ARTICLE IN PRESS

0142-9612/$ - se

doi:10.1016/j.bi

�CorrespondE-mail addr

Biomaterials 27 (2006) 2222–2232

www.elsevier.com/locate/biomaterials

Rational development of GAG-augmented chitosan membranesby fractional factorial design methodology

Yen-Lin Chena,b, Huang-Chi Chena, Hsiao-Ping Leea, Hing-Yuen Chanb, Yu-Chen Hua,�

aDepartment of Chemical Engineering, National Tsing Hua University, Hsinchu 300, TaiwanbFood Industry Research and Development Institute, Hsinchu 300, Taiwan

Received 12 July 2005; accepted 31 October 2005

Available online 21 November 2005

Abstract

To develop a novel biomaterial for chondrocyte culture, 8 glycosaminoglycan (GAG)/chitosan membranes (groups N1–N8) were

prepared, with the aid of a 2-level 24�1 fractional factorial design, by co-immobilizing chondroitin-4-sulfate (CSA), chondroitin-6-sulfate

(CSC), dermatan sulfate (DS), and heparin to chitosan membranes. The fractional factorial design allowed us to partly interpret the

effects of individual GAGs and two-way interactions between GAGs. Within the level range of �1 and +1, low CSA level (2.6mg) is

favorable for collagen synthesis but not for cell proliferation. High CSC level (1.3mg) is favorable for GAG production but not for cell

proliferation. Conversely, high heparin (0.33mg) and DS (0.13mg) levels are desired for cell proliferation but not for the production of

collagen and GAG. Moreover, the two-way interactions between GAGs influence the cell behavior. Among the 8 GAG/chitosan

membranes, N1 and N4 (containing low CSA and heparin levels) lead to the maintenance of proper chondrocyte phenotype, as judged by

the chondrocyte-like morphology, modest cell expansion, higher GAG and collagen production and proper cartilage marker gene

expression. In conclusion, this approach provides a means of rationally predicting and evaluating the proper formulation of GAG/

chitosan membranes and may facilitate the rational design of other tissue engineering scaffolds.

r 2005 Elsevier Ltd. All rights reserved.

Keywords: Chondrocyte; Chitosan; Chondroitin sulfate; Fractional factorial design; Glycosaminoglycan; Heparin

1. Introduction

Cartilage tissue engineering has been motivated by theneed to create implantable tissue equivalents for clinical use[1]. However, the development of appropriate scaffoldsthat can accommodate the chondrocytes and provideproper biological stimuli for guiding the cell differentiationis still a challenging task. To date, a wide variety of naturaland synthetic polymers are being evaluated for theirpotential as tissue scaffolds. One example of the naturalpolymer is chitosan (poly 1,4 D-glucosamine), a deacety-lated derivative of chitin abundantly present in nature.Owing to the non-toxicity, excellent biocompatibility andbiodegradability, chitosan and some of its complexes havebeen studied for use in a number of biomedical applica-tions including wound dressings, drug delivery vehicles,

e front matter r 2005 Elsevier Ltd. All rights reserved.

omaterials.2005.10.029

ing author. Tel.: +886 3 571 8245; fax: +886 3 571 5408.

ess: yuchen@che.nthu.edu (Y.-C. Hu).

space filling implants and tissue engineering scaffolds (forreview, see [2]).Glycosaminoglycans (GAGs), including chondroitin-4-

sulfate (CSA), chondroitin-6-sulfate (CSC), hyaluronan,keratin sulfate (KS), dermatan sulfate (DS), heparansulfate (HS), and heparin, are important componentsconstituting the cartilage extracellular matrix (ECM) whichconfers the cartilage desired mechanical properties. Amongthem, hyaluronan, CSA, and CSC are the predominantGAGs present in cartilage in the form of aggrecan, whileother species account for smaller fractions. The exactcomposition, however, may vary with developmental stageand age [3]. In addition to the structural role, GAGs arepivotal in assembling protein–protein complexes (e.g.growth factor receptor) on the cell surface or in theECM, and hence are involved in initiating cell signalingevents or inhibiting biochemical pathways [4]. Further,extracellular GAGs may sequester proteins and presentthem to appropriate sites for activation [4]. These functions

ARTICLE IN PRESS

Table 1

The high (coded +1) and low (coded �1) levels of each GAG component

GAG Experimental levels

Low level (coded �1) High level (coded +1)

(%)a (mg)b (%)a (mg)b

CSA 1 2.6 2 5.2

CSC 0.05 0.13 0.5 1.3

DS 0.005 0.013 0.05 0.13

Heparin 0.0125 0.033 0.125 0.33

aThe original concentrations (w/v) of the GAGs solutions added to each

well.bThe corresponding GAG amount per cm2 is expressed as amount (mg)

for simplicity, because the area of each well (1.9 cm2) and the volumes of

GAG solutions added (500ml per well) were fixed.

Y.-L. Chen et al. / Biomaterials 27 (2006) 2222–2232 2223

enable GAGs to influence the adhesion, migration,proliferation and differentiation of cells.

Due to the aforementioned features of chitosan andGAGs, a scaffold incorporating a GAG/chitosan complexmay provide a means of retaining and recruiting desirablefactors secreted by cells or from surrounding tissue fluids.This concept has been exploited by augmenting CSA [5],CSC [5,6], hyaluronan [5], or heparin [7] to chitosan viaionic cross-linkage. Although some degrees of success havebeen shown, these materials still far from resemble thenatural ECM which consists of multiple types of GAGs.Additionally, how the simultaneous presence of multipleGAG species in the biomaterial affects chondrocyteproliferation, differentiation and ECM production remainsto be explored.

