ARTICLE IN PRESS

Journal of Biomechanics 37 (2004) 763769

Short communication

Numerical simulation of tissue differentiation around loaded titanium implants in a bone chamber
L. Gerisa,*, A. Andreykivb, H. Van Oosterwycka, J. Vander Slotena, F. van Keulenb, J. Duyckc, I. Naertc
Division of Biomechanics and Engineering Design, Faculty of Engineering, Katholieke Univeriteit Leuven, Celestijnenlaan 200A, Leuven B-3000, Belgium b Faculty of Design, Engineering and Production, Delft University of Technology, Mekelweg 2, Delft 2628 CD, The Netherlands c Department of Prosthetic Dentistry, Faculty of Medicine, Katholieke Univeriteit Leuven, Kapucijnenvoer 7, Leuven B-3000, Belgium Accepted 18 September 2003
a

Abstract The application of a bone chamber provides a controlled environment for the study of tissue differentiation and bone adaptation. The inuence of different mechanical and biological factors on the processes can be measured experimentally. The goal of the present work is to numerically model the process of peri-implant tissue differentiation inside a bone chamber, placed in a rabbit tibia. 2D and 3D models were created of the tissue inside the chamber. A number of loading conditions, corresponding to those applied in the rabbit experiments, were simulated. Fluid velocity and maximal distortional strain were considered as the stimuli that guide the differentiation process of mesenchymal cells into broblasts, chondrocytes and osteoblasts. Mesenchymal cells migrate through the chamber from the perforations in the chamber wall. This process is modelled by the diffusion equation. The predicted tissue phenotypes as well as the process of tissue ingrowth into the chamber show a qualitative agreement with the results of the rabbit experiments. Due to the limited number of animal experiments (four) and the observed inter-animal differences, no quantitative comparison could be made. These results however are a strong indication of the feasibility of the implemented theory to predict the mechano-regulation of the differentiation process inside the bone chamber. r 2003 Elsevier Ltd. All rights reserved.
Keywords: Tissue differentiation; Bone chamber; Titanium implant; Finite element method; Numerical simulation

1. Introduction The last decades, many researchers tried to determine the mechanical and biological parameters that inuence the process of bone differentiation, either by means of animal experiments (Moalli et al., 2000; S^balle et al., 1992a, b; S^balle, 1993) or by means of numerical modelling (Prendergast et al., 1997; Huiskes et al., 1997; Maheshwari and Lauffenburger, 1998; Kuiper et al., ! 2000; Bailon-Plaza and van der Meulen, 2001). Prendergast and van der Meulen (2001) give an overview of the different approaches used in the numerical modelling of bone differentiation. These approaches differ from each other in the way they describe the biological and mechanical processes that underlie the differentia*Corresponding author. Tel.: +32-16-327534; fax: +32-16-327994. E-mail address: liesbet.geris@mech.kuleuven.ac.be (L. Geris). 0021-9290/$ - see front matter r 2003 Elsevier Ltd. All rights reserved. doi:10.1016/j.jbiomech.2003.09.026

tion process, and in the parameters they use to describe these underlying processes. Despite these efforts, questions remain with respect to the actual process of bone differentiation and the parameters that inuence this process. An important factor in this research is the interpretation of animal experimental results. This can be hampered by the fact that the bone adaptive response seems to be site-specic and species-specic (Carter, 1984; Bertram and Swartz, 1991). Factors like nutrition, hormonal inuences and genetic factors can inuence bone formation and resorption rates and therefore can interfere with the inquired bone response. As a consequence, this makes the validation of mathematical theories of bone differentiation more difcult. During the entire process of developing a mathematical model and creating a numerical model with the assignment of material properties and boundary conditions, there is a

