ARTHRITIS & RHEUMATISMVol. 50, No. 1, January 2004, pp 131–141DOI 10.1002/art.11433© 2004, American College of Rheumatology

Expression Profiling of Metalloproteinases andTheir Inhibitors in Cartilage

Lara Kevorkian,1 David A. Young,1 Clare Darrah,2 Simon T. Donell,2 Lee Shepstone,1

Sarah Porter,1 Sarah M. V. Brockbank,3 Dylan R. Edwards,1

Andrew E. Parker,3 and Ian M. Clark1

Objective. To profile the expression of all knownmembers of the matrix metalloproteinase (MMP),ADAMTS, and tissue inhibitor of metalloproteinases(TIMP) gene families in normal cartilage and cartilagefrom patients with osteoarthritis (OA).

Methods. Human cartilage was obtained fromfemoral heads at joint replacement for OA or followingfracture to the femoral neck. Total RNA was purified,and gene expression was assayed using quantitativereal-time polymerase chain reaction.

Results. Several members of the above gene fam-ilies were regulated in OA. Genes that showed increasedexpression in OA were MMP13, MMP28, andADAMTS16 (all at P < 0.001), MMP9, MMP16,ADAMTS2, and ADAMTS14 (all at P < 0.01), andMMP2, TIMP3, and ADAMTS12 (all at P < 0.05). Geneswith decreased expression in OA were MMP1, MMP3,and ADAMTS1 (all at P < 0.001), MMP10, TIMP1, andADAMTS9 (all at P < 0.01), and TIMP4, ADAMTS5, andADAMTS15 (all at P < 0.05). Correlation analysisrevealed that groups of genes across the gene familieswere coexpressed in cartilage.

Conclusion. This is the first comprehensive ex-pression profile of all known MMP, ADAMTS, andTIMP genes in cartilage. Elucidation of patterns ofexpression provides a foundation with which to under-

stand mechanisms of gene regulation in OA and poten-tially to refine the specificity of antiproteolytic therapies.

Osteoarthritis (OA) is a debilitating disease thataffects �80% of people over the age of 65 (1). Given thecurrent demographic trend toward an older population,OA, for which age is an important risk factor (2), will bean increasing health and economic burden on society.

Degradation of articular cartilage is a majorfeature of OA. Cartilage is made up of 2 main extracel-lular matrix (ECM) macromolecules, type II collagenand aggrecan, a large aggregating proteoglycan (3,4).The type II collagen scaffold endows the cartilage withits tensile strength, while the aggrecan, by virtue of itshigh negative charge, swells against the collagen networkas it draws water into the tissue, enabling it to resistcompression. Quantitatively more minor components(e.g., types IX, XI, and VI collagen, biglycan, decorin,cartilage oligomeric matrix protein, etc.) also have im-portant roles in controlling matrix structure and organi-zation (5).

Normal cartilage ECM is in a state of dynamicequilibrium, with a balance between synthesis and de-gradation. For the degradative process there is a balancebetween proteinases that degrade the ECM and theirinhibitors. It is generally believed that in OA, a disrup-tion of this balance, in favor of proteolysis, leads topathologic cartilage destruction.

The matrix metalloproteinases (MMPs) are afamily of 23 enzymes in humans which facilitate ECMturnover and breakdown under normal and diseaseconditions (6). The MMP family contains the onlymammalian proteinases that can specifically degradetriple-helical collagens at neutral pH. These so-calledcollagenases specifically cleave a single locus in all 3collagen chains at a point three-quarters from theN-terminus of the molecule. The “classic” collagenases

Supported by an Industrial CASE Studentship from theBiotechnology and Biological Sciences Research Council and byAstraZeneca Pharmaceuticals.

1Lara Kevorkian, BSc, David A. Young, PhD, Lee Shepstone,PhD, Sarah Porter, MSc, Dylan R. Edwards, PhD, Ian M. Clark, PhD:University of East Anglia, Norwich, UK; 2Clare Darrah, RGN, DipSci, Simon T. Donell, MD: Norfolk & Norwich University Hospital,Norwich, UK; 3Sarah M. V. Brockbank, BSc, Andrew E. Parker, PhD:AstraZeneca Pharmaceuticals, Cheshire, UK.

Address correspondence and reprint requests to Ian M. Clark,PhD, School of Biological Sciences, University of East Anglia, Nor-wich NR4 7TJ, UK. E-mail: i.clark@uea.ac.uk.

Submitted for publication June 16, 2003; accepted in revisedform September 17, 2003.

131

(MMP-1, MMP-8, and MMP-13) have differing sub-strate specificities for types I, II, and III collagen, withMMP-13 showing a preference for type II collagen (7).More recently, gelatinase A (MMP-2) and MT1-MMP(MMP-14) have also been shown to make the specificcollagen cleavage, though with less catalytic efficiencythan the classic collagenases, at least in vitro (8,9). Theenzyme(s) responsible for cartilage collagen cleavage inthe arthritides is open to debate, but it has beengenerally believed that MMP-1, produced in the syno-vium, is the primary collagenase in rheumatoid arthritis(RA), while MMP-13, produced by the chondrocyte, isthe foremost collagenase in OA. Several other membersof the MMP family have been localized to cartilage orsynovium in the arthritides (10,11).

