Hyaluronan-oligosaccharide-induced transcription of metalloproteases

Metastasising tumour cells share many common properties with embryonic cell populations such as the neural crest and cells of the adult immune system such as activated dendritic cells and lymphocytes, macrophages and neutrophils (reviewed in Sherman et al., 1996). These cells share the ability to migrate over considerable distances and to invade other tissues. Epidermal Langerhans cells provide a good example. Once these dendritic cells are activated with antigen in the epidermis, they invade and migrate through the dermis, enter lymphatic capillaries and thereby travel to the regional lymph nodes, where they present antigen to T cells (reviewed in Lappin et al., 1996). Similarly, metastatic carcinoma cells invade the surrounding tissue, enter the lymphatic system and travel to regional lymph nodes, where they form secondary tumours (reviewed in Sleeman, 2000).

The shared properties of normal cells and cancer cells are certainly based on common programs of gene expression, and it is reasonable to hypothesise that portions of the expression profiles are common. Indeed, we have previously demonstrated that splice variants of the cell surface hyaluronate receptor CD44 that play a crucial role in tumour metastasis are also required for epidermal Langerhans cell function (Weiss et al., 1997). Furthermore, matrix metalloproteases (MMPs) such as MMP-9 and MMP-13 play important roles in tumour growth and metastasis (Stamenkovic, 2000; Leeman et al., 2002). Similarly, MMP-9 and MMP-13 production is upregulated upon dendritic cell activation (Kobayashi, 1997; Chen et al., 2002). In the case of MMP-9, it has been shown that MMP-9 activity is required for invasion and migration of activated dendritic cells into regional lymph nodes (Kobayashi et al., 1999; Noirey et al., 2002).

In addition to being activated by antigen, it has recently been demonstrated that dendritic cells can be activated by degradation products of the extracellular matrix component hyaluronan (HA), but not by high-molecular-weight HA (Termeer et al., 2000). High-molecular-weight HA is synthesised and accumulated as an extracellular matrix component by most cells, particularly during proliferation (reviewed by Sherman et al., 1994; DeAngelis, 1999). RHAMM/IHABP and CD44 are specific cellular receptors for HA (reviewed in Hofmann et al., 1998). HA turnover occurs constantly but is enhanced in areas of inflammation. During inflammation, activated fibroblasts secrete hyaluronidases that degrade high-molecular-weight HA into fragments (Weigel et al., 1986; Sampson et al., 1992). This degradation can also be augmented by reactive oxygen species produced at sites of inflammation (Moseley et al., 1997; Agren et al., 1997). The small HA fragments so produced activate dendritic cells through Toll-like receptor 4 (TLR-4) (Termeer et al., 2002).

The responsiveness of dendritic cells to small HA oligosaccharides might again find its counterpart in tumours. Mounting evidence suggests that degradation of HA in tumours is associated with invasion and metastasis (Kumar et al., 1989; Godin et al., 2000; Beech et al., 2002; Patel et al., 2002). Because HA fragments activate dendritic cells and activated dendritic cells express MMP-9 and MMP-13, we therefore investigated whether HA and its degradation products can induce MMP synthesis in tumour cells.

We report here that exposure of cells to HA oligosaccharides strongly induced the expression of the metalloproteases MMP-9 and MMP-13 in Lewis Lung Carcinoma (3LL) cells and primary embryonic fibroblasts. Induction of MMP synthesis by HA oligosaccharides is mediated by an as yet unknown HA receptor that is not identical with RHAMM/IHABP, CD44 or TLR-4. NF-κB signalling pathways are activated during this induction. These data suggest that HA degradation in tumours or in areas of inflammation might promote invasion or extracellular matrix (ECM) remodelling by activating MMP expression.

