Load‐induced osteogenic differentiation of mesenchymal stromal cells is caused by mechano‐regulated autocrine signaling

Mechanical boundary conditions critically influence the bone healing process. In this context, previous in vitro studies have demonstrated that cyclic mechanical compression alters migration and triggers osteogenesis of mesenchymal stromal cells (MSC), both processes being relevant to healing. However, it remains unclear whether this mechanosensitivity is a direct consequence of cyclic compression, an indirect effect of altered supply or a specific modulation of autocrine bone morphogenetic protein (BMP) signaling. Here, we investigate the influence of cyclic mechanical compression (ε = 5% and 10%, f = 1 Hz) on human bone marrow MSC (hBMSC) migration and osteogenic differentiation in a 3D biomaterial scaffold, an in vitro system mimicking the mechanical environment of the early bone healing phase. The open‐porous architecture of the scaffold ensured sufficient supply even without cyclic compression, minimizing load‐associated supply alterations. Furthermore, a large culture medium volume in relation to the cell number diminished autocrine signaling. Migration of hBMSCs was significantly downregulated under cyclic compression. Surprisingly, a decrease in migration was not associated with increased osteogenic differentiation of hBMSCs, as the expression of RUNX2 and osteocalcin decreased. In contrast, BMP2 expression was significantly upregulated. Enabling autocrine stimulation by increasing the cell‐to‐medium ratio in the bioreactor finally resulted in a significant upregulation of RUNX2 in response to cyclic compression, which could be reversed by rhNoggin treatment. The results indicate that osteogenesis is promoted by cyclic compression when cells condition their environment with BMP. Our findings highlight the importance of mutual interactions between mechanical forces and BMP signaling in controlling osteogenic differentiation.


| INTRODUCTION
Bone fracture healing is a complex multistage process that depends on coordinated biochemical and biomechanical signals. The relevance of mechanical boundary conditions for fast and successful bone regeneration has been documented earlier (Betts & Müller, 2014;Duda et al., 2001). Due to the gained knowledge, fracture fixation systems nowadays aim to account for proper mechanobiological requirements (Märdian, Schaser, Duda, & Heyland, 2015). Interfragmentary movements at the fracture site are differentiated in axial compression of fracture fragments or relative movements resulting in shear. Whereas shear stress is widely considered detrimental, it is commonly accepted that a distinct amount of compression promotes bone regeneration and too little or too large axial movements delay healing (Claes & Heigele, 1999;Epari, Taylor, Heller, & Duda, 2006;Mehta, Duda, Perka, & Strube, 2011). Specifically, the early phase of the healing cascade is considered to be highly mechanosensitive (Klein et al., 2003). In this regard, we and others suggested that the initial mechanical condition influence the entire subsequent inflammatory, angiogenic, and endochondral healing cascades (Goodship, Cunningham, & Kenwright, 1998;Klein et al., 2003;Schell et al., 2005). To further employ the power of mechanobiology in critical healing scenarios, an improved understanding of how mechanical stimuli influence cellular processes like migration, differentiation, and growth factor signaling is needed.
Following this motivation, various test systems for the in vitro application of mechanical stimulation have been developed, including force application in cell culture well plates (Michalopoulos et al., 2012; Shin Kang et al., 2013;Sittichokechaiwut, Edwards, Scutt, & Reilly, 2010) or bioreactors (Jagodzinski et al., 2008;Kopf, Petersen, Duda, & Knaus, 2012;Ode et al., 2011;Petersen, Joly, Bergmann, Korus, & Duda, 2012). Diverse effects of mechanical stimulation on cell migration were reported. Although osteoblasts on cyclically strained 2D membranes showed an enhanced migration capacity (Bhatt et al., 2007), migration of MSCs embedded in a hydrogel was reduced in response to cyclic compression (Ode et al., 2011). Progenitor cells migrate from intact bone regions with low strain into the fracture side with high strain (Gerstenfeld, Cullinane, Barnes, Graves, & Einhorn, 2003;Klosterhoff et al., 2017) where their differentiation is controlled by the mechanobiological environment. Thus, it is of relevance to investigate potential alterations in progenitor cell migration and differentiation in response to different strain regimes.
Multiple studies examined the influence of loading on stem cell differentiation, particularly osteogenic commitment (reviewed in Delaine-Smith & Reilly, 2011). Motivated by a tissue engineering approach, the majority of these studies used osteoinductive medium supplements or bone-derived scaffolds masking effects of loading on cell fate decision.
However, studies that worked without additional osteogenic triggers reported on an induction of osteogenic differentiation by cyclic bulk material deformation ( f = 1 Hz), reflected by an increased expression of Runt-related transcription factor 2 (RUNX2; Michalopoulos et al., 2012), Alkaline phosphatase (ALP), and osteopontin (Shin Kang et al., 2013) in human MSCs. Interestingly, mechanical loading was previously shown to induce the expression of bone morphogenetic protein type 2 (BMP2) in MSCs (Wang et al., 2010). The growth factor is a potent inducer of bone formation and indispensable for fracture healing (Tsuji et al., 2006). The instance that it induces the expression of RUNX2 (Lee et al., 2000)  With the present study, we aimed at investigating the direct influence of cyclic mechanical loading on the migration and osteogenic differentiation of primary hBMSCs. Based on the predominant opinion in the literature available for MSCs, we hypothesized that cyclic mechanical loading can directly induce osteogenic differentiation and consequently influence cell migration. However, the fact that key osteogenic genes and especially RUNX2 were downregulated in response to cyclic compression, but BMP2, on the other hand, was clearly upregulated, advised us to consider the indirect load-induced autocrine BMP signaling as a mediator of osteogenic gene expression under cyclic compression.

