Low-flow intussusception and metastable VEGFR2 signaling launch angiogenesis in ischemic muscle.

Efforts to promote sprouting angiogenesis in skeletal muscles of individuals with peripheral artery disease have not been clinically successful. We discovered that, contrary to the prevailing view, angiogenesis following ischemic muscle injury in mice was not driven by endothelial sprouting. Instead, real-time imaging revealed the emergence of wide-caliber, primordial conduits with ultralow flow that rapidly transformed into a hierarchical neocirculation by transluminal bridging and intussusception. This process was accelerated by inhibiting vascular endothelial growth factor receptor-2 (VEGFR2). We probed this response by developing the first live-cell model of transluminal endothelial bridging using microfluidics. Endothelial cells subjected to ultralow shear stress could reposition inside the flowing lumen as pillars. Moreover, the low-flow lumen proved to be a privileged location for endothelial cells with reduced VEGFR2 signaling capacity, as VEGFR2 mechanosignals were boosted. These findings redefine regenerative angiogenesis in muscle as an intussusceptive process and uncover a basis for its launch.

INTRODUCTION

Lower-limb skeletal muscles can be profoundly damaged by peripheral artery disease, and amputation may be required. Accordingly, therapeutic innovations are actively sought. Promoting the development of new blood vessels has garnered particular interest, as this has the potential to mitigate muscle damage and support myofiber regeneration (). Sprouting angiogenesis, wherein endothelial cells sprout off a native vessel based on vascular endothelial growth factor (VEGF)–vascular endothelial growth factor receptor-2 (VEGFR2) signaling, has been considered particularly critical to skeletal muscle neovascularization (–), and therapies have been designed on this premise (, ). Unfortunately, however, proangiogenesis strategies have not benefitted individuals with peripheral artery disease (, ). Gaps in our understanding of regenerative angiogenesis in skeletal muscle may account for this.

Although sprouting angiogenesis has been considered central to regenerating the muscle microvasculature, it is noteworthy that direct evidence for endothelial sprouting in ischemic skeletal muscle is limited (–). Another mode of neovascularization, but one that is far less understood than sprouting angiogenesis, is intussusceptive angiogenesis. Vascular intussusception entails the internal division, or splitting, of a vessel into two daughter vessels (). There is evidence that capillaries in skeletal muscle can undergo a duplication response to vasodilators and physiologic loading (, ). However, a role for intussusceptive angiogenesis in rebuilding a vessel network after ischemic muscle injury is unknown.

Vessel splitting begins with the formation of a peculiar endothelial-based projection that crosses from one side of the lumen to the other, a structure referred to as a pillar (). Pillar formation has been mostly studied in the lung (–) and the extraembryonic circulation (–) but has been observed in muscle following forced expression of VEGF (, ). Unlike endothelial cell sprouting, little is understood about what controls the perplexing entrance of endothelial cells into a flowing blood vessel. The technical challenges in tracking individual endothelial cell events inside the lumen, the transient nature of these curious structures, and the lack of in vitro models are such that intraluminal pillar formation is largely an enigmatic process.

Herein, we report that angiogenesis after ischemic skeletal muscle injury bears little resemblance to the classical paradigms of endothelial sprouting. Instead, using real-time imaging, we found that the process entails the formation of wide-caliber, primordial conduits receiving ultralow flow, which rapidly transform into a hierarchical network by dynamic pillarization and intussusception. To dissect the driving conditions for this intussusceptive process, we sought out local VEGFR2 activation signals in vivo and used microfluidics to develop the first live-cell model of intraluminal endothelial pillar formation. The findings reframe angiogenesis in ischemic muscle and uncover a mechanosignaling-based interplay that launches the process.

RESULTS

Lost capillaries in ischemic muscle are replaced by dilated neoconduits

To delineate the early course of microvascular regeneration in ischemic skeletal muscle, we excised the femoral artery in C57BL/6J mice and histologically evaluated the extensor digitorum longus (EDL) muscle over 7 days. One day after vascular insult, the EDL muscle was widely infarcted, with pale staining myofibers and vanished nuclei. This appearance persisted until day 4, at which time there was inflammatory cell infiltration and outside-in myofiber destruction (Fig. 1A). On day 5, inflammatory cells were still abundant but now associated with clearing of necrotic muscle. Also, myocyte regeneration was evident, indicated by enlarging satellite cells with their characteristic central and bulky nuclei (Fig. 1A, arrows, day 5). Over the next 2 days, the damaged tissue progressively repopulated with multinucleated myofibers (Fig. 1A).

Fig. 1.

Vascular regeneration in ischemic muscle proceeds by forming primordial vessels that transform into a capillary network.

(A) Hematoxylin and eosin–stained sections of mouse EDL muscle in native state and after femoral artery excision. Infarction is evident on day 4, with pale, anuclear myocytes and surrounding inflammatory cell infiltration. At day 5, there is extensive inflammatory cell infiltration, clearing of necrotic muscle, and satellite cells with central nuclei (arrows). On day 7, repopulation of myofibers with central nuclei is evident. (B) Fluorescence micrographs of EDL muscle immunostained for CD31 (green), with nuclei stained with DAPI (4′,6-diamidino-2-phenylindole) and pseudocolored red. Myofibers are depicted by the dashed lines. Capillaries are evident between myofibers in the native tissue but not evident 4 days after injury. On day 5, markedly dilated, thin-walled CD31+ vessels are evident (arrows). On day 7, these evolved to capillary-like microvessels between newly formed, multinucleated muscle fibers (arrows). (C) Graph [median, interquartile range (IQR), and 10th to 90th percentile] showing microvessel lumen diameters (total number of distinct microvessels quantified is depicted and from five to seven mice at each time point). *P < 0.0001. (D) Confocal micrographs of projected optical sections (total 20-μm thickness) of EDL muscle 5 days after ischemic injury, immunolabeled for CD31 (green) and TO-PRO-3 (blue), showing a large caliber primordial vessel (left). Boxed area is enlarged on the right, illustrating densely packed and circumferentially adjacent endothelial cell nuclei (dashed lines). (E) Micrograph of projected confocal optical sections (total 30-μm thickness) of day 5 postischemic muscle immunolabeled for CD31 (green), NG2 (red), and TO-PRO-3 (blue), showing a primordial vessel extensively covered by NG2-positive pericytes.

Immunostaining for CD31 revealed that the EDL microvessels were obliterated by the ischemic insult, with no capillaries, small arterioles, or venules detected 3 and 4 days after the ischemic insult (Fig. 1B). However, on day 5, unusual wide-caliber vascular structures emerged. These were thin-walled channels with diameters up to 80 μm. The median diameter was 10.2 μm, significantly greater than that of microvessels in uninjured EDL muscle (P < 0.0001; Fig. 1C). Reconstructed confocal slices from thick muscle preparations revealed these neochannels to be lined by densely packed endothelial cells, apposed side by side and end to end (Fig. 1D). Double-immunolabeling revealed no evidence for smooth muscle (SM)–α–actin–containing cells, but, instead, the neochannels were extensively covered by NG2-positive pericytes (Fig. 1E). These dilated thin-walled vessels were transient and disappeared over the next 48 hours. By day 7, the vasculature consisted of small microvessels with a median diameter of 4.4 μm, which were located at the interface of regenerated myofibers, consistent with capillaries (Fig. 1, B and C).

Thus, ischemic injury induced a robust program of myofiber regeneration and an equally robust program of microvascular regeneration. Intriguingly, the angiogenesis program began with the formation of wide-caliber, thin-walled neochannels of little resemblance to normal microvessels.

Intravital imaging reveals ultralow flow primordial vessels that rapidly transform into a hierarchical microvasculature

To ascertain how the initial, wide-caliber neovessels were integrated into a flow network, we undertook real-time, intravital red blood cell (RBC) imaging. Video images and RBC transit maps revealed a complete absence of RBC flow across the interrogated EDL muscle regions during the first 4 days following femoral artery excision (Fig. 2, A to C). The first flow-receiving channels emerged on day 5 (Fig. 2, D and E). These were wide-caliber conduits, consistent with the channels observed histologically, with a median live-lumen diameter of 23.7 μm (range, 6.4 to 75.0 μm). They coursed as long and relatively straight channels with occasional cross-bridges (3.8 cross-bridges/mm vessel length).

