Fluid forces control endothelial sproutingJonathan W. Song and Lance L. Munn1

Edwin L. Steele Laboratory for Tumor Biology, Department of Radiation Oncology, Massachusetts General Hospital and Harvard Medical School,Charlestown, MA 02129

Edited* by Shu Chien, University of California at San Diego, La Jolla, CA, and approved August 1, 2011 (received for review April 5, 2011)

During angiogenesis, endothelial cells (ECs) from intact bloodvessels quickly infiltrate avascular regions via vascular sprouting.This process is fundamental to many normal and pathologicalprocesses such as wound healing and tumor growth, but itsinitiation and control are poorly understood. Vascular endothelialcell growth factor (VEGF) can promote vessel dilation and angio-genic sprouting, but given the complex nature of vascularmorphogenesis, additional signals are likely necessary to deter-mine, for example, which vessel segments sprout, which dilate,and which remain quiescent. Fluid forces exerted by blood andplasma are prime candidates that might codirect these processes,but it is not known whether VEGF cooperates with mechanicalfluid forces to mediate angiogenesis. Using a microfluidic tissueanalog of angiogenic sprouting, we found that fluid shear stress,such as exerted by flowing blood, attenuates EC sprouting ina nitric oxide-dependent manner and that interstitial flow, such asproduced by extravasating plasma, directs endothelial morphogen-esis and sprout formation. Furthermore, positive VEGF gradientsinitiated sprouting but negative gradients inhibited sprouting,promoting instead sheet-like migration analogous to vessel di-lation. These results suggest that ECs integrate signals from fluidforces and local VEGF gradients to achieve such varied goals asvessel dilation and sprouting.

3D angiogenesis on a chip | collagen gel | structural remodeling |alternative to animal model | vessel analog

Angiogenesis, the expansion or extension of existing vascula-ture, is necessary to deliver oxygen and nutrients to ischemic

or avascular regions in wounds and solid tumors (1), and a funda-mental understanding of the determinants of angiogenesis wouldaccelerate progress in the fields of regenerative medicine, tissueengineering, and oncology. Angiogenesis requires the coordinatedgrowth and migration of endothelial cells (ECs): each EC residingin a vessel wall integrates local signals to determine whether itremains quiescent (2, 3), participates in dilation or contraction (1),undergoes morphogenesis to form an angiogenic sprout (4, 5), orintussusceptive involution (6). Implicit in these processes are en-dothelial proliferation during expansion of the vasculature andtheir loss during vessel contraction and pruning (7).Growth factors have pleiotropic effects onECs and are undoubt-

edly key controllers of vascular morphogenesis. The best studiedvascular morphogen, vascular endothelial growth factor (VEGF)(8), stimulates EC migration (9), proliferation (10), and matrixdegradation (2). VEGF also controls vessel morphogenesis by in-ducing Delta-like ligand 4 (Dll4)—a membrane-bound ligand forthe Notch family of receptors—at the advancing front of sprouts(11), thereby promoting the formation of specialized “tip cells,”which extend protrusions or filopodia that sense growth factorconcentrations to guide sprouts (4, 5). However, given the distinctphenotypes exhibited by ECs during sprouting, dilation, contrac-tion, and quiescence—even within the same vessel segment—thereare undoubtedly codeterminants that direct EC behavior.ECs in patent blood vessels are exposed to mechanical forces

tangential to the endothelial surface due to blood flow (12–15)and across the vessel wall due to interstitial plasma flow (16, 17).Fluid shear stress imposes signals that mediate EC transcription(18), membrane fluidity (19), VEGF receptor conformationalchanges (20), tubule formation (21, 22), intraluminal morphol-

ogy (23), barrier function (24), and vessel homeostasis, bymaintaining vessel lumens (25, 26) and controlling EC pro-liferation (27) and turnover (1). In addition, shear stress caninduce significant changes in EC morphology (28) and actin cy-toskeleton rearrangement (29), whereas flow transverse to theendothelium (16) and/or through the interstitial space (30) canalso cause endothelial morphogenesis.Although there is a wealth of evidence that fluid forces affect

endothelial phenotype, no study has simultaneously examinedthe interplay of tangential shear stress, transverse interstitialflow, and VEGF gradients in mediating sprouting morphogen-esis from an intact vessel. Using a microfluidic model of angio-genic sprouting, we found that fluid shear stress inhibits vesselmorphogenesis (rearrangement of the vessel wall microanatomyresulting in sprouting or invasion into the matrix) via the nitricoxide (NO) pathway, and interstitial flow increases the rate ofmorphogenesis. Surprisingly, invading ECs not only detected thedirection of VEGF gradients but also showed dramatic differ-ences in morphogenesis, depending on the direction of in-terstitial fluid flow. Endothelial tip cell filopodia preferentiallyprotruded against the direction of interstitial flow or in the di-rection of an increasing VEGF gradient. In contrast, a de-creasing VEGF gradient promoted migration of an endothelial“sheet” in a process analogous to vessel dilation. These resultsemphasize the importance of multiple signals during angiogen-esis and suggest that fluid forces are important mediators ofvascular homeostasis and morphogenesis.

