Design and validation of a perfusion bioreactor for long-term human bone explant culture

Abstract

Ex vivo bone cultures offer a unique platform to study mechanobiology by preserving bone cells and marrow within their native extracellular matrix. We present a 3D-printable modified ZetOS bioreactor for long-term ex vivo culture of human osteoarthritic trabecular bone. Computational fluid dynamics simulations guided design optimization and supported improved average flow velocity and shear stress throughout the central region of the bone construct in the modified bioreactor design. Humeral head osteotomies from four osteoarthritic patients (age: 65-77 yr, 1 female, 3 males) were prepared within 2 h of surgery. Trabecular bone cores (9.15 ± 0.14  9.5 ± 2.0 mm) were cultured under static conditions (n = 6) or perfused at 252 μL/min (dynamic I, n = 7) or 752 μL/min (dynamic II, n = 9). Cellular metabolic activity, alkaline phosphatase (ALP), and lactate dehydrogenase levels were monitored on 4 time-points. Genomic content (DNA and cell viability) were assessed during the culture period, and micro-CT (μCT)-based morphometric analysis was performed after completion of the culture. Perfused cores maintained high metabolic activity and showed significantly higher ALP activity compared with static culture, indicating that fluid shear stress alone promotes osteogenic differentiation. Lactate dehydrogenase release decreased more rapidly in dynamic cultures, suggesting reduced cellular stress under perfusion. DNA content was higher in perfused cores with showed no significant differences. Morphometric analysis from μCT showed no significant differences, likely due to small sample size and inter-donor variability. Power analysis indicated that the current sample sizes were adequate for identifying large changes within groups but were insufficient to reliably detect small between-group differences. The modified ZetOS bioreactor proved biocompatible for long-term ex vivo culture of human trabecular bone tissue, preserving cell viability and enhancing osteogenic activity, while providing a robust platform to evaluate potential biomarkers and guide the development of bone-targeted therapies.

Graphical Abstract

Graphical Abstract.

Introduction

Bone is a metabolically dynamic tissue that continuously remodels its microstructure, adapting in response to mechanical and biochemical cues throughout life.¹ Bone remodeling is regulated by the interactions between osteocytes, osteoblasts, and osteoclasts.² Osteocytes, embedded within the mineralized matrix, are the primary mechanosensors in bone tissue that detect mechanical strains and biochemical signals within the lacunae-canalicular system and transduce this signal to bone-forming osteoblasts and bone-resorbing osteoclasts.^3,4 With aging and in diseased conditions, bones adaptive response is disrupted, and an imbalance between formation and resorption leads to bone fragility and increases the fracture risk.⁵ In OA, disrupted bone remodeling in the subchondral trabecular bone leads to a thinner and more porous microarchitecture, accompanied by the deterioration and extension of blood vessels and nerves into the cartilage.⁶ With OA progression, the breakdown of sclerotic subchondral trabecular bone results in bone cyst-like lesions, disrupting microstructure, and increasing the risk of fracture.⁷ Novel therapies for OA undergoing preclinical testing primarily depend on animal models.⁸ However, these models present ethical, financial, and translational limitations.⁹ Differences in bone microarchitecture, remodeling rates, and mechanical properties between animal and human bone can complicate the translation of experimental results to clinical populations.^8–10

Ex vivo human bone cultures have gained increasing attention, as they provide a biologically relevant platform that preserves the native three-dimensional (3D) extracellular matrix, heterogeneity of cell populations, and signaling environment that 2D in vitro cell cultures fail to capture.^9,11–16 However, human bone explants, especially under static culture, often struggle to maintain long-term viability, resulting from challenges in nutrient transport.¹⁷ Given these challenges and opportunities, ex vivo perfusion bioreactors, such as the ZetOS platform, offer a promising approach for maintaining long-term trabecular bone culture with continuous perfusion flow and mechanical stimulation to extend bone viability and promote osteogenesis ex vivo.^18–22

The ZetOS bioreactor was originally fabricated out of polycarbonate, featuring a chamber with a 5 mm height and 10 mm diameter of trabecular bone to ensure adequate perfusion through the bone construct.^18,23 Recent studies have focused on improving the system’s performance while maintaining its original design concept. Notably, one redesign made the ZetOS chamber 3D printable and increased the sample size to 10 mm height and 10 mm diameter to capture in situ microarchitecture of the trabecular bone and provide a more reliable measurement of apparent elastic modulus by reducing end-artifacts.^19,24,25 Although this study showed a significant increase in apparent elastic modulus over a 21 d compression-perfusion protocol in bovine samples, the flow rate for the new chamber design was not optimized.¹⁹ Other studies using ZetOS have also demonstrated increases in trabecular bone stiffness and osteoid deposition under dynamic loading,^19–22 yet the isolated effects of perfusion flow as a mechanical cue for osteocytes remain unexamined. Moreover, most existing explant culture studies rely on animal bones, which again limit the clinical translatability of results.^{10,15,26–28} Perfusion is particularly critical in large bone tissue cultures, as nutrients in the medium do not easily reach the inner marrow spaces due to diffusion limits and the hydrophobic nature of bone marrow.²⁹ To determine the optimal flow rate in a redesigned bioreactor, computational fluid dynamics (CFD) modeling allows for the determination of changes in flow profile introduced by the redesigned components.^30,31

Therefore, the objectives of this study were to evaluate how media flow rate within a modified ZetOS chamber influences long-term viability and osteogenesis of ex vivo human OA bone cultured for 21 d. As a preliminary investigation, this work focuses solely on fluid perfusion and does not incorporate mechanical loading, allowing the independent contribution of flow-mediated stimulation to be isolated and assessed. By experimentally applying 2 distinct perfusion rates, this study aims to determine: (1) whether controlled perfusion improves nutrient transport and supports sustained cellular viability within human trabecular bone and (2) whether increasing flow rate enhances osteogenic activity over extended culture periods. By completing CFD modeling, this study aims to quantify whether the redesigned bioreactor improves the uniformity of the flow profile and whether there is more consistent flow arising from a particular flow rate. These investigations establish foundational perfusion parameters for future studies integrating optimized fluid flow with mechanical loading in the ZetOS platform.