Fig. 2.

Architectural transformation of mother vessels into a hierarchical microvasculature.

(A to G) RBC transit maps (A, C, D, F, and G) and intravital still frames (B and E) obtained from intravital microscopy assessments of EDL muscle of mice injected with FITC (fluorescein isothiocyanate)–dextran. The RBC transit maps denote the perfused microvasculature. In the native network, parallel capillaries coursing between individual muscle fibers are evident (A). RBCs (arrow), seen in relief against bright fluorescent blood plasma, are in single file (B). See also movie S1. Four days after femoral artery excision, there is no RBC transit, indicated by the complete absence of vascular signals (C). At 5 days, flowing wide-caliber neovessels are evident (arrows, D). These neovessels contain densely packed RBCs (E). See also movie S2. On day 6 (F), a more widely distributed vasculature with smaller-caliber microvessels is evident. Capillaries (yellow arrow), arteriolar inputs (red arrow), and venular outputs (blue arrow) are evident on day 7 (G). (H) Box and whisker plot of microvessel lumen diameter (n = 164, 377, 572, and 622 for native muscle and days 5, 6, and 7 after injury, all from three to five EDL muscles per mice per time point). (I) Graph depicting the microvessel density of the native and regenerated EDL muscles (n = 28, 11, 7, and 36, distinct 0.63-mm2 muscle areas, from three to five EDL muscles per mice per time point). (J) Graph depicting branch density of the native and regenerated microvascular networks (n = 28, 11, 7, and 36, 0.63-mm2 muscle areas from three to five EDL muscles per mice at each time point). (K) RBC velocities in individual microvessels, averaged over 15 s of imaging, within native and regenerating EDL muscles. Velocities were obtained from 115, 82, 18, and 93 microvessels, within three to five EDL muscles per mice at each time point. Data in (H) and (K) are median, IQR, and 10th to 90th percentile; data in (I) and (J) are means and SE.

RBC velocity in these primary neoconduits was notably low, with a median velocity of 88 μm/s, 17% of that in the native EDL capillaries (P < 0.0001). In contrast with the single-file red cell transit in native capillaries (Fig. 2B and movie S1), neovessel hematocrit was high, with up to seven radially adjacent RBCs in the channel lumen (Fig. 2E and movie S2). Median wall shear rate was 25.4 s−1 [interquartile range (IQR), 14.4 to 65.0 s−1, n = 42 primordial vessels of diameter >20 μm], corresponding to a median wall shear stress of only 0.93 dyne/cm2 (IQR, 0.54 to 2.50 dyne/cm2), ~17% of that reported for normal EDL capillaries ().

Also noteworthy was a relative homogeneity to the day 5 neovessel network architecture, with no discernable feeding or draining vessels. However, over the next 48 hours, this nondifferentiated network converted into a hierarchical one with arterioles, capillaries, and venules (Fig. 2, F and G). This was associated with a pronounced reduction in vessel lumen diameter (P < 0.0001; Fig. 2H), a near tripling of blood vessel density (P < 0.0001; Fig. 2I), and a 54% increase in vessel branch density (P = 0.0005; Fig. 2J). Flow also increased rapidly between days 5 and 7, with a near doubling of RBC velocity (P < 0.0001; Fig. 2K). Collectively, these findings uncover a profound vascular transformation from dilated neoconduits with ultralow flow to a hierarchical microcirculatory network.

Primordial vessels give rise to a microvascular network via splitting

To discern how the primordial neoconduits, or “mother vessels,” reconfigured into a branched network, we assessed the day 5 RBC transit maps for evidence of sprouting. However, any cross-connections between mother vessels were typically of noncapillary caliber (~8 to 15 μm), making the scenario of opposing endothelial sprouts uncertain. Notably, while scrutinizing the flow patterns, we also found small (2- to 8-μm diameter) foci inside the lumen of mother vessels, foci that were entirely devoid of an RBC transit signal (Fig. 3, A and B). Real-time video sequences established these punctate zones to be physical obstructions to flow, with red cells deflecting around them and leaving a cell-free plasma stream, about 5 μm in length, immediately downstream of the obstruction (movie S3). These micro-obstructions could be found not only at Y-junctions but also within nonbranched segments of the mother vessels (Fig. 3, A and B).

Fig. 3.

Transformation of mother vessels into a microvasculature via intussusception.

(A to F) RBC transit maps of primordial vessels in EDL muscle 5 days after ischemic injury. Punctate zones completely devoid of RBC flow (micro-obstructions) are seen along the length of a vessel segment (A, cropped and vertically reflected from Fig. 2D) and at a Y-configured branch (B). A more elongated flow-devoid zone (microdivision) is shown in (C), and a “split” mother vessel is depicted in (D). Asymmetric splitting is shown in (E), with one daughter vessel being of capillary caliber (arrowheads). The vessel in (F) illustrates two microdivisions and a split in series. Micro-obstructions, microdivisions, and splits are denoted by yellow arrows. Flow direction is denoted by red arrows. See also movie S4. (G and H) Photomicrographs of semithin (0.5-μm) sections showing transluminal endothelial cell pillars/septa (arrows). The relative distance between sections is shown. Sectioned RBCs can be seen within the vessel lumens. In (G) (4.5 μm) and (H) (1 μm), finger-like cell projections cross the lumen and a small side lumen is evident (H, 3.5 μm). In (H), two opposing endothelial cells including nuclei encroach upon (4.5 and 6.5 μm) and transect the lumen (7 μm, arrow). See also movie S5. (I) Wide-field immunofluorescence images of primordial vessels containing transluminal pillars with and without a nuclei. Immunolabeling is as shown, and nuclei are stained with DAPI. (J to M) Confocal projections of neovessels 5 days after ischemic injury, immunostained for CD31 and nuclear-stained with DAPI. Image in (J) depicts filopodial-like projections consistent with endothelial sprouting. Image in (K) depicts anastomosis of two endothelial sprouts. Image in (L) depicts a CD31-positive pillar in cross section. A cell projection into the lumen is evident in (M). Schematics for each image are on the right.

There were also other RBC-free zones that were equally narrow but up to 50 μm in length. These longer RBC-free zones effectively microdivided the lumen (Fig. 3C). Also, there were mother vessels in which the lumen had definitively divided into two daughter lumens, separated from each other by the width of the parent vessel or greater. The separated vessel segments reconverged after 20 to 200 μm, giving the appearance of a split along the long axis of the vessel (Fig. 3D). The daughter vessels could be of equal or unequal diameter, and some were of capillary caliber (Fig. 3E). Micro-obstructions and microdivisions could also be found in series within a primordial vessel segment (Fig. 3F and movie S4).

This live-imaging evidence for flow micro-obstructions, luminal microdivisions, and vessel splits strongly indicates a robust program of intussusceptive angiogenesis in ischemic skeletal muscle.

Endothelial cell extensions and cell bodies enter the mother vessel lumen

To elucidate the structure of the flow obstructers, we undertook serial semithin (0.5 μm) sectioning of EDL muscle 5 days after injury (n = 5 mice). Among 50 primordial vessel segments, 50 to 200 μm in length, 24% contained one or more endothelial cell projection spanning the lumen. The morphology of these projections, or pillars, varied but with two main patterns: (i) a transluminal, finger-like endothelial extension, 1 to 5 μm in diameter (Fig. 3G), and (ii) an endothelial cell body including a nucleus that crossed the lumen (Fig. 3H and movie S5). Pillars were often asymmetrically positioned within the lumen (Fig. 3, G and H), and more than one pillar in the same primordial vessel cross section could be observed (Fig. 3H, sections at 3.5 and 4.5 μm, and movie S5), suggesting highly dynamic intraluminal endothelial cell activity. Immunostaining for CD31, vascular endothelial (VE)–cadherin, and endomucin revealed that pillars stained intensely for all three endothelial cell markers (Fig. 3I).