| Collagen scaffolds used as cell carriers
Macroporous scaffolds from porcine collagen were utilized as a 3D cell carrier material in this study (in cooperation with Matricel GmbH, Herzogenrath, Germany). Two scaffold prototypes with collagen contents of 1.1 and 1.5 wt-% were produced via directional freezing and freeze drying leading to elastic moduli of 3.4(±0.9) kPa (scaffold A) and 12.3(±1.8) kPa (scaffold B), respectively. The scaffolds were characterized by highly aligned channel-like pores with vertical pore orientation, providing nutrient and oxygen supply and at the same time offering a guiding structure for cell migration. As shown before, the differences in collagen content did not affect the pore size (1.1 wt-% = 76 μm ± 12 μm, 1.5 wt-% = 79 μm ±14 μm) and architecture as analyzed using electron microscopy (Herrera et al., 2019;Petersen et al., 2012). In brief, three rectangular scaffolds have been cut from different regions of the bulk material with the collagen content of 1.1 or 1.5 wt-%. Electron microscopy images at three different positions have been taken from the scaffolds in top view. Scaffold pore diameter was measured using a custom-made ImageJ plugin calculating the average distance between two scaffold pores. Therefore, two scaffold walls confining a scaffold pore were contoured manually in one image, and the mean pore diameter was calculated for at least three pores per image. For cell culture experiments, cylindrical shaped scaffolds with a diameter of 5 mm and a height of 3 mm were prepared.

| Bioreactor and mechanical loading parameters to mimic the early bone healing phase
A custom-made mechanobioreactor system, previously described by Petersen et al. (2012) was used to apply cyclic monoaxial compression to the cell-seeded scaffolds with pores oriented in the axial direction (Figures 1b and 1c). The system was designed to mimic the mechanical environment in the hematoma during the early phase of fracture healing. In brief, the bioreactor can be separated in two compartments, the cell culture unit and the mechanical unit. The cell culture unit can be assembled under sterile conditions and consists of a reactor chamber, a medium reservoir allowing gas exchange and a micro pump (pump rate approx. 2.5 ml/min). The micro pump is circulating the medium and is necessary for a sufficient supply of the cells inside the 3D biomaterial. The pump is running in all experimental conditions, also under control conditions (no cyclic compression). In comparison with static culture in well plates, the system provides an enhanced fluid flow in the reactor chamber.
The mechanical unit allows the application of defined loading patterns and the online measurement of the force acting on the sample. The scaffold inside the reactor chamber was subjected to a sine wave of cyclic monoaxial compression with a frequency of 1 Hz and a magnitude of 5 or 10% scaffold height (150 and 300 μm, respectively). Cyclic mechanical compression was applied periodically with 3-hr stimulation and 5-hr break in the direction of the scaffold pores. The loading parameters were selected to represent the in vivo loading situation (see Section 4).

