Introduction
Both tissue injury and the growth of solid tumors result in inflammation and deposition of a fibrin matrix1, 2, 3. Infiltrating neutrophils and cells in the monocyte–macrophage lineage remove necrotic tissue and infectious agents. Activated and proliferating fibroblasts (protomyofibroblasts) appear in the wound 2–3 d after injury. These cells are derived from local resting cells, such as fibroblasts, that migrate into the fibrin matrix from tissues surrounding the wound4, 5. Protomyofibroblasts are induced by tensile stress and contract the wound matrix through a stress fiber–rich contractile apparatus. Tissue integrity is then restored by formation of granulation tissue and its growth into the wound void at days 4–5 (refs. 5,6). Later, in response to tensile stress, ED-A fibronectin and macrophage-derived growth factors such as transforming growth factor-
1, protomyofibroblasts differentiate into highly contractile myofibroblasts that express 
-smooth muscle actin (SMA)7. Interconnected by gap junctions, myofibroblasts secrete extracellular matrix components and at the same time contract the wound by transmitting tension across intracellular actin stress fibers connected to the extracellular matrix3, 8.
The vasculature of granulation tissue is characterized by a greater number of enlarged capillaries and venules with more blood flow, as compared to normal tissues9, 10, 11, 12, 13. High capillary density assures proper oxygen supply, and postcapillary venules facilitate attachment, rolling and transmigration of leukocytes14. This specialized vascular phenotype, which gives granulation tissue its characteristic red and granular appearance, is lost during the remodeling phase, when the newly formed matrix and vessels mature into a less vascular scar3, 15.
Formation of vessels in wound and tumor granulation tissue has traditionally been explained by angiogenic mechanisms, defined as the formation of new branchpoints in the preexisting capillary network, by outgrowth of endothelial sprouts16 or vascular buds17, 18, or by intussusceptive microvascular growth19, 20, 21, 22, 23. In this report, we show that formation of granulation tissue does not require immediate angiogenesis. Instead, activated fibroblasts or myofibroblasts contract the provisional matrix and wound margins, creating tension that mediates and directs expansion of vascularized tissues. Preexisting vessels, with functional circulation preserved, are translocated into the wound void through vessel enlargement and elongation as an integral part of the expanding granulation tissue. Neovessels are then further remodeled by angiogenic splitting and sprouting as well as regression.
Results
Matrix contraction is associated with neovascularization
We used a wound healing model based on the chick chorioallantoic membrane (CAM)24, 25 to investigate early steps of nondevelopmental tissue neovascularization (Fig. 1). A provisional matrix consisting of fibrin and rat tail collagen type I was placed on the CAM at day 12 of chick development (Fig. 1a). A nylon mesh separated the CAM from the provisional matrix, which allowed discrimination between new and preexisting vessels. In response to fibroblast growth factor-2 (FGF-2), the gel gradually contracted and was partly digested by invading cells (Fig. 1b). In parallel to these physical changes in the gel, we observed ingrowth of vascularized tissue. We visualized blood-filled structures by permanent staining of erythrocytes with 3,3'-diaminobenzidine (DAB) and H2O2 (Fig. 1c–g,i,j). Subsequent clarification with benzyl benzoate–benzyl alcohol (BBBA), which renders opaque tissues transparent, visualized the entire vasculature of ingrown tissue (Fig. 1c–g,i,j). Intracardiac ink injection distinguished vessels with functional circulation from those filled with blood but with no or low circulation (Fig. 1d,e,i and Supplementary Fig. 1 online). This double staining showed that all neovessels were functionally connected with the circulation.
Figure 1: Vascularization of a fibrin and collagen matrix implanted on the CAM through elongation of the preexisting capillary network.
