell R package⁹⁶ to compute pathway activity scores. Visualization of these scores was achieved through violin plots using the AddModuleScore function. Significance statistics were calculated using the Wilcoxon rank-sum test.

Single-Cell Trajectory Analysis

We built a single-cell differentiation trajectory for macrophages across different biotin statuses using the scVelo tool⁷⁴. The initial determination of macrophage subclusters was accomplished through an unsupervised method within Seurat. These subclusters exhibited distinct tissue contributions, suggesting potential variations in their differentiation trajectories. Notably, we already possessed prior knowledge indicating that macrophage cluster 5 (c5) represented osteoclasts, we are also aware that osteoclasts are differentiated from macrophages. Consequently, our trajectory inference primarily focused on these subclusters within the macrophage population.

Customized spatial immunofluorescence quantification pipeline

Our analysis followed a multi-step pipeline: parsing per-pixel immunofluorescence text images and composing channels; DAPI-based nuclei segmentation; marker assignment and cell typing; nearest-neighbor (centroid-to-centroid) distance pairing to tumor cells; tumor-region boundary extraction; niche-labeling model sensitivity; and signed-distance/band-based area-normalized density quantification. We used outputs selectively for different aims: for example, 50 μm band–normalized densities to evaluate potential nonspecific labeling for isolated lesions, and distance analyses to interrogate diffusive lesions and characterize spatial dispersion and tumor–host interactions.

1). Image export, parsing, and multi-channel compositing.

Per-pixel intensity rasters for three immunofluorescence channels (DAPI, tumor, or one of mCherry/Biotin/ERE-GFP/F4/80/ERα signal channels) were exported from Fiji as “Text Image” format (CSV file). Files were read with Pandas and reshaped into a long table of pixel coordinates and intensities; for each file, we inferred the channel from filename tokens and an image_id stem after stripping channel tokens, then unified field names with a rename map to ensure channels overlay. Pixel coordinates were converted to physical units using the calibrated pixel size derived from the field of view. For visualization only, per-channel images were clipped and linearly rescaled to [0,1] using the 0th–99.8th intensity percentiles and rendered either as split panels or as an RGB overlay mapping. Panels were saved at high resolution, and axes were labeled in micrometers with the image origin preserved.

2). Nuclei segmentation from DAPI.

Nucleus segmentation was performed on the DAPI channel using a marker-controlled watershed tuned for aggressive splitting of touching nuclei⁹⁷. The DAPI image was robustly normalized by linear rescaling between the 1st and 99.8th percentiles, background-suppressed isotropic structuring-element radius, and lightly smoothed with a Gaussian. Foreground was obtained by adaptive (local) thresholding with a 25μm odd window, followed by removal of small objects and filling of small holes. To split clusters, we computed a Euclidean distance transform inside the mask, slightly smoothed it, and detected seeds with peak_local_max, enforcing min seed spacing=0.1μm and an h-maxima threshold. Watershed⁹⁸ was then applied on the Sobel gradient of the smoothed image. As a post-processing step, regions with area ≥ 0.5× the median nuclear area were locally re-watershedded with the same seeding/energy to further separate residual mergers. The final label image was relabeled consecutively, and per-nucleus morphometrics (area, perimeter, eccentricity, solidity, centroid, bounding box, mean/max DAPI intensity) were exported alongside nuclei boundaries (pixel and μm coordinates).

3). Channel assignment to nuclei and cell-type calling.

