The Journal of Supercritical Fluids New insights in the morphological characterization and modelling of poly( (cid:2) -caprolactone) bone scaffolds obtained by supercritical CO 2 foaming

Abstract Hierarchically porous synthetic bone grafts (scaffolds) are gaining attention in the clinical arena. Scaffolds should combine morphological (macro- and microporosity, pore interconnectivity), mechanical and biological (biocompatibility, degradation rate) properties to fit this specific use. Supercritical (sc-) foaming is a versatile scaffold processing technology. However, the selection of the optimum operating conditions to obtain a defined scaffold structure is hampered by the lack of a single characterization technique able to fully elucidate the porous features of the resulting scaffolds. In this work, the effect of soaking time (1, 3 and 5 h) on the preparation of poly(e-caprolactone) (PCL, 50 kDa) scaffolds by sc-foaming was evaluated by a combined X-ray microtomography (μ-CT) and mercury intrusion porosimetry (MIP) 3D-morphological analysis. Mechanical tests and in silico modelling for cell penetration and water permeability of the scaffolds were also conducted. Results evidenced the relevance of μ-CT and MIP as a synergistic analytical duo to fully elucidate the morphology of the sc-foamed scaffolds and the soaking time effect.


a b s t r a c t
Hierarchically porous synthetic bone grafts (scaffolds) are gaining attention in the clinical arena. Scaffolds should combine morphological (macro-and microporosity, pore interconnectivity), mechanical and biological (biocompatibility, degradation rate) properties to fit this specific use. Supercritical (sc-) foaming is a versatile scaffold processing technology. However, the selection of the optimum operating conditions to obtain a defined scaffold structure is hampered by the lack of a single characterization technique able to fully elucidate the porous features of the resulting scaffolds. In this work, the effect of soaking time (1, 3 and 5 h) on the preparation of poly(-caprolactone) (PCL, 50 kDa) scaffolds by sc-foaming was evaluated by a combined X-ray microtomography (-CT) and mercury intrusion porosimetry (MIP) 3D-morphological analysis. Mechanical tests and in silico modelling for cell penetration and water permeability of the scaffolds were also conducted. Results evidenced the relevance of -CT and MIP as a synergistic analytical duo to fully elucidate the morphology of the sc-foamed scaffolds and the soaking time effect.

