Review

Engineering next-generation oxygen-generating scaffolds to enhance bone regeneration

Hypoxia promotes angiogenesis through multifaceted signaling involving a combination of angiogenic factors and inflammatory cytokines. Although acute hypoxia appears to be essential for stimulating wound healing, prolonged or chronic hypoxia lasting for >7 days has been shown to delay the tissue repair process [16]. Despite vascular infiltration, low local O₂ tension persists longer with increasing defect size, extending from 2 weeks [17,18] to 10 weeks post-injury [19]. Since angiogenesis is crucial for bone healing and regeneration, a major concern with O₂-releasing scaffolds was that they might hinder vascular infiltration and subsequent tissue regeneration. However, O₂-generating calcium peroxide (CaO₂)/gelatin microspheres developed for treating osteonecrosis demonstrated enhanced angiogenesis, and CD31⁺ vessels were prevalent in the defect area 4 weeks post-surgery [20]. Similarly, injectable sodium alginate/carboxymethyl chitosan (CMC)/CaO₂ hydrogel led to ectopic osteogenesis in rat subcutaneous defects [21], and significantly greater blood vessel volumes were observed in scaffolds with 25% CaO₂ compared with control poly(ε-caprolactone) (PCL) scaffolds [22]. Although not statistically significant, a higher density of 'type H' vessels was observed in CaO_2–PCL scaffold groups, suggesting that these proregenerative vessel subtypes may be characteristic of the oxygenated microenvironment [22].

Hypoxia elevates hypoxia-inducible factor 1α (HIF-1α) expression and induces the recruitment of neutrophils, macrophages, endothelial cells, and subsequently fibroblasts to the wound site [23]. In hypoxic environments, macrophages switch their metabolism from oxidative phosphorylation towards glycolysis, and this biases them toward the M1 phenotype [24,25]. Although the prolonged presence of M1 macrophages causes inflammation and hinders healing, they are crucial for the initiation of bone healing through clearing cell debris and pathogens, and the recruitment of more immune cells [26]. Thus, ideally, acute hypoxia for up to 48 h might benefit bone healing. Subsequent angiogenesis alleviates hypoxia and facilitates the transition of macrophages from M1 to M2, thus promoting tissue repair [27,28]. M2 macrophages continue to be active during the ossification phase and support bone remodeling (Figure 1A) [29]. Oxygenated biomaterials with a transient O2 release profile provide an opportunity to alter immune cell behavior from proinflammatory to proregenerative within a few days post-implantation. Efforts have been directed towards investigating the immunomodulatory potential of ROS-scavenging bone scaffolds which can influence M1 to M2 macrophage polarization [27,30]. O₂ release from OGS could facilitate such a transition and stabilize bone repair through M2 polarization.

In summary, O₂ generation and/or release from an implant should maintain O₂ tension between 6.6 and 8.6%, a level that is crucial for bone formation and homeostasis, until blood vessel infiltration is stabilized – a process that typically takes at least 2 weeks (Figure 1B) [17]. An acute moderately hypoxic environment (1–5% O₂ for 24–48 h) immediately post-implantation could enhance immune response and angiogenesis [26,31,32], although this is highly dependent on the cell type and might require further consideration regarding the specific experimental conditions. As the BTE construct size increases to a clinically relevant scale, O₂ deficiency would progressively become a problem, necessitating adjustments in the O₂ delivery profile to ensure immediate O₂ availability. Consequently, the release rate of oxygenated scaffolds designed for bone repair should match the physiological O₂ demand within the defect site.