Materials and methods

Design, 3D printing, and preparation of bioreactors

The modified ZetOS bioreactor incorporated integrated luer fittings for the inlet and outlet to eliminate external connectors and reduce leakage (Figure 1). Additional internal inlet channels were added in the middle of the bone chamber, with the goal of improving fluid circulation. Bioreactors were printed with MED610 (Stratasys) and SUP706 (Stratasys), which served as support material using Stratasys J750 (Stratasys). The printing was performed using high-resolution settings, a glossy surface finish, and a 100% infill density. An optimized cleaning protocol was developed based on previous studies.^19,32,33 The cytotoxicity of the cleaned material was then assessed using a differentiated mouse osteoblastic cell line (d-MC3T3-OB)³⁴ cultured at 20 000 cells/mL in 24-well plates for 7 d [see Supplementary Material, Figures S1 and S2].

Figure 1.

(A) Bioreactor assembly. (B) Cross-section of the bioreactor assembly showing: (1) luer fittings integrated in the inlet, (2) bone chamber within the body component, (2) cap of the bioreactor, (3) cylinders positioned at the top and bottom, (4) luer fittings integrated in the outlet, and (5) X-ring/O-rings preventing fluid leakage from both ends. (C) Experimental perfusion setup with a peristaltic pump circulating culture media from a reservoir through the bioreactor at 37 °C 5% CO₂. The figure is made in biorender.com.

Visible exterior and interior support material (SUP706B) was removed manually using a pick tool. Parts were then thoroughly washed in a waterjet (Water Dental Flosser, Waterpik) operating at 2-3 bar to eliminate any remaining visible support. Parts were soaked in ultrapure (Milli-Q) water containing 1% sodium metasilicate (, 307815, Sigma-Aldrich) and 2% sodium hydroxide (, 221465, Sigma-Aldrich) for 24 h in a covered beaker placed in a fume hood. After soaking, parts were transferred to a 5% acetic acid solution for one minute to neutralize the caustic residue, followed by a rinse with Milli-Q water. The parts were then sonicated (B1500A-MT, VWR) in a covered glass beaker filled with isopropanol (IPA, 10× the volume of the parts) for 2 h at 40 kHz, with IPA replaced every hour. Following that, parts were sonicated again in Milli-Q water for 2 h at 40 kHz, with hourly solvent replacement. All parts were then rinsed with 80% ethanol, placed under UV light in a biosafety cabinet for 2 h and subsequently were sterilized by steam autoclaving at 132 °C for 4 min.

Osteoarthritic bone culture

Humeral head osteotomies were obtained from 4 patients with OA (age 65-77, female n = 1, male n = 3) who were undergoing shoulder joint replacement on a weekly basis at the Roth|McFarlane Hand and Upper Limb Centre, St. Josephs Health Care (HSREB# 113023). The humeral heads were transferred immediately after surgery into ice-cold culture medium composed of 85% Dulbecco’s Modified Eagle’s Medium/nutrient mixture F-12 Ham with 2 mM L-glutamine, 15 mM HEPES, and sodium bicarbonate (DMEM/F-12, D8437, Sigma-Aldrich), 10% v/v fetal bovine serum (FBS, F2442, Sigma-Aldrich), and 5% v/v penicillin/streptomycin (P/S, P4333, Sigma-Aldrich). Within 2 h after surgery, bone cores (9.15 ± 0.14 × 9.5 ± 2.0 mm, n = 22) were prepared using coring with a milling machine (CX600, Craftex) and custom-designed blade, and cutting with a rotary microtome (IsoMet Low Speed, Buehler), both operated at 50-150 rpm under continuous sterile phosphate-buffered saline (PBS, D8537, Sigma-Aldrich) irrigation to minimize thermal and mechanical damage (Figure 2). Following preparation, bone cores were cultured in 6-well plates under aseptic conditions in fresh medium containing DMEM/F-12, 10% FBS, and 3% P/S for 18 h of incubation, then transferred to new well plates containing complete culture medium (DMEM/F-12, 10% v/v FBS, and 1% v/v P/S with 10 μg/mL L-ascorbic acid 2-phosphate (AA, A4544, Sigma-Aldrich)) for 4 d of recovery.²¹ Following the recovery period, bone cores were randomly assigned to experimental groups and cultured in 6-well plates (static controls, n = 6) or in bioreactors preconditioned overnight with 8 mL of 50% DMEM-F12, 50% FBS, and 1% P/S, with perfusion applied via a peristaltic pump (BT100-1L-A, BMT Fluid Control Solutions) at 252 μL/min (dynamic I, n = 7) or 752 μL/min (dynamic II, n = 9). All samples were maintained for 21 d at 37 °C in 5% CO₂, with 8 mL of medium per bone core, which was changed every other day under sterile conditions inside a biosafety cabinet by manually exchanging the culture medium in the test tubes, stored at −80 °C, and pH monitored throughout the culture period at each media change with a pH/mV meter (Fisherbrand, Accumet AB315).

Figure 2.

Workflow for bone tissue culture and bioreactor setup. (A) Humeral head osteotomy was transported from the hospital to the laboratory in culture medium (85% DMEM-F12, 10% FBS, and 5% P/S) on ice. (B-D) Bone coring was performed under irrigation using a custom-made coring blade, and further cutting was carried out with a rotary microtome. (E) Bone cores were cultured in 24-well plates for 18 h in 2 mL of culture medium (85% DMEM-F12, 10% FBS, and 5% P/S), then transferred to 6-well plates containing 8 mL of complete culture medium (90% DMEM-F12, 10% FBS, 1% P/S, and 10 μg/mL AA) for 4 d of recovery. (F) Bone cores were cultured in the bioreactor system for 21 d at 37 °C and 5% CO₂.