(a) A fibrin and collagen gel placed on the CAM. (b) Response to FGF-2; the gel contracted and its opacity increased. Vascular ingrowth is visible in the periphery. (c) Neovessels in the ingrown tissue, stained for hemoglobin. (d,e) Vascularized implants after i.v. injection of India ink and staining for hemoglobin to discriminate functional vessels (black, sometimes appearing greenish) from blunt-ended or underperfused ones (red-brown). (f) The ingrowing vasculature was embedded in CAM tissue and appeared first at the implant periphery. Asterisk, center of implanted gel. (g) Enlargement of a growing tissue 'tissue bud'; it acquired its shape while growing through the mesh. (h,i) Vessels elongated within the growing tissue and formed functionally perfused macrovascular loops supporting a capillary network at the front of the ingrowing tissue. Blood perfusion in the live embryo, h; ink injection to mark functional circulation, i. Ingrowing neovessels formed a clear interface to the gel. (j) Staining of the gel with ink. Vessels did not enter the gel but were contained within a new stain-free matrix that replaced the implanted gel (n = 12). Hemoglobin stained with DAB in c–g,i,j. Vasculature was perfused with ink in d,e,i. In c–e,g,i,j the tissue and matrix were rendered transparent by BBBA treatment. Results in b–e,h,i are based on at least 20 observations. Scale bars, 1 mm.
Full size image (79 KB)Vessels grow as an integral part of the expanding tissue
After implantation of the provisional matrix, the lumens in the dense capillary network underlying the CAM implant first became enlarged, and then the entire capillary network moved toward the implant (Supplementary Fig. 2 online). The first neovessels that passed the nylon mesh were localized primarily in the outer parts of the gel (Fig. 1b–d,f,h,i). Gels were occasionally digested (11 observations), revealing that vessels were embedded in connective CAM tissue (bud-shaped structures in Fig. 1f,g). Later, the tissue with its encapsulated vessels became elongated and formed tissue tubes containing functional macrovascular loops that supported a capillary network located at the leading front of the ingrowing tissue (Fig. 1h,i). An epithelial cell layer initially surrounded the ingrowing tissue but later split up into single cells that were found in the gel matrix (Supplementary Fig. 3 online). Lumen-containing neovessels were supported by a basal lamina and smooth muscle cells, which are characteristic elements of mature vessels (Fig. 2 and Supplementary Fig. 3). The structure of the ingrowing tissue reflected the stratified vascular organization of the original CAM, with a superficial layer of fine capillaries connected with a deeper layer of macrovessels; these two layers were enclosed between two layers of chorionic and allantoic epithelium26 (Fig. 1h,i, Fig. 2c,d and Supplementary Fig. 3). Proliferating endothelial and mural cells were found in capillaries as well as in larger vessels (Supplementary Fig. 4 online). A clear interface between the implanted matrix and the neovascular tissue was identifiable at all stages of vascular ingrowth (Fig. 1i,j and Supplementary Fig. 2). Preexisting vessels hence expanded within the growing tissue and did not enter the provisional matrix as independent vessels. These observations indicate that neovascularization proceeded by expansion of the original vascular network enclosed in CAM tissue, which replaced the provisional fibrin matrix.
Figure 2: Neovascularization is preceded by ingrowth of proto- and myofibroblasts.
(a) H&E and (b) SMA staining of a cross-section parallel to the nylon mesh of an implanted gel on the CAM (horizontal section, plane shown in Supplementary Fig. 3b). Elongated eosinophilic cells formed sprouts parallel to the gel surface in peripheral parts of the gel (arrowhead). Rounded, nonsprouting cells were found throughout the gel. Vessels appeared in the outer parts of the implant (arrow). Most sprouting cells were SMA+ (b). (c–h) Staining of SMA (c,e,g; with red-brown DAB counterstaining of erythrocytes) and collagen (blue, d,f,h; with red Masson trichrome counterstaining of erythrocytes) in normal CAM (c,d; transverse section, plane in Supplementary Fig. 3a), early stages of ingrowing vascular tissue buds (e,f; horizontal section, plane in Supplementary Fig. 3b) and late stages of ingrown tissue (g,h; horizontal section, plane in Supplementary Fig. 3c). In the normal CAM (c) and early tissue buds (e), SMA+ cells were present only around vessels (arrows). The mature ingrown tissue (g) was characterized by SMA+ vascular mural cells (arrows) and elongated interconnected nonvascular cells (arrowheads) that in the outer parts of the implant were aligned parallel to the gel border (b). Except for perivascular areas, collagen was evenly distributed in the normal CAM (d). Mature ingrown tissue (h) had collagen content comparable to that of normal CAM (d). In contrast, early ingrowing tissue buds had low interstitial collagen content (f). Staining in e,f was repeated in 3 implants, a–d,g,h in 12. Scale bars, 1 mm (a,b), 100 
m (c–h).
