Structure–property-processing correlations in freeze-cast composite scaffolds

Acta Biomaterialia 9 (2013) 6338–6348

Contents lists available at SciVerse ScienceDirect

Acta Biomaterialia journal homepage: www.elsevier.com/locate/actabiomat

Structure–property-processing correlations in freeze-cast composite scaffolds Philipp M. Hunger a,b, Amalie E. Donius b, Ulrike G.K. Wegst a,⇑ a b

Thayer School of Engineering, Dartmouth College, 14 Engineering Drive, Hanover, NH 03755, USA Department of Materials Science and Engineering, Drexel University, 3141 Chestnut Street, Philadelphia, PA 19104, USA

a r t i c l e

i n f o

Article history: Received 24 July 2012 Received in revised form 21 December 2012 Accepted 1 January 2013 Available online 12 January 2013 Keywords: Composites Porosity Biopolymers Scaffolds Biomimetics

a b s t r a c t Surprisingly few reports have been published, to date, on the structure–property-processing correlations observed in freeze-cast materials directionally solidified from polymer solutions, or ceramic or metal slurries. The studies that exist focus on properties of sintered ceramics, that is materials whose structure was altered by further processing. In this contribution, we report first results on correlations observed in alumina–chitosan–gelatin composites, which were chosen as a model system to test and compare the effect of particle size and processing parameters on their mechanical properties at a specific composition. Our study reveals that highly porous (>90%) hybrid materials can be manufactured by freeze casting, through the self-assembly of a polymer and a ceramic phase that occurs during directional solidification, without the need of additional processing steps such as sintering or infiltration. It further illustrates that the properties of freeze-cast hybrid materials can independently be tailored at two levels of their structural hierarchy, allowing for the simultaneous optimization of both mechanical and structural requirements. An increase in freezing rate resulted in decreases in lamellar spacing, cell wall thickness, pore aspect ratio and cross-sectional area, as well as increases in both Young’s modulus and compressive yield strength. The mechanical properties of the composite scaffolds increased with an increasing particle size. The results show that both structure and mechanical properties of the freeze-cast composites can be custom-designed and that they are thus ideally suited for a large variety of applications that require high porosity at low or medium load-bearing capacity. Ó 2013 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved.

1. Introduction The first detailed studies on ‘‘freeze casting’’, the directional solidification of a polymer solution and the mechanisms by which a well-controlled and complex cellular architecture can be created, are probably those published by Ezekwo et al. [1] and Tong et al. [2,3]. They studied in detail the physical phenomena of directional solidification in water-based agar solutions and observed the effect of freezing-front velocity and diffusion conditions on the geometry of the ice-templated structure, drawing on extensive literature concerned with the directional solidification of metals and alloys. The first reports on the freeze casting of porous ceramics appeared at the turn of the millennium [4,5], followed by publications that described the potential of freeze-cast ceramic scaffolds for biomedical applications due to an attractive combination of connected porosity and mechanical properties [6–8]. Interestingly and despite the fact that both ceramics and metals are freeze-cast using polymeric binders to stabilise the green body

⇑ Corresponding author. Tel.: +1 603 646 3148; fax: +1 603 646 6584. E-mail address: [email protected] (U.G.K. Wegst).

before sintering, no studies have been published to date that report property values for freeze-cast ceramic composites solely glued with a polymeric phase and not further processed by sintering. One important advantage of such materials design is that multiscale composite materials can be created, which emulate the hierarchical composite structure found in stiff and strong, yet tough, natural materials such as nacre and bone [9–11]. Another advantage is that this cold processing route permits the addition of functional groups and components to the polymeric phase during scaffold formation to create functionalised biomaterials, whose activity is not compromised due to high temperature processing steps. A third advantage is that freeze casting is a highly versatile process with which highly connected and aligned porosity can be created [12]. It provides excellent control across several length-scales over the pore size and geometry as well as the pore wall architecture [13]. Thus, hierarchical scaffold architectures, including those with gradient properties in composition, structure and properties, can be produced [14]. Currently, applications of particular interest for freeze-cast materials range from hard and soft tissue scaffolds to solid oxygen fuel cells, and from catalyst carriers to electrodes in electrochemical cells [5,7,15–17]. Though little explored, great potential for

1742-7061/$ - see front matter Ó 2013 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.actbio.2013.01.012

P.M. Hunger et al. / Acta Biomaterialia 9 (2013) 6338–6348

freeze-cast materials also exists in applications that require the porosity of the scaffold to be filled with a second phase. Overall, freeze casting is a process with great promise for the synthesis of hierarchical materials that combine high porosity with excellent mechanical performance and whose properties can be tailored to a specific application. For the custom design of materials, it is important to gain a fundamental understanding of how material composition and processing affect structure and mechanical properties. The work presented here is a first attempt to explore these with a model material consisting of alumina particles in a biopolymer matrix. From it, hierarchically structured composites with a highly aligned, honeycomb-like porosity were prepared by freeze casting. At the first hierarchical level, pore size and geometry were modified through variation of the freezing rate and studied at a constant overall porosity and composition. At the second level of hierarchy, the effect of particle size and single vs. bimodal particle size distribution on the cell wall’s mechanical properties and performance were investigated, while keeping the overall cell wall material composition constant. 2. Materials and methods 2.1. Solution preparation and properties Low molecular weight chitosan (75–85% deacetylated, Sigma Aldrich, St. Louis, MO) and type B gelatin from bovine skin (Sigma Aldrich, St. Louis, MO) were used as received. Chitosan and gelatin solutions were prepared by dissolving 2.4% (w/v) chitosan and 5.5% (w/v) gelatin in 1 vol.% glacial acetic acid (VWR International, West Chester, PA). The chitosan solutions were mixed thoroughly by magnetic stirring at 60 rpm for 24 h. Gelatin solutions were mixed by magnetic stirring at 60 rpm for 12 h at a temperature of 35 °C. After initial stirring, chitosan and gelatin solutions were mixed in a volume ratio of 4:1 in a high shear SpeedMixer (DAC 150 FVZ-K, FlackTek, Landrum, SC) at a speed of 1600 rpm for 60 s to blend a 3% (w/v) polymer solution consisting of 63 wt.% chitosan and 37 wt.% gelatin [18]. The viscosity of the chitosan–gelatin solution was determined at room temperature (25 °C) on an AR 2000ex rheometer (Rheology Advantage Version 5.5.0 data analysis software, TA Instruments, New Castle, DE) [19]. A concentric cylinder geometry with a stator inner radius of 15 mm, a rotor outer radius of 14 mm, a cylinder immersion height of 42 mm and a gap of 5920 lm was used to measure the viscosity. Approximately 20 ml of the solution was pipetted into the stator of the concentric cylinder system, and the viscosity was measured in a shear rate range of 40–1100 s1. The viscosity of the 63/37 wt.% chitosan–gelatin in 1% acetic acid solution, which has a density of 1.01 g cm3, was found to be 0.0689 Pa s at room temperature.

