Research ArticleAPPLIED SCIENCES AND ENGINEERING

Extracellular matrix compression temporally regulates microvascular angiogenesis

See allHide authors and affiliations

Science Advances  21 Aug 2020:
Vol. 6, no. 34, eabb6351
DOI: 10.1126/sciadv.abb6351

Abstract

Mechanical cues influence tissue regeneration, and although vasculature is known to be mechanically sensitive, little is known about the effects of bulk extracellular matrix deformation on the nascent vessel networks found in healing tissues. Previously, we found that dynamic matrix compression in vivo potently regulated revascularization during bone tissue regeneration; however, whether matrix deformations directly regulate angiogenesis remained unknown. Here, we demonstrated that load initiation time, magnitude, and mode all regulate microvascular growth, as well as upstream angiogenic and mechanotransduction signaling pathways. Immediate load initiation inhibited angiogenesis and expression of early sprout tip cell selection genes, while delayed loading enhanced microvascular network formation and upstream signaling pathways. This research provides foundational understanding of how extracellular matrix mechanics regulate angiogenesis and has critical implications for clinical translation of new regenerative medicine therapies and physical rehabilitation strategies designed to enhance revascularization during tissue regeneration.

INTRODUCTION

Vasculature is an abundant and vital component of nearly all tissues, and revascularization is a critical step in the wound healing cascade. Angiogenesis is the primary mode of new vessel formation in wound healing. This process begins as new vessels sprout from adjacent intact blood vessels and then grow into the wounded area and begin to form microvascular networks within the first few days after an injury. Concurrent with early angiogenesis, fibroblasts deposit a collagenous provisional matrix, which appears characteristically granular due to its high density of newly formed capillaries. Granulation tissue is then remodeled into mature tissue through a complex series of chemical and physical cues, including the coordinated expression of growth factors and cytokines along with dynamically changing extracellular matrix (ECM) properties (1). Granulation tissue also experiences deformation forces depending on the type of tissue that has been injured; for example, bone experiences compression. Compressive forces regulate the progression of bone healing and can also alter the vascular networks within healing bone tissue (2, 3). However, whether angiogenesis is directly sensitive to compression or the modulatory effects are mediated by surrounding mechanosensitive cells such as osteoblasts remains unknown. Little is known about the effects of ECM deformation experienced by healing tissues on angiogenesis specifically, despite the fact that vasculature has long been recognized as mechanosensitive (4).

Angiogenesis is a highly coordinated process involving multiple phases, including sprout tip cell selection, in which a subset of endothelial cells become migratory; proteolytic ECM invasion; collective cell migration; and cell proliferation and recruitment. Many of these individual processes are known to be responsive to mechanical stimuli. For example, key molecular regulators of tip cell selection have recently been identified as mechanosensitive (5), and expression and secretion of proteases that are involved in angiogenesis are also mechanosensitive (6). Intracellular mechanotransducers, such as YAP and TAZ, also regulate a number of processes crucial to angiogenesis, including vascular endothelial growth factor (VEGF) signaling (7), cell migration (8), proliferation, and cell-cell junction formation (9). While many key components of angiogenesis have exhibited mechanosensitivity when investigated in isolation, little is known about how strain from ECM deformation regulates the highly coordinated processes of angiogenesis during wound healing.

Previous research from our laboratory has shown that compressive deformation of the ECM in vivo can have potent effects on neovascular growth—either inhibitory or stimulatory depending on the timing of load initiation. In a critical-size segmental bone defect, compressive loading of the defect was applied either immediately following the defect creation surgery, “early,” or 4 weeks after the surgery, “delayed.” When compression was applied early, concurrent with granulation tissue formation, angiogenesis and subsequent bone tissue formation were significantly inhibited. However, when compressive deformation was delayed and thus applied to newly mineralized, callus-like tissue, blood vessel growth and subsequent healing were enhanced (2), suggesting that ECM deformation regulates angiogenesis in a manner dependent on initiation time. However, the roles of other deformation parameters such as load frequency, magnitude, and mode (e.g., compression versus shear) are not well defined. Ambulatory loading typically occurs with a frequency around 1 Hz (10), and strain magnitude is an important regulator of bone formation. Peak bone tissue regeneration occurs at low strains, whereas a fibrous response occurs at high strains; 10% strain is thought to be an approximate transition point between regenerative and nonhealing responses (11). Furthermore, while compressive forces are the most widely studied in bone, ambulatory loading includes additional modes of loading such as shear, which has been shown to reduce revascularization of bone (3).

The complexity of the in vivo environment hinders comprehensive investigation of the regulatory role played by different loading parameters on the angiogenic healing environment. To exert more precise control over key factors in mechanical loading and matrix deformation and to isolate the vasculature from the complex regenerative environment, we used an in vitro three-dimensional (3D) model system of angiogenesis. Microvascular fragments are segments of mature vasculature and composed of multiple cell types, including endothelial cells, smooth muscle cells, mesenchymal stromal cells, and fibroblasts (12). The fragments are typically cultured in collagen-based hydrogels (13), which have similarities to the collagenous granulation tissue characteristic of early-stage wound healing. Temporal dynamics of microvascular fragment growth recapitulate the key stages of in vivo angiogenesis including sprout tip cell selection, matrix invasion, and neovessel elongation and branching. In addition, microvascular fragments are known to respond to mechanical cues, including ECM stiffness and tensile deformation (14, 15). Microvascular fragments cultured within collagen-based hydrogels represent a mechanically sensitive model system to recapitulate the 3D cell-matrix and cell-cell interactions critical to angiogenesis under precisely controlled mechanical loading parameters.

Here, we studied the effects of two load initiation times (early and delayed), three strain magnitudes (5, 10, and 30% strain), and two modes of compressive deformation (uniform compression and compressive indentation) on microvascular network growth. We hypothesized that vascularization would be enhanced by delayed, moderate compression and inhibited by early, high magnitude compression. Furthermore, we hypothesized that compressive indentation, which introduced greater shear stress, would inhibit angiogenesis. We found that neovascularization responded directly to dynamic matrix deformation strain magnitude and was particularly sensitive to the timing of load initiation. To elucidate the cellular and molecular changes upstream of the divergent microvascular network responses to early versus delayed loading, we investigated cell viability, cell proliferation, perivascular cell coverage, and gene expression profiles. Gene expression data particularly showed a divergent response to early versus delayed loading. This work provides a foundational understanding of how ECM mechanics regulate angiogenesis, with critical implications for synergizing regenerative therapies and rehabilitation strategies.

RESULTS

Nonloaded microvascular fragments progress through distinct stages of angiogenesis in vitro

The temporal progression of network formation by microvascular fragments cultured in decorin (DCN)–supplemented collagen hydrogels was first investigated in the absence of mechanical deformation to establish baseline time points. Under static culture conditions, microvascular fragments formed in vitro networks according to a predictable, repeatable time course. At day 0 (i.e., day of harvest), the freshly isolated fragments had characteristically rounded ends, denoted by white arrows in fig. S1. By day 3 in culture, the fragment ends adopted a pointed appearance, indicative of early sprouting and invasion of the ECM. By day 5, the initial sprouts extended and began to branch. Existing sprouts continued to elongate between days 5 and 7, and secondary branching began between days 7 and 10. Sprout and branch initiation processes occurred within the first 5 days of culture, while days 5 to 10 consisted primarily of elongation. Thus, days 0 to 5 were selected as “early loading” to coincide with initiation processes of angiogenesis and days 5 to 10 were selected as “delayed loading” to coincide with elongation processes (Fig. 1A).