OCS do not generate O₂ on their own but serve as vehicles for transporting and delivering O₂ at a desired dose to target sites. Examples include hemoglobin-based oxygen carriers (HBOCs), lipid-based O₂ microbubbles, O₂-laden nanosponges, polymeric microtanks, and perfluorocarbons (PFC). O₂ release by these systems is rate-limited by the diffusion of O₂ through the polymeric or lipid membrane. For example, O₂ was hyperbarically loaded into hollow biodegradable microtanks and incorporated into PCL for subsequent 3D printing into scaffolds. The scaffolds released O₂ for up to 8 h – a relatively short period – but enhanced the deposition of extracellular matrix (ECM) in murine calvarial and subcutaneous defects [55], suggesting that even short O₂ release at the earliest stages of healing can be beneficial. Owing to their excellent O₂ solubility, PFC-based emulsion-loaded hollow microparticles enabled controlled release of O₂ for 10 days [56]. They maintained the survival and osteogenic differentiation potential of human periosteal-derived cells under hypoxic conditions and provided a conducive environment for enhanced bone regeneration in miniature pig mandibular defects. However, for PFCs, long-term stability remains a significant concern, as the emulsions often decay during the freeze–thaw storage process or destabilize due to Ostwald ripening [57]. Oxygent, an FDA-approved PFC product, has a half-life of only 12–48 h, which limits its use primarily to temporary blood substitution during surgery [58]. These issues pose significant challenges for the clinical translation of OCS in BTE applications.

Compared with OCS, the maximum O₂ load in OGS may be two to four orders of magnitude higher, allowing O₂ release lasting from several days to a month. This longer release period potentially provides an advantage in the context of treating critical-sized bone defects. Examples include solid peroxides (sodium percarbonates, manganese oxide, CaO₂, etc.), liquid peroxides (hydrogen peroxide, H₂O₂), and photosynthetic algae [59], of which the solid peroxides are the most promising for BTE applications Sustained O₂ release from solid peroxides is achieved by their chemical reaction with the aqueous environment. O₂-generating particles are usually encapsulated within micro/nanocarriers such as liposomes, dendrimers, exosomes, nanospheres, nanocapsules, solid–lipid nanoparticles, nanofibers, or polymeric micelles [60], and O₂ release from these systems depends on water penetration and polymeric matrix biodegradation.

Among solid peroxides, CaO₂ is the most prevalently researched in BTE and has been employed in different formulations such as coatings [4,61], composite microspheres and filaments [62., 63., 64.], electrospun nanofibers [60,65,66], and hydrogel systems [3,20,21,67,68]. Used in hydrogel systems [3,67], near-complete bone regeneration in critical-sized calvaria defects was observed after 12 weeks [3], but hydrogels are not suitable for load-bearing applications. Alternatively, CaO₂ has been coated onto robocasted biphasic calcium phosphate (BCP) scaffolds [61]. When implanted into a 15 mm segmental radial defect in rabbits, a nearly twofold increase in bone formation was observed 6 months after surgery compared with uncoated scaffolds [4]. In polymer–CaO₂ composites, the polymer matrix serves as a hydrophobic barrier that slows O₂ release by limiting CaO₂–water interactions. However, in electrospinning systems [60,65,66], sustained O₂ release does not exceed 2 weeks, probably due to their high surface area. Nevertheless, CaO₂-based OGS have shown efficacy in promoting bone growth. PCL–CaO₂ microparticles developed by Suvarnapathaki and coworkers demonstrated high viability of preosteoblasts and reduced apoptosis, as confirmed by minimal in vitro caspase 3/7 activity. In vivo results from the same study also revealed that their scaffolds accelerated new bone formation, with less fibrous tissue and better tissue integration over 12 weeks, suggesting long-term safety and efficacy [3]. Gelatin–CaO₂ microparticles gave similar in vivo results over 10 weeks, and showed low immunogenicity and a significant reduction in inflammatory cells from the first to the third week post-implantation [69]. These studies collectively highlight the promise of CaO₂-based scaffolds in promoting bone regeneration for future clinical applications, while maintaining biocompatibility and reducing adverse immune responses.