6339

2.3. Freeze casting and lyophilisation Polymer solutions and ceramic slurries were directionally solidified using the freeze casting system detailed in an earlier publication [20]. Briefly, 10 ml of the polymer solution or ceramic slurry was poured into a polytetrafluoroethylene (PTFE) mould, which was sealed with a copper bottom plate and degassed at 1600 rpm for 60 s in a DAC 150 FVZ-K SpeedMixer. For freezing, the filled mould was placed with its copper bottom on the temperature-controlled copper cold finger of the freeze-caster, while the top of the mould remained open to atmospheric conditions. After precooling the slurry to 5 °C, the cold finger temperature was reduced at a constant freezing rate of either 1 or 10 °C min1 to directionally solidify the sample. Once fully frozen, the sample was removed from the cold finger, demoulded with a punch and placed for at least 72 h in a FreeZone 4.5 Liter Benchtop Freeze Dry System (Labconco, Kansas City, MO) to sublimate the ice. To determine the velocity at which the freezing front travels along the sample for a given applied cold finger cooling rate, the temperature profile was monitored and recorded during freeze casting by a mould equipped with six thermocouples mounted along its height at a spacing of 6.35 mm. To avoid any physical disturbance of the freezing process, the thermocouples were positioned in 640 lm diameter holes drilled from the outside into the PTFE mould wall, maintaining a wall thickness of 170 lm between the thermocouple and the inside of the mould. The freezing front was assumed to be positioned at the 0 °C temperature contour and the freezing front velocity calculated as the speed at which this travels along the height of the mould. Local cooling rates were determined for each thermocouple position. The density of the samples was determined by weighing each sample cube for mechanical testing and dividing it by the sample volume of 25 mm3. The relative density was calculated as the ratio of the measured density to the density of the solid wall material. The overall porosity P of the scaffolds was determined as P = 1  qrel. 2.4. Sample preparation for structural and mechanical characterisation After lyophilisation, the dry cylindrical samples were fixed with their bottoms on a ceramic plate for cutting into predetermined shapes and sizes for structural and mechanical characterisation. Cutting was performed by hand with a 220 lm diameter diamond-decorated steel wire and a wire speed of 0.7 m s1 on a Well 4240 saw (WELL Diamond Wire Saws, Inc., Norcross, GA). For structural and mechanical characterisation, cubes with a side length of 5 mm were cut at three standard heights, measured from the bottom of the sample with cube centres at 7, 17.5 and 28 mm, respectively. For each sample, four cubes were prepared and tested per sample height, and at least three samples of each composition were prepared and investigated mechanically and structurally.

2.2. Slurry preparation and properties Ceramic slurries of identical composition, but of three different particle size distributions, were prepared by suspending 27% (w/v) alumina particles of two different sizes in the aforementioned chitosan–gelatin solution using the SpeedMixer at a speed of 2500 rpm for 60 s for mixing and degassing. The particle diameter in slurry 1 was small (S), d50 = 400 nm (Sasol North America Inc. – Ceralox Division, Tuscon, AZ); the particle diameter in slurry 2 was large (L) d < 10 lm (Sigma Aldrich, St. Louis, MO). Slurry 3 had a bimodal (B) particle size distribution with 70 wt.% of d < 10 lm particles and 30 wt.% of d50 = 400 nm particles. All three slurries resulted in a final dry composition of 75 vol.% alumina, 16 vol.% chitosan and 9 vol.% gelatin.

2.5. Structural characterisation through microscopy and nitrogen sorption (BET) To determine the overall porosity of the entire cylinder before cutting and to measure the density of each specimen before structural characterisation and mechanical testing, the samples were weighed on a high precision balance (±0.01 mg; XP105 Delta Range, Mettler Toledo Inc., Columbus, OH). The 5 mm cubes were precision cut with the diamond wire saw, whose wire was positioned for cutting with a micrometre screw of 0.5 lm accuracy. The structure of the freeze-cast scaffolds was quantified using scanning electron microscopy (SEM) with a Zeiss Supra 50VP (Carl Zeiss NTS LLC., Peabody, MA, USA) at accelerating voltages between

6340

P.M. Hunger et al. / Acta Biomaterialia 9 (2013) 6338–6348

Fig. 1. Two typical stress–strain curves for the porous hybrid scaffolds of this study. The left and right curves were obtained when testing a scaffold made with small (a) and large (b) particles, respectively. Young’s modulus was determined from the slope of the initial linear region; the yield strength is the maximum stress at the end of the linear region.

2 and 10 kV. At least three images per cube of three 5 mm cubes were taken from the three layers of each specimen at identical magnification. At least 50 lamellar spacings were measured on each SEM micrograph using the digital imaging software ImageJ (ImageJ, U.S. National Institutes of Health, Bethesda, MD). The obtained values were averaged for each layer of the freeze-cast specimens. Pore size distributions were determined through nitrogen sorption in a Quadrasorb BET system (Quantachrome Instruments, Boynton Beach, FL, USA) and evaluated with the manufacturer’s quenched solid density functional theory model. This allows for a detection of pore sizes and their distribution within a range of pore diameters from 0.5 to 40 nm. 2.6. Mechanical characterisation Compression tests were conducted at ambient conditions on an Instron 4442 (Instron, Norwood, MA) with a 50 N load cell and a cross-head speed of 0.05 mm s1, corresponding to a strain rate of 102 s1. Compression tests were chosen because cellular materials, such as those described here, tend to be used in applications in which they are loaded in compression and bending, rather than in tension, and because they typically fail in compression, also when loaded in bending. Young’s modulus and yield strength were determined from compressive stress–strain curves obtained on the 5 mm cubes tested parallel to the lamellae in the structure. Typical stress–strain curves for the composite scaffolds are shown in Fig. 1. Young’s modulus and yield strength were determined as indicated. At least three cubes of each of the three layers, i.e. at least nine specimens overall, were tested from the three samples. 3. Results 3.1. Slurry sedimentation Because the density of the ceramic particles greatly differs from that of the polymer solution, sedimentation may occur during the freeze casting process, resulting in gradients in slurry composition and particle size distributions along the height of the sample [21]. The significance of sedimentation depends, for a specific particle– solution composition, on the difference between the densities of particles and solution, on the particle size and viscosity of the solution, and on additives as well as on the freezing rate at which it solidifies [20]. Sedimentation is thus another parameter with which biomaterials can be custom-made. If, for example, a homogeneous material is desired, the freezing rate or the slurry viscosity can be adjusted in such a way that sedimentation is minimised. Alternatively, a particle size distribution can be chosen with which a desired gradient in composition, structure and mechanical properties can be achieved [22].

