Bone secreted factors induce cellular quiescence in prostate cancer cells

Disseminated tumor cells (DTCs) undergo a dormant state in the distant metastatic site(s) before becoming overt metastatic diseases. In prostate cancer (PCa), bone metastasis can occur years after prostatectomy, suggesting that bone may provide dormancy-inducing factors. To search for these factors, we prepared conditioned media (CM) from calvariae. Using live-cell imaging, we found that Calvarial-CM treatment increased cellular quiescence in C4-2B4 PCa cells. Mass spectrometry analysis of Calvarial-CM identified 132 secreted factors. Western blot and ELISA analyses confirmed the presence of several factors, including DKK3, BMP1, neogenin and vasorin in the Calvarial-CM. qRT-PCR analysis of total calvariae versus isolated osteoblasts showed that DKK3, BMP1, vasorin and neogenin are mainly expressed by osteoblasts, while MIA, LECT1, NGAL and PEDF are expressed by other calvarial cells. Recombinant human DKK3, BMP1, vasorin, neogenin, MIA and NGAL treatment increased cellular quiescence in both C4-2b and C4-2B4 PCa cells. Mechanistically, DKK3, vasorin and neogenin, but not BMP1, increased dormancy through activating the p38MAPK signaling pathway. Consistently, DKK3, vasorin and neogenin failed to induce dormancy in cells expressing dominant-negative p38αMAPK while BMP1 remained active, suggesting that BMP1 uses an alternative dormancy signaling pathway. Thus, bone secretes multiple dormancy-inducing factors that employ distinct signaling pathways to induce DTC dormancy in bone.

TGFβ2, secreted from differentiated osteoblasts, have been reported to induce PCa cells to become quiescent in vitro 3 . Recently, Wnt5a, an osteoblast secreted factor that maintains hematopoietic stem cells in a quiescent state, was found to also promote PCa cell dormancy in bone 8 . In addition, Shiozawa et al. 9,10 showed that GAS6, a secreted factor that we have also identified in differentiated osteoblasts 3 , also regulates PCa dormancy in the bone marrow. These studies suggest that osteoblasts are one of the cellular components in the bone marrow that support PCa dormancy.
In this study, we looked for additional bone secreted factors that promote cellular dormancy of PCa cells. We used newborn mouse calvariae, which are enriched with osteoblasts 11 , as our study model and identified secreted proteins in the conditioned media. Several of these bone-secreted factors were examined for their effects on inducing cellular quiescence in PCa cells in vitro and for their signaling pathway(s) that leads to cellular dormancy.

Calvarial conditioned medium (Calvarial-CM) increases cellular quiescence in C4-2B4 PCa cells.
To identify bone secreted proteins, we used newborn mouse calvariae, which are enriched with osteoblasts 11 . Calvariae prepared from 2-5 day old newborn mice were cultured in BGJb medium containing 0.1% BSA for 48 h to generate calvarial conditioned medium (Calvarial-CM) (Fig. 1A). We have previously shown that this calvarial organ culture condition supports cell proliferation, calvarial bone formation and osteoblast differentiation 12 . To examine whether the Calvarial-CM contains dormancy-inducing activity for PCa cells, C4-2B4 cells were incubated with media containing either control BGJb media or Calvarial-CM and analyzed by live-cell imaging as (C) Quantification of % quiescent C4-2B4 cells that did not divide over 72 h relative to total cells examined (mean ± s.e.m.). n, number of cells monitored. N, number of independent experiments for control BGJb (N = 5), Calvarial-CM (N = 4). P values were by t test.
