The drug-eluting composite microspheres consist of porous silica (pSi) suspended in poly(lactic-co-glycolic) (PLGA) microspheres using methods previously described (Tsao et al., 2018). Unless stated otherwise, the reagents were purchased from Sigma-Aldrich (St. Louis, MO, United States). Briefly, 300 μl of tetraethyl orthosilicate was added dropwise to a mixture of 0.36% (w/v) tannic acid, 66% (v/v) ethanol, and 33% (v/v) ammonium hydroxide. The mixture was stirred continuously for 3 h before the pSi were washed and centrifuged three times in a 1:1 solution of ethanol:water. The pSi particles were then lyophilized and stored at −20°C until needed. To load the pSi with GW9662, 5 mg of lyophilized pSi were incubated in 1 mL of GW9662 (300 μg/mL) dissolved in DMSO with constant stirring for 20 min at 37°C. The GW9662-loaded pSi were washed and lyophilized before proceeding to encapsulation in PLGA microspheres. Briefly, GW9662-loaded pSi were added to a 5% (w/v) PLGA (50:50) (Lactel Absorbable Polymers, Pelham, AL, United States) solution dissolved in dichloromethane (DCM). The PLGA/pSi/DCM mixture was then added dropwise to a 2.5% poly(vinyl alcohol) (PVA) aqueous solution and stirred continuously for 6 h. The resulting microspheres were then centrifuged and washed several times with DI water before they were lyophilized and stored at −20°C until needed.

To verify the presence of pSi in PLGA microspheres, fluorescein isothiocyanate (FITC)-labeled pSi were encapsulated in PLGA microspheres and imaged by confocal microscopy (Nikon D Eclipse C1, Nikon Corporation, Japan). Liquid chromatography-mass spectrometry was used to measure the release of GW9662 from the PLGA-pSi microspheres suspended in an aqueous buffer over the course of 21 days.

Porous Coll/MgHA scaffolds with and without PLGA-pSi microspheres were fabricated as described previously with minor modifications (Minardi et al., 2015) (Supplementary Figure 1A). Type I bovine collagen (Nitta Casings, Bridgewater, NJ, United States) was chosen as the organic matrix upon which Mg-doped HA nanocrystals would nucleate during collagen fibrils self-assembly. Briefly, 200 mg of collagen was dissolved in 20 mL of acetic acid buffered at a pH of 3. To the collagen solution, 190 μL of 85% H₃PO₄ (Sigma-Aldrich) and 20 mL of DI water were added. Collagen fibrils containing HA and MgHA were reconstituted by adding the acidic collagen solution dropwise to a basic suspension consisting of 0.35 g of Ca(OH)₂ and 0.166 g of MgCl₂⋅6H₂O (Thermo Fisher, Waltham, MA, United States) in 20 mL of water. The ratio of Ca(OH)₂ and MgCl₂⋅6H₂O in the basic suspension was set to 15% as a higher ratio was needed in the precursor solution to achieve the 5–6% molar substitution of Mg²⁺ in the final lattice structure. The mixture was aged for 24 h at 37°C followed by wet crosslinking with 1,4 butanediol diglycidyl ether (BDDGE) (2.5 mM) (Sigma-Aldrich) at 4°C for 48 h at a pH of 8. The mixture was then centrifuged and washed several times with DI water to remove crosslinking solution, leaving behind a slurry-like mixture. In some cases, drug-eluting microspheres were added to achieve approximately 0.52 μg GW9662/scaffold. To fabricate a porous structure, the slurry was aliquoted into 96-well plates (70 μL/well) and lyophilized after undergoing controlled freezing (Minardi et al., 2015). The resulting Coll/MgHA scaffolds, referred to either MgHA or MgHA/pSi for brevity, were then stored at −20°C until needed.