Tension induces and directs vascularization
Before the ingrowth of neovascular tissue, the implanted matrix was populated by SMA-positive myofibroblasts (Fig. 2b,g), SMA-negative protomyofibroblasts (Fig. 2a and Supplementary Fig. 5 online) and CD45-positive leukocytes (Supplementary Fig. 3). Fusion of the gel with the underlying CAM 2–3 d after implantation was presumably due to remodeling of the tissue when these cells invaded the gel.
In a concentrically contracting, cylindrical gel with its base fixed to the underlying CAM, circumferential tension is translated into a force that is perpendicular to the CAM and is highest at the periphery. The protomyofibroblastic phenotype is induced by unshielded tension5, likely explaining why myofibroblasts appeared preferentially in the periphery of the gel implant and later invaded central parts of the implant (Fig. 2a,b). Ingrowth of vascular tissue followed the same pattern and could be observed from day 4 (Figs. 1b–d,f,h,i and 2a,b). Neovessels in the granulation tissue were positive for the endothelial marker von Willebrand factor, whereas no von Willebrand–positive cells were detected in the implanted matrix (Supplementary Fig. 3). The first vascularized tissue structures that passed through the mesh had a low content of cells and collagen (Fig. 2e,f), as would be expected if external tension was responsible for tissue expansion. The new tissue was later remodeled and enriched in cells (mainly proto- and myofibroblasts) and collagen (Fig. 2g,h).
Tissue contraction depends on both the imposed tension and compliance of the matrix to deformation. Gel contraction was a prerequisite for neovascularization, as replacement of acid-solubilized rat tail collagen type I with pepsin-solubilized collagen type I (Vitrogen) reduced implant contraction (data not shown) and impaired neovascular ingrowth (Fig. 3a). The lower cell-driven contractibility of Vitrogen is due to pepsin-mediated cleavage, which results in removal of both collagen telopeptides and cryptic binding sites for integrins27, 28. Removal of these cellular binding sites inhibits development of cell-mediated tension because cells are unable to translocate matrix molecules. We confirmed the impaired contractibility of Vitrogen in vitro using gels containing chicken-derived myofibroblasts (Fig. 3b). Lack of contraction in the Vitrogen-based gel was paralleled by a failure of cells to adopt a concentric orientation parallel to the gel surface, which is necessary to reduce the surface area by contraction29 (Fig. 3c). Vitrogen did not inhibit myofibroblast invasion or migration in vivo, as SMA-positive cells were found in the gel. These cells probably acquired their myofibroblastic phenotype as they migrated and differentiated in the activated environment of the underlying CAM but failed to spread when they entered the implanted matrix and instead appeared as rounded cells (Fig. 3d).
Figure 3: Matrix contraction is a prerequisite for vascular ingrowth on the CAM.
(a) Inhibition of FGF-2-induced matrix neovascularization by use of Vitrogen instead of rat tail collagen (*P < 0.0001). (b,c) In vitro cultures of myofibroblasts mixed with rat tail collagen were contracted (b) by cells that grew concentrically (c, left). Vitrogen cultures showed minimal contraction (b), with cells growing radially (c, right). Control, initial size of the droplet containing cells and collagen. Representative of n = 35–38 gels for each condition. (d) Staining for SMA. Collagen plus fibrin gels placed on the CAM were populated by SMA+ myofibroblasts that formed branched structures in rat tail collagen plus fibrin gels (left; representative of n = 12 experiments) but appeared as rounded cells in Vitrogen plus fibrin gels (right; representative of n = 8 experiments). (e) Inhibition of FGF-2-induced matrix neovascularization in gels containing glass fibers (*P < 0.0001). (f) The presence of glass fibers (yellow) in rat tail collagen matrices inhibited myofibroblast-mediated in vitro contraction (f, right, n = 54), whereas control collagen cultures were contracted (f, left, n = 45). (g) Example of the effect of glass fibers on implant vascularization. (h–j) Cross-linking of collagen blocked vascularization (h; P < 0.001) but allowed ingrowth and spreading of cells (j; H&E). Photographs of the CAM from below (i, top) and above (i, bottom) the mesh showed that the underlying CAM was not affected by crosslinking agents. Samples in g,i were stained with DAB and clarified with BBBA. Scale bars, 1 mm (b,c,f,g,i), 100 m (d,j).
