Engineering anatomically shaped human bone graftsWarren L. Graysona, Mirjam Fröhlicha,b, Keith Yeagera, Sarindr Bhumiratanaa, M. Ete Chanc, Christopher Cannizzarod,Leo Q. Wana, X. Sherry Liuc, X. Edward Guoc, and Gordana Vunjak-Novakovica,1

aLaboratory for Stem Cells and Tissue Engineering, Department of Biomedical Engineering, Columbia University, 622 West 168th Street, VC 12-234,New York, NY 10032; bEducell Ltd., Letaliska 33, 1000 Ljubljana, Slovenia; cBone Bioengineering Laboratory, Department of Biomedical Engineering,Columbia University, 351 Engineering Terrace, 1210 Amsterdam Avenue, MC 8904, New York, NY 12019; and dDepartment of Biomedical Engineering,Tufts University, 4 Colby Street, Medford, MA 02155

Edited by Stephen F. Badylak, University of Pittsburgh Medical Center, Pittsburgh, PA, and accepted by the Editorial Board September 1, 2009(received for review May 20, 2009)

The ability to engineer anatomically correct pieces of viable andfunctional human bonewould have tremendous potential for bonereconstructions after congenital defects, cancer resections, andtrauma. We report that clinically sized, anatomically shaped, viablehuman bone grafts can be engineered by using human mesenchy-mal stem cells (hMSCs) and a “biomimetic” scaffold-bioreactorsystem. We selected the temporomandibular joint (TMJ) condylarbone as our tissue model, because of its clinical importance and thechallenges associatedwith its complex shape. Anatomically shapedscaffolds were generated from fully decellularized trabecular boneby using digitized clinical images, seededwith hMSCs, and culturedwith interstitial flow of culture medium. A bioreactor with achamber in the exact shape of a human TMJ was designed forcontrollable perfusion throughout the engineered construct. By 5weeks of cultivation, tissue growth was evidenced by the forma-tion of confluent layers of lamellar bone (by scanning electronmicroscopy), markedly increased volume of mineralized matrix (byquantitative microcomputer tomography), and the formation ofosteoids (histologically). Within bone grafts of this size and com-plexity cells were fully viable at a physiologic density, likely animportant factor of graft function. Moreover, the density andarchitecture of bone matrix correlated with the intensity andpattern of the interstitial flow, as determined in experimental andmodeling studies. This approach has potential to overcome acritical hurdle—in vitro cultivation of viable bone grafts of complexgeometries—to provide patient-specific bone grafts for craniofa-cial and orthopedic reconstructions.

biomimetic | bioreactor | craniofacial regeneration | mesenchymal stemcells | temporomandibular joint | tissue engineering

Bone reconstructions often involve autologous tissue grafting,a method limited by harvesting difficulties, donor site mor-

bidity, and the clinicians’ ability to contour delicate 3D shapes.The availability of personalized bone grafts engineered from thepatient’s own stem cells would revolutionize the way we currentlytreat these defects. A “biomimetic” approach using stem cells,regulatory factors, and the appropriate scaffolds (1) in a wayinspired by native bone development and regeneration is re-quired to guide cell differentiation and assembly into the desir-able tissue phenotypes (2).The functionality of engineered bone grafts is evaluated

primarily by their mechanical properties and the ability of cellsto make tissue-specific proteins. Craniofacial bone grafts areunique in that their functionality is also linked to their overallgeometry. Many studies have thus focused on the ability torecapitulate the anatomical shape of cell-based constructs invitro (3–7). Bone grafts of the highest utility for reconstructivesurgery would be based on “designer scaffolds” shaped into grossgeometries specific to the patient and defect being treated (1,8–11). The temporomandibular joint (TMJ) condyle has beenwidely studied as a tissue-engineering model because it cannotbe generated easily, if at all, by current methods. The TMJcondyle is smaller than other articular joint condyles and is of

great clinical relevance, with ≈25% of the population exhibitingsymptoms of TMJ disorders (12, 13).The earliest reported cell-based TMJ grafts were formed by

seeding mature osteoblasts into polymer scaffolds, followed by“painting” the articulating surface with chondrocytes (6). Froma clinical perspective, mesenchymal stem cells (MSCs) are bettersuited than differentiated cells for use in cranial and maxillofa-cial applications (14, 15) owing to their easier accessibility,capability for in vitro proliferation, and potential for formingcartilage, bone, adipose, and vascular tissues (16). Subsequentstudies used MSCs that were predifferentiated along chondro-genic and osteogenic lineages and incorporated into photopo-lymerizable hydrogels (3–5). These studies demonstrated theformation of distinct osseous and cartilaginous regions, but thecell viability could only be maintained when cell-seeded scaffoldswere implanted in mice (3–6).In vitro control of cell viability and tissue development in

clinically sized and shaped bone tissue constructs fundamentallydetermines their utility for regenerative medicine, which at thistime remains a challenge. The enhancement of mass transportand the generation of hydrodynamic shear, which are criticallyimportant for bone development and function, depend on in-terstitial flow (17, 18). Rotating bioreactors used to cultivateporcine TMJ constructs provided convective flow around graftsurfaces but not in the graft interior (7). Bioreactors withperfusion through the cultured constructs have been preferredfor engineering bone (19–22), because they can provide micro-environmental control and biophysical stimulation of the cells inlarge constructs.We developed a tissue engineering approach for creating in