More recently, a second group of proteinases thatimpacts upon ECM synthesis and degradation has beenidentified. The ADAMTS (a disintegrin and metallopro-teinase domain with thrombospondin motifs) familycontains 19 members (12,13). These include enzymesinvolved in collagen biosynthesis as procollagen propep-tidases (ADAMTS-2, ADAMTS-3, and ADAMTS-14)(14–17). Other members of the family are so-calledaggrecanases (ADAMTS-1, ADAMTS-4, ADAMTS-5,ADAMTS-9, and ADAMTS-15), which degrade theinterglobular domain separating G1 and G2 of aggrecanat a specific Glu373–Ala374 bond (13,18–22). Cleavagewithin this interglobular domain can be mediated byboth MMPs (cleaving at Asn341–Phe342) and aggre-canases, and both activities can be detected in articularcartilage from OA and RA patients (23). Current datasupport the hypothesis that aggrecanases are active earlyin the disease process, with later increases in MMPactivity; however, the exact enzyme(s) responsible forcartilage aggrecan destruction at any stage in arthritis isunclear.

A family of 4 specific inhibitors, the tissue inhib-itors of metalloproteinases (TIMPs), has been described(24). These are endogenous inhibitors of MMPs andpotentially ADAMTS. The ability of TIMPs 1–4 toinhibit active MMPs is largely promiscuous, though anumber of functional differences have been identified(24). TIMP-3 appears to be the most potent inhibitor ofADAMTS, with subnanomolar Ki against ADAMTS-4and ADAMTS-5 (25); however, TIMP-1 was used toaffinity purify ADAMTS-4, suggesting some inhibitorycapacity against this enzyme (20). TIMP-4 is reported tobe a weak inhibitor of ADAMTS-4 but not ADAMTS-5(25), while both TIMP-2 and TIMP-3 are reported toinhibit ADAMTS-1 (19).

The current view is that the local balance of

metalloproteinase and TIMP activities is pivotal in reg-ulating cartilage homeostasis, and that disturbance ofthis balance, resulting in an excess of metalloproteinaseover TIMP, underlies pathologic cartilage destruction.The relative contribution of any one metalloproteinaseor TIMP to this balance is largely unknown (6,26).

The expression of these 3 families of genes playsan essential role in the progression of cartilage destruc-tion in the arthritides. It is therefore crucial to discoverwhich genes are expressed in normal cartilage and howthis is altered in disease states. To this end, we have usedquantitative reverse transcriptase–polymerase chain re-action (RT-PCR) to profile all of the known MMP,ADAMTS, and TIMP genes in cartilage from osteoar-thritic hip joints and to compare the results with findingsin phenotypically normal hip articular cartilage frompatients with fracture of the femur. This is the firstreported measurement of the expression of all of thesegenes together in cartilage, and provides interestingcomparative data about the role of the gene families inend-stage OA.

MATERIALS AND METHODS

Cartilage samples. Human articular cartilage was ob-tained from femoral heads of patients undergoing total hipreplacement surgery at the Norfolk and Norwich UniversityHospital. Samples from 18 patients with OA (11 women, 7men; age range 38–81 years) were compared with cartilagefrom 15 patients undergoing hip replacement following frac-ture of the femoral neck (13 women, 2 men; age range 52–93years). OA was diagnosed by clinical history and examinationalong with radiographic findings; confirmation of gross patho-logic findings was made at the time of joint removal. Thefracture patients had no known history of joint disease andtheir cartilage was free of lesions; 80% of these patientsunderwent surgery within 36 hours of fracture. Their samplesare referred to herein as normal cartilage. This study wasperformed with Ethics Committee approval, and all patientsprovided informed consent.

RNA extraction from articular cartilage. Cartilage wasdissected from the femoral head shortly (�30 minutes) aftersurgical removal and snap frozen in n-hexane cooled in liquidnitrogen. Immediately, 100–200 mg of this cartilage wasground for 3 cycles of 2 minutes grinding and 2 minutescooling, in liquid nitrogen, at an impact frequency of 10 Hz ina freezer mill (SPEX CertiPrep 6750; Glen Creston, Stanmore,UK). Total RNA was isolated from the powdered cartilageessentially as described by Price et al (27). After initialextraction in TRIzol reagent (Invitrogen, Paisley, UK), theaqueous phase was mixed with a half volume of 100% ethanoland further purified using the RNeasy Plant Mini kit (Qiagen,Crawley, UK). Purified RNA was quantified by spectropho-tometry.

Synthesis of complementary DNA. ComplementaryDNA (cDNA) was synthesized from 0.5� g of total RNA, using

132 KEVORKIAN ET AL

Superscript II RT and random hexamers in a total volume of20 � l, according to the instructions of the manufacturer(Invitrogen). Complementary DNA was stored at �20°C untilused in downstream PCR.

Quantitative real-time PCR. Oligonucleotide primersand fluorescence-labeled probes were designed using PrimerExpress 1.0 software (PE Applied Biosystems, Warrington,UK). Sequences for MMP and TIMP primers and probes wereas described by Nuttall et al (28); ADAMTS primers andprobes were as developed by two of the authors (29). In orderto control against amplification of genomic DNA, primerswere placed within different exons close to an intron–exonboundary, with the probe spanning 2 neighboring exons wherepossible. BLAST searches for all of the primer and probesequences were also conducted to ensure gene specificity. The18S ribosomal RNA (rRNA) gene was used as an endogenouscontrol to normalize for differences in the amount of totalRNA present in each sample; 18S rRNA primers and probewere purchased from PE Applied Biosystems.