We further investigated the importance of contraction for neovascularization by using contraction-resistant gels. When biochemically inert borosilicate glass fibers (10–100 m long) are mixed with the rat tail collagen matrix, hydrogen bonds form between surface silanol groups in the glass fibers and amino acids. These bonds counteract cell-mediated local gel contraction but should allow ingrowth of cells. Use of such gels resulted in complete inhibition of proto- and myofibroblast-mediated matrix contraction in vitro and inhibition of vascularization by 45% in vivo when gels were placed on the CAM (Fig. 3e–g). We also cross-linked collagen matrices with 1-ethyl-3-(3-dimethylaminopropyl) carbodiimide, which inhibits cell-mediated collagen gel contraction by increasing resistance of the matrix to applied tension30, before application onto the CAM. Cross-linking completely abolished implant contraction (data not shown) as well as vascular ingrowth (Fig. 3h,i) but allowed invasion and spreading of cells within the matrix (Fig. 3j). The lack of vascularization was not due to an inability of endothelial cells, vascular smooth muscle cells or myofibroblasts to adhere to these engineered matrices (Supplementary Fig. 6 online).
Endothelial cells themselves have been suggested to exert and follow biomechanical cues in directional sprouting31. However, when endothelial cells or myofibroblasts were embedded in rat tail collagen I for 40 d, only the myofibroblasts were able to develop tensional forces and contract the gels (Supplementary Fig. 6). These experiments showed that endothelial cells were not responsible for gel contraction and development of tensile forces. Finally, proto- and myofibroblasts might create a chemical environment that induces and directs neovascularization without a need for contraction. We ruled out this possibility by replacing the fibrin and collagen gel with polyvinyl alcohol sponges that cannot be contracted by cells. When sponges were preseeded with purified myofibroblasts, no vascularization was observed (Supplementary Fig. 7 online).
Time-lapse recording of neovascularization
We studied expansion of the neovasculature after the initial recruitment of preexisting vascular loops with 26-h time-lapse recordings (Fig. 4). Neovascularization proceeded by continuous movement of the entire vasculature toward the contracting implant. During this translocation, individual vessels were remodeled by enlargement, elongation, pruning and thinning. In addition, we observed angiogenesis by vessel splitting and loop formation (intussusception). These observations indicate that neovascularization is accomplished by nonangiogenic translocation and remodeling of preexisting vessels while angiogenesis serves to fine-tune the process.
Figure 4: Time-lapse recordings of vessel remodeling during implant vascularization on the CAM.
Neovessels that had entered the outer parts of the implant were observed by brightfield microscopy for 26 h on the living embryo beginning 6 d after implantation of an FGF-2 containing matrix. The vasculature was continuously translocated (in relation to the mesh, *) toward the contracting implant located outside the frame of the picture in the upper left corner. L1, L2 and L3, three loops that underwent substantial remodeling. The left arm of the vascular loop L1 increased and then gradually reduced its lumen diameter. The right arm of the same loop reduced its diameter and was pruned away (double arrow). During remodeling the diameters and configuration of larger vessels were normalized. Numbers 1–4, transient intravascular loops (possibly formed through intussusception) within the main loops. Scale bar, 300 M.
