Fluid shear stress threshold regulatesangiogenic sproutingPeter A. Galiea,b, Duc-Huy T. Nguyenc, Colin K. Choia, Daniel M. Cohena, Paul A. Janmeyb, and Christopher S. Chena,c,d,e,1

Departments of aBioengineering, bPhysiology, and cChemical and Biomolecular Engineering, University of Pennsylvania, Philadelphia, PA 19104; dDepartmentof Biomedical Engineering, Boston University, Boston, MA 02215; and eThe Wyss Institute for Biologically Inspired Engineering, Harvard University, Boston,MA 02115

Edited by Sheldon Weinbaum, The City College of New York, New York, NY, and approved April 21, 2014 (received for review June 7, 2013)

The density and architecture of capillary beds that form within atissue depend on many factors, including local metabolic demandand blood flow. Here, using microfluidic control of local fluid me-chanics, we show the existence of a previously unappreciated flow-induced shear stress threshold that triggers angiogenic sprouting.Both intraluminal shear stress over the endothelium and transmu-ral flow through the endothelium above 10 dyn/cm2 triggered en-dothelial cells to sprout and invade into the underlying matrix, andthis threshold is not impacted by the maturation of cell–cell junc-tions or pressure gradient across the monolayer. Antagonizing VE-cadherin widened cell–cell junctions and reduced the applied shearstress for a given transmural flow rate, but did not affect the shearthreshold for sprouting. Furthermore, both transmural and luminalflow induced expression of matrix metalloproteinase 1, and thisup-regulation was required for the flow-induced sprouting. Oncesprouting was initiated, continuous flow was needed to both sus-tain sprouting and prevent retraction. To explore the potentialramifications of a shear threshold on the spatial patterning ofnew sprouts, we used finite-element modeling to predict fluidshear in a variety of geometric settings and then experimentallydemonstrated that transmural flow guided preferential sproutingtoward paths of draining interstitial fluid flow as might occur toconnect capillary beds to venules or lymphatics. In addition, weshow that luminal shear increases in local narrowings of vesselsto trigger sprouting, perhaps ultimately to normalize shear stressacross the vasculature. Together, these studies highlight the role ofshear stress in controlling angiogenic sprouting and offer a poten-tial homeostatic mechanism for regulating vascular density.

angiogenesis | force | mechanotransduction | migration | morphogenesis

The density of capillary blood vessels varies widely acrossdifferent organs and tissues and is determined by the ability

of unmet local metabolic needs to trigger angiogenesis. Perhapsmost well characterized is the induction of VEGF expression byparenchymal hypoxia, leading to angiogenesis that persists untilthe hypoxia is relieved by subsequent enhanced tissue perfusion(1). Indeed, significant advances have been made in understandingthe mechanisms by which numerous biochemical stimuli induceendothelial sprouting (2). Importantly, excess metabolic demandalso triggers enhanced local circulation by relaxation of upstreamarterioles (3) and results in increased blood flow to these regions. Inaddition to enhanced delivery of nutrients, the increased blood flowalso increases shear stress on the luminal surface of the endothe-lium, which in some studies has been shown to induce capillarygrowth in skeletal muscle (4, 5), whereas others showed shear stressenhances endothelial barrier function and inhibits sprouting (6–9).Unlike luminal shear, transmural flow, or fluid flow exiting the

wall of the vessel, is universally accepted to induce sprouting ofendothelial cells into the extracellular matrix, at least in in vitromodels (9–11). Transmural flow is directed normal to the surfaceof cells and thus simultaneously exerts both a pressure stressagainst the apical surface of the cell and a shear stress concen-trated at the cell membrane adjacent to intercellular junctions(12–14). The present study seeks to elucidate the reasons that

transmural flow consistently induces sprouting whereas the effectsof luminal flow are more controversial, despite both flows exertingshear stress on the endothelium.Here, we use a series of microfabricated microfluidic devices to

demonstrate that a threshold of shear stress exists above whichcells will sprout regardless of whether the shear is exerted bytransmural or luminal fluid flow. Moreover, we demonstrate howthe shear threshold can guide enhanced sprouting at regions ofconstricted diameter and focus sprouting in the direction of trans-mural flow. These results underscore the importance of shear stressas a crucial parameter for the initiation of angiogenesis and theregulation of structures within a vasculature bed and highlight theinterplay between mechanical, structural, and biochemical signalsthat cells use to organize and adapt the vasculature to meet its manyfunctional demands.