CFD simulations and wall shear stress estimation

Bone chambers were meshed in ANSYS Meshing, using a patch conforming method with quadratic tetrahedral elements. The maximum edge length was 0.3 mm, corresponding to 95 568 elements for the original design, while a maximum edge length of 0.5 mm and 80 428 elements was applied for the modified design. A bone surrogate was modeled as a cylinder with a 10 mm diameter and 10 mm height, meshed with quadratic tetrahedral elements of 0.5 mm maximum edge length. The bone surrogate was defined as a porous material with permeability (κ) of 5.13e-9 m², and porosity () of 0.74.^35,36 The Darcy–Brinkman equation was used for the porous domain in simplified geometry as follows:

(1)

where ρ, u, I, p, μ, F, and denote fluid density, fluid velocity, identity tensor, pressure, dynamic viscosity, volume force, and mass source, respectively. A zero-mass source and zero-volume force were assumed due to low μ and Reynolds number (Re < 2300). Complete culture medium (DMEM + 10% FBS) was modeled as incompressible Newtonian fluid (ρ = 1009 kg/m³, μ = 9.3e-4 Pa.s at 37 °C). Two steady-state simulations were conducted per chamber, with laminar fluid flow at rates of 252 μL/min (Re = 3.6) and 756 μL/min (Re = 10.8). A mass-flow inlet was defined for each design, with zero pressure at the outlet and a no-slip condition at the walls (Ansys Fluent 2025). The effect of gravity was included, with a value of g = 9.81 m/s². To quantify average velocity and wall shear stress, 3 planes spaced 3 mm apart within the bone region were defined to analyze the solution of the following Navier–Stokes equation:

(2)

Wall shear stress () was calculated based on the methods of Gabetti et al., as follows³⁷:

(3)

where the μ and κ are the same as used within the simulation and v_ave represents the plane-specific average velocity obtained from the simulation.

Measurement of cell metabolic activity

Bone cell viability was assessed in all samples using the AlamarBlue assay (A13262, Thermo Fisher Scientific) by adding 10% v/v reagent to the culture medium on days 0, 6, 13, and 20. After 3.5 h of incubation at 37 °C and 5% CO₂, 100 μL of medium was transferred to a flat-bottom black 96-well plate, and fluorescence was measured using a plate reader (BioTek Cytation 5, Agilent/BioTek) at 530/25 nm excitation and 590/35 nm emission. Fluorescence values were corrected against medium-only controls (blank). Following each assay, samples were washed twice with sterile PBS before fresh complete medium was added.

Osteogenic activity

Alkaline phosphate (ALP) activity

Alkaline phosphate level in culture medium supernatants was evaluated on days 0, 6, 13, and 20 (MAK447, Sigma-Aldrich). Alkaline phosphate is a membrane-bound glycoprotein expressed by osteoblasts and is widely used as an early marker of osteogenic differentiation.³⁴ In this assay, ALP catalyzes the hydrolysis of p-nitrophenyl phosphate to p-nitrophenol in an alkaline environment, producing a yellow product. The rate of color development is proportional to the activity of ALP and therefore to the osteogenic potential of cells. For each sample, 20 μL of media was mixed with 180 μL of working reagent in a 96-well clear flat-bottom plate. Blanks (200 μL Milli-Q water) and calibrators (200 μL provided Tartrazine) were included in each plate. The optical density (OD) at 405 nm was measured immediately and after 4 min using a microplate reader (BioTek Cytation 5, Agilent/BioTek).

Osteoprotegerin (OPG) level in supernatant

Osteoprotegerin extracellular inhibitor of NF-κB ligand that is secreted by osteoblasts.³⁸ Osteoprotegerin levels in culture medium supernatants were measured on days 0, 6, 13, and 20 using the Human OPG ELISA Kit (BMS2021INST, Thermo Fisher Scientific). Absorbance was measured at 450 nm (BioTek Cytation 5, Agilent/BioTek), and sample concentrations were determined from the standard curve. The assay has a sensitivity of 2.5 pg/mL and reliably measures OPG concentrations in the range of 15.6-1000.0 pg/mL.

Cytotoxicity measurement

Cytotoxicity was assessed by measuring lactate dehydrogenase (LDH, MAK529, Sigma-Aldrich) release into the supernatant culture medium on days 0, 6, 13, and 20. Lactate dehydrogenase is a cytosolic enzyme released upon cell membrane damage, so its extracellular level serves as an indicator of cell membrane integrity and cytotoxicity.³⁹ For the assay, 10 μL of cell culture medium was added to 160 μL of assay reagent in each well. Blank media were included in each assay. After incubation at room temperature in the dark for 10 min, the OD was measured at 500 nm (BioTek Cytation 5, Agilent/BioTek). The average absorbance of blank controls was subtracted from all wells to calculate LDH activity as a measure of cytotoxicity.

Genomic content extraction and quantification

Genomic content, determined by quantifying DNA, was used to assess cell proliferation in the bone cores. Total DNA was extracted from all samples on day 20 of culture. Each sample was halved, and a quarter of one half was rinsed in PBS. After three freeze-thaw cycles at −80 °C and 20 min sonication at 40 kHz, 20 mg of minced tissue was transferred to a microcentrifuge tube containing 20 μL of Proteinase K (PureLink Genomic) and 180 μL digestion buffer (PureLink Genomic) and incubated at 56 °C for 8 h with gentle shaking until fully digested. The samples were then centrifuged at a maximum speed of 10 000 × g for 1 min, and the 100 μL supernatant media containing DNA was transferred to a new microcentrifuge tube. Then, 20 μL of RNase A (PureLink Genomic), 200 μL of Binding buffer (PureLink Genomic), and 200 μL of 100% ethanol were added to each microcentrifuge tube. DNA was then purified using a commercial DNA extraction kit (PureLink Genomic DNA, K182002) according to the manufacturer’s protocol and quantified using a NanoDrop 2000/2000c Spectrophotometer (Thermo Fisher Scientific) by measuring absorbance at 260/280 nm, with purity ratios ranging from 1.7 to 1.9.