One overriding objective of this study was to develop anovel biomaterial suitable for chondrocyte culture andultimately cartilage tissue engineering. Since augmentationof individual CSA, CSC, DS, and heparin to chitosan hasbeen shown as mentioned above, and multiple GAG speciesare present in the natural cartilage ECM, we hypothesizedthat co-immobilization of these 4 GAG species may exert asynergistic effect on the maintenance of chondrocytephenotype. Therefore, we cross-linked these 4 GAG speciesto chitosan membranes and then assessed the effects of GAGcomposition on membrane preparation and cell behavior.Typically, researchers develop biomaterials by examining oneparameter at a time, thus many time-consuming and laborintensive experiments would be required if the effects of 4GAG species are to be studied simultaneously. In contrast,fractional factorial design is a methodology for rationalexperimental design and allows for statistical analysis of theeffects of individual parameters and interactions betweenparameters. The potential of fractional factorial design inbiomaterial development has recently been demonstrated[8,9]. Therefore, a 4-factor, 2-level 24�1 fractional factorialdesign methodology was employed in order to minimize thenumber of experiments performed while permitting thesystematic study of the effects of GAGs.

2. Materials and methods

2.1. Materials

Chitosan (85% deacetylated, MWE200,000) from crab shells, chon-

droitin-6-sulfate (CSC, sodium salt, MWE100,000) from shark cartilage,

dermatan sulfate (DS, sodium salt, MWE60,000) and heparin (low

calcium content, MWE30,000) from porcine intestinal mucosa, were

purchased from Sigma. Chondroitin-4-sulfate (CSA, sodium salt,

MWE50,000) from bovine trachea was purchased from Fluka. N-

hydroxysuccinimide (NHS) was purchased from Merck. 1-ethyl-3-(3-

dimethylaminopropyl) carbodiimide (EDC), 2-(N-morpholino) ethanesul-

fonic acid (MES) and all other reagents were purchased from Sigma unless

otherwise noted.

2.2. Experimental design and statistical analysis

Based on the 4-factor, 2-level 24�1 fractional factorial design, 8 GAG/

chitosan compositions were formulated. All membrane preparation and

culture experiments were performed in standard polystyrene 24-well tissue

culture plates (Falcon). Because the area of each well (E1.9 cm2) and the

volumes of GAGs solutions added (500ml per well) were fixed, the

amounts of GAGs per cm2 are expressed as amounts (mg) for simplicity.

Table 1 summarizes the high (coded +1) and low (coded �1) levels of

each GAG component for GAG/chitosan membrane preparation. The

original concentrations (w/v) of the GAGs solutions added to each well

and the corresponding GAGs amounts are listed. The compositions of

these 8 membranes (N1–N8) are tabulated in Table 2.

All data analyses were performed using Design-Experts (Version 6.0,

Stat-Ease Inc.). The data from these 8 experimental groups (N1–N8) were

computed by the software using Yates’ algorithm [10] to calculate the

effect coefficients of the following regression equation:

Y n ¼ C0 þ C1 � X 1 þC2 � X 2 þ C3 � X 3 þ C4 � X 4

þ C5 � X 1X 2 þ C6 � X 1X 3 þ C7 � X 1X 4,

where Yn represents the response (e.g. cell expansion ratio, GAG and

collagen production), while X1, X2, X3 and X4 indicate the codes of CSA,

CSC, DS and heparin, respectively. X1X2, X1X3 and X1X4 denote the two-

way interactions between CSA and CSC, CSA and DS, and CSA and

heparin, respectively. C0 is the mean of response while C1, C2, C3, C4, C5,

C6 and C7 stand for the effect coefficients associated with the individual

factors or the two-way interactions. Such design and analyses enabled us

to investigate the dependence of responses on individual factors and two-

way interactions at a resolution of IV, meaning that the effects of

individual factors were not confounded by the two-way interactions. Since

this was a fractional, rather than full (requiring 16 formulations), factorial

design, the 3- and 4-way interactions were not elucidated. The statistical

difference between data sets was determined using Student’s t-test.

2.3. Preparation of GAG/chitosan membranes

To prepare the membranes in 24-well plates, 100ml of 2% (w/v)

autoclaved chitosan solution was mixed with 100ml of 2% (v/v)

autoclaved acetic acid, followed by overnight agitation and aseptic

filtration. For each well (1.9 cm2), 500ml of chitosan solution (1%) was

added and air dried in the laminar flow hood for 24 h so that E2.6mg of

chitosan (per cm2) was deposited. The GAGs were dissolved in HEPES

buffer (pH 7.4) according to the formulations shown in Table 2 and passed

through 0.2 mm filters. For each well, 500ml of GAGs solution was added

to the chitosan membrane and air dried. The cross-linking reaction was

initiated by adding 500ml of sterile EDC/NHS solution (in 50mM MES

buffer, pH 5.5) and continued for 48 h at 4 1C. The weight ratio of GAG/

EDC/NHS was adjusted to 1:1.84:0.23 in all reactions. Following the

cross-linkage, the membranes were air dried for 24 h, washed 5 times with

phosphate buffered saline (PBS, pH 7.4) to remove residual GAGs and

cross-linking reagents (EDC was washed in the form of urea to avoid

cytotoxicity). The amounts of GAGs in the spent PBS were measured [11]

ARTICLE IN PRESS

Table 2

Compositions of the 8 GAG/chitosan membranes

Experimental groups GAGs added (mg)a

CSA CSC DS Heparin Total

N1 2.6 1.30 0.130 0.033 4.063

N2 5.2 1.30 0.130 0.330 6.660

N3 2.6 0.13 0.130 0.330 3.190

N4 2.6 0.13 0.013 0.033 2.776

N5 5.2 1.30 0.013 0.033 6.546

N6 5.2 0.13 0.013 0.330 5.673

N7 2.6 1.30 0.013 0.330 4.243

N8 5.2 0.13 0.130 0.033 5.493

PSb — — — — —

aThe values indicate the amounts of GAG added to the chitosan membrane per cm2 and are expressed as amounts (mg) for simplicity. The amount of

chitosan was fixed at 2.6mg in all groups.bPolystyrene surface of the 24-well culture plate.

Y.-L. Chen et al. / Biomaterials 27 (2006) 2222–22322224

to evaluate the amounts of GAGs actually cross-linked to the chitosan

membranes. The cross-linking efficiency of GAGs was calculated as

follows:

Cross-linking efficiency ð%Þ ¼ 100� 1�GAGwashed

GAGadded

� �,

where GAGwashed and GAGadded denote the amount of GAGs in the waste

PBS and the amount of GAGs initially added, respectively.