Fig. 1 Quantification of microvascular network length and branching under 5, 10, and 30% strain loading.

(A) Microvascular fragments were cultured in DCN-supplemented hydrogels. Gels were loaded continuously for either the first 5 days of culture (early loading) or the final 5 days of culture (delayed loading). Gels were loaded at either 5, 10, or 30% strain. (B) Gels were loaded in compression with platens having a diameter greater than that of the gel or in compressive indentation by platens with a diameter less than that of the gel that also introduced greater shear stress. (C to H) Quantification of microvascular network length and branching based on 3D confocal z stacks of 200-μm depth. *, significant difference from nonloaded; one-way ANOVA, *P < 0.05, **P < 0.01, and ***P < 0.001; #, two-way ANOVA with Bonferroni post hoc test; @, overall effect of loading type; ^, overall effect of time; post hoc, #P < 0.05, ##P < 0.01, and ###P < 0.001. n = 6 per group.

Microvascular network formation exhibits sensitivity to magnitude and mode of dynamic loading

In our previous in vivo segmental defect study, early ambulatory loading disrupted vascularization within regenerating bone, whereas delayed ambulatory loading enhanced vascular network formation (2); however, the specific parameters of loading that regulate angiogenesis are poorly defined. Informed by bone regeneration literature, we identified 5% strain as proregenerative, 10% as transitional, and 30% strain as inhibitory to healing (11). These loads (5, 10, and 30% strain) were applied to microvascular fragment–seeded hydrogels with a triangle wave at a frequency of 1 Hz, corresponding to typical gait frequency (10). Gels were loaded in compression with one of two platen configurations: platens having a diameter greater than that of the gel to generate an initially homogeneous deformation approximating uniaxial compression (denoted as “compression”) or by platens with a diameter less than that of the gel (denoted as “indentation”; Fig. 1B), to introduce a more heterogeneous deformation with increased shear strain. Loading was continuously applied either early, days 0 to 5 of culture, or was delayed until days 5 to 10 of culture (Fig. 1A). Dynamic loading experiments included the following groups (n = 6 per group) at 5, 10, and 30% strain: early compression, early indentation, delayed compression, delayed indentation, and a nonloaded control.

Computational modeling revealed that stress and strain increased with the level of compression and that the range of stress and strain distributions depended on both the level of compression and the mode of loading. Compressive and shear stress values were lowly dispersed and did not differ with strain or mode of loading for 5 and 10% applied displacement (Fig. 2, A and B, and fig. S2). Application of 30% displacement greatly increased the compressive and shear stress for both modes of loading. The magnitude and dispersion of the compressive and shear strain increased with displacement for both modes of loading (Fig. 2, C and D, and fig. S2). Smoothly distributed stress and strain gradients in the vertical and radial directions are apparent at 30% indentation. Fluid flux was dependent on both the mode and magnitude of loading (fig. S3, A to C). Fluid flux differed between compression and indentation by orders of magnitude. Fluid flux generally decreased with increasing levels of compression but increased with increasing levels of indentation. This is likely because compression prevents fluid resorption to a greater degree than indentation.

Fig. 2 Simulation of DCN-supplemented collagen hydrogels subjected to 5, 10, and 30% strain loading.

Spatial distributions of compressive and shear stress and strain are plotted for compression and indentation to show (A) maximum compressive stress (third principal; Pa/mm), (B) maximum shear stress (octahedral; Pa/mm), (C) maximum compressive strain (third principal Lagrange strain; 1/mm), and (D) maximum shear strain (octahedral Lagrange strain; 1/mm). Cross-sectional visualizations of collagen hydrogels (middle, right panels) were acquired from simulations of 30% applied displacement at full depression during dynamic equilibrium. Note that applied strains are engineering strains, while the strains presented in the figure are Green-Lagrange strains.

At 5% strain, delayed loading led to significantly greater total network length [two-way analysis of variance (ANOVA), Bonferroni post hoc, P < 0.05] and number of branches (post hoc, P < 0.01) compared to early loading (Fig. 1, C and D). There was no effect of loading mode at 5% strain (i.e., compression versus indentation) and no significant interaction effect. Delayed 5% loading, both compression and indentation, increased length (one-way ANOVA, Bonferroni post hoc, P < 0.01 and P < 0.001, respectively) and branching (post hoc, P < 0.05 and P < 0.01, respectively) relative to the nonloaded control. Early 5% loading was not significantly different than the nonloaded control.

At 10% strain, delayed loading again significantly increased total network length (two-way ANOVA, overall effect, P < 0.05) and number of branches (overall effect, P < 0.05) relative to early loading (Fig. 1, E and F). At 10% strain, indentation also increased length (overall effect, P < 0.05) and branching (overall effect, P < 0.05) relative to compression. There was no significant interaction effect. Delayed indentation loading increased total length (one-way ANOVA, Bonferroni post hoc, P < 0.01) and number of branches (post hoc, P < 0.01) compared to the nonloaded control. Early 10% loading was no different than the nonloaded control.

At 30% strain, delayed loading significantly increased total network length and number of branches compared to early loading, which was especially pronounced in the indentation groups (two-way ANOVA, Bonferroni post hoc, P < 0.001; Fig. 1, G and H). Delayed indentation increased length (post hoc, P < 0.01) and branching (post hoc, P < 0.01) relative to delayed compression. There was no significant interaction effect. Furthermore, delayed indentation significantly increased length (Kruskal-Wallis, Dunn’s post hoc, P < 0.01) and branching (post hoc, P < 0.01) over the nonloaded control. At 30% strain, early loading, both compression and indentation, decreased the total network length relative to the nonloaded control (Kruskal Wallis, Dunn’s post hoc, P < 0.01 and P < 0.05, respectively). Early compression also decreased branching as compared to the nonloaded control (post hoc, P < 0.05). Qualitatively, the early loaded constructs at day 10 primarily exhibited early-stage sprouts (Fig. 3A) more comparable to the nonloaded sprouting observed at day 3 (fig. S1). Furthermore, as the strain increased for delayed loading, indentation led to an increase in network length with respect to compression (Fig. 1 and fig. S4). At 30% applied strain, the simulations demonstrated that indentation introduced a heterogeneous distribution of stresses and strains, with the median stresses from indentation greater than those induced by compression (Fig. 2). Seventy-five percent of measured strain values for indentation were less than those from compression. Furthermore, fluid flux increased with indentation but decreased with compression, affecting the resulting stress and strain and likely the convective transport of solutes and cytokines (fig. S3, A to C). Last, the fluid phase experienced greater stress compared to the biphasic mixture during indentation compared to compression. All of these differences may contribute to the improved microvessel growth under 30% delayed indentation but not compression.