ARTICLE IN PRESS
764 L. Geris et al. / Journal of Biomechanics 37 (2004) 763769

need for the validation of the hypotheses and assumptions that were made and the role they play in the nal result. This shows that the integration of experimental and mathematical models is a critical issue in gaining insight in the process of bone differentiation. Experiments provide insights and measurements, which can then be interpreted within the context of analytical frameworks. Analytical simulations permit investigation of possible explanations that require in vivo validation and will suggest further experimental investigations (van der Meulen and Huiskes, 2002). A repeated sampling bone chamber (Fig. 1) was developed. The bone chamber is a perforated hollow cylinder that is implanted in the proximal tibiae of adult rabbits. Data derived from literature (Tagil and Aspenberg, 1999; Moalli et al., 2000) and from our own pilot studies indicate that such a hollow cylinder functions as a bone chamber because of the bone ingrowth via the perforations. A test implant is positioned in the center of the bone chamber and can be loaded in a well-controlled

manner by means of an external loading device. The combination of an inner and outer chamber makes it possible to harvest the bone after each experiment without removing the entire chamber. This eliminates variability due to site- and species-specic inuences, since the different experiments can be conducted within the same site of the same animal. Moreover, because of the protected environment of the chamber, any external mechanical inuence can be excluded. So far, pilot studies with this new chamber design were performed in four rabbits (two unloaded controls, two loaded). The goal of the present study is twofold. (1) The numerical modelling of the process of tissue differentiation inside the bone chamber. (2) The assessment of the feasibility of the mathematical model for the simulation of mechanically induced tissue differentiation developed by Prendergast et al. (1997) and Huiskes et al. (1997) in situations besides fracture healing.

Fig. 1. Composition drawing of the bone chamber (upper). Picture of the bone chamber with all its separate parts (under). After insertion of the outer bone chamber (1outer diameter=1 cm), there is a healing period of 6 weeks to assure a good xation of the outer chamber. During this period the Teon cylinder (5) inhibits bone ingrowth. After 6 weeks the inner bone chamber (3) together with the Teon bearing (2) and the test implant (4) are implanted and the experiment can start.

ARTICLE IN PRESS
L. Geris et al. / Journal of Biomechanics 37 (2004) 763769 765

This second goal is accomplished by comparison of the simulation results with the histological results of the pilot studies (loaded samples).

2. Methods Like in the studies of Prendergast et al. (1997) a poroelastic (biphasic) approach was followed to model the constitutive behavior of the differentiating tissues. Both 2D and 3D poroelastic nite element models were created. Instead of using a model of the entire bone chamber and its environment, models representing only the tissue inside the bone chamber were used. The connections of the tissue with the walls of the bone chamber were modelled by imposing appropriate boundary conditions. A rigid xation between the tissue and the inner bone chamber and between the tissue and the implant was assumed. At the perforation sites, free uid ow was modelled. Calculations with a model of the entire bone chamber, including the tissue surrounding the bone chamber gave similar results as the calculations performed with a model of only the tissue inside the bone chamber. For the 2D model (Fig. 2) the assumption of axisymmetry was madewhile in reality the chamber geometry has only a 120 rotational symmetryin the 3D model (Fig. 3) the geometry of the bottom part of the chamber was simplied. Taking into account the symmetry of the displacement and uid velocity elds within the chamber, only one-sixth of the

Fig. 3. 3D Finite element model. Because of symmetry, a model of one-sixth of the tissue in the chamber is used in the simulations (highlighted). The connections of the tissue with the chamber are modelled by applying appropriate displacement (u) and pressure (p) boundary conditions (yy: ur uy uz 0; - - - - -: uz 50=160 mm, - - - - -: uy 0; mmmmm: p 0 (free uid outow)).

Table 1 Material properties of the tissues used in the simulations (Lacroix et al., 2002) Granulation tissue Youngs modulus (MPa) Poissons ratio permeability (m4(N s)1) 1 0.17 1014 Cartilage 10 0.17 5 1015 Bone 1000 0.3 1013

Fig. 2. 2D Finite element models. Model of the entire bone chamber (upper): titanium inner and outer bone chamber (1), tissue (2), Teon bearing (3) and titanium implant (4). For the simulations the model of only the tissue inside the bone chamber is used (lower). The connections of the tissue with the chamber are modelled by applying appropriate displacement (u) and pressure (p) boundary conditions (yy: ur uz 0; - - - - -: uz 50=160 mm, - - - - -: ur 0; mmmmm: p 0 (free uid outow)).

