The Molecular Signature of the Stroma Response in Prostate Cancer-Induced Osteoblastic Bone Metastasis Highlights Expansion of Hematopoietic and Prostate Epithelial Stem Cell Niches

The reciprocal interaction between cancer cells and the tissue-specific stroma is critical for primary and metastatic tumor growth progression. Prostate cancer cells colonize preferentially bone (osteotropism), where they alter the physiological balance between osteoblast-mediated bone formation and osteoclast-mediated bone resorption, and elicit prevalently an osteoblastic response (osteoinduction). The molecular cues provided by osteoblasts for the survival and growth of bone metastatic prostate cancer cells are largely unknown. We exploited the sufficient divergence between human and mouse RNA sequences together with redefinition of highly species-specific gene arrays by computer-aided and experimental exclusion of cross-hybridizing oligonucleotide probes. This strategy allowed the dissection of the stroma (mouse) from the cancer cell (human) transcriptome in bone metastasis xenograft models of human osteoinductive prostate cancer cells (VCaP and C4-2B). As a result, we generated the osteoblastic bone metastasis-associated stroma transcriptome (OB-BMST). Subtraction of genes shared by inflammation, wound healing and desmoplastic responses, and by the tissue type-independent stroma responses to a variety of non-osteotropic and osteotropic primary cancers generated a curated gene signature (“Core” OB-BMST) putatively representing the bone marrow/bone-specific stroma response to prostate cancer-induced, osteoblastic bone metastasis. The expression pattern of three representative Core OB-BMST genes (PTN, EPHA3 and FSCN1) seems to confirm the bone specificity of this response. A robust induction of genes involved in osteogenesis and angiogenesis dominates both the OB-BMST and Core OB-BMST. This translates in an amplification of hematopoietic and, remarkably, prostate epithelial stem cell niche components that may function as a self-reinforcing bone metastatic niche providing a growth support specific for osteoinductive prostate cancer cells. The induction of this combinatorial stem cell niche is a novel mechanism that may also explain cancer cell osteotropism and local interference with hematopoiesis (myelophthisis). Accordingly, these stem cell niche components may represent innovative therapeutic targets and/or serum biomarkers in osteoblastic bone metastasis.


Introduction
Prostate cancer (PCa) is the most common solid cancer in men in the western world. Despite early detection and surgical treatment of the tumor, 10-20% of PCa patients show bone metastases at diagnosis [1] and .80% of advanced PCa patients have bone metastases at autopsy [2]. Bone metastases are the most important cause of morbidity in these patients and, once developed, are incurable.
In primary and metastatic cancers neoplastic cells closely interact with different cell types and the extracellular matrix (ECM) constituting the stroma compartment. This leads to activation of the stroma and, in turn, to the secretion of additional growth factors, matrix proteins and proteases, which further favor cancer cell proliferation and invasion. These heterogeneous and bi-directional interactions within the tumor tissue are fundamental for tumor growth progression [3]. Therefore, elucidation of the mechanism(s) determining the initiation and progression of metastatic growth is essential for the identification of novel therapeutic targets for prevention and/or treatment of cancer metastases.
In bone metastases activation of the bone marrow/bone (BM/B) stroma by cancer cells alters the physiological balance between osteoblast (OB)-mediated bone formation and osteoclast (OC)-mediated bone resorption, and interferes with hematopoiesis (myelophthisis). PCa elicits predominantly an OB response, with a consequent increase in bone formation (osteoinduction) and generation of osteosclerotic lesions [4]. Instead, mammary cancer (MCa) triggers preferentially an OC reaction, resulting in exaggerated bone resorption and generation of osteolytic lesions [5]. These opposite stromal reactions may underlie different growth support requirements between pro-osteolytic and osteoinductive cancer cells.
Factors released from the bone matrix during OC-mediated bone resorption fuel the proliferation of cancer cells, which stimulate further bone resorption, thereby leading to a self-perpetuating, positive feedback loop. This mechanism, known as the ''Vicious Cycle'' hypothesis of bone resorption and tumor growth in osteolytic bone metastases is a paradigmatic example of cancer cell-stroma interaction and has provided the rationale for interfering with the bone stroma support by inhibition of bone resorption [6]. However, the molecular cues provided by OBs for survival and growth of osteoinductive cancer cells have remained largely elusive [4].
Paget's ''Seed & Soil'' hypothesis [7] postulates that cancer cells (the ''seed'') from the primary tumor can disseminate to various tissues, but succeed in establishing secondary growth only in those that are permissive for their survival and/or proliferative expansion (the ''fertile soil'') [8]. Thus, the ''Seed & Soil'' hypothesis embodies the propensity of certain cancers to metastasize to specific tissues (tissue tropism). Paget was also the first to recognize that MCa metastasizes almost exclusively to the axial skeleton, site of the red marrow [7]. This observation suggests that active hematopoiesis may represent the fertile soil for seeding of cancer cells, which may therefore mimic hematopoietic stem cells (HSCs) for homing, survival and/or expansion in the BM/B [9]. This hypothesis, implying that the BM/B metastatic niche may match with the HSC niche, has been experimentally validated by showing that PCa cells hijack the HSC niche(s) for homing in the BM/B [10].
Most attempts to decipher the cancer gene signatures highlighting signaling pathways critical for cancer progression or predicting patient outcome have been performed by gene expression profiling in clinical samples of bulk tumor tissue. Obviously, this strategy cannot discriminate between cancer-and stroma-derived gene expression. Naef and Huelsken [11] have developed a method of ''tissue compartment-specific transcriptional profiling'' (TCTP), which allows simultaneous analysis of gene expression specific for the cancer cell and the stromal compartment in situ, without prior cell separation. This approach exploits differences between human and mouse RNA sequences and selects the most species-specific oligonucleotide probes by a computational mask. By a similar strategy we generated the osteoblastic bone metastasis-associated stroma transcriptome (OB-BMST) defining for the first time the global stroma response in bone xenograft models of osteoinductive PCa cells.