Fig. 3 Representative images of nonloaded microvascular networks and networks formed under early versus delayed 30% strain loading.

(A) At 30% strain, early compression and early indentation groups exhibit only very early stage sprouts (white arrows), whereas delayed indentation appears qualitatively more densely vascularized than the nonloaded or delayed compression groups. Maximum intensity z projections (200-μm depth) of samples stained with GS-1 lectin at day 10 of culture. Scale bars, 500 μm. (B) Microvascular fragments retained perivascular coverage under both early and delayed 30% indentation. White arrows denote tips of fragments, which exhibit sprouting endothelial filopodia under early nonloaded, delayed nonloaded, and delayed 30% strain conditions. Early 30% strain leads to primarily rounded, nonsprouting ends of fragments. Representative maximum intensity z projections (25-μm depth) of DAPI (blue; nuclei)–, αSMA (green; perivascular smooth muscle cells)–, and isolectin B4 (red; endothelial cells)–stained microvascular fragments at days 3 (early) and 7 (delayed). Scale bars, 100 μm.

When all loaded groups’ length and branching were normalized to that of their respective nonloaded experimental controls, early loading exhibited significant strain magnitude dependence (fig. S4). Thirty percent compression decreased network length and branching relative to both 5% (one-way ANOVA, Bonferroni post hoc, P < 0.01) and 10% compression (post hoc, P < 0.05). Thirty percent indentation decreased network length and branching relative to 10% indentation (post hoc, P < 0.05). There were no statistically significant differences due to strain magnitude among delayed loading conditions. Together, these data demonstrated that delayed loading led to longer, more extensively branched microvascular networks than early loading over a wide range of strain magnitudes.

Time of dynamic loading initiation differentially affects proliferation but not viability or perivascular cell attachment

Many cellular-level responses may drive the observed changes in microvascular network morphology, including viability, proliferation, and cell-cell attachments. To more deeply investigate the divergent responses to early versus delayed loading, we selected early and delayed 30% strain indentation, the loading parameters that led to the greatest network morphology differences, for subsequent analyses of perivascular cell attachment, cell viability, and proliferation.

As a measure of vascular integrity, the spatial relationship between α smooth muscle actin (αSMA)+ perivascular cells and vessel endothelial cells was assessed at various time points. There were no significant differences in perivascular coverage of endothelial cells due to either early or delayed loading at any time point, and perivascular coverage was approximately 75 to 80% from day 0 to day 10 (fig. S5C). Qualitatively, there was relatively less perivascular coverage at the ends of nascent sprouts across all groups (Fig. 3B). Under nonloaded conditions, high-magnification images of early (day 3) sprouts showed forked endothelial cell extensions from both ends of the fragment. In contrast, most of the early loaded fragment tips remain rounded—a characteristic of freshly isolated, nonsprouting fragments (fig. S1). The morphological differences in sprout tip cells are less pronounced at the delayed time point. Both nonloaded and delayed indentation loaded samples had more mature and elongated sprouts that extended away from the parent microvascular fragment and established perivascular coverage. In addition, the delayed loaded samples also showed endothelial cell filopodia sprouting along the length of the vessel.

To assess the effects of dynamic loading on cell viability, a live/dead stain was performed on nonloaded and loaded samples on days 3 (early loading) and 7 (delayed loading) of culture (fig. S5A). There was no effect of loading on viability, and viability was greater at day 7 of culture than at day 3 for both nonloaded (two-way ANOVA, Bonferroni post hoc, P < 0.01) and loaded samples (post hoc, P < 0.01; fig. S5B). Viability was approximately 75% at day 3 and 90% at day 7.

Proliferation was measured by cellular 5-ethynyl-2′-deoxyuridine (EdU) incorporation also at days 3 and 7 (Fig. 4). There was no effect of early loading on proliferation, and approximately 5 to 10% of cells were proliferative at day 3. Proliferation was greater at day 7 (two-way ANOVA, overall effect, P < 0.001), and delayed loading led to increased proliferation compared to nonloaded controls (two-way ANOVA, Bonferroni post hoc, P < 0.05). Approximately 25% of cells were proliferative at day 7 under delayed loading, while approximately 15% of cells were proliferative at day 7 nonloaded. There was a significant disordinal interaction effect (P < 0.05), indicating that early loading and delayed loading have opposite effects on cellular proliferation (Fig. 4B). These data suggest that early and delayed loading differentially affect proliferation, with delayed loading having a stimulatory effect and early loading having a dampening effect. Costaining of the EdU incorporation samples with markers for perivascular support cells (αSMA) and for endothelial cells (isolectin B4) revealed proliferation of both endothelial and perivascular cells; however, a higher proportion of actively proliferating αSMA+ perivascular cells than endothelial cells were observed in both loaded and nonloaded samples (Fig. 4C). The proliferation results support the observed changes in length and branching; however, the underlying molecular mechanisms for these changes remained unclear.

Fig. 4 Effect of early versus delayed 30% strain loading on cell proliferation within microvascular networks.

Delayed loading increased the number of proliferating cells, and a significant interaction effect suggests that early loading decreased cell proliferation. (A) Image-based quantification of proliferation. Two-way ANOVA; ^^^, overall effect of time; P < 0.001; Bonferroni post hoc effect of loading, **P < 0.01. Significant interaction effect, P < 0.05. n = 3 gels per group per time point. (B) Representative maximum intensity z projections (25-μm depth) of DAPI (blue; all cells)– and EdU (red; proliferating cells)–stained microvascular fragments at days 3 (early) and 7 (delayed). Scale bars, 250 μm. (C) Representative maximum intensity z projections (5-μm depth) of microvascular fragments at day 7 of culture in either nonloaded or delayed 30% indentation conditions stained with DAPI (blue; nuclei, all cells), EdU (gray; nuclei, proliferating cells), isolectin B4 (red; endothelial cells), and αSMA+ (green; perivascular cells). EdU+ nuclei predominantly colocalize with αSMA+ perivascular cells. Scale bars, 100 μm.

Time of dynamic loading initiation differentially regulates microvascular fragment gene expression

To simultaneously probe the response of multiple key angiogenic processes [e.g., sprout tip cell selection, matrix invasion and deposition, vessel (de)stabilization and growth, adhesion and cell migration, and cell recruitment] to loading, we used a high-throughput microfluidic reverse transcription polymerase chain reaction (RT-PCR) gene expression array (fig. S6). Changes in expression of genes related to inflammation, apoptosis, and mechanotransduction were also assessed. Individual genes and corresponding functional sets are shown in table S1; genes and sets were selected based on a literature survey. We focused on early and delayed 30% strain loading, the loading parameters that led to the greatest network morphology differences, and quantified gene expression after 24 hours of loading to evaluate the early molecular changes that ultimately lead to altered network morphology.