Vascularization of mouse cornea by functional vessels
To validate the results obtained in the CAM model, we studied the course of neovascularization associated with the healing mouse cornea32. Insertion of a silk suture through the center of the cornea caused a wound with a healing reaction. This was manifested by edema, thickening and contraction of cornea, as well as by vessel outgrowth from the limbal capillary network toward the site of injury (Fig. 5 and Supplementary Fig. 1d). After injury, vessels in the limbal microvasculature33 enlarged and were translocated as vascular loops toward the corneal wound, leaving the vasculature in the limbus depleted of capillaries (Fig. 5c–e). Intracardiac injection of ink, combined with DAB and H2O2 staining of blood-filled structures, showed that neovessels were functionally connected to the limbal circulation (Fig. 5a–e). In addition, short protrusions filled with ink were located on vascular loops extending toward the wound (Fig. 5d). Sprouting SMA-positive myofibroblasts were found in and around the wound while granulation tissue originating from the limbus grew in the outer parts of the cornea stroma (Supplementary Figs. 4 and 8 online). Neovessels were supported by a basal lamina and were covered with SMA-positive mural cells (Supplementary Fig. 8). These findings support the hypothesis that tissue tension mediates and directs rapid expansion of the preexisting vasculature as loops with preserved functional circulation.
Figure 5: Neovascularization of the wounded mouse cornea by recruitment of functional microvascular loops from the limbal capillary network.
(a,b) A wound was inflicted by inserting a silk suture through the center of the cornea (arrow, b), which resulted in vessel outgrowth from the limbal capillary network toward the site of injury. Arrowhead, iris. (c) The limbal microvasculature of a normal eye. (d) Vascular loops 3 d after wounding, with short perfused vascular buds (arrowheads). (e) Depletion of capillaries in the limbus 9 d after wounding. In a–e, mice were injected with ink and corneas clarified with BBBA. (f,g) Treatment of mice with VEGFR2-blocking antibody (DC101, g) inhibited formation of vascular outgrowths (arrowheads, f) and increased pericyte coverage of vessels compared to mice treated with control (Ctrl) antibody (f). Corneas were stained for CD31 (white) and NG2 (green) 3 d after suturing. (h) Treatment with VEGFR2-blocking antibody reduced neovascularization at day 6 compared to control antibody but not at day 3 (n = 4 corneas per group). (i) Blockade of VEGFR2 reduced proliferation of vascular cells at both day 3 and day 6 (n = 8–11 corneas per group). Boxes, s.e.m.; whiskers, confidence intervals (s.e.m. 
1.96). *P < 0.01; NS, not significant. Scale bars, 0.5 mm (a–e), 100 m (f,g).
We next analyzed in detail sequential events during neovascularization by whole-mount stainings of corneas 3, 4, 5 and 6 d after suture placement, which is a period of linear increase of total vascularized area (r = 0.632; correlation P< 0.0001; Supplementary Fig. 9 online). To visualize functional vessels, the mice were perfused with fluorescein-labeled lectin; their corneas were then removed and stained as whole mount preparations with markers for endothelial cells (CD31), pericytes (NG2) and lymphatic cells (LYVE-1). Combined lectin perfusion and endothelial staining showed that most of the neovessels were functionally perfused during the 6 d of neovascularization (Fig. 6a–d). Expanding vascular loops at early stages (day 3) were enlarged, with pericytes covering the part of the vessel that was oriented toward the limbus (Fig. 6b). The pericyte-free part of the vessel was irregularly shaped, and most of the vascular outgrowths (unperfused sprouts and perfused blunt-ended buds) originated from this location. Sprouts did not contain nuclei and hence represented extensions of cells (Fig. 6c). A linear increase in total vascularized area during the 6 d of neovascularization correlated with a decreased number of vascular outgrowths and an increased pericyte coverage of vessels (Fig. 6 and Supplementary Fig. 9). The area covered by vascular outgrowths represented 4.8% 
0.43% of the total vascularized area at day 3, which decreased to 3.2% 0.35%, 2.5% 0.25% and 1.6% 0.26% at days 4, 5 and 6, respectively (means s.e.m.; analysis of variance (ANOVA), P < 0.001 for decrease in vascular sprouts and buds during 4 d of vascularization).
Figure 6: Characterization of neovascular growth during wound healing of the injured mouse cornea using whole mount staining.