entire tissue in the chamber must be modelled in the 3D model. The used elements were biphasic 8-noded axisymmetric quadrilateral elements and biphasic 20-noded bricks, respectively. At the beginning of the simulation the entire chamber was assumed to be lled with granulation tissue. The used material properties are summarized in Table 1. The loading of the implant took place in a displacement-controlled manner. This was modelled by imposing displacement boundary conditions on the nodes at the interface between the implant

ARTICLE IN PRESS
766 L. Geris et al. / Journal of Biomechanics 37 (2004) 763769

and the tissue. Two types of loading regimes were applied: a sine with a frequency of 1 Hz and an amplitude of 50 and 160 mm, respectively. These loading regimes were applied once every three and a half days for 4 weeks. The loading conditions correspond to those applied in the conducted or planned experiments (800 cycles, twice a week). For the simulation only one cycle of a sine wave per iteration step was applied (Lacroix et al., 2002). All analyses were performed using MSC Marc/Mentat (Version 2001, Palo Alto, USA) using its soil mechanics capabilities. The FE code assumes incompressibility of the solid and uid constituents. A previous study of Prendergast et al. (1996) has demonstrated the validity of this code with respect to the biphasic theory, developed by Mow et al. (1980). For the simulation of the process of tissue differentiation, the model of Prendergast et al. (1997) and Huiskes et al. (1997) was implemented. From the FE model the maximal tissue distortional strain and relative uid velocity were calculated. Depending on these stimuli a tissue phenotype for each element was predicted. High levels of these stimuli promote the differentiation of mesenchymal cells into broblasts, intermediate levels stimulate the differentiation into chondrocytes and low levels of these stimuli promote the differentiation into osteoblasts. The change in phenotype of an element is implemented through a change in material properties (modulus of elasticity, Poisson ratio and permeability). In order to avoid numerical instabilities caused by too large changes in the material properties between consecutive increments, a numerical smoothing procedure was implemented (Lacroix and Prendergast, 2000). The material properties will change gradually towards the phenotype determined by the stimulus, by distributing the change over a number of increments. The process of bone resorption is not included in the model since the simulation is only focused on the process of tissue differentiation and not on the remodelling processes afterwards. In the process of tissue differentiation, the presence of mesenchymal cells is indispensable. The mesenchymal cells can enter the bone chamber through the holes in the side and the bottom, and start proliferating and differentiating. The diffusion process of the cells in the chamber was modelled by the following equation: Dr2 n dn : dt 1

Fig. 4. Simulation scheme. Initially the entire bone chamber is lled with granulation tissue and there are no mesenchymal cells present inside the bone chamber. The diffusion of mesenchymal cells via the perforations in the walls throughout the bone chamber is calculated. After application of the loading, mesenchymal cell concentration is read and biophysical stimuli are calculated. Tissue properties are determined for each integration point using the diagram presented at the bottom-right of the gure, a rule of mixtures and a smoothing procedure.

saturated with mesenchymal cells. This was modelled by including a constant source of mesenchymal cells (constant cell density) at the perforation sites of the bone chamber. The results of the diffusion analysis were read before every loading cycle and based on the mesenchymal cell density in an element, a rule of mixtures was applied to calculate the average material properties for this element (Lacroix et al., 2000). In this way, it is possible for two phenotypes to exist within one element. The entire modelling scheme is shown in Fig. 4.

3. Results The situation in the bone chamber after 4 weeks of loading for both of the loading conditions is shown in Fig. 5. Under the rst loading condition with an amplitude of 50 mm, the granulation tissue was able to differentiate into bone in a large volume of the chamber. At the interface between the tissue and the implant, the conditions were such that the formation of cartilage was favored. At the bottom of the chamber, high strains and uid velocity inhibited any tissue differentiation beyond brous tissue. These results correspond largely with the observation of the rabbit experiments. Also the same ingrowth patterns of the bone could be seen in the simulations as in the experiments (Fig. 6). The second loading condition (amplitude of 160 mm) has not yet been experimentally investigated. The simulations predict that mainly because of the higher values for the

In this equation n represents the cell density, t is the time and D is the diffusion coefcient. This last constant was chosen in such a way that after 4 weeks a uniform mesenchymal cell density is established throughout the entire chamber. In practice, the diffusion was modelled in a separate analysis. The initial mesenchymal cell concentration in the entire bone chamber is set to zero. The tissue outside the bone chamber was assumed to be