The sedimentation speeds of the two particle sizes of this study were calculated according to Stokes’s law [23]. The viscosity and density of the chitosan–gelatin solution were measured, at a temperature of 25 °C, to be 68.9 mPa s and 1.01 g cm3, respectively. For the alumina particles, a density of 3.96 g cm3 and the particle diameters given in the manufacturer’s specifications sheet were used. For the small particles with a diameter of 400 nm, a sedimentation velocity of 3.79 nm s1 was calculated; this corresponds to 13.6 lm h1. The larger particles with a diameter of 10 lm were found to sediment significantly faster; their velocity was calculated to be 2.37 lm s1, which corresponds to 8.53 mm h1. At a freezing rate of 10 °C min1, it takes 40 min to freeze 10 ml of a composite sample which corresponds to a sample with a height of 35 mm; at a freezing rate of 1 °C min1, the solidification of the same amount of slurry takes 80 min. This means that sedimentation can be ignored in the case of the small particles, though it has to be taken into account in the case of large particles. 3.2. Pore morphology and overall porosity A longitudinal and a transverse cross-section of a typical sample frozen at 10 °C min1 are shown in Fig. 2. Parallel to the direction of solidification, the pores are highly elongated and several tens of millimetres in length, while perpendicular to it, the structure shows a lamellar arrangement of elongated cells. On average, the scaffolds had an overall porosity of 90.9 ± 0.6%. Figs. 3b and d show how the local temperature varies along the sample for the two freezing rates of 1 and 10 °C min1; temperature contours are plotted along the sample height for 0, 5 and 10 °C. The freezing front velocity is assumed to be that with which the 0 °C temperature contour travels through the sample. At an applied freezing rate of 1 °C min1, the freezing front velocity was determined to be 7.4 lm s1, whereas at 10 °C min1 it was with 27.7 lm s1 almost four times higher. In combination, Fig. 3 illustrates how the freezing rate, local temperature and freezing front velocity correlate and affect the freeze-cast material’s pore structure. For freezing rates of 1 and 10 °C min1 and a sample that has an overall height of 35 mm, Figs. 3a and c show how the lamellar spacing varies along the sample height. It decreases by 30–40% from the top to the bottom of the sample and is, on average, smaller by a factor of about 1.2–1.6 when frozen at the faster rate of 10 °C min1. 3.3. Cell wall structures Fig. 4 illustrates the differences in the solid cell wall structure that is created by self-assembly during the freezing process. Small particles and bimodal particle distributions result in a dense pack-

P.M. Hunger et al. / Acta Biomaterialia 9 (2013) 6338–6348

6341

Fig. 2. (a) Parallel alignment of lamellae with the solidification direction in longitudinal cross-section and (b) elongated pores with connecting walls between lamellae in cross-section perpendicular to the solidification direction. The blue arrows indicate the solidification direction; the white arrows highlight connecting walls between adjacent lamellae. The scale bars are 20 lm.

Fig. 3. Average lamellar spacings at different heights as determined from SEM micrographs for scaffolds made from small (S) or large (L) particles, or the bimodal (B) particle distribution freeze cast at a freezing rate of (a) 1 °C min1 or (c) 10 °C min1. Temperature profile throughout the PTFE mould during freezing at (b) 1 °C min1 and (d) 10 °C min1. The slope indicates the velocity of the freezing front.

ing in which the small particles are glued together by the biopolymer. In the case of the bimodal particle distribution, large particles are interspersed at relatively regular intervals within a wall that otherwise possesses the structure of the pure small particle wall structure. Large particles pack much less regularly, but are, in principle, aligned in the cell wall like beads on a polymer string. This string is visible in Fig. 4; it is the 400 nm thick chitosan–gelatin film that also coats the particles’ surfaces. From the SEM micrographs, there appears to be a significant amount of porosity incor-

porated in the walls with the small and bimodal particles. The cell wall thicknesses for the three particle sizes and distributions ranged from 0.3 to 5.8 lm, and are listed in Table 1. 3.4. Cell wall porosity Nitrogen sorption measurements for the porous composites with BET revealed significant differences in nanopore size distributions between the three different scaffold types. While for both

6342

P.M. Hunger et al. / Acta Biomaterialia 9 (2013) 6338–6348

Fig. 4. Individual lamellae illustrate the three different composite cell wall structures. Cell wall made with (a) small particles (S), (b) a bimodal particle size distribution (B) and (c) large particles (L). The scale bars are 5 lm. Reprinted from Ref. [11] with permission from Elsevier.

Table 1 Structural properties of the alumina composites frozen at bottom plate cooling rates of 1 and 10 °C min1. Freezing rate

Alumina particle size

Overall porosity (%)

Lamellar spacing (lm)

Wall thickness (lm)

1 °C min1

400 nm 30:70, 400 nm:10 lm 10 lm

91.4 ± 0.1 90.8 ± 0.1 90.5 ± 0.2

31.00 ± 10.85 36.10 ± 15.00 34.39 ± 9.89

5.79 ± 1.06 3.12 ± 0.67 0.55 ± 0.38

10 °C min1

400 nm 30:70, 400 nm:10 lm 10 lm

91.8 ± 0.1 90.9 ± 0.1 90.2 ± 0.1

27.40 ± 8.25 30.85 ± 11.32 27.34 ± 9.38

2.79 ± 1.00 2.09 ± 0.20 0.32 ± 0.06

freezing rates and all particle sizes a peak is present at a pore size of 3 nm, additional peaks were found at 8, 15 and 20 nm in the case of scaffolds made from small particles (S) and from the bimodal particle size distribution (B), but not in scaffolds made from large particles (L) (Fig. 5a and c). As a control, scaffolds made from the same chitosan–gelatin solution, but without alumina particles, were also investigated with BET nitrogen adsorption. For both freezing rates, these scaffolds exhibited only one peak at 3 nm, indicating that their presence is independent from the presence of particles and likely created in the polymer during the freezing or freeze-drying process. Figs. 5b and d show that, for a freezing rate of 1 °C min1, scaffolds consisting of small particles have the highest cumulative volume, at 0.049 cm3 g1, followed by scaffolds with a bimodal particle distribution, with a volume of 0.019 cm3 g1, then scaffolds consisting of large particles, with a volume of 0.012 cm3 g1. The same trend is observed when the scaffolds are frozen at 10 °C min1: small-particle scaffolds possess the highest cumulative pore volume, at 0.050 cm3 g1, followed by bimodal-particle scaffolds, at 0.017 cm3 g1, then large-particle scaffolds, at 0.011 cm3 g1.