Introduction
Bone is the second most common transplantation tissue and the harvesting of cancellous bone from the patient (allografts) is the current gold-standard surgical procedure to repair bone defects in the locomotor system. Nevertheless, the risk of infections, the intervention-associated damages and the limited availability of transplantable tissue evidence the need of new therapeutic approaches. The development of innovative synthetic grafts, the so-called scaffolds, provides a promising strategy to regenerate damaged tissues promoting the self-healing capacity of the patients. Scaffolds must display a 3D-interconnected and hierarchical porous structure and a mechanical behavior that are adapted to the anatomical target to get an appropriate performance once implanted [1].
Conventional methods for scaffold manufacturing frequently involve the use of high temperatures and/or organic solvents and may require long and tedious downstream pathways [2][3][4]. These processing conditions are usually incompatible with scaffolds containing bioactive compounds (i.e. medicated scaffolds) which do not withstand harsh treatments and may be also lost during purification. The use of medicated scaffolds is of particular interest to improve the scaffold integration and the precise tissue regeneration or to alleviate post-surgical harms. Supercritical (sc-) foaming is a versatile solvent-free green technology for the processing of biodegradable polymeric scaffolds. Sc-foaming is based on the sorption and dissolution of CO 2 in the polymeric matrix of the scaffold for a certain time period (soaking time; ST) under certain pressure (batch pressure; BP) and temperature (batch temperature; BT), followed by a pressure release (depressurization rate; DR). The latter step forces the polymer expansion and CO 2 removal in order to induce the formation of pores in the polymeric matrix. Due to the plasticizing effect of CO 2 , sc-foaming technology operates under moderate processing temperatures, i.e. compatible with the incorporation of thermolabile bioactive compounds such as growth factors [5][6][7], anti-inflammatory drugs [8][9][10] or antimicrobial agents [11][12][13] of interest for bone regeneration.
The fine control of the main operating variables of sc-foaming (BP, BT, ST, DR) allowed the manufacturing of tunable porous and interconnected architectures for polyester-based scaffolds, such as poly(D,L-lactic acid) (PLA) [14], poly(lactic-co-glycolic acid) (PLGA) [15,16], poly(-caprolactone) (PCL) [17,18] or poly(L-lactide-co-caprolactone) (PLCL) [19]. Rational design of patient-personalized and quality-reproducible scaffolds demands the development of standard operating procedures (SOP) for sc-foaming. Compilation of protocols and processing parameters is of utmost importance to obtain scaffolds fitted to the target grafting site demands, particularly in terms of pore size, morphology, distribution, throat size and interconnection that are critical for cell colonization and differentiation [20]. However, the modelling of the pore formation mechanism during sc-foaming process is still challenging but absolutely required for the definition of SOPs.
There are few studies reporting on the effects of the foaming processing conditions on the resulting polymeric scaffold 3D-architectures [18,[21][22][23][24]. Commonly, morphological characterization is carried out by scanning electron microscopy (SEM), which allows for direct measurements of pore sizes and visual estimations of pore interconnectivity although restricted to the exposed surface area. The evaluation of scaffolds using the SEM approach has severe limitations mainly related to the sample preparation itself and the method of data treatment. A physical sectioning of the scaffold is required as a previous step to expose the inner regions. This preparative step is usually performed manually using a surgical blade that makes the sample to be under compression and shear forces. These forces may cause structural damages to the structure, such as pore occlusion and deformation [25]. Moreover, flawed cuts of the scaffolds are frequent depending on the dimensions and the difficulty of handling of the scaffold; the obtained angled areas also compromise the quality and reliability of the results. Finally, the data treatment of SEM images to study the pore size distribution considers the size of a pore equivalent to the cross-section diameter of the pore in the SEM-image. This assumption usually results in an underestimation of the actual pore sizes in the case of spherical pores. For the case of elongated (cylindrical) pores, several specimens cut in the axial and coronal plane are needed to get a reliable pore size distribution from SEM images, requiring more amount of material as they are destructive tests.
Mercury intrusion porosimetry (MIP) is a well-established technique for the characterization of porous materials, based on the profile of mercury intrusion into pores when subjected to increasing pressures. Once the maximum pressure is reached, an extrusion profile is also obtained from the depressurization step. From both profiles, the pore and throat sizes of the sample can be calculated, and the pore tortuosity, compressibility or permeability may be inferred [26,27]. Despite the advantages of the technique, MIP is not a suitable characterization method for materials with large macropores due to its upper-limit measurable pore range (up to ca. 200 m) [28]. In addition, MIP does not take into account close pores and assumes perfectly cylindrical pores to correlate the volume of mercury intruded with the pore size, which do not always represent the reality of the analyzed samples [25,29].
X-ray-based microtomography (-CT) is a non-invasive method that provides quantitative and qualitative information regarding the 3D-morphology of samples and is commonly used for the analysis of trabecular bone [30,31]. The use of this technique is encouraging for scaffold characterization [32][33][34][35][36][37], although there are two main aspects to consider when performing a -CT scan: the duration of the analysis and the storage of the obtained data. Indeed, the scanning of low volume samples can last over 20 h and the generated files have sizes in the order of several terabytes depending on the -CT acquisition parameters (e.g. voxel size, rotation step) [38]. Also, the selected voxel size has a direct impact on the image resolution from the -CT scan and, subsequently, on the lowest value of measurable pore size of the porous materials. Therefore, a trade-off solution between the image resolution and the time-cost and data storage space consumption must be met.
Overall, the state-of-the-art of morphological characterization of scaffolds indicates that there is not a universal technique able to fully characterize the porous structure of polymeric scaffolds in the micro-to-macroporous range regarding pore-throat distributions and pore interconnectivities.
The combination of these complementary techniques (SEM, MIP and -CT) for the full characterization of scaffold pore structure can overcome the individual artifacts or pitfalls of each individual technique. To the best of our knowledge, it is the first time that the combination of these characterization techniques is exploited to generate information on process-structure-functionality relationships of scaffolds obtained by sc-foaming, which is of utmost interest for the definition of SOPs. Thus, the effect of the CO 2 soaking time during the supercritical foaming process on the scaffold morphology was evaluated to get the target scaffold features required for synthetic bone grafts. The morphological study of the resulting scaffolds of poly(-caprolactone) (PCL), a bioresorbable semicrystalline polymer [39][40][41][42][43][44][45], was performed based on the combination of SEM, MIP, helium pycnometry and -CT techniques. In silico studies of cell infiltration capacity and water permeability as well as in vitro mechanical tests were carried out for the sc-foamed scaffolds to predict the graft performance once implanted.