vitro an entire bone condyle containing viable cells at physio-logic density and well-developed bone matrix. Human MSCs(hMSCs) were induced to form bone on a decellularized bonescaffold that had the exact geometry of the TMJ, using an“anatomical” bioreactor with control of interstitial flow. Flowpatterns associated with the complex geometry of the bone graftprovided unique opportunity to correlate the architecture of theforming a bone with interstitial flow characteristics, undercontrollable in vitro conditions. We expect that this approachcan help provide a variety of anatomically shaped bone graftsdesigned to meet the needs of a specific patient and a specificcraniofacial or orthopaedic reconstruction.

ResultsBioreactor Cultivation. The anatomical shape of the TMJ wasdefined from digitized medical images and faithfully recapitu-

Author contributions:W.L.G. andG.V.-N. designed research;W.L.G., M.F., K.Y., S.B., M.E.C.,C.C., L.Q.W., and X.S.L. performed research; W.L.G., M.F., S.B., L.Q.W., X.E.G., and G.V.-N.analyzed data; and W.L.G., K.Y., and G.V.-N. wrote the paper.

The authors declare no conflict of interest.

This article is a PNASDirect Submission. S.F.B. is a guest editor invited by the Editorial Board.1To whom correspondence should be addressed. E-mail: gv2131@columbia.edu.

This article contains supporting information online at www.pnas.org/cgi/content/full/0905439106/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.0905439106 PNAS | February 23, 2010 | vol. 107 | no. 8 | 3299–3304

MED

ICALSC

IENCE

SSP

ECIALFEATU

RE

Dow

nloa

ded

by g

uest

on

Mar

ch 2

4, 2

020

lated by machining decellularized trabecular bone scaffold (Fig.1). This way, the cells were cultured in a scaffold that had thestructural, biochemical, and mechanical properties of nativebone (through the use of fully decellularized bone), and theactual geometry of the final graft (through image-guided fabri-cation). The void volume of decellularized bone determined bymicrocomputer tomography (μCT) analysis was >80%. SEMand histological analysis revealed pore sizes of ≈1 mm. Thesestructural features enabled efficient and spatially uniform dy-namic seeding of hMSCs into the scaffolds. Histological evalu-ation of freshly seeded scaffolds demonstrated that hMSCs linedthe internal pore walls, while leaving the pore spaces unob-structed (Fig. S1). After 1 h, the seeding efficiency was 34.0 ±7.1%, resulting in ≈3.4 × 106 cells per construct attaching in aspatially uniform manner. Constructs were cultured statically for1 week before placing in the bioreactors, enabling firm cellattachment and deposition of extracellular matrix before theexposure to hydrodynamic shear forces.Quick assembly of the perfusion units (Fig. 1 and Movie S1)

under sterile conditions enabled the maintenance of cell viabilitythroughout the process. Medium was perfused through theentire scaffold. For each flow rate, equal flow rates through thethree outlets were obtained by adjusting the pressure clamps onthe outlet tubing. The effects of flow on tissue development wereinvestigated at three flow rates: 0.4, 0.8, and 1.8 mL/min. Basedon the histological evaluation of cell density and distributionafter 5 weeks of culture, the highest flow rate of 1.8 mL/min wasselected for further studies, because it yielded the best celldistribution and most rapid tissue development (Fig. S2). Basedon the average cross-sectional area through the scaffolds in thedirection of flow that was determined by μCT analysis as ≈0.5

cm2, this flow rate corresponded to an average superficial flowvelocity of ≈0.06 cm/s.

Tissue Development and Mineral Deposition. Cells proliferated ex-tensively over the first week of culture, as evidenced by an≈7.5-fold increase in DNA content. The DNA content contin-ued to increase throughout the cultivation period under bothstatic (4.5-fold increase) and perfused (10-fold increase) cultureconditions, resulting in overall 37- and 75-fold increases, respec-tively, in cell numbers relative to initial seeding values (Fig. 2).These large increases in cell numbers with time and cultureregimen were corroborated with imaging data.Constructs cultivated under static conditions formed new

matrix primarily at the periphery (Fig. 3A), whereas bioreactor-grown constructs displayed new tissue growth throughout theirvolumes (Fig. 3E). Histological sections demonstrated starkcontrast in cell growth and osteoid formation patterns in thecentral regions between the two culture groups (Fig. 3 B and F).The new osteoid area normalized to existing bone trabeculae inthe central regions of static constructs was 0.031 ± 0.013mm2/mm2 as compared with 0.210 ± 0.022 mm2/mm2 for per-fused constructs (i.e., a 7-fold increase caused by perfusion).SEM images of the inner regions of the tissue constructs cor-

roborated these findings, yet they were uniquely instructive. Theinner regions of static constructs showed pore spaces that wereempty or only loosely packedwith the cells andmatrix (Fig. 3C andD), in contrast to perfused constructs, which showed denselypacked pore spaces throughout their entire volumes (Fig.3 G and H).The mineral deposition in pore spaces was also evident from

the 3D reconstructions of μCT images (Fig. 4) that revealedmeasurable differences between morphological parameters. Sta-tistically significant increases in bone volume were observed withtime of culture in both static (8.7%) and perfused (11.1%)constructs, with consequent increases in trabeculae number andthickness and decreases in trabecular spacing (Table 1). Thestructural model index numbers for static and perfused con-structs were lower after cultivation, indicating a trend toward theformation of plate-like trabeculae.

