Scaffold pore geometry influences bone-like tissue formation in dynamic cell culture conditions

Cells sense and respond to scaffold pore geometry and mechanical stimuli. Many fabrication methods used in bone tissue engineering render structures with poorly controlled pore geometries. Given that cell-scaffold interactions are complex, drawing a conclusion on how cells sense and respond to uncontrolled scaffold features under mechanical loading is difficult. Here, monodisperse templated scaffolds (MTSC) were fabricated and used as a well-defined porous scaffolds to study the effect of dynamic culture conditions on bone-like tissue formation. Human bone marrow derived stromal cells were cultured on MTSC or conventional salt-leached scaffolds (SLSC) for up to 7 weeks, either under static or dynamic conditions. Spinner flask bioreactors were used to provide dynamic culture conditions to apply mechanical loading on the seeded cells through wall shear stress. The influence of controlled spherical pore geometry of MTSC subjected to static or dynamic conditions on the cell response, specifically on bone-like tissue formation, structure and distribution was investigated. As revealed by collagen staining, mRNA expression levels and micro-computed tomography (micro-CT) images, dynamic conditions supported osteoblast cell differentiation and promoted a more regular extracellular matrix (ECM) formation and mineral distribution in both scaffold types compared to static conditions. The results showed that regulation of bone-related genes and the amount and the structure of mineralized ECM were dependent on scaffold pore geometry and the mechanical cues provided by the two different culture conditions. Under dynamic conditions, SLSC favored osteoblast cell differentiation and ECM formation, while MTSC enhanced ECM mineralization. The spherical pore shape in MTSC supported a more trabecular bone-like structure under dynamic conditions compared to MTSC statically cultured or to SLSC under either static or dynamic conditions. These results suggest that cell activity and bone-like tissue formation is driven not only by the pore geometry but also by the mechanical environment. This should be taken into account in the future design of complex scaffolds, which should favor cell differentiation while guiding the formation, structure and distribution of the engineered bone tissue. This could help to mimic the anatomical complexity of the bone tissue structure and to adapt to each bone defect needs. Impact statement Aging of the human population leads to an increasing need for medical implants with high success rate. We provide evidence that cell activity and the amount and structure of bone-like tissue formation is dependent on the scaffold pore geometry and on the mechanical environment. Fabrication of complex scaffolds comprising concave and planar pore geometries might represent a promising direction towards the tunability and mimicry the structural complexity of the bone tissue. Moreover, the use of fabrication methods that allow a systematic fabrication of reproducible and geometrically controlled structures would simplify scaffold design optimization.


