Extracellular matrix retention of thrombospondin 1 is controlled by its conserved C-terminal region

The thrombospondins are an evolutionarily ancient family of extracellular, calcium-binding glycoproteins with roles in cell interactions. The five TSPs of mammals, (TSP1-TSP5), have important roles in cell interactions with extracellular matrix (ECM), angiogenesis and vascular biology, synaptogenesis, immune response, wound repair, and connective tissue organisation (reviewed by Bornstein et al., 2004; Adams, 2001; Lawler and Detmar, 2004; Posey et al., 2004). TSPs are complex multidomain and multimeric proteins. In their C-terminal regions, all family members have the same domain organisation that consists of three or four EGF-like domains, a series of calcium-binding, thrombospondin type 3 repeats and a globular C-terminal domain with structural homology to L-lectin domains (Adams and Lawler, 1993; Kvansakul et al., 2004; Carlson et al., 2005) (as schematised in Fig. 2A). The N-terminal regions of individual family members have lower sequence identity and show variations in domain organisation. The five TSPs of land vertebrates divide into two structural subgroups – A and B – according to their assembly as trimers or pentamers (Adams and Lawler, 1993).

The diverse functions of TSPs depend on their presence in the extracellular milieu. However, extracellular TSPs can undergo several fates, which include uptake, proteolysis or retention within ECM. Interactions of the laminin-G-like N-terminal domain with cell-surface heparan sulphate proteoglycans and LDL-receptor-related protein (LRP) provide a mechanism for rapid endocytic uptake and intracellular proteolysis of intact TSP1 or fragments containing the N-terminal domain (Elzie and Murphy-Ullrich, 2004). Multiple TSPs bind to or are substrates of extracellular proteases (Hogg et al., 1994; Bonnefoy and Legrand, 2000; Agah et al., 2005; Lee et al., 2006). The resulting context-dependent extracellular TSP fragments can have activities distinct from the parent protein (Lee et al., 2006). The third fate of retention within the ECM appears to relate to cell phenotype and environmental conditions. Many proliferative cells accumulate extracellular TSP1 in pericellular patches (Raugi et al., 1982; Mosher et al., 1982; Jaffe et al., 1983; Vischer et al., 1985; Mumby et al., 1984). The retention of TSP1 in vascular smooth muscle cell ECM is increased under inflammatory stimulation (Kuznetsova et al., 2006). Surprisingly, despite intensive research into their structure and biological roles, there is relatively little understanding of the cellular and molecular mechanisms by which TSPs are retained within ECM. This question motivated our study. We have examined the hypothesis that the highly conserved TSP C-terminal region has a role in ECM retention.

In vivo, the presence of TSPs in ECM is well documented during mouse embryonic development and in adult human tissues. For example, TSP1 is present in epidermal basement membrane, epiretinal membranes and arterial microfibrils (Raugi et al., 1987; Wight et al., 1985; Hiscott et al., 1992; Fauvel-Lefeve et al., 1996). TSP2 is present in cartilage and various connective tissues (Bornstein et al., 2004). TSP4 is present in the ECM of the nervous system, at neuromuscular junctions, in the developing retina, adult eye and tendon (Arber and Caroni, 1995; Hauser et al., 1995; Stenina et al., 2003; Dunkle et al., 2007). TSP5 or COMP (cartilage oligomeric matrix protein) is a component of ECM in cartilage, synovium and tendon, which associates with collagen fibrils and binds collagens II and IX, matrilin-3 and fibronectin in vitro (Hedbom et al., 1992; DiCesare et al., 1995; DiCesare et al., 1997; DiCesare et al., 2002; Sodersten et al., 2005). In vivo models and human diseases provide evidence that TSPs contribute functionally to ECM organisation and can modulate the adhesive or structural properties of ECM. For example, TSP2-null mice have aberrant assembly of collagen fibrils because of reduced cell surface tissue transglutaminase, resulting in lax tendons and decreased tensile strength of the skin (Kyriakides et al., 1998; Agah et al., 2005). Interaction of Drosophila TSP (D-TSP) with αPS2 integrin on muscle cells is needed for stable tendon and muscle attachment during development (Chanana et al., 2007; Subramanian et al., 2007). Polymorphisms in the coding sequences of human THBS1 and THBS4, which predispose to cardiovascular disease, enhance platelet aggregation in the case of TSP1 and affect endothelial cell adhesion in the case of TSP4 (Stenina et al., 2007).

In vitro binding assays with purified proteins have demonstrated that TSPs can bind to multiple structural ECM components and to various growth factors that may themselves bind ECM (reviewed by Adams, 2001). These findings indicate targets for binding by extracellular TSPs, but do not address the molecular processes by which TSPs come to be retained within ECM. Knowledge of these processes is important with regard to the clear biological significance of TSPs in connective ECM, tumour biology, immune response, synaptogenesis and vascular function. Prospective therapeutic modulations or construction of synthetic ECM would also benefit from this knowledge. Until recently, the ability to analyse these mechanisms was hampered by a lack of understanding of the complex structure of TSP polypeptides, particularly the large and highly conserved C-terminal region. In recent years, structures for all the major domains of TSP1 and TSP2 have been solved at the atomic level (Tan et al., 2002; Tan et al., 2006; Kvansakul et al., 2004; Carlson et al., 2005). We demonstrate here that ECM retention is a conserved property of TSPs and present data that outline the mechanism with regard to TSP1.