To attribute tumor cell signals and Biotin/ERE-GFP/F4/80/ERα signal to individual nuclei (and a narrow perinuclear territory) we combined threshold marker masks with a nucleus-centric geometric partition. First, we computed a Voronoi band^99,100 outside the nuclear labels with width R_VORONOI = 3.0 μm and a thicker buffer zone obtained by expanding labels by R_BUFFER = 2.0 μm and extracting thick inter-label boundaries. Each marker image was preprocessed (for thresholding and voting) by percentile normalization (1–99.8%), white tophat (radius 1.0 μm), and Gaussian smoothing. Global Otsu (THRESH_MODE=‘global_otsu’) binarization identified positive pixels. Assignment proceeded in three passes: (i) inside labels: positive pixels inherit their nucleus label; (ii) Voronoi-only (non-buffer): positive pixels adopt the nearest nucleus (from an outside-distance nearest-label map); (iii) buffer zone: for each positive pixel we identified candidate neighboring nuclei within R_BUFFER and selected the winner via a weighted vote that sums normalized marker intensities inside a vote disk of R_VOTE = 3.0 μm restricted to pixels currently “owned” by the candidate nucleus; ties fall back to the nearest nucleus. This voting was Numba-accelerated when available. For each nucleus and each marker, we tabulated positive-pixel counts and sum intensity; a nucleus was called “tumor” if RFP/GFP_pos_pix ≥ 1, else label-positive if Biotin/ERE-GFP/F4/80/ERα_pos_pix ≥ 1, else negative. A filled cell-type mask was then built by taking, for each label, the union of the nucleus and any assigned positive pixels, filling holes; for negatives we also included a thin perinuclear rim (owner-map within R_BUFFER) to approximate cytoplasm.

4). Distances from target cells to tumor (cell–cell pairing).

For target cells (e.g., Biotin+/ERE-GFP +/F4/80+/ERα+ cell we computed a centroid-to-centroid Euclidean distance in micrometers to the nearest tumor cell. Centroids used the physical calibration. A cKDTree was built on tumor centroids, and for each target centroid the nearest neighbor (k = 1) was queried to obtain the matched tumor label and the distance in μm. The resulting table (target label and centroid, nearest-tumor label and centroid, pairwise distance) was exported. These distances were also used to define interaction pairs operationally as “nearest-neighbor links” (visualized as yellow straight line segments connecting target and tumor centroids); quantitative analyses rely on the distance values, not on the rendered lines.

5). Tumor region and boundary extraction.

The tumor region mask was defined from the cell-type raster by selecting pixels labeled tumor in the filled cell-type mask and performing binary hole filling to obtain a coherent region. The tumor boundary was then extracted as the 0.5 isocontour of this binary mask using skimage.measure.find_contours¹⁰¹, yielding one or more polylines which were converted from pixel coordinates to micrometers. This vector boundary served as the reference curve for all subsequent signed distance and band-based spatial analyses and for plotting overlays; no additional geometric smoothing was applied beyond the raster-to-contour extraction.

6). Labeling sensitivity measurements through identification of false-negative cells and flagging.

We focused on false-negative cells—cells called negative by marker assignment yet located in close proximity to tumor. Using the cKDTree¹⁰² nearest-tumor distances described above, each negative cell was labeled false negative if its nearest tumor centroid lay within 15.0 μm of its centroid; otherwise, it remained a true negative. For transparency, the method also records the matched nearest tumor label and the exact nearest-neighbor distance (μm) for each negative cell. Only this false-negative definition and output were used and reported for downstream analyses in this manuscript (the separate false-positive logic was computed in code but is not described or interpreted here).

7). Area-normalized densities in concentric 50 μm bands (and robust distance-bin curves).

To quantify how labeled niche cells distribute relative to the tumor edge, we constructed symmetric 50 μm spatial bands around the tumor boundary using geometric buffering: for consecutive edges [a₀, a₁) at BAND_WIDTH = 50 μm, we computed buffer (a₁) − buffer(a₀) of the boundary polyline and clipped to the field of view; each band piece was labeled tumor adjacent region (outside) by testing a representative point against the tumor mask. Band indices were signed, and for each (region, band) we accumulated area (μm² to mm²) and counts of labeled niceh nucleus centroids falling within the band (point-in-polygon with prepared geometries)¹⁰¹. Densities were reported as cells per mm²: count divided by measured band area (per region, per band). In addition, for a smoothing-free and topology-robust summary, we computed a signed Euclidean distance map in micrometers from the tumor, binned space into 10 μm slabs across the signed axis and derived per-bin areas directly from the distance histogram. Cell centroids were plotted in the same bins to yield cells/mm² vs. distance curves.