Fourier-transformed Infrared (FTIR) was performed using an Alpha-Platinum Bruker Spectrometer (Billerica, MA, United States). Prior to acquiring their FTIR spectra, the porous scaffolds were flattened into thin disks using a mortar and pestle and were then kept at 100°C in a heating block for an hour to remove residual water. Thermogravimetric analysis (TGA) was performed on a Q50 (TA Instrument, New Castle, DE, United States) by heating scaffolds (<10 mg) from 25 to 1000°C at a rate of 10°C/min. The molar ratio of Mg²⁺ to Ca²⁺ present in the mineral phase of the scaffolds was determined through inductively coupled plasma-mass spectrometry (ICP-MS) (PerkinElmer NexION 300D, Waltham, MA, United States), using a multi-element reference standard of Ca²⁺ and Mg²⁺ (ICP CAL-STD EARTH ALKALI, Inorganic Ventures, Christiansburg, VA, United States) in 1% nitric acid. For ICP-MS, scaffolds were digested overnight in 1% nitric acid. The supernatant was collected after a brief centrifugation, filtered using a 0.22 μm syringe filter, and diluted in 1:1200 in 1% nitric acid to ensure ion concentration were within standard curves. Finally, the morphology and distribution of surface pore size of the scaffolds were assessed by scanning electron microscopy (SEM) on a JCM 5000 (JEOL, Peabody, MA, United States) and analyzed using ImageJ (NIH Image). Lyophilized scaffolds stored in −20°C were kept overnight in a desiccator and were then gold sputter coated (20 mA, 60 s) to remove accumulated water prior to SEM imaging. Approximately 146 and 119 pores were analyzed for the MgHA and MgHA/pSi scaffolds, respectively.

Bone marrow-derived hMSCs were acquired from the adult stem cell distribution center at Texas A&M Health Science Center Institute for Regenerative Medicine in accordance with institutionally approved protocols. Bone marrow-derived hMSCs were expanded in complete culture media (CCM) consisting of alpha minimal essential medium (α-MEM, Invitrogen, Carlsbad, CA, United States), 20% (v/v) FBS (Atlanta Biologicals, Flowery Branch, GA, United States), 2 mM L-glutamine (Invitrogen), and 100 U/ml penicillin plus 100 μg/ml streptomycin (HyClone^TM Marlborough, MA, United States) at a seeding density of 500 cells/cm². Culture media was changed every 2–3 days until cells reached 70–80% confluency. Expanded cells were then recovered by brief trypsinization (Corning, Corning, NY, United States) and seeded at the top of each scaffold (200,000 cells/scaffold) using a 20 μL aliquot. After incubating for 10 min, scaffolds were flipped 180° and an additional 5 μL of CCM was added to achieve a more uniform distribution of cells. Scaffolds were then incubated for an additional 10 min before CCM was added to well plates, keeping track of the area in which cells were initially seeded. The cells were incubated overnight before scaffolds were switched to different culture media formulations or placed in the perfusion bioreactor. The other culture media formulations included: (1) osteogenic basal media (OBM) consisting of CCM supplemented with 50 μg/mL of ascorbic acid and 5 mM β-glycerol phosphate, (2) OBM supplemented with 10 μM GW9662 dissolved in DMSO as a positive control, and (3) OBM supplemented with an equal volume of DMSO as a vehicle control. A table depicting the scaffold/media formulation used for osteogenic assays can be found in Table 1.

Scaffolds were placed in the wells of the bioreactor (Supplementary Figures 1B,C), positioning them so that surface in which cells were initially seeded faced down toward the channels. A silicone rubber gasket with a diameter similar to the scaffolds was placed on top to block perfusate from flowing around, instead of through, the scaffolds. Once placed in the perfusion bioreactor, a multichannel peristaltic pump (Cole-Palmer, Vernon Hills, IL, United States) was used to adjust the culture media flow rate to achieve a superficial velocity through the scaffolds of approximately 116 μm/s, resulting in an estimated shear stress of about 7 mPa. Shear stress experienced by the cells was estimated using a previously described equation (Grayson et al., 2008):

where τ_w is the fluid wall stress within the pores of the scaffold, μ_m is the viscosity (∼0.77 cPa) of the culture media, u is the interstitial velocity, and d_p is the average pore diameter observed in the scaffolds by SEM imaging. Scaffolds were perfused up to 21 days, with media changes occurring every 2–3 days.

