Interstitial Flow Promotes the Formation of Functional Microvascular Networks In Vitro through Upregulation of Matrix Metalloproteinase‐2

Self‐organized microvascular networks (MVNs) have become key to the development of many microphysiological models. However, the self‐organizing nature of this process combined with variations between types or batches of endothelial cells (ECs) often lead to inconsistency or failure to form functional MVNs. Because interstitial flow (IF) has been reported to play a beneficial role in angiogenesis, vasculogenesis, and 3D capillary morphogenesis, the role IF plays during neovessel formation in a customized single‐channel microfluidic chip for which IF has been fully characterized is systematically investigated. Compared to static conditions, MVNs formed under IF have higher vessel density and diameters and greater network perfusability. Through a series of inhibitory experiments, this study demonstrates that IF treatment improves vasculogenesis by ECs through upregulation of matrix metalloproteinase‐2 (MMP‐2). This study then successfully implements a novel strategy involving the interplay between IF and MMP‐2 inhibitor to regulate morphological parameters of the self‐organized MVNs, with vascular permeability and perfusability well maintained. The revealed mechanism and proposed methodology are further validated with a brain MVN model. The findings and methods have the potential to be widely utilized to boost the development of various organotypic MVNs and can be incorporated into related bioengineering applications where perfusable vasculature is desired.


Introduction
Recent decades have witnessed an explo sive growth in the number and variety of engineered microvascular networks (MVNs). Functional vasculature is critical for the development of many microphysio logical models and for diverse applica tions in tissue engineering. Although a vasculature plays many important roles beyond just serving as conduits, its major function is to transport nutrients, oxygen, metabolic wastes, and various cells. Lacking a perfusable vasculature hinders the advancement of numerous key tech niques in bioengineering. One particular example is organoid technology. Without a perfusable vasculature, organoids in 3D culture rely solely on passive diffusion to exchange nutrients, oxygen and metabolic waste, and gradually leads to a necrotic core. [1] Tremendous efforts have been made to construct functional vasculature in vitro within various hydrogels with associated microfluidic chips. Many of those were designed following a selforganizing approach, which draws upon natural selforganization and selfassembly principles by promoting conditions in which vasculogenesis or angiogenesis, two major ways of forming and developing vascularization in vivo, can occur. [2][3][4] The major advantage of the selforganizing method is its similarity to the in vivo processes to grow and develop vasculature, thus leading to spontaneous formation of vessels mimicking in vivo capillaries in both function and morphology. However, inconsistencies in the selforganizing nature and the cells in their ability to form functional MVNs hinder its wider application and limit broader impact. Currently, perfusable MVNs have been constructed using multiple dif ferent combinations of primary endothelial and stromal cells, but with wide diversity of morphology and functionality. Although various methods have been proposed to reduce this variability, including standardization of cell sources using induced pluripotent stem cells (iPSC) or immortalized cells, improved seeding strategies, or identifying key growth factors in neovessel formation, [5,6] an easily applicable approach that can effectively boost the vasculogenesis ability of endothelial cells (ECs) and aid the robust formation of functional MVNs is still an urgent unmet need for the community.
A developing vasculature in vivo is subjected to various mechanical cues that were not incorporated in most in vitro MVN models, notably the shear stresses induced by pulsatile or unidirectional blood flow, transmural flow, and interstitial flow (IF). [7,8] Particularly, IF plays a critical role during vascu logenesis to drive nutrients, remove metabolic wastes, and provide mechanical cues to cells, before a perfusable vascula ture is established. Various microfluidic platforms have been engineered to investigate the effects of IF on ECs and vessel formation. Most of the work reported a beneficial role of IF on angiogenesis, vasculogenesis, and 3D capillary morphogen esis in vitro. [9][10][11][12][13][14] These previous reports suggest that IF could be potentially used to benefit the formation of selforganized MVNs, which is validated recently by studies using generic MVNs formed with human umbilical vein endothelial cells (HUVECs) and human lung fibroblasts (HLFs) coculture, [15] and brain MVNs with triculture of primary human brain endo thelial cells (BECs), pericytes (PCs), and astrocytes (ACs). [16] However, those studies failed to maintain a relatively stable IF during neovessel formation, and did not investigate the effects of varied IF speeds. Moreover, the underlying mechanisms by which IF lead to a better vasculature were not examined.
In this work, MVNs were grown under various physiologically relevant IF in customized singlechannel microfluidic chip, for which IF was fully characterized by computation and experi ments. After systematically investigating the beneficial role of IF in neovessel formation, we further demonstrated that MVN morphologies are determined by global IF rather than local IF with a modified singlechannel device. Moreover, with a series of inhibitory experiments, we found that the boosted vasculo genesis capacity of ECs is associated with upregulated MMP2 when cultured under IF. Based on the postulated mechanism, we developed a novel and robust approach to regulate key morphological parameters of selforganized MVNs through the interplay between IF and MMP2 inhibitor, with vascular permeability and perfusability well maintained. Our findings are further validated with a BBB model made of BECs, PCs, ACs, suggesting the identified mechanism and proposed meth odology could be more generally useful to ECs and have the potential to be broadly employed for numerous applications, especially to boost formation of various functional organotypic MVNs.

