Bone metastatic disease is the lethal end-stage of aggressive prostate cancer [1
]. For decades, almost all patients with bone-metastatic prostate cancer have been treated with androgen deprivation therapy (ADT). This reduces bone pain and temporarily retards metastatic growth, but, after some time, the disease relapses to castration-resistant prostate cancer (CRPC). New treatments for CRPC have become available in the clinic, including more efficient blockage of androgen synthesis and androgen receptor (AR) signaling, novel chemotherapies, immunotherapies, and bone-targeting treatments that all temporarily retard disease progression, although with different efficiency in different patients [2
]. Recent studies have suggested the existence of molecularly diverse subtypes of prostate cancer [3
] and our own studies have shown similar molecular diversities among bone metastases [5
]. Thus, possibilities may exist for more individualized treatment of patients with metastatic prostate cancer than previously recognized, but therapy-predictive markers are lacking.
Bone metastases in prostate cancer patients are generally classified as osteoblastic with increased bone formation, and this is also the reason why bone scintigraphy is the preferred method to assess metastasis in those patients [9
]. In contrast, other cancers such as breast, lung and renal cancer usually form osteolytic metastases with increased bone resorption [10
]. This classification, however, is probably oversimplified, as prostate cancer metastases in addition to increased bone formation also show dysregulated bone resorption [10
]. The skeleton in healthy individuals preserves its structural and functional integrity through constant bone remodeling. At the bone remodeling sites, bone mass is maintained by a tightly controlled balance between bone resorbing osteoclasts and bone forming osteoblasts [14
]. This balance is altered when prostate cancer cells colonize the bone, resulting in increased formation of immature, less mechanically competent bone, often prone to fractures [15
]. Whether a pronounced bone response is seen in all patients with metastatic prostate cancer or only in a subset of individuals with a particular molecular subtype of the disease, and whether ongoing bone remodeling influences response to different cancer treatments, are unknown.
The sclerotic phenotype of bone metastases seen in prostate cancer patients suggests a possible association between locally excessive bone formation and AR activity. Androgens stimulate AR signaling in osteoblasts, causing increased bone formation, similar to how estrogens act through estrogen receptor α (ERα) in osteoblasts [16
]. In addition, estrogens also decrease osteoclastic life span in trabecular and endocortical bone through pro-apoptotic signaling, while androgens do not seem to have any direct effects on osteoclasts [16
]. Androgens may instead be converted into estrogens and thereby indirectly inhibit bone resorption via ERα in osteoclasts [17
]. Accordingly, ADT reduces bone mineral density in prostate cancer patients [18
]. Most metastases in CRPC patients, however, maintain AR signaling despite castrate levels of testosterone in the circulation, and some also show intra-tumoral steroid levels high enough for AR activation [20
We have previously studied tumor cell AR activity in prostate cancer bone metastases [7
] and mechanisms driving AR activity in CRPC, such as intra-tumoral steroidogenesis and expression of constitutively active AR variant 7 (AR-V7) [5
]. While most prostate cancer metastases were defined as AR-driven, some (about 20%) seemed to be non-AR-driven [7
]. Patients with non-AR driven metastatic disease have few treatment options, but might benefit from therapies targeting the tumor microenvironment such as immunotherapy, bisphosphonates, receptor activator of nuclear factor kappa-Β ligand (RANKL) inhibitors, or radioisotopes [23
]. The current study specifically investigated bone cell activity in clinical bone metastases (based on expression levels of a predefined set of osteoblast, osteoclast, and osteocyte marker genes) in relation to tumor cell AR activity [7
], in order to gain novel insight into biological heterogeneities of possible importance for patient stratification into bone-targeting therapies.
Prostate cancer bone metastases are generally classified as sclerotic, due to radiological observations of increased bone volume/density in comparison to healthy bone [9
]. Nevertheless, osteolytic activity is also observed within metastatic prostate cancer patients, as judged from elevated serum levels of bone resorption markers [13
]. It has been hypothesized that the sclerotic phenotype of prostate cancer may originate from intra-tumoral steroidogenesis in castration-resistant prostate cancer and thus preserved androgen levels and AR activity [16
]. In the current study, variable bone cell activity was observed between samples of clinical bone metastases from prostate cancer patients and, strikingly, high osteoblast activity was found coupled to corresponding high osteoclast activity (and vice versa) within individual samples. In contrast to what could have been expected, bone cell activity was inversely correlated to transcriptional AR activity in tumor cells and to patient serum PSA levels. Ongoing bone formation was primarily observed in bone metastases with high RUNX2 expression, possibly stimulated by BMP signaling, as assessed and confirmed by robust BMP4 activity in bone forming metastases.