Introduction
Restoring large bone defects remains a clinical challenge [1]. Although autologous bone grafts are the gold standard treatment, problems related to donor site morbidity and limited amount of available bone tissue remain a major drawback [2]. Implantation of biocompatible scaffolds that support the migration of the patients' own cells, their proliferation and new bone tissue formation (BTF) at the defect site could represent a cost-and time-effective alternative [3].
Numerous in vitro and in vivo studies have focused on the optimization of scaffold architectures to improve the biological response and to fulfill a variety of functional requirements [4][5][6][7][8][9][10]. Despite evidence that scaffold architecture affects its performance and cellular response, it is still not clear what the best pore size, shape and interconnectivity of the scaffold architecture should be. One reason is that the selection of an optimal bone graft varies from patient to patient and relies on, the defect size and shape, on the anatomical location and the specific biomechanical or functional requirements at that site [11]. Given that cell-scaffold interactions are complex and that the biological mechanism by which the scaffold mediates BTF is still missing [12], drawing a general conclusion on how cells sense and respond to a certain architectural scaffold feature is difficult [11]. This becomes even more challenging if one considers the poorly controlled structure of scaffolds typically used for bone tissue engineering. Many of the fabrication methods used in bone tissue engineering (e.g. salt leaching, gas foaming, lyophilization) often fail to allow an independent variation of individual geometric parameters [13]. To better understand cell-scaffold interactions and thus advance on the development of new strategies to enhance BTF, it is necessary to use methods that enable the fabrication of reproducible and geometrically controlled structures. Moreover, the use of fabrication methods that allow a systematic fabrication and evaluation of the influence of individual scaffold parameters on cell behavior and promoting BTF would simplify scaffold optimization to fulfill the requirements of the patient and the mechanical conditions at the defect site.
The next generation of three dimensional (3D) scaffolds to enhance in vivo BTF aims at supporting the amount of bone tissue formed but also at closely mimicking the anatomical organization of bone tissue and its extracellular matrix (ECM) at the defect site [14]. Attention has been given to the stimulating and guiding role of the geometrical design of scaffolds [15][16][17][18][19][20][21][22][23][24][25]. In vitro and in vivo studies have suggested that the structure of engineered bone-like tissue can be modulated by controlling the scaffold pore diameter [21,24,26,27], the channel orientation [16] and the porous scaffold architecture [17,20]. In vitro, two-dimensional pore geometry and surface curvature have further shown to have a strong influence on cell viability [18], cell proliferation [15,22,25] and differentiation [19,25] or overall tissue formation [15,22,25]. Cell-independent mineral nucleation was shown to be regulated by material surface curvature [19,28]. It has been suggested that controlling the 3D scaffold pore geometry has a major impact on the bioactivity of a scaffold towards stem cell differentiation into bone tissue forming cells and subsequent ECM formation and mineralization [23]. In that study, silk fibroin (SF), a widely-used material for bone tissue engineering applications [29][30][31], was selected as material for scaffold fabrication. A new fabrication method for SF scaffolds with a highly controlled and monodisperse porous architecture was developed [23], which were referred to by the authors as inverse opal scaffolds. Here we name them, for consistency to the salt leached scaffolds based on the porogen, monodisperse template scaffolds (MTSC). A systematic evaluation of physical features and biological response of MTSC was performed relative to conventional salt-leached SF scaffolds (SLSC). The inverse opal method used poly(caprolactone) (PCL) as spherical monodisperse pore template. MTSC result in a close packed and regular spherical pore arrangement. While keeping comparable stiffness, roughness and mean pore diameter, MTSC showed a monodisperse pore diameter distribution, regular spherical shapes and lower overall porosity compared to SLSC [23]. While no differences were observed in cell proliferation, human bone marrow derived stromal cells (hBMSCs) cultured on MTSC under static conditions produced significantly more matrix mineralization compared to cells cultured on SLSC [23].
In addition to scaffold stiffness and topography (pore geometry), cells sense and respond to changes in their mechanical environment [32]. Spinner flask bioreactors have been used in tissue engineering to overcome nutrient constraints through large scaffolds but also to induce shear stresses in vitro through generation of increased fluid flow. Fluid-induced shear stress has been shown to alter the cytoskeletal composition of osteoblastic cells [33] and to induce osteoblast cell differentiation [34] and ECM mineralization [35][36][37]. Although substrate geometry and shear stresses have been shown to guide the osteoblast cell differentiation, so far studies were limited to the investigation of these two components independently.
Our goal was to assess the influence different scaffold pore geometries on fluid flow and bone-like tissue formation. More specifically, we investigated whether dynamic cell culture conditions, generated by culturing hBMSC in spinner flask bioreactors, play a role on the osteogenic cell fate and on the spatial organization of mineralized tissue within 3D SF scaffolds exhibiting either spherical and monodisperse (MTSC) or faceted and polydisperse (SLSC) porous architectures [23]. Furthermore, we investigated whether scaffold type and shear stress could have a synergistic effect on increasing cell differentiation and mineralization.

Production of SLSC
Sodium chloride (NaCl) crystals with a granule diameter between 300-400 µm were used as a faceted polydisperse porogen. 1 ml of the 17% (w/v) SF-HFIP solution was added to 2.5 g of NaCl crystals. The HFIP was allowed to evaporate for 3 days. β-sheet conformation was induced by immersing the SF-salt blocks into 90% (v/v) methanol (Merck) in UPW for 30 minutes and blocks were subsequently dried overnight [39]. NaCl was leached by washing the SF-salt blocks in UPW for 2 days. SF blocks were cut and punched (Kai medical) into disks (3 mm height, 5 mm diameter). SLSC were sterilized in phosphate buffered saline (PBS; Medicago AB) at 121°C, 1 bar for 20 minutes. The SLSC's fabrication process is depicted in figure 1A.