Magnesium peroxide (MgO₂), another solid peroxide, exhibits the slowest and most persistent release kinetics of O₂ owing to its low decomposition rate [70]. A 3D-printed scaffold composed of PCL, β-tricalcium phosphate (β-TCP), and MgO₂ created by Peng and colleagues [70] released O₂ over 21 days and facilitated the formation of new bone in a femoral condyle defect model. Mg²⁺ ions also directly aid bone formation and impede the growth of various tumors through oxidative damage to DNA, making it particularly effective for repairing bone defects after osteosarcoma resection [71]. Similarly, strontium peroxide (SrO₂)-based OGS also promoted new bone formation and inhibited bone resorption, leveraging the dual role of strontium in bone metabolism by stimulating osteoblasts and inhibiting osteoclasts [72]. Of note, strontium is recognized not only as a bioactive trace element that supports bone health but also as a component of strontium ranelate, a well-known treatment for postmenopausal osteoporosis.

ROS-responsive scaffolds have been utilized in drug delivery. Lee and colleagues developed poly(D,L-lactide-co-glycolic acid) (PLGA)-based nanoparticles that managed ischemia/reperfusion injuries via the controlled release of heparin and glutathione in response to H₂O₂, thereby enhancing bone healing [90]. BTE scaffold coatings using ROS-responsive thioketal-based polymers enabled tunable release of BMP-2 and increased bone regeneration by 50% in rat calvarial defect models [91]. Similarly, a titanium implant coated with a hydrogel responsive to ROS in femoral bone defects released Tβ4 to promote macrophage transformation and osteogenic differentiation, and improve bone healing [92]. Aleemardani and colleagues integrated ROS-mitigating quercetin, a plant flavanol, into silk fibroin–CaO₂ hydrogel–electrospun fiber–hydrogel constructs [68]. This novel construct released O₂ for 12 days while simultaneously performing ROS scavenging, and increased cell viability by 30% under normoxia and by 10% under hypoxia compared with its quercetin-lacking counterpart. Overall, harnessing and modulating ROS within OGS is a promising strategy for overcoming the challenges associated with balancing O₂ supply and demand. Continued developments will include fine-tuning their sensitivity to ROS levels. Future research could explore the incorporation of novel O₂-sensing elements, chemical bonds, or O₂ carriers within the scaffolds. In addition, the long-term stability and biocompatibility of redox-active materials and their byproducts will necessitate further research to fully understand their interactions with the biological environment [93].

Biophysical models applied to cell-seeded PLGA–sodium percarbonate OGS could predict dynamic spatiotemporal O₂ interactions during bone healing by accounting for scaffold size, O₂ release, and cellular consumption rates in avascular bone defects [94]. Finite element modeling (FEM) has been used to explore how local O₂ tension, coupled with the mechanical environment within an implanted osteochondral scaffold, impacted on tissue differentiation and repair outcomes of mesenchymal stem cells (MSCs) [95]. FEM also simulated cellular-level vascular development by mathematically linking O₂ sources and Dll4-Notch signaling in endothelial cells [96,97]. At the tissue level, FEM-based approaches have been used to simulate bone regeneration by integrating mechanical stimuli, scaffold degradation, biological responses to VEGF supplementation, vascularization, and O₂ delivery [98]. In critical-sized bone defects, spatial variations in these parameters occur during bone regeneration. For accurate FEM simulations, thousands of representative volume elements are needed, resulting in high computational costs. However, the use of neural networks to predict bone regeneration in a sheep tibia model achieved results as accurate as FEM but at one-fifth of the computational cost. The ML model was validated by longitudinal in vivo X-ray images, and showed consistent bone regeneration at 12 month based on remodeling parameters established from the first 9 months of data [99]. Similarly, Ghosh and colleagues used neural networks to substitute tedious FEM analyses in predicting bone growth over macro-textured bone implant surfaces, further demonstrating the efficiency improvements provided by ML compared with traditional FEM methods [100]. Biological parameters for both FEM and ML models are usually sourced from the literature, in vitro/in vivo data, and empirical estimation, which can be updated via iterations and new experimental data [95,99]. ML tools integrate data from in vitro and in vivo experiments, thereby providing a comprehensive framework to elucidate the underlying mechanisms by which spatiotemporal O₂ gradients and bone scaffolds enhance osteogenesis, angiogenesis, and overall bone healing. ML approaches can also guide on-demand OGS design to match a desired optimal cellular response. Integrating noninvasive imaging data with ML techniques has the potential to further enhance the simulation of biological processes such as bone regeneration and revascularization. ML models trained on extensive imaging datasets can provide more accurate and robust predictions of the in vivo O₂ distribution. Ultimately, these models would be used clinically to facilitate customized interventions tailored to the patient and their defect based on initial imaging data.