3LL cells (Suguira and Stock, 1955) were grown at 37°C and 6% CO₂ in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% foetal calf serum (FCS), 100 U ml^-1 penicillin, 100 μg ml^-1 streptomycin. Primary murine embryonic fibroblast (MEF) cells were isolated from day (d) 13.5-14 embryos of wild-type (C57BL/6), CD44-negative (C67BL6/J) (Protin et al., 1999) and RHAMM-deficient mice lacking the HA binding domain (C57BL/6) (J. Moll, unpublished). The cells were cultivated at 37°C and 6% CO₂ with 10% FCS-DMEM, supplemented with 2 mM L-glutamine, 100 U ml^-1 penicillin/streptomycin. Dermal fibroblasts were cultured from the ears of C57BL/10ScSn (wild type) and C57BL/10ScCr-(TLR 4^-/-) mice (kindly provided by C. Galanos, Max Planck Institute for Immunobiology, Freiburg, Germany) (Poltorak et al., 1998). Briefly, the ears were floated for 3 hours on 0.5% dispase (Roche Diagnostics, Mannheim, Germany) in Hanks Balanced salt Solution (HBSS) without Ca²⁺, Mg²⁺ containing 0.025 M HEPES buffer (Invitrogen, Karlsruhe, Germany) at pH 7.0 and 37°C. The epidermis was mechanically removed and the dermis was incubated in 0.4% collagenase (Roche Diagnostics, Mannheim, Germany) in HBSS (Invitrogen) overnight. After washing twice in PBS, the single cell suspension was passed through a 70 μm filter and the fibroblasts were seeded into 50 ml tissue culture flasks at 37°C in Minimum Essential Medium (MEM) with Earle's salts containing 10% foetal bovine serum (FBS), 2 mM L-glutamine, 1% w/v penicillin/streptomycin (all Invitrogen). After a minimum of two passages, homogenous confluent populations of fibroblasts were used for experiments.

Cells were starved for 12 hours before incubation with different HA preparations. The sources of HA used were Healon™ (Pharmacia, Erlangen, Germany) and rooster comb (Sigma, Deisenhofen, Germany), which was autoclaved before use. For some experiments cells were preincubated with the proteosome inhibitor N-tosyl-L-phenylalanine-chlormethyl-ketone (TPCK) (Calbiochem) at a concentration of 10 μM for 30 minutes, 0.3 mM Suramin (Bayer Leverkusen) for 45 minutes or Actinomycin D-Mannitol (2 μg ml^-1; Sigma) for 30 minutes before stimulation with HA, chondroitin sulfate A (Sigma) (see figure legends for concentrations used), 500 ng ml^-1 tumour necrosis factor (TNF) α (Sigma) or 80 ng ml^-1 12-O-tetradecanoylphorbol 13-acetate (TPA) (Sigma). The cells were harvested routinely after 9-12 hours. In the experiments using emetine, cells were pretreated for 45 minutes with 50 μM emetine before being stimulated for 10 hours with 0.5 mg ml^-1 HA and/or 80 ng ml^-1 TPA.

For blocking the HA-binding activity of CD44, 3LL cells were pretreated with 100 μg ml^-1 KM81 antibody (IgG) or with 50 μg ml^-1 KM81 (F(ab′)₂) for 45 minutes on ice. They were then plated with HA-containing medium containing 100 μg ml^-1 KM81 antibody or 50 μg ml^-1 KM81 F(ab′)₂.

Gelatinolytic activity was assayed by 8% SDS-PAGE containing 1 mg ml^-1 of copolymerised gelatin. The conditioned medium samples were collected after 48 hours and concentrated by precipitation using 12.5% trichloroacetic acid (TCA) (Sigma, Deisenhofen, Germany). The precipitates were resolubilised in Laemmli buffer and electrophoresed under nonreducing conditions. Gels were then washed twice for 30 minutes in 2.5% Triton X-100 to remove SDS and subsequently in H₂O only. The gels were then incubated at 37°C overnight in 50 mM Tris-HCl (pH 7.5), 10 mM CaCl₂, 1 μM ZnCl₂. Gels were stained with Coomassie Blue and clear zones of substrate lysis against a blue background stain indicated the presence of gelatin-degrading enzymes.

Isolation of polyadenylated RNA (poly(A)+ RNA) was performed as previously described (Rahmsdorf et al., 1987). RNAs were heat denatured in the presence of formamide and 4 μg of each sample run on 1.2% agarose gels containing 6% formaldehyde. After blotting on nylon membrane (Hybond N+, Amersham Bioscience) filters were hybridised under stringent conditions (Church and Gilbert, 1984) by using radiolabelled double-stranded cDNA probes comprising either an EcoRI/HindIII fragment of mouse Mmp-9 (Schorpp-Kistner et al., 1999), a PstI/SacII fragment of mouse Mmp-13 (Gack et al., 1994), a PstI fragment of mouse urokinase-type plasminogen activator (Sommer et al., 1987), a PstI fragment of mouse Timp-1 (Docherty et al., 1987) or a PstI fragment of rat Gapdh (Fort et al., 1985).

