BMSCs are heterogeneous populations comprising various subpopulations with diverse properties (Horwitz et al., 2005). Except for Wharton’s jelly MSCs, BMSCs from rats and mice have demonstrated dose-dependent cytotoxicity to tumor cells (Otsu et al., 2009). Thus, specific BMSCs with anticancer capacity exist and may function according to the altered expression of some cytomembrane receptors (Ridge et al., 2017). It would make sense that BMP-2 could suppress OS through the proliferation of these specific BMSCs and that a BMP-2 and Wnt pathway autocrine loop (Figure 3) may be capable of explaining this process. The Wnt pathway is involved in diverse cellular events, including mitogenic stimulation, cell fate determination, differentiation, and proliferation (Huang and Niehrs, 2014; Yao et al., 2016; Steinhart and Angers, 2018). It is not surprising that the Wnt pathway, in particular the canonical Wnt pathway (i.e., the beta-catenin–dependent pathway), plays crucial roles in osteoblastic differentiation and osteogenesis (Liu et al., 2008; Wang et al., 2014b; Lerner and Ohlsson, 2015). Although the Wnt pathway is thought to inhibit MSC proliferation (Moon et al., 2018), an activated Wnt pathway facilitating BMSC proliferation has also been reported (Zhu et al., 2014). After the canonical Wnt pathway is activated, beta-catenin translocates from the cytoplasm into nuclei. In the nuclei, a complex consisting of beta-catenin and some transcription factors—for example, lymphoid enhancer-binding factor 1/T cell-specific transcription factor (LEF/TCF)—modulates the expression of target genes, including BMP2, RUNX2, and proliferation-related genes (Zhang R. et al., 2013). Conversely, BMP-2 can stimulate the accumulation of beta-catenin in nuclei (Yang et al., 2006; Hiyama et al., 2011), thereby activating the canonical Wnt pathway in turn. BMP-2–induced cell proliferation has been reported in murine preosteoblasts, rat BMSCs, and human pulmonary artery epithelial cells (de Jesus Perez et al., 2009; Rosen, 2009; An et al., 2017). The OS suppression properties of BMP-2 might result from the positive feedback of this loop via expansion of the specific BMSCs in the OS niche. In addition, aberrant activation of Wnt/beta-catenin signaling in OS cells has been detected (Chen et al., 2015). The identification of specific BMSCs in the OS niche is a precondition for OS suppression. Unfortunately, few studies about these specific BMSCs have been conducted in OS settings, so detailed information about their characteristics is still lacking.

Cancer is a disease arising from failed cell differentiation (Honma and Akimoto, 2007). Thus, differentiation-inducing treatments have been proposed. With this strategy, tumor cells differentiate back into normal cells instead of being eliminated by chemotherapeutics and/or radiation. One well-known differentiation-inducing treatment is all-trans-retinoic acid in acute promyelocytic leukemia (Huang et al., 1988). Notably, OS is recognized as an osteoblast differentiation disruption disease (Tang et al., 2008). OS cells have characteristic properties that resemble undifferentiated osteoblasts (Carpio et al., 2001; Postiglione et al., 2003; Haydon et al., 2007), and activating RB1 transcription has reversed the disrupted osteoblastic differentiation (Thomas et al., 2001). In addition, BMP-2 has been tested for its efficacy as a differentiation-inducing treatment. Rampazzo et al. (2017) successfully induced astroglial differentiation of glioblastoma stem cells using a BMP-2 mimicking peptide. Moreover, BMP-2 has suppressed tumors and promoted bone formation simultaneously: Wang et al. (2012) indicated that renal cell cancer was inhibited and bone formation was induced with the application of BMP-2. Furthermore, BMP-2 has reduced tumor volume, attenuated OS-induced pulmonary metastases, and stimulated bone formation (Xiong et al., 2018). Applying 30 μg of BMP-2 to OS-bearing mice also increased the transcription of osteogenic genes and promoted osteogenesis (Wang et al., 2013). Taken together, these data suggest that BMP-2 may play a therapeutic role in OS by inducing osteogenic differentiation of mutated BMSCs and/or OS cells.