3.5. Mechanical properties Young’s modulus and yield strength were determined on samples frozen at freezing rates of 1 and 10 °C min1 from chitosan– gelatin polymer solutions and three alumina–chitosan–gelatin slurries. The results are illustrated in Fig. 6 and listed in Table 2. The polymer samples, made as controls, had a Young’s modulus of 0.62 ± 0.23 and 1.01 ± 0.33 MPa and a compressive yield strength of 36 ± 9 and 57 ± 12 kPa for freezing rates of 1 and 10 °C min1, respectively. The properties of the chitosan–gelatin polymer could be significantly increased when ceramic particles were added to create composite scaffolds. When made with a bimodal particle size distrubution and a freezing rate of 1 °C min1, the ceramic–polymer composites had a Young’s modulus of 9.63 ± 1.61 MPa and a

yield strength of 0.244 ± 0.044 MPa. At a freezing rate of 10 °C min1, these values were considerably higher, with a Young’s modulus of 20.05 ± 2.45 MPa and a yield strength of 0.346 ± 0.023 MPa. Also, the particle size was found to have a significant effect on the scaffold properties. The values obtained for the different particle sizes spanned almost an order of magnitude. At a freezing rate of 1 °C min1, for example, the Young’s modulus and compressive yield strength ranged from 2.6 to 14 MPa and from 0.10 to 0.27 MPa, respectively. 4. Discussion 4.1. Freezing rate and freezing front velocity The samples of this study were frozen at two different applied freezing rates: 1 and 10 °C min1. Figs. 3b and d show plots of the temperature contours of 0, 5 and 10 °C for both. Assuming that the freezing front coincides with the 0 °C temperature contour, the freezing front velocities were found to be initially constant, then slowed somewhat for 10 °C min1. These constant freezing front velocities were measured to be 7.4 and 27.7 lm s1 at the two cooling rates of 1 and 10 °C min1, respectively, and found to agree well with estimates calculated based on the Stefan problem for the solidification of flat layers, assuming a constant freezing front velocity along the mould (quasi-static conditions) [20,24]:

v

kS ¼ þ 2tk

sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi k2S kS  ðT 0 ðtÞ  T E Þ 2 t q L hE 4t2 k

ð1Þ

and a heat transfer resistance of

1 dC 1 ¼ þ k kC a

ð2Þ

where T0(t) is the cold finger temperature at time t, TE = 0 °C is the thermal transition temperature, qL = 1 g cm3 is the density of water, hE = 333.6 J g1 is the latent heat of ice, ks= 3.26 W m1 °C1 is the thermal conductivity of the solidified ‘‘packed bed’’ layer of ice and alumina particles, dC = 5 mm is the thickness of the mould

P.M. Hunger et al. / Acta Biomaterialia 9 (2013) 6338–6348

6343

Fig. 5. Pore size distribution and cumulative pore volume as determined by BET. Freezing rates: 1 °C min1 (a,b) and 10 °C min1 (c,d).

rate was maintained only until a temperature of 80 °C was reached; the temperature was then kept constant at that level. For a freezing rate of 1 °C min1, this means that a constant cooling rate was maintained for the entire 80 min long freezing process, while in the case of 10 °C min1 the cooling rate was constant only for the first 8 min of the 40 min freezing time. As a result, quasisteady-state conditions are reached much earlier at the higher freezing rate of 10 °C min1 than at the lower rate of 1 °C min1, and the freezing front velocity decreases. 4.2. Correlation between freezing front velocity, local cooling rate and lamellar spacing

Fig. 6. Young’s modulus and yield strength of the different hybrid scaffolds showing the trend of increasing mechanical properties with increasing freezing rate and increasing average particle size.

bottom copper plate, kC = 400 W m1 °C1 is the thermal conductivity of the copper plate and a = 200 W m2 °C1 is the experimentally determined heat-transfer coefficient of the experimental setup used. At t = 8 min, the calculated freezing front velocities for the two cooling rates of 1 and 10 °C min1 were 4.3 and 26.8 lm s1, respectively. At this time, the temperature that the cold finger applied to the mould bottom was T0,1(8 min) = 8 °C and T0,10(8 min) = 80 °C, for the 1 and 10 °C min1 cooling rates, respectively. The reason for the freezing front velocity slowing with time and towards the top of the sample is that an applied, linear freezing

An important correlation for materials design is that between applied freezing rate, resulting freezing front velocity v, local cooling rate T_ and lamellar spacing d, which is the short axis of pores cut perpendicular to the freezing direction. This allows for a structural control at the first level of hierarchy of the highly porous composite scaffolds. Several authors have reported a power law correlation of the form d / vn between freezing front velocity, v (lm s1), and lamellar spacing, d (lm), where values for n vary between 0.25 and 1 [25,26]. Also, in our experiments, we consistently found that the lamellar spacing and the cell wall thickness decreased with increasing freezing velocity. When decreasing the freezing rate from 10 to 1 °C min1 and the freezing front velocity from 27.7 to 7.4 lm s1, the lamellar spacing increased, on average, from 26 to 35 lm (determined from SEM micrographs), which corresponds to an exponent of n  0.2. Furthermore, a slight decrease in the freezing front velocity along the height of the freezing mould was observed, resulting in a significant reduction of the lo_ as illustrated in Fig. 7. cal cooling rate T,

6344

P.M. Hunger et al. / Acta Biomaterialia 9 (2013) 6338–6348

Table 2 Mechanical properties of the alumina composites frozen at bottom plate cooling rates of 1 and 10 °C min1. Freezing rate

1 °C min1

Alumina particle size

Relative density

Young’s modulus (MPa)

Yield strength (MPa)

0.086 ± 0.001 0.092 ± 0.001 0.095 ± 0.002

0.62 ± 0.23 2.62 ± 0.75 9.63 ± 1.61 14.22 ± 4.36

0.036 ± 0.009 0.100 ± 0.018 0.244 ± 0.044 0.270 ± 0.075

– 400 nm 30:70, 400 nm:10 lm 10 lm

10 °C min1 Relative density

Young’s modulus (MPa)