Consistent with the presence of extracellular TSP1 in tissues, one fate of TSP1 secreted by cultured cells is to be retained extracellularly as cell-associated patches, short fibrils or granules (Raugi et al., 1982; Jaffe et al., 1983; Vischer et al., 1988; Adams and Lawler, 1994; Kuznetsova et al., 2006). However, molecular mechanisms by which TSPs are retained in ECM remain poorly understood. Experiments to date have relied on the use of fixed cells or exogenously added TSP1 protein and these procedures do not reliably distinguish ECM-incorporated proteins from proteins loosely associated with the cell surface.

To address this question, we set up experimental conditions for the quantifiable analysis of ECM retention of TSP1. To identify appropriate conditions, dense cultures of C2C12 skeletal myoblasts or COS-7 cells were grown on glass coverslips for 4 days, to allow the secretion and assembly of ECM, and then the cell layers were removed by several classic methods for ECM isolation (Carter, 1982; Robinson and Gospodarowicz, 1984). The effectiveness of each method in completely removing the cell layer was assessed by DAPI staining to visualise cell nuclei and FITC-phalloidin staining to visualise F-actin-containing cellular fragments. Whereas residual cellular fragments remained after extraction of cell cultures with 2 M urea, either 8 M urea or 20 mM NH₄OH was efficient in removing all traces of cellular debris (Fig. 1A). For both cell types, the ECM layer remaining after 20 mM NH₄OH extraction was examined by staining with Oregon-Green-conjugated gelatin, which binds to multiple ECM components (Hsieh et al., 1980). The ECM of C2C12 cells contained distinct fibrils whereas the ECM of COS-7 cells was deposited in patches (Fig. 1B). Immunostaining for fibronectin demonstrated that the fibrils of C2C12 ECM contained fibronectin, whereas little fibrillar fibronectin was detected in COS-7 ECM (Fig. 1B). However, fibronectin was readily detectable in both ECMs by immunoblotting the ECM after solubilisation in hot SDS sample buffer (Fig. 1C). Skeletal myoblasts secrete abundant TSP1 that is deposited pericellularly (Adams and Lawler, 1994; Adams, 1997), and indeed TSP1 was detected in the ECM of C2C12 cells but not in COS-7 cells (Fig. 1B). The insoluble TSP1 consisted of small (0.1-0.3 μm diameter) granular patches deposited in closely spaced arrays (Fig. 1B and high-magnification inset). This pattern of deposition was reminiscent of the pericellular TSP1 patches identified previously in association with fixed cells.

To establish whether recombinantly expressed TSP1 incorporates into ECM with the same characteristics as endogenous TSP1, we heterologously expressed human TSP1 in COS-7 cells and analysed its localisation after 4 days with antibodies specific to human TSP1. The human TSP1 was present in cells at high levels in the secretory pathway (Fig. 1D, left panel) and in the NH₄OH-insoluble ECM as clustered arrays of small punctate patches (Fig. 1D, right panel; examples of individual puncta are arrowed). The areas of these arrays (mean area 1838±290 μm², n=15) were similar to the areas of well-spread single cells (mean area 1226±246 μm², n=10), indicative that the puncta were deposited underneath transfected cells. Diameters of individual puncta were in the range 0.1-0.3 μm, as for endogenous TSP1. The ECM retention of heterologously expressed TSP1 as arrays of puncta reproduced the pattern of deposition of endogenous TSP1 and was distinct from the general patterning of the ECM layer of COS-7 cells (compare the COS-7 OG-gelatin-stained ECM in Fig. 1B with COS-7 TSP1 ECM in Fig. 1D). Immunoblotting of the ECM from COS-7 transfectants established that full-length TSP1 polypeptide was present within the ECM (Fig. 1C). These results demonstrated the suitability of this assay to examine the molecular processes that regulate ECM retention of TSPs. The results obtained with 20 mM NH₄OH or 8 M urea extraction have been consistently comparable and we have routinely used extraction with 20 mM NH₄OH for the experiments described below.

TSPs have been well conserved in metazoan evolution (Adams et al., 2003; Adams, 2004) and we wished to test whether the property of ECM incorporation is conserved across the TSP family. We approached this question with regard to three TSPs from phylogenetically diverse animals and with respect to family members that have different numbers of TSP subunits per molecule (Fig. 2A). For consistency of detection, the proteins were expressed in V5-epitope-tagged forms and detected with antibody to the V5 tag (see Materials and Methods). Ciona intestinalis TSPA has a very similar domain architecture to TSP1 and 49% overall sequence identity to human TSP1 (Fig. 2A). Ciona TSPA is an unusual TSP that lacks a coiled-coil domain and multimerises as a dimer (Adams et al., 2003) (our unpublished results). Nevertheless, Ciona TSPA was retained in ECM in the form of punctate deposits (Fig. 2B). Drosophila melanogaster TSP (D-TSP) is a pentamer with a distinct domain architecture (Fig. 2A). The type 3 repeats and L-lectin domain have 41-44% and 53-57% identity, respectively, to the homologous domains of the five TSPs of tetrapods (Adams et al., 2003). D-TSP was also incorporated into the ECM as arrays of fine and very dense puncta (Fig. 2B). Quantification of the percentage of V5-expressing cells in fixed and permeabilised 4 day COS-7 cell cultures established that D-TSP and Ciona TSPA were expressed very similarly to TSP1 (Fig. 2C), yet both were retained in ECM to a greater extent than TSP1 (Fig. 2D). COMP/TSP5 has a short, 19 residue N-terminal domain and essentially comprises a pentamer of the C-terminal region (Fig. 2A) (Posey et al., 2004). Human COMP/TSP5 and human TSP1 have 54% identity across the C-terminal region (i.e. comparing EGF domain 2 with the C-terminus of COMP and EGF domain 1 to the C-terminus of TSP1). COMP/TSP5 was also deposited into ECM by COS-7 cells as arrays of circular deposits, with similar morphology to the TSP1 deposits, although the individual puncta tended to be more widely spaced than those of the other TSPs tested (Fig. 2B). Although expressed by a higher proportion of transfected cells, the retention of COMP in ECM was quantitatively lower than for other TSPs (Fig. 2C,D). An unrelated V5-tagged protein, secreted MAEA, was expressed similarly to the TSPs (Fig. 2C) and was completely negative for ECM retention (Fig. 2D). These results establish that ECM retention as punctate deposits is a well-conserved activity within the TSP family. Distinctions in the efficiency of retention and density of ECM puncta indicate a level of specificity in the properties of different family members.