Modeling Flow Patterns: Relation Between Interstitial Flow and TissueDevelopment. Theoretical modeling of flow indicated a widedistribution in the magnitude (0–0.15 cm/s) and directions offlow velocities within the constructs (Fig. 5 A and B). The flowrates were highest in the inlet and outlet regions, adjacent to theneedle ports. Because of the complex geometry of the scaffold,its flat base is not at the center of the chamber, resulting in spatial

Fig. 1. Tissue engineering of anatomically shaped bone grafts. (A–C) Scaf-fold preparation. (A and B) Clinical CT images were used to obtain high-resolution digital data for the reconstruction of exact geometry of human TMJcondyles. (C) These data were incorporated into MasterCAM software tomachine TMJ-shaped scaffolds from fully decellularized trabecular bone. (D)A photograph illustrating the complex geometry of the final scaffolds thatappear markedly different in each projection. (E) The scaffolds were seeded instirred suspension of hMSCs, to 3 million cells per scaffold (≈1-cm3 volume)and precultured statically for 1 week to allow cell attachment, and then theperfusion was applied for an additional 4 weeks. (F) A photograph of perfu-sion bioreactor used to cultivate anatomically shaped grafts in vitro. (G–I) Keysteps in bioreactor assembly (see Movie S1).

Fig. 2. Cell numbers increased with time of culture and medium perfusion.From day 1 to day 7, the cell numbers increased 7.5-fold, from 3.4 × 106 to 25 ×106 cells per construct. Over the subsequent 4 weeks, cell numbers in staticculture increased 4.5-fold to ≈110 × 106 cells per construct, whereas theincrease in perfused bioreactor culture was 10-fold to ≈250 × 106 cells perconstruct (n = 3; *, P < 0.005; **, P < 0.001)

3300 | www.pnas.org/cgi/doi/10.1073/pnas.0905439106 Grayson et al.

Dow

nloa

ded

by g

uest

on

Mar

ch 2

4, 2

020

gradients of flow distribution across the base. The lowest flowrates occurred at far-right and far-left projections of the con-structs with the velocity vector computation, indicating near zeroflow at the extremities (Fig. 5A) because the outlet needle wasnot placed at the tip of the projection. Dye studies showed thatmedium does in fact perfuse these extreme regions. Histologicalanalysis of 5-week constructs clearly demonstrated cell survivaland matrix production in these regions.SEM images demonstrated interesting morphological varia-

tions in the tissue morphology with the variation in fluid flowpattern. For example, in the middle regions where the fluid flowis unidirectional, tissue appears smooth and aligned, and thecrystalline structures can be easily seen on the surface (Fig. 5 Cand D). At the base of the projection, close to the outlet port, themodel indicates large local changes in the velocity vector,effectively resulting in swirling flow patterns. High-magnifica-tion SEM images of this region demonstrated a corresponding“swirling” of mineralized matrix structure (Fig. 5 E and F).

DiscussionWe report an approach for tissue engineering of large, anatom-ically shaped and fully viable bone grafts, starting from hMSCs.This approach is based on a bioreactor capable of (i) housinganatomically shaped tissue constructs with complex geometry,

(ii) providing controlled interstitial flow through the porespaces, and (iii) enabling the establishment of cultivation proto-cols for engineering large human bone grafts.Cell attachment to the scaffolds was maximized by 1 week of

preculture, at which time the cell-seeded scaffolds were trans-ferred into the “anatomical” bioreactor chambers and the hy-drodynamic shear was applied by starting the medium perfusion.The bioreactor was designed to support this protocol. Because oftheir modular design, the bioreactor chambers can accommo-date different geometries by simply inserting a polydimethylsi-loxane (PDMS) mold of appropriate shape, created usingdigitized medical images. The bioreactor chambers allow fornoninvasive visualization and monitoring of medium distributionwithin the tissue constructs throughout the cultivation period. Inthe ongoing modification of bioreactor design, these samechambers are being adapted for placement into imaging com-partments (μCT or MRI) without disconnecting fluid flow, fortrue longitudinal, nondestructive imaging (23).Tissue engineering of large bone constructs requires flow

through the interstices of the scaffolds for efficient transport ofnutrients and waste materials between the cells and culturemedium and for direct exposure of cells to hydrodynamic shear,which are two important factors for osteogenesis. The volumetricflow rate (1.8 mL/min) and the corresponding superficial veloc-ity (an average of 0.06 cm/s) that were selected based on thedemonstrated ability to sustain dense tissue growth throughoutthe constructs, are within the range of flow rates shown tostimulate osteogenic differentiation of hMSCs (19, 20, 24).Because of the complex distribution of flow within the TMJ

tissue constructs, the flow rates were as high as 0.15 cm/s incertain construct regions and as low as 0.0001 cm/s in otherconstruct regions. Our recent studies suggest that in the wholerange of these flow velocities, hMSCs maintain complete via-bility and exhibit characteristics of osteogenic differentiation.Within this range of velocities, we could not observe that thereis a threshold in fluid flow rate after which perfusion becomesdetrimental to hMSCs. It is thus possible that tissue growth canbe further improved by increasing the flow rates in the bioreactorabove the levels used in this study.