In Vitro Microvascular Network Formation under IF
To investigate vessel formation under IF, two types of micro fluidic device were fabricated. A common threechannel device was used to culture MVNs statically (Figure 1a) while a different device with a single rectangular gel region was fabri cated to culture MVNs under IF (Figure 1b). A pressure differ ence was established using different levels of culture medium in the reservoirs inserted at the inlet and outlet to drive IF through the entire gel region. Through Darcy's law, we can measure matrix permeability from the rate at which liquid accumulates in the outlet reservoir due to the pressuredriven flow (detailed analytical calculation can be found in the Supporting Informa tion). In the singlechannel device, IF can be maintained for 24 h, at which time culture medium was replenished to restore the pressure gradient. The relatively stable IF speed was con ( 24 ) ( 0) 74.6% = = = using analytical solutions ( Figure 1c) and measured matrix permeability (Figure 1d). To ensure fibrin gel stability and identify any changes in gel properties due to flow alone, an initial set of experiments were conducted at pressure differences of 10, 20, and 30 mm H 2 O applied to acellular fibrin gel in the device for 6 d. Average IF speeds of 1-3 µm s −1 , consistent with normal phys iological ranges observed in vivo, [17] were calculated from the volume of accumulated culture medium in the outlet reser voir each day. We found acellular fibrin gels to be stable under our experimental conditions for at least 6 d. Furthermore, the measured IF speeds agree well with our analytical calculation ( Figure 1e).
Next, we seeded HUVECs with HLFs in microfluidic devices to form MVNs. IFs were induced with 10, 20, and 30 mm height differences of culture medium between the inlet and outlet. In the first 24 h, different pressure gradi ents lead to distinct IF speeds (Figure 1f), and the measured average IF speeds agree well with the IF speeds in acellular fibrin gel, suggesting that the embedded cells did not alter matrix permeability when initially seeded. However, as the cells selforganize into MVNs and remodel the extracellular matrix (ECM), the effective matrix permeability drastically changed. Interestingly, the effective matrix permeability con tinued to decrease over the first 3 d (Figure 1g), likely due to a combination of cellsecreted ECM and matrix remodeling by HUVECs and HLFs. After this initial period, as the nascent vessels progressively connected with each other and the MVNs become gradually perfusable, luminal flow dominated over IF and equilibrated the restored pressure difference in only a few hours so it was no longer possible to measure matrix perme ability. At day 6, ECs selforganized into densely connected 3D MVNs mimicking the morphology of in vivo capillaries. A fully perfusable MVN formed across the gel region over the entire length of 15 mm, as confirmed by perfusing dextran from the inlet (Figure 1h).

IF Improves the Formation of Functional MVNs
To systematically investigate the effect of IF on vasculogenesis in vitro, HUVECs and HLFs were cocultured in the fibrin gel under static and various IF conditions. By day 6, when MVNs were interconnected and perfusable, clear morphological differences could already be observed in projected images acquired in devices cultured under static and various IF con ditions (Figure 2a). Our quantification clearly showed that vessel morphology was noticeably affected by IF treatment. [18] MVNs formed under IF covered significantly more pro jected area with perfusable vessels. Interestingly, the largest perfusable vessel coverage occurred when subjected to low IF, and kept decreasing as IF was increased (Figure 2c). We also observed fewer branch points and longer branch length as IF   (Figure 2d,e). Vessel diameters were strongly affected by IF, in comparison to MVNs formed statically. IF treatment led to vessels with significantly larger diameters ( Figure 2f). Furthermore, stronger IF engendered MVNs with significantly larger vessel diameters compared to lower IF rates.
Perfusability is one crucial factor when evaluating the quality of engineered MVNs. Usually vessels with larger diameter are easier to perfuse, so it is not surprising to find MVNs with IF treatment exhibited an overall higher perfusability. To better interpret perfusability, we defined MVNs with perfusability greater than 80% as fully perfusable, and MVNs that are not fully perfusable but with at least one perfusable vessel across the gel region as partially perfusable. When cultured under static conditions, we find only 50% of MVNs are fully perfusable and 27% are partially perfusable. In contrast, IF substantially boosted perfusability of MVNs, with all of them fully perfusable under any of the IF conditions ( Figure 2g). It also should be noted that forming perfusable vessels in the singlechannel device is more challenging because patent lumens must be connected across the entire gel region, over 15 mm, while perfusable vessels only need to be connected over the width of 3 mm across gel region when cultured statically in the threechannel device.  In our singlechannel device, nutrients are progressively consumed by cells as the IF passes along the gel region. If the gel region is excessively long, we anticipate that at some point the medium will fail to support the formation of functional MVNs near the downstream end due to depletion of nutrients.