We also evaluated three-dimensional (3D) reconstructions of CD31-immunostained confocal images from 100-μm-thick sections. From 103 mother vessel segments, we identified four ablumenal endothelial sprouts, revealing a low prevalence of sprouting off mother vessels. These endothelial cells had filopodial extensions (Fig. 3J) and could be seen to bridge two adjacent mother vessels (Fig. 3K). However, from the same set of primordial vessel segments, we identified 72 transluminal projections/pillars. Pillars were circular in cross section, akin to the micro-obstruction signal seen with intravital imaging, with filled or circumferential CD31 signal (Fig. 3L). In longitudinal sections, the structures appeared as intraluminal projections off the wall and inside the lumen (Fig. 3M). To ascertain whether there were features that distinguished vessel segments that give rise to endothelial sprouts versus pillars, we evaluated a second series of mother vessel segments from reconstructed confocal images. This confirmed the paucity of sprouts (9 of 740 vessel segments) and revealed the median diameter of the vessel from which a pillar arose was 43% higher than that for vessels from which a sprout emanated (P = 0.030).

These data establish that conventional endothelial sprouting off the mother vessel wall occurred but was a minority event. Instead, the dominant endothelial cell rearrangement event in early vascular regeneration was positioning of endothelial cell processes and endothelial cell bodies inside the lumen, effectively transecting the channel.

Pillar formation and vessel splitting are accentuated by inhibiting VEGFR2 signaling

To investigate the molecular control over endothelial pillar formation, we considered the role of VEGFR2 signaling, which is vital for sprouting angiogenesis (). We first determined whether VEGFR2 signaling was active in mother vessels, by triple-immunostaining muscle tissues for CD31, VEGFR2, and phosphorylated VEGFR2 (p-VEGFR2). p-VEGFR2 was probed using an antibody that binds to phosphorylated tyrosine residues in the activation loop (human Y1054/Y1059 and mouse Y1052/1057) that are required for, and a marker of, VEGFR2 activity (, ). This revealed little to no VEGFR2 activity in capillary endothelial cells of normal skeletal muscle but robust p-VEGFR2 signal in endothelial cells of mother vessels in regenerating muscle (Fig. 4A). Total VEGFR2 signal was similar between capillaries and primordial vessels (Fig. 4A).

Fig. 4.

Inhibition of VEGFR2 activity promotes the formation of intussusceptive angiogenic structures.

(A) Confocal micrographs of projected optical sections of native EDL muscle (top row, 20-μm-thick projection) and EDL muscle 5 days after ischemic injury (bottom row, 16 μm thick), immunostained for CD31 (green), VEGFR2 (grayscale), p-VEGFR2 (red), and DAPI (blue). Arrows depict the borders of a mother vessel with several microdivisions/early splitting zones (asterisks). Strong p-VEGFR2 immunoreactivity is present in the primordial vessels but not capillaries. (B) Epifluorescence micrographs of transverse sections of EDL muscle 5 days after ischemic injury and 16 hours after receiving vehicle (left) or apatinib (right), immunostained for CD31, p-VEGFR2, and DAPI. p-VEGFR2 signals are diminished in CD31-positive microvessels and nonvascular cells in mice receiving apatinib. (C) RBC transit maps of microvascular networks in regenerating (5-day) EDL muscle 16 hours after receiving apatinib, cabozantinib or ZM323881 (small-molecule inhibitors of VEGFR2) or DC101 (VEGFR2-blocking antibody), and respective controls. Pillars have been circled yellow (arrows) and microdivisions/splits are outlined in dashed blue lines. Graphs depicting the density of microvascular networks are shown on the right [means ± SE; n = 26/13, 17/12, 17/19, and 21/18 distinct muscle territories (0.63 mm2) in the EDL subjected to control or designated blocking agent]. The effects of cabozantinib and ZM323881 were studied using the same vehicle controls [dimethyl sulfoxide/polyethylene glycol (PEG)/saline], and P values are Bonferroni corrected. (D and E) Graphs depicting the density of pillars (D) and splits (E) in EDL muscle 5 days after ischemic injury. Data are expressed relative to the respective controls, depicted by the dashed line (means ± SE, n = 13, 12, 19, and 18).

We next determined whether pillar formation and mother vessel splitting depended on VEGFR2 signaling. For this, mice were subjected to femoral artery excision and 4.5 days later received an intraperitoneal injection of the VEGFR2-selective tyrosine kinase inhibitor (TKI), apatinib. Sixteen hours later, the EDL vasculature was interrogated for intussusception by intravital microscopy. This timed strategy ensured that primordial vessels still formed, but also resulted in reduced p-VEGFR2 signals in endothelial cells, inflammatory cells, and satellite cells at the time of intravital interrogation (Fig. 4B). To our surprise, intravital assessment revealed that apatinib did not suppress vascular network formation. Instead, it led to hypervascularization. Compared with vehicle-injected mice, there was crowding of primitive vessels and a 17% increase in vessel length density (P = 0.041; Fig. 4C). Furthermore, there was a near-tripling of micro-obstructions (pillars) per vessel length (P = 0.0002; Fig. 4D). As well, the density of vessel splits more than doubled following apatinib administration (P < 0.0001; Fig. 4E).

In an effort to validate these seemingly paradoxical findings, we undertook in vivo blocking experiments with two other VEGFR2 TKIs, cabozantinib () and ZM323881 (). Similar to apatinib, delivery of these VEGFR2 inhibitors led to crowding of mother vessels, increased length density (P = 0.039 and P = 0.021; Fig. 4C), increased transluminal pillars (P = 0.016 and P = 0.001; Fig. 4D), and increased vessel splits (P = 0.007 and P = 0.003; Fig. 4E). In contrast, endothelial cell sprouting off a mother vessel, despite being infrequent, was inhibited upon VEGFR2 blockade (fig. S1). The concordant evidence for hyperactive pillarization and accelerated vessel intussusception with three different inhibitors strongly suggests that VEGFR2 signaling has a suppressive effect on primordial vessel intussusception.

The VEGF-binding domain of VEGFR2 is not dominant for modulating pillar formation

VEGFR2 can be activated not only by VEGF ligand but also by shear forces. To shed light on which may be operant in regulating endothelial pillars, we studied the impact of the VEGFR2 blocking monoclonal antibody, DC101. In cultured endothelial cells, DC101 (10 μg/ml) fully inhibited VEGFA-mediated phosphorylation of VEGFR2, as did the TKI cabozantinib (300 nM; fig. S2). Cabozantinib also abrogated flow (shear stress of 3 dyne/cm2 for 15 min)–induced activation of junctional VEGFR2 (P < 0.0001, effect size r = 0.29). However, DC101 only modestly reduced flow-mediated activation of VEGFR2 (38% suppression, P = 0.015, effect size r = 0.15; fig. S2), and the signal remained significantly above the baseline for endothelial cells not subjected to flow (P = 0.0001). Thus, the actions of DC101 were biased toward ligand-mediated signaling, leaving VEGFR2 mechanotransduction more intact. When administered to mice subjected to femoral artery excision, at a dose shown to block sprouting angiogenesis in vivo (), DC101 did not increase neovessel density in the regenerating muscle (P = 0.446; Fig. 4C). Likewise, there was at most a trend toward increased pillar and split densities within the neovasculature (P = 0.160 and P = 0.100; Fig. 4, D and E). Together with the small-molecule inhibitor data, these findings implicate reduced shear-induced VEGFR2 signaling as a critical determinant of pillar formation.