ResultsBoth Transmural and Luminal Flow-Induced Sprouting Are Triggeredby a Common Threshold of Shear Stress. To test the effect of trans-mural flow on sprouting, we first seeded cells on a constrained2-mg/mL collagen gel of known permeability (15) formed as a plugwithin a flow chamber (Fig. S1 B, i), allowed them to attach anddevelop a confluent monolayer over 36 h, and then appliedtransmural flow to the monolayer. After 24 h of transmural flowresulting in junctional velocities greater than 5 μm/s, we observedcopious sprouting of cells into the collagen gel along the mono-layer of seeded cells (Fig. 1 A and B). Quantifying the length ofsprouts, it was evident that this effect depended sharply on thejunctional velocity (Fig. 1C). However, changing the junctional

Significance

A great deal of research has investigated the biochemical fac-tors that regulate angiogenic sprouting, but less is knownabout the role of fluid shear stress. Some studies have sug-gested distinct regulation by luminal flow within the vessel vs.transmural flow through its walls. In this paper, we demon-strate the existence of a shear stress threshold that whensurpassed, induces angiogenic sprouting regardless of whetherthe shear is applied by primarily luminal or transmural flow. Inaddition to identifying matrix metalloproteinase 1 as the rel-evant downstream effector, we use finite-element modeling topredict spatial distributions of shear stress within 3D geome-tries that experimentally caused localized patterns of sprout-ing. Together, these studies demonstrate a means by whichfluid flow can guide vasculature architecture.

Author contributions: P.A.G., D.M.C., and P.A.J. designed research; P.A.G. performed re-search; D.-H.T.N. and C.K.C. contributed new reagents/analytic tools; P.A.G. and C.S.C.analyzed data; and P.A.G. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.1To whom correspondence should be addressed. E-mail: chencs@bu.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1310842111/-/DCSupplemental.

7968–7973 | PNAS | June 3, 2014 | vol. 111 | no. 22 www.pnas.org/cgi/doi/10.1073/pnas.1310842111

velocity impacts both the pressure force normal to the luminal faceof the endothelial cells and the shear stress on cell–cell junctions.To distinguish whether the sprouting response was dictated by

velocity, pressure, or shear, we modulated junctional integrity todecouple these effects. That is, varying the distance between cellsby loosening cell–cell junctions would allow us to reduce appliedpressure drop and shear stress for a specified junctional velocity.We used atomic force microscopy (AFM) and transmissionelectron microscopy (TEM) to determine the precise geometryand spacing of the junctions between cells to estimate the re-lationship between pressure drop, junctional velocity, and shearstress magnitude at the cell–cell junctions (Figs. S2 and S3). Cellswere transduced to express a mutant VE-cadherin lacking itsβ-catenin binding domain (VEΔ). We chose VE-cadherin be-cause of its modulation of vascular permeability and angiogen-esis (16) and involvement in fluid shear mechanotransduction(17). We verified the effects of VEΔ by immunofluorescence ofmonolayers exposed to perfusion with media containing sus-pended 1-μm beads (Fig. 1 E–G). Control monolayers exhibitedlocalization of β-catenin to the cell–cell junctions (green), andthe fluorescent beads (red) settled throughout the monolayer(Fig. 1E), whereas VEΔ-treated monolayers exhibited β-cateninonly partially localized to cell–cell junctions with substantialperinuclear staining, and beads again were scattered homoge-neously (Fig. 1F). Cells treated with VEΔ required a higherjunctional velocity to sprout, but a lower applied pressure dropcompared with control monolayers (Fig. 1 C and D). Theseresults suggest that neither pressure drop nor junctional velocityis the universal quantity that determines sprouting. Because directinhibition of VE-cadherin function could impact other signaling

functions such as modulation of VEGF receptor signaling, we alsoaltered junctional geometry by examining flow-induced sproutingfrom immature monolayers. Here, we applied flow 30 min aftercell seeding, rather than waiting 36 h for a mature monolayer toform. The resulting immature monolayers exhibited hydraulicconductivities several orders of magnitude greater than valuescalculated for the mature monolayers of control and VEΔ-treated cells (5.91E-8 and 1.03E-6 cm·s−1·cm−1 H2O), whichare within the range of previous values reported for endo-thelial monolayers (18). Indeed, for the immature monolayers,beads accumulated in large gaps within the monolayer (Fig. 1G).The immature monolayers sprouted in response to substantiallyhigher junctional velocities and lower pressure drops comparedwith controls (Fig. 1 C and D), again suggesting that these arenot the physical factors transduced by cells to induce sprouting.Instead, replotting our results as a function of shear stress in-dicated that sprouting occurred at the same shear stress threshold(∼ 10 dyn/cm2) in all three conditions (Fig. 1H).To test whether luminal shear stress had similar or distinct