Seeding efficiency was approximated by recovering hMSCs attached to the well plates rather than the scaffolds the day after seeding using standard cell counting procedure with a hemacytometer. To evaluate hMSCs viability and migration into the scaffolds without the influence of soluble factors, live-dead staining—5 μM of Calcein AM and 0.1% Propidium Iodide—was performed after 8 days of static and perfused culture in MgHA/CCM scaffolds. Confocal Z-stack images of the scaffolds were captured with a Nikon D Eclipse C1 upright microscope (Nikon Corporation, Japan) from upstream face of the scaffold to a depth of 200 μm using 10 μm slices. The relative fluorescent intensity units for z-stacked images were evaluated as a function of depth using ImageJ (NIH Image). The depth at which stacked live-dead images decreased to 50% of maximum fluorescence intensity was quantified for both static and perfused samples. Afterward, scaffolds were cut to acquire cross-sectional images.

GW9662-treated hMSCs have been previously shown to deposit a unique ECM rich in collagens that are highly expressed in anabolic bone, specifically Coll VI and Coll XII. To evaluate the efficacy of the PLGA-pSi microspheres, immunostaining for Coll VI and XII was performed on scaffolds after 8 days of culture. Briefly, scaffolds were fixed overnight at 4°C in 4% formaldehyde in PBS and washed three times with PBS. Scaffolds were then blocked with 5% goat serum (MP Biomedical) and 0.3% Triton-X (Sigma Aldrich) for 1 h at room temperature before being incubated overnight at 4°C in blocking buffer supplemented with either rabbit anti-Human Coll VI or Coll XII (1:200, Novus Biological, Littleton, CO, United States). Scaffolds were then washed with PBS and were incubated at room temperature for 2 h in blocking buffer with fluorescein-conjugated goat anti-rabbit (1:500, Millipore, Burlington, MA, United States). Z-stack images were taken as described above. Maximum intensity projections was performed on the EZ-C1 software (Nikon Corporation, Japan).

After 8 and 21 days of culture, total RNA was extracted from the scaffolds with RNeasy Mini Kit (Qiagen, Hilden, Germany) supplemented with Trizol (Life Technologies, Carlsbad, CA, United States). After quantifying the purity and concentration of RNA (Infinite M200 Pro, Tecan, Männedorf, Switzerland), isolated RNA with a purity (A260/A280) above 1.9 was used for cDNA synthesis in 21 μL reaction (Superscript III kit, Invitrogen). Approximately 6.5 ng of cDNA was amplified in a 20 μL reaction containing Brilliant III Ultra-Fast SYBR Green QPCR Master Mix with Low ROX (Agilent, Santa Clara, CA, United States) on an Agilent Aria Mx Real-Time PCR System. The primer list and their sequences can be found in the Supplementary Information (Supplementary Table 1). Relative expression was calculated using the ΔΔC_T method normalized to human glyceraldehyde-3-phosphate dehydrogenase (GAPDH) levels, using hMSCs cultured for either 8 or 21 days in 2D monolayer on tissue culture plate with CCM as the calibrator control. Relative fold change in gene expression was calculated using 2^–ΔΔCt method described by Livak and Schmittgen (2001). Statistical comparisons were performed on GraphPad Prism software (San Diego, CA, United States). Briefly, a Bartlett’s test was performed on the C_t values for each gene, for each scaffold/culture condition (Supplementary Table 5) to confirm homogeneity prior to running a two-way ANOVA. A two-way ANOVA was performed on the calculated ΔΔC_T values, followed by a Tukey’s multiple comparison test (n = 3), whose results can be found in the Supplementary Information (Supplementary Tables 5–17). In addition, the number of hMSCs present in the scaffolds was measured via GAPDH expression after 8 days of static and perfused culture in all scaffold/culture conditions using known cell number standards. RNA was extracted from the cell standards in the presence of MgHA scaffolds to mimic the RNA extraction efficiency from the cultured scaffolds. Statistical testing as described above was performed on the calculated cell number (Supplementary Tables 2–4).