Cellular viability assessment

On day 20, a second quarter of each sample was used for live/dead staining with Calcein AM and Deep Red (L32250, Thermo Fisher Scientific). Samples were incubated in staining solution for 30 min at 37 °C in 5% CO₂, followed by 3 PBS washes to remove background fluorescence. Stained samples were imaged using a confocal microscope (ECLIPSE Ti2, Nikon) at excitation and emission wavelengths of 468 and 568 nm, respectively.

Micro-CT (CT) imaging and morphometric analysis

Following the culture period, the remaining half of each bone core was scanned using high-resolution μCT (inCiTe, KA Imaging) at 90 kV and 100 μA in air and reconstructed at a voxel size of 5 μm. CT images were exported as 16-bit DICOMs and processed using Dragonfly (FreeD, v2025, http://www.dragonflyit.ca). Images were filtered using a 3D Gaussian kernel (σ = 1.25), and ring artifacts were corrected using the software’s built-in removal tool. Bone tissue was segmented using a global thresholding calibrated between 600 and 1200 ,⁴⁰ with a hydroxyapatite (HA) density phantom with 5 rods ranging from 0 to 1200 (Micro-CT HA D32, QRM). The segmented masks were imported to Fiji⁴¹ (v1.54r), and trabecular morphometry was analyzed with the BoneJ plugin.⁴² Bone volume fraction (BV/TV) was calculated as the ratio of bone volume to total sample volume,⁴³ trabecular thickness (Tb.Th), and spacing (Tb.Sp) were determined using a sphere-fitting algorithm,^44,45 and degree of anisotropy (DA) was assessed via mean intercept length to describe trabecular orientation, with 0 indicating isotropic and 1 indicating highly anisotropic trabecular orientation.⁴⁶

Static histomorphometry

Following μCT scanning, the remaining half of each bone explant was fixed in 70% ethanol at +4 °C for 3 mo. Sections (200 μm) were prepared using a low-speed diamond saw (IsoMet Low Speed, Buehler), rehydrated through a graded ethanol series and soaked in PBS for 2 h to remove bone marrow prior to staining. To visualize mineralized bone matrix, sections were stained with Alizarin Red S (A5533, Sigma). A 2% (w/v) stock solution was prepared in distilled water and adjusted to pH 4.1-4.3 with 10% acetic acid. Sections were incubated in a 1% working solution for 20 min at room temperature, followed by rinsing in PBS. Stained sections were mounted in the custom-made mounting glass and imaged using a confocal microscope (ECLIPSE Ti2, Nikon) at excitation 550 nm and emission 600 nm.

Statistical analysis

All data were summarized as mean ± SD. All biological experiments were performed in duplicate, and statistical analyses were conducted in Python (v3.11). Data normality and homogeneity of variance were assessed using the Shapiro–Wilk and Levene tests, respectively. Between-group comparisons were performed using one-way ANOVA for normally distributed data with Tukey’s post-hoc test or the Kruskal–Wallis test otherwise, followed by Dunn’s multiple comparisons. Within-group changes were assessed using repeated-measures ANOVA or the Friedman test followed by post hoc paired t-tests or Wilcoxon signed-rank tests, respectively. Post hoc power analyses were performed to evaluate the sensitivity of the study to detect changes both within and among groups. Cohen’s f was used to estimate effect sizes for between-group comparisons (ANOVA or Kruskal–Wallis), whereas Cohen’s d was calculated for within-group paired comparisons over time. These effect sizes were then used to estimate the statistical power of each analysis, indicating whether the current sample sizes were sufficient to detect small, moderate, or large effects. Statistical significance was defined as α = .05.

Results

The new design enhanced media flow within the trabecular bone

Computational fluid dynamics simulations were performed to assess flow within the bone models for both the modified and original designs, where the primary metrics of interest were average velocity and wall shear stress. The velocity contours for both designs and flow rates at the 3 defined planes are visible in Figure 3.

Figure 3.

(A) Flow geometry of the original bioreactor design, where outlet flow circulates at the top prior to exit (excluded in CFD modeling). (B) Velocity contours at 252 μL/min (dynamic I) for the original design. (C) Velocity contours at 756 μL/min (dynamic II) for the original design. (D) Flow geometry of the modified bioreactor highlighted in red. (E) Velocity contours at 252 μL/min (dynamic I) for the modified design. (F) Velocity contours at 756 μL/min (dynamic II) for the modified design.

At a flow rate of 252 μL/min (dynamic I), the modified design exhibited consistently overall higher average velocities throughout the bone region compared to the original design. In the plane closest to the inlet (Plane 1), the average velocity was 2.86 × 10⁻⁵ m/s and 2.83 × 10⁻⁵ m/s for the modified and original design, respectively. Average velocities at the mid-plane (Plane 2) were 2.07 × 10⁻⁵ m/s for the modified design vs 1.40 × 10⁻⁵ m/s for the original design, corresponding to a 32.4% difference between the designs. Percent differences are only reported in Plane 2, as it represents the largest cross-sectional area of the bone region. At the plane closest to the outlet (Plane 3), the average velocity for the modified design was 2.41 × 10⁻⁵ m/s, while the original design was 1.88 × 10⁻⁵ m/s. Wall shear stress followed a similar trend, exhibiting a moderate increase in the modified design compared to the original design. The wall shear stress observed in Plane 1 was 0.47 mPa in both design configurations. In Plane 2, the modified design showed a moderate increase in wall shear stress of 0.34 mPa in the modified design relative to 0.23 mPa in the original design. This behavior persisted in Plane 3, where the wall shear stress was 0.44 mPa vs 0.31 mPa in the modified and original designs, respectively.