previously described 3 . Single cells were monitored for cell division over 72 h on a BioStation 3 . While proliferating cells typically undergo 2-3 cell divisions over 72 h under our experimental condition, dormant cells are characterized as viable, non-proliferating or slow-cycling 3,13,14 . In C4-2B4 PCa cells incubated in control media, the vast majority of control cells were observed to undergo several rounds of cell division, as illustrated by following one cell from F0 (T = 0 h) as it rounded up to divide into two F1 progenies (T = 2 h), which flattened out after cell division, to two more cell divisions into F2 (T = 43 h) and then F3 (T = 67 h) progenies (Fig. 1B, arrowheads). In contrast, there was a significant increase in the level of non-proliferating quiescent C4-2B4 cells to 12.8 ± 2.1% when incubated with Calvarial-CM relative to 4.2 ± 1.8% in control BGJb media (Fig. 1C). Immediately following live-cell imaging, cells were stained for the proliferation marker Ki67 and re-imaged on the BioStation. While proliferating cells were positive for Ki67, Calvarial-CM-treated nonproliferating C4-2B4 cells were Ki67 negative (Fig. 1B, right). These observations suggest that the Calvarial-CM contains factors that induced cellular quiescence of C4-2B4 cells.

Secretome analysis of bone conditioned medium (Bone-CM).
To identify potential dormancy-inducing factors secreted from calvariae, two independent calvarial preparations cultured in BSA-free medium, referred to as Bone-CM1 and Bone-CM2, to distinguish from Calvarial-CM that contained BSA, were concentrated 20-fold and analyzed by LC-MS/MS. Using a false discovery rate (FDR) of 1%, 416 and 244 proteins were identified from Bone-CM1 (Supplemental Table S1) and Bone-CM2 (Supplemental Table S2), respectively. Among these proteins, 114 and 109 proteins are secreted proteins from Bone-CM1 and Bone-CM2, respectively, based on UniProt mouse database. Using the UniProt database, we identified factors that are known to be secreted proteins and additional factors belonging to type I single-pass transmembrane proteins whose extracellular domain can be processed and released as a soluble fragment into the extracellular space. In this manner, 91 proteins were found in both samples, while 23 proteins were additionally found only in Bone-CM1 and 18 only in Bone-CM2 ( Fig. 2A). Thus, a total of 132 secreted proteins were identified in the Bone-CM (Table 1).

Figure 2.
Proteomics analysis of proteins from bone conditioned media. (A) Venn diagram of secreted proteins identified in Bone-CM1 versus Bone-CM2. Two independent batches of Bone-CM were subjected to LC-MS/MS analysis. Proteins were selected using 1% false discovery rate (FDR). We focused on the secreted proteins and obtained a total of 132 secreted proteins from the Bone-CM. We grouped the 132 secreted proteins into three major categories of extracellular matrix (ECM) proteins according to the Matrisome Project (http://matrisomeproject.mit.edu) 15 . We further curated our list of secreted proteins by searching against the matrisome atlas using the MatrisomeDB2.0 software 15,16 . (B) Matrisome core proteins include ECMglycoportiens, collagens and proteoglycans. (C) Matrisome-associated proteins include ECM-affiliated proteins and ECM regulators. (D) Other secreted factors include growth factors and other proteins. For a complete list of proteins identified in Bone-CM1 (n = 416) and Bone-CM2 (n = 244), see Supplemental Tables S1 and S2.  