All experiments were performed according to the German Guidelines of Animal Research on the Protection of Animals as well as the ARRIVE guidelines and were approved by the local ethical committee (No. 1455, Regierungspräsidium Tübingen, Germany). Osteoblast lineage-specific ERα-KO mice (Tg (Runx2-cre)1Jtuc x Esr1^tm1.2Mma) were generated by crossing Runx2-Cre with ERα^fl/fl mice on a C57BL/6 background, both provided by Prof. J. Tuckermann (Ulm University). Cre⁻ littermates (ERα^fl/fl) were used as control. Cre⁺ mice (ERα^{fl/fl; Runx2Cre}) were shown to lack the ERα in cells of the osteogenic lineage (Supplementary Figure S1) including osteoblasts, osteocytes and hypertrophic chondrocytes. All animals were housed in groups of up to five mice per cage with a 12 h light, 12 h dark rhythm, and received water ad libitum as well as a standard mouse feed (Ssniff R/M-H, V1535-300; Ssniff, Soest, Germany) until the day of ovariectomy/sham-ovariectomy. Subsequently, the food was switched to a phytoestrogen-free diet (Ssniff). Mouse genotyping was conducted by lysed ear punch PCR using the primers: 5′-CCA​GGA​AGA​CTG​CCA​GAA​GG-3′, 5′-TGGCTTGCAGGT ACAGGAG-3′ and 5′-GGA​GCT​GCC​GAG​TCA​ATA​AC-3′ to detect the Cre transgene, whereas the ERα loxP sites were detected using the primers: 5′- TAGGCTTTGTCTCGCTTT CC-3′, 5′- CCCTGG CAAGATAAGACAGC-3′ and 5′-AGG​AGA​ATG​AGG​TGG​CAC​AG-3′.

When aged 12 weeks, female mice (n = 24 per genotype) were randomly assigned (Table 1) to bilateral ovariectomy (n = 12 per genotype) or sham-operated (non-OVX, n = 12 per genotype) as described previously (Haffner-Luntzer et al., 2017). 4 weeks after OVX or non-OVX surgery, all mice underwent standardized unilateral femur osteotomy as described previously (Röntgen et al., 2010; Wehrle et al., 2015; Haffner-Luntzer et al., 2017). Briefly, the osteotomy was created at the right femur diaphysis using a 0.4 mm Gigli wire saw (RISystem, Davos, Switzerland) and stabilized by a semi-rigid external fixator (RISystem). Half of the OVX (n = 6 per genotype) and non-OVX mice received LMHFV. Three weeks after osteotomy surgery (day 21), all mice were sacrificed using an isoflurane overdose and cardiac blood withdrawal.

The LMHFV regimen was chosen due to our previous studies, showing that 45 Hz significantly improved fracture healing in OVX mice (Wehrle et al., 2015; Haffner-Luntzer et al., 2018a). Starting on the third day after osteotomy surgery, mice were placed on custom-made vibration platforms for 20 min per day for 5 days per week and received vertical whole-body vibration with 0.3 g sinusoidal peak-to-peak acceleration and 45 Hz frequency, as described previously (Wehrle et al., 2015) (Figure 1A). The amplitude and frequency were continuously recorded using integrated accelerometers at the platform (Sensor KS95B.100, measurement amplifier Innobeamer L2, Software Vibromatrix; IDS Innomic GmbH, Salzwedel, Germany). The control mice were sham-vibrated on the same platforms without activation of the vibration generator. All mice were allowed to move freely on the platforms during the vibration or sham-vibration treatment and afterwards returned to their home cages.