After 21 days of static and perfused cultures, scaffolds were washed with PBS and fixed overnight at 4°C with 4% formaldehyde in PBS. After washing with and aspirating excess PBS, samples were allowed to dry overnight in a desiccator. Samples were stored at −80°C until needed. Uncultured scaffolds were used as controls and were prepared as described above. Prior to acquiring μCT scans on a SkyScan 1275 X-Ray Microtomograph (Bruker, Billerica, MA, United States), samples were gently wrapped in parafilm for easier handling during imaging on the SkyScan’s rotating stage. Scaffolds were imaged over 360 degrees with a camera resolution of 18 μm, with images taken every 0.5 degrees using a 28 kV beam and a frame averaging of 3. The cross-sectional images of the scaffolds were reconstructed using NRecon (Micro Photonics) software, keeping the smoothing and beam-hardening compensation settings the same between samples at 0 (Gaussian Kernel) and 41%, respectively. Thresholding was done to better distinguish between parafilm and the scaffolds. Calcium hydroxyapatite phantoms (0.25 and 0.75 g/cm³; Bruker) were scanned using the settings described above and analyzed in the Bruker-MicroCT CT-Analyzer (CTan) software to determine their attenuation coefficient values in order to predict the bone mineral density (BMD) of the scaffolds. After confirming equal variances (Supplementary Table 18), a two-way ANOVA followed by a Tukey’s multiple comparison test were performed on the bone mineral density values (n = 3). Both the ANOVA table and the results for the multiple comparison test can be found in the Supplementary Information (Supplementary Tables 19, 20). CTan software was also utilized to measure morphological characteristics of the scaffolds, specifically percent porosity and the specific bone surface (BS/BV) (Bouxsein et al., 2010) of the scaffolds. The CTVox software was utilized to reconstruct images of the cultured samples. Identical opacity thresholding was utilized for all reconstructions. Maximum intensity projection (MIP) was enabled to highlight dense regions of mineralization.

After acquiring μCT scans, scaffolds were gently removed from the parafilm wrapping and stored at −80°C until needed for mechanical testing. The diameter and thickness of the scaffolds were measured with calipers, and underwent a uniaxial compression test to 15% strain at a rate of 5 μm/s on a DMA 850 (TA Instruments). The compressive modulus was calculated through linear fitting of the slope. Statistical comparisons were performed on the compressive modulus using two-way ANOVA followed by a Tukey’s multiple comparison test (n = 3). The ANOVA table and its corresponding multiple comparison results can be found in the Supplementary Information (Supplementary Tables 22, 23).

After 8 days of static and perfused culture, scaffolds were washed with warm followed by ice-cold PBS. Proteins were extracted using ice-cold RIPA buffer (50 mM Tris HCl, 150 mM NaCl, 1% Triton-X-100, 0.1% SDS, 0.5% Sodium Deoxylcholate, 1 mM EDTA, 1 mM Sodium Pyrophosphate) supplemented with SIGMAFAST^TM protease inhibitors (Sigma-Aldrich) through vortexing followed by centrifugation at 12,000 g for 10 min. The supernatant was recovered and stored at −80°C until total protein content was quantified through Pierce BCA Protein Assay kit (Thermo Fisher Scientific). Protein samples (15 μg) were resolved in hand-casted 10% SDS-PAGE gel (Bio-Rad, Hercules, CA, United States) and transferred to a PVDF membrane (Millipore). The membrane was blocked in 5% milk in TBST buffer (20 mM Tris-base, 150 mM NaCl, 0.2% Tween 20) for 1 h before being incubated overnight at 4°C with mouse anti human BMP2 antibody (R&D Systems, Minneapolis, MN, United States) diluted (1:500) in blocking buffer. After three washes in TBST buffer, the membrane was incubated for 1 h with an HRP-conjugated goat anti mouse secondary antibody (Protein Tech, Rosemont, IL, United States). The membrane was developed for 5 min in WesternSure Chemiluminescent Substrate (LI-COR, Lincoln, NE, United States) and imaged on a C-Digit Blot Scanner (LI-COR). After imaging, the membrane was stripped using Restore Stripping Buffer (Thermo Fisher Scientific) following the manufacturer’s instructions. The membrane was then blocked, incubated with mouse-anti-human GAPDH (1:4000, Protein Tech), and imaged as previously described.