Quantification and Statistical Analysis

The number of animals or independent replicates, along with the statistical tests used, is specified in the figure legends or figures. Data quantification utilized GraphPad v10.1.2, with results presented as Mean ± SEM in bar and curve plots. Parametric statistics were applied to normally distributed data, while nonparametric statistics were utilized for comparing metastatic burden, such as bioluminescent intensities and derived data. A significance level of p < 0.05 was used to determine statistical significance. For non-quantitative data, including images from immunofluorescent staining and flow cytometry, findings were reproduced at least three times by replicating the experiments and/or using at least three biologically independent samples. Representative results are shown.

Supplementary Material

Figure S1

Supplementary Figure 1. Validation and optimization of SAMENT in vitro, as related to Figure 1.

(A–C) Schematic and flow cytometry analysis of biotinylation in surGFP and intraGFP receiver cells. (A) Experimental setup of co-culture surGFP and intraGFP receiver cells with SAMENT-expressing cancer cells. (B) Representative flow cytometry plots showing biotinylation in intraGFP (top) and surGFP (bottom) cells. (C) Quantification of biotin⁺ percentage in surGFP and intraGFP cells (left) and biotin⁺ SAMENT cells among total SAMENT population (right) (n = 4). surGFP cells show enhanced labeling, likely reflecting increased intercellular interactions.

(D and E) Time course of biotinylation of surGFP cells co-cultured 1:1 with SAMENT-expressing cancer cells in the presence of 500 μM biotin-LPETGS (D, representative flow cytometry; E, quantification).

(F and G) Biotinylation of surGFP cells at varying biotin-LPETGS concentrations (F, representative flow cytometry; G, quantification).

(H) Biotinylation of surGFP cells by tet-on inducible SAMENT-expressing cancer cells at varying doxycycline concentrations; NC, parental LLC1 cells; SAMENT, SAMENT-expressing LLC1 cells.

(I and J) Experimental setup (I) and persistence of biotinylation (J) on surGFP⁺U937 cells after overnight labeling followed by transfer to non-labeling conditions. Percentage of biotin⁺ cells was measured over time.

(K) Flow cytometry quantification of each cell type within biotin⁺ versus biotin⁻ bone marrow populations from tumor-bearing mice with bone metastases induced by IIA injection of different cancer cell lines in the presence or absence of doxycycline: LLC1 (–Dox, n = 5; +Dox, n = 5), 4T1 (–Dox, n = 5; +Dox, n = 6), PyMT-N (–Dox, n = 8; +Dox, n = 6), B16F10 (–Dox, n = 7; +Dox, n = 6).

intraGFP: intracellular GFP; surGFP: surface GFP.

All data are shown as mean ± SEM. N indicates biological replicates. P values were assessed by two-tailed unpaired Student’s t-test in (C, K).

Figure S3

Supplementary Figure 3. Single-cell transcriptomic variations between biotin⁺ and biotin⁻ cell types across different metastatic sites, as related to Figure 3.

(A) Flow cytometry quantification of each cell type within biotin⁺ versus biotin⁻ tissue populations across organ metastases in GFP-GPI mice following IC injection of LLC1-SAMENT cells in the presence or absence of doxycycline. (Bone: –Dox, n = 9; +Dox, n = 7; Brain: –Dox, n = 6; +Dox, n = 6; Liver: –Dox, n = 5; +Dox, n = 7; Lung: –Dox, n = 6; +Dox, n = 7). P values were assessed by two-tailed unpaired Student’s t-test.

(B) UMAP visualization of the SAMENT single-cell dataset, colored by cell type, colonized tissue, and biotin-labeling group (left). Data include scRNA-seq profiles from bone (n = 10, 44036 cells), brain (n = 4, 8956 cells), liver (n = 4, 11186 cells), and lung (n = 4, 17518 cells). Cell signature gene expressions were shown in dotplot (right).

(C) GSVA analysis comparing biotin⁺ and biotin⁻ NK cells highlights regulatory effects of NK cells on macrophages. Biotin⁺ NK cells appear to modulate the tumor microenvironment by enhancing chemotaxis and inhibiting aging and apoptosis in macrophages.