Relative quantification of genes was performed usingthe ABI Prism 7700 sequence detection system in accordancewith the protocol suggested by the manufacturer (PE AppliedBiosystems). PCR reaction mixtures contained 5 ng of reverse-transcribed RNA (1 ng for 18S analyses), 50% TaqMan 2�Master Mix (PE Applied Biosystems), 100 nM of each primer,and 200 nM of probe in a total volume of 25 � l. Conditions forPCR were as follows: 2 minutes at 50°C, 10 minutes at 95°C,then 40 cycles each consisting of 15 seconds at 95°C and 1minute at 60°C.

During PCR, the TaqMan probe emits a fluorescencesignal in proportion to the amount of specific amplifiedproduct. The ABI Prism 7700 measures cycle-to-cycle changesin fluorescence and generates a kinetic profile of DNA ampli-fication over 40 cycles. The cycle number (termed “thresholdcycle [CT]) at which an increase in fluorescence is detectableabove baseline signal is an indicator of the amount of targetRNA in each sample; thus, the fewer cycles it takes to reachthe CT, the greater the initial target gene copy number. Inorder to determine the relative messenger RNA (mRNA)levels within the samples, standard curves for each gene weregenerated using the cDNA from 1 sample and making 2-foldserial dilutions across an appropriate range. Relative inputamounts of template cDNA were then calculated from CTusing the standard curves; data are presented as relative levelsof mRNA normalized to 18S rRNA. As a final quality controlfor the purified RNA samples, only those cDNA falling within�1 CT of the median value for 18S in all samples were used inthe downstream study.

The absolute CT can be used as an approximatemeasure of expression to allow comparison of expression levelsbetween genes. While PCR efficiencies do vary with primerand template, our data on 17 different genes, obtained usingthis technology, show a good correlation between copy numberas assessed by using in vitro–transcribed RNA to produce astandard curve, and CT across these genes (ref. 28 andEdwards DR: unpublished observations). The representationof CT range, as shown in Figure 1, is therefore a useful meansfor visualization of the data.

In order to confirm that the amplification product wasindeed that of the desired target gene, products were sub-cloned and sequenced. All primer and probe sets have been

shown to amplify specific products from appropriate humantissue samples (28,29).

Recovery of total RNA from stored cartilage samples.After dissection, cartilage was snap frozen in n-hexane cooledin liquid nitrogen and then either extracted for RNA purifica-tion or stored under different conditions: 1) at �80°C; 2) inliquid nitrogen; or 3) at �80°C in RNALater (Ambion,Huntingdon, UK). On days 7, 14, 50, and 100, RNA waspurified from a sample of cartilage from each storage condi-tion described above and quantified spectrophotometrically.

Statistical analysis. Levels of mRNA for each gene,obtained from standard curves, were corrected using the 18SrRNA levels as described above, and all statistical tests wereperformed using these ratios. The significance of differencesbetween the control and OA groups was determined using a2-sided Mann-Whitney U test; this nonparametric test makesno prior assumption of the distribution of the data. Tests ofcorrelations between genes were performed using Spearman’srank correlation. In order to verify potential groups of coex-pressed genes, principal components analysis was performed.

RESULTS

Purification of RNA from human cartilage. Ini-tial experiments indicated that the yield of total RNApurified from human cartilage was highly variable (datanot shown). Hence, a systematic examination of both theconditions and time of tissue storage prior to RNApurification was undertaken. Storage of frozen cartilagefor 7 days under any of the conditions tested led to an�75% decrease in yield of RNA compared with purifi-cation of RNA from fresh snap-frozen cartilage (datanot shown). This yield was then essentially unaltered

Figure 1. Levels of expression of the genes for matrix metalloprotein-ases (MMPs), ADAMTS, and tissue inhibitor of metalloproteinases(TIMPs) in osteoarthritic (OA) hip cartilage and in phenotypicallynormal (N) hip cartilage from patients with femoral neck fracture.Expression was classified according to mean threshold cycle (CT) (thepolymerase chain reaction cycle at which a positive signal is firstmeasurable above background) and grouped into ranges.

MMP, TIMP, AND ADAMTS GENE EXPRESSION IN OA 133

between day 7 and day 100. Therefore, in the currentstudy, we purified RNA from cartilage within 24 hoursof the surgical removal of the femoral head.

Differential expression of MMPs, TIMPs, andADAMTS in cartilage. Expression of every known hu-man MMP, TIMP, and ADAMTS gene (Table 1) at themRNA level was profiled in articular cartilage from thefemoral heads of patients with end-stage OA and com-pared with the expression in phenotypically normalcartilage from patients with fracture of the femoral neck.We used the CT of each gene as an approximate measureof its expression (CT � 25 � very high; 26–30 � high;31–35 � moderate; 36–39 � low; 40 � not detected)(Figure 1).

Of genes in the MMP family, only MMP20 was

not detected in any sample, although expression ofMMP7, MMP8, MMP10, MMP11, MMP15, MMP17, andMMP25 was low in both OA and normal cartilage, as wasexpression of MMP28 in normal samples (and the me-dian CT for MMP8 and MMP11 was 40 although lowlevels were detected in a few samples). Expression ofMMP1, MMP16, MMP21, MMP23, MMP24, and MMP26was moderate in OA and normal cartilage, as wasexpression of MMP9 and MMP13 in normal cartilageand MMP28 in OA cartilage. MMP2, MMP12, MMP14,MMP19, and MMP27 were highly expressed in OA andnormal cartilage, as were MMP3, MMP9, and MMP13 inOA cartilage. The only MMP gene found to be veryhighly expressed was MMP3 in normal cartilage.