Production of MTSC
A 10% (w/w) PCL in dichloromethane solution (SIGMA) was emulsified in a 2% (w/w) poly(vinyl alcohol) (PVA; SIGMA) aqueous solution within a glass capillary microfluidic device to produce monodisperse droplets [23]. The inner and the outer flow rates were adjusted to 5 ml/hr and 8 ml/hr, respectively. The droplets were collected in an excess of PVA solution and left resting for approximately five weeks until dichloromethane was completely evaporated. The particles were washed 5 times in water and once in ethanol. Subsequently, particles were placed in moulds (7 mm diameter, 3 mm height) and slightly agitated to close pack and dry. The particles were then heated to 57°C at 25% relative humidity for 1 hour to form necks between the particles. 12% SF solution was infiltrated into each particle pack and let dry. β-sheet conformation was induced by immersing the SF-salt blocks into 90% (v/v) methanol in UPW for 1 hour [40]. Following 8 hours in dichloromethane to remove the porogen particles, samples were washed once in ethanol to remove any solvent residues and left to dry. Finally, MTSC were sterilized in PBS at 121°C, 1 bar for 20 min. Due to the scaffold fragility after autoclaving, scaffolds were not punched further out. The MTSC's fabrication process is depicted in figure 1B.

Computational fluid dynamics (CFD) simulation
To quantify the wall shear stress (WSS) generated within the two scaffold types, a CFD model was developed based on the spinner flask bioreactor ( figure 2A). For saving computational cost, a computeraided design (CAD) was used for creating idealized scaffold geometries (figures 2B and C). Based on the size and shape of the NaCl and PCL particles used as porogen in SLSC and MTSC, the scaffold geometric features for CAD were determined as: a pore diameter (d) of 330 µm for both scaffold types; a cubical pore shape for SLSCs and a spherical pore shape for MTSCs. According to previous measurements, the average porosity (φ) of MTSC and SLSC was 84.7% and 92.8%, respectively [23].
Two CFD domains were meshed with 1,187,159 tetrahedron elements (SLSC scaffold) and 2,650,122 tetrahedron elements (MTSC scaffold), respectively, using the same mesh strategy as previously [41].
The density (ρ) and dynamic viscosity (µ) of the culture medium were 1000 kg/m 3

Cell culture
hBMSCs were isolated and characterized from bone marrow (Lonza) as previously described [20].

Dynamic culture conditions
Spinner flask bioreactors were chosen as a dynamic culture to induce fluid flow through and around the cell-scaffold construct. The culture medium was stirred through a magnet at approximately 70 rpm.

Histology
Constructs were fixed in 4 % formaldehyde overnight at room temperature. Samples were dehydrated, embedded in paraffin and cut into 10 µm thick sections. Weigert's hematoxylin and Sirius red were used to stain the cell nuclei and collagen simultaneously. Images were taken using a Slide Scanner Pannoramic 250 (3DHISTECH). To improve the visualization of the histological staining, contrast and brightness of figure 5 and 6 were adapted using the same setting parameters in all images.
Figures show representative images per group.

RNA isolation and Real-time PCR analysis
Total RNA was isolated using Trizol reagent (Invitrogen), according to the manufacturer's protocol. Each scaffold was disintegrated at 25'000 rpm for 10 seconds per cycle (6 cycles) using steel balls and a Minibead Beater (Biospec). Samples were placed on ice in between cycles. Total RNA (1 µg) was reverse transcribed to cDNA at 42C for 60 min using the High Capacity RNA-to-cDNA kit (Applied Biosystems), according to the manufacture's protocol. 10 µl of cDNA was diluted into 10 µl of RNAse free water (SIGMA).
Real-time PCR (Biorad CFX96) was performed using TaqMan probe detection (Applied Biosystems). Real-time PCR was done for glyceraldehyde-3-phosphate dehydrogenase (GAPDH, Hs02758991_g1), collagen type I (COL1A2-I, Hs01028956_m1), alkaline phosphatase (ALPL, Hs01029144_m1) and osteocalcin (BGLAP, Hs01587814_g1). A negative control without cDNA template was run for each assay. Values were normalized to GAPDH mRNA levels as an internal control and were presented as fold change compared to SLSC in static conditions. Data was analyzed using the comparative Cq method (2-ΔCCq) [42].

