SPARC promotes pericyte recruitment via inhibition of endoglin-dependent TGF-β1 activity

During angiogenesis, nascent blood vessels initially form as endothelial tubes that become coated with pericytes. Pericytes are mobilized from preexisting vessels by the combined activities of several proliferation- and migration-stimulating factors, including matrix metalloprotease 9 (MMP9) and PDGF-BB. Mobilized pericytes migrate to the newly formed endothelial tube and, upon contact, induce vessel maturation and stabilization (von Tell et al., 2006). This process is required for normal angiogenesis, as lack of adequate pericyte coverage results in vessel abnormalities, including leakiness and hemorrhaging (Lindahl et al., 1997; Hellström et al., 2001). Unlike normal vasculature, vessels within tumors are typically leaky, tortuous, and exhibit abnormal pericyte coverage (Helmlinger et al., 1997; Benjamin et al., 1999; Eberhard et al., 2000). Antiangiogenic tumor therapy is believed to be effective at treating some types of cancer by selectively ablating blood vessels that lack pericyte coverage, thereby increasing the efficiency of blood transport within the tumor, which increases the delivery of chemotherapeutics (Gerhardt and Semb, 2008). Targeting pericyte recruitment was shown to increase the efficacy of antiangiogenic tumor therapy in a mouse model of islet carcinoma (Bergers et al., 2003), demonstrating that modulation of pericyte behavior can be therapeutically beneficial. Further development of such approaches, however, requires a better understanding of the biological factors that control pericyte behavior.

Secreted protein acidic and rich in cysteine (SPARC) is a matricellular protein whose expression is induced during angiogenesis (Lane et al., 1994; Iruela-Arispe et al., 1995). SPARC has been implicated in cellular processes critical to angiogenesis, including migration, proliferation, and differentiation (Kupprion et al., 1998; Francki et al., 2003; Motamed et al., 2003; Chlenski et al., 2007). The activity of SPARC appears to be indirect and relies in part on its ability to influence various growth factor signaling pathways. For example, SPARC can directly interact with VEGF-A and with PDGF-BB and PDGF-AB and prevent their interaction with cell surface receptors (Raines et al., 1992; Kupprion et al., 1998). However, SPARC antagonizes FGF2/FGFR1 signaling without interacting with either the soluble growth factor or the receptor (Hasselaar and Sage, 1992; Kupprion et al., 1998; Motamed et al., 2003). In addition to controlling growth factor signaling, SPARC also orchestrates the deposition of the ECM and can modulate the interaction between cells and their substratum (Murphy-Ullrich et al., 1995; Weaver et al., 2008; Bradshaw, 2009). Using an orthotopic model of pancreatic cancer, we found that pericyte recruitment was decreased in tumors grown in SPARC-deficient mice (Puolakkainen et al., 2004; Arnold et al., 2010). Though the mechanism underlying this observation is unknown, the data suggest that SPARC facilitates pericyte behavior in vivo.

TGF-β1 is a pleiotropic cytokine expressed by vascular cells during angiogenesis. TGF-β1 is secreted in a latent complex, which must be cleaved or otherwise manipulated to expose the active protein (Derynck et al., 1986; Munger et al., 1999; Annes et al., 2004). The precise cellular responses induced by active TGF-β1 depend on the specific TGF-β receptors expressed and the level of receptor expression (Schmierer and Hill, 2007). These responses vary and include migration, apoptosis, and proliferation (Goumans et al., 2003; Guasch et al., 2007; Daly et al., 2008; Yamashita et al., 2008). TGF-β1 receptors are ubiquitous throughout all tissues; therefore, the activation of latent TGF-β1 and the bioavailability of active TGF-β1 are tightly regulated (Lyons et al., 1988; Imai et al., 1997; Saharinen et al., 1999). Active TGF-β1 inhibits pericyte migration and induces expression and secretion of basement membrane proteins; accordingly, TGF-β1 signaling in these cells is restricted during angiogenesis, occurring only upon contact with endothelial cells of newly formed vessels (Sato and Rifkin, 1989; Kojima et al., 1991; Owens, 1995). Although the mechanisms behind this regulation are not entirely clear, a plausible scenario is αV integrin–mediated control of TGF-β activation, which could occur on the surface of pericytes that have reached the nascent endothelial tube.

Pericytes express TGF-β receptor II (TβRII) and the type II TGF-β1 receptor ALK5. In addition to these signaling receptors, we show that pericytes also express the accessory receptor endoglin. Endoglin, an established regulator of TGF-β1 activity in endothelial cells, interacts with the activated TGF-β1–TGF-β receptor complex and controls endothelial cell behavior by affecting TβRII/ALK1 and TβRII/ALK5 signaling and focal adhesion assembly (Conley et al., 2004; Sanz-Rodriguez et al., 2004).

In the current study, we sought to determine the mechanism by which SPARC regulates pericyte behavior. We report that SPARC promotes pericyte migration by decreasing TGF-β1 activity. We found that SPARC is expressed by pericytes in the vasculature of the adult murine pancreas and in pancreatic ductal adenocarcinoma (PDAC), where it facilitates pericyte migration by preventing endoglin from interacting with αV integrins, thereby repressing TGF-β1 activity. SPARC’s capacity to regulate pericyte recruitment highlights its function as a critical component of tissue remodeling and angiogenesis.