At the higher flow rate of 756 μL/min (dynamic II), velocities and wall shear stress increased proportionally in both designs. In Plane 1, the average velocity for the modified design was 7.93 × 10⁻⁵ m/s while it was 1.10 × 10⁻⁴ m/s in the original design. The behavior of average velocity changed in Plane 2, where it was 6.12 × 10⁻⁵ m/s for the modified design compared to 4.21 × 10⁻⁵ m/s in the original design, corresponding to a 31% increase in the modified design. Plane 3 closely followed the behavior of Plane 1, with the modified design average velocity being 6.08 × 10⁻⁵ m/s, while the original design had a higher value of 8.21 × 10⁻⁵ m/s. Wall shear stress is dependent on average velocity; therefore, the same behavior was observed, where values were higher in the original designs for Planes 1 and 3. Wall shear stress in Plane 1 was 1.31 mPa in the modified design and 1.81 mPa in the original design. In Plane 2, the wall shear stress was 1.01 mPa for the modified design in comparison to 0.70 mPa in the original design. The modified design had a wall shear stress of 1.00 mPa in Plane 3, while the original design was 1.36 mPa.

Culture medium pH measurements

The pH of the medium (DMEM/F-12, 10% v/v FBS, 1% v/v P/S, and 10 μg/mL AA) was measured at 7.32 ± 0.02 throughout the culture period. During week 1, the pH of all groups ranged from 7.48 ± 0.39 for static culture, 7.42 ± 0.30 for dynamic I, and 7.36 ± 0.05 for dynamic II. By week 2, a slight decrease was observed in all groups, with pH values of 7.24 ± 0.12 for static culture, 7.19 ± 0.15 for dynamic I, and 7.21 ± 0.08 for dynamic II, respectively. In week 3, the pH of dynamic II was relatively low at 7.11 ± 0.03, while the static and dynamic I groups were slightly higher at 7.30 ± 0.16 and 7.22 ± 0.10, respectively.

Metabolic activity increased in all cultures

Metabolic activity increased in all cultures over time, as shown by the Friedman test (p < .001 for all groups) followed by Wilcoxon signed-rank tests, with large effect sizes observed (Cohen’s d = 2.1 for static, 1.9 for dynamic I, and 4.18 for dynamic II). Metabolic activity increased significantly over time in all groups (Figure 4A). In the static culture, metabolic activity rose from day 0 to day 6 rises from day 0 to day 6 (p = .005), day 6 to day 13 (p = .039), and day 13 to day 20 (p = .025). Dynamic I showed a modest, non-significant increase from day 0 to day 6 (p = .244), followed by a significant rise from day 6 to day 13 (p = .016), while changes from day 13 to day 20 were not significant (p = .469). Dynamic II exhibited a sharp rise by day 6 (p = .001), with no significant change from day 6 to day 13 (p = .742) and a moderate increase from day 13 to day 20 (p = .102). Between-group comparisons using the Kruskal–Wallis test were significant at the early time point (day 6: p = .037) but not at later days (day 13: p = .864; day 20: p = .465).

Figure 4.

(A) Metabolic activity increased significantly over time in all groups, with dynamic II showing the earliest and largest rise by day 6 (p = .0009). (B) Alkaline phosphatase (ALP) activity in static culture was significantly lower than dynamic I and dynamic II at days 6 and 20. (C) Normalized ALP activity to day 0, were also significantly higher in dynamic I and dynamic II groups at days 6 and 20 compared to the static group. (D) Osteoprotegerin (OPG) concentrations showed relatively stable levels in the static group, moderate increases in the dynamic I group, and the highest concentrations in the dynamic II group, with a significant difference observed between groups at day 13 (Kruskal–Wallis, p = .043), between the static and dynamic II groups (Mann–Whitney, p = .016). (E) Lactate dehydrogenase activity dropped significantly under dynamic cultures by day 6, whereas in static culture, no significant change was observed until day 13. (F) Total dsDNA content was higher in dynamic II vs static culture (p = .059), while no significant difference was found among groups at a significance level of .05. All data were tested for normality using the Shapiro–Wilk test; within-group changes over time were analyzed using Friedman test followed by the Wilcoxon signed-rank post-hoc, and between-group comparisons were analyzed using Kruskal–Wallis test. The box represents the IQR (25th-75th percentiles), with the central line indicating the median. Whiskers extend to the most extreme data points within 1.5× IQR of the first and third quartiles; ^*p < .05, ^**p < .01, ^***p < .001.

Osteogenic activity was higher in the dynamic cultures

For the ALP measurements, the Friedman test showed significant temporal changes only for static group (p = .013), whereas no significant time-dependent effects were observed in dynamic I (p = .653) and dynamic II (p = .789). At day 0, ALP levels did not differ significantly among groups (Kruskal–Wallis, p = .690) (Figure 4B and C). By day 6, static cultures showed lower ALP activity than both dynamic I (p = .033) and dynamic II (p = .0008) cultures. No significant differences were observed at day 13 (p = .106), whereas at day 20, static cultures remained lower than dynamic I (p = .043) and dynamic II (p = .015). Within-group analysis revealed a significant decrease in ALP levels in static cultures from day 0 to day 6 (p = .008) and again by day 20 (p = .025). In contrast, ALP levels in dynamic I and dynamic II did not decrease significantly. Post hoc power analyses suggested that the study had limited sensitivity to detect between-group differences at day 0 and day 13 (power ≤ 0.1), in agreement with the non-significant results. In contrast, differences observed at day 6 and day 20 were sufficiently detectable (power ≥ 0.9) for large effects. Normalized ALP levels to day 0 were also significantly higher in dynamic I and dynamic II groups at days 6 and 20 compared to the static group.