(a) Staining of endothelial cells (CD31) and pericytes (NG2, arrows) in normal limbal vessels. CD31 also stained blunt-ended lymphatic vessels (arrowhead) that were LYVE-1+ (Supplementary Fig. 9). (b,c) Limbal vessels 3 d after cornea suturing perfused with FITC-labeled Bandeiraea simplicifolia lectin and stained with CD31, NG2 and Hoechst 33342. Pericytes (red) appear yellow as they are superimposed on the lectin stain (green). Sprouts (arrows, c) did not contain nuclei (white). (d) Staining of lectin-perfused vessels with CD31 6 d after cornea suturing. Most vessels were functionally perfused. The normal limbal capillary network was replaced by enlarged feeder vessels (arrows). Unperfused structures in the center (arrowhead) are lymphatic vessels. (e) NG2+ pericytes covered most of CD31+ neovessels 6 d after cornea injury. Arrowhead, CD31+ lymphatic vessel. (f) LYVE-1 and CD31 staining at day 6. Lymphatic vessels grew as thick, blunt-ended structures ahead of blood vessels. Results in a–f based on 3 or 4 corneas. Scale bars, 120 m (a,b,d–f), 50 m (c). (g,h) Correlations between vascular ingrowth (total vascularized area, TVA), percentage pericyte coverage (PC) and number of vascular outgrowths (NO, including both unperfused sprouts and perfused buds). TVA correlated negatively with NO (r = -0.491; correlation P < 0.0003) but positively with PC (r = 0.785; P < 0.0001). PC correlated negatively with NO (r = -0.495; P < 0.0003). Dotted lines, 95% prediction bands. (i) Top row, sequential events during vascularization of an implant on the CAM. Inflammatory cells and proto- and myofibroblasts migrate into the provisional fibrin and collagen matrix (left). Proto- and myofibroblasts then contract the gel, which creates tension (arrows), highest in the periphery (middle). The CAM tissue with its vasculature thus expands in the direction of the contracting implant (right). Vessels do not grow as independent structures outside their original tissue. Instead, they expand within the growing CAM tissue as vascular loops with functional circulation. New branchpoints might be formed through intussusceptive vessel splitting or sprouting. The bottom row applies the proposed model to dermal wound healing.
To determine the importance of endothelial cell proliferation and sprouting during neovascularization, we injected mice with corneal injury with a neutralizing antibody (DC101) to vascular endothelial growth factor receptor-2 (VEGFR2) before injury and every second day. There was no difference in the size of the vascularized area at day 3 in DC101-treated mice compared to control IgG-treated mice even though the number of outgrowing vascular buds and sprouts was 70% lower and the proliferation of vascular cells 58% less in DC101-treated mice (Fig. 5f–i and Supplementary Figs. 10 and 11 online). Pericyte coverage of vascular loops at day 3 was 90% greater in mice receiving DC101 (Supplementary Figs. 10 and 11, P < 0.01). At day 6, there was still a 64% inhibition of proliferation in DC101-treated mice, which was paralleled by a 20% inhibition of neovascularization (Fig. 5h,i). No proliferating vascular cells were detected in the limbal vasculature of uninjured eyes (data not shown). These results indicate that proliferation is not critical for early stages of neovascularization. After wounding, we observed outgrowth of LYVE-1 positive lymphatic vessels from the limbus into the cornea (Fig. 6f). Treatment with DC101 reduced lymphatic outgrowth at day 6 by 21%, whereas there was no difference at day 3 (Supplementary Figs. 10 and 11). The finding that both lymphatic and vascular outgrowth occurred in parallel and were similarly affected by DC101 treatment suggests that the two are regulated by a common mechanism.
Finally, we investigated whether it is possible for sprouting cells to migrate a distance comparable to that covered during outgrowth of vessels in the cornea by using an in vitro, three-dimensional model of cell sprouting31. Human umbilical vein endothelial cells and human microvascular endothelial cells formed shorter sprouts than myofibroblasts in vitro (Supplementary Fig. 6). Myofibroblasts sprouted a distance of 203 58 m during 6 d in vitro (n = 10 spheroids). This distance was significantly shorter than the distance of vessel outgrowth in the cornea, which was 594 118 m (n = 14 corneas; P < 0.0001). These experiment indicated that sprouting cannot be the dominant mechanism for neovessel formation in granulation tissue (Supplementary Fig. 6).