Previously, we found that orthotopic pancreatic tumors grown in SPARC^−/− mice contained fewer α smooth muscle actin⁺ (α-SMA⁺) cell–invested blood vessels than tumors grown in SPARC^+/+ mice (Puolakkainen et al., 2004; Arnold et al., 2010). As α-SMA is expressed by only a subset of pericytes, we sought to confirm this finding using the more general pericyte marker NG2 (Crisan et al., 2008). NG2 is expressed by resident pericytes associated with MECA32⁺ endothelial cells of the normal adult pancreas (Fig. 1 A). NG2⁺ cells were found to express SPARC in the vasculature of the normal adult pancreas and PDAC (Fig. 1, B and C, respectively). We crossed P48Cre:LSLKras^G12D:INK4A^lox/lox mice, which develop PDAC (Aguirre et al., 2003), to SPARC^+/+ and SPARC^−/− mice. Comparison of NG2⁺ cell recruitment between SPARC^+/+ PDAC and SPARC^−/− PDAC mice revealed fewer pericyte-associated MECA32⁺ vessels in SPARC^−/− PDAC tumors, confirming previous results (Fig. 1 F). We next asked whether endogenous SPARC influenced the behavior of primary pericytes in vitro. We used anti-NG2 immunomagnetic bead separation to purify pericytes from SPARC^+/+ and SPARC^−/− pancreas digests. Purified primary cells expressed pericyte markers and induced bEnd.3 endothelial cell cord formation (Fig. 1 G). Analysis of bEnd.3 cord parameters revealed that SPARC^−/− pericytes induced fewer cords; however, these cords were on average wider than SPARC^+/+ pericyte-induced cords. To investigate the disparity in pericyte function further, we assessed focal adhesion formation in primary SPARC^+/+ and SPARC^−/− pericytes (Fig. S1). We found that SPARC^+/+ pericytes exhibited more filopodia per cell compared with SPARC^−/− pericytes when spreading on fibronectin. Focal adhesion area was also greater in SPARC^−/− pericytes, whereas the overall number was decreased.

We next assayed migration toward fibronectin, a provisional matrix protein, using a transwell assay. SPARC^−/− pericytes exhibited a significantly reduced capacity to migrate, a feature that was reversed by the addition of recombinant SPARC (Fig. 2 A). Recombinant SPARC did not enhance the migration of SPARC^+/+ pericytes, suggesting that endogenous SPARC is sufficient for optimal migration (Fig. 2 C). Furthermore, small hairpin RNA (shRNA)–mediated knockdown or IgG-mediated neutralization of SPARC in SPARC^+/+ pericytes reduced their capacity to migrate (Fig. 2 B). Thus, pericyte-derived SPARC mediates pericyte recruitment to blood vessels, at least in part, by promoting cell migration.

SPARC can regulate TGF-β1 activity in other cell types; therefore, we assessed the contribution of TGF-β1 to the migration phenotype of SPARC^−/− pericytes (Schiemann et al., 2003; Francki et al., 2004; Chlenski et al., 2007). For this, we used primary pericytes and the 10T1/2 mesenchymal cell line (Reznikoff et al., 1973). Like pericytes, 10T1/2 cells can differentiate into mesenchymal lineages (Lien et al., 2006; Crisan et al., 2008; Lee et al., 2008a; Qu et al., 2008; Boeuf et al., 2009). These cells also function as pericytes when co-cultured with endothelial cells (Hirschi et al., 1998; Darland and D’Amore, 2001). Using our transwell assay, we found that TGF-β1 could inhibit both SPARC^+/+ pericyte and 10T1/2 cell migration (Fig. 3 A). To determine whether TGF-β was responsible for the migration phenotype of SPARC^−/− pericytes, we used a TGF-β–neutralizing antibody. Surprisingly, TGF-β neutralization enhanced the migration of SPARC^−/− cells only (Fig. 3 B and Fig. S2 A). The ALK5 inhibitor SB431542 had a similar effect, suggesting that TGF-β receptor activity is enhanced in the absence of SPARC expression (Fig. 3 C). We next attempted to recapitulate our findings using 10T1/2 cells. We found that shRNA-mediated knockdown of SPARC impaired 10T1/2 cell migration in a TGF-β–dependent manner (Fig. 3 D and Fig. S2 B). Furthermore, neutralization of SPARC using a monoclonal antibody reduced 10T1/2 cell migration in a TGF-β–dependent manner (Fig. 3 E). Thus, SPARC prevents TGF-β–dependent attenuation of pericyte migration.

We found that TGF-β1 could reduce SPARC^+/+ pericyte and 10T1/2 cell migration and, therefore, hypothesized that increased TGF-β1 expression in SPARC^−/− pericytes was responsible for their migration phenotype. Surprisingly, analysis of TGF-β1 in cell lysates and conditioned media revealed no differences in TGF-β1 levels in SPARC^+/+ and SPARC^−/− pericytes (Fig. 4 A). RT-PCR also revealed no difference in TGF-β1 expression between SPARC^+/+ and SPARC^−/− pericytes (Fig. 4 B). However, we did observe an increase in PAI1 expression in SPARC^−/− pericytes, which is a canonical TGF-β1 response gene (Fig. 4 B). Using quantitative PCR (qPCR), we confirmed that SPARC reduced the expression of canonical TGF-β1 response genes in pericytes (Fig. 4 C). Interestingly, SPARC^−/− pericytes seemed to be more sensitive to exogenous TGF-β1 (Fig. S3 B). As TGF-β1 signaling is initiated at the cell surface, we hypothesized that SPARC^−/− pericytes had more surface-associated TGF-β1. Indeed, we observed more TGF-β1 on the surfaces of nonpermeabilized SPARC^−/− pericytes (Fig. S3 A). We also assessed the level of surface-associated TGF-β1 using an impermeable cross-linker. We found that SPARC^−/− pericytes had more surface-associated TGF-β1 compared with SPARC^+/+ cells, a feature that was reversed by recombinant SPARC (Fig. 4 E). We next tested the possibility that SPARC^−/− pericytes were responding to endogenous TGF-β1. For this, we examined phosphorylated SMAD2 (pSMAD2) in response to shRNA-mediated knockdown of TGF-β1. We found that knockdown of TGF-β1 reduced SMAD2 phosphorylation only in SPARC^−/− pericytes but did not inhibit their capacity to respond to exogenous TGF-β1 (Fig. S3 C and Fig. 4 D). Finally, using TGF-β1 knockdown, we found that endogenous TGF-β1 was required for the migration phenotype of SPARC^−/− pericytes (Fig. 4 F). Thus, SPARC reduces activation of endogenous TGF-β1.