The polarization of macrophages in inflammatory conditions suggests that the effect of BMSCs on OS may also transform mutually between tumor promotion and tumor suppression (Ridge et al., 2017). This hypothesis has been verified by Waterman et al. (2012) in a study that activated different cytomembrane receptors. The researchers claimed that activation of toll-like receptor-4 (TLR-4) conferred an antitumor effect on human BMSCs, which were named MSC1; after TLR-3 activation, however, the human BMSCs were converted to MSC2, which promoted tumor growth and metastasis (Waterman et al., 2012). Although myriad studies have indicated that TLR-2 and TLR-4 can enhance the expression of BMP-2 in BMSCs and accelerate bone formation (Yang et al., 2009; Su et al., 2011; Oliveira et al., 2017; Zhou et al., 2019), the effect of BMP-2 on the expression of TLRs is still equivocal. The dosage of BMP-2 and the state of BMSCs in the tumor niche may draw contrasting conclusions. Thus, the hypothesis that BMP-2 suppresses OS by affecting TLRs must be explored in more detail.

More research has reported that BMPs, especially a supra-physiological dose of BMP-2, induces tumorigenesis, not tumor suppression (Figure 4; Jin et al., 2001; Kang et al., 2011; Wu J. B. et al., 2011; Nishimori et al., 2012; Tian et al., 2017; Zhang et al., 2018). The physiological concentration of BMP is ∼2 ng/g of bone. In most clinical trials, supra-physiological doses (mg concentrations) of BMP-2 have been applied, and these doses may disturb the normal BMP-2 signal pathway (Arrabal et al., 2013; Oryan et al., 2014). After BMP-2 binds to its receptors on the cell surface, the BMP-2 signaling pathway is activated. In the canonical BMP-2 pathway, transcription of osteogenic genes, including RUNX2, and SP7 (OSTERIX), is upregulated. Although these two genes are vital for bone formation, an increasing body of evidence implies that they are also engaged in tumorigenesis. Normally, RUNX2 expresses during the cell cycle in healthy osteoblasts to disturb cell growth and induce osteoblast maturation (Pratap et al., 2003). Overexpression of RUNX2 has been found in patients with OS and is correlated to poor prognosis (Pereira et al., 2009; Sadikovic et al., 2010; Gupta et al., 2019). van der Deen et al. (2012) used chromatin immunoprecipitations to detect RUNX2 target genes in U2OS cells; results indicated that some motility-related genes were downstream of RUNX2 and that cell motility decreased after RUNX2 depletion. Furthermore, an elevated RUNX2 protein level may also be responsible for pulmonary metastasis. After RUNX2 activates SPP1 (OPN), the RUNX2 target gene encodes a secreted matricellular protein, thereby remodeling the bone matrix, which leads to tumor metastasis (Villanueva et al., 2019). RUNX2 may also account for the chemotherapeutic resistance of OS. When RUNX2 was silenced by si/shRNA, OS cells were more sensitive to doxorubicin (Roos et al., 2015). Another osteogenic gene, SP7, has not been associated with osteosarcomagenesis, but it has been described as a stimulus in other tumors (Dai et al., 2015; Yao et al., 2019; Ricci et al., 2020). This finding suggests that sustained activation of the BMP-2 pathway causing increased SP7 transcription may also precipitate OS in bone tissues.