TIMP1 was very highly expressed in OA andnormal cartilage, with TIMP3 expression bordering be-tween high and very high. TIMP2 and TIMP4 weremoderately expressed.

Of genes in the ADAMTS family, ADAMTS7 andADAMTS8 were not detected in any sample. Expressionof ADAMTS14 was low in normal cartilage, andADAMTS3, ADAMTS12, and ADAMTS16 were ex-pressed at low levels in both OA and normal cartilage.The majority of ADAMTS genes (ADAMTS2,ADAMTS4, ADAMTS9, ADAMTS10, ADAMTS13,ADAMTS17, and ADAMTS20) were moderately ex-pressed in normal and OA cartilage, with ADAMTS14expressed moderately only in OA cartilage. ADAMTS1,ADAMTS5, ADAMTS6, ADAMTS15, ADAMTS18, and

Table 1. Common or alternative names for members of the matrixmetalloproteinase (MMP) and ADAMTS families*

MMP-1 Interstitial collagenase; collagenase 1MMP-2 Gelatinase A, 72-kd gelatinase; type IV collagenaseMMP-3 Stromelysin 1; transin; proteoglycanaseMMP-7 Matrilysin; PUMP-1MMP-8 Neutrophil collagenase; collagenase 2MMP-9 Gelatinase B; 92-kd gelatinaseMMP-10 Stromelysin 2; transin 2MMP-11 Stromelysin 3MMP-12 Macrophage elastase; metalloelastaseMMP-13 Collagenase 3MMP-14 MT1-MMPMMP-15 MT2-MMPMMP-16 MT3-MMPMMP-17 MT4-MMPMMP-19 RASI-1MMP-20 EnamelysinMMP-21 –MMP-23 CA-MMPMMP-24 MT5-MMPMMP-25 MT6-MMP; leukolysinMMP-26 Matrilysin 2; endometaseMMP-27 hMMP-22MMP-28 Epilysin

ADAMTS-1 KIAA1346; METH-1; aggrecanase 3ADAMTS-2 Procollagen N-propeptidaseADAMTS-3 KIAA0366; procollagen N-propeptidaseADAMTS-4 KIAA0688; ADMP-1; aggrecanase 1ADAMTS-5 ADMP-2; aggrecanase 2ADAMTS-6 –ADAMTS-7 –ADAMTS-8 METH-2ADAMTS-9 KIAA1312; aggrecanaseADAMTS-10 –ADAMTS-12 –ADAMTS-13 VWFCPADAMTS-14 Procollagen N-propeptidaseADAMTS-15 AggrecanaseADAMTS-16–20 –

* CA � cysteine array; h � human; ADMP � aggrecan-degrading metallopro-teinase; VWFCP � Von Willebrand factor–cleaving protease.

Table 2. Genes with altered expression in hip cartilage from patientswith osteoarthritis compared with normal cartilage from patients withfemoral neck fracture

Alteration, gene P

Increased expressionMMP2 �0.05MMP9 �0.01MMP13 �0.001MMP16 �0.01MMP28 �0.001TIMP3 �0.05ADAMTS2 �0.01ADAMTS12 �0.05ADAMTS14 �0.01ADAMTS16 �0.001

Decreased expressionMMP1 �0.001MMP3 �0.001MMP10 �0.01TIMP1 �0.01TIMP4 �0.05ADAMTS1 �0.001ADAMTS5 �0.05ADAMTS9 �0.01ADAMTS15 �0.05

134 KEVORKIAN ET AL

ADAMTS19 were highly expressed in all samples pro-filed.

Regulation of gene expression in OA. Table 2summarizes the significant differences in expression ofall genes profiled in hip cartilage from OA patientscompared with normal cartilage. Figure 2 shows scatter-plots for the 6 genes in which highly (P � 0.001) alteredexpression was found.

Expression of MMP genes. The most significantincreases in expression of MMPs in cartilage from OApatients compared with normal were for MMP13 andMMP28 (P � 0.001), while the most significant decreaseswere for MMP1 and MMP3 (P � 0.001). Of the classiccollagenases, MMP8 was undetectable in most cartilage

samples (24 of 33), while MMP1 expression was signifi-cantly reduced (P � 0.001) and MMP13 expressionsignificantly increased (P � 0.001) in OA cartilage, asindicated above. Of the other MMPs shown to cleave thecollagen triple helix, MMP2 expression was increasedsignificantly (P � 0.05) in cartilage from OA patients,while MMP14 expression did not differ significantly fromthat in normal cartilage. Of the stromelysins, MMP3 wassignificantly decreased in OA cartilage (P � 0.001);MMP10 was also decreased (P � 0.01), while MMP11was unaltered. The 2 matrilysins, MMP7 and MMP26,were also unchanged in OA. Both gelatinases weresignificantly increased in OA cartilage (P � 0.05 forMMP2; P � 0.01 for MMP9). Of the membrane-type

Figure 2. Comparative expression of genes for MMPs and ADAMTS in OA and normal hipcartilage. Genes whose expression was highly (P � 0.001) altered in OA are shown (no TIMP geneshowed highly altered expression in OA). The expression level of each gene was determined asdescribed in Materials and Methods and normalized to the level of 18S ribosomal RNA geneexpression. Because units are arbitrary and primer/probe dependent, they cannot be directlycompared between genes. Horizontal bars show the means. See Figure 1 for definitions.