Total RNA was prepared from cells following standard procedures (Chomzcynski and Sacchi, 1987). RNA was reverse transcribed using oligo d(T) as a primer and Superscript Reverse Transcriptase (Gibco) (RT-PCR).

cDNAs specific for the genes analysed were amplified using the following primers. Mmp-9: 5′-CCTTGGTGTAGCACAACAGC-3′ (position 779-798), 5′-ATACTGGATGCCGTCTATGTCG-3′ (position 1321-1342). The positions refer to mRNA sequence previously described (Masure et al., 1993). Mmp-13: 5′-AGATGTGGAGTGCCTGATGTGG-3′ (position 296-317), 5′-GAGACTGGTAATGGCATCAAGG-3′ (position 880-901). Positions are taken from GenBank accession number X66473. Gapdh: 5′-AGACAGCCGCATCTTCTTGTGC-3′ (position 23-45), 5′-CTCCTGGAAGATGGTGATGG-3′ (position 282-302). The positions are based upon the GAPDH sequence for rattus norvegicus (GenBank accession number X2231). PCR conditions were as follows: initial denaturation for 1 minute at 94°C, then 35 cycles (21 cycles in the case of Gapdh) with 30 seconds at 94°C, 30 seconds at 58°C and 30 seconds at 72°C.

Conditioned medium was analysed for MMP-9 content by using an MMP-9 enzyme-linked immunosorbent assay (ELISA) kit (R&D Systems) according to the manufacturer's instructions.

Nuclear extracts were prepared and used in electrophoretic mobility shift assays (EMSAs) as previously described (Stein et al., 1989). Briefly, cells were washed twice with ice-cold PBS, detached from the plate and suspended in 1 ml PBS. The cells were sedimented and suspended in 40 μl 250 mM Tris-HCl (pH 7.8), 60 mM KCl, 1 mM dithiotreitol (DTT) and 1 mM phenyl methyl sulfonyl fluoride (PMSF). Nuclei were lysed by three cycles of freezing and thawing in liquid nitrogen and ice. The nuclear extracts were cleared by centrifugation at 13,000 g for 15 minutes.

EMSAs were performed as described previously (Stein et al., 1989), with slight modifications. Binding was performed in a volume of 20 μl with 5-8 μg protein extract in a buffer containing 12 mM HEPESKOH pH 7.8, 62.5 mM Tris-HCl (pH 7.8), 60 mM KCl, 0.6 mM EDTA, 6% glycerol, 5 mM DTT, 2 μg BSA and 1 μg poly(dI-dC). Approximately 10 fmol (40,000 cpm) of ³²P-radiolabelled double-stranded oligonucleotide (5′-AGCTTGGGGACTTTCCAGCCG-3′) derived from the human immunodeficiency virus long terminal repeat (HIV-LTR) were used per reaction. In competition experiments the protein extracts were preincubated with a ten- or 100 times excess of unlabelled HIV-LTR oligonucleotide, or for negative controls with an oligonucleotide derived from the AP-1 binding site in the mouse gene encoding Collagenase-I (5′-AGCTAAAGTGGTGACTCATCACTAT-3′). Resultant DNA-protein complexes were resolved on 5% polyacrylamide gels and detected by autoradiography.

Cells were detached from tissue culture dishes by treatment with 2 mM EDTA in PBS (without Ca²⁺, Mg²⁺). Cells were washed once in ice-cold PBS and taken up in staining solution (PBS + 3% FCS) to a density of about 5×10⁶ cells per ml. Staining and washing steps were done at 4°C. For staining of CD44, cells were incubated with 10 μg ml^-1 KM81 or IM7 8.1 (IM7) antibodies in staining solution for 1 hour. Unbound primary antibodies were washed off by three washing steps with 3% FCS-PBS. Bound anti-CD44 antibodies were stained with phycoerythrin (PE)-conjugated goat anti-rabbit antibody (DAKO, Hamburg, Germany) diluted 1:200 in staining solution. Incubation was done for 30 minutes. Finally, cell suspensions were analysed using a FACStar Plus (Becton Dickinson).

For binding studies with FITC-labelled hyaluronan (HA-FITC), cells were incubated with HA-FITC (100 μg ml^-1) on ice for 1 hour. For HA-blocking experiments, the cells were preincubated for 45 minutes with anti-CD44 KM81 monoclonal antibody (mAbs) (20 μg ml^-1) before HA-FITC was added to the incubation.