SPARC is a secreted glycoprotein; therefore, we hypothesized that control of TGF-β1 activity may be mediated through interactions with TGF-β1 receptors. SPARC can interact with a soluble form of TβRII but only in the presence of recombinant TGF-β1 (Francki et al., 2004). TGF-β1 binds sequentially to its receptors: active TGF-β1 first binds to TβRII, which can be in a heteromeric complex with endoglin, which then recruits a type I receptor, such as ALK5. We immunoprecipitated each of these receptors from 10T1/2 cells and found that SPARC specifically coprecipitated with endoglin (Fig. 5 B). SPARC was also detected in endoglin immune complexes from SPARC^+/+ pericytes (Fig. 5 B, right). We confirmed this interaction using solid-phase binding assays (Fig. 5 C). Furthermore, immunofluorescent staining of primary SPARC^+/+ pericytes revealed that SPARC and endoglin associate in distinct punctate structures (Fig. 5 A). This interaction links SPARC to TGF-β1 signaling.

The function of endoglin in TGF-β1 signaling is unclear; however, endoglin has been shown to modulate SMAD phosphorylation as well as control cell adhesion and migration, presumably through regulating the composition of focal adhesion complexes (Gougos et al., 1992; Conley et al., 2004; Koleva et al., 2006; Lee et al., 2008b). Therefore, we explored whether endoglin exhibited differences in localization and activity in SPARC-deficient pericytes. We first assessed endoglin association with focal complexes of spreading SPARC^+/+ or SPARC^−/− pericytes. Vinculin-rich focal complexes were observed in spreading SPARC^+/+ and SPARC^−/− pericytes (Fig. 6 A). We found that endoglin localized with focal complexes only in SPARC^−/− pericytes (Fig. 6 A, insets and magnified images). To determine whether endoglin aberrantly associates with focal adhesion machinery in the absence of focal complexes, we immunoprecipitated endoglin complexes from SPARC^+/+ and SPARC^−/− pericytes in suspension or seeded on plastic or fibronectin and probed for FAK, as FAK becomes incorporated into maturing focal complexes (Kornberg et al., 1992). We found that FAK was present in endoglin immune complexes from SPARC^−/−, but not SPARC^+/+, pericytes and that this association required cell adhesion (Fig. 6 B). We confirmed that FAK-associated endoglin was SPARC dependent using recombinant SPARC (Fig. 6 C). These results suggest that SPARC prevents endoglin from incorporating into focal complexes.

Next, we asked whether endoglin was required for elevated basal SMAD2 phosphorylation in SPARC^−/− pericytes. We found shRNA-mediated knockdown of endoglin resulted in decreased SMAD2 phosphorylation in SPARC^−/−, but not SPARC^+/+, pericytes (Fig. 6 D).

We then asked whether endoglin participates in the migration phenotype of SPARC^−/− pericytes. To test this, we used endoglin shRNA in 10T1/2 cells while targeting SPARC with either shRNA or a monoclonal antibody in a transwell assay. Endoglin knockdown had no effect on 10T1/2 cells when SPARC was not manipulated (e.g., control shRNA or control antibody; Fig. 6, E and F). However, knockdown of endoglin increased migration of cells transfected with SPARC shRNA or treated with an anti-SPARC antibody. Collectively, these results suggest focal complex–associated endoglin facilitates TGF-β1 activity and that SPARC functions to limit this process.

Endoglin does not directly activate TGF-β1–mediated processes (Koleva et al., 2006; Lee et al., 2008b). Association of endoglin with focal complexes and the regulation of migration and SMAD2 phosphorylation in SPARC^−/−, but not SPARC^+/+, pericytes suggested that, perhaps, in the absence of SPARC, endoglin was cooperating with factors that positively regulate TGF-β signaling. αV integrins facilitate cell adhesion and migration and can directly activate latent TGF-β1 (Delannet et al., 1994; Klemke et al., 1994; Liaw et al., 1995; Stefansson and Lawrence, 1996). Therefore, we sought to determine whether endoglin cooperated with αV integrins to enhance TGF-β1 activity in SPARC^−/− pericytes. We found that SPARC^+/+ and SPARC^−/− pericytes express similar levels of αV integrins by RT-PCR and flow cytometry (Fig. 7 A). Furthermore, SPARC^+/+ and SPARC^−/− pericytes express β integrin subunits that have been implicated to participate with αV integrin in mediating TGF-β1 activation (Munger et al., 1998; Mu et al., 2002; Ludbrook et al., 2003; Annes et al., 2004). We also found no differences in αV integrin–mediated migration between SPARC^+/+ and SPARC^−/− pericytes (Fig. 7 B). We then asked whether endoglin-associated αV integrins are required for the increased basal TGF-β1 activity observed in SPARC^−/− pericytes. Treatment of cells with an αV integrin–blocking antibody reduced SMAD2 phosphorylation in SPARC^−/−, but not SPARC^+/+, pericytes, though the extent of inhibition was not as great as that seen in cells treated with SB431542 (Fig. 7 C). We then examined whether αV integrin was present in endoglin immune complexes. αV integrin was detected in SPARC^−/− pericyte immune complexes regardless of whether cells were adherent or in suspension (Fig. 7 D). αV integrin was also detected in SPARC^+/+ pericyte immune complexes, albeit at lower levels (Fig. 7 D). As the level of αV integrin present in SPARC^−/− endoglin immune complexes was greater than that seen in SPARC^+/+ complexes, we thought that SPARC may limit αV integrin and endoglin association. Thus, we assessed the effect of adding recombinant SPARC to SPARC^−/− pericytes on the amount of endoglin-associated αV integrin. We found that exogenous SPARC reduced endoglin-associated αV integrin while having no effect on total αV integrin levels (Fig. 7 E). Collectively, these results show that SPARC, by blocking endoglin from interacting with αV integrins, can reduce TGF-β activity.