ResultsAngiogenic Sprouting in Vitro. To determine how fluid and chem-ical factors cooperate—or antagonize—to modulate sproutingfrom realistic vessel analogs in vitro, we developed a microfluidicplatform that features (i) fluid flowing through two adjacent,endothelial lined channels, (ii) contact between the vessel walland a 3D collagen matrix on the abluminal side to allow endo-thelial sprouting, (iii) controllable fluid convection through thematrix, and (iv) specifiable growth factor gradients (Fig. 1). Twoparallel channels (50 μm in height) spaced 300 μm apart andlined with confluent human umbilical vein endothelial cells(HUVECs) traverse the device (Fig. 1 B–D). Along the device,there are seven apertures of 3D collagen I gel matrix (100 μm inwidth) that allow contact between the vessel walls and theintervessel matrix (Fig. 1B). HUVEC-GFP cells that were stim-ulated with 50 ng mL−1 VEGF for 3 d migrate into the bulk of the3D extracellular matrix (ECM) space rather than along the top orbottom surface, adopting morphology distinct from the cells thatremain as a 2D monolayer in the main vessel (Fig. 1E). To moreeasily distinguish the various parameter sets for each experiment,we adopted the naming conventions in Table 1.

Author contributions: J.W.S. and L.L.M. designed research; J.W.S. performed research;J.W.S. and L.L.M. analyzed data; and J.W.S. and L.L.M. wrote the paper.

The authors declare no conflict of interest.

*This Direct Submission article had a prearranged editor.

Freely available online through the PNAS open access option.1To whom correspondence should be addressed. E-mail: munn@steele.mgh.harvard.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1105316108/-/DCSupplemental.

15342e15347 | PNAS | September 13, 2011 | vol. 108 | no. 37 www.pnas.org/cgi/doi/10.1073/pnas.1105316108

Shear Stress Attenuates VEGF-Induced Morphogenesis. Endothelialmorphogenesis and invasion into the gel were prominent frombothchannels under static conditions (SA0T0G0jSB0T0G0), but minimalwhen the cells were exposed to shear stress (SA3T0G0jSB3T0G0) at

either VEGF concentration (V5 or V50; Fig. 2). With no addedVEGF, there was little sprouting under static conditions(SA0T0G0jSB0T0G0jV0; Fig. 2C). Thus, physiological shear stress(3 dyn cm−2; Fig. S1) attenuates VEGF-driven morphogenesis.The attenuation of invasive morphogenesis by shear stress wasaccompanied by a decrease in EC proliferation (Fig S1C). In-hibition of invasion by shear was not due to proangiogenic factorsbeing washed away in the flow, as shear inhibition was also seenwith conditioned medium (Fig. S2).To determine whether VEGF gradients (Fig. S3) drive the

invasive morphogenesis, we applied identical flow conditionsto both channels but only added VEGF to channel A(SA0.1T0G−jSB0.1T0G+jV50 or SA3T0G−jSB3T0G+jV50). Theslow flow in the SA0.1T0G−jSB0.1T0G+jV50 condition appliedminimal shear stress (w0.1 dyn cm−2), but was sufficient to re-plenish nutrients and maintain a stable biochemical gradient(Fig. S3 D–F). Morphogenesis and gel invasion occurred in theslow-flow configuration from both channels, but required VEGF(SA0.1T0G−jSB0.1T0G+jV50; Fig. S4). Again, little invasion oc-curred at physiological shear stress levels, even with a VEGFgradient (SA3T0G−jSB3T0G+jV50; Fig. S4 C and D).

Attenuation of Morphogenesis by Shear Stress Requires Nitric OxideProduction. To further investigate the attenuation of HUVECsprouting by shear stress, we next blocked production of nitricoxide (NO), an important shear stress-responsive signaling mole-cule (31, 32).We infused VEGF-containing medium (50 ngmL−1)into both HUVEC channels at physiological shear stress levels (3dyn cm−2), but added the pan nitric oxide synthase (NOS) inhibitorNG-monomethyl-L-arginine monoacetate (L-NMMA) (33) (200μM) into channel A to block NO signaling (32). As expected, withshear flow in both channels (SA3T0G0jSB3T0G0jV50), minimalsprouting was observed in the channel exposed to VEGF withoutL-NMMA (Fig. 2E). However, sheared cells exposed to VEGFwith L-NMMA sprouted into the collagen gel (SA3T0G0jL; Fig. 2D and E) at a rate similar to that seen in static channels (Fig. 2C).This effect was not due to direct activity of L-NMMA: whenL-NMMA was added to standard media with no VEGF, littlesprouting was observed (Fig. 2E), and L-NMMA combined withVEGF did not enhance sprouting in static cultures, compared withVEGF alone (SA0T0G0jSB0T0G0jV50; Fig. S5). These results showthat the attenuation of VEGF-induced sprouting caused by shearstress requires NO signaling.