| Homogeneous cell seeding in collagen scaffolds
After expansion, cells were trypsinized and a cell concentration of 2.5 × 10 3 cells/μl was adjusted, which corresponds to a cell density of approximately 1.5 × 10 5 cells/scaffold. Collagen scaffolds were dipped into the cell suspension where the scaffold immediately soaked up suspension until completely filled and were subsequently transferred into a 12-well culture plate. During an incubation of 60 min at 37°C, cells were allowed to adhere to the scaffold walls.
Subsequently, cell-seeded constructs were washed once in cell culture medium to remove unattached cells and placed into a new 12-well plate filled with 1.4 ml cell culture medium. After 24 hr of incubation, the scaffolds were transferred into the bioreactor. As shown previously, uniform distribution of cells within the scaffold was achieved using this seeding method (Brauer et al., 2019).

| Fluorescent staining and imaging
Cells were fixed in 4% paraformaldehyde and quenched for 1 hr in a 50-mM ammonium chloride solution. Thereafter, the scaffolds were infiltrated with 5% gelatin solution at 37°C, which was subsequently solidified at 4°C to stabilize the scaffold structure; gentle cutting along the symmetry axis was enabled. Prior to fluorescence staining, gelatin was washed out by incubation in phosphate-buffered saline (PBS) at 37°C. Right before the staining, three washing steps with TBS-T (40 mM Tris hydrochloride, 10 mM Trizma® base, 150 mM sodium chloride, 0.025% Triton-X 100, pH 8.2) were conducted. The F-actin cytoskeleton was stained with Phalloidin-Alexa 488 (# A12379, Life Technologies, Carlsbad, CA, USA) and cell nuclei were stained with 4′-6-diamidino-2-phenylindiole (DAPI, # D1306, Life Technologies, Carlsbad, CA, USA). Images were taken using a Leica TCS SP5II confocal multiphoton microscope at 25× magnification.

| Oxygen concentration measurement inside cell seeded collagen scaffolds
The measurement was performed on the 3rd and 7th day after cell

| 3D migration assay and analysis of migration distance
The assay was performed as described previously (A. Petersen et al., 2018). In brief, a defined area for cell attachment was created using custom-made silicone rings with an inner diameter of 7 mm and a height of 3 mm, which were placed centrally into a 12-well plate.
Cells were brought into suspension in a concentration of 2 × 10 3 cells/μl, and 104 μl of the suspension was added inside the rings.
The cells were allowed to adhere in a multicell layer for 1 hr at 37°C. Thereafter, the silicone rings were removed, and the cell layer was carefully washed twice with PBS. Collagen scaffolds were dipped into DMEM and placed centrally on the cell multilayer.
Subsequently, culture medium was added and the samples were incubated for 24 hr to enable cell migration into the scaffold. On the next day, scaffolds were transferred into the bioreactor or in another 12-well plate and cultured for additional 3 days in migration medium (DMEM, D5546, Sigma Aldrich), 1 vol-% penicillin/streptomycin, 1 vol-% nonessential amino acids (K0293, Biochrom AG, Berlin, Germany), and 1 vol% Nutridoma (14244100, Roche, Basel, Switzerland). The scaffolds were fixed in 4% paraformaldehyde, incubated in ammonium chloride solution (50 mM) and infiltrated by gelatin solution (5%). After solidification of the gelatin the scaffolds were cut along the symmetry axis. The scaffold halves were transferred into a cryomold, embedded in cryoblock (41-3020-00, Medite GmbH, Burgdorf, Germany), and frozen on a liquid nitrogen-cold aluminum bridge. Cryosections of 20-μm thickness were cut using a cryostat (Leica CM3050S) and immobilized on glass slides. Cell nuclei were stained with DAPI and overview images of the sections were taken (Leica DMBR, Leica).
A region of interest was defined in the center of the scaffold (width = 3,000 px) and the x/y coordinates of each cell nuclei were obtained using ImageJ. The migration distance was calculated as the difference between Y 0 (edge of the scaffold) and the Y position of the nuclei.
Using this assay, first, the migration ability of hBMSCs under the influence of 5% and 10% cyclic mechanical compression and dependent on the substrate stiffness was investigated. Second, the influence of osteogenic differentiation on the migration capacity of hBMSCs was assessed in the cell culture incubator under static conditions. For this, hBMSCs (five donors) were osteogenically pre-differentiated by treatment with osteogenic medium (OM) in cell culture flasks over 7 days. Subsequently, scaffolds were seeded with these pre-differentiated cells for the 3D migration assay. During the migration experiment (3 days), undifferentiated and pre-differentiated hBMSCs were both cultured in migration medium to exclude a direct influence of osteogenic supplements on cell motility.

