Growth Factors Engineered for Super-Affinity to the Extracellular Matrix Enhance Tissue Healing

See allHide authors and affiliations

Science  21 Feb 2014:
Vol. 343, Issue 6173, pp. 885-888
DOI: 10.1126/science.1247663

Toward Successful Tissue Repair

The therapeutic use of growth factors in tissue regeneration has suffered from safety and efficacy issues. Reasoning that the unmet potential may be because of nonphysiological delivery, Martino et al. (p. 885) engineered growth factors to bind strongly to extracellular matrix proteins. These variants were able to induce superior tissue repair, compared to the wild-type proteins. Furthermore, unwanted side effects were decreased: For example, the engineered angiogenic growth factor VEGF showed reduced vascular permeability, a concern that has limited the therapeutic efficacy of wild-type VEGF.


Growth factors (GFs) are critical in tissue repair, but their translation to clinical use has been modest. Physiologically, GF interactions with extracellular matrix (ECM) components facilitate localized and spatially regulated signaling; therefore, we reasoned that the lack of ECM binding in their clinically used forms could underlie the limited translation. We discovered that a domain in placenta growth factor-2 (PlGF-2123-144) binds exceptionally strongly and promiscuously to ECM proteins. By fusing this domain to the GFs vascular endothelial growth factor–A, platelet-derived growth factor–BB, and bone morphogenetic protein–2, we generated engineered GF variants with super-affinity to the ECM. These ECM super-affinity GFs induced repair in rodent models of chronic wounds and bone defects that was greatly enhanced as compared to treatment with the wild-type GFs, demonstrating that this approach may be useful in several regenerative medicine applications.

Tissue repair is strongly regulated by a number of growth factors (GFs), including vascular endothelial growth factor (VEGF), which triggers angiogenesis crucial for the repair of most tissues (1); bone morphogenetic protein–2 (BMP-2), which induces the formation of new bone (2); and platelet-derived growth factor–BB (PDGF-BB), which is critical for the formation of granulation tissue and recruitment of stem cells (3). These and other GFs have been explored therapeutically, yet their translation to clinical use in regenerative medicine has been limited (4), probably in part because of their use at supraphysiological levels, as well as issues related to safety and cost effectiveness (2, 59).

Physiologically, the partitioning, availability, and signaling of GFs are orchestrated by their binding to the extracellular matrix (ECM) (10, 11). We recently characterized high-affinity GF binding sites in fibronectin (12), fibrinogen (13), and tenascin C (14), and we showed that ECM binding and co-ligation with integrins can modulate the signaling of multiple GFs (15).We took a GF protein engineering approach to enhance ECM binding, reasoning that engineering second-generation GF variants may enhance their activity and provide a simple delivery system to address these issues.

We began by screening 25 molecules from the VEGF/PDGF, transforming growth factor–β (TGF-β), fibroblast growth factor (FGF), and neurotrophin GF families for binding to six key ECM proteins: fibronectin, vitronectin, tenascin C, osteopontin, fibrinogen, and collagen I. As we previously reported, many GFs are able to bind fibronectin (12), fibrinogen (13), and tenascin C (14); here we also show that numerous GFs are also able to bind vitronectin,and osteopontin (Fig. 1). Most displayed poor binding to collagen I, which is not surprising given that no GF binding site has yet been reported. Among all the GFs screened, PlGF-2 displayed the strongest binding to all of the ECM proteins tested (Fig. 1). To the contrary, PlGF-1 did not display any binding. PlGF-2 and PlGF-1 are splice variants, with PlGF-2 but not PlGF-1 containing a heparin-binding (16) sequence (RRPKGRGKRRREKQRPTDCHL, PlGF-2123-144) near the C terminus (fig. S1A). By fusing this sequence to the model nonbinding protein glutathione-sulfotransferase (GST), we revealed that this domain is responsible for the binding characteristics of PlGF-2 to the tested ECM proteins (fig. S1, B and C); a scrambled PlGF-2123-144 sequence fused to GST did not show any specific binding (fig. S1B).