2.4. Isolation and culture of articular chondrocytes

Chondrocytes were isolated from the articular cartilages of 7-day-old

Wistar rats by 0.2% collagenase II digestion [12]. The cells were cultured

using Dulbecco’s minimal essential medium (DMEM) containing 10% of

fetal bovine serum, 100U/ml penicillin, 100mg/ml streptomycin and

2.5mg/ml amphotericin, and passaged 3 times (P3) prior to use. The P3

cells were seeded onto 24-well plates (6� 104 cells/well) with or without

GAG/chitosan membranes, and continued to be cultured for 2 weeks at

37 1C. The medium was exchanged twice a week.

2.5. Cell adhesion, morphology and proliferation

At 1-day post-seeding, the medium was withdrawn and the wells were

washed with PBS. The cells in the withdrawn medium and PBS were

pooled, centrifuged and resuspended in PBS. The number of unattached

cells was counted by a hemacytometer to calculate cell adhesion. At 7 days

post-seeding, the cell morphologies were observed and photographed

under a light microscope. At 14 days post-seeding, the cells were

trypsinized and the cell numbers in each well were counted. The expansion

ratio was calculated by comparing the final cell number and the initial

attached cell number.

2.6. Measurement of collagen and GAG production

To measure the total collagen production during the 2-week culture

period, the cells were trypsinized at day 14 and the cell suspensions were

subject to acid hydrolysis (6 N HCl for 12 h at 100 1C). Total collagen

contents were determined in triplicate spectrophotometrically from

hydroxyproline concentration after reaction with chloramines-T and p-

dimethylaminobenzaldehyde as described previously [13]. The absorbance

was measured at 550 nm and the collagen concentrations were calculated

using collagen (Sigma) as the standard.

To quantify the sulfated GAG production, the cell suspensions were

first lyophilized and then digested by papain for 24 h. The GAG

concentrations were assayed in triplicate by mixing the papain-digested

samples with 1,9-dimethylmethylene blue and reading the absorbance at

525 nm using CSA (Sigma) as the standard [11].

2.7. Reverse transcription-polymerase chain reaction (RT-PCR)

To determine the differentiation status of chondrocytes after 2-week

culture, the gene expressions of type I, type II and type X collagen and

aggrecan were analyzed by RT-PCR and normalized against that of

GAPDH. Briefly, the chondrocytes were cultured for 14 days, trypsinized

and washed with PBS. The total RNA was extracted using RNeasy Mini

Kit (Qiagen) and quantified spectrophotometrically. Approximately 0.3 mgof total RNA was used as the template for cDNA synthesis and

subsequent PCR (30 cycles) using AccuPower RT/PCR PreMix kit

(Bioneer). The sequences of the primer pairs for each gene are listed in

Table 3. The PCR products were subjected to 2% agarose gel

electrophoresis, followed by scanning densitometry using Scion Image

Software (Scion Corporation).

3. Results

3.1. Effects of GAG on the preparation of GAG/chitosan

membrane and cell adhesion

Although the preparation of GAG/chitosan membranesvia ionic complexation has been reported, this method,however, resulted in lower bond strength and release ofcomplexed GAG to the medium [2], thus changing themembrane composition over culture time. To circumventthis problem, the GAGs were covalently linked to chitosanmembranes using EDC/NHS. The maximum amount ofGAGs that could be cross-linked was first investigated byvarying the amounts of CSA added to chitosan mem-branes. As shown in Fig. 1a, the amount of CSAconjugated increased with increasing amount of CSAadded although the cross-linking efficiencies (numbersabove the bars) declined from greater than 80% to nearly50%. The amount of CSA conjugated did not appear toapproach saturation at 7.2mg, indicating sufficient func-tional groups available on chitosan for cross-linkage.

ARTICLE IN PRESS

80

93

83

68

50

0

1

2

3

4

5

0.2 0.9 1.8 3.6 7.2

Con

juga

ted

CSA

(m

g)

0

20

40

60

80

100

0 0.2 0.9 1.8 3.6 7.2

*

*

(a)

(b) CSA added (mg)

CSA added (mg)

C

ell a

dhes

ion

(% o

f ce

lls s

eede

d)

Fig. 1. Effects of the amount of CSA added on the amount of CSA

conjugated (a) and cell adhesion at 1-day post-seeding (b). The CSA/

chitosan membranes were prepared as described in Materials and methods

except that GAGs other than CSA were not included. For simplicity, the

amount of CSA per cm2 is expressed as amount (mg). The data represent

the mean7standard deviation (SD) of 3 independent experiments (6 wells

in each experiment) but only the mean values of cross-linking efficiencies

are shown above the bars in (a). The statistical difference between data sets

was analyzed using Student’s t-test. *, po0:05.

Table 3

Primer sequences used in gene expression analysis

Gene Oligonucleotides (50–30)

Collgen type I Forward CTGCTGGAGAACCTGGAAAG

Reverse TGTCACCCTTAGGACCTGGA

Collgen type II Forward GGAGACTACTGGATTGA

Reverse GCGTGAGGTCTTCTGTG

Collgen type X Forward TGCCTCTTGTCAGTGCTAAC

Reverse GCGTGCCTGTCTTATACAGG

Aggrecan Forward ATGCCCAAGACTACCAGTGG

Reverse TCCTGGAAGCTCTTCTCAGT

GAPDH Forward CTGGTCATCAATGGGAAAC

Reverse CAAAGTTGTCATGGATGA

Y.-L. Chen et al. / Biomaterials 27 (2006) 2222–2232 2225

Whether the chitosan membranes modified by CSAcross-linkage supported chondrocyte adhesion was alsodetermined. Without CSA modification, the chondrocyteattachment on the chitosan membranes was poor (onlyE40%) at 1-day post-seeding (Fig. 1b), which agreed withthe observations reported previously [14]. By modifying themembrane with as low as 0.2mg of CSA, the cell adhesionimproved to 460%. This was further enhanced to 480%

provided that 40.9mg of CSA was added. Since CSA is arepresentative GAG component in the cartilage ECM,Fig. 1 suggests that covalently cross-linking GAGs tochitosan membranes is feasible and an amount of GAGexceeding 0.9mg is desired for cell adhesion.