Yield strength (MPa)

0.082 ± 0.001 0.091 ± 0.001 0.098 ± 0.001

1.01 ± 0.33 7.08 ± 3.53 20.05 ± 2.45 12.11 ± 2.41

0.057 ± 0.012 0.157 ± 0.061 0.346 ± 0.023 0.240 ± 0.017

Reprinted from Ref. [11] with permission from Elsevier.

surface area. If the particle size was close to the polymer wall thickness, particles were incorporated into the cell walls with a brick-and-mortar structure, while particles with a diameter far above the wall thickness were included as a monolayer in a beads-on-a-string-like structure (Fig. 4). The SEM micrographs of Fig. 4 also indicate the existence of nanoporosity within the cell walls; with the help of BET measurements, the existence of this second level porosity could be confirmed (Fig. 5). The cumulative pore volume found in the scaffolds including small particles is almost five times the volume found in the scaffolds with large particles. The cumulative pore volume of the scaffolds including bimodal particles lies between the two, but is closer to those with large particles. These different cell wall porosities are responsible, together with the lower polymer film thickness, for the inferior mechanical performance of the scaffolds with small particles and the higher mechanical properties of the scaffolds with large and bimodal particles. Fig. 7. Local cooling rates determined at each thermocouple position along the height of the mould.

In addition to the above changes in lamellar spacing, SEM measurements indicated an increase in pore aspect ratio with decreasing cooling rate and freezing front velocity, thus the increase in the pore depth is stronger than that of the pore width, which is often referred to as the ‘‘lamellar spacing’’. Hence, an increasing freezing rate results in a decreasing aspect ratio and thereby brings the highly aligned lamellar structure closer to a honeycomb-like architecture.

4.3. Correlations between particle size and cell wall structure At the second level of the hierarchy, we explored the differences in cell wall structure because we expected it to greatly affect the mechanical properties of both the cell wall material itself and the overall scaffold. Knowing the surface area for the small and the large particles from BET measurements to be 8.14 and 0.39 m2 g1, respectively, and calculating it according to the rule of mixture for the bimodal distribution to be 2.72 m2 g1, it was possible to estimate an average, uniform polymer film thickness for the case of perfectly coated particles. Values of 11, 223 and 32 nm were calculated for the small and large particles and the bimodal particle distribution, respectively. These considerable differences in polymer film thickness affect the quality and quantity of particle bonding as well as the cell wall porosity. Both the small particles and the bimodal particle distribution formed thick porous walls (2-6 lm) glued by the polymer phase, while for large particles the polymer walls were thinner (300–600 nm) and the particles aligned within the polymer phase with a structure that was reminiscent of ‘‘beads-on-a-string’’. These structures were partially formed due to particle size constraints on their packing, partly due to the ratio of pore size (ice crystal size) to particle diameter and partly due to the ratio of polymer phase volume to specific particle

4.4. Correlations between cell wall structure and mechanical properties The results of the compression tests in Table 2 and Fig. 6 reveal correlations between the cell wall structure and scaffold properties. Both the Young’s modulus and the yield strength of the composite increased with increasing particle size and decreasing specific surface area. At a freezing rate of 1 °C min1, the composite with the small particles had the lowest property values, followed by the bimodal particle distribution and then the large particles. For 10 °C min1, the small particles possessed the smallest mechanical values, while the large particles had medium and the bimodal particle size distribution had the highest mechanical properties. Thus, depending on the pore size, the greatest mechanical properties were found in the large particles (with a freezing rate of 1 °C min1) or in the bimodal particle size distribution (with a freezing rate of 10 °C min1) with specific surface areas on the same order of magnitude. This trend is plausible since it parallels the increase in average polymer film thickness and a decrease in cell wall porosity with a decrease in specific surface area. Both indicate that the polymer phase and its spatial distribution determine the composite’s performance. A similar observation was made when different layers in the freeze-cast samples were compared. The bottom layers of the large particles and the bimodal particle distribution, frozen at 1 °C min1, had lower property values than the middle layers of the samples (Fig. 8a). Their particle fractions were increased due to the sedimentation of large particles and with it the surface area that required polymer coating for effective bonding increased, too. This effect is most pronounced in the case of the large particles, because of faster sedimentation, and therefore increased the particle fraction of the composite more effectively. Comparing Fig. 8a and b, it is further apparent that the influence of sedimentation is only significant for the slower freezing rate, at which the particles have more time to sediment before becoming entrapped by the progressing freezing front.

P.M. Hunger et al. / Acta Biomaterialia 9 (2013) 6338–6348

6345

Fig. 8. Mechanical properties determined for the individual layers of the three different composites produced at the two freezing rates: (a) 1 °C min1 and (b) 10 °C min1. The arrows indicate the development of the mechanical properties with decreasing sample height and the simultaneous slight decrease in pore size.

4.5. Correlations between pore architecture and mechanical properties The above results show clear correlations between material structure and properties for the different particle sizes and distributions, at an identical overall amount of material in the slurry (and the freeze-cast scaffolds). In the case of the small and bimodal slurries, we found that the higher the freezing rate, the smaller the lamellar spacing and cell wall thickness, and the higher the Young’s modulus and compressive strength. The same trends were reported earlier for freeze-cast and sintered ceramics, in which faster freezing rates also resulted in smaller lamellar spacings and increased mechanical properties [6,13]. In contrast, in the case of the large particles, the mechanical properties were nearly unaffected by these structural changes. We hypothesise that the decreased lamellar spacing in combination with a reduced aspect ratio of the pores is an important structural factor that contributes to the increase in mechanical properties. This is because they bring the overall structure closer to that of a honeycomb with its improved mechanical efficiency [27]. The increased number of stiffening and strengthening bridges stabilise each lamella through rib stiffening and reduce failure due to interlamellar shear. Additionally, the smaller lamellar spacing reduces the possible lateral motion before adjacent lamellae interlock, support each other and thus strengthen the overall structure. While all these effects are very pronounced for cell walls that incorporate small particles, they are hardly noticeable in the scaffolds made with only large particles. Their performance appears to be highly dependent on the comparably low stiffness of the polymer membrane by which the particles are joined, until the particles start to interlock and support each other. 4.6. Correlations between cell wall structure and failure modes The different particle sizes revealed not only different structures and mechanical properties, but also significantly differing failure modes when tested in compression. This is illustrated by the two typical stress–strain curves shown in Fig. 1 and the significant differences between these materials in Young’s modulus, yield strength and post yield behaviour. The small particles exhibited little plasticity before failing in a saw-tooth fashion that is typical for brittle cellular materials; the large particles failed in an elastic–plastic fashion with a well-defined yield point and a smooth plateau. This difference in failure mode can be explained by the cell wall structure. The cell walls that are composed of small particles are porous composites; stresses are transferred from particle to particle through the polymer film by which they are joined. When the cell wall bends and buckles, high stresses are generated at the particle joints. As a consequence, a crack is started, which