MMP, TIMP, AND ADAMTS GENE EXPRESSION IN OA 135

MMPs, only MMP16 was altered in OA, being signifi-cantly increased (P � 0.01) compared with normalcartilage.

No significant difference between OA and nor-mal cartilage was observed for expression of MMP7,MMP11, MMP12, MMP14, MMP15, MMP17, MMP19,MMP23, MMP24, MMP25, MMP26, and MMP27. How-ever, MMP12 and MMP17 showed trends toward de-creased expression and increased expression, respec-tively, in OA, which almost reached statistical significance.

Expression of ADAMTS genes. The most signifi-cant increase in expression of any ADAMTS in cartilagefrom OA patients compared with normal cartilage was forADAMTS16 (P � 0.001), while the most significant de-crease was for ADAMTS1 (P � 0.001). The known aggre-canases are ADAMTS1, ADAMTS4, ADAMTS5,ADAMTS9, and ADAMTS15; in addition, ADAMTS8 andADAMTS20 are potential aggrecanase candidates sincethey clustered in a phylogenetic tree published by Cal et al(12). All of these genes except ADAMTS8 were expressed

in cartilage. Of the expressed aggrecanases, ADAMTS1,ADAMTS5, ADAMTS9, and ADAMTS15 were all signifi-cantly decreased in cartilage from OA patients comparedwith normal cartilage (P � 0.001 for ADAMTS1; P � 0.01for ADAMTS9; P � 0.05 for ADAMTS5 and ADAMTS15),while ADAMTS4 showed a trend toward the same pattern,but not reaching statistical significance. The procollagenpropeptidases ADAMTS2 and ADAMTS14 were signifi-cantly increased in OA cartilage (both P � 0.01), whileADAMTS3 was unaltered. Of the remaining ADAMTSgenes, ADAMTS12 and ADAMTS16 were both signifi-cantly increased in OA cartilage (P � 0.05 and P � 0.001,respectively), while the expression of ADAMTS6,ADAMTS10, ADAMTS13, ADAMTS17, ADAMTS18,ADAMTS19, and ADAMTS20 did not differ between OAand normal cartilage and ADAMTS7 was not expressed.

Expression of TIMP genes. The expression ofTIMP1 and TIMP4 was significantly decreased in carti-lage from OA patients compared with normal cartilage

Table 3. Analysis of the correlation of expression between genes*

TS1 TS2 TS4 TS5 TS9 TS10 TS12 TS15 TS16 TS17 TS18 TS19 TS20 M1 M2

TS1 �0.001 �0.001 �0.001 �0.001TS2 �0.05 �0.001 �0.01 �0.001TS4 �0.001 �0.001 �0.05 �0.05 �0.01 �0.01TS5 �0.01 �0.05 �0.001 �0.01TS9 �0.05 �0.001TS10TS12 �0.01 �0.001TS15 �0.01 (–) �0.001 �0.001TS16 �0.05 (–) �0.001TS17 �0.01 (–) �0.05TS18 �0.05TS19 �0.05TS20M1M2M3M9M11M12M13M14M16M17M19M21M23M26M27M28T1T2T3

* The expression of each gene in all cartilage samples (osteoarthritic and normal combined) was compared and correlated with the expression of all other genes, bySpearman’s rank correlation. P values of significant correlations are shown, for all genes that had at least 1 correlation at the P � 0.001 level. TS � ADAMTS genes;M � matrix metalloproteinase genes; T � tissue inhibitor of metalloproteinases genes.

136 KEVORKIAN ET AL

(P � 0.01 and P � 0.05, respectively). TIMP3 expressionwas increased in OA cartilage (P � 0.05), while nostatistically significant difference was observed for ex-pression of TIMP2.

Coordinate regulation of genes. The correlationbetween the level of expression of each gene assayed andthe levels of all other genes assayed was analyzed usingSpearman’s rank correlation. Table 3 shows correlationsfor all genes (in the total data set, from both OA andnormal cartilage) that had at least 1 correlation at the P� 0.001 significance level. Using correlations predomi-nantly at P � 0.001, it can be seen that ADAMTS1,ADAMTS4, ADAMTS5, and ADAMTS9 were coex-pressed, along with MMP1 and MMP3 in most instances.By the same criteria, ADAMTS2 and ADAMTS12 werecoexpressed, with MMP2, MMP9, MMP13, and MMP28,along with ADAMTS16 and MMP17 in most cases.Finally, ADAMTS15, ADAMTS19, ADAMTS20, MMP12,MMP19, MMP21, MMP23, MMP26, and MMP27 were allcoexpressed, along with MMP11 in most cases. There

was a strong negative correlation between MMP3 andMMP9 expression and between MMP3 and ADAMTS16expression, and ADAMTS17 was negatively correlatedwith MMP26 and with MMP27. Figure 3 shows scatter-plots of correlations between representative pairs ofgenes in the above groups. Findings of principal compo-nents analysis, a multivariate analysis of the whole dataset, reinforced the groupings described above.