15,16 . The ECM protein atlas contains an inventory of matrisomes identified from normal lung, liver and colon tissues as well as metastatic colon, melanoma, myeloma and mammary tissues [15][16][17][18] , ovary 19 , endothelial cells 20 , pancreatic islet cells 21 , and various stem-cell derived ECMs 22 . According to the classification of the matrisome proteome 15 , the 132 bone-secreted proteins can be grouped into three major categories: "Matrisome Core" proteins (63 proteins), "Matrisome-Associated" proteins (42 proteins), and "Other Secreted Factors" (27 proteins). The 63 proteins that belong to the "Matrisome Core" are subgrouped to ECM glycoproteins, collagens and proteoglycans (Fig. 2B). The 42 proteins that belong to the "Matrisome-Associated" category are subgrouped to ECM-affiliated proteins and ECM regulator proteins (Fig. 2C). The rest of the other secreted factors are subgrouped to the IGF, TGF, cytokine, chemokine and growth factor families (Fig. 2D). Among the 27 other secreted factors, we found that 21 of the bone-secreted proteins have not been identified in other matrisomes. These 21 proteins are listed in Table 2 and include 8 growth factors and 13 other types of secreted factors. Among the growth factors are Dickkopf-related protein 3 (DKK3) 23 , also known as a tumor suppressor ("reduced expression in carcinoma REIC") [24][25][26] ; melanoma-derived growth regulatory protein (MIA) 27 , a factor that inhibits melanoma cell growth and is also known as "cartilage-derived retinoic acid-sensitive protein (CD-RAP)" with a role in cartilage differentiation 28 ; and neutrophil gelatinase-associated lipocalin (NGAL) 29 , a factor that promotes renal epithelial cell differentiation. Neogenin (NEO1), a protein that can bind and modulate BMP signaling 30,31 and is involved in the transition of proliferating cells to their differentiated state 32,33 ; leukocyte cell-derived chemotaxin-1 (LECT1), also known as chrondromodulin, a protein involved in inhibiting the www.nature.com/scientificreports www.nature.com/scientificreports/ growth of vascular endothelial cells and tumor cells 34 ; and vasorin (VASN), a protein that binds to and antagonizes TGFβ1 signaling 35 , are also found in the list of growth factors (Table 2). Together, these factors are potential candidates in the bone secretome that may play a role in regulating PCa activity in bone.

Biochemical analysis of secreted factors in Calvarial-CM. We used a 20X concentrated BSA-free
Calvarial-CM (Bone-CM) for Western blot analysis and examined the expression of five secreted proteins, i.e., BMP1 (110 kDa), DKK3 (38 kDa), MIA (15 kDa), NGAL (25 kDa), and PEDF (50 kDa). The predicted molecular weight of each protein is indicated in parenthesis. We detected NGAL and PEDF with molecular mass of 25 kDa and 50 kDa, respectively, as predicted, in Bone-CM (Fig. 3A). Western blot for BMP1 showed protein bands at 150 kDa and 110 kDa. Because BMP1 contains 5 potential N-link glycosylation sites (UniProt database), these results indicate that BMP1 in Bone-CM is likely glycosylated. DKK3 appeared as a major band at 50 kDa and a minor band at 75 kDa. As DKK3 also contains 4 potential N-link glycosylation sites (UniProt database), DKK3 in Bone-CM may also be glycosylated. MIA is a small molecular size protein that contains two disulfide bonds (UniProt database). We found that MIA also appeared slightly higher than predicted in three independent blots (see Supplemental Fig. S1), the basis of which is currently unclear. These bands were not detected in control BGJb medium (Fig. 3A), thus confirming the presence of these secreted factors in Bone-CM. Using ELISA, we further determined the concentrations of several bone-secreted factors in two independent mouse Calvarial-CM preparations, Calvarial-CM #1 and Calvarial-CM #2. We found that the Calvarial-CM #1 and #2 contains 2.7 and 3.5 ng/mL BMP1, 7.2 and 14.6 pg/mL NEO1, 0.2 and 0.6 ng/mL VASN and 10.5 and 19.4 ng/mL DKK3, respectively (Fig. 3B). The concentrations of these factors in Calvarial-CM are likely much lower than those in www.nature.com/scientificreports www.nature.com/scientificreports/ physiological or pathological conditions in vivo because in the experimental condition, the bone-secreted factors are secreted into the culture media and are vastly diluted. In contrast, PCa cells may encounter bone-secreted factors in much higher concentrations than the levels we observed by ELISA in the Calvarial-CM because PCa cells are found to be adjacent to the tumor-induced bone 36 in the in vivo bone microenvironment. Together, the Western blot and ELISA analyses confirm the presence of candidate bone-secreted factors in Calvarial-CM.