can propagate with relative ease across the porous composite wall, and the cell wall fails in a brittle manner. In the case of large particles, the structure is not a true composite; rather, the large particles are encased and joined by a polymer film for load transfer. The polymer membrane is located close to the neutral axis of the lamella when bent, and the thickness of the polymer membrane is much thinner than the diameter of the ceramic particles, but, at 300–600 nm it is about an order of magnitude thicker than the polymer ‘‘glue’’ film in the composites made from small particles. As a result, the cell wall can deform until adjacent particles touch and interlock. This allows them to carry higher loads, requiring higher stresses to strain and fracture the cell wall material, thus resulting in the measured higher property values.

4.7. Correlations between structure and mechanical properties The Gibson–Ashby model for cellular materials states that, primarily, three structural features determine a material’s mechanical and physical properties: (i) the relative density of the material; (ii) the morphology of the pores; and (iii) the solid from which the cell walls are made [28]. Our freeze-cast materials are excellent examples that illustrate all three; they also show that, even for identical composition, meaning identical weight or volume fractions of particles and polymer, and identical relative densities, freeze-cast materials offer several pathways to tailor the material’s properties and performance. First, the relative density is affected by the slurry composition. The amount of polymer and ceramic in the slurry determines the overall amount of porosity within the material. Second, the pore morphology, such as pore size, pore aspect ratio and cell wall thickness, is affected by the slurry composition, and, very importantly, the freezing rate. In particular, the decrease in pore aspect ratio, which brings the structure closer to the architecture of a honeycomb, significantly increases the mechanical efficiency. Third, the cell wall solid and its properties depend greatly on material composition, particle size and size distributions, which also determine the effectiveness of the polymer binder. With increasing average particle size, at constant composition, for example, the porosity within the cell wall material was reduced, even though the overall density of the material was constant for all three composites. This means that the amount of nano- and mesoporosity within the cell wall material differs as well as the microporosity of the freeze-cast materials. These differences in cell wall material structures, cell wall densities and properties were found, as shown, to affect both the overall mechanical properties and the failure modes of the different scaffolds. By comparing experimental results with property estimates based on models for composites

6346

P.M. Hunger et al. / Acta Biomaterialia 9 (2013) 6338–6348

and cellular materials, we next explore whether the properties of the scaffolds fall into the expected range.

ES / EW qm rel;S

ð4Þ

For this calculation, the relative density of the lamellar scaffolds

qrel,S is required; this depends on the interlamellar porosity PIL as qrel,S = 1  PIL, where PIL = P  PW. With relative densities for small,

4.8. Comparison to mechanical models A three-step approach, illustrated in Fig. 9, was taken to estimate the modulus of the freeze-cast hybrid scaffolds. In the first step, the solid modulus of the cell wall composite EC was estimated according to the Reuss model using a literature value of 370 GPa [29] for the Young’s modulus of alumina and an experimental value of 1.5 GPa, determined by testing chitosan–gelatin thin films in tension in our laboratory. At an alumina volume fraction of 75%, the cell wall solid was calculated to have a modulus of 5.93 GPa. In the second step, the porosity within the cell wall material was taken into account using the Gibson–Ashby model for equiaxed foams, where the relative density of the walls qrel,W can be calculated from the wall porosity PW as qrel,W = 1  PW:

EW / EC q2rel;W

ð3Þ

Using the wall porosities of 3%, 5% and 14% determined by BET on scaffolds made with small, large and bimodal particles, the values for the cell wall moduli EW were calculated to be 4.4, 5.3 and 5.5 GPa, respectively (Table 3). In the third step, the scaffold modulus ES was calculated using the moduli for the porous cell wall materials and according to the Gibson–Ashby models for both the equiaxed foam (m = 2) and the honeycomb architectures (m = 1):

large and bimodal particle scaffolds of 0.096, 0.096 and 0.102, respectively, the scaffold moduli were calculated, in the case of equiaxed foams (lower bound), to be 3.6, 4.4 and 5.1 MPa, respectively, and, in the case of honeycombs (upper bound), 125, 153 and 168 MPa, respectively. See Table 3 for details. The comparison of the experimentally determined moduli with the estimated moduli revealed that the freeze-cast scaffolds fell short not only of the values predicted for honeycombs, but also of those predicted for equiaxed foams. There are several reasons for this, the most important one probably being that the freezecast hybrid scaffolds are not ideal but have imperfections at each level of their structural hierarchy. At the level of the cell wall material, for example, we likely underestimated the included wall porosity, because porosity >40 nm and closed porosity cannot be detected by BET and were therefore ignored. This porosity contributes quadratically to the modulus of the cell wall solid. At the next higher level, one imperfection is a slight misalignment of the cell walls with the loading direction. This results in instantaneous loading in bending, which leads to buckling, so that a structural modulus is measured rather than that of the material, which would initially be the case in a perfectly aligned honeycomb. Additionally, the cell walls are of varying thickness, and contain flaws and inhomogeneities.

Fig. 9. Schematic illustrating the three step approach (from left to right) to estimate the Young’s modulus of the scaffolds. (1) The composite material follows the Reussmodel for a particle-reinforced polymer. (2) The Gibson–Ashby-model for an equiaxed foam was used to account for the nanoporosity in the cell walls. (3) Because of their structure, the properties of the freeze-cast scaffold are expected to fall between the property limits of an equiaxed foam and a honeycomb-like material made from the nanoporous composite.