VEGFR2 activity is selectively reduced and lateralized in pillar endothelial cells

Given the functional linkage between inhibiting VEGFR2 and pillar formation, we sought to understand the landscape of VEGFR2 activation in the pillar-forming endothelial cells themselves. Confocal reconstructions of mother vessels confirmed abundant p-VEGFR2 signal in the endothelial cells lining the vessel wall (Fig. 5A). However, the VEGFR2 activity signal in the pillar endothelial cells was different. Notably, the p-VEGFR2 signal intensity was less than the adjacent lining cells, although the CD31 signal in the pillar was pronounced (P = 0.009; Fig. 5, A and B).

Fig. 5.

Reduced and lateralized VEGFR2 activity on pillar-forming endothelial cells.

(A) Confocal micrographs of planar-projected optical sections of a primordial vessel in EDL muscle 5 days after ischemic injury immunostained for CD31 (green) and p-(activated) VEGFR2 (red), with TO-PRO-3–stained nuclei. A CD31-expressing pillar is present (arrow). There is abundant p-VEGFR2 signal in endothelial cells of the wall but not the pillar. (B) Graph depicting p-VEGFR2 signals in endothelial pillars versus wall-lining endothelium of the same vessel. N = 8 pillars from different primordial vessels. Wall and pillar endothelial cell data from a given primordial vessel are connected. A.U., arbitrary units. (C) Confocal planar projection of a primordial vessel showing a CD31-expressing pillar endothelial cell (arrow) with modest p-VEGFR2 signal lateralized to one face, orthogonal to the vessel long axis (small arrows). (D) Confocal orthogonal (XZ, YZ) reconstructions of a day 5 primordial vessel showing an endothelial cell pillar structure inside the lumen. p-VEGFR2 signal is abundant in the lining endothelium, but on the pillar p-VEGFR2 appears as a discrete lateralized signal. (E) Distribution of p-VEGFR2–positive pillars with diffuse versus lateralized p-VEGFR2 signal. (F) Confocal planar projections through a pillar within a primordial vessel in day 5–regenerated muscle showing VEGFR2 activity that is focal, lateralized, and enriched at the midzone of the pillar (arrows). The two planar projections are 3 μm apart. A schematic of the micrograph is depicted (right). (G) Distribution of p-VEGFR2 locations in lumen-crossing pillars.

Furthermore, the pillar p-VEGFR2 signal was generally neither diffuse nor circumferential, as was the case for total VEGFR2, but instead was lateralized on the pillar (Fig. 5, C and D, and fig. S3). Among 35 mother vessel pillars displaying a p-VEGFR2 signal, 83% had focal p-VEGFR2 that was lateralized on the surface (P = 0.011; Fig. 5, C to E). In addition, rather than distributing along the entire length of the pillar, the VEGFR2 activity signal was observed in the midregion, i.e., away from the wall-pillar interface. Among 19 pillars that could be observed to cross the entire lumen, 68% showed midregion enrichment (P = 0.009; Fig. 5, F and G). Thus, whereas VEGFR2 activity in primordial vessel endothelial cells was overall high, there was mosaicism, with relatively low VEGFR2 activity specifically in pillar endothelial cells that, when identifiable, was focal and lateralized.

VEGFR2-deficient endothelial cells subjected to ultralow shear stress are predisposed to pillarizing

The reduced VEGFR2 activity in pillar endothelial cells, and the pillar-promoting effect of VEGFR2 inhibition, raised the possibility that endothelial cells with reduced capacity to transduce VEGFR2 signals might be autonomously predisposed to entering the vessel lumen. To explore this possibility, we considered that an in vitro model of pillar formation was required. To our knowledge, such a model has not been reported. Therefore, we developed a microfluidic-based endothelial channel model that approximated the flow conditions of the primordial vessels we found in regenerating skeletal muscle (fig. S4). Channels were fabricated from polydimethylsiloxane (PDMS) and seeded with human umbilical vein endothelial cells (HUVECs) to yield confluent endothelial cell monolayers on all four surfaces. HUVECs were selected for study because of their early developmental status and the fact that the primordial vessel endothelial cells had attributes of venous endothelial cells. Microchannels were subjected to ultralow flow (shear rate, 37 s−1; shear stress, 0.3 dyne/cm2). Under these conditions, the lining endothelial cells remained viable, with abundant VE-cadherin–containing intercellular connections and with prominent cortical F-actin bundles (Fig. 6A). We also observed a small proportion of endothelial cells that spanned across the lumen to form a pillar. These structures were situated near corners that assumed a shoulder-like configuration (Fig. 6, B to D, and movies S6 and S7). The transluminal bridges were either slender, finger-like projections or thicker projections that contained the endothelial cell nucleus (Fig. 6B). The pillar cytoplasm contained F-actin microfilament bundles running the length of the pillars. Multiple adjacent pillars parallel to each other could also be seen (Fig. 6C). The pillars connected either to an endothelial cell lining the opposing wall, or end to end to an opposing cell projection within the lumen, with intervening VE-cadherin (Fig. 6D and movie S7).

Fig. 6.

Pillar formation by endothelial cells subjected to ultralow shear stress in a microfluidic-based model.

(A) Confocal planar projection image (120 z-slices, 0.25-μm step size) of HUVECs on the bottom surface of a PDMS microfluidic device subjected to ultralow shear (0.3 dyne/cm2) for 8 hours. Cells are stained for F-actin (phalloidin, green), VE-cadherin (red), and nuclei (DRAQ5, blue). HUVECs display prominent cortical F-actin (arrows) and variable F-actin stress fibers. (B) Orthogonal projections of confocal images of shoulder regions of HUVEC-lined microfluidic channels, stained as in (A). Top image depicts a finger-like cell extension crossing the lumen at the channel shoulder (arrow). Bottom image depicts a thicker transluminal bridge (arrow) consisting of an endothelial cell body with nucleus (smaller arrow). (C) 3D volume image of HUVEC-lined microchannel, reconstructed from confocal z-slices (120 z-slices, 0.25-μm step size, 75 μm in the y dimension), stained with phalloidin. Multiple F-actin–rich transluminal pillars at the shoulders can be seen (arrows), including three adjacent pillars, 17 and 22 μm from each other. See also movie S6. (D) 3D volume image of phalloidin- and VE-cadherin–stained microchannel reconstructed from confocal z-slices (120 z-slices, 0.25-μm step size, 70 μm in the y dimension). An F-actin–rich transluminal pillar is present with a focus of VE-cadherin (arrow) indicating connection of two endothelial cell processes within the lumen. See also movie S7.

We next determined whether low VEGFR2 activity in endothelial cells predisposed them to enter the lumen as a pillar. For this, we coseeded green fluorescent protein (GFP)–expressing endothelial cells and red fluorescent protein (RFP)–expressing endothelial cells, differentially transfected with either VEGFR2 small interfering RNA (siRNA) or control siRNA. The former displayed an 88 ± 5% knockdown in VEGFR2 transcript abundance. We first evaluated the cells under static conditions in 2D cultures. This revealed no evidence for cell protrusions out of the monolayer, as assessed from confocal z-projections (fig. S5). However, VEGFR2 knockdown endothelial cells did display unusual lateral protrusions that could cross over, or under, control endothelial cells, suggesting less morphological constraint on VEGFR2 knockdown cells when interacting with neighboring control cells (fig. S5). We then coseeded the differentially transfected endothelial cells in microfluidic devices. At confluence, the knockdown-to-control cell ratio in the channel was 2:3. VEGFR2 knockdown endothelial cells had a significantly lower length-width ratio in the flowing channel than control cells (P < 0.001; Fig. 7A). Moreover, VEGFR2 knockdown endothelial cells were 10 times more likely to form a transluminal pillar than control endothelial cells in the same device and flow conditions (P = 0.008; Fig. 7, B to E). As with nontransfected HUVECs, the VEGFR2 knockdown endothelial cell pillars were formed by either a cellular extension (Fig. 7, C and E) or the endothelial cell body itself (Fig. 7D). The transluminal bridge connected to a lining endothelial cell that was typically a control, VEGFR2-expressing cell (Fig. 7, C to E). Thus, under conditions of ultralow flow, the extent to which endothelial cells are capable of signaling through VEGFR2 affects the decision to integrate into the wall or enter the lumen as a pillar.