In the current study, we demonstrate that SPARC promotes pericyte migration by reducing TGF-β1–induced responses. To perform this function, SPARC interacts with the TGF-β1 accessory receptor endoglin. In the absence of SPARC, endoglin associates with αV integrins and enhances TGF-β1 signaling to impair pericyte migration.

During angiogenesis, pericytes migrate to nascent vessels where TGF-β1 signaling impedes further migration and triggers pericyte-induced vessel quiescence (Courtoy and Boyles, 1983; Antonelli-Orlidge et al., 1989; Sato and Rifkin, 1989; Stefansson and Lawrence, 1996; Hirschi et al., 1998; Darland and D’Amore, 2001). As TGF-β1 is present in the extracellular milieu throughout endothelial tube formation, the capacity of pericytes to respond to TGF-β1 must be regulated spatially. Mechanisms that contribute to such regulation remain unknown. We found that SPARC was expressed in pericytes during vascular morphogenesis in PDAC, and pericyte recruitment was reduced in the absence of SPARC. We propose that SPARC can block the capacity of pericytes to respond to TGF-β1 during angiogenesis and, thus, facilitate cell migration to nascent blood vessels, based on the following observations: First, SPARC-deficient pericytes migrated less in vitro, an effect that was reversed when adding back recombinant SPARC, blocking TGF-β–TGF-β receptor ligation, inhibiting ALK5 kinase activity, or knocking down TGF-β1. Second, SPARC-deficient pericytes exhibited increased basal TGF-β1–induced SMAD2 phosphorylation and activity. Third, SPARC deficiency resulted in increased TGF-β1 associated with pericyte surfaces while not effecting levels of TGF-β1 mRNA, cytosolic TGF-β1, or secreted TGF-β1. SPARC thus may control TGF-β1 perception by pericytes during blood vessel formation. Interestingly, we found that SPARC^−/− pericytes were more sensitive to exogenous TGF-β1 (Fig. S3 B). Integrins αVβ3 and αVβ5 have been shown to promote TGF-β1 signaling and target gene expression by enhancing TGF-β receptor activity through direct receptor interactions; therefore, it seems reasonable that endoglin–αV integrin complexes have an amplified response to exogenous TGF-β1 (Scaffidi et al., 2004; Asano et al., 2006a,b; Galliher and Schiemann, 2006).

In spite of our findings, the precise mechanism through which SPARC regulates TGF-β1 activity is still unclear. This is because of several independent studies with opposing conclusions on the effect of SPARC on TGF-β1 activity (Schiemann et al., 2003; Francki et al., 2004; Chlenski et al., 2007). In the current study, we found that SPARC required endoglin to regulate TGF-β1 activity in pericytes, a finding that suggests its effects may be dependent on this accessory receptor in other cells types. Studies assessing the contribution of TGF-β1 accessory receptors to SPARC activity will undoubtedly shed light on why SPARC and TGF-β1 expression temporally overlap during development and disease.

Endoglin is a critical component of the TGF-β–signaling machinery and is required for development of the vasculature. Genetic ablation of endoglin in mice results in embryonic lethality from defective vascular remodeling, a phenotype that resembles that of TGF-β1 KO mice (Dickson et al., 1995; Bourdeau et al., 1999; Li et al., 1999; Arthur et al., 2000). Interestingly, endoglin is not required for formation of the initial vascular plexus, rather vessel defects result from lack of mural cell recruitment to the early vascular network. Vascular cell expression of endoglin is increased during angiogenesis; however, the function of endoglin in pericytes has been unclear. Similarly, αV integrin is expressed by mural cells, but its function on these cells during angiogenesis is unclear (Stawowy et al., 2003; Gao and Brigstock, 2004). Our results thus place these proteins together with SPARC in a mechanism to regulate the activation of latent TGF-β1. We found that endoglin was required for the enhancement of TGF-β–induced responses we observed in the absence of SPARC. SPARC also prevented endoglin from associating with αV integrins, likely via a mechanism involving direct interaction with the extracellular domain of endoglin. We propose that the capacity of endoglin to block pericyte migration in the absence of SPARC requires its association with focal complex–associated proteins. In support of this, we found that recombinant SPARC disrupted endoglin–αV integrin and endoglin–FAK complex formation, with disruption of endoglin–FAK complexes by SPARC being more efficient. Other groups have shown that FAK and other focal adhesion proteins dissociate from integrins after internalization (Finnemann, 2003; Pellinen and Ivaska, 2006; Thomas et al., 2010). We found that endoglin–αV integrin complexes were not dependent on the formation of focal adhesions, a finding that suggests these complexes persist upon integrin internalization. Such a scenario would result in a pool of endoglin–αV complexes that were not associated with focal adhesions and might be less sensitive to SPARC-induced dissociation. These complexes would likely make it difficult to discern complex dissociation at focal adhesions in our assay. Thus, we would predict that relatively high concentrations of SPARC would be required to perturb endoglin–αV complexes (Fig. 7 E). Conversely, we found endoglin–FAK complexes only when SPARC^−/− pericytes were allowed to form focal adhesions. This suggests that prevention of endoglin from interacting with focal adhesion proteins (e.g., with SPARC) would result in the disruption of endoglin–FAK complexes. Indeed, we saw endoglin–FAK complex dissociation at a threefold lower concentration of SPARC compared with endoglin–αV complexes.