CO 2 sorption kinetics
A thermostized magnetic suspension balance (Rubotherm GmbH, Bochum, Germany) was used to evaluate the CO 2 sorption kinetics on the PCL at 140 bar and 39 • C, i.e. conditions close to body temperature and where PCL was molten according to preliminary view cell tests (not shown). The polymeric powder was dosed (80 mg) in a glass container and attached to the balance through a metal hook. Prior to the measurements, the polymer was molten at 80 • C in an oven to ensure a homogeneous CO 2 sorption along the material [46].

Processing of PCL scaffolds by sc-foaming
PCL powder was weighed (ca. 1 g) and dosed in cylindrical (length = 24.6 mm, internal diameter = 17 mm) Teflon moulds (Brand GmbH, Wertheim, Germany), followed by manual compression. Moulds were then placed in a 100 mL-stainless steel autoclave (Thar Process, Pittsburg, PA, USA) and heated to 39 • C (BT). Afterwards, the system was pressurized with CO 2 (5 g/min) until 140 bar (BP) and maintained in the static mode for a certain soaking period (ST = 1, 3 and 5 h). The system was stirred at 700 rpm during the whole process to enhance the mass transfer and to ensure a homogeneous CO 2 environment. The autoclave was then depressurized at a constant venting rate (DR = 1.8 g/min) until atmospheric pressure. Prior to their storage, the outer skin of the scaffolds was carefully removed using a surgical blade. Scaffolds were denoted as 39 ST referring to the processing temperature (BT= 39 o C) and the subscript according to the soaking time used (ST= 1h, 3h or 5h).

Structural characterization
Bulk densities ( bulk ) were obtained from the ratio between the weight and volume of each scaffold after sc-foaming. The skeletal density ( skel ) of the scaffolds was determined using a heliumpycnometer (Quantachrome, Boynton Beach, FL, USA) at room temperature (25 • C) and 1.01 bar. Values were obtained from six replicates. Overall porosity (ε) was calculated according to Eq. (1).
The morphological properties of the scaffolds were investigated by scanning electron microscopy (FESEM, ULTRA PLUS, Zeiss, Oberkochen, Germany) running at a voltage of 10 kV. Prior to their imaging, scaffolds were sliced with a scalpel and then iridiumsputtered (10 nm thickness).

MIP analyses and in silico modelling of PCL scaffolds
MIP analyses of the PCL scaffolds were performed (Autopore IV 9500 model, Micromeritics, Norcross, GA, USA) at working pressures ranging from 0.07−1800 bar to determine their pore size distributions in the 0.01−180 m range. Porosity values (ε MIP ) and pore size (MIP-Mean pore size) were determined from the intruded volume of mercury (V p MIP ) in the scaffolds with the increase of pressure using the Washburn equation [25]. A 3D-network model was generated from the MIP-cumulative curves having identical percolation properties as those of the manufactured PCL-scaffolds using PoreXpert v.1.6.567 software (PoreXpert Ltd, Plymouth, UK). This in silico model consists in a cubic structure formed by 1,000 pores (of cubic shape) connected by up to 3,000 throats of arbitrary cylindrical shape. A Boltzmann-annealed simplex algorithm was used to estimate and to simultaneously optimize the connectivity (mean number of throats per pore), pore skew, throat skew and correlation level from the MIP-cumulative curves. Water permeability (25 • C, 1.03 bar) was estimated assuming that Poiseuille flow (water) occurred in the z-direction according to Eq. (2) where l cell and A cell represent the length and the cross-sectional area of the unit cell, respectively, and ω cell (Farcs) is an averaging operator over the whole unit cell operating on the flow capacities of the pore throat-pore arcs parallel to the z-axis. PoreXpert calculates the term ω cell (Farcs) by means of the Dinic network analysis algorithm. Mesenchymal stem cells (MSCs) infiltration in the scaffolds was simulated using the filtration module from the software and assuming a cell size of 26.5 ± 5.0 m, an average value for human MSCs [47][48][49].