Osteoinductive PCa cells alter the BM/B stroma transcriptome
The strategy adopted to investigate transcriptome changes that occur specifically in the stroma compartment of bones xenografted with human osteoinductive prostate cancer cells is outlined in Fig. 1A. bioinformatics (grey) steps adopted to define a complete (OB-BMST) and a curated (''Core'' OB-BMST) stroma response signature (orange). The first two experimental steps constitute the tissue compartment-specific transcriptional profiling (TCTP). B. Heatmap showing differentially expressed probe-sets between xenografted (C4-2B and VCaP) and control (Ep156T, intact and sham) bones. The expression level is color-coded: low expression is represented in blue, whereas high expression is represented in red. C. Venn diagram showing the number of overlapping and unique genes differentially expressed in The differentially expressed probe sets clearly separate cancer-cell xenografted bones from control bones, as highlighted by hierarchical clustering (Fig. 1B). The heatmap shows homogeneous clustering within the xenografts, but the dendrogram still indicates that the C4-2B and VCaP xenografts separate in two branches. All control samples also cluster together, suggesting that surgical intervention and inoculation of non-tumorigenic, normal human prostate epithelial cells (Ep156T) did not cause major modifications in gene expression. Additionally, the heatmap indicates that in xenografts the prevalent part of genes (64%) were up-regulated.
A total of 321 genes from the OB-BMST are common to both C4-2B and VCaP xenografts (Fig. 1C, S1 Table). The scatter plot of log 2 fold change for both xenografts shows a significant correlation (R 2 50.96) (Fig. 1D), indicating a complete concordance of gene regulation for both models of osteoblastic bone metastasis. However, within the OB-BMST, 333 and 262 genes are unique for C4-2B and VCaP, respectively. Differences in osteoinductive potential (VCaP.C4-2B) and/or in growth kinetics (C4-2B.VCaP) may explain this partial diversity of stroma reaction induced by the two cell lines. Nevertheless, among the top 30 most up-regulated genes 18 are shared by C4-2B and VCaP xenografts (Fig. 1E, S1 Table), indicating that both cell lines are comparable in inducing the predominant stroma reaction.
The OB-BMST associates with myoepithelial/myofibroblast signature To identify the cell type primarily responsible for the OB-BMST expression, we compared the up-regulated genes of the OB-BMST with gene signatures previously derived from specific stroma cell populations in mammary tumors [12]. This analysis shows that, the OB-BMST mainly overlaps with the myoepithelial/myofibroblast signature and, to a minor extent, with the fibroblast and EC signatures (S1 Figure). beta (Pdgfrb), fascin homolog 1 (Fscn1) and prostate transmembrane protein, androgen induced 1 (Pmepa1) revealed that these genes are not only induced in the stroma of bones xenografted with C4-2B and VCaP cells (S2A Figure), but also in orthotopic VCaP xenografts (S2B Figure) and in ectopic (S2C Figure) xenografts of both C4-2B and VCaP cells. In addition, the expression of the 7-gene set is induced also in bones xenografted with the pro-osteolytic PCa cell line PC-3 (S2D Figure). Furthermore, ASPN and POSTN expression is not only increased in the stroma of bone metastatic human PCa, but also of primary PCa and of bone metastatic human MCa (S3 Figure).
Taken together, the results above suggest that a fraction of the OB-BMST is not specific for the BM/B response to osteoinductive PCa cells. However, the finding that proteins encoded by genes of the OB-BMST are overexpressed also in the stroma of human primary and bone metastatic tumors underscores the translational value of the OB-BMST.
A fraction of the OB-BMST is specific for the BM/B response to osteoinductive PCa cells To identify the OB-BMST component specific to the osteoblastic bone metastases, we subtracted from the OB-BMST stroma gene signatures derived from inflammatory/wound healing and desmoplastic responses, and from nonosteotropic and osteotropic cancers (Fig. 1A).
This strategy, referred to, thereafter, as ''curation'', has led to the identification of 4 major components within the OB-BMST: 1) a component putatively specific for the BM/B stroma response to osteoinductive PCa cells, from now on referred to as ''Core OB-BMST'', 2) a component shared with ''inflammatory/wound healing/desmoplastic'' response signatures, 3) a component possibly representing a ''universal'' response to cancer cells and 4) a candidate ''osteotropic'' signature.
The Core OB-BMST, covering 72.6% of the OB-BMST, consists of 336 upregulated and 298 down-regulated genes. Of these, 109 and 93, respectively, are common to both C4-2B and VCaP xenografts ( Fig. 2A, S2 Table). These genes are likely to be restricted to the BM/B stroma reaction to osteoinductive cancer cell growth. However, the specificity of the Core OB-BMST should be considered with caution since our subtractive strategy was limited to publicly available gene signatures and did not consider studies concerning single gene and/or protein expression in a variety of cancers.