3.2. Effects of GAG composition on the preparation of

GAG/chitosan membranes

As described in Materials and methods, 8 GAG/chitosanmembranes (N1–N8) of different formulations were pre-pared. As shown in Fig. 2a, in all 8 experimental groups theamounts of GAGs conjugated (black bars) varied with theGAG compositions and exceeded 0.9mg that was requiredfor optimal cell adhesion (Fig. 1b), albeit varying cross-linking efficiencies (E40–70%). This suggested that all 8GAG/chitosan membranes supported cell adhesion.Whether the GAGs remained stably conjugated to these8 membranes was further examined. Fig. 2b reveals that inall 8 groups E1–4% of GAGs was released from themembranes to the medium in the first 4 days, but therelease rate declined to E1% every 4 days thereafter. Theaccumulative amount of GAGs released after 16 daysremained below 10% of the GAGs originally added. Thesedata indicate the slow release of GAGs albeit periodicmedium change, thus confirming the stable conjugation bycross-linkage.

3.3. Effects of GAG composition on cell morphology and

proliferation

The effects of GAG composition on cell morphologyand proliferation were studied by culturing the chondro-cytes on the polystyrene surfaces (PS) and on these 8 GAG/chitosan membranes. Fig. 3 illustrates that the cellscultured on the PS continued to grow, reached confluencyand assumed a fibroblast-like morphology after 7 days. Incontrast, the cells cultured on the GAG/chitosan mem-branes did not reach confluency and assumed twomorphologies. Spread cells which appeared fibroblast-likewere observed in N2, N3, N6, N7 and N8, whereas

ARTICLE IN PRESS

71

52

6064

7041

70

72

0

2

4

6

8

10Amount of GAG added

Amount of GAG conjugated

0%

2%

4%

6%

8%

4 day

8 day

12 day

16 day

Exp. group.

Exp. group.

% o

f G

AG

rel

ease

Am

ount

of

GA

G (

mg)

(b)

(a)1 2 3 4 5 6 7 8

1 2 3 4 5 6 7 8

Fig. 2. (a) Effect of GAG composition on the amounts of GAGs

conjugated. The amounts of GAGs actually conjugated were measured

and the cross-linking efficiencies were calculated as described in Materials

and methods. The mean cross-linking efficiencies are shown above the

black bars. (b) Effect of GAG composition on GAGs release over time.

The 8 GAG/chitosan membranes were incubated with complete media for

16 days. The medium was exchanged every 4 days and the amounts of

GAGs released into the medium were measured. The percentage of GAGs

release (%) was calculated by dividing the amounts of GAGs in medium at

different time points by the amounts of GAGs initially immobilized. The

data represent the mean7SD of 3 independent experiments (6 wells in

each experiment).

Y.-L. Chen et al. / Biomaterials 27 (2006) 2222–22322226

chondrocyte-like round cells were observed in N1, N4and N5.

The expansion ratio analysis (Fig. 4) shows that the cellscultured on the PS for 14 days grew E12-fold, whereas thecells cultured on the GAG/chitosan membranes grew onlyE1–4 fold. Comparing Figs. 3 and 4, the cells assumingfibroblast-like morphology (N2, N3, N6, N7 and N8) alsoexhibited higher expansion ratios (from 1.85 to 3.92). Incontrast, the chondrocyte-like cells (N1, N4 and N5)exhibited lower expansion ratios (from 0.98 to 1.43).

3.4. Effects of GAG composition on collagen and GAG

production

Whether the GAG compositions influenced subsequentcollagen and GAG production by chondrocytes wasdetermined at day 14. As shown in Fig. 5, all experimentalgroups, except N8, resulted in significantly higher collagenand GAG production (po0:05) compared to the controlgroup (PS), indicating that these GAG/chitosan mem-

branes enhanced ECM production. In accordance with thedata for cell morphology and proliferation, the extent ofECM production varied widely among experimentalgroups. Groups N2, N3, N7 and N8 in which cellsappeared fibroblast-like exhibited lower production ofcollagen (E0.16–0.69 mg) and GAG (E41–80 mg). All theseare signs of de-differentiation and may limit the potentialsof these GAG compositions for chondrocyte culture.In comparison, groups N1, N4, and N5 whereby cellsappeared chondrocyte-like resulted in significantly higherproduction of collagen (E1.03–1.49 mg) and GAG(E104–161 mg). When comparison was made betweenN1, N4 and N5, no significant difference in collagenproduction was observed (p40:05). The GAG productionwas significantly higher in N5 than in N1 and N4 (po0:05),but not significantly different between N1 and N4(p40:05).

3.5. Gene expression in chondrocytes cultured on the GAG/

chitosan membranes

Whether N1, N4 and N5 adopted chondrocytic differ-entiation pathway was investigated by analyzing the geneexpression profiles by RT-PCR (Fig. 6a). The expressionlevels of all genes were normalized against that of GAPDH(Fig. 6b). In comparison with the control group (PS), theexperimental groups (N1, N4 and N5) expressed signifi-cantly higher aggrecan and type II collagen but lower typeI collagen (po0:05), again confirming the superiority ofthese GAG/chitosan membranes relative to PS. Althoughsome degrees of de-differentiation (expression of type Icollagen) were observed, this was likely due to the use of P3chondrocytes that have been shown to undergo substantialde-differentiation [15].The comparison between the experimental groups

revealed that N5 expressed significantly lower levels ofaggrecan and type II collagen than N4 (po0:05). The geneexpression data seemed conflicting to the biochemical assaydata (Fig. 5) which show similar collagen production buthigher GAG production in N5 compared to N1 and N4.This disparity arose probably because N5 underwenthigher degree of hypertrophy as judged by the significantlyhigher type X collagen expression (compared to N4,po0:05). Note that the biochemical assays only measuredthe total collagen and GAG but could not distinguish thetypes. The higher type X collagen expression in N5probably compensated for the lower type I and type IIcollagen expression. Likewise, the higher GAG productionin N5 (Fig. 5b) could be attributed to the switch ofproteoglycan production from aggrecan to other smallerproteoglycans, as the expression of biglycan and fibromo-dulin was up-regulated more than that of aggrecan inhypertrophic cells [16]. Namely, although the total GAGproduction was higher in N5, the proportion of aggrecanpresumably decreased and the GAG composition mighthave deviated from that of natural cartilage.