Direction of VEGF Gradient with Interstitial Flow Affects SproutMorphology. We next evaluated the effect of simultaneous appli-cation of physiological levels of shear stress (3 dyn cm−2) and in-terstitial flow (2.5–35 μm s−1; Fig. S6) (16, 30) on sproutingmorphogenesis. Either pushing VEGF-containing medium intochannel A (SA3TwG−jSB0TaG+) or pulling non–VEGF-containingmedium into channel B (SA0TwG−jSB3TaG+) creates interstitialflow (Fig. S6) and a VEGF gradient (Figs. S7 and S8) from A to B.

Fig. 1. Microfluidic device with localized 3D ECM for fluid force-mediatedangiogenic sprouting and morphogenesis. (A) Multilayer fabrication of thepoly(dimethylsiloxane) PDMS microfluidic device featuring localized regionof collagen gel (blue). The top PDMS layer contains the channel features (50μm in height) and the bottom layer provides a planar surface. (B) HUVECsseeded into two channels separated by multiple collagen gel apertures vi-sualized under phase microscopy. (C) Immunofluorescence staining for VE-cadherin expression (red) demonstrates integrity at intercellular junctions ofthe HUVEC monolayer. Blue depicts nuclei stained with DAPI. (D) Cross-sec-tion view of one of the HUVEC channels. HUVEC-GFP cells seeded on top,bottom, and Side faces mimicking a fully-lined blood vessel. (E) HUVEC-GFPcells sprouting into 3D collagen gel demonstrate clear morphological dif-ferences, with HUVECs invading the bulk of the gel rather than along thetop or bottom surface. (F) Close-up view of boxed area in A showing sevenapertures that allow connection of the two HUVEC channels (green) throughthe collagen barrier (blue). Each HUVEC channel has independent input andoutlet ports, allowing strict control over flow in both channels (SA and SB)and across the collagen matrix (T). As drawn, the nomenclature for this flowconfiguration is SA3TwG−jSB0TaG+. (Scale bars, 100 μm.)

Table 1. Notation for flow and VEGF conditions

Notation Definition Superscripts

SA, SB Shear stress in channel A (Upper channel)or channel B (Lower channel),respectively (Fig. 1)

0 indicates approximately no flow in the channel; 0.1 indicates w0.1 dyn cm−2;and 3 is w3 dyn cm−2. For example, SA3 and SB0 indicate that channel Awas exposed to 3 dyn cm−2, whereas channel B was static.

T Transverse convection or interstitial flow(2.5–35 mm s−1; Fig. S7)

0 indicates no convection; a indicates convection against the directionof endothelial invasion; and w means convection with thedirection of endothelial invasion.

G VEGF gradient 0 indicates no VEGF gradient; + indicates invasion toward theVEGF source or a positive gradient; and − indicates invasionaway from the VEGF source or a negative gradient.

V The concentration of exogenously added VEGF 0 indicates no VEGF; 5 indicates 5 ng ml−1; and 50 indicates 50 ng ml−1.

Song and Munn PNAS | September 13, 2011 | vol. 108 | no. 37 | 15343

MED

ICALSC

IENCE

S

As before (Fig. 2A), cells in the sheared channels, which were alsocontinually stimulated with VEGF-containing media, exhibitedvery little invasion (Fig. 3). In contrast, HUVECs in the non-sheared channels invaded into the collagen gel (Movie S1), andthe area of invasion increased with VEGF concentration (V0, V5,and V50; Fig. 3, P < 0.0001). However, HUVECs in the twoconfigurations invaded the gel in opposite sense relative to thedirection of the VEGF gradient and interstitial flow and exhibitedstriking differences in sprout morphology (Fig. 4 A and B). WhereHUVECs invaded toward the VEGF source (positive VEGFgradient) and against the direction of interstitial flow(SB0TaG+jV50), the number of filopodia or tip cell projections (5,34) was significantly greater than in the configuration whereHUVECs invaded away from the VEGF source (negative VEGFgradient) and with the direction of interstitial flow (SA0TwG−jV50;Fig. 4 A, B, and E, gray vs. orange bar). The prominent filopodiaprojected by cells moving toward the VEGF source and againstflow (SB0TaG+jV50 configuration) were characteristic of sproutingfrom a preexisting vessel. On the other hand, HUVECs movingdown the gradient, with the flow (SA0TwG−jV50 configuration)maintained a relatively smooth boundary as they moved into thegel. This process more closely resembled vessel dilation thansprouting.