The overall similarities in the ECM retention properties of phylogenetically remote TSPs led us to hypothesise that the C-terminal region, which is the most conserved region between TSP family members, has a major role in mediating ECM retention. Indeed, for several TSP family members, domains or motifs within this large conserved region bind to structural components of the ECM (Rosenberg et al., 1998; Holden et al., 2001; Narouz-Ott et al., 2000) (reviewed by Adams, 2004). To understand which domains of TSP1 are required for its incorporation into ECM, we analysed the activities of a set of deletion proteins. The coiled-coil oligomerisation domain of TSP1 is located adjacent to the N-terminal domain, which can thus be expressed conveniently in trimeric form as a truncation mutant (No, Fig. 3A) (Sottile et al., 1991). To examine the specific activities of the conserved C-terminal region, we built on new structural knowledge of the C-terminal regions of TSP1 and TSP2 (Kvansakul et al., 2004; Carlson et al., 2005) to engineer novel, trimeric C-terminal recombinant proteins (Fig. 3A). These were prepared in two forms: (1) as a deletion of the von Willebrand factor_C domain (vWF_C) and type 1 repeats, by fusing the encoding sequence for the signal peptide, N-terminal domain and coiled-coiled domain of human TSP1 (aa 1-297) in-frame with the encoding sequence for the C-terminal region (aa 531-1152) (NoTSP1C, Fig. 3A), and (2) as a chimaeric, fluorescently tagged protein in which the TSP1 signal peptide and N-terminal domain sequence was substituted by the signal peptide of BM40 and the encoding sequence for monomeric red fluorescent protein (mRFP), fused in frame to the encoding sequence of the coiled-coil domain and vWF_C domain (aa 241-360) and C-terminal region (aa 531-1152) of human TSP1 (mRFPovTSP1C, Fig. 3A). The vWF_C domain was included in this construct because its deletion has been found to affect trimer assembly (Lawler et al., 1992). An mRFP-tagged construct was similarly prepared for expression of the trimeric vWF_C domain and type 1 repeats (TSR) (aa 241-530) (mRFPovTSR, Fig. 3A). Constructs for expression of secreted mRFP, secreted mRFP fused to the coiled-coil domain of TSP1 (mRFPo) and mRFP-tagged C-terminal region monomer (mRFPTSP1C) were generated as controls (Fig. 3A). All the proteins were secreted efficiently. Proteins containing the TSP1 oligomerisation domain trimerised as expected (Fig. 3B,C). The mRFPo protein showed reduced fidelity of trimerisation, with higher order oligomers also secreted (Fig. 3C).

The No protein was essentially negative for ECM retention, although rare examples of small deposits of widely spaced puncta were detected (Fig. 4A and quantified in Fig. 4B,C). By contrast, NoTSP1C was retained in ECM as dense arrays of puncta very similar to those formed by intact TSP1 (Fig. 4A, quantified in Fig. 4B,D). A previously characterised monomeric C-terminal fragment corresponding to type 3 repeats 5 to 7 and the L-lectin domain (type3_5-7GCT) (Kvansakul et al., 2004) (Fig. 3A) was not deposited into ECM (Fig. 4A; quantified in Fig. 4B,D). Within the set of mRFP-tagged proteins, no incorporation of mRFP, mRFPo, mRFPovTSR or mRFP-TSP1C monomer into ECM was detected, even though these proteins were all well secreted (Fig. 4A and data not shown, quantified in 4B,D, see also Fig. 3C). By contrast, trimeric mRFPovTSP1C was strongly retained in the ECM as arrays of puncta (Fig. 4A,C,E). The appearance of these puncta, detected by fluorescent excitation of RFP, was indistinguishable from the arrays of puncta formed by intact TSP1 or NoTSP1C, detected by indirect immunofluorescence with a monoclonal antibody to human TSP1 (Fig. 4A and Fig. 1). When ECM-incorporated mRFPovTSP1C was analysed by dual imaging for RFP signal and by staining with an antibody reactive with an epitope close to the C-terminus of human TSP1, the two signals colocalised, indicating that the full-length mRFPovTSP1C protein was present in the puncta (data not shown). Immunoblotting of the ECM confirmed that the full-length mRFPovTSP1C subunit was present (Fig. 4F). Lower molecular mass fragments were also detected, although it cannot be excluded that fragmentation of mRFPovTSP1C occurred during ECM solubilisation.