Tumor cell-produced factors can influence bone remodeling, either by stimulating bone resorption or by activating osteoblasts, both processes ultimately resulting in release of factors stimulating tumor growth. This vicious cycle potentially enhances metastasis growth and aggressiveness via growth factors such as BMPs, Endothelin-1, PDGF and TGFβ [11
]. BMPs activate specific serine-threonine kinase receptors that transmit signals by activating receptor-specific SMADs that complex with SMAD4 and translocate into the nucleus to activate transcription [39
]. In complex with RUNX2 the SMADs activate transcription of genes, such as ALPL
, associated with osteoblastogenesis and bone formation [40
]. We found BMP4
and other BMP related factors, accompanied by increased expression levels of all the above mentioned genes, to be associated with detectable, ongoing bone formation in the examined metastases. The net effect of high bone cell activity on bone formation, however, was not clarified in the current study. Evaluation of molecular evidence of bone cell activity in relation to corresponding bone scans might help in drawing conclusions, as would comparisons to serum levels of bone remodeling markers (reviewed in [41
]), but such data were unfortunately not available.
It would also be interesting to measure intra-metastatic androgen and estrogen levels in the examined cases. Androgens and estrogens are important for maintaining the balance between bone resorption and bone formation [17
]. In osteoblasts, activation of the AR or the estrogen receptor α (ERα) prevents these cells from undergoing apoptosis. In mature osteoclasts, activation of the ERα promotes apoptosis and activation of ERα or AR in osteoblasts downregulates the expression of RANKL and thereby decreases osteoclastogenesis (reviewed in [42
]). The AR, ERα, and RUNX2 also interact to modulate transcription of bone related genes (summarized in [43
]). By all these mechanisms, sex hormones enhance bone mass and ADT would in general result in reduced bone formation and increased bone resorption. The observed situation in clinical bone metastases from CRPC patients, with excessive bone formation in many cases, might be explained by restored androgen levels due to local steroidogenesis and a parallel activation of osteoblast and inactivation of osteoclasts [44
]. The net effect on bone formation might also depend on the presence of tumor-derived factors stimulating osteoblast differentiation and matrix formation, such as the BMPs. It is noteworthy that bone metastases with ongoing bone formation showed higher tumor cell AKR1C3 immunoreactivity [6
] than cases without detectable bone formation, possibly due to intra-tumoral conversion of androstenedione to testosterone and subsequent osteoblast activation.
Besides bone formation, BMPs might be involved in a diverse range of developmental processes, including cell proliferation, differentiation, apoptosis, and angiogenesis [47
]. BMP2 has been implicated in promoting epithelial-to-mesenchymal transition and in inhibiting tumor cell apoptosis [49
]. BMP4 has been shown to promote prostate tumor growth in bone by stimulating osteogenesis [50
] and, specifically, to stimulate endothelial to osteoblast conversion leading to osteoblastic bone formation in prostate cancer patients [51
]. In the current study, we identified the tumor cells as the predominant source of BMP4 in metastasis with ongoing bone formation, while the cell origin of BMP2 was not examined. Another possible tumor-secreted factor found at high expression levels in metastases with ongoing bone formation was SPP1
. This gene encodes osteopontin, which is a major non-collagenous protein in bone matrix [52
]. Osteopontin facilitates the binding of osteoclasts to bone, and subsequently promotes bone resorption [53
]. Over-expression of osteopontin in the LNCaP prostate cancer cell line was shown to increase proliferation and invasiveness [54
] and, in a breast cancer model, osteopontin instigated the growth of otherwise quiescent metastases [55
]. BMP2 signaling is known to activate SPP1
transcription via the RUNX2-SMAD complex [40
] and, accordingly, we found SPP1
expression to be positively correlated to RUNX2 immunoreactivity in examined bone metastases. Besides BMP, the WNT signaling system has proven important for bone formation [56
]. However, in the metastatic tissues examined here, we found unexpectedly low expression levels of factors involved in WNT signaling. Notably, PTHRP
expression levels were also low in all cases examined, indicating no obvious role for this well-known osteolytic factor [57
] in stimulating bone cell activity in prostate cancer bone metastasis.
We have previously identified two subgroups among CRPC bone metastases, those that have high AR and metabolic activity, but show low cellular immune responses, and those with low AR and metabolic activity, but more prominent immune responses [7
]. The current study adds information to this classification of prostate cancer bone metastasis by finding high bone cell activity primarily in patients with low AR activity and ongoing inflammation (Table 2
). Based on this, we hypothesize that patients with non-AR-driven bone metastases who respond poorly to standard AR inhibiting treatment may instead be specifically suited for bone- and/or immune-targeting therapies [7
], while AR-driven metastases in addition to AR-targeting might benefit also from metabolic targeting [58
], including targeting the cholesterol pathway [60
]. Metastases with high bone cell activity may be clinically identified by relatively low serum PSA and raised levels of bone remodeling markers (see above). For a more thorough discussion on how the molecular findings of this paper could be translated into clinical practice, please see the conclusions below that further discusses limitations of the current study, but also possibilities for the future.