Hyaluronan for clinical application (Healon, endotoxin content lower than 0.1 ng mg^-1) was either sonified or enzymatically digested with bovine-testis hyaluronidase (Sigma, Deisenhofen, Germany) for 12 hours at 37°C in 1 M sodium acetate buffer pH 5.0. The fragments were separated on a Biogel P10 (Bio-Rad) 3.5×115 cm column overnight. Samples were collected from the column with a Pharmacia/LKB FRAC-100 fraction collector for 12 hours, 20 minutes each fraction. The determination of HA-fragment size in each sample was determined by 8-aminonaphthalene 1,3,6-trisulfonic acid (ANTS)-labelling technique (Lee and Cowman, 1994; Termeer et al., 2000). In brief, 100 μl of each sample was dried in a speed vac™ vacuum drier (Life Sciences International, Frankfurt, Germany) and 5 μl of 0.15 M ANTS and 5 μl 1 M NaCNBH₄ dissolved in dimethylsulfoxide (DMSO) were added (all from Sigma, Deisenhofen, Germany). After 16 hours at 37°C, the samples were dried, resuspended in 50 μl 25% glycerin solution and analysed by 30% acrylamide gel electrophoresis.

The HA concentration of each sample was analysed according to Bitter and Muir (Bitter and Muir, 1960). Samples of 100 μl were added to 600 μl of 0.0025 M bisodium-tetraborate in concentrated H₂SO₄ and stirred for 10 minutes at 90°C. The product was cooled to 4°C and 20 μl 0.1% carbazol in ethanol was added. The sample was stirred again for 10 minutes at 90°C. After staining had developed, the concentration of uronic acid was measured photometrically at 520 nm against distilled water. Fractions were adjusted to a concentration of 1 mg ml^-1.

To investigate whether HA can induce expression of gelatinolytic MMPs such as MMP-9, we incubated 3LL cells with commercially available mixed-size and heat treated HA from rooster comb and then examined the conditioned medium from these cells for proteolytic activity using zymograms. To be certain that any effects were not due to contaminants in the HA, the HA was heat treated before use. Heat-stable endotoxin was found to be less than 0.139 endotoxin units ml^-1. This concentration of endotoxin did not induce gelatinolytic activity (not shown). We found that HA-treated cells were strongly induced to synthesise and secrete gelatin-degrading activity in a dose-dependent manner. Fig. 1A shows conditioned medium from 3LL cells that was incubated with rooster comb HA for 48 hours and was then concentrated and resolved on a gelatin zymogram. Although the 3LL cells produced very little proteolytic activity spontaneously, two bands of secreted gelatin-digesting activity were detected upon HA treatment: a major ∼100 kDa activity and a minor 55-60 kDa activity (Fig. 1A). Another glycosaminoglycan, chondroitin sulfate, could not induce gelatinolytic activity at similar concentrations. No gelatinase activity was detected when the cells were incubated with less than 300 μg ml^-1 HA (data not shown).

To investigate whether HA induced enhanced levels of progelatinolytic activity intracellularly and/or whether HA treatment resulted in release of stored gelatinolytic activity, 3LL cell lysates together with their corresponding conditioned media were analysed for gelatinolytic activity with and without HA treatment. As shown in Fig. 1B, no gelatinolytic activity could be observed in cell lysates with or without HA treatment, even though gelatinolytic activity was strongly induced in the culture medium of HA-treated cells. These data suggest that HA treatment does not cause the release of stored gelatinolytic activity in 3LL cells, but rather induces the synthesis of the enzyme that is released immediately into the medium. The controls also demonstrate that the induced gelatinolytic activity is not present in the added HA.

Gelatin is predominantly digested by metalloproteases. From their sizes, the predominant gelatin-degrading activity in the zymogram in Fig. 1A could represent MMP-9 (92 kDa form), whereas the weaker gelatinolytic activity of around 55-60 kDa could represent either MMP-3, MMP-10 or MMP-13. To confirm the identity of the enzymes and to determine whether the enzyme activities were newly synthesised or activated from presynthesised precursor enzyme, we isolated poly(A)+ RNA at various times after HA addition and performed northern blots (Fig. 1C) and RT-PCR (Figs 2, 3, 4, 5, 6). Both methods revealed enhanced RNA levels of Mmp-9 and Mmp-13 in response to HA. Two MMP-9 gene transcripts of 2.5 kb and 3.2 kb (Masure et al., 1993) were first detectable at 6 hours after HA addition to 3LL cells (Fig. 1C), compatible with the kinetics of appearance of the major HA-induced gelatinolytic activity (data not shown). The amount of Mmp-13 RNA was also elevated in response to HA, as was that of the urokinase plasminogen activator (uPa) (Fig. 1C). Furthermore, the amount of Timp-1 mRNA was slightly increased, albeit from an already high spontaneous level (Fig. 1C). Northern blot analyses did not provide any evidence that Mmp-2, Mmp-3 or Mmp-10 expression is induced in response to HA (not shown). These data strongly suggest that the predominant gelatinolytic activity in the zymograms must be the 92 kDa MMP-9, whereas the weaker activity is MMP-13.