The top 10 genes of the Core OB-BMST list (Fig. 2B, S2 Table) are conserved from the top 30 most up-regulated genes of the OB-BMST (Fig. 1E, S1 Table). Among the top 30 genes of the Core OB-BMST, 12 are common to C4-2B and VCaP xenografts, while 13 and 5 are unique of VCaP and C4-2B, respectively. Thus, in contrast to the OB-BMST top 30-gene list, the Core OB-BMST is primarily induced by VCaP cells. Most likely, this is the consequence of the more robust OB response induced by these cells, as compared to C4-2B (Fig. 1A).  Table), genes of the grey area represent a potential osteotropic signature (complete list reported in S6 Table). B. Top 30 up-regulated genes of the Core OB-BMST derived from C4-2B (black bars) and VCaP (grey bars) xenografted bones. Angiogenesis and osteogenesis are the key processes represented in the OB-BMST and Core OB-BMST GO terms analysis shows that up-regulated genes in both the OB-BMST and Core OB-BMST common to VCaP and C4-2B (''common'' OB-BMST/Core OB-BMST) are highly associated (FDR,0.05) with annotations terms related to angiogenesis, skeletal system development and enzyme-linked receptor protein signaling pathway (Fig. 3 A and B). ECM organization and cell adhesion are also highly represented in the OB-BMST, while TGFb receptor signaling appears as additional term in the Core OB-BMST. Down-regulated genes are mainly grouped in GO terms related to cell cycle ( Fig. 3A and B). Most likely, the complete lack of hematopoiesis in the BM spaces invaded by cancer cells (myelophthisis) is responsible for this phenomenon.
The analysis of over-represented sequence motifs in the promoters of the genes of the Core OB-BMST shows that only the up-regulated genes have significant motif over-representation, namely for Foxo4, Meis1 and Maz ( Table 2). The processes associated with the Foxo4-, Meis1-and Maz-linked genes are angiogenesis and vasculogenesis (P510E-09), cell junction assembly and organization (P510E-05) and collagen fibril organization and extracellular matrix organization (P510E-03).
Collectively, the findings above strongly suggest angiogenesis and osteogenesis as major biological processes and Tgfb as the major signaling pathway involved in osteoblastic bone metastasis.
The OB-BMST and Core OB-BMST partially overlap with ''stem cell niche'' signatures The literature survey for 96 of the most up-regulated genes of the Core OB-BMST shows that 32 genes (33.3%) are involved in OB recruitment/function, 6 genes (6.3%) in OC recruitment/function, 34 genes (35.4%) in EC recruitment/function and 13 genes (13.5%) in MSC differentiation/function. Notably, 10 genes (10.4%) are documented HSC niche components and 9 genes (9.4%) cancer cell niche components. It has to be considered that single genes could be assigned to more than one category. Forty-one genes (42.7%) are unrelated to any of the categories above ( Table 3, S4 Table).
To further corroborate the contribution of SC niche-related genes to the osteoblastic response, we compared both the OB-BMST and the Core OB-BMST with two SC niche signatures publicly available for the HSC [13] and for the developing prostate (uro-genital mesenchyme, UGM) [14]. The OB-BMST contains 14 genes (45%) of the 31 HSC-gene signature [13] and 208 genes (14.8%) of the 1405 UGM-gene signature ( Fig. 4A and B). As an effect of the curation, 5 (16%) of the HSC genes and 141 (11%) of UGM genes are retained in the Core OB-BMST ( Fig.4C and D, Table 4 and S5 Table). Importantly, 37% and 42% of the up-regulated genes of OB-BMST and Core OB-BMST, respectively, represent genes up-regulated also in the UGM signature.
The Core OB-BMST SC niche gene Epha3 is specific for the BM/B stroma reaction in osteoblastic bone metastasis The up-regulation of 4 representative SC niche genes, namely pleiotrophin (Ptn), Eph receptor a3 (Epha3), Cd109 and Slit homolog 3 (Slit3) in VCaP and C4-2B bone xenografts is confirmed by RT-qPCR analysis (Fig. 4E). Slit3 and, to a minor extent, Cd109, are significantly up-regulated also in the stroma of orthothopic VCaP (Fig. 4F). Ptn and, Cd109 are significantly up-regulated also in the stroma of subcutaneous C4-2B xenografts (Fig. 4G). In contrast, none of the 4 SC niche genes are significantly induced in the stroma of subcutaneous VCaP xenografts (Fig. 4G). A significant induction of these genes is also shown in the stroma of bone xenografts of the pro-osteolytic PC-3 cells (Fig. 4H), yet this is negligible when compared to the induction in the stroma of bones xenografted with osteoinductive C4-2B and VCaP cells. These results indicate that, at least in cancer cell xenografts, induction of Epha3 expression is the most specific for the osteoblastic response. Expression of the Core OB-BMST proteins PTN, EPHA3 and FSCN1 is restricted to human PCa bone metastasis In order to verify the translational value of Core OB-BMST genes, we investigated the protein expression of the two SC-niche components PTN and EPHA3, and of the most up-regulated gene FSCN1 in normal prostate and bone, and in primary and bone metastatic PCa.
Normal bone and hematopoietic marrow ( Strongly EPHA3-immunoreactive stellate-like cells are found scattered within normal BM (Fig. 5B) and, with increased density, in the tumor stroma of bone metastases (Fig. 5E). OBs, osteocytes and OCs are negative, as well as the majority of cancer cells. In normal prostate approximately half of the acini are EPHA3 negative, whereas in the other half discrete portions of the basal layer are positive (Fig. 5H). Positive staining is also found in rare myofibroblasts. In contrast, in primary PCa, cancer cells of all neoplastic acini are invariably EPHA3 negative, with no modification of the staining pattern in the stroma compartment (Fig. 5K). In normal bone stellate cells within the hematopoietic marrow are strongly FSCN1-positive ( Fig. 5C), but OBs, lining cells, osteocytes and OCs are FSCN1negative. In contrast, in PCa bone metastasis (Fig. 5F), active OBs, osteocytes and fibroblast-like stromal cells surrounding areas of cancer cell growth are FSCN1positive, while cancer cells are negative. In normal prostate, (Fig. 5I) FSCN1 immunoreactivity is detected in ECs of small vessels and in scattered, fibroblastlike cells. In PCa (Fig. 5L) the expression pattern of FSCN1 is similar to normal prostate tissue.
These findings further support the translational value of the OB-BMST and demonstrate that the Core OB-BMST contains genes that are specifically induced in the stroma of bone metastatic, but not of primary PCa.
The OB-BMST also contains gene signatures not unique to osteoblastic bone metastasis As outlined above, the curation strategy has led to the definition of 3 additional components within the OB-BMST.