BMP-2 may promote OS progression through modulation of the tumor microenvironment (TME), which plays an indispensable role in tumor progression (Hui and Chen, 2015; Yang et al., 2020). The bone microenvironment where OS grows is composed of hematopoietic stem cells, lymphoid progenitors, mature immune cells, bone cells, MSCs, mineralized extracellular matrix, and more (Tsukasaki and Takayanagi, 2019; Corre et al., 2020). The crosstalk in these items modulates the OS TME, which affects OS progression. Cancers are identified as “wounds that never heal” (Dvorak, 1986), so it is not surprising that MSCs are involved in tumor development, given the central role of MSCs in repairing wounds by altering the local inflammatory environment and secreting growth factors, immunoregulatory factors, and chemokines (Caplan and Correa, 2011; Wang et al., 2014a; Shi et al., 2017) after the tumor-specific tropism of MSCs (Kidd et al., 2009). However, MSCs are not always beneficial for healing; the fluctuation of their function depends on the milieu where they reside (Wang et al., 2014a). In the TME, MSCs can be converted into tumor-associated MSCs that have vast differences from normal MSCs (Le Nail et al., 2018) and that can promote tumor proliferation, migration, immunosuppression, and angiogenesis through extracellular vesicles (Quante et al., 2011; Baglio et al., 2017; Shi et al., 2017; Whiteside, 2018). In the OS niche, interleukin-6 (IL-6) and vascular endothelial growth factor (VEGF) secreted from BMSCs have been involved in OS progression (Tu et al., 2012; Zhang P. et al., 2013); BMSCs promoted pulmonary metastasis of OS by increasing the expression of CXC chemokine receptor 4 (CXCR4) and VEGF (Fontanella et al., 2016). Furthermore, extracellular vesicles, such as exosomes from BMSCs, are loaded with certain miRNAs involved in OS aggression and development (Xie et al., 2018). BMP-2, as a member of the TGF-β superfamily with the ability to recruit MSCs to inflammatory surroundings and the TME (Spaeth et al., 2008), may recruit BMSCs to OS, and BMP-2-induced chemotaxis has been reported in other conditions (Hiepen et al., 2014; Simões Sato et al., 2014; Pardali et al., 2018). BMP-2, particularly at high doses, induces inflammation (James et al., 2016), which may cause MSC homing as a result of inflammatory cytokines; in addition, MSCs have been recruited by BMP-2 through CXCR4, accelerating bone formation (Zwingenberger et al., 2014). Thus, BMP-2 might recruit BMSCs toward the OS phenotype. Together, these results suggest a tentative hypothesis. After BMSCs are recruited by BMP-2 to the OS niche, they will be educated directly or indirectly by OS cells. Afterward, the emergence of the educated BMSCs that can secrete some cytokines and growth factors will promote OS proliferation, migration, angiogenesis, and more.

Diverse OS cell lines applied in the research contribute to the confusion about results. Histologically, several OS subtypes with distinct characteristics have been confirmed. At the cellular level, various in vitro OS cell lines have been used in research; great differences in these cell lines have been verified. Saos2 cells appear more identical to normal osteoblasts than other OS cell lines, as osteoblastic markers can be detected in these cells. Conversely, osteocalcin, an important marker in bone mature, was hardly expressed in MG-63 and U2OS cells. However, matrix metalloproteinase-9 (MMP-9), a well-known cytokine for tumor migration and metastasis (Deryugina and Quigley, 2006), was positive in most MG-63 cells (Pautke et al., 2004). In other research, researchers (Mohseny et al., 2011) compared differences in differentiation, tumorigenesis, and protein expressions among 19 OS cell lines. Only OSA, IOR/OS9, and IOR/OS18 could differentiate into osteoblasts, chondrocytes, and adipocytes; 13 of the 19 cell lines could differentiate toward osteoblasts. This finding may explain why some researchers claimed that OS cell lines could not be induced into osteoblasts by BMP-2, whereas other studies reported opposite results (Haydon et al., 2007). Moreover, in these 19 OS cell lines, HOS-14B cells had the greatest capacities of tumorigenesis and metastasis. These inherent disparities between various OS cell lines, to some extent, account for the conflicting conclusions about the role of BMP-2 in OS progression.

Variations in MSCs are also ubiquitous. MSCs are heterogeneous populations consisting of a few subtypes with diverse characteristics; the differences may come from individual differences and species differences (Peister et al., 2004). The proposed definition of MSCs suggests that they must (1) adhere to plastic, (2) express special surface markers, and (3) differentiate along the osteogenic, chondrogenic, and adipogenic lineages (Dominici et al., 2006; Lindsay and Barnett, 2017). Commonly, CD34, CD31, and CD45 are negative on both human and mouse MSCs (Dominici et al., 2006); some markers, such as STRO-1 and CD271, are only detected on human MSCs (Lv et al., 2014); these are specific and can be found on other cell types (Kuhn and Tuan, 2010). CD29, CD51, CD73, CD90, CD105, and CD146 are universal in human and mouse MSCs (Sacchetti et al., 2007; Zhang et al., 2019). BMSCs are the most used MSCs in research; they are heterogeneous as well, which complicates the research and weakens the conclusions. Although some specific isolation kits based on the cell surface markers have been applied to clarify results, it remains hard to purify the homogeneous BMSCs, as MSCs share cell-surface markers and localization with pericytes (Crisan et al., 2008). With the development of biotechnology, the function and characteristic identification of a single cell are practicable. Single-cell RNA sequencing has been used to detect immune cell heterogeneity (Papalexi and Satija, 2018); Zhou et al. (2020) assayed the intratumoral heterogeneity and immunosuppressive microenvironment in advanced OS and demonstrated the complex variations in OS.