Fig. 7.

VEGFR2 deficiency and reduced NO availability predispose endothelial cells in ultralow flow conditions to form pillars.

(A) Graph depicting the aspect ratio of control and VEGFR2 knockdown endothelial cells lining a PDMS microfluidic device subjected to ultralow shear stress (0.3 dyne/cm2). A total of 394 cells with control siRNA and 379 with VEGFR2 siRNA from four different devices were quantified. Median values and all data points are shown. (B) Pillar content, defined as the proportion of endothelial cells at microchannel shoulders that formed a pillar, for differentially labeled control and VEGFR2 knockdown endothelial cells. Data are from four independent experiments and depicted as mean and SE. (C to E) Confocal microscopy volume projections of differentially labeled control and VEGFR2 knockdown endothelial cells lining a microfluidic device and subjected to ultralow shear stress. In (C) and (D), control-siRNA cells express GFP, and VEGFR2-siRNA cells express RFP. In (E), the labeling is reversed. Nuclei were stained with DAPI. In (C) and (E), pillars are composed of VEGFR2-siRNA cell projections that span across the lumen and connect with a control-siRNA cell on the opposite wall. In (D), a VEGFR2-siRNA cell body crosses the lumen and is connected on either side with control siRNA cells lining the device walls. (F) Pillar content of microfluidic channels lined by HUVECs subjected to ultralow shear stress (0.3 dyne/cm2) and infused with L-NAME (10 μM) or water vehicle. (G) Pillar content of endothelial cell–lined microchannels under ultralow shear conditions infused with DETA NONOate (1 μM), ZM323881 (10 nM), or the combined interventions. Data are means and SE.

VEGFR2 deficiency–driven pillar formation depends on nitric oxide availability

Nitric oxide is a well-established downstream mediator of VEGFR2 signaling, including the signaling response to flow (). We therefore asked whether endothelial cell nitric oxide was also a controller of pillar formation in the low-flow environment. Given that suppressed VEGFR2 signaling increased pillar formation, we first assessed whether low nitric oxide (NO) had a similar effect. HUVEC-lined channels in the microfluidic device were subjected to ultralow flow (0.3 dyne/cm2) in the presence of 50 μM L-NAME or vehicle. This revealed a 3.1-fold increase in pillar formation with NO inhibition (P = 0.0003), implicating NO in stabilizing the endothelial monolayer and suppressing pillarization (Fig. 7F). We next tested whether reduced NO was a downstream effector of low-flow pillarization by endothelial cells with suppressed VEGFR2 signaling. Infusion of the VEGFR2 selective inhibitor, ZM323881 (10 nM), augmented pillar formation, recapitulating the effect of VEGFR2 mRNA knockdown. Furthermore, the increase in pillar formation upon inhibiting VEGFR2 was blunted in the presence of the NO donor, DETA NONOate (Fig. 7G). These data implicate endothelial NO as a downstream determinant of the decision to form a pillar or consolidate within the vessel wall.

DISCUSSION

We have found that vascular regeneration following ischemic muscle injury in mice does not follow the canonical processes of sprouting angiogenesis. Instead, reconstruction of a vascular network after femoral artery excision began with the formation of large-caliber neoconduits receiving ultralow blood flow. These primordial conduits transformed over 48 hours into a branched and hierarchical network and did so primarily by intussusception. Intussusception, in turn, was launched by endothelial cells with reduced VEGFR2 signaling that entered the flowing lumen as a transluminal pillar, where they received VEGFR2 activation signals. These findings reframe regenerative angiogenesis in skeletal muscle as an intussusceptive process and uncover the mechanistic basis of its launch.

A program of primordial vessel formation and intussusception in regenerating muscle was ascertained by real-time microscopy of red cell transit, from serial semithin sections and from 3D confocal reconstructions of thick muscle sections. The latter strategy also revealed that endothelial cell sprouting events, with characteristic filopodial extensions, were still part of the early vascular regenerative process. However, there was a 40-to-1 ratio of intraluminal positioning of endothelial cells in comparison to extraluminal sprouting. This bias toward intussusception may be critical for regenerating the vasculature in the high-demand environment of regenerating muscle. Intussusception is faster and more energy efficient that sprouting angiogenesis (, ). Moreover, splitting of mother vessel tracks resolves the architectural challenge of forming long, parallel neocapillaries aligned with myofibers (). This vessel architecture contrasts with the capillary meshwork produced by sprouting angiogenesis, as exemplified by the developing retina (, , ). Intussusception might also explain how cross-connections between capillaries are formed, because cross-bridges would be a natural end point of two opposing fronts of splitting. Because some endothelial cell sprouts reaching across parallel neovessels were also found, we propose that the angiogenesis program in regenerating skeletal muscle is dominated by intussusception and refined by sprouting.

Being the first vessels to receive flow during muscle regeneration, the wide-caliber primordial channels are of great interest. Flow in these channels was remarkably slow, with median shear rates less than 10% of that in the venous circulation (). The cellular constituents of these slow-flow conduits are also noteworthy. They were lined by densely packed endothelial cells that expressed CD31, VE-cadherin, and endomucin, a profile described for capillary and vein endothelial cells (). The origin of these endothelial cells is speculative, but one consideration is tissue-resident endothelial progenitor cells. Endothelial cell progenitors located in the vessel lining, with the capacity to regenerate a vascular network, have been identified in a number of adult tissues (–). Surviving, local endothelial progenitors as a source of vessels that emerge after ischemic injury would constitute an interesting parallel with the regeneration of skeletal myofibers from satellite cells and merits study. Also noteworthy was our finding that, although the primordial channels were much larger than capillaries, they were heavily invested by pericytes. This mural cell coverage of the channels is consistent with their ultimate fate as capillaries. The pericytes can also be expected to support the viability of the primordial conduits despite the slow-flow (), ensuring the transformation to a differentiated microvessel network can proceed.

Our finding of robust pillar formation under ultralow shear conditions is consistent with computational simulations of the chorioallantoic membrane (CAM) vasculature, which predicted pillar formation at sites of low shear stress (). Low and variable shear rates have also been identified in intussusception-prone tumor vessels () and in channels that pathologically dilate after VEGF overexpression (). In each of these contexts, the vasculature can be considered unstable. We propose that the primordial, slow-flow channels in ischemic muscle exist in a metastable state. They are resistant to regression, but the lining endothelial cells are not in a steady-state relationship with blood flow. In this setting, undergoing one or more rounds of vessel splitting may be a programmed response designed to quickly bring endothelial cell shear rates toward that of a normal muscle microcirculation. In doing so, a capillary landscape suitable for oxygen delivery is rapidly generated.

Beyond the permissive environment of low shear stress, how intussusceptive pillars form has been enigmatic. As a paradigm, it has been tempting to consider endothelial pillar formation as a mirror-image phenomenon to endothelial sprouting. The current data indicate that the mechanistic similarities between pillar formation and sprouting may, in fact, be limited, particularly in the context of VEGF-VEGFR2 signaling. Whereas activation of VEGFR2 in select endothelial cells is fundamental to sprouting (), we found no evidence that VEGFR2 activity mediated intraluminal entry of endothelial cells. None of four different VEGFR2 blockers, delivered when mother vessels were emerging, inhibited pillar formation. This strongly suggests that pillar formation is not a scenario of simply reversing the polarity of VEGFR2-mediated signaling.

However, blocking VEGFR2 signaling was not neutral for the intussusceptive process. Upon inhibiting VEGFR2 signaling, mother vessels aggressively pillarized and split. This was an unusual finding, recognizing that ischemic, regenerating skeletal muscle is a high-VEGF environment () and our finding of high VEGFR2 activity in mother vessel endothelial cells. However, although overall VEGFR2 signaling in these vessels was high, there was cellular mosaicism. Those endothelial cells that entered the lumen as pillars displayed lower p-VEGFR2 signals than those that lined the lumen. Together with the pro-pillar consequences of inhibiting VEGFR2, this raised the possibility that having inherently low VEGFR2 signaling capacity may underlie a cellular decision to form a pillar.