Micro X-ray computed tomography image acquisition, reconstruction and analysis
Micro X-ray computed tomography (-CT) scans were acquired (in the local mode) using a Phoenix v|tome|x tomograph (GE, Boston, MA, USA) equipped with a Varian PaxScan detector (1920 × 1536 pixels). Voxel resolution was set equal to 12 m.
3D images were reconstructed and analyzed using Avizo v.2019.1 software (Thermo Fisher Scientific, Waltham, MA, USA). A general scheme of the workflow from the image acquisition to complete 3D reconstruction and analysis is depicted in Fig. 1. Firstly, a specific region of interest (ROI) of cubic shape (4.8 × 4.8 × 4.8 mm 3 ) and representative of the entire scaffold was isolated to evaluate the influence of the working parameters on the resulting morphologies. Afterwards, the calibration of the thresholding of the grey-scale was performed to differentiate the void fraction (pores and pore interconnections, i.e. throats) from the solid material. This image thresholding is considered a crucial step prior to 3D modelling that influences the subsequent analysis and structural plots [50]. Once the void volume was precisely identified, porosity ( -CT ) of the scaffolds was calculated in the ROI and expressed as percentage of void voxels with respect to the total number of voxels. Connected porosity of the scaffolds was calculated (in percentage) as the volume fraction of the largest group of interconnected pores with respect to the total pore volume in the ROI of the material. Tortuosity was calculated from the mean pathway distance of a particle moving from one face of the ROI to the opposite one divided by the length of the edge of the cubic ROI (i.e. the shortest possible pathway).
A 3D-network model of balls and struts was also obtained from the -CT data, where the pores are represented as balls and the pore throats as cylindrical struts. The sizes of balls and struts were calibrated according to a color scale to visually identify morphological differences between scaffolds. In addition, pore size and pore throats distributions and their mean values were obtained based on the former 3D model. Finally, simulated MIP-data and pore volume distributions of the PCL-scaffolds were obtained from the 3D model generated from -CT using Xlib plugin from Fiji-ImageJ software [51,52].

Mechanical properties
PCL scaffolds (in triplicate) were subjected to unidirectional compression tests in a tensile bench with a 30 kg load cell Firstly, a specific region is isolated from the entire scaffold. Through the image thresholding step, the void fraction is converted to a 3D-network model of balls and struts, whose sizes are calibrated according to a color scale, and the solid material in gray. Based on the 3D model, pore size and throats distribution plots are obtained. (TA.XTPlus, Stable Micro Systems, Ltd., Godalming, UK) at a crosshead speed of 1 mm/min. All the experiments were performed at room temperature (25 • C), atmospheric pressure and 45 % relative humidity. Elastic deformation was calculated from the ratio between the initial height and the height of the sample bearing the highest applied physical stress. The Youngś modulus (E) was calculated from the stress-strain plots previous conversion to engineering stress and engineering strain.

Statistical analysis
All results were expressed as mean ± standard deviation. Statistical analyses of the mechanical results (1-way ANOVA) were performed using Statistica v8.0 software (StatSoft Inc., Tulsa, OK, USA) followed by the post-hoc Tukey HSD multiple comparison test.