Osteoprotegerin levels varied across the culture period in all experimental groups. In the static group, concentrations showed minor fluctuations, ranging from 27.76 to 56.99 pg/mL at day 0 and 24.38 to 66.19 pg/mL at day 20, remaining within a relatively narrow range throughout the experiment (Figure 4C). In the dynamic I group, OPG levels exhibited moderate increases, with concentrations ranging from 39.01 to 63.75 pg/mL at day 0 and 32.82 to 102.81 pg/mL at day 20, with some samples exceeding 100 pg/mL during the culture period. The dynamic II group showed the highest OPG concentrations and the greatest variability, ranging from 29.45 to 103.08 pg/mL at day 0 and 31.13 to 140.47 pg/mL at day 20, with several samples exceeding 130 pg/mL. Statistical analysis using the Kruskal–Wallis test indicated no significant differences between groups at day 0 (p = .679) or day 6 (p = .083). A significant difference between groups was observed at day 13 (p = .043). Mann–Whitney post-hoc analysis revealed a significant difference between the static and dynamic II groups (p = .016), while differences between static and dynamic I (p = .198) and dynamic I and dynamic II (p = .289) were not significant. By day 20, no significant differences between groups were detected (p = .116).

Cytotoxicity decreased earliest in the perfused conditions

For cytotoxicity measurements, Wilcoxon signed-rank tests against the day 0 were conducted to evaluate cellular response relative to the initial baseline. Lactate dehydrogenase release in supernatant culture media decreased significantly over time in all groups, with the earliest reduction observed in the bioreactor-perfused cultures (p < .05, Figure 4E). In the static group, LDH levels remained comparable between days 0 and 6 (p = .542) but decreased significantly by days 13 (p = .001) and 20 (p = .002). Dynamic I showed earlier reductions, with significant decreases at day 6 (p = .035), day 13 (p < .001), and day 20 (p < .001). Dynamic II consistently exhibited the lowest LDH levels, with significant reductions at day 6 (p = .039), day 13 (p < .001), and day 20 (p = .004).

Cell viability and proliferation

DNA content showed a trend toward differences among groups but did not reach statistical significance (Figure 4E). A Kruskal–Wallis test indicated a near-significant group effect (p = .064), with an approximate power of 0.43. Post hoc Dunn tests with Bonferroni correction did not identify significant pairwise differences, although a trend was observed between the static and dynamic II groups (p = .059). Representative confocal microscopy images of bone cores stained for live (Calcein AM, green) and dead (Deep Red, red) cells are shown in Figure 5A-F. Tissue viability was maintained after 21 d, with high cell survival along surfaces exposed to coring and cutting on day −4.

Figure 5.

Representative live/dead confocal images (1) showed cell viability throughout trabecular cores under (A) static, (B) dynamic I, and (C) dynamic II cultures, showing that bone cutting and coring did not compromise cell survival at the bone surfaces on day −4. Cell morphology is visible in the images (10), showing attached, elongated cells: (D) static, (E) dynamic I, and (F) dynamic II cultures.

Morphological measurements

Morphometric parameters of trabecular bone cores are summarized in Table 1. Shapiro–Wilk normality tests indicated that the data were normally distributed across groups (static, dynamic I, and dynamic II). Levene’s test confirmed equal variances among groups. BV/TV ranged from 0.152 ± 0.061 to 0.206 ± 0.141, with the highest mean value observed in the static group. Tb.Sp values were comparable across groups (p = .804), ranging from 0.745 ± 0.138 in static to 0.854 ± 0.121 in dynamic I. Degree of anisotropy ranged from 0.408 ± 0.054 in dynamic I to 0.460 ± 0.103 in dynamic II and 0.453 ± 0.080 in static, with no statistically significant differences observed among groups (p = .998). Substantial variability was observed at the individual donor level; statistical analyses for individual donors were not conducted due to small sample sizes. Post hoc parametric power analysis for the ANOVA indicated limited sensitivity to detect between-group differences with estimated power of 0.18 for BV/TV, 0.23 for Tb.Th, 0.34 for Tb.Sp, and 0.06 for DA.

Table 1.

Three-dimensional morphometric parameters of trabecular cores.

Groups/donor	BV/TV	Tb.Th (mm)	Tb.Sp (mm)	DA
Static (n = 6)	0.206 ± 0.141	0.191 ± 0.087	0.745 ± 0.138	0.453 ± 0.080
Dynamic I (n = 7)	0.152 ± 0.061	0.173 ± 0.044	0.854 ± 0.121	0.449 ± 0.059
Dynamic II (n = 9)	0.160 ± 0.066	0.153 ± 0.026	0.805 ± 0.095	0.460 ± 0.103
Female_1 (age: 74, n = 2)	0.306 ± 0.222	0.277 ± 0.105	0.751 ± 0.319	0.436 ± 0.005
Male_1 (age: 65, n = 6)	0.165 ± 0.065	0.169 ± 0.028	0.833 ± 0.142	0.546 ± 0.408
Male_2 (age: 77, n = 5)	0.139 ± 0.062	0.158 ± 0.057	0.790 ± 0.052	0.408 ± 0.054
Male_3 (age: 65, n = 9)	0.160 ± 0.066	0.153 ± 0.026	0.805 ± 0.095	0.460 ± 0.103

Group sample sizes (n) are indicated. Individual donor data (age in years) are shown.

Abbreviations: BV/TV, bone volume fraction; DA, degree of anisotropy; Tb.Sp, trabecular separation; Tb.Th, trabecular thickness.

Mineralized bone matrix

Representative confocal microscopy images of Alizarin Red S-stained bone explant sections are shown in Figure 6. Mineralized bone regions were visualized in both static, dynamic I and dynamic II cultures, showing areas of newly formed mineralized bone.

Figure 6.

Representative confocal microscopy images of Alizarin Red S-stained bone explant sections under static, dynamic I, and dynamic II cultures, showing areas of newly formed mineralized matrix (10). Mineral deposition is observed in all groups, indicating bone formation under both static and dynamic conditions; however, a greater extent of mineralized regions was observed in the dynamic groups.