A B C D

G HE F

Fig. 3. Bone formation was markedly enhanced by perfusion, in a manner dependent on the fluid flow pattern. (A–D) Constructs cultured under staticconditions. (E–H) Constructs cultured with medium perfusion. (A and E) Trichrome staining of entire cross-section of scaffolds showing differences in the newmatrix distribution (red) compared with the original scaffold (green) for the static (A) and perfused (E) culture groups. (B and F) Major differences were observedin osteoid formation (arrows) in the central regions of constructs cultured statically (B) and in perfusion (F). (C, D, G, and H) SEM images of the central constructregions. (C and D) Statically cultured constructs exhibit empty pore spaces and loosely packed cells. (G and H) Constructs cultured in perfusion demonstrateformation of dense and confluent lamellae of bone tissue that filled the entire pore spaces. (Scale bars: A and E, 5 mm; B, C, F, and G, 1 mm; D and H, 500 μm.)

Fig. 4. Architecture of the mineralized bone matrix developed with time andin a manner dependent on culture conditions. The reconstructions of 3D μCTimages demonstrate the changes in pore structure (relative to the initial state)that were evident at the end of the 5-week cultivation period. Bioreactorconstructs exhibit more rapid deposition of new mineral matrix as comparedwith static constructs (see Table 1). (Scale bar: 5 mm.)

Grayson et al. PNAS | February 23, 2010 | vol. 107 | no. 8 | 3301

MED

ICALSC

IENCE

SSP

ECIALFEATU

RE

Dow

nloa

ded

by g

uest

on

Mar

ch 2

4, 2

020

The study of DNA content and tissue histomorphology pro-vided important guidance for developing the bioreactor cultiva-tion protocols. Rapid, spatially uniform proliferation of hMSCswas facilitated by the high porosity and large pore sizes indecellularized bone scaffolds. Extensive growth of hMSCs hasbeen observed in other bioreactor applications (25) and isprobably related to a combination of enhanced nutrient transferand biophysical stimulation (26). Final cell densities were 105–

210 × 106 cells per mL, and these very high cell densities areimportant for functional bone tissue formation, because itheavily depends on cell–cell interactions (27).For statically cultured constructs, the loose packing of cells

(indicated by SEM) and only minimal osteoid formation (indi-cated by histology) provided evidence of limited functionaldifferentiation of the hMSCs in the inner regions of theseconstructs. For bioreactor cultured constructs, various imagingmodalities confirmed that cells formed dense tissues throughoutthe construct volumes, leading to larger increases in bonevolume. The changes in bone volume (up to 20 mm3 in bioreactorsamples) and microstructure could easily be measured by μCTwithin 5 weeks of cultivation with osteo-progenitor cells, whichis a relatively short time compared with the time scale of boneremodeling during fracture healing. For example, repair of a ratsegmental defect with a biomaterial revealed only 5 mm3 of newbone after 4 weeks (10 mm3 when scaffold was used in conjunc-tion with growth factors) and 17 mm3 after 16 weeks (28).Understanding the role of interstitial flow patterns on the

development of engineered bone constructs is critically impor-tant for bioreactor design and for determining the upper sizelimits of construct that can be grown in vitro. A computationalmodel was developed in our study to investigate the patterns ofmedium flow (amplitude and direction of interstitial velocity)through constructs and correlate these patterns to the experi-mental observations of bone formation (amounts and morphol-ogies of bone matrix). Predictive models developed in this wayare generally useful for designing bioreactors for engineeringtissues with complex geometries. However, we have not weighedthe relative contributions of fluid flow to the enhancement ofmass transport and the biophysical stimulation of cells byhydrodynamic shear. Additional analysis will be required todistinguish the effects of these two velocity-dependent factors onthe rates and patterns of bone formation. The ability to modelflow within the complex TMJ geometry, established in thepresent study, provides a powerful tool for in-depth investiga-tions of the role of flow on bone structure for a range oftissue-engineering applications.In summary, there is an acute need for bone grafts engineered

in vitro in a way that alleviates tissue morbidity (associated withthe use of autografts) and the incompatibility and diseasetransmission (associated with the use of allografts). The abilityto engineer with great precision large and viable bone graftscustomized to the specific patient and defect treated presentsunique opportunities to meet the clinical demands of diversecraniofacial and orthopaedic applications.To establish a robust method for engineering anatomically

shaped and fully viable human TMJ bone grafts, we integratedseveral tissue engineering concepts: (i) imaging guided fabrica-tion of anatomical scaffolds, (ii) use of decellularized bone as anosteo-inductive scaffold, (iii) use of multipotent MSC popula-