Micro-computed tomography
For quantification and visualization of the engineered mineralized bone-like tissue, scaffolds were analyzed using a micro-CT imaging system (µCT40, Scanco Medical AG). Samples were noninvasively scanned weekly for up to 7 weeks. All bioreactors were scanned at 36 µm resolution, 45 kVp and 177µA. An integration time of 200 ms and 2-fold data averaging was applied. Cylindrical volumes of interest were selected for the analysis. A constraint Gaussian filter of 1.2 and a filter support of 1 was used to suppress noise. Mineralized tissue was segmented from non-mineralized tissue using a global thresholding procedure [43].

Statistics
Data (n=3) represent the mean ± standard deviation (SD). SPSS® 22.0 was used for statistical analysis. A one-way ANOVA with post-hoc Bonferroni correction was performed to compare differences among groups. P-values ≤ 0.05 were considered statistically significant. Figure 5 shows upper median samples based on the BV/TV values of the whole scaffold.

Dynamic conditions up-regulated the expression of osteoblastic differentiation markers in SLSC
and MTSC compared to static conditions. This was evidenced by a significant increase of the ALPL mRNA levels by 3.1-fold (p = 0.00045, figure 7B) or of the BGLAP mRNA levels by 3.5-fold (p = 0.00500, figure 7C) in cells cultured on MTSC and on SLSC, respectively, compared to static conditions.
The expression of these genes was also influenced by pore geometry. Under dynamic conditions, these genes were differentially expressed in both scaffolds. After 7 weeks of culture, ALPL mRNA levels were markedly up-regulated in MTSC compared to SLSC (8.4-fold, p = 0.00002) while BGLAP mRNA levels were significantly down-regulated (51-fold, p = 0.00043, figure 7B, 7C). Under static conditions, COL1A2 mRNA levels were up-regulated on SLSC compared to MTSC (p = 0.03, figure 7A).

ECM mineralization
Micro-CT was used to evaluate the influence of scaffold pore geometry and culture condition on mineralized bone-like tissue formation.