| RNA isolation, reverse transcription, and quantitative real-time PCR
After 6 days of bioreactor culture with and without cyclic compression, cell-seeded scaffolds were snap-frozen on dry ice. Thereafter, scaffolds were mechanically pulverized in the gas phase of liquid nitrogen utilizing a custom-made piston/cylinder device to enhance the RNA yield. Total RNA was isolated and purified using the PureLink® RNA Mini Kit (12183018A, Life Technologies, Carlsbad, USA), and total DNA digestion was performed using On-column PureLink® DNase (12185-010, Invitrogen, Carlsbad, USA). For each sample, 500 ng mRNA were transcribed into cDNA using iScript™ cDNA Synthesis Kit (#170-8891, BIO-RAD, Hercules, USA). Gene expression was measured by SYBR green-based quantitative real-time PCR (qPCR) and calculated using the efficiency corrected ΔΔC T -method. The mean normalized expression ratios were determined using hypoxanthine -guanine phosphoribosyl transferase 1 (HPRT1) as the reference housekeeping gene. The sequences of all primer are deposited in Table 1. Gene expression of cells treated with 5% or 10% cyclic compression and seeded in 3.4 kPa or 12.3 kPa scaffolds were normalized to the noncompressed control of the respective scaffold type.

| Quantitative measurement of hBMP2 in the culture medium
Bioreactor cell culture media collected from samples stimulated with and without cyclic compression for 6 days were either concentrated 5-fold (R low conditions) utilizing the Amicon Ultra® centrifugal filters (UFC501096, Sigma Aldrich), diluted 3-fold (recombinant human BMP2 [rhBMP2] added conditions), or used directly (R high conditions), depending on the experimental condition. The BMP2 concentrations were determined using a human BMP2 ELISA kit (900-M255, PeproTech, Rocky Hill, USA) detecting human BMP2 as well as rhBMP2 produced in E.coli. The assay does not exhibit cross reactivity to BMP-3, −5, −6, −7, and − 13 and less than 5% cross reactivity to BMP4 and is therefore highly specific to BMP2. Experiments were conducted according to the manufacturer´s instruction.

| Measurement of alkaline phosphatase (ALP) activity
Human BMSCs were cultured in 2D for 7 days in OM or normal expansion medium. Thereafter, medium was aspirated, cells were washed once in PBS (37°C) and subsequently in ALP-buffer (37°C; 100 mM sodium chloride, 100 mM Tris [hydroxymethyl] aminomethane, 1 mM magnesium chloride, pH 9). ALP-substrate (4-Nitrophenylphosphat (pNPP) in 1 M Diethanolamine, N4645, Sigma Aldrich) was added and incubated for 10 min at 37°C. Reaction was stopped with 1 M sodium hydroxide. The absorbance was measured in duplicates at 405 nm using a microplate reader (Tecan). The 4-Nitrophenolate concentration, as a measure for the ALP activity, was calculated using the following equation: c [mol/L] = (E-E 0 )/ε*d with E = extinction, E 0 = extinction blank, ε = 18,450 L/mol*cm (molar extinction coefficient), d = 0.329 cm (plate bottom thickness). Samples were incubated for 4 min in the lysis buffer on ice, vortexed, sonicated for 30 sec, and vortexed again. Subsequently, scaffolds were centrifuged through a tip of a 0.5-to 10-μl pipette, thereby collecting the lysate, while the scaffold remained dry inside the tip.