ARTICLE IN PRESS

Fig. 3. Effect of GAGs composition on cell morphology. The P3 chondrocytes were cultured on the polystyrene surfaces (PS) of 24-well plates or on the 8

GAG/chitosan membranes. The cell morphologies were photographed at 7 days post-seeding. S, spread cell; R, round cell; magnification, 320� .

Y.-L. Chen et al. / Biomaterials 27 (2006) 2222–2232 2227

On the other hand, aggrecan, type II and type X collagenwere expressed at similar levels (p40:05) in N1 and N4.The gene expression data agreed with the biochemicalassay data (Fig. 5) and confirmed that N1 and N4maintained reasonably high expression of cartilage-specificmarkers and lower degree of hypertrophy.

3.6. Effects of individual GAG species and two-way

interactions

To further elucidate why N1, N4 and N5 were superior,the dependences of responses (i.e. expansion ratio andcollagen and GAG production) on the factors (i.e. CSA,CSC, DS, heparin and two-way interactions) were analyzedusing Yates’ algorithm. Table 4 summaries the effectcoefficients for the independent factors in the regressionequations. Within the design level range, positive coeffi-cients indicate that a higher level (higher amount) isfavorable for the response, whereas negative coefficientsindicate that a lower level (lower amount) is desired. The

magnitudes of the coefficients indicate the relative con-tributions of the factors to the response.With respect to expansion ratio (Y1), Table 4 reveals

positive coefficients for CSA (C1 ¼ þ0:61), DS(C3 ¼ þ0:10), and heparin (C4 ¼ þ0:53), indicating thatin the level range between �1 and +1 higher levels of CSA,DS and heparin favored cell proliferation. However, higherCSC level (C2 ¼ �0:40) was not desired. As to collagenproduction (Y2), higher levels of CSA (C1 ¼ �0:24), DS(C3 ¼ �0:18) and heparin (C4 ¼ �0:20) exerted inhibitoryeffects, but the small coefficient for CSC (C2 ¼ þ0:03)implied that collagen production was virtually independentof CSC in the level range between �1 (0.13mg) and +1(1.3mg). As to GAG production (Y3), higher DS(C3 ¼ �29:60) and heparin (C4 ¼ �18:60) levels exertedstrong inhibitory effects, but higher CSC (C2 ¼ þ10:40)level only exerted a relatively weak positive effect. TheGAG synthesis was nearly independent of CSA(C1 ¼ þ1:90) in the level range between �1 (2.6mg) and+1 (5.2mg).

ARTICLE IN PRESS

3.922.302.66

11.68

3.011.851.430.981.04

0

4

8

12

16

1 2 3 4 5 6 7 8 PS

cell

expa

nsio

n ra

tio

*

Exp. group

Fig. 4. Effect of GAG composition on expansion ratio. The P3

chondrocytes were cultured on the PS or on the 8 GAG/chitosan

membranes for 14 days. The expansion ratio was obtained by dividing the

final cell number by the number of cells initially attached. The data

represent the mean7SD of 4 independent experiments (3 wells in each

experiment) but only the mean values are shown above the bars. The

Student’s t-test revealed a significant difference in expansion ratios

between N5 and N7 (po0:05). *po0:05.

1.29

0.16

0.65

1.03

0.19 0.32

0.69

1.49

0.71

0

0.5

1

1.5

2

2.5

PS

** **

*

4146

80

109104 118

1140

161

0

50

100

150

200

250

** *

(a)

(b)

Exp. group

Exp. group

GA

G p

rodu

ctio

n (µ

g)C

olla

gen

prod

uctio

n (µ

g)

1 2 3 5 6 7 84

PS 1 2 3 5 6 7 84

Fig. 5. Effects of GAG composition on the production of collagen (a) and

GAG (b) by chondrocytes during 2-week culture. The cells cultured on the

PS were included as a control. The total amounts of GAGs and collagen in

each well were measured as described in Materials and methods. The

amount of GAGs in the blank (the well that contained identical GAG/

chitosan membranes and incubated with medium for 2 weeks) was

measured and subtracted from the total amount of GAGs in the well to

obtain the GAG production. The total amounts (mg) of collagen and

GAGs produced in each well were converted to mg/106 cells based on the

cell number at day 14, but are expressed as amounts (mg) for convenience.The data represent the mean7SD of 4 independent experiments (3 wells in

each experiment) but only mean values are shown above the bars.

*po0:05; **p40:05.

Y.-L. Chen et al. / Biomaterials 27 (2006) 2222–22322228

In addition to the influences from single factors, theregression equations allowed for unraveling the roles oftwo-way interactions between factors. Note that the two-way interaction only means that the co-existence of the twoGAG components may impose synergistic or subtractiveeffects on the response, but does not necessarily imply thatthe two GAG species physically interact with each other.By setting the CSC and heparin levels at 0 while varying theCSA and DS levels at �1 or +1, a standard procedure inthe analysis, the effect of the two-way interaction betweenCSA and DS on collagen production was predicted fromthe regression equation and is shown in Fig. 7a. Assumingno interactions between CSA and DS, lower CSA and DSlevels would be desired for collagen production asmentioned above. When the two-way interaction wastaken into account (Fig. 7a), however, high (+1) or low(�1) DS levels led to similar collagen production levelswhen CSA was present at low level (�1). Likewise, at lowCSA level (�1) the collagen production levels were similarregardless of the CSC level (Fig. 7b). In sharp contrast,Fig. 7c depicts that at low CSA level, low heparin level(�1) yielded considerably higher collagen production thanhigh heparin level (+1). The same approach was used toelucidate how the two-way interactions influenced GAGsynthesis. Although low DS level was desired for GAGproduction when considering DS alone, at low CSA level(�1) not much difference in GAG production wasobserved at high (+1) and low (�1) DS levels (Fig. 7d).Similarly, at low CSA level the GAG production levelswere similar regardless of the CSC levels (Fig. 7e).Analogous to Fig. 7c, Fig. 7f delineates that at low CSAlevel, low heparin level (�1) resulted in higher GAGproduction than high heparin level (+1). These analyses, incombination with those regarding the influences of singlefactors, collectively suggested that lower CSA and heparinlevels (as in N1 and N4) were desired for ECM production.Under these circumstances, ECM production was similarregardless of the CSC and DS levels. These data explainedwhy N1 and N4 yielded similar collagen and GAGproduction levels.The two-way interactions also aided in interpreting other

data. When comparing N1 and N5, Fig. 7a suggests thatN1 (low CSA and high DS levels) and N5 (high CSA andlow DS levels) resulted in fairly close collagen production,but Fig. 7d suggests that N1 yielded significantly lowerGAG production than N5. These comparisons againagreed with the data shown in Fig. 5.