Expression of secreted factors from different types of cells in calvariae.
The newborn mouse calvariae contain osteoblasts as well as osteoclasts, osteocytes and fibroblasts. We next determined whether the candidate bone-secreted factors identified by MS/MS are mainly expressed in mouse osteoblasts or other cell types within the calvariae. Primary mouse osteoblasts (PMOs) were isolated from newborn mouse calvariae by digestion with collagenase plus trypsin as described in Materials and Methods (see Fig. 1A). RNA was prepared from PMOs or total calvariae. The expression of the candidate bone-secreted factors in PMOs or calvariae was analyzed by quantitative RT-PCR using mouse-specific primers (Table 3). A higher PMO/calvariae mRNA ratio indicates that a factor is likely to be expressed by osteoblasts. Using this criteria, we found that neogenin, vasorin and DKK3 are mainly expressed in PMOs, while MIA, LECT1 and NGAL are mainly expressed in other cells in the calvariae (Fig. 3C). We also analyzed the expression of two known matrisome-associated ECM regulator  www.nature.com/scientificreports www.nature.com/scientificreports/ proteins that have important functions in bone: BMP1, a metalloproteinase that regulates the formation of the bone matrix 37,38 , and pigment epithelium-derived factor (PEDF), a potent antiangiogenesis factor 39 that was also shown to promote osteoblast differentiation and mineralization of the bone matrix (Table 1) 40 . We found that BMP1 is mainly expressed in PMOs while PEDF is expressed by other cells in the calvariae (Fig. 3C). As controls www.nature.com/scientificreports www.nature.com/scientificreports/ for known osteoblast-expressed proteins, we showed that two TGFβ family members TGFβ1 and TGFβ2 are expressed by PMOs. TGFβ2 was previously shown to be a dormancy-inducing factor while TGFβ1 was not 3,14 . TGFβ1 also did not affect cellular dormancy in HNSCC HEp3 cells 14 , breast cancer cells 20 or PCa cells 3 .

Dormancy-inducing activity of bone-secreted factors.
To examine the potential dormancy-inducing activity of candidate bone-secreted factors, recombinant proteins corresponding to select factors were first screened for their ability to induce cellular quiescence in PCa cells. We conducted a dose response analysis for each selected bone-secreted factors. We used factor concentrations reported in the literature [41][42][43][44][45][46][47] and tested several concentrations to ensure that the dosages used do not cause toxicity, i.e., cell death, in PCa cells using live-cell imaging. Based on the response, we selected the following concentrations for further study: DKK3 (10 µg/ mL), BMP1 (0.4 µg/mL), vasorin (0.25 µg/mL), neogenin (0.5 µg/mL), MIA (0.1 µg/mL), or NGAL (0.25 µg/mL) (Fig. 4A, arrowheads). Live-cell imaging showed that there was an increase in non-dividing quiescent cells when C4-2B4 cells were incubated with recombinant human DKK3, vasorin, neogenin or BMP1 (Fig. 4B, left), as illustrated by examples of cells that exhibited a "dormant" phenotype after treatment with DKK3 (cell a), vasorin (cell b), neogenin (cell c), or BMP1 (cell d). In contrast, the majority of control C4-2B4 PCa cells underwent several rounds of cell division, as illustrated by following one cell from F0 (T = 0 h) as it divided into F1 (T = 1 h), F2 (T = 22 h) and then F3 (T = 70 h) progenies (Fig. 4B, asterisks). Of note, although these non-proliferating quiescent cells did not divide, they exhibited a low level of movement with cell shape changes during time-lapse recording (Fig. 4B, left).