To evaluate the mechanical properties of intact and osteotomized femurs explanted on day 21, a non-destructive three-point bending test was performed as described previously (Röntgen et al., 2010; Wehrle et al., 2015). Briefly, following removal of the fixator, an axial load with a maximum of 2 N was applied on top of the callus side in a cranio-lateral position using a material-testing machine (1454, Zwick GmbH & Co KG, Ulm, Germany). The bending stiffness was calculated from the slope of the load-deflection curve as described previously (Röntgen et al., 2010). If the bones were too fragile for reaching a preload of 0.05 N, the three-point bending test was not performed and a flexural rigidity value of 0 was given. Following biomechanical testing, femora were fixed in 4% phosphate-buffered formaldehyde solution and scanned using a μCT scanning device (Skyscan 1172 version 1.5; Skyscan, Kontich, Belgium) at a resolution of 8 μm using a peak voltage of 50 kV and 200 μA. Analyses and calibration steps were performed according to the guidelines of the American Society for Bone and Mineral Research (Bouxsein et al., 2010). The volume of interest was defined as the entire callus between the fractured cortices. Within each scan, two phantoms with a defined density of hydroxyapatite (250 and 750 mg hydroxyapatite/cm³) were included to determine the bone mineral density. To distinguish between mineralized and nonmineralized tissue, a threshold for cortical bone set at 641.9 mg hydroxyapatite/cm³ was used (Morgan et al., 2009). The trabecular bone of the intact left femora was assessed 200 μm proximal of the metaphyseal growth plate over a length of 280 μm in the distal femur, excluding the cortex. For assessing the trabecular bone parameters, a threshold set at 394.8 mg hydroxyapatite/cm³ was used (Morgan et al., 2009). Analyses were performed by means of Skyscan software (NRecon version 1.7.1.0, DataViewer version 1.5.1.2, and CTAn version 1.17.2.2). Bony bridging score was determined in two perpendicular planes with one bridged cortex counting for one scoring point. A bony bridging score of 3 or 4 represented successful healing.

Following μCT analysis, bone specimens were subjected to decalcified histology as described previously (Haffner-Luntzer et al., 2014). Sections of 7 μm were stained with Safranin O for histomorphometric tissue quantification. The amounts of bone, cartilage and fibrous tissue in the fracture gap were determined using an image analysis software (Leica DMI6000 B; Software MMAF Version 1.4.0 MetaMorph^®; Leica, Heerbrugg, Switzerland). The number and surface of osteoclasts (NOc/BPm, OcS/BS) were quantitatively assessed using tartrate-resistant alkaline phosphatase (TRAP) staining. Osteoclasts were defined as TRAP-positive cells with two or more nuclei, directly located on the bone surface with a visible resorption lacuna between the bone matrix and the cell. The number and surface of osteoblasts were determined using Safranin-O staining. Osteoblasts were defined as cubic-shaped cells with visible cytoplasm, located directly on the bone surface. Bone cells and surface were evaluated in a rectangular area (650 × 450 μm) using the Osteomeasure system (Osteometrics, Decatur, United States). Images were obtained using a Leica microscope (DMI 6000B). All analyses were performed according to the American Society for Bone and Mineral Research guidelines for bone histomorphometry (Dempster et al., 2013).

For immunohistochemical staining, paraffin-embedded sections of 4 µm were deparaffinized, enzymatically demasked by trypsin, blocked with 3% peroxidase followed by a serum block with 5% goat serum in TBS-T for 2 h. Staining for ERα was performed at 4°C overnight using the following primary antibody: rabbit anti-mouse ERα (1:75, # PA5-16440, Invitrogen). Rabbit IgG was included as isotype control to confirm specific staining. Next, sections were washed with TBS and incubated with a goat anti-rabbit secondary antibody (1:100, #B2770, Life Technologies) for an hour at RT. After another washing step, horseradish peroxidase (HRP)-conjugated streptavidin (#PK-6100, VECTASTAIN^® Elite ABC-HRP Kit, Peroxidase, Vector Laboratories) was applied according to the manufacturer`s guidelines. NovaRED (#SK-4800, Vector^® NovaRED^® Substrate Kit, Peroxidase (HRP), Vector laboratories) was used as chromogen and the sections were counterstained with hematoxylin (1:5; #2C-306, Waldeck, Münster, Germany) and rinsed. This staining protocol was previously established by using biological negative controls (bone sections from ERα-general KO mice). In ERα-stained sections, osteoblasts were identified by their cubic shape and direct contact to the bone trabeculae, whereas osteocytes were defined as ellipsoidal shaped cells embedded in the mineralized bone matrix. Hypertrophic chondrocytes were identified as spherical or polygonal shaped cells within a population of other chondrocytes.

