To determine whether osteolineage cells may be present in the TME, we crossed the established tetracycline-dependent Tg(Sp7-tTA,tetO-EGFP/cre)1Amc/J (Osx-cre) to the B6.Cg-Gt(ROSA)26Sortm9(CAG-tdTomato)Hze/J (TdT) to generate the Osx-cre;TdT reporter mouse model (Rodda and McMahon, 2006). When constitutively activated, TdT marks the entire osteolineage, including bone surface osteoblasts, osteocytes and bone marrow cells with mesenchymal stem and osteoprogenitor cell features; while delaying Osx-cre expression until postnatally restricts TdT targeting to committed osteoblasts and osteocytes (Mizoguchi et al., 2014; Fontana et al., 2017). Therefore, Osx-cre;TdT mice and control animals carrying only the TdT transgene (WT;TdT) were kept on standard chow to allow constitutive embryonic transgene activation, or fed a doxycycline (doxy)-containing diet until weaning to suppress transgene activation until one month of age (Figure 1A). We previously reported that doxy-fed mice display less than 1% of spontaneous recombination in the bone residing osteoblasts at weaning, but full transgene activation 1 month thereafter (Fontana et al., 2017).

Osx-cre;TdT reporter mice and WT;TdT fed a normal diet (no doxy) were inoculated with either 10⁵ B16-F10 melanoma cells subcutaneously or 10⁵ PyMT breast cancer cells in the mammary fat pad (MFP). To account for the possible growth delay sometimes observed in no-doxy fed Osx-cre mice (Davey et al., 2012), tumor cells were injected at 12 weeks of age, when the growth delay is fully recovered. Importantly, no differences in tumor growth were observed in age and sex matched Osx-cre;TdT and WT;TdT mice, excluding any potential confounding effect of the Osx-cre (Figure 1—figure supplement 1). To assess the presence of osteolineage Osx+ derived cells within the tumor stroma of no doxy-fed mice, we performed flow cytometry analysis on B16-F10 and PyMT tumors isolated 2 weeks post-inoculation, and found that about 10–18% of the total cells were TdT^OSX+ (Figure 1B and D). Histological analysis of frozen sections from the same tumors confirmed the presence of TdT^OSX+ cells. These cells had either round or elongated shape, indicating morphologic heterogeneity within the TdT^OSX+ population (Figure 1C and E). These TdT^OSX+ cells within the tumor stroma of no-doxy mice may derive from resident Osx-expressing embryonic progenitors or from mobilization of bone marrow Osx+ cells.

Remarkably, TdT^OSX+ cells were also present in the B16-F10 and PyMT TME of doxy-fed animals 4 weeks post weaning, although their numbers were lower than in tumors isolated from no doxy-treated mice (Figure 1F and H). Fluorescence micrographs also confirmed the presence of TdT^OSX+ cells in the stroma in both tumor models (Figure 1G and I), with the same pleiotropic morphology as seen in no doxy-treated mice. Thus, embryonic and adult-derived osteolineage Osx+ cells are present in the stroma of extra-skeletal tumors. Based on these findings, the doxy-fed Osx-cre;TdT mice were used for all the subsequent studies.

To determine the phenotype of TdT^OSX+ cells more in depth, we isolated TdT^OSX+ cells by FACS sorting 14 days after inoculation of B16-F10 tumor cells into doxy-fed Osx-cre;TdT reporter mice (Figure 2A). Due to the limited number of TdT^OSX+, tumors from 2 to 4 mice were pooled for FACS sorting and reported as one experimental replicate (single data point). Messenger RNA from an immortalized murine CAF cell line (i-CAFs) and cortical long bone devoid of bone marrow cells were used as controls. Real-Time PCR confirmed Sp7 expression in TdT^OSX+ cells but not in the TdT^OSX– fraction, which includes the remaining stroma and the tumor cells (Figure 2B). Interestingly, Sp7 expression was also detected in i-CAFs (Figure 2B), and as expected in bone extracts in higher abundance. Expression of osteoblast-specific markers such as Runx2, Bglap (Ocn) (Osteocalcin), Ibsp (Bsp, bone sialoprotein) and Alpl (Tnap, tissue non-specific alkaline phosphatase) were negligible in TdT^OSX+ cells and undetectable or barely detectable in the i-CAFs (Figure 2C–F). However, S100a4(Fsp1) (Fibroblast-specific protein 1) and Acta2 (α-SMA) (α-smooth muscle actin), two CAFs specific markers, were detected in TdT^OSX+ cells (Figure 2G and H). TdT^OSX+ also expressed Col1a1 (Collagen Type Ia1) and Col1a2 (Collagen Type Ia2) similar to i-CAFs and bone extracts (Figure 2I and J). Expression of matrix metalloproteinases, Mmp2 and Mmp9, involved in matrix remodeling and highly expressed in CAFs and osteoblasts respectively, was also detected in the TdT^OSX+ cells (Figure 2K and L).