Discussion
This is the first comprehensive transcriptome analysis defining the BM/B stroma reaction in xenograft models PCa-induced osteoblastic bone metastasis. The stroma specificity of this transcriptome, designated as OB-BMST, has been provided by the TCTP [11], a method allowing the analytical dissection of tumor stroma and cancer cell transcriptomes in situ, without physical separation of the two compartments.
Osteogenesis and angiogenesis are the key processes in the OB-BMST and Core OB-BMST The GO terms, the principal upstream regulators and the major effector cells of the OB-BMST and Core OB-BMST strongly indicate that osteogenesis and angiogenesis are the predominant processes in the BM/B stroma reaction of PCainduced, osteoblastic bone metastasis. The scarcity of genes related to OCs (4 genes encoding inhibitors of OC recruitment/activity and only two genes promoting OC recruitment) in the Core OB-BMST and the lack of genes encoding master factors stimulating OC recruitment (Rankl, Csf-1 and IL-8) in the OB-BMST further support the notion that, in osteoblastic bone metastases, there is no increase in OC recruitment [34].

Osteogenesis
The contribution of osteogenesis to the OB-BMST/Core OB-BMST could be anticipated and confirms the robustness of the two mouse models of osteoblastic bone metastasis adopted in this study. Markers and/or effectors of MSC and OB recruitment and function are highly enriched in the Core OB-BMST, where they represent 40% of the 96 most upregulated genes. This indicates that the Core OB-BMST better illustrates the bone context-specific, stromal response to osteoinductive cancer cells than the original OB-BMST. Furthermore, the finding that the osteogenesis-related genes among the top 30 Core OB-BMST genes are predominantly induced by VCaP cells seems to underscore the higher osteoinductive potential of these cells, as compared to C4-2B.
Cancer-associated fibroblasts (CAFs) are an additional cell population contributing to the OB-BMST as indicated by: a) the association with a previously retrieved myofibroblasts/fibroblasts gene signature [12], b) the enrichment of CAF markers (i.e., Pdgfrb and Sparc) and recruiting factors (Tgfb1, Tgfb3, Fgf2 and Pdgfbb) [35], c) the pivotal position of CAF-derived ECM proteins (Fn1 and collagens) [36] in the protein network and d) the expression of ASPN and POSTN by myofibroblasts in primary and bone metastatic PCa. This finding is in agreement with a previous study reporting that a CAF signature is overrepresented in bone metastases, as compared to lung, liver and brain metastases [37]. CAFs and OBs share a common cell-of-origin (MSC/pericyte) and some degree of marker expression, and their recruitment is induced by identical growth factors. Therefore, it is tempting to speculate that OBs are an additional, bonespecific population of CAFs and that the osteoblastic response is a tissue-specific manifestation of desmoplastic response to cancer cell invasion.
Fscn1, one of the 7-gene set, is the top up-regulated gene of the Core OB-BMST. Its encoded protein is critical for cell-matrix adhesion, cell interactions and cell motility. In normal adult tissue, it is expressed exclusively by dendritic cells [38]. In solid cancers, FSCN1 has been mainly associated with cancer cells and its stromal expression has been generally overlooked [39]. Here we show for the first time that, in bone metastasis, the expression of this protein is induced de novo in OBs and osteocytes, and in CAFs. This finding suggests FSCN1 as a candidate biomarker for osteoblastic bone metastasis.