Interstitial Flow Enhances Sprouting Morphogenesis. To testwhether interstitial flow affects HUVEC invasion and sproutmorphology independent of a VEGF gradient, we connectedboth ports of channel B to the pump and pulled media (28.5 μms−1) through the intervessel matrix from reservoirs connected tochannel A, while exposing both channels to negligible levels oftangential shear (SA0TwG0jSB0TaG0). This configuration alsoeliminated VEGF gradients (Figs. S8 and S9 A–C). Morphogenicinvasion occurred both with and against the direction of flow tothe same extent (SA0TwG0jSB0TaG0jV50 condition; Fig. S9 Dand E; P > 0.1). Fig. 4 D and E consolidate and compare the data

with interstitial flow, without and with VEGF gradients (pro-duced with the previous flow configurations). Interstitial flowalone significantly increased the area of invasion by w75% forboth directions (Fig. 4D). Furthermore, the presence of a VEGFgradient (both positive and negative) enhanced interstitial flow-guided invasion in an additive manner for both directions (Fig.4D). However, in the absence of a VEGF gradient, sproutsmoving against the direction of interstitial flow still had morefilopodia than those moving with the flow (Fig. 4E, yellow vs.purple bar). Considering only the sprouts moving against theinterstitial flow from channel B, the presence of the positiveVEGF gradient did not result in more filopodia, compared withthe uniform VEGF condition (Fig. 4E, purple vs. gray bar). Forinvading cells moving with the direction of flow from channel A,the negative VEGF gradient inhibited filopodia formation (Fig.4E, orange vs. yellow bar). These results show that interstitialflow and VEGF both enhance morphogenic invasion, andsprouts moving against the direction of interstitial flow or upa VEGF gradient extend more filopodia.

DiscussionThe primary role of the vasculature is to deliver blood efficiently.Therefore, it is not surprising that forces exerted by blood andplasma would provide feedback control for formation of thatvasculature. It has long been known that blood flow mediatesremodeling of immature embryonic networks (35, 36) and helpscontrol vascular tone and collateral formation in muscle tissue(37, 38). However, the role of fluid forces in the initiation ofangiogenesis is not well understood, mainly due to a lack of ap-propriate tools for controlling the relevant parameters. Commonapproaches either lack intraluminal flow (e.g., the “tube forma-tion” and microbead sprouting assays) or by virtue of plating cellson a solid substrate, prohibit the EC reorganization necessary forvessel dilation or sprouting (e.g., parallel plate flow chambersand poly(dimethylsiloxane), PDMS channels). In vivo analyses,

Fig. 2. Shear stress attenuates VEGF-induced HUVECinvasion. (A) To examine the role of shear stress, identi-cal flow (3 dyn cm−2) and VEGF conditions (50 ng mL−1)were applied to the Upper and Lower channels(SA3T0G0jSB3T0G0jV50), resulting in no interstitial flow orVEGF gradient across the collagen gel (see Table 1 forexplanation of nomenclature). Invasion was minimalfrom both channels. (B) Without flow, application of 50 ngmL−1 VEGF-containing media in both HUVEC channelsunder static conditions results in dramatic invasion(SA0T0G0jSB0T0G0jV50 configuration). Solid blue VEGF barindicates uniform VEGF concentration in the collagen gel.(C) Normalized area of invasion in the 3D collagen gel fromsheared (SA3T0G0jSB3T0G0) and static (SA0T0G0jSB0T0G0)channels with various VEGF concentrations. (D) In theSA3T0G0jSB3T0G0jV50

flow configuration, the pan-NOS in-hibitor L-NMMA was added to the medium in the Upperchannel (SA3T0G0jL) resulting in significant invasion. (E)Normalized area of invasion in the 3D collagen gel for theflow configuration in D. Under shear flow, HUVEC invasionrequires both VEGF stimulation and NO inhibition by L-NMMA. Duration of all experiments was 3 d. Data pointson the graphs represent mean values and error bars depictSEM. Sample populations were compared using two-wayANOVA (row factor was day of treatment; column factorwas treatment condition). Statistical outcome indicated fortreatment condition (e.g., SA0T0G0jV50 vs. SA3T0G0jV50). n =21–28 per condition per day. ***P < 0.0001; ns, P = 0.33.Images represent a mosaic of 4 separate 10× fields ac-quired along the length of the device, spliced togetherautomatically using the Photomerge command in AdobePhotoshop (SI Materials and Methods, Image Acquisitionand Processing). (Scale bars, 100 μm.)

15344 | www.pnas.org/cgi/doi/10.1073/pnas.1105316108 Song and Munn

on the other hand, face the currently insurmountable challenge ofquantifying or controlling all of the various growth factors andmechanical forces in the microenvironment of a sprouting vesselsegment. Improvements to existing models of vessel developmentor repair are needed to understand why certain ECs sprout,whereas others are quiescent or are involved in vessel dilationor extension.By controlled application of VEGF and fluid forces in a

microfluidic device that accurately mimics angiogenic sprouting,we found that VEGF gradients and fluid forces cooperate tocontrol endothelial sprouting and morphogenesis. A majorfinding is that physiological shear stress attenuates VEGF-in-duced sprouting through NO signaling (Fig. 2 C and E). Thisfinding is consistent with the inhibition of EC proliferation bysteady laminar shear stress in vitro (27, 39, 40). In vivo, thisshear-controlled mechanism would tend to stabilize maturevessels to ensure that sprouting is initiated from vessels with lowor no flow, such as damaged or occluded vessels (41), or blind-ending sprouts (42). The fact that tumor blood vessels havehighly abnormal flow patterns (7, 43, 44), characterized by anabundance of segments with stagnant or slow flow, suggests thatshear-mediated sprout inhibition may be lacking in many vessels,thereby contributing to tumor angiogenesis. Although eNOS andNO have been implicated in tumor vessel remodeling (33), thelink between shear stress and morphogenesis has not beenestablished in tumor vessels. However, there is evidence thateNOS expression is sensitive to temporal or spatial gradients ofshear stress profiles in vitro (29, 45). In tissues where angio-genesis is occurring, we would expect temporal changes in shearstress—as vessels dilate, become leaky, or are occluded—andspatial gradients, which would be prominent at new branchpoints or those leading to occluded segments. It remains to beseen whether shear stress magnitude or spatial gradients of shearstress are more important in determining EC behavior. Furthercharacterization of the stress-sensing mechanisms may providenew targets for controlling sprouting angiogenesis in tumors.We also demonstrated that VEGF can cause either vessel