To determine whether MVNs formed in our singlechannel device were affected by nutrient depletion, we quantified the morphology at inlet, middle, and outlet region separately under various IF conditions ( Figure S1, Supporting Information). We found similar morphologies in different regions when cultured under certain IF condition (Figure 2h), indicating adequate nutrients were supplied for MVNs formation over the entire gel region in our experimental setup.
Overall, MVNs formed under IF exhibited distinct morpho logies with a larger projected areas and larger vessel diameters, compared with MVNs grown statically. Further more, the variability seemed to be greatly reduced for all morphological parameters quantified in IF cases, indicating that IF regulates the in vitro neovessel formation in a robust and replicable manner. Moreover, IF treatment significantly boosted the perfusability of selforganized MVNs, leading to a robust method of forming fully perfusable MVNs in vitro.

IF Affects Global MVN Structures within Microfluidic Device
To better understand the role IF plays during neovessel forma tion, we fabricated another singlechannel device with a large circular middle region and two narrow neck regions (Figure 3a). With the same setup of reservoirs as before to maintain a pres sure gradient along the hydrogel, we were able to generate inhomogeneous IF within the gel. Setting matrix permea bility as the value we measured for acellular fibrin gel in this study, we computed the distribution of IF within the entire device using COMSOL (Figure 3c). Four regions each with the same area were chosen as ROIs, shown as white rectangles in  morphology. Distinct MVN morphologies between 100 and 200 Pa pressure difference were still obvious though, following the same trend as we reported in the previous section: MVNs formed under stronger IF were associated with larger vessels, smaller coverage area and fewer branches.
We further examined the effect of local IF on morphology. Interestingly, we observed MVNs in the neck region subjected to 100 Pa endtoend pressure (higher local IF speed) were asso ciated with narrower vessels, larger coverage area and more branch points compared to the MVNs in the middle region subjected to 200 Pa endtoend pressure (lower local IF speed), which is opposite to what we found before. These findings sug gest that MVN morphologies are determined by global IF rather than local IF. One possible explanation is that the secretomes of the ECs that we find are regulated by IF and are crucial for neovessel formation. Secreted factors therefore would exert their influence globally as they are transported by IF and diffu sion. This knowledge is particularly important for microfluidic designs incorporating IF to boost MVN formation. Even aiming for engineering MVNs with uniform morphology, there is no need to require that the design accommodate for spatial homo geneity of IF.

IF Enhances the Vasculogenesis Capacity of EC through Upregulation of MMP-2
In order to selforganize into functional MVNs, ECs must remodel the surrounding ECM, mostly through cleavage of ECM components by secreted enzymes from embedded cells. Since we consistently observed larger diameter vessels in MVNs formed under IF, we hypothesized that IF elevates the produc tion of proteases, thus enhancing the vasculogenesis capacity of ECs. In our systems, however, HLFs were also seeded as supporting cells with HUVECs, making it difficult to decouple the effect of IF on each cell type. We therefore decided to elimi nate the HLFs from the system and test protease function using a monoculture of HUVECs to form MVNs. In our experience, only rarely are HUVECs able to selforganize into perfusable MVNs in the absence of supporting cells. And even in those rare instances in which they do, the vessels tend to be leaky and regress rapidly. [2,3,19] Consistent with those earlier findings, when we seeded HUVECs alone in our threechannel device cultured without IF, we could find no perfusable vessels. Even though ECs seemed to form a connected network, they were not able to form continuous patent lumens (Figure 4a). In con trast, a HUVEC monoculture of the same concentration seeded in a singlechannel device subjected to IF exhibited significant improvement in neovessel development. IF drastically altered the behavior of HUVECs seeded in the device, leading to robust formation of fully perfusable and interconnected MVNs by day 5 (Figure 4b).
To reveal the underlying mechanism, we performed quan titative PCR to evaluate gene expression levels in HUVECs extracted from the hydrogel at day 2 during neovessel forma tion cultured under both static and IF conditions. Following our hypothesis, we further carried out a series of protease inhibitory experiments to identify the specific enzymes that are responsible for the significant difference induced by IF treatment (Figure 4c). Since fibrin was used as our 3D gel scaf fold in this study, we initially focused on plasminogen activator (PA)-plasmin system because it constitutes the most common pathway for fibrinolysis. Tissue plasminogen activator (tPA) and urokinase (uPA) are two factors that convert plasminogen to active plasmin, which is the protease mainly responsible for fibrinolysis. However, we did not find clear upregulation of the PA-plasmin system when we quantified expression levels of several related genes (PLAT, PLAU, PLAUR, SERPINE1, ANXA2, S100A10, THBD) ( Figure 4d). Moreover, supplementing culture medium with high concentration plasminogen activator inhibitor1 (inhibiting both uPA and tPA) or aprotinin (inhibiting plasmin) did not adversely affect the development of fully perfusable MVNs by HUVEC mono culture under IF conditions. Those findings clearly suggest that HUVECs remodel fibrin gel during neovessel formation in a plasminindependent manner in our experimental setup, and that there must be other protease pathways mediated by IF that regulate fibrinolysis during vasculogenesis.