Table 1. μ CT evaluation of morphological parameters

Static Bioreactor

Parameter Initial Final % Δ Initial Final % Δ

Bone volume 190.3 ± 20.7 207.0 ± 23.9* 8.72 ± 0.75 190.9 ± 20.5 211.9 ± 21.7* 11.07 ± 0.66†

Trabecular number 2.055 ± 0.332 2.087 ± 0.360 1.47 ± 1.37 1.946 ± 0.281 1.999 ± 0.272 2.76 ± 1.55Trabecular thickness 0.101 ± 0.010 0.104 ± 0.009* 2.91 ± 0.98 0.104 ± 0.004 0.109 ± 0.004 4.23 ± 2.22Trabecular spacing 0.481 ± 0.075 0.472 ± 0.077* −1.93 ± 0.72 0.502 ± 0.075 0.486 ± 0.069 −3.20 ± 1.22Structural model index 1.055 ± 0.128 0.897 ± 0.159* N/A 1.084 ± 0.154 0.949 ± 0.159* N/A

Each scaffold was evaluated before and after cultivation to determine changes in bone volume, trabecular number, thickness, and spacing, and structuralmodel index in static or perfused bioreactor constructs. Paired t test was used to determine statistical changes in constructs relative to day 0 values. Student’s ttest was used to evaluate statistical differences in the percentage change between experimental groups. (n = 3 per group and time point). N/A, not applicable.*Statistically different from day 0 samples.†Statistically different from the other group at week 5.

Fig. 5. Bone matrix morphology correlated to the patterns of mediumperfusion flow. (A and B) Computational models of medium flow through TMJconstructs during bioreactor cultivation. (A) Color-coded velocity vectors in-dicate the magnitude and direction of flow through the entire constructbased on experimentally measured parameters. (B) Construct is digitally sec-tioned, and the color-coded contours are used to indicate the magnitude offlow in the inner regions. (C–F) Correlation of the medium flow pattern (bycomputational modeling) with the structural features of the new bone tissue(by SEM). (C and D) Flat and well-aligned layers of tissue with regular depo-sition of mineral crystals was observed in the central construct region whereflow is unidirectional. (E and F) A swirling flow in a region close to the outletfluid port resulted in the formation of a swirling tissue structure. (Scale bars:C and E, 1 mm; D and F, 100 μm.)

3302 | www.pnas.org/cgi/doi/10.1073/pnas.0905439106 Grayson et al.

Dow

nloa

ded

by g

uest

on

Mar

ch 2

4, 2

020

tions, applicable in either autologous or allogeneic fashion, (iv)perfusion based environmental control and biophysical stimu-lation of cultured bone constructs, and (v) a computationalmodeling optimization of bioreactor design.For TMJ as our bone graft model, this approach enabled the

formation of geometrically complex bone constructs of high struc-tural and biologic fidelity. Computational modeling of fluid flowprovided important insights into tissue responses to biophysicalstimuli. The engineered graft does not replicate the entire jointanatomy inclusive of the cartilage layer and the TMJ disc that areoften damaged in TMJ disorders. However, this approach couldimpact developmental biology (where high-fidelity tissue modelscan be used to study bone formation) and bioengineering andclinical translation (by providing surgeons with large and viableanatomically shaped bone grafts for treating craniofacial or ortho-pedic defects). Engineering of vascularized bone in a way that allowsimmediate connection to the vascular supply of the host, an issuecritical for the clinical application of large bone grafts, is a majorfocus of ongoing studies.

Materials and MethodsCreating Anatomically Shaped Constructs. Clinical CT scans were used to obtainthe exact image of a human TMJ condyle. 2D optical slices of the head in thecoronal plane (with 1-mm spacing between slices) were loaded into Mimics 9.0for 3D reconstruction, and the hard tissue threshold was selected so that onlybone was reconstructed from the slices. The reconstructed skull was seg-mented to include only the region that contained the mandibular condyle.The 3D geometry of the TMJ condyle was reconstructed as an IGESfile that wasimported into a 3D computer-aided-design (CAD) program Solid Works andused to design both the scaffold and the bioreactor chamber for tissueengineering of an anatomically correct TMJ. The 3D geometry was importedinto the MasterCam computer-aided-manufacturing (CAM) software used togenerate a set of toolpaths for fabrication on a Bridgeport three-axis com-puter-numerical-control (CNC) milling machine (Hardinge). Cylindrical piecesof trabecular bone harvested from the knee joints of calves were placed in arotary fixture in the CNC machine and machined into the exact shape of thepatient’s TMJ condyle (Fig. 1A). The final scaffolds had dimensions of ≈15 ×15 × 5 mm3 and a volume of ≈1 cm3 as determined by μCT and Solid Works.