We found that an increase in synthesis and release of MMPs was not a unique property of 3LL tumour cells. For example, primary MEFs expressed almost no detectable MMP-9 and MMP-13 in the absence of HA. However, rooster comb HA induced MMP-9 and MMP-13 expression in these cells even more strongly than in 3LL cells (Fig. 2).

Commercially available rooster comb HA contains HA oligosaccharides of different sizes. By using ANTS labelling techniques (Termeer et al., 2000), we determined that approximately two-thirds of the rooster comb HA preparation contained molecules of 60-400 kDa, whereas the remaining third consisted of low-molecular-weight degradation products (data not shown). To determine whether the size of HA influences MMP induction, we compared high-molecular-weight HA (Healon) with lower-molecular-weight fragments of HA for their ability to induce MMP gene expression. Because MEFs are more sensitive to HA-induced MMP gene expression than 3LL cells, they were used for these experiments. Healon HA of molecular weight above 600 kDa (High HA) cannot induce Mmp-9 or Mmp-13 expression (Fig. 2A). When HA of this size was fragmented by sonification to 60-200 kDa (Int HA) and added to cells, still no expression of MMP genes was observed (Fig. 2A). Healon was digested with hyaluronidase followed by Biogel P-10 fractionation to produce HA fragments of four to six residues in size (Low HA). These fragments proved to be highly effective inducers of both Mmp-9 and Mmp-13 expression (Fig. 2A). Similarly, although 3LL tumour cells express low levels of Mmp-9 and Mmp-13 spontaneously, Low HA strongly enhanced expression (Fig. 2B). These data suggest that small fragments of HA, but not high-molecular-weight HA, can promote MMP gene expression.

Cells might sense the presence of HA oligosaccharides either through a specific surface receptor or through endocytic uptake of the HA into the cell. To distinguish between these two possibilities, we used the receptor blocking ability of Suramin, which can completely block HA-induced MMP gene expression (Fig. 3). At the concentrations of suramin used, receptor-mediated TNFα-induced transcription of genes encoding MMPs was reduced by only 50%, emphasising the likelihood that HA-mediated MMP gene expression is receptor mediated. This finding suggests the existance of one or more cell surface receptors capable of binding to HA oligosaccharides and transducing intracellular signals in response.

3LL cells carry abundant CD44 on their surface (Fig. 4A). FITC-labelled HA was bound to 3LL cells and this binding was abolished by an antibody specific for the CD44 domain carrying the binding motif (Fig. 4A), indicating that CD44 is the major HA receptor in 3LL cells. However, the antibody barely, if at all, affected the induction of Mmp-9 (Fig. 4B). Thus CD44 is certainly not the major receptor responsible for induction of the Mmp-9 promoter. Consistent with this result, embryonic fibroblasts from CD44^-/- mice (Protin et al., 1999), similar to wild-type MEFs, responded well to HA by increasing Mmp-9 and Mmp-13 transcription (Fig. 4C).

We investigated whether two other HA receptors might be responsible for mediating HA-induced MMP gene expression. RHAMM/IHABP is an intracellular HA binding protein but, according to one laboratory, can also appear on the cell surface (reviewed by Hofmann et al., 1998). Embryonic fibroblasts from mice with a disrupted gene encoding RHAMM/IHABP (J. Moll, unpublished) responded to HA with similar efficiency to wild-type fibroblasts (Fig. 4C). TLR-4 is a cell surface receptor for small HA fragments (Termeer et al., 2002). Fibroblasts carrying a disruption of both alleles of TLR-4 can still be induced by HA to produce MMP-9 and MMP-13 (Fig. 5). In these fibroblasts, the induced expression of Mmp-9 was low, but it was clearly detectable. For Mmp-9, these data were confirmed by ELISA, which showed that both wild-type and TLR-4^-/- fibroblasts secreted approximately tenfold more MMP-9 into their culture medium when treated with HA. To confirm that the different receptors do not substitute for each other, we blocked the HA-binding domains of CD44 in TLR-4^-/- cells using antibodies. Induction by HA was unaffected (not shown).