The dimensionality of gene expression data was reduced using partial least squares discriminant analysis (PLSDA) to construct gene expression profiles of nonloaded versus loaded microvasculature. For both the early and delayed time points, PLSDA generated a latent variable (LV1) that significantly separated nonloaded from loaded samples (one-way ANOVA on score on LV1, Bonferroni post hoc, P < 0.001; Fig. 5, A and C). LVs are composed of a weighted average of genes, and each individual gene’s relative contribution can be visualized with an LV loading plot. In general, expression levels of many genes across functional groups were down-regulated by early loading (negative values in Fig. 5C), while expression levels of many genes across functional groups were up-regulated by delayed loading (positive values in Fig. 5D). To analyze the effect of loading on functional gene groups, we performed principal components analysis (PCA) on each nonoverlapping functional gene set (e.g., sprout tip cell selection, matrix invasion, and deposition) and then assessed whether the values of principal component 1 significantly differed by group (nonloaded, compression, and indentation). We also analyzed the effect of loading on individual genes using one-way ANOVA.

Fig. 5 Effect of early versus delayed 30% strain loading on gene expression.

Gene expression array data were analyzed with PLSDA to reduce their dimensionality to an LV, which is composed of a weighted average of each individual gene. The gene expression profiles of loaded versus nonloaded microvasculature, which are represented by LV1, were significantly different for both early (A) and delayed (B) 30% strain loading. Plots of early loading LV1 (C) and delayed loading LV1 (D) demonstrate the weighted average relative contribution of each individual gene to the LV1. In both cases, genes with more positive values are more highly expressed by loaded microvascular networks, and genes with more negative values are more highly expressed by nonloaded microvascular networks. Individual genes are color-coded by angiogenic processes (e.g., sprout tip cell selection) they are known to be involved in. (E) When interleaved, the loading plots of early loading LV1 versus delayed loading LV1 reveal differential regulation of a number of genes. Genes strongly down-regulated by early loading (negative values in black bars) are often instead up-regulated by delayed loading (positive values in red bars), such as Mmp14, Timp3, and Cxcr4. n = 6 gels per group. Error bars represent mean ± SD of Monte Carlo subsampling without replacement.

When considered in aggregate, expression of gene sets known to be involved in sprout tip cell selection (one-way ANOVA on score on principal component 1, Bonferroni post hoc, P < 0.01) and matrix invasion and deposition (post hoc, P < 0.01) was significantly down-regulated by early loading (fig. S7). Genes associated with matrix invasion (e.g., proteases and protease inhibitors) were also down-regulated by early loading, whereas matrix deposition genes were up-regulated, as evidenced by the loading plot of principal component 1 (fig. S7D). These data suggest that a more quiescent sprout tip cell phenotype may be induced by early loading. Gene sets known to be involved in cell recruitment (post hoc, P < 0.05) and mechanosensation (overall P < 0.05) were up-regulated by early loading.

When considered at the individual gene level, Tie1, Mmp14, Timp3, Cxcr4, Cxcl12, Mmp9, and Itgav were all significantly down-regulated by early loading. Cyr61, Ctgf, Vegfa, and Fgf2 were all significantly up-regulated by early loading (fig. S8A). Tie1, an orphan receptor that regulates angiogenic sprouting through Angpt/Tie2 signaling, is expressed by active sprout tip cells and is strongly down-regulated in quiescent endothelial cells (16), suggesting that early loading shifts endothelial cells to a more quiescent state. Two other genes associated with tip cells were also down-regulated by early loading: Cxcr4 and Mmp14. Cxcr4 is expressed by activated tip cells (17) and may also play a role in sprout anastomosis (18). The ligand for Cxcr4, Cxcl12, or Sdf1 was also down-regulated by early loading. Mmp14 is expressed by tip cells that lead the invasion of surrounding matrix (18). Timp3 is able to inhibit all matrix metalloproteinases (MMPs) (19), and the fact that it was also down-regulated by early loading suggests that the homeostatic balance of MMP-Timp (tissue inhibitor of matrix metalloproteinase) activity may be perturbed by early loading. Although early loading was primarily characterized by a down-regulation of gene expression, Cyr61, Ctgf, and Vegfa were strongly up-regulated. While Vegfa is a necessary component of the sprouting process, it alone is not sufficient to induce sprouting; the balance of other factors, especially angiopoietins 1 and 2, is also a key determinant of whether angiogenesis will occur (20). The increase in Vegfa may be a compensatory response of microvascular fragments pushed into a more quiescent state by early loading. The overall increase in cell recruitment genes including Fgf2 may also reflect a compensatory response. Alternatively, there is evidence that endothelial cells can produce an antiangiogenic isoform of Vegfa (21). The two genes most strongly up-regulated in response to early loading were Ctgf and Cyr61, which are both canonical targets of the YAP mechanotransduction pathway (8).

When considered in aggregate, gene sets known to be involved in cell recruitment (ANOVA on principal component 1, Bonferroni post hoc, P < 0.001) and mechanosensation (post hoc, P < 0.01) were up-regulated by delayed loading (fig. S7). At the individual gene level, Flt1, Ctgf, Itgb1, Cxcr4, Timp3, and Tgfb1 were all significantly up-regulated by delayed loading, and Cxcl12 was significantly down-regulated by delayed loading (fig. S8B). The five genes most strongly up-regulated by delayed loading were Flt1 or Vegfr1, Ctgf, Itgb1, Cxcr4, and Timp3. By increasing expression of Vegfr1, delayed loading may increase the sensitivity of microvascular fragments to proangiogenic VEGF signaling. This is in contrast with early loading, in which Vegfa up-regulation was not accompanied by Vegfr1 or Vegfr2 up-regulation. Delayed loading led to strong up-regulation of both Cxcr4 and Timp3, which is also in contrast with early loading. Although Cxcr4 was differentially affected by early versus delayed loading, its ligand Cxcl12 or Sdf1a was strongly down-regulated by both loading scenarios. Although Cxcl12 can function as a potent cell recruitment molecule, it may play a different role in our system. Tgfb1 is also considered a potent cell recruitment signal and was up-regulated by delayed loading. Itgb1 is an adhesion molecule essential for angiogenesis (8).

Delayed compression and delayed indentation were also significantly separated along LV1 (one-way ANOVA, Bonferroni post hoc, P < 0.05; Fig. 5B). Of the 43 genes tested, only Itga2 expression was significantly higher in indentation relative to compression (one-way ANOVA, Bonferroni post hoc, P < 0.05) and to the nonloaded control (post hoc, P < 0.01), despite the large morphological differences observed at day 10.

A number of genes were differentially regulated by early versus delayed loading (Fig. 5E); the strongest down-regulated contributors to early loading LV1 were instead up-regulated by delayed loading, and the strongest up-regulated contributors to delayed loading LV1 were instead down-regulated by early loading (e.g., Tie1, Mmp14, Timp3, Flt1 or Vegfr1, Itgb1, and Cxcr4). There were two strongly up-regulated contributors to both early LV1 and delayed LV1, Ctgf and Cyr61, which are canonical targets of the YAP mechanotranduction signaling pathway (8).