dilation or sprouting, depending on the nature of its local gra-dient (Fig. S10). Negative VEGF gradients resulted in a processresembling vessel dilation in our device (although restricted tothe regions of the apertures), whereas positive VEGF gradientsinduced filopodial extensions analogous to tip cell sprouts seen

in vivo (4, 5) (Fig. 4 A and B). This suggests that individual ECsaround the vessel circumference can sense their orientationrelative to the VEGF source to determine whether they shouldinitiate sprouting or simply participate in vessel dilation to in-crease overall flow to the tissue.Even more interesting is the finding that interstitial flow

encourages morphogenesis and filopod formation. Interstitialflow alone enhances the rate of EC invasion irrespective of thedirection of flow or VEGF gradient (Fig. 4D). The mechanism ofthe mechanotransduction is not clear, but previous studies havesuggested that transverse flow can act on mechanosensors locatedat cell–cell junctions of endothelial monolayers (46). Otherstudies have shown that shear stress applied to cells scattered ina 3D matrix can promote morphogenesis by activating membrane-bound sensors in either ECs (30) or vascular smooth muscle cells(SMCs) (47). The fact that more filopodia were produced by cellsmoving against the direction of interstitial flow in our study—evenin the absence of a VEGF gradient (Fig. 4D)—suggests thatfilopodia originating from tip cells at the leading edge of sproutsmay also be involved in probing the local flow environment. Wespeculate that, whereas VEGF guides sprouts toward hypoxicregions (1, 5, 9), interstitial flow directs them to other vessels,against the direction of interstitial flow, and consequently, towardvessels that have higher microvascular pressure than their own.The difference in hydrostatic pressure, which drives interstitialflow between the vessels, would also ensure effective perfusionupon connection of those two vessels.An example of how the various findings in our study would

apply during tissue revascularization is depicted in Fig. 5. In is-chemic, hypoxic tissue, some stagnant vessels are embedded inthe source of VEGF; ECs in these vessels can see a negativeVEGF gradient across the vessel wall and initiate dilation. An-other vessel farther away (Lower) sees a positive VEGF gradienton one side, but a negative gradient on the other. ECs in thissegment sprout or dilate, respectively. Other ECs, far enoughaway from the VEGF source or with sufficient flow, resiststructural changes (Upper). By following both VEGF and in-terstitial flow cues, the extending sprouts infiltrate the hypoxicregion, but also seek the leaky, higher pressure vessels instead offorming self-connections, which are less productive in terms oftissue perfusion.

Fig. 3. Shear stress attenuates HUVEC invasion irrespectiveof the direction of interstitial flow and VEGF gradient. (A) Inthe flow configuration SA3TaG+jSB0TwG−jV50, positive pres-sure shear flow (3 dyn cm−2) in channel A (Upper) results ininterstitial flow of 50 ngmL−1 VEGF-containingmedium at arate of 2.5 μm s−1 and a VEGF gradient from channelA to channel B (Lower). (B) In the flow configurationSA0TwG−jSB3TaG+jV50, negative pressure shear flow (3 dyncm−2) in channel B results in interstitial flow at a rate of 35μm s-1 and a VEGF gradient from A to B. Solid arrows in-dicate direction of axial flow in the HUVEC channels; dashedarrows indicate direction of interstitial flow. Blue gradientbar indicates VEGF gradient from A to B in the collagen gel.Images represent a mosaic of four separate 10× fields ac-quired along the length of the device, spliced together au-tomatically using the Photomerge command in AdobePhotoshop (SI Materials andMethods, Image Acquisition andProcessing). (Scale bars, 100 μm.) (C) Normalized area of in-vasion in the 3D collagen gel for the SA3TaG0jSB0TwG0jV0,SA3TaG+jSB0TwG−jV5, and SA3TaG+jSB0TwG−jV50

flow config-urations. (D) Normalized areaof invasion in 3D collagen spacefor the SA0TwG0jSB3TaG0jV0, SA0TwG−jSB3TaG+jV5, andSA0TwG−jSB3TaG+jV50

flow configurations. HUVEC invasionoccurs mainly from the nonsheared channels and irrespective of the direction of interstitial flow and VEGF gradients in channels A and B. Duration of allexperiments was 3 d. Data points represent mean + SEM. Statistical outcome indicated for treatment condition (V0 vs. V5 vs. V50). n = 21–35 per day per condition.***P < 0.0001; ns, P > 0.44.