ARTICLE IN PRESS
L. Geris et al. / Journal of Biomechanics 37 (2004) 763769 767

Fig. 5. Tissue types inside the bone chamber after the simulated time for the loading regimes with a displacement amplitude of 50 mm (upper) and a displacement amplitude of 160 mm (lower). The results for the simulations in 2D (left) and 3D (right) are shown.

maximal distortional strain, the tissue will only differentiate into bone in the perforations of the chamber. At the interface of the tissue with the implant, the favored phenotype is brous tissue. About 200 mm from this interface, cartilage will be formed. Bone formation can take place via endochondral or intramembranous bone formation. In these simulations it mostly happened via intramembranous bone formation. Only at some places in the bone chamber endochondral bone formation was observed. Verication of the stimulus for bone resorption demonstrated that the conditions for resorption were never met during the entire process, so no resorption would have taken place. A sensitivity analysis was performed in order to determine the relative importance of the model parameters (uid velocity and maximal distortional strain) and their inuence on the predicted differentiation patterns. Fig. 7 shows the contributions of the maximal tissue distortional strain and the relative uid ow to the value of the stimulus. The inuence of the relative uid ow on the differentiation patterns is only perceptible in the bottom part of the bone chamber. Results of simulations ignoring the contribution of the relative uid velocity (not shown) differ slightly from the results

Fig. 6. Progression of tissue ingrowth in the bone chamber. The upper gures indicate the location in the chamber for which (numerical and experimental) results are shown. Comparison of results of the rabbit experiments (histological sections, stained with Stevenels blue and Von Giesons picrofuchsin red, middle) and the numerical simulation (lower). The histological sections are from two different animals after a period of 4 weeks. The results of the simulations show the situation after 1 week (left) and 4 weeks (right). The tissue inside the perforations of the outer bone chamber is not present in the histological sections. Since it cannot be harvested together with the inner chamber, it is removed separately and not included in the histology. The simulation model however includes the tissue inside the perforations of the outer chamber.

in Fig. 5 at the bottom of the chamber and at the interfaces between the different phenotypes.

4. Discussion To simulate the process of tissue differentiation inside the bone chamber, a number of simplications and assumptions were made. First, both the 2D and the 3D FE model are a simplied reection of the reality. For the 2D model axisymmetry was assumed. For the 3D model the bottom of the bone chamber was not accurately modelled. The latter simplication is based on observations made during the animal experiments. Since no tissue was present in the hole in the bottom of the bone chamber during the entire duration of the experiments, this simplication is justied. When comparing the results of the 2D and 3D simulations, a strong resemblance concerning predicted tissue phenotype, maximal distortional strain and uid ow is seen near the implant interface. These results suggest that a

ARTICLE IN PRESS
768 L. Geris et al. / Journal of Biomechanics 37 (2004) 763769

Fig. 7. Sensitivity analysis to determine the inuence of the relative uid velocity in the entire differentiation process. The upper part of the gure shows the contributions of the two parameters to the total stimulus for a 2D analysis with a 50 mm displacement loading regime. The stimulus is predominantly determined by the value of the maximal distortional strain. In the simulation results (not shown), the inuence of the relative uid velocity is perceptible in the bottom half of the chamber as well as at the interfaces of different phenotypes.

2D model is sufcient for the prediction of tissue differentiation at the implant interface. However, differences arise near the chamber walls, where only the 3D model succeeds in correctly predicting the nal tissue phenotype. Therefore, a 3D model is needed in order to capture the correct ingrowth pattern of the tissue during the experiment. Second, to model the tissue differentiation the mechano-regulation model of Prendergast et al. (1997) was used with the threshold values determined by Huiskes et al. (1997). The threshold values were determined from the experiments conducted by S^balle et al. (1992a) on dogs. These thresholds might be species-specic, so that for rabbits different values should have been used. For the modelling of the mesenchymal cell diffusion, it was assumed that the initial cell concentration inside the bone chamber is zero and that the tissue outside the bone chamber is saturated with cells. Furthermore, it was assumed that the mesenchymal cells will spread homogeneously throughout the bone chamber in a period of 4 weeks. Cell mitosis, cell removal, cell death and the inuence of growth factors on the differentiation process (Mahesh! wari and Lauffenburger, 1998; Bailon-Plaza and van der Meulen, 2001) were not yet modelled in these simulations. Further investigations into this matter will be made by our group in the near future. Despite these limitations, the model does successfully predict several features of the tissue differentiation process inside the bone chamber. There is a qualitative resemblance in the process predicted by the simulations and observed in the rabbit experiments (Fig. 6). The