DISCUSSION

The findings of initial experiments in this studyhighlight the need to purify RNA from cartilage on theday of tissue collection, or as soon as possible thereafter.The first time point at which we purified RNA afterstorage was 7 days, and we found a substantial dropoff inyield over this very short period. It is, of course, possiblethat other storage conditions may result in more effec-tive stabilization of RNA, but based on our preliminaryfindings we purified RNA within 24 hours of surgical

Table 3. (Cont•d)

M3 M9 M11 M12 M13 M14 M16 M17 M19 M21 M23 M26 M27 M28 T1 T2 T3

TS1 �0.001 �0.01TS2 �0.05 (–) �0.001 �0.01 �0.05 �0.01 �0.001 �0.05 �0.01 �0.001 �0.05TS4 �0.05TS5 �0.05 �0.01TS9 �0.001 �0.05 (–) �0.05 (–) �0.05TS10 �0.01 �0.05 �0.01 �0.05 �0.001TS12 �0.001 �0.001 �0.05 �0.001 �0.01TS15 �0.01 �0.001 �0.05 (–) �0.05 �0.01 �0.001 �0.001 �0.001 �0.001 �0.05 (–)TS16 �0.001 (–) �0.001 �0.001 �0.001 �0.05 �0.001 �0.01TS17 �0.05 (–) �0.05 �0.01 (–) �0.001 (–) �0.001 (–) �0.05TS18 �0.05 �0.001 �0.05TS19 �0.01 �0.05 �0.01 �0.001 �0.001TS20 �0.001 �0.05 (–) �0.05 �0.001 �0.05 �0.001 �0.001M1 �0.001 �0.05 �0.001M2 �0.001 �0.001 �0.05 �0.01 �0.001 �0.05 �0.05M3 �0.001 (–) �0.01 (–) �0.05 �0.01 (–) �0.01 (–) �0.001M9 �0.001 �0.01 �0.05 �0.001M11 �0.001 �0.05 �0.001 �0.01 �0.05 �0.01M12 �0.05 (–) �0.01 �0.001 �0.001 �0.001 �0.001 �0.001M13 �0.01 �0.05 (–) �0.001 �0.05 �0.05M14 �0.001 �0.05 �0.05 �0.05 �0.05M16 �0.05 �0.001 �0.05M17 �0.05 �0.05 �0.01M19 �0.001 �0.001 �0.01 �0.001M21 �0.001 �0.001 �0.001M23 �0.01 �0.01M26 �0.001M27M28 �0.01 �0.01T1 �0.001T2 �0.001T3

MMP, TIMP, AND ADAMTS GENE EXPRESSION IN OA 137

removal of the femoral head in the remainder of thestudy, allowing us to obtain reproducible yields.

This study is the first to profile the expression of

all genes from the MMP, ADAMTS, and TIMP familiesin tissue from a cohort of patients with arthritis, andcompare the results with findings in normal samples.The simultaneous profiling of all of these genes poten-tially allows for better targeting of antiproteinase ther-apies, provides information that better enables us toexamine the mechanisms controlling expression of thegenes, and identifies proteinases that have not previ-ously been implicated in cartilage metabolism. The useof OA cartilage from the hip as being representative ofOA cartilage in general and the use of patients withfracture as the source of normal cartilage are justified bythe predominantly similar patterns of gene expression incartilage observed in a preliminary study of postmortemspecimens and specimens from patients with knee OA(see below). It should be emphasized that the OAcartilage in our study was from patients with end-stagedisease, and gene expression patterns may be distinctfrom those occurring at disease initiation or early inprogression.

Examination of collagenase MMPs showed thatMMP8 is unlikely to play a role in cartilage destructionin OA; in most samples the expression of this gene wasundetectable, even after 40 cycles of PCR. MMP13 wasincreased in OA cartilage, consistent with the currentbelief that it is the predominant collagenase in OA andwith its substrate preference for type II collagen(9,10,30). Interestingly, MMP1 expression was decreasedin cartilage from patients with hip OA, suggesting eitherthat this collagenase has no role in cartilage collagendegradation in OA of the hip, or that its role is earlier inthe disease process. In a smaller preliminary study of afurther 6 samples of cartilage from normal and OA hip,we also observed these patterns of MMP1 and MMP13expression (27).

We also conducted a preliminary study using OAknee cartilage and normal knee cartilage and found thatboth MMP1 expression and MMP13 expression weresignificantly raised in the OA samples (P � 0.01 and P �0.05, respectively). This suggests that regulation of theMMP1 gene may be different in cartilage from the kneecompared with the hip, and the role of MMP-1 in thedisease process may therefore be distinct in each joint.Another recent study showed no significant change inMMP1 expression in OA knee cartilage compared withnormal knee cartilage (31). This latter study used real-time PCR technology, similar numbers of subjects as inthe current study, GAPDH as a housekeeping gene fornormalization, and a different (though similarly vali-dated) primer and probe set.

Of the other MMPs that have collagenolytic

Figure 3. Correlation analysis of gene expression. Plots show repre-sentative samples of the correlation between genes in each of the 3groups of coexpressed genes (see Results). Correlation coefficientswere as follows: ADAMTS5 versus ADAMTS1 0.846, MMP2 versusADAMTS12 0.852, MMP21 versus MMP12 0.894 (P � 0.001 for allpairs of genes shown). See Figure 1 for definitions.

138 KEVORKIAN ET AL

activity, MMP2 expression is significantly increased inOA, although there is no evidence that it functions as acollagenase. MMP14 is highly expressed, but not regu-lated in disease, suggesting a possible role in tissuemaintenance rather than destruction.