Furthermore, the dose and the delivery strategy of BMP-2 affect the research conclusions (Wu G. et al., 2011). Most of the reported disadvantages of BMP-2 result from overdosage. The effective dose of BMP-2 in osteoblastic differentiation of MSCs, which is dose-dependent, is just 25–100 ng/mL in vitro (Rickard et al., 1994; Lecanda et al., 1997). However, the working concentration of BMP-2 for in vitro or in vivo research is not distinguished, and most doses are supra-physiological, which may confound the results and cause adverse effects. The delivery pattern of BMP-2 is also crucial. A continuous and slow release, rather than a burst stimulation, is more bionic and more closely resembles physiological conditions. Most recent research has administered rhBMP-2 protein directly into the culture medium or intravenously, which may cause stress conditions for cells and tissues. The advantages of a sustained, low-dose release of BMP-2, including less inflammation and ectopic ossification, have been verified (Wildemann et al., 2004; Ji et al., 2010; Seo et al., 2017; Berkmann et al., 2020; Xin et al., 2020). The mitigatory inflammatory surroundings can reduce the risk of tumorigenesis as well, which makes low-dose BMP-2 application more reasonable.

Currently, most in vitro studies are carried out on traditional two-dimensional (2D) culture models (i.e., flask- and petri-dish-based cultures). However, these 2D models hardly mimic tumor cell biology because of tumor heterogeneity and different responses to secreted cytokines, growth factors, and methylation states of the cells. Moreover, the 2D cell culture systems cannot sufficiently simulate a three-dimensional (3D) physiological microenvironment, so they fail to provide physiologically relevant information regarding cell–cell interactions, cell–extracellular matrix interactions, growth factor synthesis, or physical and chemical cues to oncogenesis (Hickman et al., 2014; Berkmann et al., 2020). Furthermore, the results obtained from gene expression analysis and drug resistance also differ substantially between 2D and 3D cell culture models (Zhao et al., 2014; Costa et al., 2016; Henriksson et al., 2017; Zhou et al., 2017; Fontoura et al., 2020; Mao et al., 2020; Sun et al., 2020; Xie et al., 2021). The disadvantages of the 2D culture reduce attempts to understand the authentic role of BMP-2 that may play in the formation and pathology of OS.

In most OS studies, rodents, such as mice or rats, have been used as experimental animal models in addition to the patient-derived xenograft or cell line-derived xenograft models. Normally, OS is rare in mice and rats, and these models may present limited information or misinformation. The OS incidence in dogs is ∼27-fold higher than in humans, which makes the canine model a more useful model to the human OS for research (Simpson et al., 2017). To date, preclinical research using dogs as animal models has suggested that a combination of canine BMSCs together with rhBMP-2 treatment suppressed OS by increasing p53 and some other pro-apoptotic proteins (Rici et al., 2012, 2018). However, using dogs as animal models to study the effects of BMP-2 on OS development is not well accepted in Western countries because of social and cultural reasons.

Large-scale and multicenter cohort studies for evaluating BMP-2 treatment effects on OS progression remain unavailable. Although some clinical retrospective studies have suggested that BMP-2 used in spine fusion surgery was not involved in tumorigenesis (Fahim et al., 2010; Cooper and Kou, 2013; Lad et al., 2013), these studies were performed with small sample sizes and had insufficient follow-up times. Large-scale and multicenter cohort studies are needed to draw a scientific conclusion and establish the effects of the BMP-2 on patients living with cancer.