YAP is involved in microvascular response to delayed loading

The genes with the largest induction by mechanical loading, regardless of loading mode, were Ctgf and Cyr61. Early loading led to a nearly threefold increase in Cyr61 expression for both compression and indentation, and delayed loading led to a 3.4- and 5-fold increase in Ctgf expression for compression and indentation, respectively. Ctgf and Cyr61 are both canonical target genes of the mechanosensitive transcriptional coactivators YAP and TAZ (8). To determine whether the mechanoactivation of Ctgf and Cyr61 was YAP/TAZ dependent, we used a pharmacological inhibitor of YAP, verteporfin (VP; 5 μM). VP has been shown to inhibit YAP/TAZ activity by disrupting YAP’s formation of a transcriptional complex with TEA domain (22) and by sequestering YAP in the cytoplasm, thus preventing nuclear translocation (23). Thus, we hypothesized that the addition of VP would abrogate the increased expression of target genes Cyr61 and Ctgf due to loading. Because the gene expression profiles of compression and indentation were nearly identical in our initial array, only indentation was studied with VP.

At the early time point, loading increased expression of Ctgf (overall effect, P < 0.01; Fig. 6), but there was no statistically significant effect of VP on early gene expression of either Ctgf or Cyr61. At the delayed time point, delayed loading without VP increased expression of both Ctgf (two-way ANOVA, Bonferroni post hoc, P < 0.001) and Cyr61 (post hoc, P < 0.001) relative to the nonloaded control. VP significantly abrogated the increased expression of Ctgf (post hoc, P < 0.05) and Cyr61 (post hoc, P < 0.001) induced by loading, suggesting that YAP mediates the mechanotransductive gene induction by delayed loading. A similar abrogation of Ctgf and Cyr61 expression by VP was not observed in early loaded samples despite their induction in the initial gene expression array.

Fig. 6 Expression of YAP/TAZ target genes Ctgf and Cyr61 due to 30% strain loading and to YAP/TAZ inhibition.

Early 30% indentation loading induced significant up-regulation of Ctgf, and delayed 30% indentation loading induced significant up-regulation of both Ctgf and Cyr61. YAP inhibitor, VP (5 μM), abrogated the delayed loading-induced up-regulation of both target genes. Expression is shown relative to mean expression of housekeeping genes. Two-way ANOVA, overall effect of loading, ##P < 0.01; Bonferroni post hoc, *P < 0.05, **P < 0.01, and ***P < 0.001. Delayed Ctgf demonstrated significant interaction effect. n = 5 to 6 per group.

DISCUSSION

In this set of studies, we demonstrated that dynamic ECM deformation profoundly influences microvascular network formation and that the initiation time, mode, and magnitude of compressive strain are all critical regulatory parameters. Across all conditions tested, delayed compressive loading (initiated at day 5 following initial sprout and branch initiation) enhanced vessel network formation compared to early loading (initiated at day 0). These morphological differences (i.e., longer, more extensively branched networks due to delayed versus early loading) were mirrored by increased cell proliferation in response to delayed loading and divergent regulation of genes associated with active angiogenic sprouts, where many of the same genes were down-regulated by early loading but up-regulated by delayed loading. Together, these data implicate the timing of load initiation as a critical determinant of vascular network formation and suggest that the early stages of angiogenic sprouting are exquisitely sensitive to the bulk ECM deformation that would be experienced by healing tissues.

Load magnitude is often implicated as the parameter of mechanoregulation that primarily dictates regenerative responses. However, our previous results have shown that the timing of load application, even loads of similar magnitudes, has a profound effect. Consistent with our hypothesis and with previous in vivo results (2), delayed loading produced greater network length and number of branches at all strain magnitudes tested. At low strains (5 and 10%), early loading did not affect vessel network formation; however, at a much higher strain (30%), early loading significantly decreased network length and branching. These data suggest that low magnitude strain may be permissive to early-stage angiogenesis, whereas higher magnitude strains can disrupt this process. Inhibitory effects of delayed loading were not observed, even at 30% strain, suggesting that once a critical stage of vascular network maturity is reached, even high magnitude loading is permissive to vessel growth. Day 5 was chosen as the inflection point of early versus delayed loading in the present studies to distinguish between the distinct processes of sprout and branch initiation versus elongation, suggesting that the initial formation of sprouts/branches may be this critical stage. This is also evidenced by the tip/sprout morphology shown in Fig. 3B. Future work may include extending the gene expression analysis by staining for proteins that are more highly expressed by endothelial tip cells than stalk cells, such as Dll4 (18).

In addition to load magnitude, previous work has also investigated the effects of different magnitudes of loading, and shear stress has been shown to be detrimental to neovascularization of regenerating bone tissue (3). Thus, we hypothesized here that increased shear stress in the indentation loading groups may directly inhibit angiogenesis. Contrary to our hypothesis, compressive indentation that also introduced greater stress increased vessel length and branching overall compared to compression alone at strains above 5%. Previous work that demonstrated a reduction in bone neovascularization due to shear focused on the effects of translational shear, i.e., perpendicular to the axis of compression (3), whereas in the current study, we used indentation platens that were designed to introduce a zone of interfacial shear. Although there were no differences observed in the microvascular networks at the interface of contact between the platen and gel and there was no difference in the spatial distribution of vessels, the indentation platens did introduce shear stresses between two and four times greater than those introduced by the compression platens (Fig. 2B). This may suggest that the microvascular growth was not sensitive to the local distribution of strains. However, the indentation platens also profoundly changed the local stress and strain gradients as well as fluid flow patterns and the share of pressure between the fluid and solid phases throughout the gels (fig. S3, A to C), all of which are known to also influence tissue-level remodeling. A previous study similarly found that new bone tissue formation can localize specifically at sites of high strain gradients (24). In contrast, there was virtually no strain gradient within the uniformly compressed gels (Fig. 2). The effects of strain gradients may also explain the differing effects of strain magnitude in uniform compression versus compressive indentation. Delayed compressive indentation increased vascular network length and branching at all three tested strain magnitudes, including the high 30% strain. In contrast, delayed uniform compression exerted a significant stimulatory effect at 5% strain, a positive significant overall effect but not statistically significant post hoc difference at 10% strain, and no effect at 30% strain. The trend observed in uniform compression was more consistent with our hypothesis and the bone regeneration literature focused on strain magnitude; lower strains tend to promote bone regeneration, while 30% compressive strain is sufficiently high that it is not often found under physiologic conditions and disrupts bone healing (11). Our work suggests a critical role for strain gradients in regulating angiogenic processes and motivates further investigation of the effect of strain gradients on enhancing vascularization. Future computational work may more precisely define these effects.

Although the delayed compression and delayed indentation environments produced differences in terms of vascular network length and branching, gene expression from the 43 genes tested in our array was similar between the conditions, with only one (Itga2) showing differential expression between the conditions (fig. S8). This may suggest a critical role for Itga2 in the mechanosensation of strain gradients and/or initiating the proangiogenic response to delayed loading. The lack of differences in gene expression profiles between these groups may be due to the sampling of a single time point after 24 hours of loading; gene expression and subsequent morphologies may diverge later between loading conditions. Future work may also include an unbiased analysis of additional genes using techniques such as RNA sequencing. Alternatively, the similar gene expression profiles may suggest that later (after 5 days of loading rather than 24 hours), differences may be primarily mediated by parameters that were not assessed in this system, such as convective fluid flow patterns, which are difficult to decouple from in vitro loading systems. The uniform compression platen generally reduced fluid flux, while the indentation platen generally increased fluid flux (fig. S3, A to C). Differences in fluid flux are partially responsible for the differences in stress, strain, and likely convective solute and cytokine transport, which may contribute to the enhanced microvascular growth under 30% delayed indentation as compared to compression. However, if all effects observed in these experiments were due to an increase in fluid flow and thus nutrient transport, early 30% loading would likely have also had a beneficial effect rather than the inhibitory effect that we observed.