To validate expression of Sp7 in CAFs from primary tumors, we orthotopically injected B16-F10 or PyMT cells in 8 week old WT mice and the CAFs were isolated as the adherent fraction of a single cell suspension of the tumor mass. We detected Sp7 expression only in CAFs but not in the tumor cell lines nor in the non-adherent fraction of the single cell suspension from the tumor mass (Figure 2M and N). Importantly, corroborating our findings in mice, using Oncomine (Rhodes et al., 2004) we found that microarray analysis of human breast stromal cells collected from patients with invasive breast carcinoma from the Finak gene set (Finak et al., 2008) showed significantly higher Sp7 expression in the tumor associated stroma compared to the adjacent normal breast tissue (Figure 2O). Thus, a subset of cells in the TME of mice and humans express Sp7, an osteolineage cell marker, but also markers associated with the CAFs. TdT^OSX+ cells may represent a subpopulation of CAFs with specific features.

To determine whether tumor resident TdT^OSX+ cells can affect tumor growth as do CAFs, TdT^OSX+ cells were sorted from primary B16-F10 tumors inoculated in 8 week old doxy-fed Osx-cre;TdT mice and re-injected along with B16-F10 tumor cells at 5:1 ratio into age-matched WT recipient animals (Figure 3A). Remarkably, co-injecting TdT^OSX+ with tumor cells resulted in larger tumors (Figure 3B) compared to injection of B16-F10 cells alone.

Because TdT^OSX+ cells normally reside in the bone microenvironment (Fontana et al., 2017; Strecker et al., 2013), we next examined bone sections from 8 week old doxy-fed Osx-cre;TdT reporter mice, naïve or bearing soft tissue B16-F10 or PyMT tumors. We confirmed presence of TdT^OSX+ cells on the surface of cortical and trabecular bone, TdT^OSX+ osteocytes, and also abundant staining in the bone marrow. TdT was not detected WT;TdT mice (Figure 3C and Figure 3—figure supplement 1A). We further analyzed the percentage of TdT^OSX+ cells in the bone marrow by FACS and found that TdT^OSX+ cells accounted for about 10% of total marrow cells in naïve mice and their number increased in mice injected with B16-F10 or PyMT tumor cells (Figure 3D and Figure 3—figure supplement 2B). Similar findings were observed in the Tamoxifen (TAM) inducible Sp7-creER^T2; Rosa26^{<fs-TdTomato>} (Osx-creER^T2;TdT) mouse model. TAM was administered to 7 week old mice via IP injections starting 3 days prior to B16-F10 tumor inoculation, and 1, 6, and 9 days post tumor cell injection (Figure 4—figure supplement 1A). No tumor bearing mice were used as control. Similar to the Tet-OFF Osx-cre model, we found that bone marrow TdT^OSX+ cells significantly increased in presence of a tumor (Figure 3E).

To test their capacity to promote tumor growth, bone marrow-derived TdT^OSX+ cells were isolated from tumor bearing mice and re-injected with B16-F10 tumor cells into WT mice (5:1 ratio). In contrast to what observed with tumor-derived TdT^OSX+ cells, bone marrow-derived TdT^OSX+ cells did not enhance tumor growth compared to mice injected with tumor cells alone; if anything, there was a trend towards a smaller tumor size (Figure 3F). This result further prompted us to isolate bone marrow-derived TdT^OSX+ cells and analyze the expression levels of the same mesenchymal genes analyzed in the tumor-derived TdT^OSX+ cells (Figure 2). Interestingly, TdT^OSX+ cells from the TME expressed higher levels of Sp7, CAF markers S100a4 and Acta2, and matrix proteins Col1a1 and Ibsp compared to TdT^OSX+ from the bone marrow (Figure 3—figure supplement 2). Bone marrow TdT^OSX+ cells instead expressed significantly higher levels of the early osteoblast transcription factor Runx2 and of the protease Mmp9, while no differences in other osteoblast markers such as Bglap and Alpl were detected between the tumor and bone marrow sorted TdT^OSX+ cells (Figure 3—figure supplement 2).

These results suggest that Osx+ cells infiltrating a tumor are functionally and phenotypically distinct from bone marrow resident Osx+ cells.