| Statistical analysis
Statistical significance between groups was analyzed using the

| Cell morphology, proliferation, and oxygen concentration inside scaffolds cultured in the bioreactor
The in vitro setup used in this study was specifically designed to resemble the mechanical environment during the early phase of bone healing as it was observed in vivo. The utilized scaffolds are characterized by low elastic moduli mimicking the soft tissue matrix in the fracture gap. The bioreactor was used to simulate interfragmentary compression that occur as a consequence of weight bearing in the rage of reported data for external fixation in bone healing in sheep (Klein et al., 2003;Petersen et al., 2012;Schell et al., 2008). At first, to verify that neither bioreactor culture nor cyclic compression have major influences on cell morphology and number after 7 days, confocal imaging was performed and scaffolds cultured in the bioreactor (with and without cyclic compression) were compared with scaffolds cultured in the cell culture incubator (static; Figure 1). Images of hBMSC-seeded scaffolds showed no differences in cell distribution and morphology between different culture time points (1 or 7 days) and conditions (static, bioreactor without cyclic compression, or bioreactor with cyclic compression). Cells were homogeneously distributed throughout the scaffold and showed

| HBMSC migration is reduced by cyclic mechanical compression and osteogenic differentiation
The migration behavior of hBMSCs inside the collagen scaffolds was investigated by an inwards-migration assay where the median migration distance e D d into the 3D biomaterial was quantified 3 days post-seeding ( Figure S1).
In response to cyclic compression, e D d decreased in a magnitudedependent way with the strongest reduction of 25% observed for scaffold B (E = 12.3 kPa) and a compression magnitude of 10%.
Although the same trend was observed for the softer scaffold A (E = 3.4 kPa), the reduction of cell migration was less pronounced and did not reach statistical significance (Figure 2a). In contrast, the stiffness of the collagen scaffold did not affect migration in the unstimulated control.
Based on these results, we asked whether that the reduction in migration is the consequence of a differentiation process induced by cyclic compression. This was based on, firstly, the previously   (Figure 2b). Thus, the result supports a putative connection between the cell's migration behavior and differentiation status.

| Cyclic mechanical compression downregulates the expression of key osteogenic marker genes but upregulates BMP2 expression
To investigate whether the reduction of migration in response to cyclic compression is a result of osteogenic differentiation, the impact of cyclic mechanical compression on the mRNA expression of osteogenic marker genes was quantified by qPCR ( Figure 3)  (1) Human BMSCs were seeded in collagen scaffolds (12.3 kPa) and cultured with or without cyclic compression (f = 1 Hz, ε = 10%) for 6 days.
(3) The cell number was increased 5 times and the medium volume was reduced from 27 ml to 12 ml ("high cell-to-medium ratio"). (b) Human BMP2 ELISA of collected bioreactor media from all loading experiments were conducted. The relative gene expressions of (c) RUNX2, (d) BMP2, (e) BMP4, BMP6, and Noggin (n = 5 hBMSC donors). (f) Validation of rhNoggin efficiency by western blot analysis of p-Smad1/5 level after 60 min of rhBMP2 stimulation. Phosphorylation was normalized to GAPDH (n = 3, one donor). (g) Scaffolds were cultured under high cell-to-medium ratio with or without 10% cyclic compression and with or without rhNoggin stimulation (100 ng/ml, added at Days 1, 3, and 5), and RUNX2 and ID1 expression were analyzed. Gene expressions were analyzed by qPCR. HPRT1 was used as the reference gene, and expressions were normalized to the untreated control (low cell-to-medium-ratio, without cyclic compression; n ≥ 4 for one donor). Statistical significance was tested via Mann-Whitney test (two-sided) with Bonferroni correction, *p ≤ 0.05, **p ≤ 0.01 [Colour figure can be viewed at wileyonlinelibrary.com] mechanical compression. However, BMP2, a growth factor known to be a potent inducer of osteogenic differentation, was clearly upregulated under 10% cyclic compression.