4. Discussion

Although a number of studies have reported theconjugation of GAG to chitosan, the modification islimited to conjugating single GAG type. In contrast toprevious studies, we co-immobilized 4 GAG species (CSA,CSC, DS and heparin) and prepared 8 GAG/chitosanmembranes with the aid of fractional factorial designmethodology. Notably, the GAG composition profoundly

ARTICLE IN PRESS

0.0

0.5

1.0

1.5

2.0

N1 N4 N5 PS** *

**

*

**

***

*

N1 N4 N5

Aggrecan

Collagen II

Collagen I

Collagen X

GAPDH

Aggrecan collagen II collagen I collagen X

rela

tive

gene

exp

ress

ion

PS

(a)

(b)

Fig. 6. (a) Gene expression profiles of GAPDH, aggrecan, type I, type II and type X collagen as analyzed by RT-PCR. The chondrocytes were cultured on

N1, N4, N5 or PS for 14 days. (b) The relative expression levels of aggrecan, type I, type II and type X collagen among N1, N4, N5 and the control group

(PS). The expression level of each gene was normalized against that of GAPDH. The data represent the mean7SD of 3 independent experiments (3 wells

in each experiment). *po0:05; **p40:05.

Table 4

Effect coefficients of the regression equationsa

Response Effect coefficients

C0 C1 C2 C3 C4 C5 C6 C7

Y1b +2.15 +0.61 �0.40 +0.10 +0.53 �0.31 �0.02 �0.00

Y2c (mg) +0.79 �0.24 +0.03 �0.18 �0.20 +0.09 �0.14 +0.16

Y3d (mg) +87.39 +1.90 +10.40 �29.60 �18.60 +3.86 �16.14 +6.89

aY n ¼ C0 þ C1 � X 1 þ C2 � X 2 þ C3 � X 3 þ C4 � X 4 þC5 � X 1X 2 þ C6 � X 1X 3 þ C7 � X 1X 4,where Yn represents the response while X1, X2, X3 and X4

denote the codes of CSA, CSC, DS and heparin, respectively. X1X2, X1X3 and X1X4 denote the two-way interactions between CSA and CSC, CSA and DS,

and CSA and heparin, respectively. C0 is the mean of response while C1, C2, C3, C4, C5, C6 and C7 are the effect coefficients. The analyses enabled us to

investigate the dependence of responses at a resolution of IV, meaning that the effects of individual factors were not confounded by the two-way

interactions. No 3- or 4-way interactions were elucidated.bY1 (n ¼ 1) denotes cell expansion (fold increase)cY2 (n ¼ 2) denotes collagen production (mg) per 106 cells but is expressed as collagen production (mg) for convenience.dY3 (n ¼ 3) denotes GAG production (mg) per 106 cells, but is expressed as GAG production (mg) for convenience.

Y.-L. Chen et al. / Biomaterials 27 (2006) 2222–2232 2229

dictates the cell morphology and differentiation. Amongthe 8 GAG/chitosan membranes, N2, N3, N6, N7 andN8 partly improve the characteristics of chondrocyteculture as opposed to PS, but fail to reverse the de-differentiation of P3 cells used in this study. N5 appearssimilar to N1 and N4 in many aspects, but causes a higherdegree of hypertrophy. In contrast, N1 and N4 lead tochondrocyte-like morphology, modest cell expansion,higher GAG and collagen production and proper cartilagemarker gene expression. These clearly indicate the main-tenance of proper chondrocyte phenotype and implicate

the potentials of N1 and N4 formulations in cartilage tissueengineering.In addition to facilitating rational experimental design,

the factorial design approach makes it possible to elucidatethe effects of individual GAG components and two-wayinteractions between GAGs. Within the level range of �1and +1, high heparin level strongly supports cellproliferation but greatly suppresses the production ofcollagen and GAG. This may be linked to its high bindingaffinity to various growth factors, including basic fibroblastgrowth factor (bFGF) [17]. bFGF has been shown to alter

ARTICLE IN PRESS

0

0.5

1

1.5

-1 10

50

100

150

-1 1

0

0.5

1

1.5

-1 10

50

100

150

-1 1(b)

(a)

coll

agen

(µg

)

level of CSA level of CSA

level of CSAlevel of CSA

(d)

GA

G (µ

g)G

AG

(µg)

(e)

0

50

100

150

-1 1

level of CSA

GA

G (µ

g)

(f)

CSC -1 level

CSC +1 level

DS-1 level

DS+1 level

DS-1 level

DS+1 level

coll

agen

(µg

)

0

0.5

1

1.5

-1 1(c) level of CSA

coll

agen

(µg

)

CSC -1 levelCSC +1 level

heparin -1 levelheparin +1 level

heparin -1 levelheparin +1 level

Fig. 7. Effects of two-way interactions on the production of collagen (a–c) and GAG (d–f). The predicted production levels were obtained from the

regression equations by setting the levels of two GAG species at +1 or �1 while setting the levels of the other two at 0.