MMP is another common protease family consisting of more than 20 members, each of which exhibits substrate specificity for numerous ECMs. Although not specifically functioning as fibrinolytic enzymes, it has been reported that several MMPs, including at least MMP1, MMP2, MMP3, and MMP14, have the capacity to degrade fibrin gel. [20,21] Furthermore, con sistent with our finding, other groups reported that ECs invade fibrin ECM in a plasminindependent manner and require the activity of MMPs. [20,22,23] To confirm that MMP activity facili tates neovessel formation in a fibrin scaffold, we grew HUVEC monocultures under IF in our system adding 50 × 10 -6 m mari mast, a highly potent broad spectrum MMP inhibitor, to the medium. We found that inhibition of MMPs completely dis rupted the vessel formation process. Even after 5 d of culturing with IF, HUVECs were only able form short discontinuous seg ments with no signs of lumen formation. ECs are known to produce multiple MMPs, including MMP1, MMP2, MMP9, MMP13, MMP14, but with variations in species and micro environment. [24] To identify specific MMPs that are crucial for vessel formation, quantitative PCR was performed on a series of MMPs from HUVECs cultured in 3D fibrin gel under static or IF conditions at day 2, and found that IF upregulated expres sion levels of MMP1 and MMP2 by 1.6 and 2.7fold, respec tively. MMP14, also known as membrane type1 MMP or MT1 MMP, was downregulated by IF (Figure 4d). However, we still included MMP14 in the subsequent inhibitory experiments because MMP14 has been reported to be involved in the pro cesses of EC invasion and network formation through localized basement membrane degradation. [21] In our following inhibitory tests, ARP100 (IC 50 values are 12, 200, 4500, >50 000, and >50 000 × 10 −9 m for MMP2, MMP9, MMP3, MMP1, MMP7, respectively) and NSC 405020 (selec tive inhibitor of proteolytic activity of MMP14 without affecting the catalytic activity of cellular MMP14) were used as specific inhibitors for MMP2 and MMP14. Because narrow band inhibitors with exclusive specificity for MMP1 are not avail able at this time, we used high concentration antiMMP1 neutralizing antibody to inhibit its activity. Our experiments showed that blocking activity of MMP1 does not affect vascu logenesis of HUVEC monoculture in fibrin gel under IF; fully Unlike MMP1 or MMP14, MMP2, which is also the most upregulated MMP under IF, turns out to be crucial for vasculogenesis in fibrin scaffolds. Supplementing with 20 × 10 -6 m of MMP2 specific inhibitor ARP100 significantly altered the morphology of MVNs formed with HUVEC mono culture under IF. HUVECs were only able to selforganize into MVNs with extremely narrow vessels that were not perfusable (Figure 4e). With even higher concentrations of ARP100 (50 × 10 -6 m), HUVECs barely migrated in the 3D gel scaffold due to the strong suppression of MMP2 activity (Figure 4e). At day 5, most HUVECs were still not able to connect with their neighboring cells, leading to complete failure in developing a network structure.
Since it has also been reported that hypoxia promotes the production of MMP and enables clusterbased vasculogenesis of endothelial progenitor cells, [25] we further evaluated HIF1A with qPCR. Similar gene expression levels of HIF1A were found in MVNs cultured statically and under IF (Figure 4d)  through upregulation of a specific member of the MMP family, MMP2.

Regulating Morphology of MVNs through Interplay between IF and MMP-2 Inhibitor while Maintaining Functionality
Concentration of MMP2 specific inhibitor was observed to influence distinct behaviors of HUVECs for the entire dura tion of in vitro vessel formation (images captured at day 1, 3, and 5 under 10, 20, and 50 × 10 -6 m ARP100 can be found in Figure S2, Supporting Information). Inspired by the observa tion that MMP2 inhibitor noticeably altered MVNs formed using HUVEC monoculture under IF in a dosedependent manner (Figure 5a), we hypothesized that balancing MMP2 activity through interplay between IF and MMP2 inhibitor could be potentially used to regulate critical morphological parameters while maintaining the perfusability of engineered MVNs. Using MVNs developed by HUVEC monoculture under IF, we indeed found that appropriate concentration (10 × 10 −6 m ARP100 in this case) of MMP2 inhibitor effectively reduced the ECM remodeling capacity of HUVECs, leading to the formation of MVNs with significantly narrower vessels, reduced coverage area, and increased branches, compared with MVNs in the control group (Figure 5b-d). Most importantly, unlike supplementing with higher dose of MMP2 inhibitor, perfusability of MVNs treated with 10 × 10 −6 m MMP2 inhib itor are well maintained (Figure 5e). We further validated this approach using HUVEC and HLF coculture, which is probably the most common combination used to engineer functional MVNs in vitro (Figure 5f). Again, a clear dosedependent effect of MMP2 inhibitor was observed, following the same trend as we characterized in MVNs formed with HUVEC mono culture (Figure 5g-j). However, MVNs formed by HUVEC and HLF coculture can maintain high perfusability with a higher con centration (20 × 10 −6 m ARP100) of MMP2 inhibitor when cultured under the same IF.