| Limited biochemical conditioning of the culture medium during bioreactor culture
To study whether the observed increase in BMP2 gene expression resulted in an increased protein expression and secretion, BMP2 protein concentrations β(BMP2) were analyzed using an enzyme-linked immunosorbent assay (ELISA) specific for human BMP2 (Figure 4b, left).
In agreement with the expression data, a slight increase of BMP2 was detected in culture media of samples stimulated with 10% compression

| A higher cell-to-medium ratio significantly increased BMP2 concentrations in the culture medium
With the goal to promote medium conditioning and autocrine BMP2 signaling during bioreactor cultivation (6 days), the ratio between cell number and medium volume was increased in subsequent experiments.
The number of scaffolds per bioreactor was increased from one to five, and the medium volume was reduced from 27 ml to 12 ml (minimal filling volume of the bioreactor), resulting in an increase in cell-tomedium ratio from R low = 0.56 × 10 4 to R high = 6.25 × 10 4 cells/ml. Next, we analyzed the concentration of BMP2 in the conditioned media collected from bioreactors that were supplemented with rhBMP2. Also here, mechanical loading increased BMP2 concentration slightly but nonsignificantly (Figure 4b, dark vs. light orange box) e β BMP2 ð Þ 0% . It has to be mentioned that the detected BMP2 concentrations were overall very low compared with the initially added amount of rhBMP2 (135 ng/ml). To understand this discrepancy, we analyzed the BMP2 stability in the bioreactor. Therefore, 135 ng/ml rhBMP2 was added to the bioreactor experiment (conducted without cells), and medium samples were collected after 30 min and at Days 1, 3, 5, and 7 (supplementary Figure 5). Already after one day, BMP2 concentration decreased to about one third and declined further during culture. Together, this indicated that BMP2 stability is transient under cell culture conditions.

| Mechanical loading enhances RUNX2 mRNA expression only in a BMP-enriched environment
Signaling initiated by BMP2 directly stimulates the expression of RUNX2 and balances its transcriptional activity (Nishimura, Hata, Matsubara, Wakabayashi, & Yoneda, 2012;Yang et al., 2016). The biomaterial used in this study was made from fibrillar collagen, a component of the extracellular matrix that is a relevant cell substrate throughout the bone healing process (Petersen et al., 2018). The macroporous architecture assures three-dimensional cell morphology and provides enhanced nutrient and oxygen supply for the cells even in the center for the scaffold. Collagen crosslinking during production protects the scaffolds from fast enzymatic degradation.
Both, collagen crosslinking and its elastic deformation behavior allowed repetitive compression without major shape-changes even over long periods of time (Brauer et al., 2019;Petersen et al., 2018). By changing the solid content, the wall stiffness was tuned without affecting the pore architecture ( Figure 1a). As the stiffness of the substrate that cells adhere to is known to be an important regulator influencing cellular behavior (Engler, Sen, Sweeney, & Discher, 2006), scaffolds with bulk stiffnesses of 3.4 kPa and 12.3 kPa were used in this study. The utilized scaffolds are characterized by low elastic moduli mimicking the physical environment in the fracture gap early after bone injury where a soft tissue matrix is present within the fracture gap.
The early healing phase is especially sensitive to mechanical signals and is believed to lay the ground for the entire repair process.
Indeed, allowing a limited interfragmentary movement in the early phase was shown to enhance fracture healing in a sheep model (Klein et al., 2003). Axial interfragmentary movement was reported to be the main loading regime in animal osteotomy models with an external fixator (Bottlang & Feist, 2011). Interfragmentary compression that occurs as a consequence of weight bearing in experimental bone healing studies was reported to range between 10 and 33% (Klein et al., 2003;Schell et al., 2008) and 2 and 20% (Claes & Heigele, 1999;Klosterhoff et al., 2017)