Early versus delayed gene expression results mirrored the morphological results, with early loading resulting in decreased vascular network formation and decreased overall gene expression and delayed loading instead resulting in enhanced vascular network formation and increased gene expression. PCA analysis revealed that expression of the gene set known to be involved in sprout tip cell selection was significantly decreased due to early loading. This suggests that early-stage sprout tip cell selection, which is regulated by a balance of Notch interaction with proangiogenic Jagged-1 and antiangiogenic Dll4, is perturbed by loading. The Notch-Jagged signaling axis has recently been shown to be mechanically sensitive, including the down-regulation of proangiogenic Jag1 in response to mechanical stretch (5). In response to early loading, we observed not only a down-regulation of Jag1 but also an overall down-regulation of the sprout tip cell selection gene set, including antiangiogenic Dll4. We also observed an overall down-regulation of promigratory protease expression but an increase in matrix deposition genes. Together with the strong down-regulation of Tie1, characteristic of quiescent endothelial cells (16), our data suggest the induction of a more complex quiescent, nonsprouting phenotype in response to early loading. In contrast, when loading was delayed until after initial sprout tip cell selection has already occurred, we saw an overall up-regulation of gene expression, including many of the same genes that were instead down-regulated by early loading: Tie1, Mmp14, Timp3, Flt1 or Vegfr1, Itgb1, and Cxcr4. Of these genes, Tie1, Mmp14, Flt1 or Vegfr1, and Cxcr4, in particular, are known to be expressed by actively migrating sprout tip cells (16, 17). Together, these divergent effects of early versus delayed loading suggest that tip cell activity may be the mechanism through which loading affects neovascularization; early loading depresses sprout tip cell selection signaling, while delayed loading increases expression of genes associated with active tip cells and cell proliferation.

In our mechanically stimulated system, we probed the expression of multiple known elements of mechanotransduction in angiogenesis, including cell-matrix coupling integrins, cell-cell coupling Notch and Jagged (5), and transcriptional targets of intracellular mechanotransducers YAP and TAZ (8). As discussed above, we observed gene expression changes in each, and each one represents a potential avenue of future work. Cyclic strain has been shown to regulate canonical Notch-Jagged signaling (5, 25). In endothelial cells, the Notch1 receptor has been shown to mediate increased angiogenic network formation due to cyclic strain (25), and in vascular smooth muscle cells, expression of the proangiogenic ligand Jag1 decreases with increasing strain (5). Here, we observed a down-regulation of Notch1, proangiogenic Jag1, and antiangiogenic Dll4 due to early loading and an up-regulation of both Jag1 and Dll4 due to delayed loading. As adhesion molecules, integrins play an important role in transducing forces from the ECM to the cellular cytoskeleton (26). We observed a broad up-regulation of integrins due to delayed loading, especially Itgb1, and down-regulation of integrins due to early loading. The one molecule that was differentially expressed by delayed compression versus indentation was also an integrin, Itga2. The a2 and b1 integrin subunits form a complex that can bind both collagen (27) and DCN (28), the two components of our gel matrices. In our system, integrins a2 and b1 may couple cells to the mechanically dynamic matrix and thus act as an element of the mechanotrasduction pathway. However, the largest and most notable changes in expression that we observed due to mechanical loading, regardless of loading time or mode, were Ctgf and Cyr61—two canonical targets of the YAP/TAZ signaling pathway (8) and known regulators of angiogenesis (29). Both integrins and the Notch signaling pathway can interact with the YAP/TAZ signaling pathway (8, 30). Thus, as a first step toward unraveling the molecular mechanism truly driving the response to loading, we chose to follow up on YAP/TAZ signaling.

YAP and TAZ have recently emerged as vascular mechanotransducers of oscillatory shear stress (31)—one of the most well-studied examples of vascular mechanosensitivity (4). In addition, YAP and TAZ are known to promote sprouting angiogenesis, even in systems that are not directly mechanically stimulated (9), through molecular regulators of sprout tip cell selection (30). In our studies, we observed a strong up-regulation of YAP/TAZ target genes Cyr61 and Ctgf due to loading; this response was significantly abrogated by YAP/TAZ inhibitor VP only under the delayed loading condition. Although we also observed significant up-regulation of Cyr61 and Ctgf in the initial gene expression array due to both early and delayed loading, we did not observe the same significant up-regulation due to early loading in the VP experiment. These samples contained dimethyl sulfoxide (DMSO) to appropriately control for the delivery of VP within DMSO, and DMSO may have altered the baseline response of microvascular fragments to early loading. DMSO has been previously reported to have antiangiogenic effects (32), which also prevented functional analysis of microvascular network formation in the presence of VP; the authors acknowledge this as a limitation of the present studies. These results suggest a potential role for YAP signaling. The regulatory role played by YAP/TAZ may be determined by other factors concurrently affected by early versus delayed mechanical loading, and the role of YAP/TAZ in regulating divergent responses requires future work. Itgb1, which was up-regulated by delayed loading and is known to be involved in YAP/TAZ signaling (8), may transduce ECM deformation forces into the cell, activate YAP/TAZ transcriptional coactivation of target genes including Cyr61 and Ctgf, and, in turn, lead to an increased activity of sprout tip cells (30) and increased proliferation of smooth muscle cells (33) to enhance vascularization. In future work, we will assess the effect of YAP/TAZ inhibition of the vascular morphology in response to loading to definitively establish a causal link between up-regulated YAP/TAZ target gene expression levels and the altered morphology in response to loading and continue to explore the full molecular pathway.

Here, we used a simplified in vitro model system to isolate the effects of mechanical ECM deformation on angiogenesis. While our system preserves the 3D cell-cell and cell-matrix interactions of multiple vascular cell types, any in vitro system inherently lacks the complexity of true in vivo healing. For example, the ECM contains multiple different protein constituents, and in the bone healing environment that initially motivated these studies, the stiffness of the ECM surrounding the vasculature increases with time as healing progresses. Delayed in vivo loading applied to stiffer, more callus-like tissue enhanced vascularization (2), despite the fact that a stiffer ECM has been shown to hinder vascularization (14). Now that we have decoupled these two factors in vitro, future computational and in vivo work can begin to understand their combinatorial effects. In addition, the early-stage healing environment contains a myriad of inflammatory cytokines and cell types. Inflammation and angiogenesis often occur contemporaneously and have known cross-talk, including macrophages mediating vessel anastomosis (18) and cytokines such as interleukins modulating both pro- and antiangiogenic responses (34). Future work in more complex in vitro systems and ultimately in vivo will build upon the foundational results presented here and provide additional insight.