Angiogenesis
The contribution of angiogenesis to the stroma response in osteoblastic bone metastasis is substantiated by: a) the up-regulation of markers of ECs of endosteal, sinusoidal BM vessels, namely endomucin [40] and laminin B1 [41] and of pericytes, namely Mcam [42] and nestin [43] in the OB-BMST, b) the prominent representation (35%) of markers and/or effectors of EC recruitment and function among the 96 most up-regulated genes of the Core OB-BMST, c) the presence of VEGFA and FGF2, the most important factors in EC recruitment [44], of PDGFB, a critical recruitment factor for pericytes in normal and tumor-induced neovasculature [45] and of CTGF, a factor coordinating angiogenesis in bone [46], among the activated upstream regulators of the OB-BMST. The finding that angiogenesis is a prominent process in osteoblastic bone metastasis is not surprising considering that, in physiological bone remodeling, it is tightly coupled to osteogenesis [40] and is required for primary and metastatic tumor growth [44]. It is also consistent with recent studies highlighting the relevance of sprouting BM neovasculature as a metastatic niche supporting cancer cell growth [47]. Nevertheless, it is surprising that the role of angiogenesis in osteoblastic bone metastasis is almost unexplored [4]. To our knowledge, only two studies have investigated the impact of anti-angiogenic therapy in mouse models of osteoblastic bone metastasis and shown inhibition of both osteoblastic response and tumor burden [48,49]. Further investigation is warranted for determining the relevance of angiogenesis and the efficacy of anti-angiogenesis in osteoblastic bone metastasis.
Ephb4 is an up-regulated OB-BMST gene that also emerges as one of the upstream regulators of the Core OB-BMST. The ephrinB2/Ephb4 axis is involved in coupling bone resorption to bone formation whereby OC-derived ephrinB2, by binding to Ephb4 on MSCs, induces their differentiation into OBs [50]. In line with this, interference with the ephrinB2/EphB4 axis by myeloma cells represses bone formation [51]. This axis plays also an essential role in angiogenesis [52]. Accordingly, the ephrinB2/EphB4 axis may mediate also coupling of angiogenesis and osteogenesis in osteoblastic bone metastases.

Osteoinductive PCa cells amplify the HSC niche
The dominant angiogenic and osteoblastic responses found in the OB-BMST/ Core OB-BMST are paralleled by up-regulation of several genes encoding proteins controlling homing/mobilization, self-renewal, dormancy or expansion of HSC and, therefore, to be considered as components of the HSC niche. Furthermore, the transcriptional components predicted to be active in the Core OB-BMST (Maz, Meis1 and Foxo4) have been implicated in both hematopoiesis [53][54][55] and angiogenesis [56][57][58].
The endosteal (OBs) and vascular (EC/pericytes) are considered the two most relevant HSC niches in the BM (reviewed in [59]) and both angiogenesis and osteogenesis are required for generation of the HSC niche [60]. Very recently, elegant studies on the spatial organization of HSC niches have revealed that the vascular and endosteal niches constitute a single structural and functional entity [41,60]. Collectively, these studies indicate that physiological angiogenesis and osteogenesis are merged spatially and temporally to build the HSC niche. Our findings suggest that, in osteoblastic bone metastasis, the angiogenic and OB responses are also finalized to the expansion of the HSC/metastatic niche. Therefore, they seem to validate a model of cancer cell-metastatic niche interaction, whereby cancer cells not only occupy and activate, but also amplify a pre-existing SC niche, thus fuelling further tumor growth [61].
Several molecules acting at the HSC niche interface, namely, kit-ligand (Kitlg or stem cell factor, SCF) [62], growth-arrest specific 6 (GAS6) [63], annexin A2 (Anxa2) [64], chemokine (C-X-C motif) ligand 12 (CXCL12), N-cadherin (Cdh2) [65,66] and trombospondin 1 (Thbs1) [47] have been shown to function also as components of the metastatic niche in models of osteolytic bone metastases. Only one of these genes, namely Cdh2, is present in the Core OB-BMST gene list. Most likely, this poor agreement reflects the fact that the HSC niche components of the Core OB-BMST are specific for the osteoblastic bone metastasis. In line with this, preliminary results from our laboratory show that the stroma transcriptome from xenograft models of osteolytic bone metastases differs substantially from the OB-OBMST and shares only 4 out of the 14 HSC genes present in the Core OB-BMST (JH, AW and MGC, unpublished).
Among the genes encoding components of the HSC niche, Ptn and EphA3 are especially interesting. Ptn has been reported to be secreted by OBs [67] and ECs [68] and to be integrated in new bone matrix [67]. Here we confirm PTN expression by active OBs, but not by the BM vasculature of human samples of osteoblastic bone metastasis. While its effects on the OB lineage are still disputed [67], Ptn is known to promote tumor angiogenesis, CAF recruitment and ECM remodeling [69]. Notably, Ptn has been demonstrated to be a component of the BM vascular niche regulating HSC self-renewal and retention in vivo [68].
Bidirectional signaling between Eph receptors and their membrane-bound ligands, ephrins, has been shown to play a role in cell migration and adhesion, and patterning of vascular, nervous and skeletal systems [70]. Notably, Ephs/ephrins are expressed in adult, epithelial SC niches where they modulate SC function [71]. However, the involvement of EphA3 in these processes is still unexplored. EphA3 mRNA expression in cultured mouse OBs has been reported, but with no indication of its function [72]. EphA3, together with EphA2 and EphA4, has been found to be expressed by mouse BM stromal cell lines and involved in homing and mobilization of HSCs [73]. We observed EphA3 protein expression by stellate-like cells within normal, human BM and in the tumor stroma of human bone metastasis, but not in bone or other BM stromal cells. This discrepancy may be possibly due to species differences and/or in vitro versus in vivo detection.
Our results and the observations above propose PTN and EPHA3 as HSC niche components that may also be part of the BM metastatic niche.