Since vascular permeability is another critical aspect regarding functionality of engineered MVNs, we measured and compared vascular permeability using media containing 40 kDa dextran under various conditions in which perfusable vessels can be robustly formed ( Figure S5, Supporting Information). Interest ingly, we found no clear difference in vascular permeability across all the conditions (Figure 5k), despite the MVNs exhibiting dis tinct morphologies. Moreover, all the measurements were on the order of 10 −8 cm s −1 , comparable to vascular permeability meas ured in vivo using a 40 kDa dextran. [26] These results suggest perfusable MVNs formed under various conditions in this study exhibit normal barrier functions, even those formed without supporting fibroblasts. This contrasts with previous reports that MVNs formed by HUVEC monoculture are more permeable than those formed with various types of stromal cell. [19] The excellent barrier function of MVNs formed by HUVEC monoculture under IF cannot be explained alone by our findings regarding MMP2 upregulation under IF, and there are likely to be other pathways working synergistically. We further compared gene expression levels in connection with ECM secretion, cell migration, adhesion, and angiogenesis on HUVECs seeded in devices cultured under static and IF conditions at day 2. However, none of those genes were obviously upregulated by IF treatment (Figure 4d), sug gesting that further evaluation on a more extensive gene library is needed in the future.
Next, we assessed the longevity of MVNs developed with HUVEC monoculture under IF with or without MMP2 inhib itor. Surprisingly, these monoculture MVNs remained perfus able from day 5 until at least day 26 when the experiment was terminated ( Figure S3, Supporting Information). During this entire time, MVN morphologies only moderately changed, while vascular permeabilities were well maintained. It should be noted that IF cannot be maintained after MVNs become fully perfus able; as mentioned above, luminal flow will take over IF and lead to rapid equilibration of the pressure difference between inlet and outlet. The mechanisms responsible for this superior longevity of MVNs formed under IF remain to be investigated.
Taken together, perfusable MVNs formed with HUVEC monoculture under IF exhibit outstanding functionality in terms of perfusability, permeability and longevity. Moreover, some critical morphological parameters of the selforganized MVNs can be robustly regulated through an interplay between IF and MMP2 inhibitor, with functionality well maintained.

IF Enhances Formation of Fully Perfusable Brain Microvascular Networks
To validate that our findings and methods are generalizable and can be used to boost the formation of other organotypic MVN models, we investigated the effect of IF treatment on vasculogenesis of an in vitro BBB model consisting of brain endothelial cells (BECs), brain pericytes (PCs), and astrocytes (ACs) that we had previously developed. [27] In static condition, BECs selforganized into MVNs with extremely narrow vessels by day 6 (Figure 6a). Those vessels appeared to lack continuous patent lumens, which was evidenced by failed dextran perfu sion. We then seeded the triculture in a singlechannel device and applied IF. Consistent with our other experiments, BECs developed into a fully perfusable MVN and exhibited excel lent barrier function (Figure 6d). IF also drastically altered the morphology of brain MVNs, significantly increasing vessel diameters leading to substantially perfusable networks, fol lowing the same trend we found in HUVECHLF experiments (Figure 6b,c). When cultured without IF, we observed perfus able vessels only rarely. In contrast, all MVNs treated with IF were fully perfusable, further validating the effectiveness of IF in forming functional MVNs in vitro (Figure 6c). Immu nostaining revealed that PCs and ACs reside in the interstitial space surrounding microvessels formed by BECs in both static and IF conditions (Figure 6e,f). PCs and ACs were in direct contact with endothelium, like described in other similar brain MVN models. [16,19] Quantitative PCR was further performed on BECs isolated from the triculture within a fibrin gel at day 2 during neovessel formation to evaluate gene levels of MMP2. We confirmed that MMP2 expression was increased (1.7 ± 0.26 fold) under IF, consistent with what we found in MVNs formed with HUVECs, suggesting that IF triggers the same down stream effect in both cases among different types of ECs.
Finally, we tested whether MMP2 specific inhibitor can be employed in the brain MVN model to antagonize the effect www.afm-journal.de www.advancedsciencenews.com 2206767 (9 of 14) induced by IF. Obvious differences were observed during brain MVN formation. On day 3, a higher concentration of MMP2 inhibitor (20 × 10 −6 m) led to thinner and sparser vessels compared with the ones treated with a lower concentration (10 × 10 −6 m) (Figure 6g). At day 6, brain MVNs with a low concentration of MMP2 inhibitor became perfusable with much thinner vessels compared to untreated ones. In contrast, networks cultured with the higher concentration of MMP2 inhibitor exhibited disconnected thin network segments with no perfusable vessels (Figure 6h).