This possibility of an endothelial cell-autonomous decision to form a pillar was established using a microfluidic model. To our knowledge, the endothelial-lined channel system we used for this constitutes the first in vitro scenario of transluminal positioning of endothelial cells inside a flowing lumen. Under conditions of ultralow shear stress, endothelial cell projections and entire endothelial cell bodies were found bridging the lumen, with configurations very similar to those observed in mother vessels in vivo. The microfluidic system also allowed us to model a heterogeneous mix of endothelial cells with differing capacities to signal through VEGFR2, using differential labeling. Under the same culture and flow conditions, those endothelial cells with diminished VEGFR2 preferentially formed a pillar. They did so while making, or retaining, VE-cadherin–associated connections with other endothelial cells.

The data also implicate the shear-sensing role of VEGFR2 as being central to the pillar-forming decision of endothelial cells. Endothelial cells detect and align with shear forces via a mechanosensory complex composed of CD31, VE-cadherin, and VEGFR2 (). VEGFR2 is phosphorylated during this response (), and VEGFR2 knockdown reduces alignment of endothelial cells in the direction of shear stress (). We found that, even under the ultralow flow conditions generated in the microfluidic conduit, VEGFR2-low endothelial cells were less elongated than control cells, a potential precursor to repositioning into the lumen. That a relative lack of shear sensing contributed to pillar formation was also suggested in vivo. Delivery of DC101, an anti-VEGFR2 antibody that binds the extracellular domain of VEGFR2 () and was biased toward blocking ligand-mediated signaling, had little effect on pillar formation in mother vessels despite VEGFR2 TKIs accentuating pillar formation. Thus, impaired shear sensing by VEGFR2, rather that impaired responsivity to VEGF ligand, appears to be an important condition for entering the lumen as a pillar.

Nitric oxide is a downstream mediator of VEGFR2 signaling in endothelial cells (). It was thus noteworthy that, under conditions of ultralow flow, inhibiting eNOS with L-NAME promoted pillar formation in the microfluidic model. Moreover, supplying NO with DETA NONOate partially abrogated the pillarization response to inhibiting VEGFR2 signaling. These findings implicate the VEGFR2-NO axis, specifically a deficiency, in the pillar-forming cascade. Both NO donor-induced and flow-induced S-nitrosylation of cytoskeletal proteins in endothelial cells have been identified (, ). We speculate therefore that suppressed NO in VEGFR2-deficient endothelial cells might affect the cytoskeletal architecture, which may underlie the permissive condition for egressing out of the monolayer.

In addition to the permissive environment afforded by ultralow flow and a reduced VEGFR2-NO axis, might an underdeveloped shear-sensing system directly promote pillarization? Endothelial cells with immature shear-VEGFR2 signaling, and with ultralow shear inputs, would be at an extreme, low end of a shear stress signaling spectrum. We propose that repositioning into the lumen is a bold cellular attempt to normalize this signaling axis. Several findings support this. Real-time imaging of RBC transit in primordial vessels demonstrated a unique hemodynamic environment for pillar endothelial cells. Red cells bombarded the pillar, deflected around it, and left a red cell–free plasma stream downstream. Notably, these in vivo findings are concordant with computationally modeled shear stresses around intussusceptive pillars in the mouse mucosal plexus, which predicted elevated shear stresses along the upstream sides of the pillar and a relative shear stress “dead zone” on the downstream surface (). Moreover, our identification of local p-VEGFR2 signal in endothelial pillars in vivo provided a live-cell readout of this putative shear distribution. Any p-VEGFR2 signal that was detected in pillar cells was not dispersed, as for lining endothelial cells, but lateralized to one face of the pillar. In addition, the p-VEGFR2 signal on mother vessel pillars tended to be confined to the midregion, away from the wall-pillar interface (see Fig. 5F), which is also remarkably similar to computational shear mapping data (). Collectively, the findings indicate that, even with a suppressed capacity to signal, pillar cells inside the lumen can receive a threshold level of shear to locally activate VEGFR2.

Accordingly, we speculate that movement of endothelial cells from a position lining the wall to inside the lumen as a pillar may reflect shear-seeking behavior. This cellular decision to reposition as a pillar could be stabilizing particularly for those endothelial cells with a diminished capacity to respond (e.g., activate VEGFR2) to the low shear conditions of primordial blood vessels. Shear-seeking behavior has been documented as a bona fide endothelial cell attribute, observed during vessel branch regression wherein cells in low-shear conditions migrate against the direction of blood flow (, ). Intussusceptive pillar formation may thus share this flow-seeking principle, but in a different context and with a very different consequence.

Vascular regeneration in ischemia-injured skeletal muscle is fundamental to restoring muscle function and to informing therapies for vascular disease. We have established that the microvascular regenerative program in ischemic skeletal muscle is dominated, not by sprouting, but by the formation of large, metastable channels that rapidly transform into a neocirculation via intussusception. This intussusceptive program is launched by endothelial cells with underdeveloped VEGFR2 signaling capacity that bridge across the lumen, seemingly to exist in a more favorable environment for mechanically activating VEGFR2. These findings argue for a reconsideration of angiogenesis paradigms in ischemic muscle.

METHODS

Mouse hindlimb ischemia

Experiments were conducted in accordance with the University of Western Ontario’s Animal Care and Use Subcommittee, which follows the policies set out by the Canadian Council on Animal Care. Male, 12-week-old C57BL/6J mice (Jackson Laboratory, Bar Harbor, ME) were anesthetized with ketamine (80 mg/kg) and xylazine (10 mg/kg). Hindlimb ischemia was induced by ligating the right femoral artery above and below the profunda femoris branch using 6-0 silk sutures and excising the intervening 5- to 6-mm portion of artery (, , ).

Histology and immunostaining

The hindlimb anterior muscle bundle, including the EDL, tibialis anterior, and peroneus longus muscles, was fixed overnight in tris-buffered zinc or 4% paraformaldehyde and embedded in paraffin. Five-micrometer cross sections through the proximal, middle, and distal regions were stained with hematoxylin and eosin.

Immunofluorescence staining was undertaken following heat-mediated antigen retrieval with citrate buffer. Antibodies used were (i) rabbit polyclonal anti-CD31 (1:100; RB-10333-P1, Thermo Fisher Scientific), detected with biotinylated donkey anti-rabbit immunoglobulin G (IgG) (1:200; Jackson ImmunoResearch) and DyLight 549–conjugated streptavidin (1:200; Vector Laboratories); (ii) rat monoclonal anti-CD31 (1:20; Dianova, Clone SZ31), detected with Alexa Fluor 488–conjugated goat anti-rat IgG (1:100; Thermo Fisher Scientific); (iii) goat polyclonal anti–VE-cadherin antibody (1:100; R&D Systems, AF1002) and Alexa Fluor 488–conjugated donkey anti-goat IgG (1:200; Thermo Fisher Scientific); and (iv) rat monoclonal anti-endomucin antibody (1:100; Santa Cruz, Clone V.7C7, sc-65495), detected with Alexa Fluor 488–conjugated goat anti-rat IgG (1:200; Thermo Fisher Scientific). Also, VEGFR2 activation in endothelial cells was assessed by double immunolabeling with rabbit polyclonal anti-VEGFR2 p-Y1054/1059 antibody (p-VEGFR2, 1:50; Abcam, ab5473) and either rat anti-endomucin or rat anti-CD31 monoclonal antibodies. The anti–p-VEGFR2 antibody binds to activation loop tyrosine residues (mouse Y1052/1057) that are required for, and a marker of, VEGFR2 activity (, ). Bound antibodies were detected using Alexa Fluor 594–conjugated goat anti-rabbit IgG (1:100; Thermo Fisher Scientific) and Alexa Fluor 488–conjugated goat anti-rat IgG. Nuclei were visualized with 4′,6-diamidino-2-phenylindole (DAPI)–containing Fluoromount-G (SouthernBiotech, 0100-020). Thin sections were imaged by wide-field microscopy (Olympus BX-51) with Northern Eclipse (EMPIX Imaging Inc.) software or a Leica TCS SP8 confocal microscope and Leica Application Suite X software (confocal microscopy detailed below).