Here, we have demonstrated that vasculature is directly sensitive to ECM deformation forces, independent of tissue-specific cells such as osteoblasts. Furthermore, the magnitude, mode, and initiation time of ECM loading are all critical regulators of angiogenesis. Across all tested magnitudes and modes, delayed loading enhanced vessel network formation relative to early loading. Morphological differences were mirrored by increased cell proliferation, especially by αSMA+ perivascular cells, in response to delayed loading and divergent regulation (down-regulated by early loading and up-regulated by delayed loading) of genes associated with active angiogenic sprouts. Together, these data implicate time of load initiation as a critical determinant of vascular network morphology and suggest that therapeutic loading should be delayed until after initial angiogenic sprouting can occur. While we were initially motivated by bone tissue regeneration, a number of other tissues also experience ECM deformation forces; for example, ligaments and tendons undergo tension, venous ulcers are often treated with compression bandages, and even cutaneous wounds experience tension during closure. By providing increased foundational understanding of the time-dependent mechanical regulation angiogenesis, this work begins to enable mechanical loading to be leveraged as a therapeutic component of future tissue engineering and physical rehabilitation approaches.

MATERIALS AND METHODS

Microvascular fragment isolation and culture

Microvascular fragments were isolated as previously described (13) and in compliance with the Georgia Institute of Technology Institutional Animal Care and Use Committee. Briefly, epididymal fat pads of retired breeder Lewis rats were harvested, minced, and digested in a collagenase solution. Microvascular fragments were obtained through selective filtration to retain multicellular structures between 20 and 200 μm. The fragments were suspended at a density of 20,000 fragments/ml in 3% collagen gels supplemented with DCN (50 μg/ml) to improve construct dimensional stability (35). Gels were formed by 15 to 20 min of incubation at 37°C in custom polycarbonate molds to create gels with a diameter of 5 mm and a height 4 mm. Microvascular fragment–containing gels were cultured in serum-free medium supplemented with recombinant human VEGF (10 ng/ml; R&D Systems, Minneapolis, MN) (15). Medium was changed on days 3, 5, and 7 of culture, and gels were fixed with 4% paraformaldehyde on day 10.

Separate batches of microvascular fragments were isolated for each of the 5, 10, and 30% strain vascular network quantification experiments (Fig. 1), for the staining-based analyses (viability, proliferation, and perivascular coverage; Figs. 3B and 4), for the gene expression analysis (Fig. 5), and for the YAP/TAZ inhibition study (Fig. 6).

Dynamic loading

Microvascular fragment–containing gels were loaded using ElectroForce 5500 with a multispecimen compression chamber containing a 24-well plate loading assembly (TA Instruments, New Castle, DE). Loading was applied in a triangle wave with amplitudes corresponding to 5, 10, or 30% strain (0.2, 0.4, and 1.2 mm, respectively) at a frequency of 1 Hz. Gels were loaded in homogeneous compression using polyetheretherketone (PEEK) platens with a diameter greater than that of the gel (1 cm) or in heterogeneous compression with an interface zone that produced a larger and more heterogeneous distribution of shear strain (indentation) using PEEK platens with a 3-mm gel-contacting diameter. Loading was applied continuously, breaking only for medium changes, for either the first 5 days of culture, early loading, or the final 5 days of culture, delayed loading. To ensure that gels remained centered within the well, gels sat within the inner diameter of 1-mm-thick 3D printed poly(lactic-co-glycolic acid) rings during loading.

To study the effects of loading mode and time of initiation, dynamic loading at 5, 10, and 30% strain experiments included the following groups (n = 6 per group): early compression, early indentation, delayed compression, delayed indentation, and a nonloaded control.

Computational simulation of dynamic loading

Finite element simulations were carried out to examine the time-dependent stress and strain distributions during compressive and indentation loading of the hydrogels. The hydrogel geometry for the models was based directly on the experimental dimensions of the platens, indenter, and the molds that produced the hydrogels. The hydrogels were represented as anisotropic and biphasic (poroelastic) materials under time-varying finite deformation. The constitutive model for the solid phase of the biphasic material consisted of a neo-Hookean ground matrix reinforced by an ellipsoidal fiber distribution (fig. S3, D and E) (35, 36). To obtain material coefficients for the constitutive model, stress relaxation data of DCN-supplemented collagen hydrogels from a previous publication were used (35). Permeability and fibril modulus were determined via a constrained nonlinear least squares method, using the parameter optimization module of FEBio during a simulation of stress relaxation to provide the function evaluations (37). The solid volume fraction was approximated based on the effective specific volume of collagen at 3 mg/ml (38). To calculate stress and strain distributions of dynamically loaded DCN-supplemented collagen hydrogels, a 90° wedge geometry was meshed with radial biasing away from the center and vertical biasing away from the contact surface to better accommodate high strains and capture pressure gradients in the poroelastic material. The geometry was divided into 10 circumferential divisions providing 9° resolution for each element. Symmetry boundary conditions were enforced by fixing nodes along the x axis in the y direction and nodes along the y axis in the x direction, restricting deformation to lateral directions. Free draining surfaces were modeled by prescribing zero fluid pressure at the radial edge of the wedge and on the exposed portion of the top surface for the case of compressive indentation. Nodes along the bottom of the gel were fixed in the vertical direction (fig. S3E and table S2). Gels were deformed using rigid body contact to peak strains of 5, 10, or 30% strain. To achieve convergence in the presence of high strain rates, dynamic loading was ramped with displacement prescribed at 2/3 of peak strain at 0.2 Hz for 25 cycles, 9/10 of peak strain at 0.2 Hz for 25 cycles, and finally peak strain at 1 Hz until less than a 0.1% change in the third principal stress and shear stress was achieved between cycles for all models. The volume average for stress and shear was determined at full depression during the last cycle by weighting the elemental values by the current volume of each element.

Staining, imaging, and image-based analyses

To assess network morphology at day 10, fixed gels were stained with rhodamine-labeled Griffonia simplicifolia (GS-1) lectin (Vector Labs, Burlingame, CA) at a concentration of 5 μg/ml in phosphate-buffered saline overnight at 4°C (n = 6 per group). Gels were imaged using a Zeiss 700 confocal microscope with a 5× objective. The entire diameter of each gel was imaged to a depth of 200 μm from the loaded surface. Confocal z stacks were median-filtered, deconvolved, and thresholded using Amira for Life Sciences (Thermo Fisher Scientific, Waltham, MA). Islands smaller than 30 voxels (e.g., single cells, debris, or noise) were also removed using Amira. Thresholded images were exported for skeletonization and quantification of length and branching using the 4-D open snake method of the Farsight Toolkit (39).

Viability and proliferation of nonloaded versus loaded microvascular fragments were measured at day 3 for early loading and at day 7 for delayed loading. Viability was determined using a live/dead assay kit performed according to the manufacturer’s instructions (Thermo Fisher Scientific). Gels were imaged at ×10, and three randomly selected fields within 200 μm of the loaded surface were imaged per gel to a z stack depth of 25 μm, which is the approximate depth of a single vessel. Maximum intensity z projections were created to quantify viability using Fiji’s Analyze Particles feature. Percent viability was calculated as the pixel area of live cells (green channel) over the pixel area of all cells (green + red channels). Proliferation was assessed using the Click-iT EdU Alexa Fluor 594 Imaging Kit (Thermo Fisher Scientific; n = 3 per group per time point). EdU was added to medium at a concentration of 10 μM at days 2 and 6 of culture and incubated for 24 hours before gels were fixed at days 3 and 7. Gels were imaged at ×10 to a depth of 25 μm, and four images were taken to capture the diameter of each gel. Proliferation was quantified as the number of EdU+ nuclei over the total number of nuclei.