We further confirmed the dormancy status of the non-dividing quiescent cells by co-immunostaining for Ki67 and p27. Proliferating cells express the proliferation marker Ki67 in all phases of the cell cycle (G1, S, G2, M) but not during quiescence (G0). In quiescent breast cancer cells, the cell cycle inhibitor p27 has been shown to be upregulated and translocated into the nucleus 48,49 . Thus, cellular dormancy is characterized by Ki67 negativity and p27 positivity. Control cells that underwent several rounds of cell division over 72 h were found to be Ki67 positive and p27 negative (Fig. 4B, right). In contrast, quiescent cells a -d that did not divide upon treatment with DKK3, vasorin, neogenin or BMP1 were Ki67 negative and showed nuclear translocation of p27 (Fig. 4B, right). The combination of these morphological, cell cycle and biochemical analyses suggests that DKK3, vasorin, neogenin and BMP1 induced cell cycle arrest leading to cellular dormancy in C4-2B4 cells.
The levels of dormancy induced by the select bone-factors were quantified directly by cell counting. By combining 72 independent live-cell imaging experiments, around 1,700 control cells and 3,500 cells treated with various recombinant factors were monitored at the single cell level (Fig. 4C). Amongst the selected factors, we found that DKK3 significantly increased the level of quiescent, non-dividing C4-2B4 cells by about 3-fold, to 20% when compared to 7% in untreated control cells (Fig. 4C). This level of DKK3-mediated dormancy induction was comparable to that observed with the dormancy-inducing factor BMP7 50 , which was used as a positive control. Two other factors, vasorin and neogenin, were also effective in inducing cellular quiescence, though to a lesser extent than DKK3. Of the three factors expressed by other calvarial cells, MIA and NGAL exerted a modest effect on cellular quiescence on C4-2B4 cells while PEDF effects did not reach stastistical significance relative to control cells. BMP1, which differs from BMP2 -BMP16 in that it is not a TGFβ-like protein, but is important for the formation and mineralization of bone 38,51 , was found to induce a significant 18% of C4-2B4 cells to become quiescent. These analyses show that a variety of bone-secreted factors display different extent of dormancy-inducing activity on C4-2B4 cells.
We tested the dormancy-inducing activity of bone-secreted factors on another PCa cell line C4-2b 52 . Although C4-2b and C4-2B4 cells are both sublines derived from LNCaP cells 53,54 , we previously observed that these cells show differences in cell morphology and motility under the same culture conditions 3 . That C4-2b and C4-2B4 cells have distinct properties is supported by recent reports that these cells exhibit differences in their epithelial-to-mesenchymal transition state 55 . We found that DKK3, BMP1 and vasorin induced significant cellular quiescence to 18-24% relative to 7% in control untreated C4-2b cells (Fig. 4D), while neogenin's effects were modest. Taken together, these results show that multiple factors secreted by different cell types within bone can induce cellular dormancy in PCa cells.
Phospho-p38MAPK in dormancy signaling. Activation of p38MAPK through phosphorylation is one of the major signaling pathways utilized by dormancy-inducing factors 14,56 . We examined whether these newly-identified dormancy factors induce dormancy through p38MAPK phosphorylation 3 . Immunofluorescence imaging followed by quantification showed that phospho-p38MAPK (p-p38) became significantly enriched in the nucleus of C4-2B4 cells in response to DKK3 (Fig. 5A), vasorin (Fig. 5C) and neogenin (Fig. 5D) stimulation, indicating that these factors induced p38MAPK activation. Both DKK3 and vasorin stimulated p-p38 nuclear translocation within 3 h with sustained activation maintained up to 24-48 h in C4-2B4 cells (Fig. 5A,C), while neogenin stimulated a transient p-p38 nuclear translocation with a maximum at 3 h followed by a decline to basal levels by 48 h (Fig. 5D). Interestingly, although BMP1 activated dormancy induction (Fig. 4C), BMP1 did not stimulate p-p38 nuclear translocation in C4-2B4 cells (Fig. 5B), suggesting that an alternative signaling pathway for dormancy induction is used by BMP1. A similar pattern of p-p38 nuclear translocation in response to DKK3 (Fig. 5E), vasorin (Fig. 5G), and neogenin (Fig. 5H) was observed in C4-2b cells. As also observed in C4-2B4 cells, BMP1 did not stimulate p-p38 nuclear translocation in C4-2b cells (Fig. 5F), although BMP1 was effective in stimulating dormancy induction in C4-2b cells (Fig. 4D). These results suggest that DKK3, vasorin and neogenin activated canonical p-p38MAPK dormancy signaling, albeit with differences in the kinetics of p38MAPK activation, and BMP1 employs an alternative signaling pathway to dormancy induction.