(D-G) Tissue-specific top enriched pathways in biotin⁺ immune subsets compared with their biotin⁻ counterparts. (D) Biotin⁺ immature B cells in brain. (E) Biotin⁺ monocyte precursors in lung. (F) Biotin⁺ neutrophil precursors in lung. (G) Biotin⁺ GMPs in bone.

(H) Validation of published tissue-specific macrophage signatures in the SAMENT dataset. Reported microglia markers⁸⁵ include Tmem119, Trem2, Fcrls, Slc2a5, P2ry12, and Cx3cr1; alveolar macrophage markers⁸⁶ include Cd163⁻, Cd11b (Itgam)⁻, Cd206 (Mrc1)^int, Spp1⁺, SiglecF⁺, and Cd11c (Itgax)⁺; osteoclast markers⁵² include Trap1, Ctsk, Acp5, Oscar, Arg1, Spp1, Mmp9, and Atp6v0d2; Kupffer cell markers⁸⁷ include Cd11b (Itgam)^low, F4/80 (Adgre1)^high, and Clec4f⁺; and monocyte-derived macrophage markers^87,88 include Cd11b (Itgam), F4/80 (Adgre1), Ly6c (Ly6c1), Cx3cr1, Csf1r, and C5ar1.

(I) Feature plots showing the expression of additional tissue-specific macrophage markers across the indicated populations.

Figure S2

Supplementary Figure 2. Additional Comparative Data of Labeling Systems, as related to Figure 2.

(A) Representative fluorescence staining of ROIs from multiple fields of view for each niche-labeling system, corresponding to the analysis shown in Figure 2D. Left: ROIs showing isolated bone metastatic lesions with clearly defined tumor boundaries and adjacent regions. Middle: Cell type annotations for each ROI. Right: Classification of labeled niche cells within intratumoral and tumor-adjacent regions. Tumor-adjacent areas were segmented into 50-μm concentric bands extending outward from the tumor edge, and labeled niche cell populations were normalized to the area of each band and visualized as cell density distributions.

(B) Representative fluorescence staining of ROIs from multiple fields of view for each niche-labeling system, corresponding to the analysis shown in Figure 2E. Left: ROIs showing diffusive bone lesions lacking clearly defined tumor boundaries. Right: Corresponding cell type annotations and nearest tumor–niche cell pairs, indicated by yellow dashed lines. (Distance definition: the distance between the centroids of a labeled niche cell nucleus and its nearest tumor cell nucleus).

(C) Labeling efficiency analysis for diffusive bone lesions. Representative ROIs (top) from SAMENT and uLIPSTIC samples show niche cells in contact with tumor cells but lacking biotin labeling (yellow circles). The corresponding adjacent labeled cell proportions quantified from multiple ROIs are summarized in the bottom scatter plot (SAMENT, n = 12; uLIPSTIC, n = 14).

(D and E) Comparison of niche-labeling system controllability. Ex vivo (D) and in vivo (E), CD45⁺ bone marrow cell labeling (top) and fraction of tumor cells expressing SAMENT, uLIPSTIC, and sLP-mCherry (bottom). SAMENT and uLIPSTIC do not label cells in the absence of doxycycline and biotin–LPETGS, whereas sLP-mCherry labels cells autonomously, without tight controllability, and its labeling efficiency increases with tumor cell number. Sample sizes: ex vivo, n = 4 per group; in vivo, n ≥ 3 per group and per time point.

(F) Representative immunofluorescence images illustrate nonspecific or leaked labeling in the three systems. Top: SAMENT and uLIPSTIC do not label without doxycyclin and biotin–LPETGS, despite the successful induction of SAMENT and uLIPLSTIC machineries (red color). Bottom: sLP-mCherry exhibited progressively labeling surrounding niche cells as tumors develop (green: tumor cells, red: labeled cells by secreted sLP-mCherry).

All data are shown as mean ± SEM. Scale bar, 100 μm. P values were assessed by two-tailed unpaired t-test in (C), by ordinary one-way ANOVA followed by least significant difference (LSD) test in (D, E).