Perfusion Bioreactor. A perfusion bioreactor system (Fig. 1 E and F) has beendevelopedforcontrollingtheperfusionpaththroughageometricallycomplexbonescaffold. Key elements of the design were the anatomically shaped chamber, perfu-sion flow, maintenance of sterility throughout the cultivation period, and imagingcompatibility. The bioreactor had the external diameter of 7.5 cm and height of 5 cmand was machined from acrylic and polyetherimide plastics. Scaffolds were placedinto a PDMS mold, which was created by pouring PDMS around a CNC milled delrin(acetal copolymer) generated from digital images to exactly correspond to the TMJcondyle and allowing it to cure before cutting out the delrin.

An exploded view of the bioreactor is shown in Fig. 1G. The bone scaffoldand PDMS mold are assembled and placed into the polypropylene casing. Ametal rod placed through the preformed holes in the PDMS mold and polypro-pylene casing aligns the assembly with the groove in the acrylic chamber,thereby maintaining the scaffold in its correct orientaton (Fig. 1H). The entireassembly (scaffold–PDMS–casing) is inserted intotheacrylic chamber,which is tightlycappedwith a polyetherimide top (Fig. 1 andMovie S1).

The acrylic chamber and cap also serve to compress the PDMSmold aroundthe TMJ scaffold, forcing culture medium to flow through the entire scaffoldrather than channeling around the periphery. The acrylic chamber has sixradial cylindrical ports, each serving as a guide for controlling the exactposition and depth of insertion of the 23-G needle into the scaffold. Thelocation of each port is determined in CAD, which was used for the 3Dreconstruction of the patient’s TMJ. In this study, three of the six ports wereused as outlets. An additional port, aligned along the central axis of thechamber, was connected to the tubing via a Luer connector and served as asingle inlet for medium to enter the scaffold.

The flow rate ofmedium exiting through the outlet ports was regulated tobe equal for each outlet, by adjusting clamps on the tubing. Culture mediumwas pumped back into a reservoir, which also acted as a bubble trap. Mediumwas recirculated by a low-flow multichannel digital peristaltic pump (Isma-tec).Mediumchangesweredone through the tubing thatwas connected toanadditional port at the reservoir and extended from the incubator to theadjacent biological safety cabinet. A syringe was used to remove and addmedium sterilely and without disturbing the bioreactor operation.

Decellularized Bone Scaffolds. Bone scaffolds were prepared by using a mod-ification of a method we previously described (22). Briefly, trabecular bonewas cut out from the subchondral region of the knee joints of 2-week-old to4-month-old cows by using a bandsaw. The blocks were placed on a lathe andmade into cylinders of variable length and 1.75 cm in diameter. Cylindricalbone pieces were then CNC-milled into the geometry of the TMJ condyle.

The grafts were washed with high-velocity streams of water to remove themarrow from the pore spaces. Then scaffolds were washed for 1 h in PBS with0.1% EDTA (wt/vol) at room temperature followed by sequential washes inhypotonic buffer [10 mM Tris, 0.1% EDTA (wt/vol)] overnight at 4 °C, deter-gent [10 mM Tris, 0.5% SDS (wt/vol)] for 24 h at room temperature andenzymatic solution (50 units/mL DNase, 1 units/mL RNase, 10 mM Tris) for 3–6h at 37 °C to remove any remaining cellular material. After SDS treatment,several (>7) washes with PBS were performed until no more bubbles wereseen in the PBS duringwashing. At the end of the process, decellularized boneplugs were rinsed repeatedly in PBS and freeze-dried. The dry weights of thescaffolds were measured and used to calculate the scaffold density. Theinternal pore structure varied between grafts, hence standardization wasachieved by selecting for a specified density range of 271.6 ± 34.1 mg/mL.Scaffolds were sterilized in 70% ethanol for 1 h, rinsed in PBS, and incubatedin culture medium overnight before seeding cells.

hMSCs. Fresh bone marrow aspirates were obtained from Cambrex. Bonemarrow-derived hMSCs were isolated by their attachment to the plasticsurfaces. Cellswere expanded inhigh-glucoseDMEMsupplementedwith 10%FBS, 1% pen-strep, and 0.1 ng/mL bFGF (control medium). Isolated cells werecharacterized by their ability to differentiate into osteogenic, chondrogenic,and adipogenic lineages (Fig. S3). The hMSCs were cultured up to the thirdpassage and then used for seeding the scaffolds. After seeding, constructswere cultured with osteogenic medium (high-glucose DMEM supplementedwith 10% FBS, 10 nM dexamethasone, 10 mM sodium-β-glycerophosphate,and 0.05 mM ascorbic acid-2-phosphate).

Construct Seeding and Culture. The hMSCs at the third passage weretrypsinized and resuspended in culturemediumat adensityof 106 cells permL.The TMJ scaffolds were fixed unto 26-G needles, and three scaffolds weresuspended in an inverted position in 30 mL of cell suspension that was stirredat 300 rpm for 1 h at 37 °C. [Stirring was done with an Isotemp analogmagnetic stirrer from Fisher Scientific (catalog no. 11-601-16SQ).] Constructswere then placed into 50-mL tubes containing one scaffold per tube in 10 mLof osteogenic medium, with daily medium change. After 7 days, the timeallowed the cells to deposit extracellular matrix and become firmly attachedto the scaffold, the constructs were transferred to bioreactors and perfused ata flow-rate of 1.8 mL/min or placed into static culture for an additional 4weeks. Constructs were provided with 40 mL of medium with 50% mediumchange every 3 days. Constructs were harvested for analyses after 1 and 5weeks of cultivation.