Table 3 Comparison of Young’s moduli computed with a three-step model approach and the experimentally determined moduli of the composite scaffolds. Scaffold type

Composite modulus EC, Reuss model (GPa)

Cell wall porosity PW (%)

Cell wall modulus, equiaxed foam EW (%)

Interlamellar porosity PIL (%)

Scaffold modulus ES, equiaxed foam (MPa)

Scaffold modulus ES, honeycomb (MPa)

Measured average scaffold modulus (MPa)

Small Bimodal Large

5.93 5.93 5.93

14 5 3

4.39 5.32 5.54

90.4 90.4 89.8

40 49 57

419 512 563

4.8 14.8 13.2

P.M. Hunger et al. / Acta Biomaterialia 9 (2013) 6338–6348

6347

4.9. Comparison of freeze-cast hybrid performance with scaffolds of similar composition Even if the freeze-cast hybrid scaffolds do not yet reach the properties that an ideal scaffold of the same composition could theoretically possess, they compare very favourably with others of similar composition described in the literature, all of which have considerably lower porosities. At an overall porosity of about 91% and a mineral content in the cell wall solid of 75 vol.%, the freeze-cast hybrid scaffolds with the best mechanical properties had a Young’s modulus of 20 MPa and a compressive yield strength of 0.35 MPa. In the absence of literature values for the mechanical properties of similar freeze-cast polymer–ceramic hybrid scaffolds—it appears that this is the first study on freeze-cast hybrids—we compared our materials with literature values for freeze-dried (nondirectionally frozen) composite foams that had been made with the aim of application as tissue scaffolds. Fig. 10 shows how these freeze-dried foams compare with the freeze-cast honeycombs of this study. For a porosity of 85%, Harley et al. [30] reported a Young’s modulus of 0.76 MPa and a compressive strength of 0.09 MPa for collagen–glycosaminoglycan–calcium phosphate scaffolds. At the same porosity of 85%, Kim et al. [31] achieved a modulus of 4 MPa with collagen–hydroxyapatite (HAp) scaffolds. Zhang and Zhang [32] produced scaffolds with a porosity of 87% from chitosan and b-tricalcium phosphate that had a compressive strength of 0.2–0.3 MPa and a modulus of up to 3 MPa. A similar material was produced by Li et al. [33], who produced 83% porous chitosan–HAp scaffolds which had a compressive strength of 0.1 MPa. The freeze-cast hybrids of this study exceed these values considerably; at a porosity that is 5–10% higher and a mineral content that is up to 40% higher or up to 20% lower, the freeze-cast hybrids’ moduli are about one order of magnitude greater and their yield strengths higher by a factor of 1.2. When the porosity and ceramic volume fraction of the freeze-cast scaffolds was reduced to 87% and 70%, respectively, to parallel the values of the freezedried foams, the modulus was increased by a factor of 2.5 to 50 MPa and their compressive yield strength increased by a factor of 3.4 to 1.2 MPa, respectively, as shown in Fig. 10. Both the experimental results and the property estimates using the Reuss and Gibson–Ashby models show that the mechanical

Fig. 10. Mechanical properties of the model materials of this study in comparison with literature values for materials of similar composition and porosity. Young’s modulus is shown as squares; triangles mark the yield strength. A considerable property range can be covered through the careful choice of particle size distribution and freezing rate. Decreasing the porosity from 92% to 87% while decreasing the scaffold’s ceramic volume fraction from 75% to 70% further increases both modulus and yield strength (orange symbols).

Fig. 11. Mechanical properties of a biocompatible chitosan–HAp scaffold freeze cast at 10 °C min1 in comparison to the alumina–chitosan–gelatin scaffolds made with different particle sizes. The mechanical properties of the biocompatible scaffolds are very similar to those of the alumina scaffolds made with a bimodal particle size distribution.

properties of the freeze-cast hybrid scaffolds are polymer-dominated and thus little affected by variations in ceramic filler material. The alumina particles of the model material could, therefore, be replaced by HAp, which has a Young’s modulus and a yield strength that are 4 times lower, with little effect on the scaffold’s mechanical properties. In fact, first results show that the mechanical properties of freeze-cast chitosan–HAp scaffolds made with a particle diameter of 2.3 lm fall, as expected, in the range of those of the alumina–chitosan–gelatin scaffold made from the bimodal particle size distribution (Fig. 11). 5. Conclusions This study revealed that highly porous (>90%) hybrid materials can be manufactured by freeze casting, through the self-assembly of a polymer and a ceramic phase that occurs during directional solidification alone, without the need of additional processing steps such as sintering or infiltration. It further illustrates that the properties of freeze-cast hybrid materials can independently be tailored at two levels of their structural hierarchy, allowing for the simultaneous optimisation of both mechanical and structural requirements. At the first hierarchical level, the overall porosity of the scaffold is determined by the amount of solid material in the starting slurry, and the lamellar spacing and pore aspect ratio are determined by the freezing rate. At a given composition, an increasing freezing rate results in increased mechanical properties through decreases in lamellar spacing and pore aspect ratio. At the second hierarchical level and with an identical composition, variations in the particle size determine the Young’s modulus, compressive yield strength and failure mode. Small particles result in a brittle failure of the material, whereas large particles and a bimodal particle distribution show an elastic–plastic response, with properties that exceeded the small-particle composite by nearly an order of magnitude. These results are thought to reflect the polymer ‘‘glue’’ thickness between the particles and cell wall porosities, a key factor that controls the mechanical performance of the hybrid scaffolds. The results of the work presented here further reveal that the directional solidification process creates honeycomb-like structures with properties that are considerably higher and compare favourably with those obtained by other processing techniques, such as non-directional freeze-drying. They also reveal that scaffold performance is dominated by the polymer phase, and that