We confirmed the ECM retention activity of mRFPovTSP1C in a second cell type. A549 lung carcinoma cells also deposited mRFPovTSP1C into ECM as arrays of punctate patches. The formation of these patches was specific to mRFPovTSP1C and was not detected upon expression of the mRFPo control protein (supplementary material Fig. S1). Together, these results establish that the C-terminal region in trimeric form is sufficient to mediate ECM retention and that this is a general property of cells from different tissue sources.

Deposition of mRFPovTSP1C into ECM was examined in more detail by time course experiments. ECM retention of mRFPovTSP1C became established between 1 and 2 hours after plating. At this time, cells had attached but only partially respread and the areas of deposition of mRFPovTSP1C puncta were correspondingly small and sparse (Fig. 4G). Between 4 and 8 hours after plating, in correlation with cell spreading, the arrays of punctate mRFPovTSP1C deposits increased in number and size (Fig. 4G). Between 8 and 24 hours, the area of the arrays continued to increase. With incubation for up to 7 days, the density of puncta continued to increase within the arrays, such that puncta sometimes appeared merged together as amorphous regions (Fig. 4G, arrowed, shown for the 3 day time point). At 3 days, confocal Z-stack images demonstrated that mRFPovTSP1C was present throughout the secretory pathway of intact cells (Fig. 4G). Examination of confocal X-Z sections confirmed the presence of intracellular perinuclear mRP-containing vesicles in intact cells and the presence of discrete punctate deposits in the insoluble ECM after NH₄OH extraction (Fig. 4H).

We next wished to identify specific motifs within the TSP1 C-terminal region that are important for ECM retention. For other families of ECM macromolecules that directly contribute to ECM structure, ECM assembly depends on cell-surface interactions (Colognato and Yurchenco, 2000; Mao and Schwarzbauer, 2005). Although TSP1 is not a regular structural component, we hypothesised that ECM retention would involve its cell attachment properties. The seventh type 3 repeat of TSP1 contains a conserved RGD, integrin-binding motif that mediates calcium-dependent interaction with αvβ3 integrin (Lawler et al., 1988). Interactions with β1 integrins have also been reported (reviewed by Adams, 2004). The C-terminal L-lectin-like domain is also known as a locus for cell interactions, but the molecular basis for these is currently not well understood. Two peptide motifs previously characterised to bind the cell-surface glycoprotein CD47 are located on largely buried β-strands in both the TSP1 and TSP2 crystal structures (Gao et al., 1996; Kvansakul et al., 2004; Carlson et al., 2005) (Fig. 5A) and thus are unlikely to be active for receptor binding under normal physiological conditions.

With regard to the structural homology of the TSP1 C-terminal globular domain to animal L-type lectins, such as ERGIC-53 (Velloso et al., 2002) and VIP36 (Satoh et al., 2007), we noticed that the calcium-dependent carbohydrate-binding site of VIP36 coincides with the double calcium-binding site in the TSP1 L-lectin domain (Fig. 5A). The L-lectin domains of TSP1 and VIP36 share only 10% sequence identity, but 164 Cα atoms can be superimposed with a root-mean-square deviation of 2.7 Å (program SSM) (Krissinel and Henrick, 2004). The two calcium ions in TSP1 are bound by a DDD motif (Kvansakul et al., 2004) that is conserved in all TSPs identified to date (Adams, 2004), suggesting an important function. This face of the L-lectin domain is not involved in interactions with the type 3 repeats in the TSP2 structure (Carlson et al., 2005) and is likely to be available for ligand binding in all TSPs (Fig. 5A). We hypothesised that the cell-attachment site of the TSP L-lectin-like domain may involve the calcium ion-binding DDD motif and the adjacent surface-exposed loop between the seventh and eighth β-strands (Fig. 5A).

To test the hypothesis, we first introduced point mutations of the RGD or DDD motifs into the monomeric TSP1 C-terminal fragment, type3_5-7GCT, of known structure (Fig. 5A). The RGD motif was mutated to VGD because valine is frequent at the equivalent position in other type 3 repeats. The DDD motif provides the only negatively charged metal ligands at the double-calcium site (Kvansakul et al., 2004) and mutation to AAA is predicted to abolish calcium binding at this site. Importantly, the two mutant proteins were efficiently secreted and their secondary structure and oligomeric status in the presence of 2 mM calcium ions were indistinguishable from wild-type type3_5-7GCT (supplementary material Fig. S2). Thermal unfolding of all three proteins, as monitored by the CD signal at 215 nm, occurred at ∼50°C (data not shown). Thus, the mutations did not affect the folding of TSP1 C-terminal region.