The use of bisphosphonates, which function to inhibit osteoclasts and osteolysis, has been proven beneficial for patients with prostate cancer bone metastases by reducing the incidence of skeletal-related events [23
]. This justifies the concept of targeting bone resorption in prostate cancer patients with bone metastases. Another inhibitor of bone resorption used in the clinic is the RANKL inhibiting antibody denosumab. RANKL is mainly expressed by bone stromal cells, including osteoblasts and osteocytes, and as described above, functions as a major mediator of osteoclastogenesis and bone resorption. In the current study, expression of TNFSF11
and its neutralizing decoy receptor TNFRSF11B
(OPG) was undetectable in most samples. Immunohistochemical staining of RANKL and OPG might be able to determine if the ligand is present in the metastasis samples and if expression differs in relation to osteoclast activity. Another attractive therapy for prostate cancer patients with high bone remodeling may be radium-223, a short range α particle-emitting agent that acts as a calcium mimetic, meaning that it accumulates in areas of high bone turnover where it is incorporated into bone. Given the important functions of BMPs in both bone remodeling and cancer, they could also be considered attractive therapeutic targets. The natural BMP inhibitor noggin has been found to inhibit both BMP2 and BMP4, and to inhibit the expansion of PC-3 cells in vivo [62
]. Dorsomorphin, a small molecule inhibitor of type I BMP receptors, has been found to inhibit ovarian cancer cell growth in vivo [63
]. Notably, the present study found indications that patients undergoing chemotherapy might develop primarily osteolytic metastases, since their TRAP levels were unchanged, while their RUNX2 levels were markedly decreased. This very interesting observation deserves to be further explored in a larger cohort of patients.
4. Materials and Methods
4.1. Patient Samples
Bone metastasis samples were obtained from a series of fresh-frozen and formalin fixed paraffin embedded biopsies collected from patients with prostate cancer operated for metastatic spinal cord compression at Umea University Hospital (2003–2013). The patient series and the tissue handling have been previously described [5
]. Clinical and pathological characteristics for patients included in the current study are summarized in Table 1
. All patients gave their informed consent for inclusion before they participated in the study. The study was conducted in accordance with the Declaration of Helsinki, and the protocol was approved by the local ethic review board of Umea University (Dnr 03-158, 13-05-2003; Dnr 04-26M, 11-05-2007).
4.2. RNA Extraction and Gene Expression Analysis
Total RNA isolation and gene expression analysis of metastasis samples were previously performed, according to details given in [5
]. Briefly, RNA was extracted from representative areas of fresh frozen bone metastases sections using the Trizol (Invitrogen, Carlsbad, CA, USA) or the AllPrep DNA/RNA/Protein Mini Kit (QIAGEN, Hilden, Germany) protocols. Nucleic acids were quantified by absorbance measurements using a spectrophotometer (ND-1000 spectrophotometer; NanoDrop Technologies, Inc., Wilmington, DE, USA). The RNA quality was analyzed with the 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA) and verified to have an RNA integrity number ≥6. Gene expression array analysis was performed using the human HT12 Illumina Beadchip technique (Illumina, San Diego, CA, USA) with version 3 in [5
] and version 4 in [7
]. In the current study, data from the two gene expression studies were combined and samples were included if they contained a minimum of 10% bone tissue, as determined from the hematoxylin-eosin stained sections taken before RNA isolation and originally controlled to have a tumor cell content of at least 30% [5
] or 50% [7
]. Four patients were excluded due to previous bisphosphonate treatment. With these criteria, bone metastasis samples from 7 treatment-naïve and 21 castration-resistant patients (from originally 13 and 56 cases, respectively) were left for analysis of transcriptomic data. The bone tissue content of those metastases was further defined as 10–25% (n
= 17) or 25–50% (n
= 11) by visual assessment of histological sections. Beadchip data was included for all probes with average signals above two times the mean background level in at least one sample per study array. Arrays were individually normalized using the quantile method (GenomeStudio V2011.1, Illumina) and each probe was centered by the median of the samples in the corresponding dataset. Normalized datasets were merged by mapping Illumina ID and Hugo gene symbol, leaving 13,846 probes for further analysis. Redundant replication for transcripts were kept in the multivariate analysis, but removed in ontology analysis.