Discussion
Selforganized in vitro MVNs are now widely utilized as models or platforms to study development, diseases as well as basic physiological processes, and demonstrate considerable poten tial in drug screening and development. Various hydrogels have been employed as a 3D scaffold to accommodate engi neered MVNs, including natural materials ranging from fibrin, collagen and Matrigel to synthetic biocompatible materials such as PEG and its derivatives. [28] Among these, fibrin is the most widely used to engineer selforganized MVNs. Understanding the dynamics between seeded ECs and fibrin ECM promises to provide valuable insight regarding the mechanism of vas culogenesis and engineering better synthetic ECM supporting formation of functional MVNs in vitro. Moreover, it could also inspire new strategies to regulate selforganizing in vitro vasculogenesis to have a degree of control over morphological parameters and perfusability in a much more robust manner. Interestingly, we found that inhibiting the PA-plasmin system, the most common means of inducing fibrinolysis, had little Figure 6. IF enhances formation of fully perfusable brain MVNs. a) Brain specific MVNs formed statically or under IF induced by 10 mm H 2 O pressure difference. Green: GFP BECs, magenta: 40 kDa Texas Red dextran. Scale bar is 1 mm. b,c) Statistical analysis of vessel diameter and perfusability of brain MVNs formed statically or under IF. n = 6 devices for static group; n = 3 devices for IF group. d) Vascular permeability measurements using 40 kDa dextran. n = 2 devices, two measurements in each device at different ROIs. e) Fluorescent confocal images of brain specific MVNs cultured statically at day 6 with zoomed-in view. f) Fluorescent confocal images of brain specific MVN cultured under IF induced by 10 mm H 2 O pressure difference at day 6, with zoomed-in view. Green: GFP BECs, magenta: S-100b staining for ACs, yellow: PDGFR staining for PCs. Scale bar is 1 mm for the left panel, and 100 µm for expanded view. g) Day 3 brain MVNs cultured under IF, supplemented with 10 or 20 × 10 −6 m ARP-100. Scale bar is 1 mm. h) Day 6 brain MVNs cultured under IF, supplemented with 10 or 20 × 10 −6 m ARP-100. Green: GFP BECs, magenta: 40 kDa Texas Red dextran. Scale bar is 1 mm. effect on vasculogenesis in fibrin scaffolds. Further investiga tion revealed that MMP2, a specific member of the MMP family that we found to be upregulated by applying IF, plays a critical role during the neovessel formation of engineered MVNs. Our findings are consistent with previous reports that EC invasion and capillary morphogenesis in fibrin were not affected by plasmin inhibitors, and in contrast, these pro cesses were completely inhibited by natural and synthetic MMP inhibitors. [20,22,23] A simple singlechannel device was fabricated to grow MVNs under IF. We extensively characterized the IF with both experi ments and computations to ensure that a relatively stable IF could be maintained during neovessel formation. With this approach, we then systematically investigated the effect of IF on engineered MVNs and found that IF significantly affects neovessel formation, leading to MVNs with larger and more uniform diameters compared with those cultured under static conditions. As a consequence, perfusability, a crucial attribute of engineered MVNs, was dramatically improved. Further PCR analysis and a series of carefully designed inhibitory tests directly relate IF treatment with MMP2 activities of ECs. MVNs cultured under IF responded in a dosedependent manner when subjected to various concentrations of MMP2 specific inhibitor. With a suitable dose, we are able to balance MMP2 activity through the interplay between IF and MMP2 inhibitor, to obtain MVNs with small diameter but still fully perfusable vessels. This method has been successfully used in the current study to produce such vessels in generic MVNs using HUVEC monoculture or HUVECHLF coculture (Figure 4), and BBB models with triculture of BECs, PCs, and ACs ( Figure 5). Fur thermore, although the geometry of spontaneously formed MVNs cannot be predefined, our methodology confers a certain level of control over many important morphological para meters during selforganized neovessel formation, thereby enabling production of MVNs with desired morphology for various applications.