Endoglin can control cell migration, at least in part, via intracellular interactions. Previous studies have demonstrated direct interaction between the cytoplasmic PDZ-interacting motif of endoglin and GIPC1. This interaction resulted in retention of surface-associated endoglin in focal complexes, increased SMAD phosphorylation, and reduced cell migration (Lee et al., 2010; Ray et al., 2010). The cytoplasmic domain of endoglin also facilitates blockade of zyxin and ZRP-1 recruitment into maturing focal complexes; thus, endoglin may reduce migration as a result of suboptimal focal adhesion assembly (Conley et al., 2004; Sanz-Rodriguez et al., 2004). In addition to intracellular interactions, our results demonstrate that endoglin can influence cell migration via extracellular interactions. Determining whether the effect of SPARC on endoglin activity requires the cytoplasmic tail of endoglin and subsequent changes in focal complex assembly will provide insight on the mechanism of SPARC control of pericyte behavior and may help explain how SPARC regulates cell morphology in general.

Activated TGF-β receptors regulate pericyte migration via their inherent kinase activity; for example, TGF-β1 inhibits migration by stimulating ALK5-dependent phosphorylation of SMADs and p38 (Feinberg et al., 2004). TGF-β1 signaling is potentiated upon recruitment of TGF-β receptors into focal complexes, though the mechanisms that drive receptor recruitment to these structures are unclear (Scaffidi et al., 2004; Asano et al., 2006b; Galliher and Schiemann, 2006; Lee et al., 2010). In addition to interacting with intracellular components of focal adhesions, endoglin also interacts with TβRII in the absence of a bound ligand, a feature that suggests it may serve to bridge the gap between TGF-β1 signaling and focal complexes (Barbara et al., 1999; Guerrero-Esteo et al., 2002). We found that in the absence of SPARC, endoglin facilitated increased SMAD2 phosphorylation. As endoglin does not exhibit kinase activity, this was likely the result of TGF-β receptor recruitment to focal complexes.

TβRII/ALK5 signaling is required for the decreased migration observed in the absence of SPARC; however, it remains to be determined whether endoglin localization to focal complexes is dependent on TβRII/ALK5 activity. In our model, endoglin links these receptors to TGF-β1, which predicts that association of endoglin with focal complexes is upstream of any kinase activity.

The question then becomes: how does focal complex–associated endoglin enhance TGF-β1 activity? We propose that endoglin bridges TGF-β receptors and αV integrin–associated TGF-β1. Our results support this proposal based on the following: First, αV integrins were found in the endoglin-enriched focal complexes observed in the absence of SPARC. Second, recombinant SPARC increased migration of SPARC^−/− pericytes while decreasing endoglin–αV integrin complex formation. Third, αV integrins enhanced ALK5/TβRII activity but only in the absence of SPARC. These results are in line with other studies demonstrating that SPARC can interact with and regulate focal adhesion proteins (Murphy-Ullrich et al., 1995; Motamed and Sage, 1998; Shi et al., 2004; Barker et al., 2005; Weaver et al., 2006, 2008; Nie et al., 2008). Recently, SPARC was shown to interact with β1 integrins (Nie et al., 2008; Weaver et al., 2008). Our data do not rule out the possibility that SPARC interacts with endoglin and β integrins to control αV integrin–TGF-β1 signaling. Indeed, αVβ1 integrin can bind to the latency-associated peptide (LAP), though actual activation of latent TGF-β1 has yet to be demonstrated clearly (Munger et al., 1998).

TGF-β1 LAP contains an RGD domain to which all five αV integrins can bind (Munger et al., 1999; Mu et al., 2002; Ludbrook et al., 2003). Furthermore, interaction of latent TGF-β1 with αVβ3, αVβ5, αVβ6, and αVβ8 results in the presentation of TGF-β1 to its receptors. In our model, in the absence of SPARC, αV integrins interact with latent TGF-β1 and present the active protein to endoglin-associated TGF-β receptors, facilitating the inhibition of migration. In wild-type (WT) cells, SPARC interacts with endoglin and prevents it from recruiting TGF-β receptors to αV integrin–latent TGF-β1 complexes, promoting pericyte migration (Fig. 8).

There are currently two known mechanisms of αV integrin–mediated activation of TGF-β1, one requiring protease activity and the other being protease independent. Protease-independent activation of latent TGF-β1 has been shown to occur with αVβ3, αVβ5, and αVβ6 integrins (Ludbrook et al., 2003; Annes et al., 2004; Wipff et al., 2007). In this mechanism, traction is proposed to induce the release of active TGF-β1 from the latent complex. Binding of active TGF-β1 requires TGF-β receptors to be in close proximity, a feature that explains why release of active TGF-β1 into the culture media is never observed. Interestingly, we failed to detect changes in active TGF-β1 in culture media, suggesting this mechanism may be involved.