To gain insights on the possible origin of tumor-derived TdT^OSX+ cells, we performed FACS analysis of circulating cells in naïve mice and two weeks post B16-F10 or PyMT tumor inoculation. Interestingly, we found 5–8% of TdT^OSX+ cells in the blood of tumor-free doxy-fed Osx-cre;TdT mice, and this percentage was 3- to 4-fold higher in tumor-bearing animals (Figure 4A and B). The unexpected high number of circulating TdT^OSX+ cells prompted us to determine whether they may express the immune marker CD45, which is also present in fibrocytes. Surprisingly, 95% of TdT^OSX+ cells in circulation of tumor bearing mice were also positive for CD45, and they represented about 13–18% of total blood cells (Figure 4C and D). Similarly, about 92–95% of total TdT^OSX+ cells in the bone marrow were CD45+ (Figure 4E and F), representing 12–20% of total marrow cells. Such finding explained the low expression levels of mesenchymal markers in the bone marrow-derived TdT^OSX+ cells (Figure 3—figure supplement 2). Intriguingly, also the majority of TdT^OSX+ cells in the TME expressed CD45 (Figure 4G and H), data further confirmed by RT-PCR in TdT^OSX+ fraction sorted from B16-F10 tumors (not shown).

To exclude a possible off-target effect of the doxy-dependent Tet-OFF system, we turned to the TAM inducible Osx-creER^T2;TdT model. FACS analysis confirmed presence of both CD45- and CD45+ TdT^OSX+ populations in the bone marrow of B16-F10 bearing Osx-creER^T2;TdT mice, but not in the cre negative mice (Figure 4—figure supplement 1B).

Next, to confirm that expression of CD45 in the TdT^OSX+ cells was not due a tight interaction between a CD45+ immune cell and TdT^OSX+ mesenchymal cell, we performed CD45 immunostaining on paraffin-embedded B16-F10 tumor sections and on bone marrow single cell suspension from Osx-cre;TdT reporter mice. We found co-localization of Osx-driven TdTomato and CD45 in the same cell, confirming that CD45 is indeed expressed in subsets of TdT^OSX+ cells (Figure 4I and Figure 4—figure supplement 2). Thus, Osx not only marks mesenchymal cells, but also cells expressing the hematopoietic marker CD45, suggesting that they could represent bone marrow fibrocytes or an immune cell subset.

To further characterize the phenotype of the CD45- and CD45+ TdT^OSX+ populations, we next performed RNAseq analysis. We subcutaneously injected the GFP-labeled PyMT-BO1-tumor line in Osx-cre;TdT reporter mice. Two weeks post tumor inoculation, the stromal cells were separated from the GFP+ tumor cells and sorted based on the expression, or lack of thereof, of TdT and CD45 markers. We sorted 4 groups of cells: double negative (TdT^OSX-;CD45-), representing the non-immune tumor stroma, CD45 single positive (TdT^OSX-;CD45+), representing the tumor infiltrating immune populations, TdT^OSX single positive (TdT^OSX+;CD45-), and double positive (TdT^OSX+;CD45+).

Principal components analysis of RNA-seq expression patterns across all four groups revealed clustering that was uniquely dependent on CD45 expression and suggested that there were only two very distinct cell populations regardless of TdT^OSX (Figure 5A). Inspection of each cell population in a one versus all other approach for only statistically significant up-regulated genes (FDR >= 0.05, log2 fold-change >= 2) to identify robustly expressed biomarkers, showed that all four groups shared many genes in common based on CD45 expression alone. The CD45 negative populations shared 1248 genes in common with 536 in the TdT^OSX single positive and 477 in the double negative cells uniquely up-regulated (Figure 5B). Likewise, the CD45 single and double positive cells shared 495 significantly up-regulated genes with an additional 77 and 316 uniquely expressed, respectively.

Subsequent Gene Ontology (GO) analysis of the log₂ fold-changes for each cell population revealed that the CD45 negative subsets showed upregulation of GO pathways related to cellular responses to growth factors, extracellular structure and matrix organization and skeletal system development, confirming their mesenchymal nature (Figure 5C). Among the top upregulated genes in the TdT^OSX single positive cells versus the double negative (Table 1) we found several markers expressed by osteolinage cells, such as fibromodulin (Fmod), a binding protein regulating bone mineralization (Gori et al., 2001) and also involved in TGFβ signaling during cancer pathogenesis (Pourhanifeh et al., 2019), Collagen 24a1 (Col24a1), known to modulate collagen chain trimerization (Koch et al., 2003; Matsuo et al., 2006), Frem1, a protein involved in the formation and organization of basement membranes (Vissers et al., 2011), and Msx1, a transcription factor important in craniofacial bone development (Orestes-Cardoso et al., 2002). TdT^OSX single positive cells also expressed high levels of fibulin 7 (Fbln7), a cell adhesion molecule overexpressed in glioblastoma by pericytes and involved in neovascularization (de Vega et al., 2019; Ikeuchi et al., 2018). Such result confirmed the osteolineage nature of the TdT^OSX+ cells and indicated their tumor supporting role.