| Cell migration is downregulated by mechanical loading independent of osteogenic differentiation
In this study, a significant reduction of the migration capacity of hBMSCs was observed in response to the application of 10% cyclic compressive loading. Previous studies have already shown that cell migration is influenced by external mechanical forces. Contrary results were reported depending on the cell type and experimental system used. A reduction of the migration capacity in response to cyclic stretch or compression has been demonstrated for alveolar epithelial cells (Desai, Chapman, & Waters, 2008) and rat bone marrow MSCs, respectively (Ode et al., 2011). However, a stimulating effect of cyclic mechanical stretch was seen for osteoblasts (Bhatt et al., 2007) and fibroblasts (Raeber, Lutolf, & Hubbell, 2008). So far, migration has been studied mainly in two dimensional environments (Bhatt et al., 2007;Desai et al., 2008;Li, Wernig, Hu, & Xu, 2003). A direct comparison with these studies is difficult because integrin expression pattern may be essentially affected by the dimensionality of the environment (Hong & Stegemann, 2008). In this regard, the present work is compared with the study by Ode et al. (2011) in which rat MSCs seeded in a 3D fibrin gel matrix were subjected to cyclic compression. In line with the data presented here, the migration capacity was significantly reduced by about 40% after the application of 20% compressive strain ( f = 1 Hz, 72 hr). However, for the evaluation of MSC migration, the cells had to be removed from the 3D fibrin matrix and transferred into 2D transwell plates. In the present study, a cell migration assay was used that allows to study the direct impact of cyclic compression on cell migration in a 3D environment, mimicking the in vivo situation.
Interestingly, the strongest reduction of the migration distance in response to 10% mechanical strain could be seen in the stiffer scaffold B (25% reduction in scaffold B vs. 12% reduction in scaffold A). Because the median migration distance of uncompressed control hBMSCs was found to be independent of scaffold stiffness, this indicated a stiffness-dependent cell response to load. Cells on stiffer substrates may sense extrinsic mechanical stimulation differently because, in stiffer scaffolds, the influence of the cell's own forces on the extrinsically induced wall deformation is limited and the signal might be transmitted more directly to the cells. However, for a deeper understanding of the force-interplay and the molecular consequences, more mechanistic investigations are needed.
Based on our findings it can be concluded that the migration of hBMSCs is strongly affected by in vivo-motivated mechanical loading regimes. An accumulation of attracted bone marrow MSCs at locations of increased mechanical strain could represent a conceivable physiological mechanism in healing of bone fractures and bone defects. Previous in vivo and in vitro studies reported that mechanical loading induces the differentiation of osteoprogenitors into osteoblasts (Jagodzinski et al., 2008;Michalopoulos et al., 2012;Shin Kang et al., 2013;Sittichokechaiwut et al., 2010;Turner, Owan, Alvey, Hulman, & Hock, 1998). Concerning our finding that hBMSC migration is reduced in response to cyclic compression, this led to the assumption that a reduction in migration is caused or at least accompanied by an onset of osteogenic differentiation.
This was initially supported by our observation that the migration capacity of hBMSCs was clearly reduced after induction of osteogenic differentiation using OM. Also, previous studies have found reduced migration with advancing differentiation (Ichida et al., 2011;Sliogeryte et al., 2014). However, the fact that cyclic compression failed to induce osteogenic differentiation in our study indicated that loading and differentiation interfered with hBMSC migration independent of each other.