effects on sprouting, we seeded cells in a cylindrical channel withinthe collagen hydrogel, using a method adapted from Tien and co-workers (19) (Fig. S1 B, ii). In this configuration, computationalmodeling verified that flow along the channels resulted in primarilyluminal shear with minimal transmural shear (Fig. S4). We againobserved a threshold for sprouting similar to that of transmural flow(Fig. 1 I, i and ii). This finding indicates that the location whereshear stress was exerted, whether at cell–cell junctions or alongthe luminal (apical) face of the cell, did not affect the thresholdrequired for sprouting. Moreover, VEΔ-treated monolayerssprouted at the same threshold of luminal shear as untreated,

Immature

B

H Flow

E F G

I,i

I,ii

A TransmuralTransmural Flow FlowStaticStatic

DAPIPhalloidin

DAPIBeta-catenin

Beads

MatureMature VEVE

I

StaticStatic

Luminal FlowLuminal Flow

ImmatureImmature

FlowFlow

Transmural Shear Stress ( )dyn cm2

Spro

ut

Len

gth

(um

)

Luminal Shear Stress ( )dyn cm2

Spro

ut

Len

gth

(um

)

*

Spro

ut

Len

gth

(um

)

Junctional Velocity (um/s)

DC

Spro

ut

Len

gth

(um

)

Pressure Drop (kPa)

VE (M)Mature

Immature

0 10 20 30 200 400 600 0 0.005 0.5 1.0 1.50

20

40

60

80

100

120

0

20

40

60

80

100

120

MatureVE

Immature

MatureVE

J

Fig. 1. (A and B) Schematic and representative image of a mature endothelial cell monolayer plated on the surface of a collagen gel (A) and sprouting inresponse to perfusion with transmural flow (B). (Scale bar, 50 μm.) (C and D) Mean sprout length of mature untreated and VEΔ monolayers and immaturemonolayers exposed to transmural flow of several junctional velocities (C) and pressure drops (D). Linear ANOVAs followed by Tukey tests were used todetermine that the sprouting responses of control and VEΔ groups were significantly different (*P < 0.05). (E–G) Monolayer of mature untreated (E) and VEΔ(F) and immature (G) conditions. Green, β-catenin; blue, nucleus; red, 1-μm beads. The white arrowheads point to the distributed beads accumulated on themonolayer surface. (Scale bar, 10 μm.) (H) Mean sprout length as a function of transmural shear for nontreated mature, VEΔ mature, and immature mon-olayers. Again, post hoc Tukey tests were conducted to calculate the significant difference in the reduced serum group (*P < 0.05). (I) Surface of cylindricalchannel for static (I, i) and luminal flow (I, ii) conditions. Green, F-actin; blue, nucleus. (Scale bar, 100 μm.) (J) Mean sprout length as a function of luminal shearfor nontreated and VEΔ cells and cells exposed to 0.5% serum.

Galie et al. PNAS | June 3, 2014 | vol. 111 | no. 22 | 7969

ENGINEE

RING

APP

LIED

BIOLO

GICAL

SCIENCE

S

mature monolayers, confirming our previous conclusion thatwidening cell–cell junctions suppressed transmural flow-inducedsprouting for a given junctional velocity by decreasing shear stressat the cell–cell junctions, not by disrupting the ability of cells torespond to mechanical cues.The observed shear threshold of ∼10 dyn/cm2 is lower than the

physiological shear that occurs in nonsprouting vessels in vivo,suggesting that additional aspects of the in vivo context may in-crease the threshold for the sprouting response. One possibility isthat in vitro culture uses a high level of serum, which containsproangiogenic factors from lysed platelets, whereas in vivo plasmais relatively replete of growth factors. We therefore perfused flowusing medium with reduced serum (0.5% vs. 2%) and observedan increase in the threshold of shear above which sproutingoccurs (20–30 dyn/cm2) (Fig. 1J). These results suggest thatalthough we have identified a previously unappreciated shear stressthreshold for inducing angiogenic sprouting, the value of thatthreshold may shift, depending on the environmental context.