Murine MC3T3-E1 cells were purchased from the American Type Culture Collection (ATCC) and seeded at a density of 4,000 cells per cm². On day 3 after seeding, cells were differentiated by adding 50 mg/ml ascorbic acid and 10 mM β-glycerophosphate to the phenol-red free culture medium containing 10% charcoal-stripped (estrogen-free) fetal calf serum (FCS), 1% L-glutamine and 1% penicillin/streptomycin (all ThermoFisher Scientific). Half of the cultures were subjected to LMHFV on a custom-made vibration platform at 0.3 g peak-to-peak acceleration/45 Hz for 20 min/day for 5 days. The other half of the cultures (sham-vibrated) were placed on the same platform without turning on the vibration device. Total RNA was isolated as described previously (Haffner-Luntzer et al., 2018b) and microarray-based gene expression analysis with mouse Gene 1.0 ST GeneChip^® arrays was performed as published previously (Mödinger et al., 2018). Differentially expressed probesets were determined by t test and considered statistically significant when p < 0.05 and fold change ≤1.5, as described previously (Zoller et al., 2017). To identify the involved molecular pathways, the GoMINER tool (Zeeberg et al., 2003) was used, whereas functional protein interaction networks were identified using the STRING 10 program (http://string-db.org/). Complete microarray data are available in the Supplementary Material. For the Cox2 antagonist experiment, celecoxib was obtained from Sigma (SML3031) and prepared using dimethyl sulfoxide (stock: 2 mg/ml). AH23848 (EP4 antagonist) was purchased from Cayman Chemical (19023) and dissolved in dimethyl sulfoxide (stock: 5 mg/ml). Both antagonists were used at a concentration of 1 × 10⁻⁶ M.

Human bone marrow-derived mesenchymal stem cells (hMSCs) were obtained by Lonza (PT-2501) and seeded in 24-well plates (10,000 cells/well) 2 days prior to the differentiation experiments. The medium was changed to differentiation medium (phenol-red free α-MEM, 10% FCS + 1% penicillin/streptomycin + 1% L-glutamine, 50 mg/ml ascorbic acid, 10 mM β-glycerophosphate) at day 0 and the cells were allowed to differentiate for 10 days. At day 10, the medium was switched to estrogen-free medium and vibration was performed 4 h after the medium-switch for 20 min at 45 Hz and 0.3 g. Thirty minutes after the end of the vibration treatment, the experiment was stopped and RNA was isolated (RNeasy Mini kit, Qiagen) for analyzing gene expression by qPCR and human Gene 1.0 ST GeneChip^® arrays.

The SensiFAST SYBR Hi-ROX One-Step Kit (Bioline, Mempis, United States) was used according to the manufacturer’s guidelines to perform quantitative PCR. B2M (murine F: 5′-ATA​CGC​CTG​CAG​AGT​TAA​GCA-3′, murine R: 5′-TCA​CAT​GTC​TCG​ATC​CCA​GT-3′, human F: 5′-CTC​ACG​TCA​TCC​AGC​AGA​GA-3′, human R: 5′-GGA​TGG​ATG​AAA​CCC​AGA​CA-3′) was used as the housekeeping gene. Relative gene expression of Lef1 (murine F: 5′-TCA CCT ACA GCG ACG AG-3′, murine R: 5′-TGA​CAT​CTG​ACG​GGA​TGT​GT-3′, human F: 5′ GAG​ATT​TCT​CTG​TAT​GGC​ACC-3′, human R: 5′-CTG​CAA​TGA​GAC​ACT​TTC​TC-3′), Ptgs2 (murine F: 5′-AGG​GGT​GTC​CCT​TCA​CTT​CT-3′, murine R: 5′-CAT​TGA​TGG​TGG​CTG​TTT​TG-3′, human F: 5′-TAA​GGG​GAG​AGG​AGG​GAA​AA-3′, human R: 5′-CTG​CTG​AGG​AGT​TCC​TGG​AC-3′) and Mdk (human F: 5′-TGC​CCT​GCA​ACT​GGA​AGA​A-3′, human R: 5′-GCCTGTGCCCCC ATCAC-3′) was calculated using the delta-delta CT method.