Next, the set of proteins were tested for cell attachment activity under calcium-replete conditions, as a known activity of wild-type type3_5-7GCT. The type3_5-7GCT fragment supported attachment of C2C12 and human arterial smooth muscle cells (HASMC) (Fig. 5B). In the calcium-replete state, the attachment of C2C12 cells to intact TSP1 or type3_5-7GCT is not inhibited by RGD peptide (Adams and Lawler, 1994; Kvansakul et al., 2004). We found that C2C12 cells attached to type3_5-7GCT/VGD protein almost as well as to wild-type type3_5-7GCT (Fig. 5B). By contrast, HASMC attached poorly to type3_5-7GCT/VGD (Fig. 5B). The attachment of both cells to type3_5-7GCT/AAA was significantly less than to the wild-type protein, indicating that the DDD motif participates in an attachment activity that is distinct from the RGD motif (Fig. 5B). To investigate whether cell attachment to type3_5-7GCT depends on CD47, we compared attachment of fibroblasts from strain-matched wild-type or CD47-null mice (Lindberg et al., 1996). The CD47-null cells attached equally as well as wild-type cells to type3_5-7GCT. Furthermore, both wild-type and CD47-null fibroblasts had significantly reduced attachment to type3_5-7GCT/VGD and type3_5-7GCT/AAA (Fig. 5C). Thus, CD47 is not essential for cell attachment to type3_5-7GCT. This is consistent with previous data that CD47 is dispensable for assembly of fascin protrusions in response to intact TSP1 (Adams et al., 2001).

We introduced the same point mutations into mRFPovTSP1C. The mutant proteins were expressed, secreted and trimerised with comparable efficiency to the wild-type mini-trimers (Fig. 5D, supplementary material Fig. S3) yet both point mutant proteins were decreased for ECM retention (Fig. 5E). Thus, both cell attachment sites are implicated in the mechanism of ECM retention of TSP1.

We next examined the role of CD47 and β1 integrins in ECM retention of TSP1 C-terminal region trimers. Wild-type and CD47-null mouse fibroblasts were found to equivalently deposit arrays of mRFPovTSP1C puncta of similar morphology (Fig. 6A). Endogenous TSP1 was also detected in the ECMs of both cell types (data not shown). The wild-type fibroblasts were severely impaired for deposition of VGD or AAA mutants, to the extent that meaningful comparison of the CD47-null cells with these mutant proteins could not be made (data not shown). For an independent evaluation of CD47 binding to the C-terminal region of TSP1, analysis was made by surface plasmon resonance. CD47 is a well-characterised ligand of signal regulatory protein α (SIRPα) and the interaction of the extracellular domains of CD47 and SIRPα has been extensively established (Seiffert et al., 1999; Vernon-Wilson et al., 2000; Hatherley et al., 2007; Liu et al., 2007). The interaction of previously characterised recombinant proteins corresponding to the extracellular regions of human CD47 and SIRPα (Hatherley et al., 2007) with type3_5-7GCT was examined by surface plasmon resonance. As expected, the SIRPα extracellular region as analyte bound to the immobilised extracellular region of CD47 but not to itself (Fig. 6B). Similarly, the CD47 extracellular region bound to immobilised SIRPα extracellular region, but not to itself. The CD47 preparation contains some multimers (Hatherley et al., 2007), accounting for a slow rate of dissociation from SIRPα (Fig. 6C). By contrast, type3_5-7GCT as analyte produced the same signal on the control protein SIRPα as on CD47 (Fig. 6D). The increase in response units upon injection of type3_5-7GCT was due to the different refractive index of the buffer used for type3_5-7GCT compared with the running buffer. Thus, under conditions in which CD47 clearly binds its ligand SIRPα, no interaction with TSP1 L-lectin domain was detected.

The contribution of the RGD site in ECM retention suggested that integrins could play a role in the ECM incorporation of TSP1. To test whether β1 integrins are essential, we examined the ability of GD25, β1-integrin-null fibroblasts to retain heterologously expressed mRFPovTSP1C in ECM. These cells incorporated mRFPovTSP1C into ECM puncta, whereas control mRFPo was not deposited (Fig. 7). However, in comparison to GD25 cells re-expressing β1A integrin, fewer punctate deposits were present, even though the percentage of RFP-expressing cells was similar for both cell lines (Fig. 7A,B). Thus, β1 integrins are not essential for ECM retention of mRFPovTSP1C but may increase the efficiency of the process. The major integrin heterodimers on GD25 cells are αvβ3, αvβ5 and, at a lower level, α6β4 (Wennerberg et al., 1996). To establish whether αvβ3 has an essential role, we examined ECM retention by CHO cells, which express αvβ5 but not αvβ3 or αvβ1 integrins (Triantafilou et al., 2000). CHO cells also assembled ECM puncta of mRFPovTSP1C, demonstrating that αvβ3 integrin is not essential for the process (data not shown).

Fibronectin is a known ECM binding partner of both TSP1 and certain β1 integrin heterodimers (Homandberg and Kramer-Bjerke, 1987; Dardik and Lahav, 1989; Sottile and Hocking, 2002; Hynes, 2002; Mao and Schwarzbauer, 2005). GD25β1A cells establish a denser fibrillar fibronectin ECM than GD25 cells (Wennerberg et al., 1996). To investigate the hypothesis that increased ECM retention of mRFPovTSP1C by GD25β1A cells is due to fibronectin acting as a bridge between β1 integrins and mRFPovTSP1C, we compared the ECMs of GD25 and GD25β1A cells for possible colocalisation of fibronectin and mRFPovTSP1C. The ECMs of both cell lines contained examples of either nonoverlapping or colocalised mRFPovTSP1C and fibronectin. The frequency of colocalised deposits was not significantly increased in GD25β1A cells. There was no obvious correlation between colocalisation with fibronectin and the density of mRFPovTSP1C deposits (representative examples from GD25β1A cells are shown in Fig. 7C). We conclude that TSP1 can be retained in ECM in proximity to fibronectin, but the deposition of fibronectin is not essential. Colocalisation of mRFPovTSP1C and laminin was not observed (data not shown).