Y.-L. Chen et al. / Biomaterials 27 (2006) 2222–22322230

cell morphology, induce chondrocyte de-differentiation[18] and cause ossification and matrix degradation [19].Besides, bFGF promotes cell proliferation while suppressesthe synthesis of type II collagen and proteoglycans [20].Consequently, high heparin level in the GAG/chitosanmembranes might facilitate the recruitment of bFGFtoward the cells, thereby signaling the chondrocyteproliferation and de-differentiation. This, in part, explainswhy the membranes with high heparin amounts (N2, N3,N6, and N7) lead to higher cell expansion ratios and lowerECM production.

Similarly, Table 4 shows that higher DS level favors cellproliferation but strongly discourages collagen and GAGproduction, which accounts for the poor maintenance ofchondrocyte phenotype and lower ECM production insome GAG/chitosan membranes containing high DS levels(N2, N3 and N8). Although N1 also contains high DSlevel, the negative effects might be offset by the low CSAand heparin levels which favor the maintenance ofchondrocyte phenotype. The stimulatory effect of DS oncell proliferation concurs with the previous finding that DSenhances the proliferation of a leukemia cell line (U-937)but induces a sharp decrease of differentiation markers[21]. In addition, DS stimulates cell proliferation viabinding to FGF-2 and FGF-7 [22,23]. As such, it is likelythat DS promotes chondrocyte proliferation by recruitingFGF to the matrix. Comparatively, the effects of DS on

cartilage-specific GAG and collagen production remainlargely unknown. One relevant clue comes from the studyof mucopolysaccharidosis (MPS) type VI articular chon-drocytes. MPS is a disease characterized by excessiveaccumulation of DS-containing proteoglycans which con-tributes to marked proteoglycan and collagen depletion inthe diseased chondrocytes [24]. This suggests the down-regulation of GAG and collagen production by DS andagrees with our data.Furthermore, higher CSA level promotes cell prolifera-

tion but inhibits collagen synthesis in the level rangebetween �1 and +1. GAG production is nearly unaffectedby CSA level within this range. Conversely, higher CSClevel is adverse to cell proliferation but promotes GAGproduction. Chondroitin sulfate is an important constitu-ent of ECM and its involvement in modulating celldifferentiation, morphogenesis and brain and centralnervous system development has been well documented(for review see [25]). However, very little is known aboutthe role of chondroitin sulfate in the regulation of articularcartilage formation and how CSA and CSC differ in termsof modulating chondrocyte differentiation and ECMproduction. One huge impediment to exploring thebiological functions of chondroitin sulfate is the dauntingtask in extraction of chondroitin sulfate in a more nativeform [26]. Consequently, it is difficult to interpret our datareadily. Nevertheless, by conjugation to type I collagen

ARTICLE IN PRESSY.-L. Chen et al. / Biomaterials 27 (2006) 2222–2232 2231

scaffolds, CSA helps maintain the bovine chondrocytephenotype [27]. Similarly, ionic complexation of CSA tochitosan membranes elevates the expression of cartilage-specific type II collagen by bovine chondrocytes [14].Consistent with the previous findings, our data demon-strate that CSA supports chondrogenesis. Besides, CSCbinds and induces the dimerization of pleiotrophin (PTN)[28] which inhibits chondrocyte proliferation while stimu-lates proteoglycan synthesis [29]. The dimerization andsubsequent activation of PTN by CSC might account forthe positive effects of CSC on GAG production.

It is also noteworthy that two-way interactions betweenGAG components are involved in mediating the celldifferentiation and ECM production. To our best knowl-edge, this is the first report exploring the roles of two-wayinteractions between GAGs when designing a GAG-augmented biomaterial for tissue engineering. Note, how-ever, that the two-way interaction does not necessarilymean that the two GAG species have physical interactions.Instead, it suggests that co-existence of the two GAGcomponents may exert synergistic or subtractive effects oncell proliferation and differentiation by, for example,recruiting or inhibiting the binding of certain biologicalfactors. Because information concerning how the relativeamounts of two GAG species affect cell behavior is notavailable in literature, in-depth discussions are not possibleat present. Nonetheless, the factorial design, together withthe statistical analyses, allows us to partly interpret theeffects of individual GAGs and two-way interactionsbetween GAGs. This may pave the way for designing anexperimental system to investigate the importance of GAGcompositions on cartilage formation and elucidate how thedynamic change of GAG compositions in diseasedcartilages leads to disease progression.

Importantly, when interpreting our data, care should betaken that the effects of these factors are valid only withinthe level range of �1 and +1. Although the experimentaldata suggest the use of low CSA (2.6mg) and heparin(0.033mg) levels (as in N1 and N4), one should not bemisled to conclude that these two GAGs ought to beremoved from the membrane because CSA helps maintainchondrocyte differentiation while heparin stimulates cellproliferation, both are required for cartilage tissue en-gineering. The analysis only suggests that higher CSA andheparin levels are not recommended within the design levelrange. This is logic because the effects of heparin,chondroitin sulfate and DS on chondrogenesis are dose-dependent [30] and the incorporation of excess GAGsgreater than an optimal amount may disrupt the delicatebalance between cell proliferation and differentiation.Further optimization of the GAG formulation can beperformed by Response Surface Methodology.

5. Conclusion

To our best knowledge, this is the first report employingthe factorial design methodology for the development of

GAG-augmented biomaterials and exploring the roles oftwo-way interactions between GAGs. This approachprovides a means of rationally predicting and evaluatingthe proper formulation of GAG/chitosan membranes andmay facilitate the rational design of other tissue engineer-ing scaffolds.

Acknowledgment

The authors gratefully acknowledge the financial sup-port from Ministry of Economic Affairs (TechnologyDevelopment Program for Academia Grant, MOEA 93-EC-17A-17S1-0009), Taiwan.

References

[1] Langer R, Vacanti JP. Tissue engineering. Science 1993;260:

920–6.

[2] Suh JK, Matthew HW. Application of chitosan-based polysaccharide

biomaterials in cartilage tissue engineering: a review. Biomaterials

2000;21:2589–98.

[3] Plaas AHK, Wong-Palms S, Roughley PJ, Midura RJ, Hascall VC.

Chemical and immunological assay of the nonreducing terminal

residues of chondroitin sulfate from human aggrecan. J Biol Chem

1997;272:20603–10.