To assess the degree of perivascular coverage of endothelial cells by smooth muscle cells over time, nonloaded gels were fixed at days 0, 3, 5, 7, and 10 (n = 3 per time point), early loaded samples were analyzed at days 3, 5, and 10 (n = 3 per time point), and delayed loaded samples were analyzed at days 7 and 10 (n = 3 per time point). Following fixation, gels were stained with Alexa Fluor 488–conjugated anti-αSMA antibody (ab184675, Abcam, Cambridge, UK) at a 1:100 dilution, DyLight 649–conjugated GS-1 isolectin B4 (DL-1208, Vector Labs) at 5 μg/ml, and 4′,6-diamidino-2-phenylindole (DAPI; Thermo Fisher Scientific) at a 1:1000 dilution. Gels were imaged at ×40 to a depth of 25 μm. A region of interest (ROI) was drawn around vascular structures, and signal overlap between αSMA and isolectin B4 was quantified for the ROI using Manders coefficients as determined by the Fiji plugin coloc2 (40).

Gene expression analyses

To assess gene expression changes due to loading, RNA was harvested from nonloaded and loaded constructs 24 hours after load initiation (i.e., after 24 hours of culture total for early loaded samples and their nonloaded controls and after 6 days of culture total for delayed loaded samples and their nonloaded controls). RNA was collected from nonloaded and loaded gels at both early and delayed time points (n = 5 to 6 per group per time point). RNA was extracted using Qiagen MinElute kits, and complementary DNA (cDNA) was made using Qiagen RT2 First Strand kits (Qiagen, Hilden, Germany). RNA concentration as determined by NanoDrop spectrophotometer (Thermo Fisher Scientific) was used to ensure that cDNA concentrations were equivalent. TaqMan probes were used to assess gene expression of 43 genes known to be involved in various stages of angiogenesis and five housekeeping genes (table S1). Gene expression was quantified using a Biomark real-time PCR integrated fluidic circuit array (Fluidigm, South San Francisco, CA). Rat universal cDNA (Gene Scientific, Rockville, MD) and ultrapure water (Thermo Fisher Scientific) were used as positive and negative controls, respectively. Data were normalized on a per-sample basis to the mean of three housekeeping genes that did not have significantly different levels of expression across groups (Gapdh, Ubc, and Hrpt1) using the ∆Ct method.

Multivariate analysis of gene expression data

PLSDA was performed in MATLAB (MathWorks, Natick, MA) using Cleiton Nunes’s partial least squares algorithm (MathWorks File Exchange). To avoid biasing results with the absolute magnitude of different genes’ expression levels, data were z-scored before being analyzed with PLSDA. Orthogonal rotations were applied to the z scores to maximally separate groups (nonloaded, compressive loading, and indentation loading) based on LV1 and LV2 created by the PLS algorithm. LV loading plots show the mean and SD of each gene’s relative contribution to the LV; mean and SD were calculated using Monte Carlo subsampling that iteratively excluded a randomly chosen 15% subset of the data 1000 times (41).

PCA was performed on nonoverlapping gene sets known to be involved in sprout tip cell selection, matrix invasion and deposition, vessel (de)stabilization and growth, adhesion and cell migration, cell recruitment, inflammation and apoptosis, and mechanotransduction using the MATLAB pca command. Loading plots for principal component 1, which explains the greatest amount of variation in the data, represent the relative contributions of each gene within the set.

YAP inhibition

The YAP inhibitor VP (MilliporeSigma, Burlington, MA) was dissolved in DMSO and added to serum-free medium at a concentration of 5 μM. Non-VP controls received an equal volume of DMSO. VP and DMSO were added to microvascular fragment–containing gels 30 min before initiation of loading to allow diffusion throughout the gel. After 24 hours of loading, RNA was collected from early and delayed samples as above. The groups for the YAP inhibition study were indentation + VP, indentation + DMSO carrier only, nonloaded + VP, and nonloaded + DMSO at both early and delayed time points (n = 5 to 6 per group). All culture of samples containing light-sensitive VP (22) was conducted in the dark.

Statistical analysis

For day 10 microvascular network morphology data, a two-way ANOVA was used to compare early versus delayed loading and compression versus indentation loading. A one-way ANOVA was used to compare early loading to the nonloaded control and delayed loading to the nonloaded control; a Kruskal-Wallis test with Dunn’s post hoc test was used in cases where the variances significantly differed among groups. To directly compare the effects of different load magnitudes, loaded groups were normalized to their respective nonloaded controls and analyzed within time point and loading mode (e.g., early indentation compared at 5, 10, and 30% strain) using a one-way ANOVA. Viability, proliferation, and perivascular coverage data were analyzed by two-way ANOVA. Gene array data were studied in aggregate using PLSDA as detailed above, and statistical significance was determined using a one-way ANOVA on each group’s mean score on LV1. The expression levels of individual genes were compared within time point using a one-way ANOVA. The effect of loading on entire gene sets was assessed with PCA as detailed above, and statistical significance was determined using a one-way ANOVA on each group’s mean score on principal component 1. The effect of VP on gene expression of nonloaded versus loaded constructs was analyzed with a two-way ANOVA. Bonferroni’s post hoc test followed all ANOVAs. All statistical analyses were performed in GraphPad Prism 5 with α = 0.05. Data are presented as mean ± SEM.

SUPPLEMENTARY MATERIALS

Supplementary material for this article is available at http://advances.sciencemag.org/cgi/content/full/6/34/eabb6351/DC1

https://creativecommons.org/licenses/by-nc/4.0/

This is an open-access article distributed under the terms of the Creative Commons Attribution-NonCommercial license, which permits use, distribution, and reproduction in any medium, so long as the resultant use is not for commercial advantage and provided the original work is properly cited.

REFERENCES AND NOTES

Acknowledgments: We wish to acknowledge the core facilities at the Parker H. Petit Institute for Bioengineering and Bioscience at the Georgia Institute of Technology for the use of their shared equipment, services, and expertise. Funding: This work was supported by funding from the NIH (grant R01 AR069297). This material is the result of work supported with resources and the use of facilities at the Atlanta VA Medical Center along with funding from the VA (grant 5 I01 RX001985); the contents do not represent the views of the U.S. Department of Veterans Affairs or the U.S. government. Author contributions: M.A.R. and E.A.E. performed in vitro experiments and analyzed in vitro data. S.A.L. performed computational simulations and analysis with guidance from J.A.W. L.B.W. guided and performed multivariate gene expression data analysis with M.A.R. M.A.R., L.K., J.A.W., J.D.B., R.E.G., and N.J.W. designed the experiments. All authors edited and approved the final manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data related to this paper may be requested from the authors.
View Abstract

Stay Connected to Science Advances

Navigate This Article