Figure S4

Supplementary Figure 4. ERα is selectively activated in metastatic niche macrophages, as related to Figure 4.

(A) Flow cytometry quantification of each cell type among biotin⁺ and biotin⁻ populations at different stages of bone metastasis in GFP-GPI mice bearing LLC1-SAMENT tumors. Control, cancer cells without doxycycline (SAMENT not expressed); Labeled, cancer cells treated with doxycycline (SAMENT-mediated labeling active). (Early: control, n=5; labeled, n=5; Intermediate: control, n=4; labeled, n=5; Late: control, n=4; labeled, n=5).

(B) Boxplot showing GOBP_Response_to_Estrogen pathway activity. Average pathway scores per mouse were derived from scRNA-seq data and compared between biotin⁺ and biotin⁻ groups, grouped by cell type and ranked left to right by ascending p values. ncMo: non-classical monocyte.

(C) Heatmap showing top five differentially expressed genes for each myeloid cell type per biotin group. Only macrophages display distinct expression differences between biotin⁺ and biotin⁻ groups (outlined with green dotted circles). Ctsd and Hpgds, two estrogen-regulated genes, are highlighted in red.

(D–F) Esr1 RNA levels (D); ERα protein levels (E); RNA levels of estrogen receptor target genes (F), assessed in sorted biotin⁺ and biotin⁻ macrophages from LLC1-SAMENT bone metastases.

(G–I) Bone marrow–derived macrophages expressing the ERE-Luc-T2A-GFP reporter were transplanted into macrophage-depleted WT mice. (G) Representative immunofluorescence images show intra-tumoral and extra-tumoral regions, with an algorithm identifying ERE-reporter macrophages. (H) Quantification includes the percentage of ERE^high macrophages among total cells (left) and total macrophages (right), and (I) the distance between ERE^high and ERE^low macrophages to their nearest tumor cells. Scale bar, 50 μm.

(J–L) Immunofluorescence staining and quantification of ERα⁺ macrophages in the similar transplantation model as (G-I). (J) Representative immunofluorescence images show intra-tumoral and extra-tumoral regions, with an algorithm identifying ERα⁺ macrophages. (K) Quantification includes the percentage of ERα⁺ macrophages among total cells (left) and total macrophages (right), and (L) the distance between ERα⁺ and ERα⁻ macrophages to their nearest tumor cells. Scale bar, 50 μm.

(M) Representative immunofluorescence images showing co-localization of ERα (white) and CD68 (red, human macrophage marker) in bone metastasis samples from patients with diverse primary tumors, except for adenocarcinoma of the rectum/colon, which represents a primary lesion. CD68⁺ macrophages in the colon primary tumor (rightmost panel) display minimal ERα expression. Scale bar, 50 μm. Cervical SCC, Cervical Squamous Cell Carcinoma.

All data are shown as mean ± SEM. N indicates biological replicates. P values were assessed by two-tailed unpaired Student’s t-test in (H, K), by two-tailed paired t-test in (B, D, F, I, L).

Figure S5

Supplementary Figure 5. Depletion of Esr1 in monocytes and macrophages does not regulate bone formation or remodeling, as related to Figure 5

(A) Heatmap showing pairwise Spearman correlations among HALLMARK pathway activities in macrophages across all samples from the scRNA-seq dataset. Color indicates correlation strength, and asterisks denote adjusted P values (Benjamini–Hochberg correction). Estrogen-related pathways are labeled in red, and lipid-related pathways in green. Left: biotin⁺ populations. Right: biotin⁻ populations.

(B)Western blot of ERα, ERβ, and α-Tubulin in BMDMs from Esr1^fl/fl and Esr1^LysM mice.

(C) GSVA of CD8⁺ T cells comparing Esr1^fl/fl (n=12) and Esr1^LysM (n=12) samples from the scRNA-seq dataset. Representative CD8⁺ T-cell stimulatory and tumor-targeting pathways are shown. Samples (columns) are hierarchically clustered based on GSVA scores.