Although we identified MMP2 as the major downstream effector responsible for enhanced vasculogenesis capacity of ECs when cultured under physiologically relevant IF, the detailed underlying mechanism is currently under further investigation. We hypothesize this IFmediated response involves signaling at cell-matrix interfaces: IF leads to clus tering of corresponding integrins through activation of focal adhesion kinase (FAK), which further induces elevated metallo proteinase expression, thus allowing rapid and robust vascular network formation. Our hypothesis is well supported by several relevant works. For example, it has been shown that fluid shear stress results in increased FAK/Src phosphorylation and integrin clustering. [29,30] Furthermore, FAK participates in recruitment of vinculin and the formation of large focal adhesions that lead to the upregulation of MMPs, thereby enabling the cell sprouting and migration necessary for vascular morphogenesis. [31] Proper barrier function is an essential aspect of functional MVNs. Measurements of vascular permeability were thus carried out on various MVNs formed under different conditions with 40 kDa fluorescent dextran. Excellent barrier function was demon strated with our measurements (permeability ≈ 10 −8 cm s −1 ), which was shown to be up to two orders of magnitude lower than that measured in static transwell cultures and comparable to values measured in vivo. [26] No noticeable differences in vascular permeability were observed between MVNs cultured stat ically or under IF, suggesting IF could be employed as a robust method to boost perfusability of engineered MVNs without weakening barrier function. During those experiments, we find our singlechannel device is ideal for vascular permeability measurement. The long gel region eliminates contamination from free diffusion of dextran through porous hydrogel, which is an issue for the typical threechannel design and some com mercial MVN chips. Also because of the large gel region and the application of IF, the length of the perfusable MVNs formed is 1.5 cm, one order of magnitude larger than typical selforganized MVNs formed statically. Our approach might therefore be useful in engineering largescale vascularized bulk tissues by vascular selfassembly and applied more broadly in various applications combined with IF.
In most microfluidic chips, maintaining a homogenous stable IF during neovessel formation is quite challenging, even with the help of pumping system sometimes. To test if our approach can be broadly applied to boost formation of fully per fusable MVNs, we employed same setup to establish pressure gradient in commonly used threechannel device by hooking syringes in four media reservoirs ( Figure S4a, Supporting Infor mation). Due to the small width of the gel region (3 mm), the pressure gradient rapidly dissipates. Our simulation confirmed IF speed decreases 50% in about 2.5 h after the initial pres sure difference is established ( Figure S4b, Supporting Infor mation). However, this transient IF still sufficiently boosted the formation of fully perfusable MVNs with HUVEC mono culture ( Figure S4c, Supporting Information), similar to MVNs cultured under relatively stable IF. Furthermore, we confirmed applying IF at a later time point could rescue defective MVNs and make it perfusable. MVNs cultured statically for 3 d using HUVEC monoculture exhibited abnormal morphologies of disconnected narrow vascular segments. Morphology of the MVNs gradually improved after applying transient IF from day 3, and eventually the network became perfusable at day 9 ( Figure S4d, Supporting Information).
Recent years have seen explosive growth in the number and variety of engineered MVNs using various sources of ECs and stromal cells to mimic specific in vivo vascular microenviron ments. It is now widely recognized, however, that different types of ECs, or even a single type of ECs but from different donors, often exhibit widely divergent levels of vascular morphology and perfusability. Our approach thus offers the potential to confer vesselforming capacity among cells from various sources in order to ensure robust formation of functional MVNs. Specifi cally, our methodology can be used to boost the formation of certain organotypic MVNs that are difficult to form, or not yet available, saving tremendous time and resources. Moreover, MVNs are often engineered to study interactions with other physiological systems, such as neural or muscular tissues, or used as a platform to host various developing organoids. [32][33][34][35][36] In those multiculture systems, the choice of hydrogel scaffold and composition of culture medium must be carefully designed to support the development and growth of all the cellular com ponents. With our methodology, MVN development in multi culture systems could be substantially improved by employing IF to boost vasculogenesis, significantly broadening the applica tion of selforganized MVNs.

Conclusion
We systematically investigated the effect of IF on neovessel formation of selforganized MVNs in vitro using a customized singlechannel device, for which IF was fully characterized. Gene expression analysis and inhibitory tests revealed that the vasculogenic capacity of ECs induced by IF results from upregulation of MMP2. Balancing MMP2 activity through the interplay between IF and an MMP2 specific inhibitor, we managed to regulate many key morphological para meters of selforganized MVNs, while maintaining vascular perme ability, perfusability and longevity. Our findings were further validated with an organspecific MVN (brain) comprised of BECs, PCs and ACs, demonstrating that the identified mech anism and proposed methodology can be utilized to pro mote the formation of various functional MVNs with broad applications.

Experimental Section
Device Design and Fabrication: Two types of microfluidic device were made for this study: i) a three-channel device, featuring a central gel channel and two adjacent medium channels for culturing MVNs under static conditions (Figure 1a). A partial wall separates the central gel channel and side medium channels, allowing for surface-tensionassisted filling of cell-laden fibrin gels. ii) A single-channel device for growing MVNs under various IF conditions (Figure 1b). Single-channel devices consist of two layers: a bottom layer with a single gel channel and a top layer with larger holes at inlet and outlet. Molds of the devices were designed in AutoCAD (Autodesk, Inc.), and imported in Fusion 360 (Autodesk, Inc.) to generate corresponding tool paths, followed by milling a delrin block with a micro-CNC milling machine (Bantam Tools). Polydimethylsiloxane (PDMS) based microfluidic devices were then fabricated using the procedure outlined previously. [27] For singlechannel devices, 2 mm holes for the gel injection ports were punched in bottom layer, and 4 mm holes at inlet and outlet on top layer for attaching syringes to establish various IFs.