Discussion

The modified ZetOS bioreactor developed in this study enabled long-term culture of ex vivo human osteoarthritic humeral head, where viability was assessed and maintained. The ZetOS bioreactor has been widely used to investigate the effects of mechanical loading on trabecular bone viability and adaptation.^18–22 Early work by Endres et al. demonstrated that bovine trabecular bone cores could be maintained viable ex vivo in ZetOS under controlled loading signals.¹⁸ Similarly, Vivanco et al. showed that dynamic loading increased the apparent elastic modulus of ex vivo bovine trabecular bone using ZetOS.²² Meyer et al. further reported that bovine sternal cores in ZetOS exposed to dynamic loading exhibited enhanced osteogenesis, resulting in a higher apparent elastic modulus.²⁰ More recently, a study using the ZetOS bioreactor on ex vivo human trabecular bone showed bone adaptation under mechanical loading.²¹ Kunath et al. redesigned the ZetOS chamber to be 3D printable and accommodate bone samples with double the height of those used in previous works, minimizing end-artifact effects during elastic modulus measurements.¹⁹ However, perfusion flow in the new ZetOS systems was not fully optimized for the larger trabecular bone cores, and the specific contribution of perfusion alone in maintaining viability remains unclear. To address this gap, the present study experimentally employed 2 distinct perfusion rates to quantify the effects of media flow rate on trabecular bone cell viability, osteogenic activity, and proliferation over a 21-d culture period. The present study also completed concurrent CFD modeling to observe the effects of the redesigned system on the flow profile under the same two flow rates.

Computational fluid dynamics has not been completed in ZetOS systems; however, the effect of flow rate has been previously explored in perfusion bioreactors. Bancroft et al. showed that a flow rate of 3000 μL/min through 3D tissue-engineered scaffolds increased mineralized matrix deposition by inducing fluid shear stress.⁴⁷ Computational modeling of media flow in a perfusion bioreactor indicated that 0.05 mPa shear stress promotes cell proliferation, whereas higher stress levels upregulate osteogenic gene expression.⁴⁸ Davidson et al. further demonstrated the value of perfusion-based culture systems by maintaining rat femoral bone ex vivo for 14 d under static and dynamic conditions using a perfusion rate of 1000 μL/min.⁴⁹ In their study, dynamic perfusion resulted in significantly higher metabolic activity, osteogenic activity, and DNA content compared to static culture.⁴⁹ Yamada et al. completed CFD simulations using a perfusion rate of 800 μL/min in a porous scaffold, where an average velocity of 1.5 × 10⁻⁴ m/s was observed, along with a wall shear stress of 0.7 mPa.⁵⁰ In another study, CFD modeling demonstrated similar values for a porous scaffold at 1500 μL/min, where average velocity was 4.0 × 10⁻⁴ m/s, and the wall shear stress was 1 mPa.⁵¹ These values are generally aligned with what was observed in the redesigned system of the current investigation. Specific differences in velocity and wall shear stress can be attributed to variations in the design of each system. Further, the flow was applied along the longitudinal axis of the scaffold in previous CFD investigations, which is important to consider when comparing velocities, as there is a uniform cross-sectional area in this configuration. Due to the desire to include compressive loading in ZetOS systems, perfusion is applied normal to the longitudinal axis, causing decreased uniformity in velocity throughout the bone region.

The CFD simulations demonstrated that the modified design exhibited more consistent flow throughout the entire bone region in the simulation. Simulated flow rates were within the reported optimal range for osteoblastic differentiation, between 200 and 1000 μL/min.^17,52,53 Computational fluid dynamics is a valuable tool, as it allows precise measurement of fluid velocity at multiple locations in the bone region. The modified design introduced additional inlet flow channels at the middle of the bone core (Plane 2), a region of particular interest. This region exhibits the largest cross-sectional area due to the normal orientation of the longitudinal axis of the core relative to the inlet, making it challenging to maintain uniform velocity. Under both simulated flow rates, the modified design demonstrated higher Plane 2 velocities, suggesting improved fluid distribution resulting from the updated geometry. The results of the CFD of the modified design did not demonstrate notable differences in fluid behavior dependent on inlet flow rate, except for an expected proportional increase in velocity and therefore wall shear stress. When considering osteogenic and metabolic impacts observed experimentally, there were no significant differences between these metrics based on flow rate for the dynamic I and dynamic II groups. Therefore, in combination, the simulation and experimental results suggest that the modified design provides sufficient perfusion at either flow rate tested.

In terms of cell viability, the bioreactor system did not negatively affect the cells compared with static controls, confirming that the setup is fully biocompatible with human osteoarthritic bone core culture. The early rise in metabolic activity observed in dynamic II likely reflects enhanced fluid flow and improved nutrient transport, which then plateaued from day 6 to days 13 and 21, suggesting that the cells may have reached a maximal density and metabolic capacity under higher perfusion rate.⁵³ Alkaline phosphate levels were significantly higher in the dynamic cultures, suggesting that flow-induced shear stress alone contributes substantially to osteogenic differentiation.⁵⁴ Although ALP activity decreased over time in all groups, this reduction was statistically significant only in static cultures at all timepoints (p < .01, power = 0.92). The elevated ALP activity observed at day 0 likely reflects an early cellular response to mechanical stress from coring and cutting of the bone samples.⁵⁵ Nevertheless, the high proportion of viable cells along the coring and cutting surfaces at day −4 suggests that initial mechanical damage was minimal and that cells recovered during the culture period (Figure 5A-F). While the observed increases in ALP levels in the supernatant suggest enhanced osteogenic potential under dynamic culture, they do not provide definitive evidence of complete osteogenic differentiation. Osteoprotegerin levels showed minor fluctuations across all groups throughout the culture period; however, OPG was significantly higher in dynamic II compared with static culture on day 13. The relatively stable OPG levels observed in the dynamic cultures suggest sustained osteoblast activity and continued antiresorptive signaling in response to flow-induced shear stress. The supernatant media pH was slightly higher during week 1 in all groups, which may have supported ALP activity consistent with the known role of alkaline pH in promoting osteogenic function.^56–58 A decline in pH was observed across all groups by weeks 2 and 3, likely reflecting increased cellular metabolic activity and accumulation of acidic by-products, which may promote localized extracellular acidification known to strongly stimulate osteoclast activity and modulate bone remodeling.⁵⁹ Despite this decrease, pH values remained within a range compatible with cell viability and osteogenic function (7-7.4).^59,60 Histological analysis confirmed mineralized tissue formation (Figure 6B and C). Although qualitative assessment did not reveal significant differences in the total mineralized area between groups, high-intensity Alizarin Red staining was observed on the trabecular surfaces that indicates regions of dense calcium deposition, likely representing active mineralization fronts, a hallmark of osteoblast activity. Lactate dehydrogenase activity decreased significantly by day 6 in the dynamic cultures, whereas this early reduction was not observed in the static condition, which suggests prolonged cellular stress in the absence of perfusion.^16,39 Although not statistically significant, dynamic cultures showed higher total DNA content than the static culture, suggesting enhanced cell proliferation under perfusion conditions, consistent with observations reported in previous studies.^15,61