Discussion
Our results suggest that translocation of the preexisting vasculature is responsible for the initial rapid formation of functional vessels in granulation tissue. Vessels expand as an integral part of the growing granulation tissue in response to mechanical forces. This revitalizes and supports the concept of biomechanical regulation of tissue vascularization34, 35, 36 that was originally proposed by Thoma more than 100 years ago (discussed in ref. 37). We now show that these tensional forces are generated by proto- and myofibroblast-mediated contraction of the provisional matrix and the surrounding tissue, which physically guides and mediates neovascularization.
The concept that tissue remodeling is accompanied by angiogenesis through endothelial16 and vascular17 sprouting is widely accepted. These models are derived mainly from investigations of angiogenesis during embryonic development and in vitro studies. However, embryonic and adult neovascularization cannot be directly compared because these processes are driven by different mechanisms38. Developmental angiogenesis is a genetically controlled process in which vessels form de novo together with the surrounding tissue in a manner that is spatially and temporally reproducible. In contrast, neovascularization during tissue healing is stimulated by unpredictable wound-induced liberation of local factors and occurs in already differentiated tissues. Also, current understanding of the cellular processes underlying capillary growth—including formation of a sprout lumen, connection to another sprout or vessel and incorporation or differentiation into functional capillaries, veins and arteries—during postdevelopmental neovascularization, in vivo remains incomplete39.
Endothelial sprouting depends mainly on the relatively slow process of stalk cell proliferation and migration40 and thus can hardly explain the rapid appearance of fully functional vessels in granulation tissue as early as 24 h after wounding6. Also, the presence of an interface between the fibrin clot and the invading granulation tissue, which is observed in healing wounds6, is difficult to explain based on models that depend on endothelial sprouting. Finally, damage of vessels adjacent to a wound inhibits revascularization9 instead of facilitating the process, as would be expected if a natural tissue barrier for the sprouting endothelium was removed. A different type of sprouting, the outgrowth of lumen-containing dead-ended vascular sprouts (buds), has been suggested to occur during tumor vascularization17. However, no mechanism has been defined to explain how these complex multicellular structures are translocated into the avascular zone, connect and later differentiate into functional circulation. Finally, intussusceptive angiogenesis permits formation of instantly functional neovessels in granulation tissue at relatively low energy costs19, 40, but it is not known how intussusceptively formed vessels are translocated from the surrounding tissue into the avascular wound.
In this report, we propose a mechanism of neovascularization that explains the early appearance of functional vessels in granulation tissue (Fig. 6i). Enlarged preexisting capillaries moved as vascular loops with continuous functional circulation. Translocation of functional vascular loops at the leading edge of ingrowing granulation tissue has previously been described for tumor-induced neovascularization of mouse cornea41 and in healing human wounds42. In our CAM model, neovessels were not found in the implanted matrix, which implies that they did not enter the gel as independent entities. Instead, they were contained within the expanding granulation tissue that eventually replaced the provisional matrix. As a result, there was a clear demarcation between the ingrowing granulation tissue and the implanted gel, analogous to the interface between the growing granulation tissue and the fibrin clot in healing wounds42, 43.
Even though proto- and myofibroblast mediated wound contraction is a well-known process5, 44, it has not been directly linked to the synchronous ingrowth of vessels in granulation tissue. In both of the models that we studied, ingrowth of vascular tissue was associated with contraction. This contraction was likely caused by proto- and myofibroblasts that populated the matrix implanted on the CAM and the area surrounding corneal wounds. Contraction was a prerequisite for neovascularization, as vascular ingrowth correlated with the contractibility of different collagen preparations. Prevention of gel contraction by reinforcing matrix rigidity with glass fibers or increasing matrix stiffness by chemical cross-linking reduced neovascularization but allowed ingrowth and spreading of cells. Factors released by invading cells were not sufficient to induce vessel ingrowth in the absence of contraction, as there was no vascularization of noncontractile sponges seeded with myofibroblasts. Although growth factors are necessary to induce neovascularization by activating, for example, cell proliferation and migration, the spatial pattern of vessel ingrowth that we observed suggests that neovascularization is dependent on mechanical forces rather than gradients of growth factors. Hence, the capillary network expanded in the superficial layer of the cornea stroma, where the highest torque (angular force) is anticipated, and not along the shortest route through deeper corneal parts, as would be expected if chemoattraction of vessels predominated. Similarly, neovessels first entered peripheral parts of the CAM implant, where tension is expected to be highest.