Cell Culture: HUVECs and normal HLFs (Lonza) were cultured in VascuLife VEGF Endothelial Medium (Lifeline Cell Technology) and FibroLife S2 Fibroblast Medium (Lifeline Cell Technology), respectively. HUVECs were transduced to express cytoplasmic GFP, as described earlier. [5] Primary human BECs, PCs, and ACs were purchased from ScienCell, cultured in corresponding growth medium (ScienCell) on a poly-l-lysine (Sigma) coated flask. All the cells were maintained in a humidified incubator (37 °C, 5% CO 2 ), with the culture medium replenished every 2 d. All cell types were used between passages 6 and 8.
Microvascular Network Formation and Establishing IF: Cells were cultured until near-confluent prior to detachment, and seeded into the chip as previously described. [27] Briefly, cells were concentrated in VascuLife containing thrombin (4 U mL −1 , Sigma). Cell mixture solution was then further mixed with fibrinogen (3 mg mL −1 final concentration, Sigma). The cell-laden fibrin solution was quickly injected into the device through the gel loading port. Various combinations of cells were used in this study to engineer different types of MVNs. For MVNs formed with HUVECs and HLFs, the final concentration is 7 m HUVECs mL −1 and 1 m HLFs mL −1 . For MVNs formed with HUVEC monoculture, the final concentration is 8 m HUVECs mL −1 . For brain MVNs, the final concentration is 6 m BECs mL −1 , 1 m ACs mL −1 , and 0.5 m PCs mL −1 . After seeding, the devices were placed in incubator for 15 min to allow complete polymerization, before adding VascuLife as culture media. Using syringes (3 mL sterile luer slip tip syringe, BH Supplies), a hydrostatic pressure difference was established to generate IF across the gel in the single-channel device, which can be maintained for 24 h (Figure 1c and Supporting Information). Culture medium was replenished every 24 h to restore the pressure gradient. All devices were kept in incubator with daily change of culture media for 5-7 d, until perfusable MVNs were formed.
Measurement of Matrix and Vascular Permeability: The difference in height of culture medium in the syringes attached at the inlet and outlet drives IF through the gel region of the single-channel device. Through Darcy's law, matrix permeability from liquid volume accumulated in the outlet reservoir at a given time can be measured (for detailed analytical calculation please see the Supporting Information). Vascular permeability was measured for various perfusable MVNs engineered in this study following published protocols. [26,27] Briefly, 40 kDa Texas red dextran (Invitrogen) was perfused into the microvascular networks by generating a slight pressure gradient across the gel of the device. Confocal images were captured with a 5 µm step size at 0 and 12 min, from which permeability was calculated ( Figure S5, Supporting Information). Several regions of interest (ROIs) were captured via timelapse volumetric imaging for each device. Dimensions of the ROI are 1.6 mm × 1.6 mm (640 × 640 pix).
Microvascular Network Perfusion, Imaging, and Analysis: To confirm the perfusability of MVNs, 40 kDa Texas red dextran solution was perfused into the microvascular networks by generating a slight pressure gradient across the gel region. Confocal images were then acquired using an Olympus FLUOVIEW FV1200 confocal laser scanning microscope with a 10× objective. Z-stack images were acquired with a 10 µm step size. Morphological parameters were analyzed and quantified based on collapsed z-stack images using AutoTube. [18] In addition, perfusability of MVNs was calculated as the ratio between dextran projected area (perfusable vessels) and GFP ECs projected area (all the vessels), presented as a percentage. If not otherwise specified, ROIs are 4 mm × 1.6 mm (2000 × 800 pix, with stitching) and 1.6 mm × 1.6 mm (800 × 800 pix) for single-channel devices and three-channel devices, respectively.
Treatment of Inhibitors: All inhibitor experiments were performed using the single-channel device with IF induced by 10 mm H 2 O pressure difference. Individual inhibitor was supplemented in the culture media for the whole duration of MVN formation. Control experiments with DMSO and corresponding isotype control antibody were also performed. Cell Liberation from MVNs, Flow Cytometry, and RT-PCR: Gene expression levels of the MVN cell components were measured to assess changes due to IF treatment, following a published protocol. [27] In brief, cells were isolated from devices 2 d after seeding using liberase (Sigma). Gels from three single-channel devices and three 3-channel devices were pooled to ensure that enough ECs can be isolated via fluorescenceactivated cell sorting (FACS) to obtain suitable RNA levels necessary for appropriate gene quantification by PCR.
Corresponding HUVECs or BECs were sorted using their intrinsic GFP expressions. Total RNA was isolated using TRIzol (Life Science). Reverse transcription was performed using High-Capacity RNA-to-cDNA Kit (Thermo Fisher Scientific). TB Green Premix Ex Taq II (Takara) was used for RT-PCR, performed on a 7900HT Fast Real-Time PCR System (Applied Biosystems). Glyceraldehyde 3-phosphate dehydrogenase (GAPDH) was used as the control housekeeping gene. Primers were synthesized by Genewiz. All primers used in this study are listed in the Supporting Information.