The microstructure of MgHA and MgHA/pSi scaffolds were evaluated via SEM and μCT. SEM imaging of the scaffold surfaces revealed an isotropic porous structure (Figure 1A). Cross-sectional μCT scans of the scaffolds internal microstructure indicated interconnected pores (Figure 1B). When quantifying surface pore size, a wide distribution of pore diameter of 50–200 μm was observed (Figure 1C). Analysis of morphological characteristics showed that inclusion of the drug-eluting microspheres had no effect on either specific bone surface ratio (Figure 1D) or porosity (Figure 1E).

TGA analysis was performed to quantify the mineral-to-protein ratio of the scaffolds. The initial loss in mass observed, approximately 10%, was attributed to residual water in the scaffolds. Thus, the collagen matrix and the mineral phase contributed approximately 33 and 66% of the scaffolds dry weight (Figure 2A). Scaffold composition was further characterized using FTIR (Figure 2B). Large bands were observed in 550 cm^–1 and 1000 cm^–1, corresponding to phosphates in hydroxyapatite. For the organic component, vibrational bands characteristic to amide groups were observed around 1250 cm^–1, 1540 cm^–1, 1650 cm^–1, and 3300 cm^–1. Carbonate ion substitutes within the crystalline structure were also detected, as indicated by the small single peak at 870 cm^–1 and the double bands around 1430 cm^–1 and 1450 cm^–1, which indicate a type-B carbonated substitution in the PO₄^3– position (Landi et al., 2008; Kourkoumelis and Tzaphlidou, 2010). XRD analysis showed both scaffolds shared a similar profile to that of native bone tissue reported in the literature, specifically large peak at 32 degrees and a smaller peak at 26 degrees (Figure 2C). The Mg/Ca substitution ratio was evaluated to be approximately 4.8 and 4.2% for MgHA and MgHA/pSi scaffolds, respectively (Figure 2D).

A seeding efficiency above 93% was observed for each scaffold composition (Figure 3A). After 8 days of static culture in MgHA scaffold with CCM, live-dead staining revealed closely packed cells near the surface of the scaffolds, with a majority of cells staying within 100 μm from the surface of the scaffolds. In contrast, hMSCs perfused for 8 days invaded more deeply into the scaffold (Figure 3B). Semi-quantitative analysis of confocal images showed a rapid drop in fluorescence in the static samples, with a 50% reduction approximately 60 μm into the scaffold, as compared to 150 μm for perfused scaffolds (Figure 3C).

Next, cell number was quantified by GAPDH expression for the eight culture conditions (Figure 3D). Two-way ANOVA indicated that perfusion increased the average cell number in all scaffolds and a lack of interaction between perfusion and the scaffold/culture media formulation (Supplementary Table 3). Multi-comparison testing indicated that, in perfused cultures, GW9662 administered locally, increased cell number significantly relative to the DMSO control (Supplementary Table 4).

Fluorescently labeled pSi were used to characterize their distribution within PLGA microspheres (Figure 4A). Images indicated that the pSi were distributed uniformly within individual microspheres and that pSi content appeared uniform when comparing different PLGA microspheres.

The rate of release of GW9662 from the composite pSi/PLGA microspheres was quantified over a 21-day period since osteogenic differentiation of hMSCs using GW9662 involves continuous application of GW9662 for this duration (Krause et al., 2010). A 14% burst release occurred in the first 24 h, followed by a constant rate of release over the remainder of the experiment, with 82% of the drug released after 21 days (Figure 4B).