Discussion
We have identified a unique set of bone-secreted factors that can induce cellular dormancy in PCa cells. These factors belong to different protein families. While many of these factors are found to be expressed by osteoblasts, some are expressed by other stromal cells in the calvariae. Most of these factors induce cellular dormancy in PCa cells through activation of the p38MAPK signalling pathway, albeit with different kinetics of p38MAPK activation. Interestingly, BMP1-induced cellular dormancy is mediated through a p38MAPK-independent pathway. Thus, dormancy induction from the bone microenvironment involves factors secreted from multiple cell types with various signal transduction mechanisms. Such variations may result in varied dormancy duration seen in the clinical setting. These observations may contribute to the future development of strategies for the prevention or treatment of bone metastasis.
The transforming growth factor-β (TGFβ) family proteins are the first group of proteins shown to exhibit dormancy-inducing activity 57 . These proteins include TGFβ2 3,14 , GDF10 (BMP-3b) 3 , and BMP7 50 . Other factors that exhibit dormancy-inducing activities include GAS6 9,10 , thrombospondin (TSP-1) 20 , Wnt5a 8 and LIF 58 . The panel of bone-secreted factors identified in our studies expands this growing family of dormancy-inducing factors. One of the secreted factors is DKK3, a divergent member of the DKK family of secreted Wnt signaling antagonists 59 . Overexpression of DKK3 has been shown to inhibit prostate tumor growth and metastasis by limiting TGFβ/Smad signaling 24,25,60 . DKK3 was also found to be downregulated in a range of human cancers 59,61 . Interestingly, DKK3 does not seem to act as a Wnt signaling antagonist 59,62 . Recent studies in a bone fracture and repair model using a DKK3-GFP reporter mouse suggest that DKK3 is expressed in the mesenchymal progenitor cells in the periosteum and in the bone marrow and is required for cartilage cell development by blocking osteogenesis 63 . Our studies revealed a potential role for DKK3 as a dormancy-inducing factor that signals through p38MAPK to modulate PCa growth in bone. Vasorin, another bone-secreted factor, is a type I single-pass transmembrane protein 35 that contains tandem leucine-rich repeats, an epidermal growth factor (EGF)-like motif, and a fibronectin type III-like motif in the extracellular domain, which are motifs involved in protein-protein interactions. It has been shown that the extracellular domain of vasorin can be cleaved into a soluble ectodomain that binds directly to TGFβ1 and antagonizes TGFβ1 signaling 35,64 . Although our study showed that vasorin induces PCa cell dormancy through p38MAPK in vitro, whether antagonizing TGFβ1 signaling contributes in part to how vasorin induces cellular quiescence of PCa cells is unknown. Another dormancy-inducing factor is BMP1. The ECM-regulator BMP1 is a member of the BMP1/Tolloid-like family of metalloproteinase and is found in many matrisomes (Table 1). In bone, BMP1 is highly expressed in areas of bone matrix formation, and mutations in BMP1 result in osteogenesis imperfecta or "brittle bone disease" in humans 51 . BMP1 is reported to affect BMP/TGFβ pathways indirectly through the release of latent binding proteins that tether BMP2/4 as well as TGFβ1-3 in the ECM matrix 38,65 . Our studies showed that BMP1 contributes to dormancy in PCa cells through a p38MAPK-independent pathway. Whether BMP1's metalloproteinase activity is involved in its dormancy-inducing activity is not known.