Biophysical simulations and ML also offer powerful solutions to address the technical limitations of OGS designs by predicting the effects of different materials on O₂ generation, diffusion, and distribution. For example, O₂ gradients within CaO₂-based scaffolds supporting the high O₂ demand of islet cells were simulated using FEM [101., 102., 103.]. These models, based on oxygenated reaction, diffusion, and consumption kinetics, successfully guided the optimization of cell loading [103]. There are limited studies highlighting the potential of ML for advancing biomaterial designs for BTE applications. In the Polymer Genome Project, computational methods were used to predict polymer properties, such as glass transition temperature and solubility parameters, thereby enhancing material selection and design [104]. ML accelerated peptide discovery for self-assembling hydrogels by combining Monte Carlo tree search and random forest algorithms with molecular dynamics simulations, and efficiently identified sequences with high self-assembly potential [105]. In addition, ML could predict the material properties and bone-forming ability of metals using artificial neural networks (ANNs) [106] or in ceramics [106,107]. ML has been used to predict scaffold biodegradability, which affects O₂ release kinetics, tissue integration, and scaffold longevity [108]. Bioceramics and natural substances (e.g., collagen, fibrin), commonly used as OGS binding materials, exhibit complex degradation patterns owing to their inherent heterogeneity and bioactivity. Predicting these patterns using traditional mathematical principles is challenging. However, random forest algorithms have been trained on histological images to predict collagen-based scaffold biodegradation [109]. In addition, neural networks trained on the relationships between mechanical properties, crosslinking, and swelling characteristics and degradation rates have predicted gelatin-based bone scaffold degradation [110].

Finally, ML has been used to create bone scaffolds with controlled porosity, Young's modulus, and compression strength, and achieved <5% error from design targets [111,112]. Although integrating FEM and computational fluid dynamics (CFD) helps to simulate mechanical properties and O₂ permeability concurrently to iteratively adjust scaffold design [113,114], recent advances have utilized a more efficient ML approach for inverse design, such as supervised learning to generate Voronoi lattices with targeted mechanical properties that were validated through compression tests and simulations [112]. Another study used FEM simulations to create a neural network dataset linking microstructural parameters to the elastic matrix of bone [111]. ML also provides significant benefits to scaffold accuracy and consistency in image-based computer-aided design (CAD) and additive manufacturing (AM)-based biomanufacturing. Creating patient-specific bone scaffolds relies on accurate bone segmentation from computerized tomography (CT) scans. Minnema and colleagues improved this step by using automated convolutional neural networks (CNNs), and achieved a 92% similarity to gold standard segmentation tools in shorter times [115]. During the subsequent bioprinting process, ML models analyzed data signatures from in situ IR sensors in a feedforward process to predict print quality metrics and adjust process parameters in real time, thereby preventing manufacturing flaws and improving print quality [116].

Challenges to ML for OGS include the need for extensive, high-quality data for model training. The variability in the experimental conditions and requirements for preclinical versus clinical applications require collaborative efforts to build comprehensive datasets for OGS materials. Currently, there are limited publicly available biomaterial databases, and most groups still rely on extracting descriptors from multiple sources, which reduces scalability and increases computing power requirements [120]. High-quality, standardized data collection protocols will be needed in the future to ensure accurate and reproducible experimental data. In addition, the 'black-box' nature of many ML models limits the interpretability of ML-generated designs for in vivo applications. Therefore, ML should be considered as an aid, and not as a replacement for empirical trials, to be accompanied by rigorous experimental validation of OGS performance. Enhancing the transparency of ML algorithms through explainable artificial intelligence (AI) techniques will increase their clinical acceptance among medical professionals [121]. Creating effective ML models for OGS requires interdisciplinary expertise in materials science, tissue engineering, and computer science, which would further advance bone scaffold and next-generation OGS design.