| Towards a deeper understanding how cyclic compression influences osteogenic differentiation
To study the impact of mechanical loading on cell behavior, multiple different experimental setups and loading devices were developed. Here, only studies with 3D cell cultures and comparable mechanical loading regimes (cyclic biomaterial compression) are discussed.
RUNX2 (Cbfa 1), a member of the RUNT domain gene family, is an indispensable transcription factor for osteoblast differentiation (Ducy, Zhang, Geoffroy, Ridall, & Karsenty, 1997). In our study, the expression of RUNX2 in hBMSCs was found to be downregulated upon cyclic compressive loading (Figure 3). This observation stands in contrast to previous reports in which comparable experimental setups were used (Jagodzinski et al., 2008;Michalopoulos et al., 2012 ; Shin Kang et al., 2013 ;Sittichokechaiwut et al., 2010). It is known that RUNX2 regulates the expression of osteoblast-specific genes such as collagen type 1, bone sialoprotein, osteocalcin, and RUNX2 itself (Franceschi & Xiao, 2003). Thus, the downregulation of COL1A2, which encodes the pro-alpha2 chain of type I collagen and osteocalcin that we observed in our study is most likely a consequence of the downregulation of RUNX2.
Only one study was identified that reported an inhibitory effect of mechanical stimulation on the RUNX2 expression and, consequently, on other osteogenic markers. In this study, continuous application of mechanical loading over up to 10 days might be responsible for the negative impact (Shi et al., 2011). However, this can be excluded as an explanation in our study because, here, only intermittent loading (repeated cycles of 3-hr cyclic compression and 5-hr break) was applied.
In contrast to RUNX2, COL1A2, and osteocalcin, we found a significant upregulation of osteopontin mRNA expression in response to 10% compression in the softer scaffold A (E = 3.4 kPa). A possible explanation for this discrepancy is that osteopontin expression is regulated by an alternative mechanism independent of RUNX2. The expression of osteopontin was previously found to be sensitive to mechanical stimulations (Klein-Nulend, Roelofsen, Semeins, Bronckers, & Burger, 1997;Sittichokechaiwut et al., 2010). Osteopontin is an abundant noncollagenous protein in the extracellular matrix of bones and serves as a cell attachment point mediated through integrin binding (Toma, Ashkar, Gray, Schaffer, & Gerstenfeld, 1997). It is conceivable that hBMSCs establish stronger attachments to their substrate in response to the cyclic deformation of the walls by increased osteopontin secretion.
In summary, aside from osteopontin, all investigated osteogenic marker genes were found to be downregulated in response to cyclic compression. This surprising finding contradicts the majority of literature on this topic as well as our original study hypothesis. A reason for the discrepancy may be found in the specific experimental conditions. In comparison with others, we can exclude previously reported indirect effects of mechanical loading on oxygen concentration and nutrient supply inside the biomaterial due to the chosen macroporous architecture (Witt et al., 2014). A further major difference is the low total cell number in relation to the volume of cell culture medium in the bioreactor (1.5 × 10 5 cells/27 ml medium) compared with the cultivation of 3D cell seeded constructs in well plates with low medium volume (Michalopoulos et al., 2012;Shin Kang et al., 2013;Sittichokechaiwut et al., 2010). This, in combination with the small but decisive fluid flow in the bioreactor, leads to a strong dilution of signaling proteins secreted by the cells and hinder autocrine biochemical self-stimulation.

| Cyclic compression possess an osteoinductive potential only in a BMP-enriched environment
As an indicator for an osteogenic response to cyclic compressive loading, and in contrast to the osteogenic genes mentioned above, BMP2 expression and secretion were found to be upregulated. In previous studies, BMP2 was already described to be a target of mechanotransduction. An upregulation in response to mechanical loading was reported by several research groups (Rui et al., 2011;Sumanasinghe, Bernacki, & Loboa, 2006). The growth factor BMP2 is known to be an indispensable player during bone repair (Tsuji et al., 2006), and its administration leads to bone defect healing (Wulsten et al., 2011). In fact, BMP signaling regulates the transcription of RUNX2 through the activation of the Smad transcription factors (Nishimura et al., 2012). Based on this connection, we hypothesized that cyclic compression does not induce osteogenic differentiation per se but only in presence of BMP.

SUPPORTING INFORMATION
Additional supporting information may be found online in the Supporting Information section at the end of the article.