The Downstream Effector for Flow-Induced Sprouting Is MatrixMetalloproteinase 1. Because luminal flow acts on the apicalsurface of the endothelial cell whereas transmural flow acts on thecell–cell junctions, it was not clear whether the mechanisms forsprouting in the two settings were shared or distinct. Here, weexplored by quantitative (q)RT-PCR whether transmural and lu-minal flows altered expression of selected members of the matrixmetalloproteinase (MMP) family, whose expression is critical in3D cell migration and sprouting (20). Cells first were exposed todifferent levels of transmural flow for 24 h in the same config-urations detailed in Fig. 1 and lysed for analysis of MMP expres-sion. MMP1 was the only member of the MMP family significantlyup-regulated in response to flow (Fig. 2A). Importantly, the degreeof MMP1 expression was sensitive to the level of applied shearstress such that it increased most dramatically at the samethreshold when sprouting became substantial. None of the othermeasured MMPs changed expression in response to shear stress.

Interestingly, in cells exposed to reduced serum, MMP1 up-regulation again mirrored the sprouting response. More luminalshear stress was required to significantly increase MMP1 ex-pression in cells treated with reduced serum (Fig. 2B). Luminalshear stress also up-regulated MMP1 only at the established shearthreshold of ∼10 dyn/cm2 for 2% serum. (Fig. 2C). These resultsconfirmed that not only do both transmural and luminal shearstresses induce sprouting at the same threshold, but also the up-regulation of MMP1 correlated with the sprouting response.To examine whether the shear stress-induced MMP1 up-

regulation was causally required for the sprouting response,we transiently transfected a dicer substrate interference RNA(DsiRNA) targeting MMP1 into endothelial cells and exposedcells to 22 dyn/cm2 of transmural shear stress. Silencing the MMP1message blunted the increased sprout length in response totransmural shear. Conversely, a scrambled DsiRNA constructdid not have any significant effect on the sprouting response(Fig. 2D). Mamaristat, a broad spectrum MMP inhibitor (21),similarly blocked flow-induced sprouting and confirmed thatMMP1 is the crucial matrix metalloproteinase for flow-inducedsprouting. To validate the knockdown of the DsiRNA, we usedqRT-PCR to analyze the expression of MMP1 in response tocontrol and 22 dyn/cm2 transmural shear stress conditions. Fig.2E indicates that in response to flow, the siRNA knockdownprevents the augmented expression of MMP1, whereas a scram-bled construct does not.

Fig. 2. (A) MMPmessage levels for MMP1,MMP2, MMP9, MMP10, andMMP14double normalized to GAPDH and static controls in response to transmural shearstress. Two-sample t tests were performed for each MMP primer to determinestatistical significance between the applied shear and static controls (*P <0.05). (B) MMP message levels for cells exposed to luminal flow and reducedserum (*P < 0.05). (C) MMP message levels in response to luminal flow (*P <0.05). (D) Mean sprout length for cells treated withMMP1 DsiRNA, a scrambledDsiRNA control, and Mamaristat. Two-sample t tests were calculated to test fora significant change from untreated controls (*P < 0.05). (E) Validation ofsuccessful knockdown by qRT-PCR. Transmural flow only significantly increasesmessage levels of MMP1 (*P < 0.05).

DC

E

B **

*

A

0 - 24 hr.flow

24 - 32 hr.no flow

32- 56 hr.flow

0 - 24 hr.flow

24 - 32 hr.flow+Mm.

32- 56 hr.flow-Mm.

Spro

ut

Len

gth

(um

)

Spro

ut

Len

gth

(um

)

F

0

50

100

150

200

0

50

100

150

200

Fig. 3. (A and B) Mean sprout length after stopping and restarting fluidflow (A) and temporary Mamaristat treatment (B). Two-sample t tests werecalculated to determine a significant change from the initial 24-h sproutingresponse (*P < 0.05). (C and D) Representative image of sprouts after 8 h ofstatic conditions (C) or Mamaristat treatment (D). Green, F-actin; blue, nu-cleus. (Scale bar, 50 μm.) (E and F) Representative image of sprouts after the24-h period following resumption of flow (E) or washing out of Mamaristat(F). White arrowheads point to one of the single-cell sprouts observed in thehalted-flow group (E) and to the multicellular structures observed in thecontinuous-flow group (F). Green, F-actin; blue, nucleus. (Scale bar, 50 μm.)