Conversely, both CD45+ subsets were significantly up-regulated for GO pathways related to regulation of lymphocyte activation and proliferation, and innate immune cell responses (Figure 5D). Interestingly, among the top genes upregulated in the TdT^OSX+ CD45+ subsets versus the CD45 single positive populations (Table 2), we found many genes expressed by tumor promoting immune cells, such as regulatory T cells (i.e. Foxp3) and gamma-delta T cells (i.e. Tcrg-C1). The most upregulated gene in the TdT^OSX+;CD45+ subset (>80 fold versus CD45 single positive) was Cd163l1, a marker of gamma-delta IL-17 producing cells (Tan et al., 2019) and M2 macrophages (González-Domínguez et al., 2015), two tumor-promoting immune populations. We confirmed Cd163l1 higher expression by RT-PCR together with other two genes, Wdr78 and Spock2, also upregulated in the double positive cells (Figure 5—figure supplement 1). The second most expressed gene was Lymphotoxin beta (Ltb), a cytokine produced by lymphocytes and NK cells and associated with carcinogenesis (Wolf et al., 2010). The third one, Klrg1, negatively regulates cytotoxic lymphocytes and is associated with lymphocyte senescence and dysfunction (Henson and Akbar, 2009). Thus, these analyses revealed that two main transcriptional programs predominate based on CD45 expression, and suggest that TdT^OSX+;CD45- comprise a subset of CAFs with some characteristics of skeletal cells, while TdT^OSX+;CD45+ cells represent a heterogeneous subset of tumor-promoting immune cells.

Next we performed flow cytometry analysis to better characterize the TdT^OSX+;CD45+ population and validate some of the RNAseq findings using the B16-F10 melanoma model. We confirmed that the double positive TdT^OSX+;CD45+ represented about 20% of the total CD45+ cells in both the primary tumor and the bone marrow of doxy-fed Osx-cre;TdT mice (Figure 6A and B). Next, we analyzed the percentage of TdT^OSX+ expressing the common myeloid and lymphoid markers (gate strategy in Figure 6—figure supplement 1), such as CD11b (monocytes), F4/80 (macrophages), Gr1 (granulocytes/neutrophils), CD3 (T cells, further divided into CD4+ and CD8+), and NK1.1 (Natural Killer cells). To standardize the comparison between the TdT^OSX+;CD45+ and TdT^OSX-;CD45+ (single positive) populations within each sample, data were represented as percentage of either total TdT^OSX+;CD45+ or total CD45 single positive cells (Figure 6C and D). TdT^OSX+;CD45+ subset expressed both myeloid and lymphoid markers, with a similar pattern as the CD45 single positive cells. In both subsets, the majority of the tumor immune infiltrate included myeloid populations, which represented about 60% of the TdT^OSX+;CD45+ and over 70% of the total CD45 single positive cells. Lymphoid cells appeared to be more abundant among the TdT^OSX+;CD45+ (32.3 ± 6.5%), compared to the TdT^OSX-;CD45+ cells (23.3 ± 4.3%). Based on this observation, we calculated the ratio of lymphoid over myeloid populations and observed that the tumor-infiltrating TdT^OSX+;CD45+ double positive population was enriched in lymphoid cells (Figure 6E). Flow cytometric analysis of the bone marrow from tumor bearing Osx-cre;TdT mice revealed similar pattern of distribution between the TdT^OSX+;CD45+ and the CD45 single positive immune populations (Figure 6F and G), although the frequencies of each immune subset in the bone marrow differed compared to the cells at tumor site. These data suggest that Osx marks multiple immune cell types but that TdT^OSX+-derived immune cells infiltrating a tumor are skewed towards lymphoid populations.

Since adoptive transfer of tumor-derived TdT^OSX+ cells increases tumor growth (Figure 3B), we asked whether TdT^OSX+;CD45+ immune populations contribute to tumor progression. We isolated by FACS sorting TdT^OSX+;CD45+ and the TdT^OSX-;CD45+ from B16-F10 subcutaneous tumors from doxy-fed Osx-Cre;TdT mice. Sorted cells were then re-injected along with B16-F10 tumor cells (ratio 5:1) into new WT recipient mice. The co-injection of CD45 single positive cells decreased tumor growth relative to B16-F10 injected alone at day 15, while TdT^OSX+;CD45+ did not show anti-tumor effects (Figure 6H).

Collectively, these data indicate that Sp7 marks cells of both mesenchymal and hematopoietic origin to support the development of tumor promoting populations.