predicted tissue phenotypes (Fig. 5) correspond largely to those observed in the experiments. Due to the limited number of animals (four: two loaded, two unloaded) for this rst series of experiments and the observed interanimal differences, only a qualitative comparison is possible. The experimental result for rabbit 1 after 4 weeks (Fig. 6, left) shows a qualitative resemblance with the numerical result after a simulated period of 1 week. This might suggest that the differentiation process in the chamber of this animal was not yet as advanced as in the other animal at the time of harvesting. These interanimal differences could be explained by factors as initial growth factor release and initial mesenchymal cell concentration (Aspenberg et al., 1996; Linkhart et al., 1993). Since these factors are not yet included in the mathematical model, this hypothesis could not be tested. The sensitivity analysis performed on the parameters of the mechano-regulation model, showed that the inuence of the relative uid ow on the predicted differentiation patterns is only perceptible in the bottom part of the bone chamber and at the interfaces between the different phenotypes. This raises the question whether other mathematical models like those developed by Claes and Heigele (1999) and Carter et al. (1998), which do not use biphasic modelling, are also capable of providing a good estimate of the events inside the bone chamber. This question will be addressed by means of further simulations and animal experiments. Among others, a second type of bone chamber will give us more information about the inuence of uid ow on the differentiation process. In this bone chamber, the loading will take place through compression rather than shear. We expect this will cause a much more substantial uid ow inside the bone chamber. The next steps in this research will be the extension of the differentiation model. The present model only takes into account the concentration of mesenchymal stem cells and its rate of change is only dependent on diffusion. Moreover, it is assumed here that the differentiation process can be entirely described by only taking into account mechanical inuences, while in reality initial growth factor release plays an important ! role as well (Bailon-Plaza and van der Meulen, 2001). The development of more mechanistic models that integrate the biological complexity of cell proliferation, differentiation and apoptosis and their interaction with mechanical loading, constitutes a major task for the future.

5. Conclusion The presented study is a rst step in the simulation of the process of peri-implant tissue differentiation inside a bone chamber. The simulation with the presented FE models and mathematical model (Prendergast et al.,

ARTICLE IN PRESS
L. Geris et al. / Journal of Biomechanics 37 (2004) 763769 769

1997; Huiskes et al., 1997) succeeds in predicting this process in a qualitative way. The results demonstrate the capability of the mathematical model to simulate the process of tissue differentiation in another situation and for another animal species than the one it was originally derived from.

Acknowledgements This study was supported by the Research Council K.U. Leuven. Liesbet Geris is an aspirant of the Fund for Scientic Research Flanders, Hans Van Oosterwyck and Joke Duyck are postdoctoral fellows of the Fund for Scientic Research Flanders. The authors wish to thank Prof. P.J. Prendergast, Department of Mechanical Engineering at Trinity College in Dublin for his scientic advice.