Experimental CO 2 sorption kinetics in PCL
Solubility and diffusivity of CO 2 within a polymer can be modulated by tuning the working temperature and pressure leading to dramatic changes in the resulting foam morphologies [53]. The CO 2 sorption profile in PCL under an atmosphere of sc-CO 2 at 140 bar and 39 • C is shown in Fig. 2. The amount of CO 2 absorbed in PCL was greatly influenced by the soaking time with a fast CO 2 sorption kinetics followed by a slower stage after ca. 60 min of exposure with values in the near-saturation range (0.23−0.27 grams of CO 2 absorbed per gram of PCL). The solubility values of CO 2 at 35 • C and 130 bar, and 40 • C and 150 bar were previously reported to reach 0.22 and 0.40 g CO 2 /g PCL at the saturation stage, respectively [17,54]. The broad range between these two values indicates the proximity of the phase transition and shows that the reported Table 1 Morphological and textural properties of PCL scaffolds obtained by sc-foaming. Density values expressed as mean values ± standard deviation (n = 6).  values in the literature [54,55] are in fair agreement with this work, in spite of the differences in the molecular weight and crystallinity of the used PCL and in working pressures. It is reported that CO 2 dissolves to a higher extent in the amorphous regions of polymers and in polymers of lower molecular weights [56,57]. Finally, CO 2 solubility usually increases at higher pressures [17,54,58]. According to these results, the soaking times selected for further foaming trials were set at 1, 3 and 5 h to evaluate morphological differences depending on the increased amount of CO 2 dissolved in the PCL (0.238, 0.261 and 0.270 g CO 2 /g PCL, respectively)

Sc-foaming process development and morphological characterization of the scaffolds
Cylindrical and highly porous PCL scaffolds ( = 68-72 %) were obtained through sc-foaming, matching the human trabecular bone porosity values [59,60]. A dome-like top ending was observed for scaffolds processed with soaking times above 1 h in contact with the CO 2 at the foaming pressure ( Figure S1). All specimens presented a non-porous skin of 100−140 m thickness due to the rapid CO 2 diffusion upon depressurization from the surface of the PCL matrix [61]. Longer soaking times favored the polymeric expansion upon depressurization, slightly decreasing the bulk density values ( bulk ) of the manufactured scaffolds ( Table 1).
The morphological analysis of the scaffolds from SEM microscopy images unveiled smooth surfaces with subtle differences among them regarding the pore geometries and densities (Fig. 3). 39 1h scaffolds presented the less homogenous pore morphologies with small pores (<100 m) combined with larger ones (Fig. 3A and B). Enhanced cell adhesion, migration, proliferation and differentiation are reported for polymeric scaffolds with a porous population in the 20-50 m range [62,63]. Longer ST (5 h) favored the homogeneity and sphericity of the pores (39 5h scaffolds in Fig. 3E and F). Qualitatively, scaffolds processed during 3 h pre- Table 2 Pore throat and interconnectivities of PCL scaffolds processed by sc-foaming obtained from MIP and -CT data analysis.

MIP
-CT sented more void spaces along the analyzed area (39 3h scaffolds in Fig. 3C,D) than 39 5h scaffolds. All scaffolds had interconnected pores as highlighted by the presence of inner pores within the large pore cavities (white arrows in Fig. 3B, D and F). Pore interconnectivity strongly influences the performance of the scaffolds since low levels can hamper the cell colonization phenomena and the diffusion of nutrients and removal of waste products from the cells [1,4]. The pore interconnectivity level of scaffolds obtained by sc-foaming can be modulated by modifications on the depressurization rate [18,64].  (Table 1). This divergence in values can be attributed to the presence of pores either larger than 180 m or closed. Namely, the lowest overall porosity was obtained for the 39 3h scaffold. The mean pore size calculated from MIP measurements unveiled that scaffolds processed at longer ST presented larger pores increasing up to 6 % for 39 5h scaffold ( Table 1). The pore throat diameters were virtually identical, although differences in the degree of pore connection (interconnectivity) were obtained, mainly related to the different tortuosity of the scaffolds (Table 2). Overall, the manufactured scaffolds present good pore interconnectivity (above 70 %) for regenerative medicine purposes, regardless of the working parameters.