To determine whether Sp7 may be activated at early stages of hematopoiesis as a potential explanation for the heterogeneity of TdT^OSX+;CD45+ cells, we performed flow cytometric analysis for hematopoietic stem cell (HSC) markers in the bone marrow of doxy-fed Osx-cre;TdT and TAM-induced Osx-creER^T2;TdT mice bearing B16-F10 tumors subcutaneously. We analyzed the lineage negative (Lin^-) LSK and LK populations (Figure 7A) gated based on the expression of Sca1 and c-kit and found 10.9–11.45% TdT^OSX+ LSK and 7.08–8.94% TdT^OSX+ LK in Osx-cre;TdT and Osx-creER^T2;TdT mice, respectively (Figure 7B and D). The LSK population was further divided into HSC and multipotent progenitor MPP(1-4) subsets (Figure 7A and gate strategy in Figure 7—figure supplement 1). Interestingly, TdT^OSX+ cells represented about 30% of MPP1, and between 7–11% of the MPP2-MPP3 and MPP4 subsets in Osx-cre;TdT mice (Figure 7C). Similar results were also obtained in Osx-creER^T2;TdT mice (Figure 7E). The slight difference in frequency between the two mouse models could depend on the different timing of Cre activation, but both models succeeded to find TdT^OSX+ cells in HSCs.

Importantly, RT-PCR of isolated HSCs from the bone marrow of 7–8 week old wild-type naïve mice showed Sp7 expression compared to the TdT^OSX- fraction isolated from the tumor mass, used as negative control (Figure 8A). Similarly, Sp7 transcripts were found in the MPP1-4 subsets, and in the common lymphoid progenitors CLP, but not in the common myeloid progenitors CMP (Figure 8A). In contrast, RT-PCR analysis of mature immune populations from naïve mice (macrophages, dendritic cells and T cells) failed to detect Sp7 expression (Figure 8B). Finally, we performed immunostaining for Osx to confirm protein expression on a single cell level. To enrich for Osx+ cells, we FACS sorted TdT^OSX+ HSCs from the bone marrow of Osx-cre;TdT reporter mice after enrichment for c-kit+ cells. Sorted cells were plated on serum-coated coverslips for two days before immunostaining for Osx and with DAPI for nuclear localization. We confirmed Osx localization in the nuclei of the majority of TdT^OSX+ HSCs (Figure 8D). As positive control for Osx staining, we used bone marrow stromal cells isolated from WT mice cultured for 4 days in osteogenic medium (Figure 8C). Whole bone marrow was used as additional control confirming that only a small number of cells expressed Osx while the majority was negative (Figure 8E). Thus, for the first time we show that subsets of HSCs express the transcription factor Sp7, giving rise to tumor infiltrating immune populations.

B16-F10 (C57BL/6 mouse melanoma cell line, ATCC # CRL-6475-LUC2), PyMT and PyMT-BO1-GFP (Su et al., 2016) (C57BL/6 mouse breast cancer cell lines), kindly provided by Prof. Katherine N Weilbaecher (Washington University in St. Louis), were cultured at 37°C in complete media (DMEM supplemented with 2 mM l-glutamine, 100 μg/ml streptomycin, 100 IU/ml penicillin, and 1 mM sodium pyruvate) containing 10% FBS. Immortalized CAF cell line was kindly provided by Prof. Gregory D Longmore (Washington University in St. Louis). All the cell lines used have been tested for mycoplasma and were mycoplasma free. CAFs were isolated from primary tumors as described in Corsa et al., 2016. Tumors were minced, digested and plated as single cell suspension on a tissue plastic dish for 30 min to separate the adherent fraction, composed of CAFs, from the cells in suspension, comprising of tumor cells, tumor infiltrating immune populations and endothelial cells, among few others. CAFs were then left 24 hr in culture and RNA was extracted afterwards.

Bone marrow stromal cells (BMSCs) were cultured in complete alpha-MEM (without ascorbic acid) containing 10% FBS and differentiated for 4 days in osteogenic medium (complete alpha-MEM supplemented with 50 μg/ml ascorbic acid and 10 μM beta-glycerophosphate) to obtain pre-osteoblasts.

Sorted hematopoietic stem cells (HSCs) were plated overnight on a glass coverslip coated with serum in the presence of StemPRO (Gibco) media and then fixed and used for immunostaining.