Fig. 1 GF binding to ECM proteins, measured by enzyme-linked immunosorbent assay.

A signal over 0.1 (gray box) was considered to be significant. PlGF-2 strongly bound all ECM proteins tested (red bars). Bovine serum albumin was used as a control. n ≥ 3 experiments, mean ± SEM.

Based on the knowledge of the high affinity and promiscuous ECM protein-binding domain within PlGF-2, we used rational protein engineering to incorporate PlGF-2123-144 into GFs that bear clinical translation limitations (59), namely VEGF-A, PDGF-BB, and BMP-2. Because VEGF-A and PlGF are structurally related, the alignment of VEGF-A165, VEGF-A121, and PlGF-2 sequences (fig. S2A) suggested substitution of the heparin-binding domain of VEGF-A165 with PlGF-2123-144 to generate the engineered putative super-affinity variant VEGF-A/PlGF-2123-144. We fused PlGF-2123-144 to the C terminus of PDGF-BB to obtain PDGF-BB/PlGF-2123-144, and we similarly generated BMP-2/PlGF-2123-144*, fusing a version of PlGF-2123-144 with a substitution of cysteine-142 by serine (PlGF-2123-144*), because the presence of the cysteine impaired production of the fusion protein (fig. S2B).

Insertion of PlGF-2123-144 into GFs did not alter their ability to activate their receptors, as determined by phosphorylation of VEGFR-2 in endothelial cells (ECs) and phosphorylation of PDGFR-β or induction of alkaline phosphatase in mesenchymal stem cells (fig. S3). Insertion of the PlGF-2123-144 domain conferred super-affinity for ECM proteins and heparan sulfate, as determined by measurement of binding affinity [dissociation constant (KD) value] (Table 1). KD values of the PlGF-2123-144–fused GFs for ECM proteins were driven to similar values as those displayed by PlGF-2, resulting in 2- to 100-fold elevations in affinity as compared to the wild type, depending on the particular GF (Table 1 and figs. S4 and S5). Moreover, PlGF-2123-144–fused GFs could be strongly retained in a fibrin matrix mimicking a clot (fig. S6), and they could be released by the protease plasmin, in addition to association-dissociation kinetics (fig. S7).

Table 1 Affinity (KD is shown) of wild-type versus PlGF-2123-144–fused GFs for ECM proteins and heparan sulfate (HS).

n = 3 enzyme-linked immunosorbent assays, mean ± SEM.

View this table:

After having verified that the engineered GFs could bind to and be retained by ECM molecules in vivo (fig. S8), we tested whether delivering GFs in the context of the strong association of the PlGF-2123-144–fused domain with endogenous ECM would significantly enhance their capacity to induce tissue healing at doses where wild-type GFs are usually not effective. As a first model, we used skin wound healing in the diabetic db/db mouse, which is a well-established and relevant experimental model of impaired wound healing (17). Taking advantage of the synergistic and combinatorial effect between VEGF-A165 and PDGF-BB (18), we reasoned that VEGF-A165 would induce EC recruitment, whereas vessel stabilization by smooth muscle cells (SMCs) and pericytes as well as granulocyte recruitment would be driven by PDGF-BB (15). Because 20 μg per wound of VEGF-A165 or 10 μg per wound of PDGF-BB applied topically for five consecutive days were known to promote wound healing in the db/db mouse (19, 20), we treated full-thickness back-skin wounds with a roughly 40- to 250-fold lower dose of GFs (200 ng each of PDGF-BB and VEGF-A, combined) delivered once in a fibrin matrix or simply applied topically three to four times. Low doses of wild-type PDGF-BB and VEGF-A did not significantly enhance wound healing as compared to untreated or fibrin alone–treated wounds, as indicated by either extent of wound closure (indicated by re-epithelialization) (Fig. 2, A and F) or amount of granulation tissue (Fig. 2, B and G). In contrast, wounds treated with PlGF-2123-144–fused PDGF-BB and VEGF-A led to significantly faster wound closure and more granulation tissue, both topically (Fig. 2, A to C) and in fibrin (Fig. 2, F to H). Because angiogenesis is a crucial step in sustaining newly formed granulation tissue (21), we compared how angiogenesis differed between the treatments. Immunohistological analysis for CD31 (highly expressed by ECs) and desmin (expressed by SMCs) revealed that angiogenesis within the granulation tissues was much more pronounced when PlGF-2123-144–fused GFs were delivered (Fig. 2, D, E, I, and J; and figs. S9 to S11). In light of these results, EC response to VEGF-A/PlGF-2123-144 may be different from that to the wild-type forms in vivo, because VEGF-A binding to the ECM critically controls angiogenesis via modulation of VEGF-A signaling kinetics (22).