References
. Aspenberg, P., Jeppsson, C., Wang, J.S., Bostrom, M., 1996. Transforming growth factor beta and bone morphongenetic protein 2 for bone ingrowth: a comparison using bone chambers in rats. Bone 19 (5), 499503. ! Bailon-Plaza, A., van der Meulen, M.C.H., 2001. A mathematical framework to study the effects of growth factor inuences on fracture healing. Journal of Theoretical Biology 212, 191209. Bertram, J.E.A., Swartz, M.S., 1991. The law of bone transformation: a case of Crying Wolff? Biology Reviews 66, 245273. Carter, D.R., 1984. Mechanical loading histories and cortical bone remodeling. Calcied Tissue International 36 (Suppl. 1), S1924. Carter, D.R., Helms, J.A., Tay, B.K., Lotz, J.C., Beaupr, G.S., 1998. Stress and strain distributions predict tissue differentiation patterns in distraction osteogenesis. Transactions of the Orthopaedic Research Society 23, 234. Claes, L.E., Heigele, C.A., 1999. Magnitudes of local stress and strain along bony surfaces predict the course and type of fracture healing. Journal of Biomechanics 32 (3), 255266. Huiskes, R., Van Driel, W.D., Prendergast, P.J., S^balle, K., 1997. A biomechanical regulatory model for periprosthetic brous-tissue differentiation. Journal of Material Science: Materials in Medicine 8, 785788. Kuiper, J.H., Ashton, B.A., Richardson, J.B., 2000. Computer simulation of fracture callus formation and stiffness restoration. In: Proceedings of the 12th Conference of the ESB. European Society of Biomechanics, Dublin.

Lacroix, D., Prendergast, P.J., 2000. A homogenization procedure to prevent numerical instabilities in poroelastic tissue differentiation models. In Proceedings of the Eighth Annual Symposium: Computational Methods in Orthopaedic Biomechanics. Orthopaedic Research Society, Orlando, FL. Lacroix, D., Prendergast, P.J., Li, G., Marsh, D., 2000. A 3D nite element model of a tibia to simulate the regenerative and resorption phases during fracture healing. In: Proceedings of the 12th conference of the ESB. European Society of Biomechanics, Dublin. Lacroix, D., Prendergast, P.J., Li, G., Marsh, D., 2002. Biomechanical model to simulate tissue differentiation and bone regeneration: application to fracture healing. Medical and Biological Engineering and Computing 40, 1421. Linkhart, T.A., Mohan, S., Baylink, D.J., 1993. Growth factors for bone growth and repair: IGF, TGFb and BMP. Bone 19 (Suppl. 1), S112. Maheshwari, G., Lauffenburger, D.A., 1998. Deconstructing (and reconstructing) cell migration. Microscopy Research and Technique 43, 358368. Moalli, M.R., Caldwell, N.J., Patil, P.V., Goldstein, S.A., 2000. An in vivo model for investigations of mechanical signal transduction in trabecular bone. Journal of Bone and Mineral Research 15, 13461353. Mow, V.C., Kuei, S.C., Lai, W.M., Armstrong, C.G., 1980. Biphasic creep and stress relaxation of articular cartilage in compression: theory and experiments. Journal of Biomechanical Engineering 102, 7384. Prendergast, P.J., van der Meulen, M.C.H., 2001. Mechanics of bone regeneration. In: Cowin, S.C. (Ed.), Bone Mechanics Handbook. CRC Press LCC, Boca Raton, FL, pp. 32.132.13. Prendergast, P.J., Van Driel, W.D., Kuiper, J.-H., 1996. A comparison of nite element codes for the solution of biphasic poroelastic problems. Proceedings of the institution of mechanical engineers. Journal of Engineering in Medicine, Part H 210, 131136. Prendergast, P.J., Huiskes, R., S^balle, K., 1997. Biophysical stimuli on cells during tissue differentiation at implant interfaces. Journal of Biomechanics 30, 539548. S^balle, K., 1993. Hydroxyapatite ceramic coating for bone implant xation. Acta Orthopaedica Scandinavia 64 (S255), 148. S^balle, K., Rasmussen, H.B., Hansen, E.S., Bunger, C., 1992a. . Hydroxyapatite coating modies implant membrane formation (controlled micromotion in dogs). Acta Orthopaedica Scandinavia 63 (2), 128140. S^balle, K., Hansen, E.S., Rasmussen, H.B., Jorgensen, P.H., Bunger, . C., 1992b. Tissue ingrowth into titanium and hydroxyapatitecoated implants during stable and unstable mechanical conditions. Journal of Orthopaedic Research 10 (2), 285299. Tagil, M., Aspenberg, P., 1999. Cartilage induction by controlled mechanical stimulation in vivo. Journal of Orthopaedic Research 17 (2), 200204. van der Meulen, M.C.H., Huiskes, R., 2002. Why mechanobiology? Journal of Biomechanics 35, 401414.