Group size was n = 6 per group and per genotype (ERα^fl/fl, ERα^{fl/fl; Runx2Cre}). In one animal, the histological sections were not evaluable, therefore analyzing this group with n = 5 (Figures 3E–H, second boxplot). Data were tested for normal distribution using the Shapiro-Wilk test, and data sets were normally distributed. Statistical testing between two groups was done using two-tailed Student’s t-test, whereas comparisons between more than two groups were performed by one-way ANOVA with post hoc Tukey’s test for adjusting the p-value for multiple comparisons using GraphPad Prism 8.4.3 (GraphPad Software, La Jolla, CA). The level of significance was set at p ≤ 0.05. Results are presented as box-and-whisker plots (with median and interquartile range) from max. to min., showing all data points.

All mice received the selected treatments including ovariectomy/sham-operation, fracture and LMHFV/sham-vibration as illustrated in Figure 1A. The body weight of the animals was not significantly influenced by the treatments or the genotype during the entire experimental period (Figures 1B,C). To control for the success of OVX in ERα^fl/fl and ERα^{fl/fl; Runx2Cre} mice, uterus weight at day 21 after osteotomy was assessed. In both mouse strains, OVX resulted in a significantly lower uterus weight (Figures 1D,E). Additionally, we performed µCT analysis of the intact left femora to determine if our mice developed an osteoporotic bone phenotype after OVX treatment. We found that non-vibrated and vibrated OVX groups from both genotypes displayed lower bone volume fraction and trabecular number, either significantly or by strong trend (Supplementary Figure S2).

In non-OVX mice that received LMHFV treatment, a significant reduction of flexural rigidity (p < 0.0001) and bony bridging (p = 0.0047) of the fracture gap was displayed, whereas the bone mineral density and the relative bone volume ratio did not differ significantly (Figures 2A–D). Furthermore, histomorphometrical analysis revealed a by trend reduced relative bone area (p = 0.0923), an increased cartilage area (p = 0.0564) as well as a significantly reduced osteoblast count (p = 0.0539) and a significantly lower osteoblast activity (p = 0.0159) in vibrated non-OVX ERα^fl/fl mice (Figures 3A–F). All these data indicated impaired fracture healing at day 21 in non-OVX mice subjected to vibration treatment.

Bone regeneration was also compromised in OVX mice compared to non-OVX mice, as indicated by an inferior flexural rigidity (p < 0.0001), significantly reduced bony bridging (p = 0.0041) and by trend increased amount of cartilage (p = 0.1469) in the fracture callus (Figures 2, 3). Similarly, osteoblast number (p = 0.0576) and surface (p = 0.0296) were also diminished in OVX mice compared to non-OVX mice (Figures 3E,F).

Bone healing was improved by LMHFV in OVX mice in comparison to OVX animals as indicated by significantly enhanced flexural rigidity (p = 0.0374), bone mineral density (p = 0.0108) and bony bridging (p = 0.0155) of the fracture gap and a by trend reduced cartilage area (p = 0.2248) (Figures 2, 3). Furthermore, LMHFV resulted in a significantly increased number (p = 0.0031) and activity of osteoblasts (p = 0.0012) in OVX ERα^fl/fl mice compared to OVX animals without vibration (Figures 3E,F). The callus size as well as the number and surface of osteoclasts did not differ significantly between all the ERα^fl/fl groups (Figures 3D,G,H). Overall, these data demonstrated that OVX and LMHFV both disturb bone regeneration in control ERα^fl/fl mice, whereas in combination both treatments resulted in an improved fracture healing.

Although mice with a specific deletion of the ERα in the osteoblast lineage (ERα^{fl/fl; Runx2Cre}) were previously shown to have a preexisting bone phenotype with decreased trabecular bone mass in the spine and tibiae (Seitz et al., 2012), we found that this does not influence fracture healing (p = 0.439) when directly comparing non-OVX non-vibrated animals of both genotypes (Supplementary Figure S3).