Discussion
In this study, we used a manufacturing method to systematically fabricate scaffolds with highly controlled monodisperse porous architecture. The aim of the current study was to compare two scaffold manufacturing methods on how scaffold pore geometry influences the response of cells to culture conditions. Evaluated were the distribution and osteogenic cell differentiation as well as the amount, structure and spatial distribution of the mineralized tissue formed. Dynamic culture conditions were used to overcome the nutrient flow limitations and to apply mechanical loading through WSS. Cells (hBMSCs) differentiated towards the osteogenic lineage when cultured either under static or dynamic conditions for both scaffold pore geometries. Nevertheless, distinct responses to fluid flow in gene expression, the amount, structure and spatial distribution of the mineralized ECM formed were shown to be dependent on scaffold pore geometry.
The applied SLSC and MTSC specimens were produced following the same procedure described by Sommer et al 2016 [23]. Geometrical features reported by the authors were thus used here to aid the interpretation of the obtained results [23]. Despite their similar local stiffness and a comparable mean pore diameter of about 200 µm, SLSC and MTSC differed in the pore diameter distribution and pore geometry. This resulted in a lower porosity in the MTSC (84.7 ± 2.1 %) compared to SLSC (92.8 ± 0. 6%) (figure S1) [23]. We used histological staining's to evaluate at the influence of these differences of scaffold porosities on cell proliferation capabilities. The histological results showed that cells could adhere well on both scaffold types but that the ability of cells to proliferate was reduced in the MTSC scaffolds under static conditions, which could indeed be due to a lower pore interconnectivity. Similarly, previous studies have shown that SF scaffolds with similar pore diameter and porosities were able to better support cell proliferation in vitro [9,48,49]. Assuming a similar MTSC porosity as described by Differences in the up-regulation of mid (ALPL) or late (BGLAP) differentiation markers between MTSC and SLSC under dynamic conditions suggest that cells on SLSC were at a more matured differentiation stage. The magnitudes of mechanical forces have been shown to influence gene expression in osteoblastic cells [53]. According to the computational model, the SLSC had a higher surface area fraction (98.21%) exposed to the WSS range under dynamic conditions for osteogenesis than the MTSC (82.3%). In line, dynamic SLSC upregulated the expression BGALP which is a late osteogenic-related marker involved in the bone mineralization. While the underlying mechanisms are not fully clear, we speculate that cells reacted distinctly to the different WSS developed in the two pore geometries.
Osteoblast cell differentiation is characterized by the formation of a mineralized ECM. Although bonelike tissue was formed in all groups, the amount, the structure and distribution of the mineralized ECM was dependent on the pore geometry and whether the constructs were grown statically or dynamically.
Interestingly and in contrast to the gene expression profile observed, MTSC under dynamic conditions had a significantly increased BV/TV compared to SLSC under dynamic conditions. This is in agreement with our observation that pore geometry can lead to differences in the bone-like structure along the scaffold under dynamic conditions. On the other hand, the computational prediction showed that a larger surface area fraction of the SLSC was exposed to a WSS range for stimulating mineralization than that of the MTSC, which is in contrast to results obtained experimentally. One reason for this discrepancy was that the CFD model is based on empty scaffold geometry neglecting tissue growth. Recent used for making the pores in MTSC, the rock-salt crystal structure of NaCl rendered more faceted pore surfaces in the SLSC scaffolds. It has been reported that surface concavities act as a strong template for in vitro surface mineralization, even in the absence of living cells [28]. Surface mineralization is probably favored within pore concavities as compared to planar surfaces because nuclei are less exposed to shear and are thus able to grow larger in sites where the local fluid velocity is low [19,28]. We hypothesize that MTSC with more regular and concave surfaces favor a more homogeneous deposition of mineral due to inhibited fluid motion close to regularly distributed pore walls. By contrast, the more planar features present in SLSC might result in an unevenly distributed mineral deposition with higher mineral formation at concave surfaces and lower mineral formation at planar surfaces of the pore.
Differences in fluid flow next to concave and planar surfaces might then modulate the local ion concentration affecting the nucleation process and therefore mineral tissue formation. The decrease in BV/TV for SLSC under dynamic compared to static conditions could be due to the higher flow velocities at planar surfaces, which in turn could lead to more fluctuations in the local concentrations of Ca 2+ and PO4 3 , as previously suggested by Bianchi et al 2014 [28]. This effect might be enhanced by the high porosity of the SLSC [23].
In conclusion, our results suggest that scaffold pore geometry and porosity generate different flow velocities/WSS within the scaffolds in the bioreactor, which in turn directly affect cell differentiation and bone-like tissue formation. While SLSC enhanced osteoblast cell differentiation, the spherical and more regular concave pore geometry in MTSC could probably favor nucleation of ions increasing the amount of bone-like tissue formation. These findings point out the role of pore geometry on the stimulation of cell differentiation and indirectly on guiding the mineral deposition process. This suggests the importance of 1) the fluid flow imposed by the scaffold's pore geometry on the cell response and the formation of an engineered bone-like tissue, 2) control and tune the scaffold pore geometry to modulate the amount and structure of engineered bone-like tissue. Fabrication of complex scaffolds, by additive manufacturing methods, comprising regions with more concave pores with others with more planar pore geometries might thus represent a promising direction towards the tunability of the BTF and to mimic the structural complexity of the bone tissue and its spatial distribution to each specific need.
Our findings also demonstrate the relevance of the chosen mechanical loading as loads applied can overrule or affect the influence of the scaffold pore geometry. These conditions should be carefully chosen and included for quality assurance during assessment of new scaffolds.

Acknowledgements
The authors thanks to Trudel Inc. for the supply of the silk cocoons and ScopeM for providing access to their histology equipment. The research leading to these results has received funding from the European Union Seventh Framework Programme (FP7/2007-2013) under grant agreement n°. 329389, 262948 and 336043.

Author Disclosure Statement
The authors declare that no competing financial interests exist.