-CT imaging of PCL scaffolds obtained by sc-foaming using different soaking times
The presence of numerous pores with diameters larger than 200 m (as observed by SEM) encouraged the use of -CT in order to effectively assess the effect of CO 2 soaking time on the resulting morphologies. Porosity values determined from -CT images (ε −CT ) were in the 67-68 % range, being close to the overall poros-  ity values calculated from Eq. (1) ( Table 1). Although all scaffolds had similar porosity values, remarkable differences in their morphology were observed in the 2D -CT images as a function of the processing time (ST). An increase in the number of pores and a reduction of the pore size was recorded as the ST increased ( Figs. 4 and 5). An overestimation of the mean pore size was obtained from the analysis of the 2D axial plane sections when compared to the bulk structure, since the foams present elongated pores (Table 1). On the other hand, the total specific pore volume of the scaffolds calculated from -CT data underestimated in 5-10 % the values obtained from MIP analysis, since the volume contribution of small macropores and mesopores cannot be considered when -CT is used due to resolution limitations (Fig. 6).
The architecture of each scaffold was further analyzed from the data of the 3D-model obtained from the library of 2D slices (Fig. 7). Longer soaking times led to a reduction in the mean pore sizes of the scaffolds and narrower pore size distribution. Particularly, 39 5h scaffolds presented a narrow pore size distribution with values falling in the ideal size range (1−500 m) for bone tissue engineering [65,66]. These results are in line with previous studies with neat PCL scaffolds showing a remarkable reduction of the pore size with the increase of the soaking time [18]. Conversely, other study [67] reported the opposite effect during the supercritical foaming of pure PCL scaffolds, where the lowest pore diameters (11.75 m) were obtained after ST = 1 h. Overall, the reduction of the pore size of the foamed scaffolds obtained in this work follows the classical nucleation-growth theory of pore formation [68,69]. At longer processing times, higher initial amounts of CO 2 are dissolved within the polymer (Fig. 2) lowering the interfacial tension of the PCL-CO 2 system. This reduction in interfacial tension facilitates the formation of more nucleation sites since the initial critical nucleation radius is reduced [70][71][72]. Then, upon depressurization, scaffolds of higher cell densities with smaller pores are typically obtained when processed at longer soaking times [18,71]. The reduction in pore size is a consequence of the spatial limitation to the growth of pores from these nucleation sites due to their high abundance. Moreover, longer soaking times allow for a more efficient CO 2 distribution along the polymeric matrix resulting in more uniform and narrower pore distributions [32]. Both effects are easily appreciated in Fig. 7 as the soaking time increased.
Scaffolds allowing the infiltration of cells within their porous structures and the transport of nutrients, metabolites and wastes through them are needed to match the demands of the bone regeneration process. These properties are strongly dependent on the morphological properties of the scaffolds regarding pore and throat size distribution as well as the pore interconnectivity. The Table 3 In silico predicted values of water permeability and MSCs infiltration on sc-foamed PCL scaffolds.

Scaffold
Water permeability (m 2 ) Cell infiltration (%) degree of connected porosity in the scaffolds and how the pores are interlinked were characterized by the pore throat analysis. These features are of utmost importance since inhibition effects on the tissue differentiation process were reported for porous implants of narrow pore throats [73]. The increase in the soaking time resulted in scaffolds showing a reduction on the mean throat diameters and narrower throat size distribution, following a similar trend to the pore sizes discussed above ( Table 2 and Fig. 7). Furthermore, the decrease in the mean throat size resulted in an increase on the tortuosity values, although the degree of connected void fraction was almost 100 % for the longer soaking time tested (39 5h scaffold).