We present novel data that advance concepts of the mechanism by which secreted TSPs are retained within ECM. Importantly, we demonstrate that ECM retention is an evolutionarily conserved property of TSPs, shared by both trimeric and pentameric TSPs. Thus ECM retention appears likely to be an ancestral activity of TSPs. This is borne out by the recently reported functional role of the single TSP of Drosophila, D-TSP, in the ECM of muscle-tendon attachment sites and the localisation of a prawn TSP to specialised ECM structures – oocyte cortical rods (Yamano et al., 2004; Chanana et al., 2007; Subramanian et al., 2007).

Studies of cell-attachment activities of TSP1 and TSP2 have relied on purified TSPs coated onto glass or plastic surfaces, where they probably form a uniform layer of individual trimers. Using classic procedures for isolation of insoluble ECM, we demonstrate that TSP1 and other TSPs are retained in ECM in the physical form of small, closely spaced puncta. This is consistent with descriptions of punctate, granular or amorphous deposits of TSP1 around fixed cultured cells (Jaffe et al., 1983; Vischer et al., 1988; Raugi et al., 1982; Adams and Lawler, 1994; Kuznetsova et al., 2006). For the cell types tested here, association of TSP1 puncta with fibrillar ECM did not appear obligatory, although partial colocalisation with fibronectin was detected (Fig. 7). Associations of TSP1 with elastin microfibrils and several ECM proteoglycans have been documented (Vischer et al., 1997; Kuznetsova et al., 2006). These interactions have all been linked to TSP1 N-terminal domain and endocytosis of TSP1. The literature contains little evidence for self-association of TSP1 molecules, which suggests that a bridging molecule such as fibronectin might be involved. Alternatively, post-secretion cell-surface interactions might provide conditions suitable for TSP1 self-association. Prior studies indicated multiple binding sites for fibronectin on TSP1 (Dardik and Lahav, 1989; Sipes et al., 1993), but the role of the C-terminal region in fibronectin binding remains unclear and further analysis will be needed. ECM retention of TSP1 may be a complex process involving more than one component of the structural ECM.

By investigating the mechanism of ECM retention for TSP1, we provide novel evidence that the trimeric C-terminal region is necessary and sufficient for ECM retention activity. Recent advances in understanding the structures of TSPs have revealed that there are extensive interactions between the EGF domains, type 3 repeats and L-lectin-like domain, such that the whole multidomain region should be considered a single structural unit (Kvansakul et al., 2004; Misenheimer and Mosher, 2005; Carlson et al., 2005). These findings are corroborated by immunochemical analyses (Annis et al., 2007). For this reason, we chose to develop a recombinant trimer that included all these domains for functional studies. The “mini-trimer” protein is secreted efficiently by cells and should be a valuable general tool for analysing oligomer-dependent functions of the C-terminal region of TSP1 and, in principle, other TSPs.

The two cell-attachment sites within the C-terminal region are located near the C-terminus, in the last type 3 repeat and L-lectin domain. However a TSP1 C-terminal region (aa 531-1152) monomer that includes these sites was not sufficient for ECM incorporation, establishing the importance of trimerisation for ECM retention (Fig. 4). The RGD motif in the last type 3 repeat of TSP1 has a well-documented role in binding integrins (reviewed by Adams, 2004). Our results with a novel VGD point mutation demonstrate that the RGD motif contributes to cell-attachment activity and also contributes to retention of TSP1 C-terminal trimers in ECM. This was corroborated by the demonstration that β1 integrins increase the efficiency of ECM retention. Neither β1 integrins nor αvβ3 were essential, suggesting that several integrin heterodimers can redundantly mediate the role of the TSP1 RGD site in ECM retention. This is consistent with the multiple integrins known to support RGD-dependent cell attachment to TSP1 (reviewed by Adams, 2004). As discussed above, our results do not exclude the possibility of an indirect role for integrins, for example by binding an RGD-containing ligand such as fibronectin, which also binds TSP1.

The L-lectin domain is recognised to have independent cell-attachment activity; however, new structural knowledge has necessitated the reappraisal of the physiological role of previously identified peptide motifs (Gao et al., 1996; Kvansakul et al., 2004; Carlson et al., 2005). A novel candidate functional site was suggested by the striking structural homology of TSP L-lectin domain to the calcium-dependent lectins ERGIC53 and VIP36, which bind glycoproteins in the secretory pathway. The well-characterised oligosaccharide binding site of these lectins (Satoh et al., 2007) coincides with a double calcium site that is conserved in all TSPs and suitably positioned for ligand binding (i.e. on the surface and away from the type 3 repeats that wrap around the L-lectin domain; Fig. 5A). Mutation of the DDD motif in TSP1 (AAA mutant) resulted in impaired cell attachment activity for all cells tested and significantly reduced ECM retention. These data provide strong initial evidence that the conserved DDD motif participates in a ligand-binding site of general physiological importance in TSP1 and potentially other TSP family members. Lectins are typically multivalent and the need for trimerisation of the L-lectin domain for ECM retention may reflect a similar requirement for clustering of low-affinity interaction sites to achieve high local avidity (Brewer et al., 2002). It is intriguing that the RGD and DDD motifs are physically well separated (∼50 Å distance) (Fig. 5A). Thus, the TSP1 C-terminal region might act to assemble multiprotein complexes, bringing integrins into lateral proximity with other adhesion receptors or ECM components. In our analyses of cell attachment, ECM retention and direct binding we did not obtain evidence for a role of CD47 (Figs 5 and 6). Recent genetic analyses in knockout mice indicate co-dependency of TSP1 and CD47 in endothelial and smooth muscle responses to nitric oxide (Ridnour et al., 2005; Isenberg et al., 2006). Interpretation of these in vivo data is complicated by the multiple roles of CD47 as an integrin-signalling co-receptor and a transcellular ligand for SIRPα, which is itself regulated by integrin activation (Brown and Frazier, 2001; Johansen and Brown, 2007).