Osteoinductive PCa cells also induce their own niche
Remarkably, 37% and 42% of the up-regulated genes of the OB-BMST and Core OB-BMST, respectively, are part of the developing prostate SC niche (UGM) signature [14]. Yet, approximately 20% of these genes are also functionally linked to the HSC niche. An example for these is Ptn, overexpressed in the Core OB-BMST, and component of both the HSC [13], where it modulates HSC maintenance [68], and prostate SC [14] niches, where it regulates branching morphogenesis [74]. Interestingly, binding sites for Maz, one the transcription factors predicted to be active in the Core OB-BMST, have been also found enriched in the UGM [14].
Collectively, these findings suggest that osteoinductive PCa cells not only amplify a pre-existing, BM HSC niche, but also induce de novo an ectopic epithelial SC niche reminiscent of the organ of origin. A similar mechanism has been demonstrated in MCa metastasis to the lung, whereby tumor initiating cells educate stromal cells of the target organ to express ECM components of the SC niche of the developing mammary gland, such as POSTN, which then become components of the metastatic niche supporting stem-like cancer cells maintenance [75].
The combinatorial niche association of HSC and epithelial SC niches may create a ''soil'' suitable for stem-like cancer cell growth, but hostile for HSCs, thus explaining the hematopoietic aplasia (myelophthisis) occurring in bone metastasis [76] and in myeloproliferative neoplasia [77]. It may also correspond to the growth requirements specific for osteoinductive PCa cells. Furthermore, the capacity to induce de novo prostate epithelial SC niche components in the BM, but not in other tissues, as shown here, could also be a mechanism explaining the osteotropism of PCa cells.
The OB-BMST also contains gene signatures not unique to osteoblastic bone metastasis Inflammatory/wound healing/desmoplastic response A first fraction of the OB-BMST contains genes of inflammation/wound healing/ desmoplasia-related signatures. This is not surprising in view of the similarities between wound healing and cancer [78] and of the persistent activation of wound-healing and inflammatory programs in tumors [79]. Clearly, genes contained in this fraction are not cancer-restricted, thus limiting their use as cancer biomarkers. POSTN, one of these genes, has been proposed as a serum biomarker for bone metastasis [80]. However, its validity in this regard should be proven versus a clinical scenario of inflammation.
''Universal'' stroma response to cancer A second fraction of the OB-BMST contains genes shared by cancers with no or low propensity to metastasize to bone. This component is species, tissue and cancer cell type independent and thus seems to be a ''universal'' stroma response to cancer cells. Accordingly, genes part of this OB-BMST component could represent general biomarkers of cancer, but not of a specific cancer type.

Stroma osteomimicry in primary tumor as possible determinant of cancer osteotropism
An additional fraction of the OB-BMST contains genes shared by primary PCa and MCa known for their high propensity to metastasize to bone (osteotropism).
It has been hypothesized that the osteotropism of PCa is conferred by the capacity of PCa cells to aberrantly express OB-restricted proteins, such as osteopontin and osteocalcin. This phenomenon, named osteomimicry, would allow the cancer cells to thrive specifically in the bone microenvironment [81]. Recently, it has been suggested that osteomimicry by the stroma of the primary tumor, rather than by cancer cells, may also explain the osteotropism of MCa [37]. Our results may extend this notion to PCa and expand the list of molecules potentially involved.
Mcam, also known as Cd146, is of special interest in this regard. It is a cell adhesion molecule expressed at the intercellular junction of ECs and in activated T lymphocytes. A homophilic MCAM interaction between these cells mediates lymphocyte trafficking at inflammatory sites [82]. MCAM expression in human BM marks specifically pericytes of the sinusoids [42], an essential cell population of the HSC niche [43]. MCAM expression by cancer cells has been reported in most cancer types [83]. Increased stromal expression of MCAM parallels metastatic potential in osteotropic cancers, including PCa and MCa [84,85]. Importantly, MCAM is present in a gene signature predicting bone metastatic progression in MCa [86]. Therefore, MCAM-mediated homophilic interaction between cancer and BM stromal cells may represent a possible molecular mechanism of osteotropism.
The fraction of the OB-BMST common to the stroma response of primary PCa and MCa may contain a gene signature that may predict, similarly to MCAM, disease progression as bone metastasis. Therefore, it warrants to be verified in gene expression datasets, once available, of primary PCa from patients with or without bone metastatic relapse.