The most highly statistically significant change inexpression of any gene in late-stage hip OA cartilagecompared with normal cartilage was that of MMP3.Expression of MMP3 was decreased in diseased tissue,consistent with the findings of Bau et al (31) in knee OAcartilage. Both this and the decrease in MMP1 expres-sion in cartilage from patients with hip OA are surpris-ing in view of the reported increase in both genes in RA.Levels of MMP-1 and MMP-3 are increased in rheuma-toid synovium (e.g., by immunohistochemistry) (32), andalso in synovial fluid (33,34) and serum (35,36). Instudies in which neoepitope antibodies were used toexamine aggrecan turnover in human cartilage, theconcentration of the MMP-generated VDIPEN neo-epitope was constant in normal cartilage from donorsages �25 years and older (23). In cartilage from OApatients there was a trend toward increased staining forVDIPEN with greater severity of disease. Such data aredifficult to ascribe to any one MMP since many MMPswill cleave the susceptible Asn–Phe bond to create theVDIPEN neoepitope (37). Also of note is the fact thatMMP3 is the most highly expressed MMP gene innormal cartilage and its levels remain comparativelyhigh even in OA. This is in accordance with datareported by Bau et al (31) and may reflect a mainte-nance function of this enzyme in normal cartilage me-tabolism, which is dysregulated in OA.

MMP10 and MMP11, the other so-called strome-lysins, are expressed at a much lower level than MMP3,with MMP10 also decreased in OA cartilage comparedwith normal. MMP12 (metalloelastase), an enzyme pre-viously shown to be expressed by chondrocytes (38), ishighly expressed in normal cartilage and tends towardbeing decreased in OA cartilage, although the differencecompared with normal cartilage is not statistically signif-icant.

The expression of MMP7 was not significantlyaltered in OA cartilage in the current study; this iscontrary to the results of an earlier study using conven-tional RT-PCR (39). Among the OA samples we stud-ied, only 3 of 18 had raised MMP7 expression.

Expression of both MMP2 and MMP9 (the gela-tinases) was significantly increased in OA cartilage. Thisis consistent with findings of previous studies in bothhuman tissue and animal models of OA (11,40,41).

Of the membrane-type MMPs, only the expres-

sion of MMP16 (MT3-MMP) was altered (increased) inOA. Interestingly, the expression of MMP16 has beenshown to increase progressively in an in vitro model ofchondrogenesis from human adult stem cells (42).Hence, this enzyme may have a role in chondrocytedifferentiation. As stated above, MMP14 is the mosthighly expressed gene of this group, although it is notregulated in OA; this is in accordance with earlierfindings (31). MMP-14–deficient mice exhibit musculo-skeletal defects related to their inability to turn overcollagen at important stages of development (43,44);however, it seems unlikely that this is an important rolefor this enzyme in OA.

Of the remaining MMPs, the expression ofMMP28 was significantly increased in OA cartilage.MMP-28 (epilysin) has been implicated in wound repair,being produced by proliferating keratinocytes upon in-jury to the epidermis (45). The role of MMP-28 in theOA disease process is unknown and, since its substratesare also currently unknown, it is difficult to speculateupon this. Importantly, the expression of MMP28 wasalso significantly (P � 0.01) increased in cartilage frompatients with knee OA in our preliminary study.

Examination of genes encoding the ADAMTSenzymes that have been shown to have aggrecanaseactivity showed that the expression of ADAMTS1,ADAMTS5, ADAMTS9, and ADAMTS15 was decreasedin cartilage from OA patients compared with normalcartilage. ADAMTS4 followed the same trend, but notreaching statistical significance. This pattern of expres-sion was replicated in our preliminary study of cartilagefrom patients with knee OA compared with normalcartilage. In hip cartilage, the overall levels of expressionof ADAMTS1, ADAMTS5, and ADAMTS15 were simi-lar, and higher than those of ADAMTS4 and ADAMTS9;this is similar to findings in the preliminary study usingknee cartilage, except that the expression of ADAMTS15was lower in knee cartilage than in hip cartilage. Aprevious study has shown a small but significant increasein ADAMTS4 and ADAMTS5 expression in cartilagefrom patients with end-stage knee OA compared withnormal cartilage (31); however, that report does statethat “. . .no strong, highly significant increase in anyaggrecan-degrading protease . . . could be identified.”

Two possible explanations for our results are 1)that the OA cartilage used was from patients withend-stage disease and aggrecan loss may be at its highestin the earlier stages; and 2) that these ADAMTS genesare not regulated strongly at the level of steady-statemRNA. This latter possibility is supported by the find-ings of other studies demonstrating, e.g., that the activityof ADAMTS-4 is induced by interleukin-1 without a

MMP, TIMP, AND ADAMTS GENE EXPRESSION IN OA 139

concomitant increase in the level of ADAMTS-4 protein(46). However, it has also been shown that a steady-statelevel of mRNA for ADAMTS4 and ADAMTS5 can beinduced by cytokines that induce loss of aggrecan fromcartilage (47). One aspect of our future work will be toinvestigate the temporal expression of these genes inSTR/ort mice, to explore changes in gene expressionduring early disease in a model of spontaneous OA.

ADAMTS2 and ADAMTS14 were both signifi-cantly increased in OA cartilage. These genes are bothprocollagen N-proteinases (17), processing the propep-tides of fibrillar collagens to generate collagen moleculeswhich assemble into fibrils. The increase in expression ofthese genes reflects the increase in collagen synthesis bychondrocytes in OA cartilage (48), presumably in anattempt to repair the cartilage matrix. The expression ofADAMTS12 and ADAMTS16 was also elevated in OAcartilage, but the function of these enzymes is as yetunknown.