4.3. Univariate Statistics
Correlations between variables were investigated using the Spearman rank test. Groups were compared using the Mann–Whitney U test for continuous variables and the Chi square test for categorical variables. Statistical analyses were performed using the Statistical Package for the Social Sciences, SPSS 24 software (SPSS, Inc., Chicago, IL, USA).
4.4. Multivariate Data Analysis
Multivariate principal component analysis (PCA) was applied to create an overview of the variation in bone remodeling among metastasis samples, here based on gene expression levels of the predefined osteoblast (ALPL
), osteoclast (ACP5
) and osteocyte (SOST
) marker genes (reviewed in [33
]). The bone cell activity was represented by the first score vector (t
), capturing the largest variation in the data as a linear combination of transcript levels in the selected genes. Orthogonal projections to latent structures (OPLS) were utilized to find genes co-varying with the bone cell activity by relating the entire expression data (X) to the activity score, represented by the above t
 (Y), in a linear multivariate model. In the case where OPLS was used to capture class separation, OPLS discriminant analysis (OPLS-DA) was performed. For multivariate analysis, data were centered and scaled to unit variance (UV). Sevenfold cross-validation and cross-validated analysis of variance (CV-ANOVA) testing were used to measure the statistical significance of models. Multivariate statistical analyses were performed with SIMCA software version 14.0 (Umetrics AB, Umea, Sweden).
4.5. Functional Enrichment Analysis
Analysis was performed using the MetaCore software (Clarivate analytics, Philadelphia, PA, USA) in order to identify enriched process networks in the data. The significance of the association between the list of molecules in the data (here gene products showing positive, p
(corr) ≥ 0.5, or negative, p
(corr) ≤ −0.5, correlation loadings to the score vector (t
) representative for bone cell activity in Figure 1
) and process networks defined from the literature were assessed by: (i) the ratio of molecules in the data that mapped to a specific pathway in relation to the total number of molecules included in the network and (ii) the false discovery rate when applying the Fisher’s Exact test to determine the probability that the relationship between the molecules in the data set and the networks is explained by chance.
Upstream analysis was used to identify regulators with a probability to be responsible for the observed enriched process networks, based on the p-value for a calculated connectivity ratio between actual and expected interactions with objects in the data.
4.6. Immunohistochemistry and Bone Histomorphometry
Briefly, tissue sections were deparaffinized in xylene and rehydrated through graded ethanol. For histological examinations, sections were stained with hematoxylin-eosin and/or van Gieson solution. Samples with adequate tumor and bone tissue content were further immunohistochemically stained as described below.
Immunostaining for TRAP and RUNX2 was performed using the Benchmark Ultra system (Ventana, Oro Valley, AZ, USA) and the ultraView Universal DAB Detection Kit (760-500, Ventana), with antigen retrieval in the CC2 buffer and the following primary antibodies: MABF96, diluted 1:400 (Millipore, Burlington, MA, USA) and ab81357, diluted 1:100 (Abcam, Cambridge, MA, USA). BMP4 immunostaining was performed using the ab39973 antibody (Abcam, diluted 1:100) and the Envision HRP Rabbit detection system. Negative control sections were prepared by performing immunostaining procedures without adding primary antibodies and for BMP4 also by pre-incubating the primary antibody with excess (1:10 w/w) of blocking peptide (ab40140, Abcam). Immunostained sections were scanned using the Pannoramic 250 FLASH scanner and evaluated using the Pannoramic viewer 1.15.2 software (3D HISTECH, Budapest, Hungary).
TRAP and RUNX2 immunoreactivity of cells lining bone surfaces was examined in ten randomly selected fields, with bone in close proximity to tumor cells, at 40x magnification. TRAP- and RUNX2-positive bone surface and total bone surface was measured using the ImageJ 1.50i software (NIH, Bethesda, MD, USA). A total of 35 metastases were analyzed (8 treatment-naïve and 27 castration-resistant cases), of which 24 were included in the gene expression analysis.
In corresponding van Gieson-stained sections, bone morphology was assessed by evaluating the amount of non-lamellar, newly formed, osteocyte-rich bone in close vicinity to tumor cells. The amount of such bone tissue in the biopsies was graded on a scale from 0 to 5, which reflects gradually increasing areas in relation to the total area of the biopsies. Active bone formation was evaluated as the presence or absence of cuboid osteoblasts on the surface of the newly formed bone and registered as ongoing or not ongoing.
BMP4 immunostaining was performed in 21 available samples and evaluated by scoring intensity (0 = negative; 1 = weak; 2 = moderate; 3 = intense staining) and fraction of stained cells (1 = 1–25%; 2 = 26–50%; 3 = 51–75%; 4 = 76–100%). A total IR score ranging from 0 to 12, was obtained by multiplying the staining intensity and fraction scores.