6348

P.M. Hunger et al. / Acta Biomaterialia 9 (2013) 6338–6348

the mechanical properties of freeze-cast chitosan–HAp hybrids are almost identical to those made with the model material alumina. Because of the combination of excellent mechanical properties at unusually high porosities, freeze-cast hybrid scaffolds are particularly promising for biomedical applications, in which a mimic of the natural hybrid material bone, with a high porosity (on the nano- and microscales) and a composite structure of mineral particles embedded in a polymer matrix, is sought. Hence, freeze-cast hybrid materials possess tremendous potential in the field of tissue engineering since their porosity and structure, as well as their mechanical performance, amount of open nanoporosity, failure modes and composition, can be tailored to create custom-designed scaffolds for a given surgical site. Acknowledgements This work was supported by the NIH-NIDCR through Grant 5R01DE015633-05 and by the NSF through NSF-IGERT (A.E.D.). The authors wish to thank Thao Vi Le, Marjorie S. Austero, John McDonough and Boris Dyatkin for experimental assistance, and Dr. Anthony Lowman, Dr. Yury Gogotsi and Dr. Giuseppe R. Palmese (Drexel University) for the permission to use their facilities. The authors further wish to acknowledge the use of the Centralized Research Facilities in the College of Engineering at Drexel University. U.G.K.W. expresses her gratitude to Anne L. Stevens for the generous support of her research and group while at Drexel University. Appendix A. Figures with essential colour discrimination Certain figures in this article, particularly Figures 1–3, 5–11, are difficult to interpret in black and white. The full colour images can be found in the on-line version, at http://dx.doi.org/10.1016/ j.actbio.2013.01.012. References [1] Ezekwo G, Tong H-M, Gryte CC. On the mechanism of dewatering colloidal aqueous solutions by freeze–thaw processes. Water Res 1980;14:1079–88. [2] Tong HM, Noda I, Gryte CC. Formation of anisotropic ice–agar composites by directional freezing. Colloid Polym Sci 1984;262:589–95. [3] Tong HM, Gryte CC. Mechanism of lamellar spacing adjustment in directionally frozen agar gels. Colloid Polym Sci 1985;263:147–55. [4] Fukasawa T, Deng ZY, Ando M, Ohji T, Goto Y. Pore structure of porous ceramics synthesized from water-based slurry by freeze-dry process. J Mater Sci 2001;36:2523–7. [5] Fukasawa T, Deng Z-Y, Ando M, Ohji T, Kanzaki S. Synthesis of porous silicon nitride with unidirectionally aligned channels using freeze-drying process. J Am Ceram Soc 2002;85:2151–5. [6] Deville S, Saiz E, Tomsia AP. Freeze casting of hydroxyapatite scaffolds for bone tissue engineering. Biomaterials 2006;27:5480–9. [7] Fu Q, Rahaman MN, Dogan F, Bal BS. Freeze-cast hydroxyapatite scaffolds for bone tissue engineering applications. Biomed Mater 2008;3:1–7.

[8] Fu Q, Rahaman MN, Bal BS, Brown RF. In vitro cellular response to hydroxyapatite scaffolds with oriented pore architectures. Mater Sci Eng C 2009;29:2147–53. [9] Buehler MJ. Tu(r)ning weakness to strength. Nano Today 2010;5:379–83. [10] Munch E, Launey ME, Alsem DH, Saiz E, Tomsia AP, Ritchie RO. Tough, bioinspired hybrid materials. Science 2008;322:1516–20. [11] Hunger PM, Donius AE, Wegst UGK. Platelets self-assemble into porous nacre during freeze casting. J Mech Behav Biomed Mater. http://dx.doi.org/10.1016/ j.bbr.2011.03.031. [12] Fu Q, Rahaman MN, Dogan F, Bal BS. Freeze casting of porous hydroxyapatite scaffolds. I. Processing and general microstructure. J Biomed Mater Res B Appl Biomater 2008;86B:125–35. [13] Munch E, Saiz E, Tomsia AP, Deville S. Architectural control of freeze-cast ceramics through additives and templating. J Am Ceram Soc 2009;92:1534–9. [14] Jung H-D, Yook S-W, Kim H-E, Koh Y-H. Fabrication of titanium scaffolds with porosity and pore size gradients by sequential freeze casting. Mater Lett 2009;63:1545–7. [15] Schoof H, Apel J, Heschel I, Rau G. Control of pore structure and size in freezedried collagen sponges. J Biomed Mater Res 2001;58:352–7. [16] Moon J-W, Hwang H-J, Awano M, Maeda K. Preparation of NiO–YSZ tubular support with radially aligned pore channels. Mater Lett 2003;57:1428–34. [17] Mukai SR, Nishihara H, Tamon H. Formation of monolithic silica gel microhoneycombs (SMHs) using pseudosteady state growth of microstructural ice crystals. Chem Commun 2004:874–5. [18] Meghri N, Donius A, Riblett B, Martin E, Clyne A, Wegst U. Directionally solidified biopolymer scaffolds: mechanical properties and endothelial cell responses. JOM 2010;62:71–5. [19] Le TV. Stainless steel scaffolds by freeze casting for biomedical applications. Philadelphia, PA: Drexel University; 2010. [20] Wegst UGK, Schecter M, Donius AE, Hunger PM. Biomaterials by freeze casting. Philos Trans R Soc A 2010;368:2099–121. [21] Munch E, Franco J, Deville S, Hunger P, Saiz E, Tomsia A. Porous ceramic scaffolds with complex architectures. JOM 2008;60:54–8. [22] Chen R, Wang C-A, Huang Y, Ma L, Lin W. Ceramics with special porous structures fabricated by freeze-gelcasting: using tert-butyl alcohol as a template. J Am Ceram Soc 2007;90:3478–84. [23] Richardson JF, Zaki WN. The sedimentation of a suspension of uniform spheres under conditions of viscous flow. Chem Eng Sci 1954;3:65–73. [24] Baehr HD, Stephan K. Heat and mass transfer. 2 ed. Berlin: Springer; 2006. [25] Deville S, Saiz E, Tomsia AP. Ice-templated porous alumina structures. Acta Mater 2007;55:1965–74. [26] Zhang H, Hussain I, Brust M, Butler MF, Rannard SP, Cooper AI. Aligned twoand three-dimensional structures by directional freezing of polymers and nanoparticles. Nat Mater 2005;4:787–93. [27] Gibson LJ, Ashby MF. Cellular solids: structure & properties. 2nd ed. Oxford: Pergamon Press; 2001. [28] Ashby MF. The properties of foams and lattices. Philos Trans R Soc A 2006;364:15–30. [29] Shackelford JF, Alexander W. CRC materials science and engineering handbook. 3rd ed. Boca Raton, FL: CRC Press; 2000. [30] Harley BA, Lynn AK, Wissner-Gross Z, Bonfield W, Yannas IV, Gibson LJ. Design of a multiphase osteochondral scaffold. II. Fabrication of a mineralized collagen–glycosaminoglycan scaffold. J Biomed Mater Res A 2010;92A:1066–77. [31] Kim H-W, Knowles JC, Kim H-E. Hydroxyapatite and gelatin composite foams processed via novel freeze-drying and crosslinking for use as temporary hard tissue scaffolds. J Biomed Mater Res A 2005;72A:136–45. [32] Zhang Y, Zhang M. Synthesis and characterization of macroporous chitosan/ calcium phosphate composite scaffolds for tissue engineering. J Biomed Mater Res 2001;55:304–12. [33] Li H, Zhou CR, Tian JH, Rong JH. Preparation and characterization of homogeneous hydroxyapatite/chitosan composite scaffolds via in-situ hydration. J Biomater Nanobiotechnol 2010;1:42–9.