It has been previously shown that hMCSs treated with GW9662 secrete an anabolic bone ECM rich in Coll VI and XII, which not only improves retention of hMSCs at an injury site, but also upregulates various osteogenic and angiogenic factors (Clough et al., 2015). To validate the efficacy of the controlled release of GW9662 from PLGA/pSi microspheres and to explore the effects of the fluid shear stress on the expression and deposition of this anabolic bone matrix, qRT-PCR analysis and immunostaining for Coll VI and Coll XII was performed after 8 days of static and perfused culture. The length of culture time was chosen to coincide with the formation of fibrocartilaginous callus seen during fracture repair (Wang and Yeung, 2017).

Consistent with previous findings (Clough et al., 2015), hMSCs cultured under static conditions in media supplemented with GW9662 expressed Coll VI in significantly higher amounts than in the absence of GW9662 and cultures in which GW9662 was provided using microspheres (Figure 5A). Interestingly, the expression for Coll VI was increased by perfusion culture for each condition, though this difference was not significant in the cultures in which GW9662 was added directly to the media. Furthermore, perfusion resulted in comparable levels of Coll VI expression between DMOS, GW-pSi, and GW-media scaffolds. The simple interaction—diminishing returns of GW9662-treatment on Coll VI gene expression by perfusion—was confirmed to be significant by two-way ANOVA (Supplementary Table 6). Immunostained samples indicated comparable amounts of Coll VI in the cultures in which GW9662 was administered via controlled release (Figure 6B) or direct addition to the media (Figure 6C), while the Coll VI was barely detectable in the cultures lacking GW9662 (Figure 6A). Perfusion increased amount of Coll VI detected by immunostaining in each condition, with the levels noticeably higher for samples containing GW9662 (Figures 6E,F) than samples lacking GW9662 (Figure 6D).

Under static conditions, Coll XII gene expression was significantly higher in hMSCs cultured in media supplemented with GW9662 than in cultures lacking GW9662 (Figure 5B). In addition, controlled release of GW9662 from microspheres resulted in comparable expression levels to exogeneous GW9662 administration, highlighting the utility of the composite microspheres as an in situ drug release platform capable of replacing traditional media treatments. It was also noted that despite the lack of GW9662, Coll XII gene expression was positively expressed in DMSO scaffolds (Figure 5B). Furthermore, in contrast to the trend observed for Coll VI gene expression, Coll XII gene expression either stayed consistent or decreased in response to perfusion, suggesting at a suppressive interaction between perfusion and GW9662 treatment on Coll XII gene expression (Supplementary Table 8). Immunostaining of samples cultured under static conditions confirmed comparable deposition of Coll XII between cultures in which GW9662 was administered directly to the media (Figure 6I) and controlled release (Figure 6H). In contrast, Coll XII deposition was scarce in samples lacking GW9662 (Figure 6G). Perfusion increased the amount of Coll XII detected via immunostaining in all conditions, with noticeably higher depositions in samples treated with GW9662 (Figures 6K,L) than in samples lacking GW9662 (Figure 6J).

To assess the osteoinductive potency of the scaffolds, the relative expression of the early osteogenic markers ALP and BMP2 were examined after 8 days of culture, while the late markers osteocalcin (OCN) and osteopontin (OPN) were investigated after 21 days of culture by qRT-PCR analysis. Briefly, no significant difference was observed in ALP expression between GW-media and DMSO scaffolds under static conditions. In contrast, ALP expression increased significantly in GW-pSi scaffolds in comparison to both the DMSO and GW-media scaffolds (Figure 7A). Perfusion further upregulated ALP expression in all scaffold compositions (Figure 7A). No significant interaction between the effects of perfusion and media/scaffold formulation on ALP expression was observed (Supplementary Table 10). BMP2 gene expression in the absence of perfusion increased in response to GW9662 treatment, with both GW-pSi and GW-media scaffolds having significantly higher expression than the DMSO scaffolds (Figure 7B). A significant increase in BMP2 gene expression was observed in the vehicle controls in response to perfusion compared to its static counterpart. In contrast, BMP2 gene expression was negatively impacted in both exogenous and drug-eluting GW9662 samples under perfusion (Figure 7B). The interaction between perfusion and media/scaffold formulation was found to be significant (Supplementary Table 12). To further investigate these trends at the protein level, immunoblotting against BMP2 was performed. Immunoblotting revealed GW9662 treatment and perfusion had a synergistic effect on BMP2 expression at the translational level (Supplementary Figure 2).