Fig. 2 VEGF-A/PlGF-2123-144 and PDGF-BB/PlGF-2123-144 induce greater skin wound healing and angiogenesis than wild-type VEGF-A and PDGF-BB.

(A to J) Low doses of VEGF-A/PlGF-2123-144 and PDGF-BB/PlGF-2123-144 promoted skin wound healing in diabetic mice, whereas wild-type GFs did not. Full-thickness back-skin wounds (6 mm in diameter) were treated with GFs (200 ng of each, combined) delivered either topically three to four times or in a fibrin matrix once. After 10 and 15 days [topical groups, (A) and (B)], or 7 and 10 days [fibrin groups, (F) and (G)], wound closure and granulation tissue formation were evaluated by histology. n ≥ 8 wounds per group per time point, mean ± SEM. Analysis of variance (ANOVA) with Bonferroni post hoc test for pairwise comparisons; *P < 0.05, **P < 0.01, ***P < 0.001. (C) and (H) Representative histology at 10 days for the fibrin groups and at 15 days for the topical groups (hematoxylin and eosin staining). Black arrows indicate wound edges; red arrows indicate tips of the healing epithelium tongue. Scale bar, 1 mm. (D), (E), (I), and (J) Quantification of the angiogenesis within the granulation tissue. After 10 and 15 days [topical groups, (D) and (E)], or 7 and 10 days [fibrin groups, (I) and (J)], wound tissues were stained for ECs (CD31+ cells) and SMCs (desmin+ cells); dual staining indicates stable vascular morphology. n ≥ 4 wounds per group per time point, mean ± SEM. For statistics: ANOVA with Bonferroni post hoc test for pairwise comparisons; *P < 0.05, **P < 0.01, ***P < 0.001. (K to M) VEGF-A/PlGF-2123-144 induces much less vascular permeability than the same dose of wild-type VEGF-A165 (500 ng). (K) The graphs show measurement of vascular permeability in the mouse ear skin. n = 8 ears, mean ± SEM. Mann-Whitney test; *P < 0.05. (L) and (M) Representative images of the mouse ear skin vasculature 25 min after VEGF-A application. Induced permeability is visualized by the red-labeled dextran leaking from the vessels. Scale bar, 0.2 mm.

We also explored whether our approach could resolve a major problem that has arisen in translating VEGF-A to clinical use. VEGF-A has been shown to rapidly induce vascular permeability, which leads to systemic hypotension and edema; this phenomenon has been the dose-limiting toxic response in peripheral and cardiovascular applications (5) and presents serious issues in regenerative medicine. Because VEGF-A/PlGF-2123-144 displays an enhanced capacity to bind endogenous ECM, we explored whether VEGF-A/PlGF-2123-144 might induce less vascular permeability. In a model of dextran extravasation from vessels in the skin of the mouse ear (23), the rate of leakage due to application of VEGF-A/PlGF-2123-144 was only 10% of that induced by application of wild-type VEGF-A165 (Fig. 2, K to M, fig. S12, and movies S1 to S4), even though they showed equivalent activities regarding phosphorylation of VEGFR-2 (fig. S3A). This effect was shown not to depend on binding to the co-receptor neuropilin-1 (figs. S2C, S3A, and S12). The engineering of VEGF-A to tightly bind the ECM appears to decouple angiogenesis from hyperpermeability, potentially solving a major problem with VEGF-A’s clinical translation.