Although the initial steps of wound revascularization can be fully explained by expansion of the preexisting vascular network, we emphasize that this model does not eliminate the necessity of angiogenesis during later stages of tissue repair to increase the number of vessel branchpoints. In support of this role for angiogenesis, time-lapse recordings of late remodeling events during neovascularization of CAM implants revealed formation of intravascular loops, indicative of angiogenic intussusceptive vessel growth. These observations in real time also revealed dynamic remodeling, with enlargement, elongation and pruning of individual vessels while the whole neovascular bed moved in the direction of the contracting implant. A similar pattern of events was found in the healing cornea in which neovascularization was observed from day 3 to day 6; in this system, enlarged and perfused vascular loops translocated toward the contracting wound. In addition, we identified short perfused vascular buds and unperfused endothelial sprouts on growing vascular loops. Both sprouts and buds might eventually connect and form functional vessels through angiogenesis, although we could not identify intermediate structures (small loops, comparable in size to the short sprouts or buds) that would be the result of such fusions.
In the healing cornea model, we observed a linear increase in total vascularized area, but the number of sprouts and vascular buds decreased. Endothelial sprouts were derived from the immature pericyte-free part of expanding vascular loops, and the percentage of pericyte-covered mature vessels increased with time. Hence, sprouting was not a main source of neovessels as a decreasing number of immature vessels did not affect vessel growth. Vascular endothelial growth factor-A has been postulated to control angiogenic sprouting by activating VEGFR2 expressed on endothelial cells45, and we therefore inhibited VEGFR2 with a blocking antibody. Similarly to results of previous studies, inhibition of VEGFR2 reduced formation of sprouts46, 47, increased pericyte coverage48 and hence normalized vessels49. Inhibition of VEGFR2 did not influence vessel growth during the first 3 d after injury but partially inhibited vascularization at day 6. Taken together, these results suggest that early steps of vascularization are independent of sprouting.
Endothelial cell proliferation is commonly assumed to be an indicator of sprouting angiogenesis50. However, because endothelial mitogenesis accompanies any vascular expansion process, it cannot distinguish between sprouting angiogenesis and processes of vessel elongation or enlargement. In agreement with previous studies of corneal neovascularization51, we found proliferating endothelial and mural cells in capillaries as well as in larger neovessels. This arteriogenesis-like process is an efficient mechanism that could explain the rapid appearance of functional vessels during tissue granulation, as only one cell division by every cell of the vessel is required to double vessel length. Our model also implies that redistribution of existing endothelial and mural cells from the limbal vessels of the cornea could initially allow elongation of pre-existing vessels without cellular proliferation, as has been previously suggested52. In support of this hypothesis, we found that inhibition of VEGFR2 signaling markedly reduced endothelial cell proliferation at day 3 without affecting the degree of neovascularization. Neovascularization over longer distances would likely require proliferation of vascular cells, as we found that decreased cell proliferation was associated with reduced vascularization at day 6.
Normal tissue contains a dense network of capillaries that may be quickly recruited through capillary enlargement and elongation to form the vasculature of the healing wound. We propose that mechanical tension is both necessary and sufficient to mediate and direct expansion of vascular tissue. At early stages of tissue healing, nonangiogenic processes of vessel remodeling are sufficient for the formation of granulation tissue vasculature, whereas angiogenesis by intussusceptive splitting and sprouting serves to fine-tune neovascularization by increasing the number of branch points during later stages of tissue repair. The degree to which angiogenic splitting and sprouting, as well as lymphangiogenesis, are also regulated by biomechanical mechanisms awaits further studies. Biomechanically driven neovascularization might explain phenomena ranging from acceleration of wound healing by vacuum-assisted wound closure devices53 to complex mechanisms of tissue regeneration54. We show that proto- and myofibroblast mediated tension may represent an important target for development of new treatment strategies for diseases that are characterized by expansion of vascularized tissue, such as tumor growth, proliferative retinopathies and rheumatoid arthritis.
Note: Supplementary information is available on the Nature Medicine website.