Song and Munn PNAS | September 13, 2011 | vol. 108 | no. 37 | 15345

MED

ICALSC

IENCE

S

Thus, by integrating flow and VEGF signals, endothelial cellscan more efficiently determine (i) whether they reside ina damaged vessel and are candidates for sprouting (driven bychanges in luminal shear stress), (ii) in which direction to extendto revascularize hypoxic regions (driven by VEGF gradients),and (iii) which direction sprouts should extend to reach nearbyvessels with different pressure (driven by interstitial flow). Abetter understanding of how these signals are integrated to co-ordinate sprouting should lead to therapeutic strategies formodulating pathological angiogenesis.

Materials and MethodsThe microfluidic devices were fabricated from PDMS, using standard softlithography techniques (48). A collagen gel solution (3 mg mL−1 type I col-lagen, 10 μg mL−1 fibronectin) was localized to defined regions and thenpolymerized for at least 48 h at 37 °C and hydrated conditions. HUVECs wereseeded onto fibronectin-coated regions (10 μg mL−1). Shear stress levels inmicrofluidic channels were estimated using previously published data forchannels of comparable heights and aspect ratios (49, 50). Interstitial flowacross the collagen gel was measured experimentally by tracking fluid dis-placement over time in both the perfused and nonperfused sides of the flowconfiguration with 1 μm fluorescent beads. The chemical gradient profileswere measured experimentally in HUVEC-lined channels using TRITC-BSA(MW 66 kDa) dissolved in cell culture medium as fluorescent surrogate forVEGF. Filopodia formation was determined by two independent reviewerswho blindly analyzed and counted confocal projection images. To system-atically investigate whether VEGF and mechanical fluid forces cooperateduring the initiation of angiogenesis, we use the nomenclature summarizedin Table 1. Additional and more detailed methods are described in SIMaterials and Methods.

ACKNOWLEDGMENTS. We are grateful to P. Au, G. Cheng, and J. Tse forperforming the retroviral transfections of the HUVECs provided to us by theCenter for Vascular Excellence at Brigham and Women’s Hospital. We thankR. Samuel for her assistance with counting filopodia, A. Jain for helpfuldiscussions, and R. Jain for his invaluable insight and suggestions. Fundingwas provided by the National Cancer Institute (L.L.M).

Fig. 4. Sprout morphogenesis is affected by the direction of interstitial flowand VEGF gradient. (A–C) Filopodia formation in sprouting HUVECs in the(A) SB0TaG+jV50, (B) SA0TwG−jV50, and (C) SA0TwG0jSA0TwG0jV50

flow con-figurations. Interstitial flow rates were (A) 2.5, (B) 35, and (C) 28.5 μm/s. Eachimage depicts a single aperture imaged 2–3 d after initiation of experiment.(Scale bars, 100 μm.) (D) Quantification of the number of filopodia persprouting area produced by sprouting HUVECs. n = 5–21. *P < 0.05; **P <0.001; ns, P = 0.96. Gradient blue and solid blue VEGF bars indicate VEGFgradient and uniform VEGF concentration, respectively, in the collagen gel.Dashed arrows indicated direction of interstitial flow. (E) Isolated effects ofinterstitial flow and VEGF gradient on sprouting area, comparing the nor-malized sprout area at day 3 from each sprouting direction (with interstitialflow and a negative VEGF gradient or against interstitial flow with a positiveVEGF gradient). Interstitial flow alone significantly enhances sprout area forboth sprouting directions (***). Addition of a VEGF gradient to interstitialflow significantly increases sprout area compared with interstitial flow only,for both sprouting directions (***9). Furthermore, the normalized area ofsprouting for interstitial flow only (B/A: 39 ± 5; A/B: 37 ± 6) plus VEGFgradient only (B/A: 28 ± 5; A/B: 23 ± 4) is comparable to the area whenboth VEGF gradient and interstitial flow are present (B/A: 65 ± 10; A/B:56 ± 7), suggesting that these components are additive in enhancing sproutarea for both sprouting directions. n = 21–49. Data points represent mean +SEM. *** or ***9P < 0.0001.

Fig. 5. Putative integration of flow forces and VEGF in damaged tissue. (A)Hypoxic cells (blue) residing in an ischemic region produce VEGF (small bluedots). The damaged, occluded vessel within this region lacks shear stress, butfills with VEGF, and its ECs sense a negative VEGF gradient. This vessel dilates(walls outlined in pink) and becomes leaky in response (arrows; note thatP1 > P2). Proximal ECs in the other occluded vessel (Lower) see a positiveVEGF gradient across the vessel wall (orange outline), whereas those in theopposite wall see a negative gradient (pink outline). The former sprout,whereas the latter dilate. Morphogenesis of nondamaged, well-perfusedvessels is inhibited by shear stress (green outline). (B and C) Interstitial flowsupports efficient revascularization. If no fluid forces were involved withmediating revascularization and instead relied solely on VEGF and otherchemical factors, then many self-connections would be made between thesprouting vessels, leading to inefficient reperfusion (B). However, interstitialflow originating from the leaky and dilated central vessel serves as an im-portant cue that guides the sprouts toward this higher-pressure vessel toensure more uniform revascularization of the central region (C).