The five TSPs of humans are associated with multiple genetic and acquired disease processes. The cellular and molecular roles of the type 1 repeats of TSP1 and TSP2 have been intensively investigated, and this knowledge led to the development of novel anticancer agents currently in clinical trial (e.g. Markovic et al., 2007). Based on recent advances in understanding the structural complexity of the TSP C-terminal region, we establish here for TSP1 that the trimeric C-terminal region is required for ECM retention and that two sites proximal to the C-terminus of TSP1 mediate cell attachment and ECM retention activities. The mRFP-tagged mini-trimers will enable further investigations of the possible significance of ECM retention for other important functions of TSP1, which include modulation of endothelial cell proliferation, vascular wall remodelling and the immune response. The mini-trimers will also be generally useful novel tools to examine parameters of TSP localization in culture and enable better detection of TSPs in vivo. In summary, knowledge of the cell-mediated mechanisms necessary for ECM retention of TSP1 has important implications for understanding the normal and pathological activities of the TSP C-terminal region.

COS-7, A549 and CHO cells were from ATCC. 293-EBNA cells were from Invitrogen. GD25 and GD25β1A cells were from Reinhard Fässler (Wennerberg et al., 1996). Dermal fibroblasts were prepared by trypsinisation from dissected hindlimbs of neonatal wild-type or CD47-null mice, gifts from Frederick Lindberg (Lindberg et al., 1996). pCEP-Pu plasmid was as described (Kvansakul et al., 2004). pcDNA3/D-TSP.V5His was as described (Adams et al., 2003). cDNA encoding mRFP was as described (Campbell et al., 2002). cDNA encoding human COMP (GenBank BC125092) was obtained from Open Biosystems and ligated into pCDNA3.V5His by TOPO cloning. cDNA encoding Ciona intestinalis TSPA was obtained from Noriyuki Satoh, Kyoto University (Satou et al., 2002). We sequenced this cDNA in its entirety on both strands by automated DNA sequencing, carried out by the CCF Molecular Biotechnology Core, using vector and cDNA-specific oligonucleotide primers. The complete nucleotide sequence was deposited as GenBank AY465896. Oligonucleotide primers used to generate the various constructs are listed in supplementary material Table S1. All oligonucleotides were synthesised by Sigma-Genosys and oligonucleotides for mutagenesis reactions were HPLC-purified. Mutations in TSP1 C-terminal region were introduced using QuikChange XL kit (Stratagene). FITC-phalloidin was from Sigma. Oregon-Green-gelatin was from Molecular Probes. The following mouse monoclonal antibodies were used: V5 epitope (GKPIPNPLLGLDST) (Southern et al., 1991) (Invitrogen); DsRed (Clontech); clone D4.6 to mouse and human TSP1 (Labvision) (Annis et al., 2006), and MAI and MAII to human TSP1 (Lawler et al., 1985) (a gift from Jack Lawler, Harvard). Rabbit polyclonal antibody to fibronectin was from Sigma and FITC-conjugated goat polyclonal IgG to V5 tag from Abcam. Alkaline-phosphatase-conjugated secondary antibodies were from Applied Biosystems and FITC-conjugated secondary antibodies were from ICN.

COS-7 cells (2×10⁵) were transiently transfected with 2.5 μg of the indicated plasmids using Polyfect transfection reagent (Qiagen) according to manufacturer's instructions. After 18 hours, cells from each transfection were replated onto glass coverslips in two replicate 35-mm dishes and incubated for 3 days to allow for secretion and deposition of ECM. One dish was rinsed in PBS and fixed in 2% paraformaldehyde (Electron Microscopy Sciences). These cells were used to quantify the percentage of transfected cells expressing the protein. The parallel dish was treated with 20 mM ammonium hydroxide for 5 minutes (Robinson and Gospodarowicz, 1984), rinsed in PBS, and fixed in 2% paraformaldehyde. Coverslips from cultures expressing mRFP constructs were mounted directly on slides with Vectashield DAPI mounting medium (Vector Laboratories, Burlingham, CA). Other samples were processed for indirect immunofluorescence before mounting. Treatments with 2 M or 8 M urea for 10 minutes were used as alternative procedures to isolate insoluble ECM (Carter, 1982). To screen for cellular material after the various extractions, samples were stained with FITC-phalloidin to visualize F-actin or DAPI as a nuclear stain. Two coverslips were scored for each sample in each experiment and a minimum of three independent experiments were carried out for each condition.