(D) Heatmap showing estrogen receptor transactivation in BMDM-ERE reporter cells co-cultured with various cancer cell lines, quantified by luciferase reporter activity. Luciferase signals were normalized to macrophage monoculture (NC) and converted to Z-scores. Three biological replicates per group are shown.

(E–G) Bioluminescence and ERα protein/Esr1 RNA levels in RAW264.7 cells treated with CM from cancer cells or normal medium (E-F) or with BSA-palmitate complex or BSA control (G).

(H-J) Knockdown efficiency of FASN in LLC1 cells at protein and RNA levels (H) and corresponding effects on bioluminescence (I) and ERα protein/Esr1 RNA expression (J) in RAW264.7 cells treated with CM from LLC1 cancer cells after FASN knockdown or scramble shRNA.

(K-M) Knockdown efficiency of RAB27a in LLC1 cells at protein and RNA levels (K) and corresponding effects on bioluminescence (L) and ERα protein/Esr1 RNA expression (M) in RAW264.7 cells treated with CM from LLC1 cancer cells after RAB27a knockdown or scramble shRNA.

(N) UMAP visualization of the human bone metastasis scRNA-seq dataset, showing dimensional reduction colored by cell type.

(O) UMAP showing unbiased subclustering of all myeloid cells from the SAMENT scRNA-seq dataset (left) and projection of human OC signature scores (derived from N) onto the UMAP (right).

(P) Estrogen responsive pathway activity scored across the myeloid subclusters (O). Gene set combines both “HALLMARK Estrogen Response Early” and “HALLMARK Estrogen Response Late”.

(Q) Trajectory analysis of myeloid lineages, separated by biotin-labeling groups. The least-action path (LAP; color gradient with directional arrows) illustrates that osteoclast precursor formation (Mφ_c5) proceeds similarly in both biotin groups.

(R) Representative TRAP-stained sections (left) confirming osteoclasts and immunofluorescence images (right) showing ERα colocalization with RANK⁺ (top) or CTSK⁺ (bottom) osteoclasts in bone metastases.

(S and T) TRAP⁺ (S) and CTSK⁺ (T) osteoclasts in 8-week-old tumor-free mice. Representative images (left) and quantification (right). Scale bar, 100 μm.

(U) Representative immunofluorescence images depicting ERα expression in tumor-adjacent macrophages (F4/80⁺) within bone metastatic lesions from Esr1^fl/fl and Esr1^LysM mice.

(V) Representative confocal images (left) and quantification (right) of new bone formation in Esr1^fl/fl(n=3 mice) and Esr1^LysM naïve mice (n=3 mice). Four independent images per mouse were analyzed for quantification.

(W and X) Distal femoral trabecular bone of 8-week-old female Esr1^fl/fl and Esr1^LysM naïve mice. (W) Representative 3D μ-CT images. (X) Quantification. (n=5 per group).

(Y and Z) Midshaft femoral cortical bone of 8-week-old female Esr1^fl/fl and Esr1^LysM naïve mice. (Y) Representative 3D μ-CT images. (Z) Quantification. (n=5 per group).

TRAP: tartrate resistant acid phosphatase; CTSK: Cathepsin K; Tb: Trabecular bone; BS: bone surface; BV: bone volume; TV: tissue volume; Tb.N: Trabecular number; Tb.Th: Trabecular thickness; Ct: Cortical bone.

All data are shown as mean ± SEM. N indicates biological replicates. P values were assessed by two-tailed unpaired Student’s t-test in (E, F, S, T, I, X, Z), by repeat measure one-way ANOVA followed by LSD test in (D, G-M).

Figure S6

Supplementary Figure 6. Depletion of Esr1 in different bone stromal or myeloid lineage cells results in variable effects in experimental and spontaneous metastasis assays, as related to Figure 6.