7970 | www.pnas.org/cgi/doi/10.1073/pnas.1310842111 Galie et al.

Transmural Flow Is Required to Sustain Sprouts. Having establishedthe role of shear stress in inducing endothelial cells to sproutfrom a monolayer, we next examined whether flow is necessaryto sustain a sprout after it has already initiated. We thereforeapplied transmural flow to cells for 24 h at a flow rate resulting in11 dyn/cm2 of shear stress, a quantity known to induce sprouting.After quantifying sprout length, we halted the flow for 8 h. Weobserved a significant retraction of the sprouts (Fig. 3 A and C).After restarting flow for an additional 24 h, the sprouts recoveredand continued to grow. Because we identified MMP1 as thedownstream effector of transmural shear stress, we sought todetermine whether blocking MMP activity could produce thesame dynamic response as the stop/restart response. After ap-plying 11 dyn/cm2 of transmural shear stress for 24 h, we addedMamaristat to the perfusing medium for 8 h. Unlike the flow-stopresponse, Mamaristat simply prevented further extension of sproutswithout apparent retraction (Fig. 3 B and D). After washing outthe inhibitor, application of transmural flow for an additional 24 hrescued continued extension.Confocal microscopy indicated that the structure of the dy-

namic sprouting response also differed between changes in flowvs. MMP inhibition. Following resumption of flow, endothelialcells recommenced sprouting into the collagen gel, as shown inFig. 3 A and E, but mostly as single cells or sprouts containingonly two cells (Fig. 3E). In contrast, after washing the Mamaristatfrom the system, the endothelial cells also continued to sprout butdid so in the same classical multicellular structures as observed incontrol sprouts (Fig. 3 B and F). This result indicates that theaction of stopping and resuming fluid flow disrupted the co-ordinated migration of endothelial cells required to form a mul-ticellular sprout. Interestingly, at the end of both protocols (stop/restart flow, transient Mamaristat treatment), the final meansprout lengths at 56 h were not significantly different from eachother or from those of sprouts under uninterrupted flow (Fig. 3 Aand B). This implies that the rate of sprouting following re-sumption of flow was greater than the rate during the initial 24 hof transmural flow application. Overall, these results suggest that

shear stress from fluid flow not only induces endothelial sproutingbut also sustains the sprout and prevents retraction.

Three-Dimensional Focusing of Transmural and Luminal Flows CausesDirected Sprouting. Because one can engineer spatially focusedpatterns of flow and shear, we explored whether shear-inducedsprouting can drive spatial patterns of sprouting. To focus trans-mural flow, we generated closed-ended voids on opposing sides ofa collagen gel such that flow would follow the shortest path acrossthe gel, entering the closed-end, cell-seeded channel, and exitthrough the wall closest to an acellular void. We constructed finite-element models to predict the fluid shear stress distributions forthe different configurations. Fig. 4A shows the resulting shearstress distribution of one geometry: two acellular voids placedopposite the cell-seeded void, focusing flow on two corners of thevoid. Cells were seeded in this geometry and subjected to trans-mural flow for 48 h. The computational model verified thattransmural shear stress exceeded luminal shear stress by ap-proximately two orders of magnitude (Fig. S5). Sprouting ap-peared to track along the predicted regions of highest transmuralshear (Fig. 4D) symmetrically along the two corners of the void.Because the frequency of sprouting is somewhat limited, westacked multiple images of different constructs into a single plot(Fig. 4G), showing the increased frequency of sprouting at bothcorners symmetrically in response to fluid flow. To further il-lustrate this effect, we generated a second geometry in whichonly one acellular void is placed opposite the cell-seeded void,such that the transmural flow is focused asymmetrically at onlyone corner, and the inlet velocity was tuned so that the peakshear stress matched that of the first configuration (Fig. 4B). Asbefore, experimental measurements of endothelial sprouting inthis configuration again matched the flow profile, in which thehighest sprouting frequency was skewed to the corner adjacent tothe acellular void (Fig. 4E). Stacking multiple images of differentconstructs again demonstrated the increased frequency of sproutingat the corner adjacent to the void (Fig. 4H).We were also able to spatially pattern the magnitude of lu-

minal flow, by using a channel with a progressively decreasing

Fig. 4. (A–C) Finite-element model predictions ofshear stress for the two-void, transmural flow con-figuration (A), the single-void configuration (B), andthe endothelial-lined nozzle (C). (D–F) Representa-tive images of the flow-induced sprouting for thethree configurations: two-void (D), single-void (E),and nozzle (F). Green, F-actin; blue, nucleus. (Scalebar, 200 μm.) (G–I) Frequency plots generated by atleast five images of sprouting in the three config-urations: two-void (G), single-void (H), and nozzle (I).