Thick-section whole-vessel immunostaining

Mice were euthanized by isoflurane overdose and perfused at physiologic pressure with phosphate-buffered saline (PBS) followed by 4% paraformaldehyde via a left ventricular cannula. Following cryoprotection with sucrose, muscles were embedded in OCT embedding medium (Tissue-Tek) and longitudinally sectioned at 100-μm thickness. Sections were permeabilized with 0.5% Triton X-100 in PBS and double immunostained for CD31 and either NG2 or p-VEGFR2, using as primary antibodies biotinylated rat anti-CD31 antibody (1:25; 553371, BD Biosciences), rabbit polyclonal anti-NG2 antibody (1:200; AB5320, Millipore), or rabbit polyclonal anti–p-VEGFR2 Y1054/1059 antibody. Bound antibodies were visualized using DyLight 488–conjugated streptavidin (1:100; Vector Laboratories) and Alexa Fluor 546–conjugated donkey anti-rabbit IgG (1:100 or 1:200; Thermo Fisher Scientific). Nuclei were visualized with TO-PRO-3 iodide (Thermo Fisher Scientific). Triple immunostaining was performed using rat monoclonal anti-CD31 antibody (1:20; Clone SZ31, Dianova), rabbit polyclonal anti–p-VEGFR2 Y1054/1059 antibody, and goat polyclonal anti-VEGFR2 antibody (1:100; AF644, R&D Systems). Bound antibodies were visualized using Alexa Fluor 488–conjugated donkey anti-rat IgG (1:100; Thermo Fisher Scientific), Alexa Fluor 546–conjugated donkey anti-rabbit IgG, and Alexa Fluor 633–conjugated donkey anti-goat IgG (1:200; Thermo Fisher Scientific). Following labeling, thick sections were transferred to positively charged glass slides and mounted between 100-μm-thick plastic coverslip spacers (Thermo Fisher Scientific) with Fluoromount-G (0100-20, SouthernBiotech).

Serial semithin sectioning

Semithin sectioning of hindlimb anterior muscles was performed as described (). After perfusion fixation with 4% paraformaldehyde, anterior hindlimb muscles were immersed in Karnovsky’s fixative (15731-10, Electron Microscopy Sciences), infiltrated with plastic embedding mixture, and sectioned into 0.5-μm-thick serial ribbons using a Histo Jumbo diamond knife (Diatome AG, Biel, Switzerland) and ultramicrotome (Reichert-Jung Ultracut E). Up to 1000 serial sections for each muscle specimen were subjected to polychromatic staining with basic fuchsine and methylene blue as previously reported () and mounted with Eukitt mounting medium (03989, Sigma-Aldrich). Stained serial sections were scanned using a ScanScope CS bright-field digital slide scanner with a 40× objective (Aperio Technologies). Individual serial TIFF images were generated (Sedeen Viewer Software, Pathcore), rotated, and aligned (Photoshop, Adobe). Videos were generated from serial images in 3D Slicer.

Intravital video microscopy

RBC transit within EDL muscle microvessels in a 50-μm-deep zone was assessed by epifluorescence intravital video microscopy as described previously (). Mice were anesthetized with ketamine and xylazine, and body temperature was maintained at 37°C with a heat lamp. The EDL muscle was separated from the tibialis anterior and peroneus longus muscles, covered with a glass coverslip, and positioned face-down on the stage of an inverted Olympus IX81 microscope. After a 20-min stabilization period, mice received an intravenous injection of fluorescein isothiocyanate (FITC)–labeled dextran (2 × 106 Da; Sigma-Aldrich), and RBC transit was visualized by blue light epi-illumination (, ). Video recordings (696 × 520 pixels, 21 frames per second) were captured using a cooled charge-coupled device camera (Rolera-XR, QImaging) for 15 s. The entire EDL surface zone was recorded in 7 to 10 fields of view. Video sequences were digitized as uncompressed AVI files for postprocessing using custom acquisition software (Neo Vision) and in-house software written in the MATLAB (MathWorks) ().

Analysis of microvascular network architecture

RBC transit maps of EDL microvascular networks were generated from “sum of absolute differences” images, which is based on the cumulative sum of the absolute differences in light intensity values at each pixel between consecutive video frames (). From these maps, vessel lumen diameter and vascular density [microvasculature length (μm)/vascularized EDL area (μm2)] were quantified, the latter by tracing of all vessel centerlines and normalizing total length to vascularized EDL area. Bifurcation density was determined by counting all bifurcations and expressing relative to total microvasculature length. The density of intraluminal punctate obstructions, microdivisions, and vessel splits were similarly quantified and expressed relative to total microvasculature length. Punctate obstructions were defined as circular signal voids, 2- to 8-μm diameter, within flowing microvessels. Microdivisions were defined as segments where transiting RBCs diverged into two distinct paths, separated by no more than 7 μm, and then reconverged within 50 μm. Vessel splits were defined as two daughter lumens separated from each other by at least the width of the parent vessel and reconverging within 200 μm. All outcomes were evaluated in three to six independent mice per condition.

Analysis of RBC velocity and vessel wall shear

RBC velocities [VRBC (micrometer per second)] in individual microvessels were quantified from video files, and generating 2D plots of RBC location change with time (space-time images), as described previously (, ). This enabled unbiased quantification of RBC transit velocities within microvessels throughout the entire EDL muscle. Mean VRBC averaged over a 15-s imaging time frame within each microvessel was determined in a total of 114 uninjured capillaries and 193 postischemia regenerated microvessels. Vessel wall shear rate was calculated using the formula, vessel shear rate = 8V/D, where D is the vessel lumen diameter and V is vessel fluid velocity, which we equated to RBC velocity. Vessel wall shear rate was ascertained in 42 primordial vessels 5 days after ischemic injury ranging from 20 to 46 μm in diameter. Wall shear stress (dyne per square centimeter) was determined by multiplying wall shear rate by blood viscosity [3.84 centipoise (cP)] ().

Laser scanning confocal microscopy of muscle vasculature

Vessels were imaged with (i) a Zeiss LSM 510 Meta Confocal Microscope using a 40× water immersion objective and Argon2 (488 nm), HeNe1 (543 nm), and HeNe2 (633 nm) lasers, generating up to 50, 1-μm-thick z-slices at a pixel resolution of 90 nm; (ii) a Leica TCS SP8 Confocal Microscope using a 40× oil immersion objective and Diode 405, OPSL 488, OPSL 552, and Diode 638 lasers, generating up to 100, 0.8-μm-thick z-slices at a pixel resolution of 70 nm; or (iii) a Nikon A1R Confocal Laser Scanning System using a 40× oil immersion objective and 488-nm laser, generating up to 80 z-slices with a z-step size of 0.20 μm and x-y resolution of 100 nm/pixel using a Resonance scanner. Z-slices were reconstructed into a 3D volume through maximum intensity projections with Zen software (Zeiss), LAS X software (Leica), or NIS Elements software (Nikon).

VEGFR2 activation in endothelial cells was assessed in thick tissue sections from the immunofluorescence signal intensity for p-VEGFR2. To compare p-VEGFR2 signals in pillar endothelial cells with those in cells lining the channel, 3D volumes were generated. p-VEGFR2 signal intensity was measured in three adjacent optical sections through the pillar and two sections of the mother vessel wall, 20 to 25 μm on either side of the pillar. Lateralization of pillar-associated p-VEGFR2 was assessed from orthogonal XZ and YZ projections of 3D reconstructions. p-VEGFR2 signal was considered to be lateralized on a pillar if the signal was restricted to half or less of the perimeter of the pillar and crossed the pillar midline orthogonal to the vessel long axis.