(A–E) IIA-mediated experimental bone metastasis of LLC1 cancer cells in female mice with cell type–specific Esr1 deletion: Cx3Cr1-CreERT2 (macrophages, n=18; control n=20), Osx-tTA-Cre (osteoprogenitors, n=24; control n=28), NG2-CreERT2 (mesenchymal stem cells/pericytes, n=9; control n=9), S100A8-Cre (neutrophils, n=15; control n=16), and Itgax-Cre (dendritic cells, n=12; control n=13). Left: schematic diagram; middle: normalized tumor growth curves; right three: quantification of ex vivo bioluminescence in lungs and hind limb bones from experimental bone metastasis.

(F–K) Spontaneous metastasis of LLC1 cancer cells in female mice with cell type–specific Esr1 deletion: Cx3Cr1-CreERT2 (macrophages, n=13; control n=15), Osx-tTA-Cre (osteoprogenitors, n=31; control n=32), Col1a1-CreERT2 (differentiated osteoblasts, n=13; control n=16), NG2-CreERT2 (mesenchymal stem cells/pericytes, n=23; control n=20), S100A8-Cre (neutrophils, n=20; control n=17), and Itgax-Cre (dendritic cells, n=6; control n=8). Left: schematic diagram; middle: quantification of subcutaneous LLC1 tumor weights; right three: quantification of ex vivo bioluminescence in lungs and hind limb bones from spontaneous metastases.

Each dotted curve represents an individual animal, whereas the highlighted curve shows the mean growth for each group in (A-E, middle). All data are shown as mean ± SEM. N indicates biological replicates. P values were assessed by LSD test post repeat measure two-way ANOVA for all tumor growth curves in (A-E, middle), and by the Mann–Whitney test for all scatter plots.

Figure S7

Supplementary Figure 7. T cell modulation by macrophage-specific Esr1 deletion differs from that induced by anti-PD1 treatment, as related to Figure 7.

(A) Flow cytometry quantification of each cell type within biotin⁺ and biotin⁻ bone marrow populations from Esr1^fl/fl–GFP-GPI (n=12) and Esr1^LysM–GFP-GPI (n=11) mice with bone metastases induced by IIA injection of LLC1-SAMENT cells, assessing the impact of macrophage-specific Esr1 deletion.

(B) CD8⁺ T cell frequency in bone metastases of LLC1-injected mice: Esr1^fl/f ± anti-PD1 (n=5 each) and Esr1^LysM (n=5).

(C) Quantification of Biotin⁺ CD8⁺ T cell in bone metastases of GFP-GPI mice treated with fulvestrant (n=10), anti-PD1 (n=11), or control (n=10) following IIA injection of LLC1-SAMENT cells.

(D) Representative immunofluorescence images of showing CD8⁺ T cells (white), F4/80⁺ macrophages (red), and LLC1 tumor cells (green). Infiltrating CD8⁺ T cells indicated by yellow arrows. Scale bar, 100 μm.

(E and F) Quantification of cancer cells (E) and monocytes/macrophages (F) in bone metastases of LLC1-injected mice following IIA injection. Esr1^LysM (n=5) and Esr1^fl/f controls (n=5), with or without anti-PD1 treatment.

All data are shown as mean ± SEM. N indicates biological replicates. P values were assessed by two-tailed unpaired Student’s t-test in (A), by ordinary one-way ANOVA followed by LSD test in (B, C), by two-way ANOVA followed by LSD test in (E, F).

Table S1. DNA sequences of the plasmids used in the study. Related to STAR Methods.

RESOURCE AVAILABILITY

Lead Contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Xiang H.-F. Zhang(xiangz@bcm.edu).

Materials Availability

All unique/stable reagents generated in this study are available from the Lead Contact with a completed Materials Transfer Agreement.

Data and Code Availability

Raw and cellranger processed scRNA-seq data have been deposited in the NCBI Gene Expression Omnibus (GEO) under accession number GSE263383 and GSE296725. Processed scRNA-seq objects (Seurat and Scanpy formats), along with analysis scripts and intermediate visualization files used to generate the figures, are available at Zenodo (https://zenodo.org/records/15374531). Script for creating the figures is also deposited at the same Zenodo site and GitHub (https://github.com/xzhanglab/SAMENT-scrnaseq).