[4] Raman R, Sasisekharan V, Sasisekharan R. Structural insights into

biological roles of protein–glycosaminoglycan interactions. Chem

Biol 2005;12:267–77.

[5] Denuziere A, Ferrier D, Damour O, Domard A. Chitosan–chon-

droitin sulfate and chitosan–hyaluronate polyelectrolyte complexes:

biological properties. Biomaterials 1998;19:1275–85.

[6] Chupa JM, Foster AM, Sumner SR, Madihally SV, Matthew HW.

Vascular cell responses to polysaccharide materials: in vitro and in

vivo evaluations. Biomaterials 2000;21:2315–22.

[7] Madihally SV, Matthew HW. Porous chitosan scaffolds for tissue

engineering. Biomaterials 1999;20:1133–42.

[8] Hedberg EL, Shih CK, Lemoine JJ, Timmer MD, Liebschner MAK,

Jansen JA, et al. In vitro degradation of porous poly(propylene

fumarate)/poly(DL-lactic-co-glycolic acid) composite scaffolds. Bio-

materials 2005;26:3215–25.

[9] Fisher JP, Dean D, Mikos AG. Photocrosslinking characteristics and

mechanical properties of diethyl fumarate/poly(propylene fumarate)

biomaterials. Biomaterials 2002;23:4333–43.

[10] Montgomery DC. Design and analysis of experiments. New York:

Wiley; 1991.

[11] Enobakhare BO, Bader DL, Lee DA. Quantification of sulfated

glycosaminoglycans in chondrocyte/alginate cultures, by use of 1,9-

dimethylmethylene blue. Anal Biochem 1996;243:189–91.

[12] Chen H- C, Lee H- P, Sung M- L, Liao C- J, Hu Y- C. A novel

rotating-shaft bioreactor for two-phase cultivation of tissue-engi-

neered cartilage. Biotechnol Progr 2004;20:1802–9.

[13] Komsa-Penkova R, Spirova R, Bechev B. Modification of Lowry’s

method for collagen concentration measurement. J Biochem Biophys

Methods 1996;32:33–43.

[14] Sechriest VF, Miao YJ, Niyibizi C, Westerhausen-Larson A,

Matthew HW, Evans CH, et al. GAG-augmented polysaccharide

hydrogel: a novel biocompatible and biodegradable material to

support chondrogenesis. J Biomed Mater Res 2000;49:534–41.

[15] Darling EM, Athanasiou KA. Rapid phenotypic changes in passaged

articular chondrocyte subpopulations. J Orthop Res 2003;23:

425–32.

[16] Adams ME, Matyas JR, Huang D, Dourado GS. Expression of

proteoglycans and collagen in the hypertrophic phase of experimental

osteoarthritis. J Rheumatol Suppl 1995;43:94–7.

ARTICLE IN PRESSY.-L. Chen et al. / Biomaterials 27 (2006) 2222–22322232

[17] Trowbridge JM, Gallo RL. Dermatan sulfate: new functions from an

old glycosaminoglycan. Glycobiology 2002;12:117R–25R.

[18] Inoue H, Kato Y, Iwamoto M, Hiraki Y, Sakuda M, Suzuki F.

Stimulation of cartilage-matrix proteoglycan synthesis by morpholo-

gically transformed chondrocytes grown in the presence of fibroblast

growth factor and transforming growth factor-beta. J Cell Physiol

1989;138:329–37.

[19] Chintala SK, Miller RR, McDevitt CA. Basic fibroblast growth

factor binds to heparan sulfate in the extracellular matrix of rat

growth plate chondrocytes. Arch Biochem Biophys 1994;310:180–6.

[20] Holland TA, Mikos AG. Advances in drug delivery for articular

cartilage. J Control Release 2003;86:1–14.

[21] Volpi N, Petrini M, Conte A, Valentini P, Venturelli T, Bolognani L,

et al. Effects of glycosaminoglycans on U-937 leukemia cell

proliferation and differentiation: structure-function relationship.

Exp Cell Res 1994;215:119–30.

[22] Penc SF, Pomahac B, Winkler T, Dorschner RA, Eriksson E,

Herndon M, et al. Dermatan sulfate released after injury is a potent

promoter of fibroblast growth factor-2 function. J Biol Chem

1998;273:28116–21.

[23] Taylor KR, Rudisill JA, Gallo RL. Structural and sequence motifs in

dermatan sulfate for promoting fibroblast growth factor-2 (FGF-2)

and FGF-7 activity. J Biol Chem 2005;280:5300–6.

[24] Simonaro CM, Haskins ME, Schuchman EH. Articular chondrocytes

from animals with a dermatan sulfate storage disease undergo a high

rate of apoptosis and release nitric oxide and inflammatory cytokines:

a possible mechanism underlying degenerative joint disease in the

mucopolysaccharidoses. Lab Invest 2001;81:1319–28.

[25] Sugahara K, Mikami T, Uyama T, Mizuguchi S, Nomura K,

Kitawara H. Recent advances in the structural biology of chondroitin

sulfate and dermatan sulfate. Curr Opin Struct Biol 2003.

[26] Knudson CB, Knudson W. Cartilage proteoglycans. Semin Cell Dev

Biol 2001;12:69–78.

[27] van Susante JLC, Pieper J, Buma P, van Kuppevelt TH, van

Beuningen H, van Der Kraan PM, et al. Linkage of chondroitin-

sulfate to type I collagen scaffolds stimulates the bioactivity of seeded

chondrocytes in vitro. Biomaterials 2001;22:2359–69.

[28] Bernard-Pierrot I, Heroult M, Lemaitre G, Barritault D, Courty J,

Milhiet PE. Glycosaminoglycans promote HARP/PTN dimerization.

Biochem Biophys Res Commun 1999;266:437–42.

[29] Tapp H, Hernandez DJ, Neame PJ, Koob TJ. Pleiotrophin inhibits

chondrocyte proliferation and stimulates proteoglycan synthesis in

mature bovine cartilage. Matrix Biol 1999;18:543–56.

[30] San Antonio JD, Winston BM, Tuan RS. Regulation of chondrogen-

esis by heparan sulfate and structurally related glycosaminoglycans.

Dev Biol 1987;123:17–24.