Animals were housed in a pathogen-free animal facility at Washington University (St. Louis, MO, USA). Age and sex-matched animals were used in all experiments according to IACUC guidelines. B6.Cg-Tg(Sp7-tTA,tetO-EGFP/cre)1Amc/J (catalog #006361; The Jackson laboratory, ME USA) mice, which carry a tetracycline-responsive Osx promoter driving Cre (Osx-cre) were mated with reporter mice B6.Cg-Gt(ROSA)26Sor^{Tm9(CAG-tdTomato)Hze}/J (TdT) (catalog #007909; The Jackson Laboratory), to generate Osx-cre;TdT mice. To suppress Cre expression, 200ppm doxycyline (doxy) was added to the chow (Test Diet #1816332–203, Purina, MO USA) and fed to some groups of mice until weaning (P28); pups were switched to standard rodent chow at weaning. Sp7-creER^T2/Rosa26^{<fs-TdTomato>} mice were kindly provided by Prof. Matthew D Silva (Washington University in St. Louis). Tamoxifen (#T5648, Millipore Sigma, MO USA) was dissolved in corn oil and injected at 100 mg/kg intra-peritoneally. Wild-Type (WT) C57BL/6 mice were purchased from The Jackson Laboratory.

Subcutaneous (sq) injections were performed using 10⁵ B16-F10 tumor cells suspended in 100 µl of sterile PBS and Matrigel Matrix (Corning #354234), while 10⁵ PyMT were suspended in 50 µl of sterile PBS and Matrigel Matrix and injected into the MFP. Adoptive transfer and co-injection with tumor cells were performed using a 5:1 ratio of TdT^OSX+:B16-F10 tumor cells. The number of cells injected in each experiment varied depending upon the number of cells obtained from sorting in each experiment (hence, the differences in tumor size among the different experiments), but within (2.5×10⁵:5×10⁴) and (5×10⁵:1×10⁵) cells. B16-F10 cells only were injected as control using the same number of tumor cells co-injected with TdT^OSX+ cells. Tumors were monitored by caliper measurements and mice were sacrificed between day 14 and 16 post tumor inoculation. Tumor volume was calculated according to the formula: 0.5236*length*(width [Kalluri and Zeisberg, 2006]).

Single cell suspensions were prepared from fresh bone marrow and tumors upon sacrifice. Bone marrow cells were obtained from femurs and tibias by centrifugation, while tumors were minced and digested with 3.0 mg/ml collagenase A (Roche, Basel Switzerland) and 50 U/ml DNase I (Millipore Sigma) in serum-free media for 45 min at 37°C. Cells were filtered through 70 μm strainers, red blood cells (RBC) were then removed with RBC lysis buffer (Millipore Sigma), washed twice in PBS and stained in FACS buffer (0.5% BSA, 2 mM EDTA, 0.01%NaN₃) with the following antibodies: CD45 (clone 30-F11, eBioscience), CD11b (clone M1/70, eBioscience), Ly6G (Gr-1) (clone RB6-8C5, eBioscience), F4/80 (BM8, Biolegend), CD3 (145–2 C11, eBioscience), CD4 (RM4-5, BD Biosciences), CD8 (clone 53–6.7, BD Biosciences), NK1.1 (clone PK136, Biolegend). Once stained, cells were sorted for the adoptive transfer, RT-PCR and RNAseq studies or were fixed for flow cytometry (gate strategy in Figure 7—figure supplement 1) in BD Cytofix fixation buffer (BD Biosciences, 554655). Samples were acquired using the BD LSR-Fortessa cytometer and analyzed with FlowJo version 9.3.2.

Hematopoietic stem cell subsets were analyzed from fresh bone marrow isolated from femurs, tibias and hip bones crushed with mortar and pestle in 3 ml of PBS, filtered in a 70μm-nylon strainer to remove bone fragments and lysed of red blood cells (RBC). 10 million cells were stained for each sample using the following antibodies: CD3, Gr1, B220 (clone RA3-6B2, eBioscience), Ter119 (clone TER119, e Bioscience), CD41 (clone eBioMWReg30, eBioscience) for lineage negative gating, Sca-1 (clone D7, BD Biosciences), CD117/c-kit (clone ACK2, eBioscience), CD16/32 (clone 2.4G2, BD Biosciences), CD34 (clone RAM34, eBioscience), CD150/SLAM (clone TC15-12F12.2, Biolegend), CD48 (clone HM48-1, eBioscience), CD135/Flt3 (clone A2F10, eBioscience), gate strategy in Figure 7—figure supplement 1. Once stained, cells were fixed in BD Cytofix fixation buffer, acquired using the Bio-Rad ZE5 (YETI) cytometer and analyzed with FlowJo version 10.4.1.

Isolation of hematopoietic progenitor subsets for Real Time analysis was obtained by FACS sorting using the following markers: HSCs (Lineage- c-Kit+Sca-1+CD48-CD150+), MPPs (Lineage- c-Kit+Sca-1+CD48-CD150-), CLPs (Lineage- c-Kit+Sca-1+IL7rα+Flt3+Ly6D-) and CMPs (Lineage- c-Kit+Sca-1-CD34+CD16/32). For IF, TdTomato was also used to sort Osx+ HSCs.