In vivo blockade of VEGFR2 signaling

To block VEGFR2 signaling in vivo, mice were intraperitoneally injected with small-molecule inhibitors of VEGFR2 tyrosine kinase, a VEGFR2-blocking antibody, or their respective vehicles, 4.5 days following surgery. Blocking agents were (i) apatinib [median inhibitory concentration (IC50) = 1 nM, 60 mg/kg S2221; Selleckchem] in 50% dimethyl sulfoxide (DMSO) in phosphate buffer (140 mM NaCl, 9 mM Na2HPO4, 1.3 mM NaH2PO4, pH 7.4) (, ); (ii) cabozantinib (IC50 = 0.035 nM, 30 mg/kg, S1119, Selleckchem) (); (iii) ZM323881 HCl (IC50 < 2 nM, 60 mg/kg, S2896, Selleckchem) in 2% DMSO, 30% PEG-300 (Sigma-Aldrich), 5% Tween 80 (Sigma-Aldrich), and 63% ddH2O (); and (iv) DC101 (40 mg/kg, BE0060, rat IgG1, Bio X Cell) and nonspecific rat IgG1 control antibody (40 mg/kg, BE0088, Bio X Cell) in PBS (pH 7.0) (, ). Intravital microscopy of the EDL muscle was performed 16 to 18 hours after reagent injection. Thick-section confocal microscopy for quantifying pillars and sprouts after in vivo blockade of VEGFR2 was performed on tissues fixed immediately thereafter, using 400 × 400 μm tiled fields of view, with 16-μm depth.

In vitro blockade of VEGFR2 signaling and flow response

HUVECs (CC-2519, Lonza) cultured to confluence were in serum-free EBM media (Lonza) for 16 hours and incubated with cabozantinib (300 nM in DMSO, 30 min), DMSO alone, DC101 (10 μg/ml, 4 hours), or isotype-matched control antibody. Cells were then subjected to either VEGFA (30 ng/ml; Sigma-Aldrich, for 5 min) or to laminar shear stress at 3 dyne/cm2 using Flexcell Streamer Fluid Shear Stress Device (Flexcell International). Five minutes after VEGFA stimulation and 15 min after the onset of flow, cells were fixed with 4% paraformaldehyde and immunostained for p-VEGFR2 1054/1059 (1:200). p-VEGFR2 signals were visualized with Alexa Fluor 546–conjugated donkey anti-rabbit IgG (1:400; Thermo Fisher Scientific). The intensity of p-VEGFR2 per cell was quantified using ImageJ.

Microfluidic analysis of endothelial cells subjected to ultralow flow

A microfluidic device with channels of 300 μm by 1000 μm by 1 cm (h, w, l) or 100 μm by 100 μm by 1 cm were fabricated from a micromachined mold via soft lithography using PDMS (SYLGARD184, Sigma-Aldrich). The patterned PDMS was then irreversibly sealed to glass-bottomed culture dish (MatTek 50 mm, 30-mm glass bottom) spin coated with PDMS (). Channels were infused with fibronectin (100 μg/ml; F1141, Sigma-Aldrich) in PBS using a programmable syringe pump (NE-300, New Era Pumps) for 1 hour and then left static at 37°C and 5% CO2 for another hour. Channels were then washed with microvascular endothelial growth media (EGM-MV) BulletKit (CC-3125, Lonza) containing l-ascorbic acid (0.1 μg/ml; A4034, Sigma-Aldrich) before infusing HUVECs. HUVECs were selected for study because of their embryologic/developmental status and the venous attributes of primordial vessel endothelial cells.

HUVECs stably expressing either RFP or GFP (cAP-0001RFP, cAP-0001GFP, Angio-Proteomie) were cultured in EGM-MV. Fluorescent HUVECs were transfected with 42 nM of either KDR/VEGFR2 siRNA or universal scrambled negative control siRNA (SR302557 and SR30004, respectively, Origene) with Lipofectamine RNAiMAX (13778075, Thermo Fisher Scientific) for 6 hours and allowed to recover overnight. The efficiency of VEGFR2 knockdown was confirmed by reverse transcription quantitative polymerase chain reaction (PCR) using SYBR green chemistry and a ViiA 7 Real-Time PCR System (Thermo Fisher Scientific). Primers (Origene) for human KDR gene were 5′-GGAACCTCACTATCCGCAGAGT-3′ and 3′-CCAAGTTCGTCTTTTCCTGGGC-5′.

GFP-HUVECs and RFP-HUVECs were mixed in equal amounts and introduced into fibronectin-coated PDMS microchannels at a bulk concentration of 3 × 106 cells/ml in EGM-MV containing l-ascorbic acid and 8% Dextran-500 (31392, Sigma-Aldrich). Cells were left under static fluid conditions at 37°C and 5% CO2 for 24 hours, which allowed them to reach confluence. Channels were then perfused with fresh media (MiniPuls3, Gilson pump) at low flow (116 μl/min) and then ultralow flow (33 μl/min) for 8 hours. These flow rates corresponded to a shear stress of 1.0 and 0.3 dyne/cm2, respectively, calculated using the Navier-Stokes and continuity equation–derived formula, τ = 6Qμ/(bh2), where τ is shear stress, Q is flow rate, μ is viscosity, and b and h are the width and height of the channel, respectively. The culture media viscosity value used was 0.78 cP ().

The effects of L-NAME (50 μM), ZM323881 (10 nM), and DETA NONOate (1 μM) on pillar formation were assessed with unlabeled HUVECs in 100 μm by 100 μm channels. Reagents and respective vehicle controls were delivered to the channels upon transitioning from a shear stress of 1.0 to 0.3 dyne/cm2 and evaluating pillar content after 24 hours.

Cell configurations were assessed after in situ fixation with 4% paraformaldehyde. Cells were permeabilized with 0.1% Triton X-100 in PBS, blocked with 5% normal donkey serum in PBS, and triple labeled for VE-cadherin, F-actin, and nuclei. Specifically, channels were infused with anti–VE-cadherin antibody (1:200; Clone D87F2, Cell Signaling Technologies) followed by Alexa Fluor 546–conjugated donkey anti-rabbit IgG (1:400; Thermo Fisher Scientific), as well as with Alexa Fluor 488–conjugated phalloidin and DAPI.

Confocal microscopy imaging and analysis of endothelial pillars in microfluidic channels

Endothelial cells in microfluidic channels were imaged with a Nikon A1R Confocal Laser Scanning System using a 40× oil immersion objective and 405-, 488-, and 561-nm lasers, generating up to 250 z-slices with a z-step size of 0.20 or 0.25 μm and x-y resolution of 310 nm per pixel using a resonance scanner, or up to 120 z-slices with z-steps of 0.40 or 0.50 μm and the same pixel resolution using a Galvano scanner. Z-slices were reconstructed into 3D volume through maximum intensity projections with NIS Elements software (Nikon). Pillar quantification was performed on 3D reconstruction of Z-slices of 14 to 20 volumes of view (320 μm × 320 μm × 60 μm, x-y-z) from each of four different channels. The proportion of endothelial cells that formed a pillar within a pillar-competent zone, defined as a 40 μm by 40 μm by 1 cm shoulder zone, was determined and expressed relative to the total endothelial cell number in that zone.

Statistics

Statistical analyses were performed using Prism 8 (GraphPad Software). All values passing D’Agostino and Pearson omnibus normality test are presented as means ± SE. Those not passing the normality test are presented as median and IQR, with 10th and 90th percentiles shown on graphs. Comparisons among normally distributed variables were made by t test or analysis of variance with Bonferonni’s post hoc test. Comparisons among non-normally distributed variables were made by Mann-Whitney test or Kruskall-Wallis test with Dunn’s post hoc test. Distributions of p-VEGFR2 signals on endothelial pillars were tested against a random distribution by chi-square test with Yates’ correction. All statistical tests were two sided, and significance was set at P < 0.05.