μCT. μCTwas performed by using amodified protocol from Liu et al. (29). Tissueconstructs were imaged longitudinally, after the decellularization process (be-fore cell seeding and cultivation) and at the end of static or perfused culture. Inthis way, the changes in bone volume and morphological parameters were forthesameconstructatdifferent timepointsandevaluatedwithapaired t test. Forimaging, the constructs were stabilized with wet gauze in a 15-mL centrifugetube and aligned along their axial direction. The tube was clamped in thespecimen holder of a vivaCT 40 system (SCANCO Medical AG) and the sampleswere scannedat 21-μmisotropic resolution. Thebone volumewasobtained fromthe application of a global thresholding technique so that only the mineralizedtissue was detected. Spatial resolution of this full-voxel model was consideredsufficient for evaluating the microarchitecture of the bone tissue.

DNA Content. For evaluating DNA content, constructs were harvested imme-diately after seeding (day 0), after the first week of static culture (day 7), andat the end of 5 weeks of cultivation in static or bioreactor conditions (day 35)(n = 3 constructs per group and time point). Tissue constructs were washed inPBS, placed in 1 mL of digestion buffer (10 mM Tris, 1 mM EDTA, and 0.1%Triton X-100) with 0.1 mg/mL proteinase K in centrifuge tubes, and incubatedovernight at 50 °C. The supernatants were drawn off and diluted 10-fold.Picogreen dye (Molecular Probes) was added to the samples in duplicate in96-well plates (100 μL of dye: 100-μL sample) and read in a fluorescent platereader (excitation 485 nm: emission 530 nm). Comparisons of DNA contentswere carried out with one-way ANOVA followed by Tukey's posthoc analysisusing STATISTICA software. P < 0.05 was considered significant.

Grayson et al. PNAS | February 23, 2010 | vol. 107 | no. 8 | 3303

MED

ICALSC

IENCE

SSP

ECIALFEATU

RE

Dow

nloa

ded

by g

uest

on

Mar

ch 2

4, 2

020

Hard Tissue Histology. Upon harvest, tissue constructs were washed in PBS,fixed in 10% formalin (1–2 days), dehydrated with sequential washes inethanol (2 days at 70%, 2 days at 100% ethanol with twice daily solutionchanges) and toluene (2 days with once daily solution change). Subsequently,the samples were washed in activated methyl methacrylate (MMA) with dailychanges of MMA solution for 4 days at 4 °C and then placed at 32 °C until theMMA cured. Plastic-embedded sections were sectioned at 8 μm by using aLeica hard tissue microtome. Staining for the osteoid formation was done byusing the traditional Goldner’s Masson trichrome stain. A bone histomor-phometry program, OsteoMeasure (Osteometrics), was used to quantify thebone formation in 9-mm2 regions throughout the construct volume.

SEM. Tissue constructs were washed in PBS, fixed in 2.5% glutaraldehyde inPBS overnight at 4 °C, washed in buffer, and freeze-dried overnight in alyophillizer. Before imaging, the samples were coated with gold/palladium.The constructs were cut, and different pieces were used to systematicallyevaluate both the inner and outer regions.

Flow Modeling. Flow patterns through the complex 3D geometry wereinvestigated with the aid of the FloWorks module of Solid Works. Velocityfields were generated by assuming flow of an incompressible fluid througha porous medium with no-slip boundary conditions at construct surfaces.The porosity of constructs was determined by μCT. An order-of-magnitudeapproximation for pressure drop was derived from published values (30).Inlet flow was assumed to be fully developed and set at 1.8 mL/min. Laminarflow and ideal walls (vanishingly small shear stresses at wall surfaces)were assumed throughout the construct, and Darcy’s equation for flowthrough porous medium was applied. Equal flow through each of theoutlet ports was assumed to correlate with the experimentally determinedparameters.

ACKNOWLEDGMENTS. This work was supported by National Institutes ofHealth Grants R01 DE161525 and P41 EB02520 (to G.V.-N.) and the MandlFoundation (to W.L.G.).

1. Feinberg SE, Hollister SJ, Halloran JW, Chu TMG, Krebsbach PH (2001) Image-basedbiomimetic approach to reconstruction of the temporomandibular joint. Cells TissuesOrgans 169:309–321.

2. Grayson W, Chao PHG, Marolt D, Kaplan D, Vunjak-Novakovic G (2008) Engineeringcustom-designed osteochondral tissue grafts. Trends Biotechnol 26:181–189.

3. Alhadlaq A, et al. (2004) Adult stem cell driven genesis of human-shaped articularcondyle. Ann Biomed Eng 32:911–923.

4. Alhadlaq A, Mao JJ (2003) Tissue-engineered neogenesis of human-shaped mandibu-lar condyle from rat mesenchymal stem cells. J Dent Res 82:951–956.

5. Alhadlaq A, Mao JJ (2005) Tissue-engineered osteochondral constructs in the shape ofan articular condyle. J Bone Joint Surgery Am 87:936–944.