Samples for fluorescence microscopy were examined under a Leica DMIRE2 inverted microscope (Heidelberg, Germany), equipped with electronically controlled shutters and filter wheel. Images were captured under 63× objective using a Hamamatsu camera controller C4742_95 run by Improvision Openlab software (version 3.1.5). Some samples were examined under a Leica DM RXE(TCS-SP/SP-AOSB) confocal laser scanning microscope with HCX Plan Apo 63× NA 1.4 oil-immersion objective lens, using Leica confocal software version 2.5. Confocal images were acquired as Z stacks, X-Y, or X-Z sections at zoom 1 or 2, at room temperature. Projections of image stacks were prepared in Volocity 3.7.0. In each experiment, the percentage of transfected cells expressing the protein of interest was calculated by direct scoring of cells expressing mRFP-tagged proteins as a fraction of DAPI-stained nuclei, or by permeabilisation in 0.5% Triton X-100 and indirect immunofluorescent staining for the non-RFP proteins. Ten 63× fields were scored per coverslip. Cell area measurements were made from phalloidin-stained samples. To quantify ECM retention, the number of arrays of TSP puncta per 63× field was scored from 30 fields per coverslip. The areas of ten arrays that corresponded to the production of single cells were measured in Improvision Openlab 3.1.5 and a mean area of deposition per single cell was calculated. This mean area was used to calculate the number of cells producing larger arrays, and thereby the total number of single-cell arrays per field.

The TSP1 type3_5-7GCT construct with a C974S mutation (Kvansakul et al., 2004) was used as template for strand overlap extension PCR to introduce R908V (VGD mutant) and D1001A/D1002A/D1003A (AAA mutant). Sequence-verified inserts were subcloned into pCEP-Pu vector coding for proteins bearing an N-terminal His-Myc-tag and transfected into 293-EBNA cells using Fugene reagent (Roche). Stable transfectants were selected with 1 μg/ml puromycin. Secreted proteins were purified from conditioned serum-free medium as described (Kvansakul et al., 2004). Gel filtration was carried out using an Äkta chromatography system and a 24 ml S200 column (GE Healthcare). Running buffer was 20 mM Na-HEPES pH 7.5, 150 mM NaCl, 2 mM CaCl₂ and the flow rate was 0.5 ml/minute.

Cell attachment was measured as described (Kvansakul et al., 2004).

TSP proteins were collected from the conditioned medium of transfected cells using TALON metal affinity resin (Clontech), or heparin-Sepharose (Amersham Biosciences, Sweden) in the case of proteins containing the TSP1 N-terminal domain, according to the manufacturer's instructions. 50 μl packed slurry was incubated with 3 ml medium for 1 hour at 4°C. Beads were collected by centrifugation, washed three times and bead-bound proteins released by boiling in SDS-PAGE sample buffer, without or with 100 mM DTT. Polyacrylamide gel electrophoresis and immunoblotting were carried out as described (Adams et al., 2003).

Recombinant human SIRPα consisting of the extracellular domain (aa 1-349) followed by the sequence TRHHHHHH was produced as described (Hatherley et al., 2007). Recombinant extracellular regions of human CD47 (residues 1-136) or SIRPα with a biotinylation site and polyhistidine tag were prepared by transient expression in 293T cells using the pEF-BOS vector (Mizushima and Nagata, 1990) and biotinylated with Bir enzyme as described (Brown et al., 1998). CD47 extracellular domain without a biotinylation site was expressed by CHO K1 cells from the pEE14 plasmid (Hatherley et al., 2007). Recombinant proteins were purified by nickel affinity chromatography using chelating Sepharose fast flow (GE Healthcare Life Sciences) and buffer exchanged into 10 mM HEPES pH 7.5, 150 mM NaCl. Protein purity and was as previously described (Hatherley et al., 2007). Protein interactions were analysed using a BIAcore™ 2000 at 25°C, with 10 mM HEPES pH 7.5, 150 mM NaCl, 2 mM CaCl₂ as running buffer. Briefly, ∼1500 RU of streptavidin were coupled to a CM5 research grade chip using amine coupling. Biotinylated CD47 or SIRPα were bound to separate flow cells (1001.4 and 1134.8 RU, respectively). The following were passed over both flow cells at a flow rate of 5 μl/minute: 5 μl of type3_5-7GCT (in 10 mM Tris-HCl pH 7.5, 150 mM NaCl, 2 mM calcium acetate) at 70 μM, 5 μl of SIRPα at 10 μM, and then 5 μl of CD47 at 47 μM at 100 μl/minute.

Near-UV circular dichroism (CD) spectra were obtained for type3_5-7GCT and its VGD and AAA mutants. Proteins were dialysed against 5 mM MOPS pH 7.5, 100 mM NaCl, 2 mM CaCl₂ and adjusted to ∼5 μM concentration. CD spectra were recorded at 25°C in 0.1 mm quartz cuvettes on an Applied Photophysics Chirascan instrument.

We thank Naa Amponsah for technical assistance, Saffi Hussain for help with protein expression and purification, Yosuke Hashimoto for help with confocal microscopy, the CCF Imaging core for confocal microscopy training and CCF Molecular Biotechnology and Genomics cores for scientific support. We thank Noriyuki Satoh, Kyoto University, and the Academica DNA Sequencing Center, Center for Genetic Resource Information, Japan for providing the Ciona TSPA cDNA. Supported by NIH grant GM068073 to J.C.A. and a Wellcome Senior Research Fellowship to E.H. D.H. is supported by the Medical Research Council of Great Britain.