Furthermore, LMHFV did not exert negative effects on fracture healing in non-OVX ERα^{fl/fl; Runx2Cre} mice, because there were no significant differences in flexural rigidity, bone content or bridging of the fracture gap in comparison to ERα^{fl/fl; Runx2Cre} mice that did not receive LMHFV (Figure 4). In addition, histomorphometrical parameters were unaltered by LMHFV treatment in non-OVX mice except the osteoblast number (p = 0.0005) and surface (p = 0.0004) that were significantly increased in vibrated non-OVX mice lacking the ERα in osteoblasts (Figure 5).

ERα^{fl/fl; Runx2Cre} OVX mice without vibration were protected from impaired fracture healing due to OVX, because a similar flexural rigidity, bony bridging and bone formation in the fracture callus was observed compared to non-OVX ERα^{fl/fl; Runx2Cre} mice (Figure 4). Furthermore, histomorphometrical analysis did not reveal any significant differences in callus composition or cellular parameters compared to non-OVX ERα^{fl/fl; Runx2Cre} mice (Figure 5).

Interestingly, reversed effects of vibration on bone formation during fracture healing in OVX ERα^{fl/fl; Runx2Cre} mice were observed, which resulted in a considerably reduced flexural rigidity (p = 0.001), by trend diminished bone formation (p = 0.1975) and significantly reduced bony bridging (p = 0.0032) compared to OVX ERα^{fl/fl; Runx2Cre} mice (Figures 4A,C,D) in contrast to the improved fracture healing seen in vibrated OVX control ERα^fl/fl mice. Histomorphometrical analysis further revealed a significantly increased cartilage area (p = 0.0114) in the fracture callus of OVX ERα^{fl/fl; Runx2Cre} animals subjected to vibration treatment (Figure 5B). In addition, the osteoblast surface (p = 0.0039) and number (p = 0.0188) were significantly decreased in vibrated OVX ERα^{fl/fl; Runx2Cre} mice (Figures 5E,F), whereas osteoclast activity was significantly reduced (p = 0.0315) (Figure 5H).

We performed whole genome microarray-based gene expression analysis of vibrated and sham-vibrated samples and found that 304 genes were differentially regulated upon LMHFV treatment (Figure 6). Among the 184 upregulated genes, the mechanosensitive enzyme Cox2, encoded by the gene Ptgs2, appeared in our array (Table 2), as well as bone remodeling and extracellular matrix-associated genes (Table 3). Downregulated genes were found to be involved in Wnt/β-catenin signaling (Table 4), with most of the genes encoding for proteins that act as Wnt inhibitors (Axin2, Sostdc1, Dkk2) except the gene encoding for the Wnt receptor Fzd9. STRING analysis was performed to investigate the interactions of the respective pathways (Figures 6A,D). As expected, the Wnt pathway genes were clustering as well as the genes for extracellular matrix and Cox2/PGE₂ signaling (Figure 6D). Confirming the microarray findings, gene expression of Cox2 was validated by qPCR (Figure 6C). Consequently, we were interested whether the Cox2/PGE₂ pathway might interact with canonical Wnt signaling. To investigate whether the signaling via the Cox2/EP4 axis might affect downstream signaling of the Wnt pathway, we performed further in vitro experiments with MC3T3-E1 cells pretreated with either the EP4 inhibitor AH23848 or the Cox2 inhibitor celecoxib. We demonstrated that the prostaglandin pathway via EP4 positively modulates the Wnt target gene Lef1 in response to LMHFV, because the EP4 and Cox2 inhibitors abolished the LMHFV-induced Lef1 upregulation (Figures 6E–G). Because we aimed to study the effects of LMHFV also in human cells, hMSCs were subjected to vibration and gene expression was assessed. We demonstrated that Cox2 in hMSCs was also upregulated by vibration (Figure 6H). Furthermore, the Wnt inhibitor Mdk was downregulated upon LMHFV, whereas the Wnt transcription factor Lef1 was significantly upregulated (Figures 6I,J). In addition, microarray-based gene analysis of vibrated vs. non-vibrated hMSCs revealed that a further 56 genes were differentially regulated upon vibration. After excluding non-coding transcripts, 26 probesets remained for analysis. Among them, there were five micro RNAs (miRNAs) that are known to be involved in regulating bone metabolism, whereof three were upregulated and two downregulated (Table 5).