6. Weng YL, Cao YL, Silva CA, Vacanti MP, Vacanti CA (2001) Tissue-engineered compos-ites of bone and cartilage for mandible condylar reconstruction. J Oral Maxillofac Surg59:185–190.

7. Abukawa H, et al. (2004) Reconstruction of mandibular defects with autologoustissue-engineered bone. J Oral Maxillofac Surg 62:601–606.

8. Hollister SJ (2005) Porous scaffold design for tissue engineering. Nat Mater 4:518–524.9. Hollister SJ, Levy RA, Chu TM, Halloran JW, Feinberg SE (2000) An image-based

approach for designing and manufacturing craniofacial scaffolds. Int J Oral MaxillofacSurg 29:67–71.

10. Hollister SJ, Lin CY (2007) Computational design of tissue engineering scaffolds.Comput Meth Appl Mech Eng 196:2991–2998.

11. Hollister SJ, Zysset PK, Guldberg RE, Chu TM, Halloran JW (2001) Engineering micro-structures to evaluate and replace trabecular bone. Adv ExpMed Biol 496:199–211.

12. Wang L, Detamore MS (2007) Tissue engineering the mandibular condyle. Tissue Eng13:1955–1971.

13. Detamore MS, Athanasiou KA, Mao J (2007) A call to action for bioengineers anddental professionals: Directives for the future of TMJ bioengineering.Ann Biomed Eng35:1301–1311.

14. Mao JJ, et al. (2006) Craniofacial tissue engineering by stem cells. J Dental Res85:966–979.

15. Shanti RM, Li WJ, Nesti LJ, Wang X, Tuan RS (2007) Adult mesenchymal stem cells:Biological properties, characteristics, and applications in maxillofacial surgery. J OralMaxillofac Surg 65:1640–1647.

16. Pittenger MF, et al. (1999) Multilineage potential of adult human mesenchymal stemcells. Science 284:143–147.

17. Coletti F, Macchietto S, Elvassore N (2006) Mathematical modeling of three-dimensional cell cultures in perfusion bioreactors. Ind Eng Chem Res 45:8158–8169.

18. Pathi P, Ma T, Locke BR (2005) Role of nutrient supply on cell growth in bioreactordesign for tissue engineering of hematopoietic cells. Biotechnol Bioeng 89:743–758.

19. Bancroft GN, et al. (2002) Fluid flow increases mineralized matrix deposition in 3Dperfusion culture of marrow stromal osteloblasts in a dose-dependent manner. ProcNatl Acad Sci USA 99:12600–12605.

20. Sikavitsas VI, Bancroft GN, Holtorf HL, Jansen JA, Mikos AG (2003) Mineralized matrixdeposition by marrow stromal osteoblasts in 3D perfusion culture increases withincreasing fluid shear forces. Proc Natl Acad Sci USA 100:14683–14688.

21. Sikavitsas VI, et al. (2005) Flow perfusion enhances the calcified matrix deposition ofmarrow stromal cells in biodegradable nonwoven fiber mesh scaffolds. Ann BiomedEng 33:63–70.

22. Grayson WL, et al. (2008) Effects of initial seeding density and fluid perfusion rate onformation of tissue-engineered bone. Tissue Eng A 14:1809–1820.

23. Porter BD, Lin ASP, Peister A, Hutmacher D, Guldberg RE (2007) Noninvasive imageanalysis of 3D construct mineralization in a perfusion bioreactor. Biomaterials28:2525–2533.

24. Holtorf HL, Jansen JA, Mikos AG (2005) Flow perfusion culture induces the osteoblasticdifferentiation of marrow stromal cell-scaffold constructs in the absence of dexameth-asone. J BiomedMater Res A 72:326–334.

25. Zhao F, Ma T (2005) Perfusion bioreactor system for human mesenchymal stem celltissue engineering: Dynamic cell seeding and construct development. BiotechnolBioeng 91:482–493.

26. Martin I, Wendt D, Heberer M (2004) The role of bioreactors in tissue engineering.Trends Biotechnol 22:80–86.

27. Stains JP, Civitelli R (2005) Gap junctions in skeletal development and function. BiochimBiophys Acta Biomembr 1719:69–81.

28. Oest ME, Dupont KM, Kong HJ, Mooney DJ, Guldberg RE (2007) Quantitative assess-ment of scaffold and growth factor-mediated repair of critically sized bone defects.J Orthop Res 25:941–950.

29. Liu XWS, Sajda P, Saha PK, Wehrli FW, Guo XE (2006) Quantification of the roles oftrabecularmicroarchitecture and trabecular type in determining the elasticmodulus ofhuman trabecular bone. J Bone Miner Res 21:1608–1617.

30. Hui PW, Leung PC, Sher A (1996) Fluid conductance of cancellous bone graft as apredictor for graft-host interface healing. J Biomechanics 29:123–132.

3304 | www.pnas.org/cgi/doi/10.1073/pnas.0905439106 Grayson et al.

Dow

nloa

ded

by g

uest

on

Mar

ch 2

4, 2

020