15346 | www.pnas.org/cgi/doi/10.1073/pnas.1105316108 Song and Munn

1. Carmeliet P, Jain RK (2000) Angiogenesis in cancer and other diseases. Nature 407:249e257.

2. Davis GE, Bayless KJ, Mavila A (2002) Molecular basis of endothelial cell morpho-genesis in three-dimensional extracellular matrices. Anat Rec 268:252e275.

3. Ingber DE (2002) Mechanical signaling and the cellular response to extracellularmatrix in angiogenesis and cardiovascular physiology. Circ Res 91:877e887.

4. Carmeliet P, De Smet F, Loges S, Mazzone M (2009) Branching morphogenesis andantiangiogenesis candidates: Tip cells lead the way. Nat Rev Clin Oncol 6:315e326.

5. Gerhardt H, et al. (2003) VEGF guides angiogenic sprouting utilizing endothelial tipcell filopodia. J Cell Biol 161:1163e1177.

6. Patan S, Munn LL, Jain RK (1996) Intussusceptive microvascular growth in a humancolon adenocarcinoma xenograft: A novel mechanism of tumor angiogenesis. Mi-crovasc Res 51:260e272.

7. Jain RK (2005) Normalization of tumor vasculature: An emerging concept in anti-angiogenic therapy. Science 307:58e62.

8. Harris AL (2002) Hypoxia—a key regulatory factor in tumour growth. Nat Rev Cancer2:38e47.

9. Ferrara N, Gerber H-P, LeCouter J (2003) The biology of VEGF and its receptors. NatMed 9:669e676.

10. Yancopoulos GD, et al. (2000) Vascular-specific growth factors and blood vessel for-mation. Nature 407:242e248.

11. Hellström M, et al. (2007) Dll4 signalling through Notch1 regulates formation of tipcells during angiogenesis. Nature 445:776e780.

12. Davies PF (1995) Flow-mediated endothelial mechanotransduction. Physiol Rev 75:519e560.

13. Frangos JA, Eskin SG, McIntire LV, Ives CL (1985) Flow effects on prostacyclin pro-duction by cultured human endothelial cells. Science 227:1477e1479.

14. Shyy JY, Chien S (2002) Role of integrins in endothelial mechanosensing of shearstress. Circ Res 91:769e775.

15. Topper JN, Cai J, Falb D, Gimbrone MA, Jr. (1996) Identification of vascular endo-thelial genes differentially responsive to fluid mechanical stimuli: Cyclooxygenase-2,manganese superoxide dismutase, and endothelial cell nitric oxide synthase are se-lectively up-regulated by steady laminar shear stress. Proc Natl Acad Sci USA 93:10417e10422.

16. Hernández Vera R, et al. (2009) Interstitial fluid flow intensity modulates endothelialsprouting in restricted Src-activated cell clusters during capillary morphogenesis.Tissue Eng Part A 15:175e185.

17. Tada S, Tarbell JM (2000) Interstitial flow through the internal elastic lamina affectsshear stress on arterial smooth muscle cells. Am J Physiol Heart Circ Physiol 278:H1589eH1597.

18. Garcia-Cardeña G, Comander J, Anderson KR, Blackman BR, Gimbrone MA, Jr. (2001)Biomechanical activation of vascular endothelium as a determinant of its functionalphenotype. Proc Natl Acad Sci USA 98:4478e4485.

19. White CR, Frangos JA (2007) The shear stress of it all: The cell membrane andmechanochemical transduction. Philos Trans R Soc Lond B Biol Sci 362:1459e1467.

20. Wang Y, et al. (2007) Selective adapter recruitment and differential signaling net-works by VEGF vs. shear stress. Proc Natl Acad Sci USA 104:8875e8879.

21. Kang H, Bayless KJ, Kaunas R (2008) Fluid shear stress modulates endothelial cell in-vasion into three-dimensional collagen matrices. Am J Physiol Heart Circ Physiol 295:H2087eH2097.

22. Tressel SL, Huang RP, Tomsen N, Jo H (2007) Laminar shear inhibits tubule formationand migration of endothelial cells by an angiopoietin-2 dependent mechanism. Ar-terioscler Thromb Vasc Biol 27:2150e2156.

23. Hueck IS, Rossiter K, Artmann GM, Schmid-Schönbein GW (2008) Fluid shear at-tenuates endothelial pseudopodia formation into the capillary lumen. Microcircula-tion 15:531e542.

24. Price GM, et al. (2010) Effect of mechanical factors on the function of engineeredhuman blood microvessels in microfluidic collagen gels. Biomaterials 31:6182e6189.

25. Hudlicka O, Brown MD, May S, Zakrzewicz A, Pries AR (2006) Changes in capillaryshear stress in skeletal muscles exposed to long-term activity: Role of nitric oxide.Microcirculation 13:249e259.