Protease-dependent activation of TGF-β1 has only been demonstrated with αVβ8: αVβ8 binds latent TGFβ1 and recruits MT1-MMP, which then releases active TGF-β1 via proteolytic cleavage of LAP (Mu et al., 2002). As we failed to detect changes in TGF-β1 in culture media, it is unlikely that an αVβ8/MT1-MMP axis is involved in SPARC-mediated regulation of pericyte behavior.

Endoglin is an established regulator of endothelial cell behavior. We describe here a novel mechanism in which endoglin cooperates with SPARC to regulate pericyte responses to TGF-β1. Our findings suggest that during angiogenesis, SPARC functions to restrict pericyte perception of TGF-β1 in the angiogenic milieu through its interaction with endoglin. Such control is critical, as TGF-β1 is present early in the angiogenic cascade, yet TGF-β1 signaling in pericytes must occur after formation of endothelial tubes.

SPARC was expressed in both resting and mobilized pericytes, an observation that suggests the function of SPARC is also controlled. Indeed, previous studies have demonstrated that SPARC can be cleaved by plasmin, whose own expression is temporally regulated throughout angiogenesis (Lane et al., 1992, 1994; Iruela-Arispe et al., 1995). Furthermore, the integrin profile of pericytes changes during angiogenesis, a phenomenon that may also affect SPARC-mediated blockade of TGF-β1. Experiments addressing how these events regulate SPARC activity will undoubtedly yield important insights on how matricellular proteins, such as SPARC, regulate critical physiological processes.

PDAC (P48Cre⁺:LSLKrasG12D⁺:INK4A^lox/lox) mice were crossed with SPARC^+/+ or SPARC^−/− mice to produce WT PDAC or SPARC-null PDAC (knockout [KO] PDAC) mice. For tumor analyses, mice were sacrificed once becoming moribund, with at least six mice per group. Tumors were preserved in formalin or snap frozen using liquid nitrogen. Animal experiments were performed at the University of Texas Southwestern Medical Center at Dallas in compliance with the Animal Welfare Act, the Public Health Service Policy, and the U.S. Government Principles Regarding the Care and Use of Animals.

Antibodies to the following proteins were used for indirect immunofluorescent microscopy: α-SMA (NeoMarkers), NG2 (AB5320; Millipore), desmin (Ab907; Millipore), vinculin (V4139; Sigma-Aldrich), MECA32, SPARC (R&D Systems), total TGF-β (SC146; Santa Cruz Biotechnology, Inc.), and endoglin (MJ7/18). For function-blocking assays, a pan–TGF-β–neutralizing antibody was purchased from R&D Systems (1D11), an αVβ6-blocking antibody was a gift from D. Sheppard (University of California, San Francisco, San Francisco, CA), and an αV integrin–blocking antibody was purchased from BioLegend (RMV-7). The hybridomas that produce mAb293 and mAb303 were grown in our laboratory and purified by protein A chromatography. For Western blots, ALK5 (SC-398; Santa Cruz Biotechnology, Inc.), TβRII (SC-220; Santa Cruz Biotechnology, Inc.), endoglin (clone MJ7/18), αV integrin (AB1930; Millipore), FAK (3285; Cell Signaling Technology), phospho-SMAD2 (AB3849 serine 465/467; Millipore), and total SMAD2 (tSMAD2; 3107; Cell Signaling Technology) were used. For solid-phase binding assays, endoglin (MJ7/18) and SPARC (mAb 236) were used. For immunoprecipitations, SPARC (mAb303), ALK5, TβRII (SC-220), and endoglin (MJ7/18) were used. The hybridomas MECA32 and MJ7/18, developed by E.C. Butcher (Stanford University, Palo Alto, CA), were obtained from the Developmental Studies Hybridoma Bank, developed under the auspices of the National Institute of Child Health and Human Development, and maintained by The University of Iowa.

Mouse pancreata from 4-wk-old SPARC^+/+ and SPARC^−/− mice were minced and then subjected to digestion with 1% collagenase type 1, DME, 10 mM Hepes, 1% fetal bovine serum, and PBS at 37°C until a single-cell suspension was obtained. Cell suspensions were centrifuged at low speed to pellet large debris, resuspended in wash buffer, and passed through a 70-µm cell strainer. The resulting cell suspension was then incubated with sheep anti–rabbit IgG-conjugated magnetic Dynabeads (Invitrogen) and rabbit anti-NG2 IgG (Millipore) at 4°C. Dynabeads were preincubated with anti-NG2 IgG overnight at 4°C on a nutator and then washed three times in wash buffer to remove NaN₂. Bead-bound cells were separated from unbound cells using a cell separation magnet (IMagnet; BD). Primary pericytes were maintained in 10% fetal bovine serum–supplemented DME and used between passage 1 and 7 for experiments. 10T1/2 cells were used before 10 passages and maintained in 10% fetal bovine serum–supplemented DME. Primary pericytes were transfected using Lipofectamine 2000 (Invitrogen), whereas 10T1/2 cells were transfected using Fugene (Roche). For shRNA knockdown of SPARC, endoglin, and TGF-β1, shRNA expression plasmids were purchased from Sigma-Aldrich (MISSION). 2 µg plasmid DNA was mixed with 3 µl transfection reagent and used to transfect 150,000 cells. Cells were used 48 h after transfection. For negative control transfections, a nontargeting shRNA expression plasmid was used (MISSION).