Isolation of hematopoietic stem cells for Osx immunostaining was performed as following: whole bone marrow was isolated from femurs, tibias and iliac crests and incubated with mouse CD117-conjugated microbeads (Miltenyi Biotec #130-091-224). CD117-enrichment was then carried out using the AutoMACS Pro Seperator (Miltenyi Biotec). Post CD117 enrichment, the positive cell fraction was stained to sort phenotypically defined TdTomato+ HSCs (Lineage- c-Kit+Sca-1+CD48-CD150+) by FACS.

Tumors and bones (femur and tibia) were fixed in 4% PFA over-night (ON) at 4°C. The fixed bones were partially decalcified in 14% EDTA for 3 days, washed in a sucrose gradient (1 hr in 10% sucrose, 1 hr in 20% sucrose, ON in 30% sucrose) before snap-freezing them in OCT embedding medium. Tumors were directly put in sucrose gradient and then embedded in OCT. Frozen sections were cut at 5 μm thickness and kept at −20°C until analysis.

Thawed sections were washed in PBS and mounted with VECTASHIELD Mounting Medium containing DAPI (Vector Laboratories, CA USA).

Images were taken using the Leica DMi8 Confocal Microscopy at the Musculoskeletal Research Center (Washington University in St Louis), magnification 200X.

Tissues were fixed in 10% neutral-buffered formalin for 18 hr, embedded in paraffin after graded-ethanol dehydration, and sectioned into 6 μm sections using a microtome. Automated staining was carried out on the BondRxm (Leica Biosystems). Following dewaxing and citrate-buffered antigen retrieval, sections were stained with the primary antibody for 1 hr at RT. Sequential chromogenic detection was performed with the Bond IntenseR (Rat primary) and Bond Polymer Refine Red (Rabbit primary) detection kits (Leica Biosystems). The primary antibodies used were rat anti-CD45 (Invitrogen clone 30-F11, #14-0451-82) followed by rabbit anti-RFP/TdT (Rockland, # 600-401-379). Stained sections were dehydrated in graduated ethanol and xylene washes then mounted with xylene-based Cytoseal (Thermo Fisher).

Stained slides were imaged at 20X magnification with the Zeiss Axioscan slide scanner. For image analysis, Halo v3 (Indica Labs) was used to deconvolve dual stained images into single channels and pseudo colored with blue representing nuclei (hematoxylin), green representing CD45+ cells (DAB stained), and red representing Osx+ cells (RFP stained).

For colocalization of TdT+ cells and CD45 in B16-F10 soft tissue tumors, subcutaneous tumors were dissected and fixed in 10% neutral-buffered formalin for 18 hr, embedded in paraffin after graded-ethanol dehydration, and sectioned into 6 μm sections using a microtome. Sections were dewaxed in xylene and then hydrated through graded ethanol washes. Endogenous peroxidases were quenched by incubating in hydrogen peroxide (1% in PBS) for 10 min at RT. Antigen retrieval was then performed by microwave treatment in a citrate buffer for 15 min. Sections were washed in TBS-T (1X TBS with 0.05%Tween-20) followed by blocking for 30 min at RT in blocking buffer (5% goat serum, 2.5% BSA in TBS). Slides were then blocked using the Avidin/Biotin Blocking Kit (Vector Labs). For staining, slides were incubated overnight in a humidified chamber at 4°C with 1:200 rat anti-CD45 (Invitrogen clone 30-F11, #14-0451-82) and 1:400 rabbit anti-RFP (Rockland, #600-401-379) antibodies diluted in 50% blocking buffer. After primary staining, slides were washed in TBS-T and then incubated with HRP-conjugated anti-rat IgG secondary (1:500 in 50% blocking buffer) for 30 min at RT. For fluorescent detection of CD45, slides were washed and then incubated in 1:50 FITC Tyramide reagent (PerkinElmer, # NEL701A001KT) for 8 min at RT. For fluorescent detection of RFP, slides were washed and incubated in AlexaFluor 594-conjugated anti-rabbit IgG secondary (1:500 in 50% blocking buffer) for 30 min at RT. Stained slides were washed and mounted using ProLong Gold with DAPI (Invitrogen).

Total RNA was extracted with TRIzol (Invitrogen, CA USA) and quantified on a ND-1000 spectrophotometer (NanoDrop Technologies). The cDNA was synthesized with 1 μg RNA using High Capacity cDNA Reverse Transcription Kit (#4368814, Applied Biosystems, CA USA).