A trademark of terminally differentiated hMSCs is the expression of late osteogenic markers. Thus, expression of osteopontin (OPN) and osteocalcin (OCN) at the transcription level was investigated after 21 days of culture. OPN was positively expressed in all scaffolds and media conditions, with the lowest level expressed in the DMSO scaffolds (Figure 7C). Although OPN expression increased in both the exogeneous and GW9662-eluting microspheres scaffolds relative to the DMSO control under static conditions, the trend was not significant. Perfusion significantly upregulated OPN expression in both GW-pSi and GW-media scaffolds relative to the static DMSO scaffolds. The interaction between perfusion and scaffold/media formulation and their effects on OPN expression was not significant (Supplementary Table 14). OCN expression was positively expressed in DMSO scaffolds and responded strongly in a dose-depended manner to GW9662 treatment. Briefly, GW-media and GW-pSi had significantly higher levels of expression relative to the DMSO control (Figure 7D). The application of perfusion stimulated higher levels of OCN expression in GW-pSi and DMSO scaffolds while decreasing OCN expression in GW-media scaffolds. However, fold expression in perfused GW-media scaffolds were still significantly higher than the static vehicle control. The interaction between perfusion and scaffold/media formulation on OCN was found to be significant (Supplementary Table 16).

Mineralization within the scaffolds was detected with μCT. The degree of radio-opacity of reconstructed μCT images of the scaffolds harvested after 21 days of culture (Figures 8A–F) is summarized in Figure 8G. DMSO (Figure 8A) scaffolds cultured under static conditions were largely radiolucent, while larger volume of GW-pSi (Figure 8B) and GW-media (Figure 8C) scaffolds were detected. Furthermore, denser regions, which correlate to degree of mineralization, were detected in all perfused samples, with larger regions detected in the GW-pSi (Figure 8E) and GW-media (Figure 8F) scaffolds. When quantified using the attenuation coefficient method, significantly higher levels in bone mineral density were detected in scaffolds cultured under perfusion relative to their static counterparts in all three scaffolds/media conditions (Figure 8G). The interaction between perfusion and scaffold/media formulation and their effects on BMD was not significant (Supplementary Table 19). Under static conditions, GW9662 treatment resulted in significantly higher BMD levels compared to the DMSO scaffolds, regardless of GW-source. A similar trend was observed under perfusion (Figure 8G).

We then evaluated whether the augmented bone mineral density translated to improved mechanical properties. Although the variances in scaffold compressive moduli were not homogenous due in part to an outlier in GW-pSi perfused samples, a two-way ANOVA and Tukey’s multiple comparison test was still performed to ensure similar analysis could be performed between assays (Supplementary Table 21). Mechanical testing revealed a general trend in improved compressive modulus in response to perfusion (Figures 9A–C). Although the compressive modulus increased in response to perfusion in DMSO and GW-media scaffolds in comparison to their static counterparts, the trend was not significant (Figure 9D). In contrast, a significant difference was observed between perfused and static GW-pSi scaffolds (Figure 9D). Furthermore, the combined treatment of GW9662, through either exogenous means or microspheres, and perfusion resulted in compressively stiffer scaffolds compared to the static MgHA/DMSO scaffold (Figure 9D). Lastly, a strong correlation between bone mineral density and compressive modulus was observed for the 18 pooled samples, with a Person’s correlation r value of 0.696 and a two-tailed p-value of 0.014 (Figure 9E).