We note that the empirically-determined concentrations of recombinant factors used in our studies are about 1,000 times higher than those found in Calvarial-CM. As PCa cells are found adjacent to the tumor-induced bone 36 , it is likely that PCa cells may encounter bone-secreted factors in higher concentrations in vivo. Indeed, the high factor concentrations used in our studies are within the range reported in the literature. As an example, similar concentrations of human DKK3 at 10-50 µg/mL were used to block the proliferation of human kidney KPK1 cancer cells 41 ; human BMP1-3, an alternatively-spliced variant of BMP1, at 150-1000 ng/mL was observed to increase the expression of p21, a cell cycle inhibitor, in human embryonic kidney 293 (HEK293) cells 42 ; human MIA at 50 ng/mL was used to inhibit the invasive activity of melanoma cells 45 ; and PEDF at 250 ng/mL was used to induce osteoblastic differentiation of human mesenchymal stem cells 47 .
We found that different cell types, besides osteoblasts, within the tumor microenvironment are also involved in the induction of tumor cell dormancy. That bone marrow stromal cells secrete factors to induce cellular dormancy has been previously reported. Kobayashi et al. 50 showed that HS5 bone marrow stromal cells secrete BMP7 that inhibits PC3 PCa cell proliferation. We also confirmed the dormancy-inducing activity of BMP7 on C4-2B4 PCa cells (Fig. 4C), but found that BMP7 is not expressed by either osteoblasts or other calvarial cells (data not shown). Ghajar et al. 20 showed that endothelial cells in mature vessels secrete TSP-1 that leads to breast tumor cell dormancy. Price et al. 66 showed that dormant breast cancer cells are tethered to the bone marrow vascular niche through C-X-C chemokine receptor type 4 (CXCR4) interactions with stromal cell-derived factor (SDF-1) that is www.nature.com/scientificreports www.nature.com/scientificreports/ enriched in the perisinusoidal vascular region. Thus, various dormancy-inducing factors secreted by different cell types within the tumor microenvironment are involved in the induction of tumor cell dormancy.
We have previously performed proteomics analysis of osteoblast exosomes, which we termed osteosomes, isolated from undifferentiated/proliferating versus differentiated/mineralizing primary mouse osteoblasts 67 . It is possible that the proteomics analysis of Calvarial-CM contains both bone-secreted soluble factors and proteins derived from osteosomes. We compared the 132 secreted proteins identified in Bone-CM (Table 1) with the 373 proteins previously identified in osteosomes 67 . We found that 25 proteins (18.9%) in the Bone-CM were also found in osteosomes (Supplemental Table S3). These include the cytokine MIF; chemokines S10A9, S10AB and S10A6; ECM glycoproteins TENN, AEBP1 and LAMB1; and ECM regulator proteins SERPH, MMP13, CD109, CATB and A2MP (Supplemental Table S3). Notably, the bone-secreted factors DKK3, BMP1, vasorin, neogenin, MIA and NGAL that exhibited dormancy-inducing activity in PCa cells are not found in the osteosomes. Whether osteosomal proteins can mediate dormancy induction in PCa cells remains to be determined.