26. Jones EAV, le Noble F, Eichmann A (2006) What determines blood vessel structure?Genetic prespecification vs. hemodynamics. Physiology (Bethesda) 21:388e395.

27. Lin K, et al. (2000) Molecular mechanism of endothelial growth arrest by laminarshear stress. Proc Natl Acad Sci USA 97:9385e9389.

28. Helmlinger G, Geiger RV, Schreck S, Nerem RM (1991) Effects of pulsatile flow oncultured vascular endothelial cell morphology. J Biomech Eng 113:123e131.

29. Blackman BR, García-Cardeña G, Gimbrone MA, Jr. (2002) A new in vitro model toevaluate differential responses of endothelial cells to simulated arterial shear stresswaveforms. J Biomech Eng 124:397e407.

30. Helm CL, Fleury ME, Zisch AH, Boschetti F, Swartz MA (2005) Synergy between in-terstitial flow and VEGF directs capillary morphogenesis in vitro through a gradientamplification mechanism. Proc Natl Acad Sci USA 102:15779e15784.

31. Adamo L, et al. (2009) Biomechanical forces promote embryonic haematopoiesis.Nature 459:1131e1135.

32. Sessa WC (2004) eNOS at a glance. J Cell Sci 117:2427e2429.33. Kashiwagi S, et al. (2008) Perivascular nitric oxide gradients normalize tumor vascu-

lature. Nat Med 14:255e257.34. Suchting S, et al. (2007) The Notch ligand Delta-like 4 negatively regulates endothelial

tip cell formation and vessel branching. Proc Natl Acad Sci USA 104:3225e3230.35. Djonov V, Schmid M, Tschanz SA, Burri PH (2000) Intussusceptive angiogenesis: Its role

in embryonic vascular network formation. Circ Res 86:286e292.36. Jones EAV, Baron MH, Fraser SE, Dickinson ME (2004) Measuring hemodynamic

changes during mammalian development. Am J Physiol Heart Circ Physiol 287:H1561eH1569.

37. Gruionu G, Hoying JB, Gruionu LG, Laughlin MH, Secomb TW (2005) Structural ad-aptation increases predicted perfusion capacity after vessel obstruction in arteriolararcade network of pig skeletal muscle. Am J Physiol Heart Circ Physiol 288:H2778eH2784.

38. Skalak TC (2005) Angiogenesis and microvascular remodeling: A brief history andfuture roadmap. Microcirculation 12:47e58.

39. Akimoto S, Mitsumata M, Sasaguri T, Yoshida Y (2000) Laminar shear stress inhibitsvascular endothelial cell proliferation by inducing cyclin-dependent kinase inhibitorp21(Sdi1/Cip1/Waf1). Circ Res 86:185e190.

40. Levesque MJ, Nerem RM, Sprague EA (1990) Vascular endothelial cell proliferation inculture and the influence of flow. Biomaterials 11:702e707.

41. Yamada H, et al. (1999) Hyperoxia causes decreased expression of vascular endo-thelial growth factor and endothelial cell apoptosis in adult retina. J Cell Physiol 179:149e156.

42. Anderson CR, Hastings NE, Blackman BR, Price RJ (2008) Capillary sprout endothelialcells exhibit a CD36 low phenotype: Regulation by shear stress and vascular endo-thelial growth factor-induced mechanism for attenuating anti-proliferative throm-bospondin-1 signaling. Am J Pathol 173:1220e1228.

43. Kamoun WS, et al. (2010) Simultaneous measurement of RBC velocity, flux, hemat-ocrit and shear rate in vascular networks. Nat Methods 7:655e660.

44. Munn LL (2003) Aberrant vascular architecture in tumors and its importance in drug-based therapies. Drug Discov Today 8:396e403.

45. Kuchan MJ, Frangos JA (1994) Role of calcium and calmodulin in flow-induced nitricoxide production in endothelial cells. Am J Physiol 266:C628eC636.

46. Tarbell JM, Demaio L, Zaw MM (1999) Effect of pressure on hydraulic conductivity ofendothelial monolayers: Role of endothelial cleft shear stress. J Appl Physiol 87:261e268.

47. Shi Z-D, Wang H, Tarbell JM (2011) Heparan sulfate proteoglycans mediate interstitialflow mechanotransduction regulating MMP-13 expression and cell motility via FAK-ERK in 3D collagen. PLoS ONE 6:e15956.

48. Duffy DC, McDonald JC, Schueller OJA, Whitesides GM (1998) Rapid Prototyping ofMicrofluidic Systems in Poly(dimethylsiloxane). Anal Chem 70:4974e4984.

49. Gaver DP, 3rd, Kute SM (1998) A theoretical model study of the influence of fluidstresses on a cell adhering to a microchannel wall. Biophys J 75:721e733.

50. Zeng Y, Lee TS, Yu P, Roy P, Low HT (2006) Mass transport and shear stress in a mi-crochannel bioreactor: Numerical simulation and dynamic similarity. J Biomech Eng128:185e193.

Song and Munn PNAS | September 13, 2011 | vol. 108 | no. 37 | 15347

MED

ICALSC

IENCE

S