In silico modelling of water permeability and human MSCs infiltration
The microstructure of scaffolds plays a critical role in cell infiltration and distribution of biological fluids [74]. Despite the intrinsic limitations of MIP technique previously presented, it widely covers pore populations below 10 m, unreachable by both -CT (with the selected acquisition parameters) and MIP in silico simulation based on -CT data (Fig. 6). Therefore, the simulation of the permeability and cell infiltration values of the obtained scaffolds was based on experimental MIP data.
Water permeability values of 39 1h and 39 3h scaffolds (Table 3) were of the same order of magnitude to those reported in silico for PCL scaffolds obtained by sc-foaming at 37 • C and the same pressure [75], and in vitro for low molecular weight-PLGA scaffolds (ca. 10 −13 m 2 ) with similar porous structures [76]. The increase of soaking time augmented water permeability and 39 5h scaffolds displayed one order of magnitude larger values. Overall, the obtained in silico permeability values for the manufactured scaffolds were close to the lowest experimental data reported for cancellous bone (10 -12 to 10 -8 m 2 ) [77,78]. Nevertheless, scaffold permeability was based on a water flow in a single axis direction and differences are expected as the structures have anisotropic properties. For instance, not only directional but also spatial differences in permeability calculations were reported for human [79], porcine [80] and bovine [81] cancellous bone. Despite the abovementioned limitations of the in silico model, permeability values were of the same order of magnitude to those reported for natural bone by experimental perfusion methods [78].
The cell infiltration capacity in the PCL scaffold was evaluated through the spread capacity of particles with defined sizes (corresponding to human MSCs dimensions). 39 5h scaffold had cell infiltration values over 90 %, suggesting that this structure may have full accessibility for cells once implanted, allowing a homogeneous tissue ingrowth instead of restricting it to the outer surface of the scaffold. Lower cell infiltration values were obtained for 39 1h and 39 3h scaffolds. Results should be sparingly considered since the model in this work was based on individual and non-interacting particles. For instance, the formation of undesired cell aggregates or clusters during the cell seeding period may occur, being a source of variability in the in vitro determination of cell infiltration [82].
Overall, 39 5h scaffold is the most promising candidate regarding its further biological performance as synthetic bone grafts in terms of water permeability and cell infiltration capacity.

Mechanical properties
Mechanical performance of scaffolds directly correlates to the structural modifications induced by the variation of processing conditions. During the foaming process, the CO 2 was vented in a single direction and the polymer expansion was forced to occur preferentially in the vertical axis, obtaining PCL foams with elongated pores mimicking the natural bone anisotropy (Fig. 3) [83]. Scaffolds presented an elastic deformation of ca. 25 % ( Figure S2) and the obtained Young's moduli were in the 5−8 MPa range (Table 4), being in the reported range for Table 4 Mechanical characterization of the sc-foamed PCL scaffolds. Mean values and standard deviation (n = 3). Results were statistically identical (1-way ANOVA; p < 0.05).

Conclusions
The architecture of porous PCL scaffolds produced by supercritical CO 2 foaming can be tailored by modifications of the processing conditions to provide highly adaptable scaffolds. From the -CT analysis, realistic scaffold reconstructions were obtained, being particularly useful to analyze the effect of the processing conditions on the resulting morphologies. Longer soaking times permit more CO 2 to be dissolved in the polymeric matrix, leading to higher density of pores with lower sizes. In addition, more homogeneous scaffolds and higher degree of pore interconnection were obtained with longer soaking times. On the other hand, MIP allowed the characterization of the meso-and low macro-pore population and of the degree of pore interconnection. This is of particular interest since the graft performance after implantation can be highly affected by these pore populations. In this sense and based on MIP measurements, the in silico modelling of cell infiltration capacity and water permeability of the obtained scaffolds constituted a potential screening tool for further in vitro/in vivo biological tests. The combination of complementary characterization techniques (-CT and MIP) coupled to the modelling of the generated data, offers not only broader and more realistic information regarding the manufactured scaffolds but also regarding their potential use as synthetic bone grafts. Overall, the herein presented supercritical CO 2 foaming process allows the manufacture of PCL scaffolds meeting the structural and mechanical requirements for bone tissue regeneration purposes. This work represents a step forward towards the knowledge on process-structure-functionality relationships in synthetic bone grafts for the definition of standard operating procedures in the manufacturing of poly(-caprolactone) scaffolds by supercritical CO 2 foaming.

Declaration of Competing Interest
Authors declare no conflict of interest.