Only a limited number of signaling pathways that mediate dormancy responses have been reported. A change in mitogenic signaling from the Ras-induced extracellular signal-regulated kinase (ERK) pathway to the p38MAPK pathway has been shown to lead to a switch from proliferation to cellular dormancy [68][69][70] . Our studies together with reports by others showed that the p38MAPK signaling pathway is involved in PCa cell quiescence mediated by TGFβ2, GDF10, BMP7, DKK3, vasorin and neogenin (Figs. 5 and 6) 3,14,50 . Interestingly, we note that factors that elicited a more sustained p-p38MAPK activation and nuclear translocation for up to 24-48 h, such as DKK3 and vasorin, generated a higher level of dormancy induction than those that generated a transient p-p38MAPK response, such as neogenin, in PCa cells (Fig. 5). While the significance of the different kinetics of p38MAPK activation is currently unclear, they may have effects on the duration of the dormancy state. How p38MAPK activation imparts specificity to dormancy induction is also unknown. In our previous study, we found that p38MAPK phosphorylates RB at the novel N-terminal S249/T252 sites 71 to block PCa cell proliferation, a process that leads to an increase in the cell cycle CDK inhibitor p27, followed by G0-G1 cell cycle arrest and cellular quiescence 3 . While p38MAPK represents a canonical dormancy signaling pathway, other pathways that signal to cellular dormancy also exist. Johnson et al. 58 reported that the leukemia inhibiting factor LIF, a member of the IL-6 cytokine family, induces breast cancer cell dormancy by activating STAT3 signaling molecules. Notably, we found that BMP-1 induces cellular dormancy through a p38MAPK-independent pathway (Fig. 6), which remains to be identified.
How dormancy-inducing factors regulate the dormancy gene network is not clear. Aguirre-Ghiso 72 and Segall's 73 groups have reported that p38MAPK signaling reprograms tumor cells to acquire a quiescence program through regulating a network of transcription factors and target genes ("dormancy signature genes") whose expression promotes quiescence of metastatic head and neck squamous cell carcinoma HEp3 cells and breast cancer cells, respectively. These genes constitute a dormancy transcription network that confers cellular dormancy. The transcription factors include NR2F1/COUPTF1 73,74 , BHLHE41/Dec2/Sharp-1 14,73 and FOXM1 72,73 , which may serve as a "switch" for converting active tumor cells into "dormant" cells. BHLHE41 has been shown to suppress the metastasis of aggressive triple-negative breast cancers through the degradation of hypoxia inducing factor HIF that promotes tumor angiogenesis 75,76 . How p38MAPK signaling leads to BHLHE41 gene expression and which dormancy signature genes might be regulated by BHLHE41 in PCa cells remain to be elucidated.
Delineating the mechanisms that lead to exit from dormancy is important for preventing tumor relapse. Similar to dormancy entry, dormancy exit can be due to intrinsic properties of the tumor cells that overcome dormancy and/or extrinsic signals encountered in the tumor microenvironment 77,78 . An increase in the expression of the BMP inhibitor Coco has been shown to render highly metastatic MDA-MB-231 breast cancer cells non-dormant 79 as well as to promote metastatic outgrowth of 4TO7 mammary tumor cells in lung by inhibit-ing BMP signaling, thus allowing the reactivation of tumor cells from dormancy 79 . Increased expression of VCAM1 in mammary tumor cells has been shown to induce bone lysis, resulting in the release of growth factors from bone that enable tumor cells to exit dormancy 4,80 . Increased signaling through β1 integrin was shown to induce both D2A1 breast cancer cells 48 and LuCaP PDX PCa cells 81 to transition from quiescence to proliferation. Additionally, alterations in the tumor microenvironment as a result of hormonal changes, aging or therapies 77,78 may lead to a loss of dormancy-inducing factors. Together, these observations suggest that entry into and exit out of cellular dormancy may involve distinct pathways.
Currently, therapy strategies that target dormant tumor cells, which are resistant to chemotherapies 82,83 , are limited 77,83 . It seems that keeping dormant tumor cells dormant indefinitely or eradicating dormant tumor cells while they are dormant would be safer than reactivation followed by chemotherapy. In this context, maintaining the dormancy-conferring microenvironment would be an important aspect of the prevention strategy. Identification of mechanisms that support tumor cell survival during prolonged periods of dormancy may elucidate strategies to target dormant tumor cells. Because dormant tumor cells are responsible for cancer relapse, further understanding of tumor cell dormancy mechanisms is warranted.