No statistically significant differences were observed among groups in terms of morphometric parameters, which may be attributed to the small sample size and high inter-donor variability. Trabecular morphometric parameters are dependent on OA disease stage; early OA is often marked by thinner, less connected trabeculae,⁶² while later stages involve irregular bone formation, thickened and sclerotic trabeculae, and the development of cyst-like lesions.^63,64 However, the stage of OA in the tissues used in this study was unknown. The samples from the female donor showed higher BV/TV and Tb.Th compared to other donors, likely reflecting advanced OA,^65–67 as evidenced by the small, irregular shape of the humeral head osteotomy, which limited the number of cores that could be obtained. Therefore, the large standard deviations observed in the measured microstructural parameters may be due to intrinsic differences in the trabecular network at different locations within the humeral head, as well as the disease state of the donors. Furthermore, the cores were not systematically cut at the exact same location within the humeral head, which we acknowledge as a limitation. Future studies could further reduce variability by standardizing core orientation and location.

Independent bone cores from 4 donors were collected weekly to ensure biological replication. Each measurement was performed in duplicate or triplicate, and post hoc power analyses indicated that within-group changes were detectable with sufficient sensitivity for large effects (power ≥ 0.8), whereas between-group differences were underpowered. Based on observed variability, approximately 25-30 samples per group would be required to reliably detect statistically significant differences between groups. Paired t-tests or Wilcoxon signed-rank tests were used to evaluate within-group changes over time rather than repeated-measures ANOVA or the Friedman test. This approach was chosen because our primary interest was in specific pairwise comparisons between time points within each group. Paired tests directly account for repeated measurements on the same samples, thereby providing greater statistical power to detect within-group changes, particularly when sample sizes are limited. However, multiple pairwise comparisons raise the possibility of Type I errors (false positives).

Cramer et al. demonstrated that longitudinal μCT can be used to visualize bone formation and resorption in porcine osteochondral explants cultured ex vivo for 28 d.¹⁶ The current study examined ex vivo human osteoarthritic trabecular bone, in which bone remodeling is known to be impaired.⁶ Therefore, μCT-detectable mineralization was not anticipated during the 21-d culture period, and longitudinal μCT imaging was therefore not performed. Nevertheless, future studies integrating longitudinal μCT imaging will be important to capture microstructural changes in long-term ex vivo bone cultures. Complementary mechanical testing would also be valuable to assess tissue-level behavior. In the current study, dynamic histomorphometry,²⁸ using fluorochrome labels (Calcein blue and Alizarin red S) was not performed due to the potential interference with the live/dead fluorescence staining used to assess cell viability within the explants. Instead, endpoint histology was performed on ethanol-fixed bone sections to visualize mineralized bone regions. These static histological analyses were not used to quantify mineralization rate or bone formation rate but provided a qualitative assessment of newly formed bone regions within the explants. A limitation of this study is that the bone was modeled as a homogeneous porous material, which does not account for its complex microarchitecture or spatial variations in permeability. As a result, the predicted wall shear stress and velocity distributions may differ from the true local mechanical environment experienced by individual cells within the trabecular network. Future studies could incorporate high-resolution microstructural data into CFD simulations to more accurately predict local fluid flow patterns. Additionally, integrating cyclic mechanical loading in combination with perfusion, along with longitudinal imaging, would provide insights into dynamic bone adaptation and remodeling over time. Repeated live/dead staining throughout the culture period in all groups would also allow direct tracking of cell viability changes over the culture period. Despite these limitations, this study highlights the potential of the modified ZetOS bioreactor as a platform for long-term ex vivo human bone culture. By enabling the investigation of osteoarthritic trabecular bone mechanobiology under perfusion and controlled osteogenic conditions, this system can help to evaluate potential biomarkers and inform strategies for bone-targeted therapies.

Author contributions

Mahsa Zojaji (Conceptualization, Methodology, Investigation, Formal analysis, Writing—original draft), Annelise Grace Pageau (Software, Validation, Formal analysis), Teodora Maluckov (Investigation, Validation, Resources), Johannes Eichwalder (Investigation, Validation, Resources), Mariya Peskova (Conceptualization, Methodology, Validation), Corey Smith (Resources, Project administration), George S. Athwal (Resources, Project administration), Maud Gorbet (Conceptualization, Methodology, Validation), Louis Miguel Ferreira (Resources, Project administration), and Nikolas Knowles (Conceptualization, Supervision, Project administration, Writing—review & editing)

Funding

Funding for this work was provided in part by the Arthritis Society Canada (#23-0079) and the Natural Sciences and Engineering Research Council of Canada (NSERC) (#RGPIN-2023-003401).

Conflicts of interest

None declared.

Data availability

The data underlying this article will be shared on reasonable request to the corresponding author.

Ethics statement

Ethics approval was received for use of human tissue (HSREB #113023).