10,000 bEnd.3 endothelial cells were plated onto matrigel-coated 3-well chamber slides (BD) in the presence or absence of either 10,000 SPARC^+/+ or SPARC^−/− pericytes and allowed to self-assemble into cords for 17 h in DME supplemented with 0.75% fetal bovine serum at 37°C before visualization by fluorescent microscopy. Before use in the assays, bEnd.3 cells and pericytes were stained with either the red fluorochrome PKH26 or the green fluorochrome PKH67, respectively (Sigma-Aldrich). Experiments were performed three times and in triplicate. Images were taken at a 4× magnification, with five images taken per well. The peripheral zone of matrigel was avoided during image acquisition so to avoid cord artifacts associated with changes in surface elevation. Cord lengths and widths were calculated using NIS-Elements software (Nikon). For cord width measurements, widths were taken halfway into the length of each cord.

Transwell inserts with 8-µm pores were used for migration assays. Inserts were placed in 24-well tissue-culture plates for the duration of experiments. The bottom sides of the insert membranes were coated with 10 µl of 1-µg/µl fibronectin for 1 h at 37°C. Inserts were then used immediately for experiments (Sigma-Aldrich). 7,000 or 5,000 primary pericytes or 10T1/2 cells were added into the insert reservoir in DME in a total volume of 125 µl, whereas DME containing 0.1% fetal bovine serum was added into the tissue-culture plate well. Experimental conditions were always added to both the top and bottom of the transwell. Cells were allowed to migrate to the fibronectin-coated side of the insert membrane for 6 h. Cells on the noncoated side of the insert membrane were removed. Cells that migrated to the underside of the membrane were fixed in formalin and manually counted. Experiments were performed in triplicate and repeated two or three times as indicated in the figure legends.

A TGF-β1 ELISA kit (TGF-β1 EMAX Immunoassay G7591) that detects the active form of TGF-β1 was purchased from Promega. Sample preparation for active TGF-β1 ELISA was performed as follows: primary pericytes were seeded at 150,000 cells per well in 6-well tissue-culture plates and cultured in 0.75% fetal bovine serum–supplemented DME in triplicate. Cells did not exhibit any expansion and remained subconfluent and viable for the duration of the experiment. Conditioned media and cell lysates were collected at 24, 48, 72, and 96 h after seeding. A mammalian protein extraction reagent (M-PER) cell lysis buffer supplement with protease inhibitor (Complete Mini) was used for lysate preparation (Roche). ELISA was performed according to kit instructions. Active TGF-β1 concentrations were calculated by interpolating values onto a standard curve generated with TGF-β1 accompanying the kit. For qPCR expression analyses, fold change was calculated using the ΔΔ cycle threshold method, in which WT at 0 pg/ml was the reference sample, and GAPDH was the reference gene. Sample preparation for qPCR expression analyses were as follows: serum-starved primary pericytes were seeded at 100,000 cells per well in fibronectin-coated 6-well tissue-culture plates in triplicate. Active TGF-β1 purchased from PeproTech was added to serum-starved pericytes at final concentrations of 0, 50, and 5,000 pg/ml. Cells were then incubated for 17 h at 37°C. RNA was harvested using TRIZOL reagent (Sigma-Aldrich). cDNA was synthesized using iScript (Bio-Rad Laboratories). 12.5 ng cDNA was used per 96-well PCR plate well, with each tissue-culture plate well represented in three individual PCR plate wells. The following primer sets were used for qPCR or RT-PCR: CTGF forward, 5′-AGCCTCAAACTCCAAACACC-3′, and reverse, 5′-CAACAGGGATTTGACCAC-3′; PAI-1 forward, 5′-GACACCCTCAGCATGTTCATC-3′, and reverse, 5′-AGGGTTGCACTAAACATGTCAG-3′; BIGH3 forward, 5′-TGATAAGAGGGGACGGTTTG-3′, and reverse, 5′-ATTGGTGGGAGCAAAAACAG-3′; and GAPDH forward, 5′-AGAAGGCTGGGGCTCATTTG-3′, and reverse, 5′-AGGTCGGAGTCAACGGATTTG-3′.

To assess the effect of SPARC on SPARC^−/− pericyte transcription, pericytes were cultured for 72 h in the presence or absence of recombinant SPARC or BSA control. Media were replaced with fresh SPARC- or BSA-containing media every 24 h before RNA extraction.

Cells were incubated in 0.75% fetal bovine serum–supplemented DME overnight before RNA extraction and cDNA synthesis. The following primer sets were used for RT-PCR: endoglin (L-endoglin) forward, 5′-GCACTCTGGT­ACATCTATTCT­CACACACGTGG-3′, and reverse, 5′-GGGCACTACGCCATGCTGCTGGTGG-3′; SPARC forward, 5′-CTGCGTGTGAAGAAGATCCA-3′, and reverse, 3′-TGGGACAGGTACCCATCAAT-3′; ALK5 forward, 5′-GGCGACGGCATTACAGTGTT-3′, and reverse 5′-TGTACATACAAATGGCCTGT-3′; TβRII forward, 5′-GCAAGTTTTGCGATGTGAGA-3′, and reverse, 5′-GGTATCTTCCAGAGTTGAAGC-3′; TGF-β1 forward, 5′-TTGCTTCAGCTCCCACAGAGA-3′, and reverse, 5′-TGGTTGTAGAGGGCAAGGAC-3′; αV integrin: itgav forward 5′-GGGTGATCATCTTGGCAGTT-3′, and reverse, 5′-GAACTTGGAGCGGACAGAAG-3′; β1 integrin: itgb1 forward, 5′-GTGACCCATTGCAAGGAGAAGGAC-3′, and reverse 5′-GTCATGAATTATCATTAAAAGTTT-3′; β3 integrin: itgb3 forward, 5′-CTGGTGTTTACCGATGCCAAG-3′, and reverse, 5′-TGTTGAGGCAGGTGGCATTGAAGG-3′; β6 integrin: itgb6 forward, 5′-CCGGCTGGCCAAAGAGATGT-3′, and reverse, 5′-AGTTAATGGCAAAATGTGCT-3′; RPS6: rps6 forward, 5′-AAGCTCCGCACCTTCTATGAGA-3′, and reverse, 5′-TGACTGGACTCAGACTTAGAAGTAGAAGC-3′; and β-actin: actb forward, 5′-ATATCGCTGCGCTGGTCGTC-3′, and reverse, 5′-AGGATGGCGTGAGGGAGAGC-3′.