Of the TIMP genes, TIMP1 was the most highlyexpressed in cartilage. Particularly TIMP1, but alsoTIMP4 expression, was significantly reduced in OAcartilage. TIMP3 was significantly increased in OA, withTIMP2 following the same trend but without reachingstatistical significance. The destruction of cartilage ECMin OA is driven by an imbalance between proteinase andinhibitor activities (6,26), and our data indicate that thereduction in TIMP-1 and TIMP-4 levels may play a keyrole in this process.

Correlation analysis yielded fascinating insightsinto coregulation of sets of genes within the MMP andADAMTS families. Aggrecanase enzymes ADAMTS1,ADAMTS4, ADAMTS5, and ADAMTS9 were all coex-pressed (and therefore potentially coregulated), and thisseparates them from ADAMTS15, the other knownaggrecanase, even though the expression of all of thesegenes was reduced in OA cartilage. Several of the genesthat were increased in OA cartilage were alsocoexpressed, with strong correlation between levelsof MMP2, MMP9, MMP13, MMP28, ADAMTS2,ADAMTS12, and ADAMTS16. MMP16 and ADAMTS14were increased in OA, but did not correlate with thisgroup. Finally, the levels of a large group of genes thatwere predominantly unaltered in OA compared withnormal cartilage also correlated strongly, although these(ADAMTS15, ADAMTS19, ADAMTS20, MMP11,MMP12, MMP9, MMP21, MMP23, MMP26, andMMP27) have no obvious functional connection. Thecoregulation of members of the ADAMTS gene family isnot the same in all tissues. Our recent study of ADAMTSexpression in human breast carcinoma showed coexpres-

sion of ADAMTS1, ADAMTS3, ADAMTS5, ADAMTS8,ADAMTS9, ADAMTS10, and ADAMTS18 and also ofADAMTS2, ADAMTS4, ADAMTS6, ADAMTS7,ADAMTS12, and ADAMTS14 (29).

This is the first comprehensive screen of all of theknown MMP, ADAMTS, and TIMP genes in cartilage,providing an interesting comparison between normalcartilage and that from patients with end-stage osteoar-thritis. Data from this study form a foundation on whichto build knowledge of the details of how these genes areregulated in OA, to define functions of the more re-cently discovered gene products, and potentially torefine the specificity of antiproteolytic therapies.

ACKNOWLEDGMENTS

The authors would like to thank Jo Price, Rob Nuttall,Caroline Pennington, Mark Thomas, and Mark Downey-Jonesfor their invaluable help and advice on aspects of this study.

REFERENCES

1. Petersson IF, Jacobsson LT. Osteoarthritis of the peripheral joints.Best Pract Res Clin Rheumatol 2002;16:741–60.

2. Karlson EW, Mandl LA, Aweh GN, Sangha O, Liang MH,Grodstein F. Total hip replacement due to osteoarthritis: theimportance of age, obesity, and other modifiable risk factors. Am JMed 2003;114:93–8.

3. Prockop DJ. What holds us together? Why do some of us fallapart? What can we do about it? Matrix Biol 1998;16:519–28.

4. Hardingham TE, Fosang AJ. Proteoglycans: many forms and manyfunctions. FASEB J 1992;6:861–70.

5. Eikenberry EF, Bruckner P. Supramolecular structure of cartilagematrix. In: Seibel MJ, Robins SP, Bilezikian JP, editors. Dynamicsof cartilage and bone metabolism. San Diego: Academic Press;1999. p. 289–300.

6. Murphy G, Knauper V, Atkinson S, Butler G, English W, HuttonM, et al. Matrix metalloproteinase in arthritic disease. ArthritisResearch 2002;4:39–49.

7. Knauper V, Lopez-Otin C, Smith B, Knight G, Murphy G.Biochemical characterization of human collagenase-3. J BiolChem 1996;271:1544–50.

8. Ohuchi E, Imai K, Fujii Y, Sato H, Seiki M, Okada Y. Membranetype 1 matrix metalloproteinase digests interstitial collagens andother extracellular matrix macromolecules. J Biol Chem 1997;272:2446–51.

9. D’Ortho MP, Will H, Atkinson S, Butler G, Messent A, GavrilovicJ, et al. Membrane-type matrix metalloproteinases 1 and 2 exhibitbroad-spectrum proteolytic capacities comparable to many matrixmetalloproteinases. Eur J Biochem 1997;250:751–7.

10. Konttinen YT, Ainola M, Valleala H, Ma J, Ida H, Mandelin J, etal. Analysis of 16 different matrix metalloproteinases (MMP-1 toMMP-20) in the synovial membrane: different profiles in traumaand rheumatoid arthritis. Ann Rheum Dis 1999;58:691–7.

11. Tetlow LC, Adlam DJ, Woolley DE. Matrix metalloproteinase andproinflammatory cytokine production by chondrocytes of humanosteoarthritic cartilage: associations with degenerative changes.Arthritis Rheum 2001;44:585–94.

12. Cal S, Obaya AJ, Llamazares M, Garabaya C, Quesada V,Lopez-Otin C. Cloning, expression analysis and structural charac-terization of seven novel ADAMTSs, a family of matalloprotein-