Translational significance
The OB-BMST/Core OB-BMST will serve as a reference list of physiological SC niche components to be validated as novel components of the BM/B metastatic niche. Besides their mechanistic significance, these molecules may represent additional therapeutic targets allowing interference with cancer cell homing, survival and growth in the BM/B. Indeed, it has been shown that cancer cells can be mobilized from the HSC niche into the blood by using HSC mobilizing agents [87]. Furthermore, anti-angiogenesis, by limiting the size of the SC/metastatic niches, may also interfere with survival and growth of osteoinductive PCa cells in the BM/B.
Modifications of the cancer-associated plasma proteome seem to be predominantly derived from the tumor microenvironment [88]. Therefore, the Core OB-BMST could also provide novel stroma-derived, serum biomarkers for the detection of bone metastatic cancer progression.

Ethics statement
The Ethical Committee of the Canton of Bern, Switzerland, approved the overall study protocol and tissue collection from patients (Nr 06/03). A written informed consent was obtained from each patient.
The Committee for Animal Experimentation and the Veterinary Authorities of the Canton of Bern, Switzerland approved the experimental animal protocols, anesthesia, surgical procedures and post-surgical analgesia (Permit Number: 15/ 07 and 6/10). Mice were housed in individual ventilated cages in strict accordance to the Swiss Guidelines for the Care and Use of Laboratory Animals. Autoclaved water and sterile mouse chow were provided ad libitum. For surgical manipulation, mice were anesthetized with a cocktail of medetomidin (1 mg/kg body weight), midazolam (10 mg/kg) and fentanyl (0.1 mg/kg) [34]. Postoperative analgesia with buprenorphine (0.1 mg/kg) was performed for 3 days following surgical intervention. Animals xenografted with human cancer cells were carefully monitored for signs of pain, distress and loss of body weight. Development of bone lesions was followed by radiography at two weeks intervals for VCaP and C4-2B and at weekly intervals for PC-3 cells. The size of subcutaneous and orthotopic tumors never exceeded 200 mm 3 . At the experimental endpoint mice were sacrificed by CO 2 euthanasia.

Xenografts
For intra-osseous xenografts, PCa cells with low/medium (C4-2B cell line) or high (VCaP cell line) capacity to induce an osteoblast response (osteoinductive cancer cells), and PCa cells (PC-3 cell line) with high capacity to induce an osteoclast reaction (pro-osteolytic cancer cells) and immortalized, non-tumorigenic human prostate epithelial Ep156T cells were inoculated in the BM cavity of the left tibia of male CB17 SCID mice [34]. Sham-operated animals (sham) and animals not subjected to surgery (intact) were used as controls. Development of bone lesions was monitored by radiography. Mice xenografted with VCaP and Ep156T cells were sacrificed after 6 weeks, and those inoculated with C4-2B after 8 weeks. PC-3 xenografts and their Ep156T controls animals were sacrificed after 33 days. Sham and intact animals were sacrificed at the 3 time points above. Xenografted and control tibiae were used either for RNA isolation or immunohistochemistry.
VCaP cells were used for orthotopic (intra-prostate) implantation, while C4-2Bluc and VCaP cells were used for subcutaneous implantation. For both types of xenografts, a suspension of 10E06 cells was mixed with collagen (Collagen Type I from rat tail, BD Biosciences, Allschwil, Switzerland). Prostates implanted with collagen pellets and dermal tissue served as control. Subcutaneous tumors were excised after 9 weeks and orthotopic tumors after 10 weeks. For more details see S1 File.

Microarray hybridization and data analysis
Labeled cRNA was prepared according to Affymetrix protocols and hybridized to Mouse Genome 430A 2.0 arrays (Affymetrix Ltd., High Wycombe, UK). The hybridization was performed with 30 mg cRNA for VCaP and Ep156T xenografts, and for intact, sham bones. Double the amount of cRNA was used for the C4-2B xenografts in order to correct for the high human RNA content in these xenografts. Three replicates were performed for all experimental groups except sham-operated bones and Ep156T xenografts, which were done in duplicates. Quality control of the microarray data was performed using RReportGenerator [92] and confirmed that all arrays used in the study were of good and consistent quality. Raw data are available on GEO (GSE22813). We computed statistical significance using the standard moderated t-test approach of the limma package [93] and computed local false discovery rates (FDR) [94] to adjust statistical significance to multiple-testing.
To reduce the impact of cross-species hybridization on gene expression signals, we re-defined probe-sets in order to use only probes considered mouse-specific (for a detailed description of the identification of cross-hybridising probe-sets see S1 File. After exclusion of cross-hybridizing probe-sets, the 3 control groups were analyzed for gene expression changes using a FDR of 0.2. Only a single gene, namely MMP12, was differentially expressed between intact bones and shamoperated/Ep156T-xenografted bones. Subsequently, differential gene expression between xenografted bones and sham-operated bones was calculated. The different human to mouse RNA ratio for the different samples was taken into account when selecting highly stringent thresholds in the statistical testing. Two different FDR thresholds were selected for VCaP (FDR#3E-05) and C4-2B (FDR#1E-05) due to different distribution of values.