Pericytes were seeded at 100,000 cells per well of 6-well culture plates and cultured in 1.5% fetal bovine serum–supplemented DME for 17 h before being lysed in 300 µl sample buffer (62.5 mM Tris-HCl, pH 6.8, at 25°C, 2% SDS, 10% glycerol, 50 mM DTT, and 0.01% bromophenol blue). Lysates were subjected to SDS-PAGE and Western blotting for tSMAD2 and pSMAD2 (serine 465/467) immediately thereafter.

10T1/2 cells were lysed in modified radioimmunoprecipitation assay buffer (0.5% deoxycholate, 0.5% SDS, 1% Triton X-100, 10 mM sodium phosphate, pH 7.2, 150 mM sodium chloride, and protease inhibitor [Complete Mini]). Pericytes were lysed in a milder buffer containing 1% NP-40, 10 mM sodium phosphate, pH 7.2, 150 mM sodium chloride, and protease inhibitor (Complete Mini). Lysis was performed on serum-starved adherent cells after washing with chilled PBS. Cells were scraped using 1 ml modified radioimmunoprecipitation buffer. Lysates were allowed to rotate at 4°C on a nutator for 1 h and then vortexed several times before centrifuging at 13,000 rpm for 10 min to pellet any insoluble material. Lysates were then precleared with protein A/G beads (Thermo Fisher Scientific). 200 µg cellular protein in 1 ml lysis buffer was used per immunoprecipitation reaction. 1 µg of the appropriate IgG was added with 20 µl protein A/G bead slurry to each sample; each sample was then allowed to rotate overnight at 4°C on a nutator. Immunoprecipitated complexes were washed twice in lysis buffer and then boiled in sample buffer and subjected to SDS-PAGE and Western blot analysis.

Wells of 96-well clear-well assay plates were coated with recombinant human SPARC, recombinant human endoglin (R&D Systems), or serum (EastCoast Bio), blocked, and incubated with recombinant endoglin or recombinant SPARC. Bound endoglin or SPARC was detected with antiendoglin (MJ7/18) or anti-SPARC (mAb 303) antibodies or detected with horseradish peroxidase–conjugated secondary IgG. Assays were developed using tetramethylbenzidine reagent (Thermo Fisher Scientific). Samples were added in triplicate, and the experiment was repeated three times.

Primary pericytes were grown to 80% confluency and then switched to 0.75% fetal bovine serum–supplemented DME. Cells were then labeled with a cell surface protein isolation kit (Sulfo-NHS-SS-Biotin; Thermo Fisher Scientific) according to the manufacturer’s instructions. 4–10-cm dishes per pericyte genotype were used per fractionation. Fractionations were performed twice with identical results.

Epifluorescent images were taken using a microscope (Eclipse E600; Nikon) and a camera (CoolSNAP HQ; Photometrics). Images were acquired and analyzed using NIS-Elements software. For visualization of immunofluorescently stained cells, images were thresholded so as to not include a signal caused by the nonspecific binding of the fluorophore-conjugated secondary antibody alone and analyzed as JPEG 2000 files. Confocal images were taken using either a TCS-SP5 confocal microscope (Leica) or an Eclipse TE2000E confocal microscope (Nikon). Leica images were taken using the Imaging Application for Confocal SP5 software (Leica). Images were saved as Leica Image Files (.LIF) and analyzed using ImageJ software (National Institutes of Health). Contrast and brightness were adjusted equally in all channels using Photoshop (CS3 Extended; Adobe). Nikon images were taken using a camera (CoolSNAP ES) and EZ-C1 3.8 software (Photometrics). Images were saved in the native ICS/IDS format. Images were processed using NIS-Elements software. Channels were thresholded so as to not include autofluorescence from the assay medium (10% FBS in DME containing phenol red). Nikon epifluorescent and confocal objectives (plan fluorite) had the following numerical apertures: 10×, 0.3; 20×, 0.5; 40×, 0.75; and 100×, 1.3 in oil. Leica confocal images were taken using a 63× objective with a 1.4 numerical aperture in oil. Fluorescent staining was performed using cyanine (Cy3) or fluorescein isothiocyanate–conjugated secondary antibodies (Jackson ImmunoResearch Laboratories, Inc.).

Student’s t test analysis or analysis of variance was performed for all experiments.

Fig. S1 shows vinculin and phalloidin staining used to assess focal adhesion formation in primary pericytes. Fig. S2 shows exogenous SPARC blocked the anti-TGF-β–induced enhancement of SPARC^−/− pericyte migration. Fig. S3 shows that SPARC^−/− pericytes exhibit enhanced basal TGF-β activity.

We gratefully acknowledge the support from the members of the Brekken laboratory and guidance from Drs. Amy Bradshaw, Lance Terada, Philip Shaul, Joseph Albanesi, Michelle Tallquist, Shae Padrick, and Dean Sheppard.

This study was supported, in part, by the Effie Marie Cain Scholarship in Angiogenesis Research and a grant from the National Institutes of Health (R01CA118240 to R.A. Brekken). L.B. Rivera was supported, in part, by a training grant from the National Institutes of Health (T32 GM008203).