For purified hematopoietic precursors (HSC, MPP, CLP and CMP) total RNA was extracted using the NucleoSpin RNA XS kit (Macherey-Nagel #740902) and reverse transcribed with the SuperScript VILO kit (Invitrogen #11754–050).

The amount of each gene was determined using Power SYBR Green mix on 7300 Real-Time PCR System (Applied Biosystems). Cyclophilin mRNA was used as housekeeping control. Specific primers for mice were as follows: Cyclophilin, 5’-AGC ATA CAG GTC CTG GCA TC-3’ and 5’-TTC ACC TTC CCA AAG ACC AC-3’; Osterix, 5’-AAG GGT GGG TAG TCA TTT GCA-3’ and 5’-CCC TTC TCA AGC ACC AAT GG-3’; S100a4 (Fsp1), 5’-TGA GCA ACT TGG ACA GCA ACA-3’ and 5’-TTC CGG GGT TCC TTA TCT GGG-3’; Acta2 (a-SMA), 5’-GTC CCA GAC ATC AGG GAG TAA-3’ and 5’-TCG GAT ACT TCA GCG TCA GGA-3’; Col1a1, 5’-GGC CTT GGA GGA AAC TTT GC-3’ and 5’-GGG ACC CAT TGG ACC TGA AC-3’; Bglap (Osteocalcin), 5’-GGA CTG AGG CTC TGT GAG GT-3’ and 5’-CAG ACA CCA TGA GGA CCA TC-3’; Runx2, 5’-GTT ATG AAA AAC CAA GTA GCC AGG-3’ and 5’-GTA ATC TGA CTC TGT CCT TGT GGA-3’; Ibsp (Bsp), 5’-AGG ACT AGG GGT CAA ACA C-3’ and 5’-AGT AGC GTG GCC GGT ACT TA-3’; Alpl (Tnap), 5’-GGG GAC ATG CAG TAT GAG TT-3’ and 5’-GGC CTG GTA GTT GTT GTG AG-3’. The relative quantification in gene expression was determined using the 2^-ΔΔCt method.

Single cell suspensions were prepared from fresh tumor upon sacrifice. Tumors were minced and digested as described above. Tumor stromal cells were separated from the GFP+ tumor cells and sorted based on the expression, or lack of thereof, of TdT and CD45 markers.

Library preparation was performed with 10 ng of total RNA and integrity was determined using an Agilent bioanalyzer. ds-cDNA was prepared using the SMARTer Ultra Low RNA kit for Illumina Sequencing (Clontech) per manufacturer’s protocol. The cDNA was fragmented using a Covaris E220 sonicator using peak incident power 18, duty factor 20%, cycles/burst 50, for a 120 s. cDNA was blunt ended, had an A base added to the 3’ ends, and then had Illumina sequencing adapters ligated to the ends. Ligated fragments were then amplified for 12 cycles using primers incorporating unique index tags. Fragments were sequenced on an Illumina HiSeq 3000 using single end reads extending 50 bases. Basecalls and demultiplexing were performed with Illumina’s bcl2fastq software and a custom python demultiplexing program with a maximum of one mismatch in the indexing read. RNA-seq reads were then aligned to the Mus musculus Ensembl release 76 top-level assembly with STAR version 2.0.4b (Dobin et al., 2013). Gene counts were derived from the number of uniquely aligned unambiguous reads by Subread:featureCount version 1.4.5 (Liao et al., 2014). Isoform expression of known Ensembl transcripts were estimated with Sailfish version 0.6.13 (Patro et al., 2017). Sequencing performance was assessed with RSeQC version 2.3 (Wang et al., 2012).

All gene counts were then imported into the R/Bioconductor package EdgeR (Robinson et al., 2010) and TMM normalization size factors were calculated to adjust for samples for differences in library size. Ribosomal genes and genes not expressed in at least two samples greater than one count-per-million were excluded from further analysis. The TMM size factors and the matrix of counts were then imported into the R/Bioconductor package Limma (Ritchie et al., 2015) and weighted likelihoods based on the observed mean-variance relationship of every gene and sample were then calculated with Limma’s voomWithQualityWeights (Liu et al., 2015). The data were then fitted to a Limma generalized linear model to test for changes in a single cell population relative to the mean of all other populations to find only genes that were uniquely up-regulated with Benjamini-Hochberg false-discovery rate adjusted p-values less than or equal to 0.05. For each group of cells, global perturbations in known Gene Ontology (GO) terms were then measured using the R/Bioconductor package GAGE (Luo et al., 2009) to quantify the mean log two fold-changes of ll genes in a given term versus the background log two fold-changes of all genes found outside the respective term. Only globally up-regulated GO terms with Benjamini-Hochberg false-discovery rate adjusted p-values less than or equal to 0.05 were considered for comparison across populations.