In conclusion, although we favour the idea that a cell-surface receptor is responsible for the signal transduction to the MMP genes, none of CD44, RHAMM/IHABP or TLR-4 seem to be responsible.

Which signal-transduction pathways are responsible for HA-induced MMP gene expression? The HA-dependent induction of Mmp-9 or of Mmp-13 RNA was transcriptional because it could be inhibited by actinomycin (Fig. 6A). Treatment with emetine, an inhibitor of protein synthesis, caused strong enhancement of Mmp-9 but not Mmp-13 RNA levels (Fig. 6C). Further increase in Mmp-9 transcription in response to HA was not discriminated with certainty because of the strong induction upon emetine treatment. However, other experiments suggest that the HA response of the Mmp-9 promoter is direct, not involving synthesis of a mediating transcription factor (see below).

To define the signal-transduction pathway triggered by HA treatment, we first addressed the transcription factors responsible for the induction. The Mmp-9 promoter carries several consensus AP-1-binding sites (Masure et al., 1993) and the enhancement of Mmp-9 RNA levels by protein synthesis inhibitors resembles the activation described for the transcription factors JNK and AP-1. However, AP-1 does not seem to be the HA-responsive factor because phorbol ester treatment led to strongly enhanced Fos (not shown) and Mmp-13 (Fig. 6A) expression, but did not influence Mmp-9 in either embryonic fibroblasts (Fig. 6A) or 3LL cells (not shown). Increased expression of Mmp-13 is mediated by upregulation of FOS and JUN (Porte et al., 1999). In agreement with the notion that AP-1 transcription factors are not involved in the HA response of Mmp-9, 3LL cells express high levels of JUN spontaneously that are not further enhanced by HA (not shown). Furthermore, the response cannot be inhibited by one of several inhibitors of the MEK-ERK and p38 pathways (not shown). Only genestein at 20 μM caused partial inhibition. Also consistent with these results, western blot analyses showed that HA did not activate ERK1 and ERK2, JNK or p38 (not shown).

The Mmp-9 promoter carries a functional NF-κB site, as demonstrated by experiments with inflammatory cytokines (Himelstein et al., 1997; Bond et al., 1998; Ikebe et al., 1998). Consistent with other reports (Oertli et al., 1998), treatment with HA caused activation of NF-κB (p65/p50) in both embryonic fibroblasts (Fig. 7) and 3LL cells (not shown), as shown by band-shift experiments. First signs of specific p65/p50 DNA binding were detected 60 minutes after HA addition. From the data in Fig. 7, it is impossible to determine whether p50 homodimers are also enhanced. Treatment of cells with the proteasome inhibitor TPCK blocked the appearance of p65/p50 (Fig. 7) and the induction of Mmp-9 (Fig. 6B) suggesting that NF-κB is involved in HA-induced transcription. As a control, TPCK also blocked the response to TNFα (not shown).

Together, these data provide evidence that NF-κB rather than AP-1 signalling pathways are responsible for HA-induced MMP expression.

HA-derived oligosaccharides, but not high-molecular-weight HA, deliver a specific signal to tumour cells and fibroblasts that leads to the activation of transcription factors such as NF-κB and to the synthesis and secretion of metalloproteases. This example might represent a larger programme of gene expression that tumour cells and non-transformed cells switch on in areas of inflammation, within tumours or at sites such as lymph nodes where HA is degraded.