Reverse transcription quantitative PCR (RT-qPCR)
RNA was reverse transcribed with M-MLV-RT (Promega, Wallisellen, Switzerland) and random primers (Roche Diagnostics, Rotkreuz, Switzerland) in the presence of a RNase inhibitor (Promega). mRNA expression was measured by RT-qPCR using ABI Prism Sequence Detection Systems (Applied Biosystems, Rotkreuz, Switzerland). The mouse and human specific gene expression assays are listed in the S1 File. RNA expression data were analyzed with two-tailed, unpaired t-test using GraphPad Prism version 6.0 (GraphPad Software Inc., La Jolla, USA).
Samples of normal bone were obtained from patients with coxarthrosis. PCa bone metastasis samples were obtained from iliac crest and femur, while MCa bone metastasis samples were obtained from the humerus. Samples of normal prostate tissue were obtained from cystoprostatectomy for bladder cancer. Samples of PCa of Gleason grades 3/4 were obtained from radical prostatectomy.

Validation of the stroma specificity of the OB-BMST
We confirmed the absence of 25 accepted pan-epithelial and/or prostate epithelial cell-specific markers in the OB-BMST. Furthermore, a literature review of 15 selected genes strongly up-regulated in the OB-BMST confirmed their stromal origin. The differential expression and stroma specificity of 7 representative genes was validated by RT-qPCR using mouse-specific probes. We also confirmed the stromal expression of two proteins (Aspn and Postn) encoded by up-regulated genes of the OB-BMST. More details concerning the stroma specificity of the OB-BMST are available in the S1 File.
Taken together these results show the reliability of our approach to analyze specifically the stroma compartment.
The upstream regulators of the ''common'' OB-BMST and Core OB-BMST were predicted by using default options of the Ingenuity Pathway Analysis software.
Genes of the ''common'' Core OB-BMST were analyzed for the overrepresentation of sequence motifs in their promoters within 2-kilobases of the transcription start site. Enrichment was calculated against a matched number of randomly selected genes and represented as p values versus this random set and also as an enrichment factor, which is the frequency of the transcription factor binding motif within the gene cluster versus its frequency within the background gene set as previously reported [14]. We report transcription factor motif using enrichment factors of 1.2-1.5 and p values ,10E-05. The motif database used for this analysis is supplied by TRANSFAC [96]. Results were further validated using the accessible web-based tool called GeneCodis (http://genecodis.cnb.csic.es/) [97].
Supporting Information S1 Figure. The OB-BMST overlaps with myoepithelial/myofibroblast signature and, to a lesser extent, with fibroblast and endothelial cell signatures.  Figure. A fraction of the OB-BMST is not specific for the BM/B response to osteoinductive PCa cells. Relative expression levels of Postn, Aspn, Sparcl1, Mcam, Pdgfrb, Fscn1 and Pmepa1 mRNA in intra-osseous, orthotopic and ectopic xenografts. A. VCaP intra-osseous xenografts (grey, n53) and corresponding sham-operated bones (white, n53) and C4-2B xenografts (black, n54) and corresponding sham-operated bones (white, n53). Values are shown as foldchange (mean ¡ SD) relative to contralateral bones. B. VCaP orthotopic xenografts (grey, n55) and sham (white, n54). Values are shown as fold-change (mean ¡ SD) relative to intact prostate. C. VCaP (grey, n53) and C4-2B (black, n55) subcutaneous xenografts. Values are shown as fold-change (mean ¡ SD) relative to intact skin. D. PC-3 intra-osseous xenografts (light grey, n56) and sham (white, n54). Values are shown as fold-change (mean ¡ SD) relative to contralateral bones (n53-4). *, P,0.01; **, P,0.001; ***, P,0.0001; ****, P,0.0001, ns 5 not statistically significant. Abbreviations: Postn, periostin; Aspn, asporin; Sparcl1, SPARC-like 1; Mcam, melanoma cell adhesion molecule; Pdgfrb, platelet derived growth factor receptor beta; Fscn1, fascin homolog 1; Pmepa1, prostate transmembrane protein, androgen induced 1. Normal prostate (G) is devoid of POSTN immunoreactivity both in the stroma and epithelial compartment. In contrast, in PCa (I) strong POSTN immunoreactivity is found in myofibroblasts over the entire tumor stroma, while cancer cells are negative. The myofibroblast identity of the POSTNimmunoreactive cells was confirmed in PCa by co-staining with a-smooth muscle actin (not shown). In normal bone (B), ASPN immunoreactivity is detected in OBs at sites of active bone formation, while lining cells, osteocytes and OCs are negative. Spindle-like cells within the hematopoietic marrow are also positive. In PCa (D) and MCa (F) bone metastases, strong ASPN immunoreactivity is detected in active OBs, and additionally in lining cells, osteocytes, and OB precursors. Stromal cells within areas of cancer cells are also ASPN-positive whereas cancer cells are ASPN-negative. In normal prostate (H) ASPN immunoreactivity is found in fibroblast-like cells and EC of small vessels, but not in epithelial cells. In the prostate, ASPN expression is also detected in cells, identified, in sequential sections, as neuroendocrine by expression of chromo-granin-A and synaptophysin and in Schwann cells (not shown). In PCa (J) the number of ASPN-positive, fibroblast-like cells is increased. In some specimens rare PCa cells are stained for ASPN (not shown). Insets represent a higher magnification of selected areas. Scale bar550 mm. Abbreviations: POSTN, periostin; ASPN, asporin. doi:10.1371/journal.pone.0114530.s003 (TIF) S1 Table Table. Overlapping genes of the OB-BMST with wound, desmoplastic response, non-osteotropic and osteotropic gene lists.