In the context of bone repair, we tested whether PlGF-2123-144–fused BMP-2 and PDGF-BB could drive bone regeneration at low doses. Again, taking advantage of a hypothetic combinatorial effect between GFs (2, 3), we reasoned that PDGF-BB could induce progenitor cell recruitment, whereas the differentiation to bone tissue would be driven by BMP-2 (15). As a relevant model to illustrate translational potential, we used the critical-size calvarial defect in the rat (24). Because delivering micrograms of wild-type BMP-2 is usually barely sufficient to repair such calvarial defects (25), we tested a combination of BMP-2/PlGF-2123-144* and PDGF-BB/PlGF-2123-144 (200 ng of each) delivered in a fibrin matrix or delivered topically to the dura before surgical skin closure at a somewhat higher dose (1 μg of each, combined). After 4 weeks, bone healing—characterized by bone tissue deposition and coverage of the defects—was analyzed with microcomputed tomography (μCT). Delivery of wild-type GFs alone or within fibrin slightly increased bone healing when compared to the healing of defects without treatment or treated with fibrin only (Fig. 3, A to D, F, and I). In contrast, treatment with PlGF-2123-144–fused GFs led to a marked increase of bone tissue deposition as to wild-type GF (Fig. 3, A to D, G, and J), yielding coverage at 96% when delivered in fibrin and at 74% when simply administered on the dura. The improved tissue regeneration with PlGF-2123-144–fused GFs most likely involves elevated recruitment of progenitor cells, because we could detect more mesenchymal stem cells/pericytes in the defects treated with PlGF-2123-144–fused GFs than in those treated with wild-type GFs (fig. S13).

Fig. 3 Delivering PDGF-BB/PlGF-2123-144 and BMP-2/PlGF-2123-144* induces greater bone regeneration than wild-type PDGF-BB and BMP-2.

Critical-size calvarial defects (6 mm in diameter) were treated with GFs delivered topically to the dura (1 μg of each GF, combined) or in a fibrin matrix (200 ng of each GF, combined). (A to D) Four weeks after treatment, bone repair was measured by μCT as bone volume and coverage of the defect [(A) and (B) show topical groups; (C) and (D) show fibrin groups]. (E to J) Representative calvarial reconstructions. (E) Saline vehicle; (F) BMP-2 + PDGF-BB; (G) BMP-2/PlGF-2123-144* + PDGF-BB/PlGF-2123-144; (H) fibrin only; (I) fibrin with BMP-2 + PDGF-BB; (J) fibrin with BMP-2/PlGF-2123-144* + PDGF-BB/PlGF-2123-144. The defect area is shaded with a red dotted outline. n = 6 defects per condition, mean ± SEM. ANOVA with Bonferroni post hoc test for pairwise comparisons; **P < 0.01, ***P < 0.001.

In conclusion, we found that PlGF-2, through PlGF-2123-144, displays extraordinarily strong and promiscuous binding to the ECM. When this domain was conferred to other GFs, we could dramatically improve their efficacy and reduce their dosing in preclinical models of skin and bone repair. We further show that a critical limitation of VEGF-A, its induction of vascular hyperpermeability, may be ameliorated through this protein engineering concept. Because localized GF delivery and dose reduction are critical for optimal efficacy and clinical safety, this simple and broadly applicable approach to engineering second-generation ECM super-affinity GFs may be useful in a number of applications in regenerative medicine.

Supplementary Materials

Materials and Methods

Figs. S1 to S13

Table S1

References (2635)

Movies S1 to S4

References and Notes

  1. Acknowledgments: The Protein Expression, Proteomics, and Histology Core Facilities of the Ecole Polytechnique Fédérale de Lausanne and M. Pasquier, X. Quaglia, and C. Dessibourg provided technical assistance. Funding was from the European Community's Seventh Framework Programme in the project Angioscaff, the Swiss National Science Foundation, and the Fondation Bertarelli. J.A.H., M.M.M., and P.S.B. are named as inventors on a patent application filed by the Ecole Polytechnique Fédérale de Lausanne that covers the technology described in this paper.
View Abstract

Navigate This Article