Since their discovery as a collagen-degrading activity, the known metalloproteases have not only increased in number (to 24 by 2002) (Overall and Lopez-Otin, 2002) but have also been assigned many functions (Bergers and Coussens, 2000; Vu and Werb, 2000). These include the processing of growth-factor proforms (Stamenkovic, 2000), the release of growth factors from the extracellular matrix (Whitelock et al., 1996), cleavage of plasminogen (which releases angiostatin) (Pozzi et al., 2000) and roles in several complex physiological processes (reviewed by Vu and Werb, 2000). The coordinated upregulation of Mmp-9 and Mmp-13 in response to small HA fragments indicates the possibility that these oligosaccharides induce collagen degradation in the context of tumours. MMP-13 is an interstitial collagenase that can initiate the degradation of fibrillar collagen. The resulting cleavage products can then be further degraded by MMP-9. However, in addition to its ability to degrade gelatin and basement membrane collagen type IV, MMP-9 can also cleave the TGFβ and IL-1β proforms (Yu and Stamenkovic, 2000; Schönbeck et al., 1998), might produce angiostatin (Pozzi et al., 2000) and can inactivate the α1-proteinase inhibitor (Sires et al., 1994). MMP-13 can also digest tenascin, aggrecan core protein and fibronectin, as well as a broad range of collagen types. These data suggest that MMP-9 and MMP-13 production in response to fragmented HA might lead to a range of consequences that are highly relevant to the processes of invasion and metastasis.

Many studies provided correlative and functional evidence for MMP-9 having a role in tumour invasion and metastasis. Thus, MMP-9 has repeatedly been reported to be upregulated in metastatic cancer (Stearns et al., 1993; Sato and Seiki, 1993; O-Charoenrat et al., 2000; Horikawa et al., 2000; Gong et al., 2000; Papathoma et al., 2001; Nestl et al., 2001). Recent experiments using Mmp-9-null mice also support a role for MMP-9 in the early stages of tumour growth and angiogenesis (reviewed in McCawley and Matrisian, 2001). Mmp-13 expression serves as a progression marker for a number of human tumour types and is strongly associated with invasive tumours (Brinckerhoff et al., 2000).

We have observed that 3LL tumour cells can produce MMP-9 and MMP-13 in response to small HA fragments. In the context of tumours, significant evidence also suggests a tumour-induced production of these MMPs in stromal cells (reviewed in Powell and Matrisian, 1996). Thus, it is significant that we observed MMP-9 and MMP-13 production induced by HA degradation products in fibroblasts, the major cellular constituents of the tumour stroma.

3LL tumour cells can also produce the serine protease uPa in response to small HA fragments (Fig. 1). uPA is involved in invasion and metastasis (reviewed in Sheng, 2001) and cleaves plasminogen to generate plasmin. In turn, plasmin is a broad-acting serine protease that can degrade fibrin, fibronectin and laminin, and can also activate latent MMPs. A picture therefore emerges in which HA degradation products co-ordinately activate the expression of a range of proteases that are capable of matrix remodelling.

In dendritic cells several pathways are switched on after HA-induced TLR-4 activation. These include MAP kinase and NF-κB pathways that are essential for dendritic cell maturation and TNFα release (Termeer et al., 2002). Our data are also consistent with the notion of a pathway that is activated by fragmented HA, involves NF-κB and leads to Mmp-9 transcription. Furthermore, others have shown that HA fragments induce expression of nitric oxide synthase through NF-κB (McKee et al., 1997).

The receptor for HA oligosaccharides responsible for the induced gene expression observed here remains to be determined. Our data exclude the possibility that CD44 is the receptor in the cells we have examined. However, recent experiments using CD44-null mice suggest that CD44 is partly responsible for removing HA fragments from the inflamed lung (Teder et al., 2002), suggesting that, in some contexts, CD44 might be involved directly or indirectly in the metabolism of HA degradation products. TLR-4 is the receptor for HA degradation products in dendritic cells (Termeer et al., 2002), but we show here that TLR-4 does not serve this function in 3LL tumour cells or primary fibroblasts. Thus, there must be another receptor for low-molecular-weight HA fragments. Other candidate receptors include LYVE-1 (Banerji et al., 1999) and Layilin (Bono et al., 2001). Future work will try to identify the receptor for low-molecular-weight HA fragments in 3LL tumour cells and primary fibroblasts that regulates Mmp-9 transcription.

The fascinating conclusion from this study is that a simple carbohydrate such as hyaluronan can have significantly different biological effects depending on its size. We have previously discovered that high-molecular-weight HA above certain concentrations causes cell-cycle arrest, a process mediated by the transmembrane receptor CD44 and its intracellular partner protein Merlin (Morrison et al., 2001). Upon degradation of this HA, a tumour would benefit in two ways. Digestion of high-molecular-weight HA would allow cells to escape from HA-induced cell-cycle arrest. At the same time, the oligosaccharide-induced program triggered by HA degradation products would endow new properties on tumour or stromal cells, including the production of MMPs.

We thank Nathalie Decker for excellent technical assistance.