Galie et al. PNAS | June 3, 2014 | vol. 111 | no. 22 | 7971

ENGINEE

RING

APP

LIED

BIOLO

GICAL

SCIENCE

S

radius (henceforth referred to as a “nozzle”). In this configura-tion, shear stress changes continuously as a function of the ra-dius, allowing for focusing of luminal shear at certain regions ofthe nozzle surface. The finite-element model indicated that lumi-nal shear stress exceeded unintended transmural shear stress byabout an order of magnitude (Fig. S5). Fig. 4C shows the shearstress exerted by luminal flow within the nozzle. We used an inputflow rate that resulted in the shear threshold occurring approxi-mately midlength of the nozzle. The images and resulting contourplot (Fig. 4 F and I) show that the frequency of sprouting wassubstantially increased once the shear stress threshold was sur-passed within the nozzle. These results have important implica-tions for flow-induced sprouting in vivo, in addition to verifying theexistence of a shear stress threshold, using a different geometry.

DiscussionThis study reveals the existence of a shear stress threshold abovewhich sprouting is induced. This threshold response occurs re-gardless of whether the flow is transmural or luminal, in apparentcontrast to previous studies that have reported enhanced sproutingwith transmural flow (9–11, 22, 23), but reduced sprouting withluminal shear (9, 24). It is now clear that the reason for the distinctresponses was that transmural shears were high whereas the lu-minal shear used was low. Given that we observed sproutingwhether the shear was applied across the apical surface or at cell–cell junctions, it is not surprising that junctional proteins such asVE-cadherin do not participate in transducing the shear stress intothe sprouting response, beyond impacting the width between cellsthrough which shear is generated. Although this shear-inducedsprouting shares some superficial similarities to the luminal shearstress-induced alignment response of endothelial cells, in that bothinvolve transduction of shears into cytoskeletal reorganization,luminal shear transduction requires a VE-cadherin–containingcomplex (17, 25). Interstitial flows around single cells embeddedwithin matrices also trigger directed migration, but in such settings,shear stresses as low as 0.1 dyn/cm2, two orders of magnitudebelow our reported threshold, are sufficient to drive the response(26–29). Taken together, these studies suggest the conservation ofa fundamental link whereby cell movement is stimulated by flow,although the detailed triggers and mechanisms have diverged fordifferent situations.The mechanisms by which shear stress induces sprouting are

likely multifaceted. Here we show that shear stress above the10 dyn/cm2 threshold specifically up-regulates the expression ofMMP1 but not that of other MMPs. Interestingly, interstitialflow has been reported to up-regulate MMP1 also in migratingsmooth muscle cells (30, 31), although several previous reportshave shown that shear stress decreases expression of MMP2and MT1-MMP (32–35). Again, this disparity may result fromdifferences in the levels of shear stress applied to cells or thecontext in which they have been applied. Nonetheless, the MMP1

response we observed in our studies is clearly relevant, as the up-regulated MMP1 was necessary for shear-induced sprouting. Al-though it is possible that MMP1 expression is also sufficient todrive sprouting, it is more likely that the development of thecomplex cytoskeletal structures that give rise to directed invasionis independently triggered by shear stress, and sprouting occursthrough the coordination of both processes, matrix degradationand cellular invasion. What remains unclear is how the cyto-skeletal processes are regulated and what the molecular basisis for the sharp threshold-like response to shear stress.The localized concentration of sprouts in different flow pat-

terns reveals the potential significance of this shear-induced re-sponse. Our computational model revealed that transmural flowinto underlying matrix is highest along the shortest paths (corre-sponding to least resistance) between a source channel (artery) anda sink channel (lymphatic or venous drain), providing a mechanismfor how vessels could sprout along low paths of flow resistance tobridge previously unconnected vessels. For luminal flow, our resultssuggest that endothelial cells sprout once a threshold of wall shearis surpassed. Thus, if the metabolic needs of a tissue increase, thenblood flow is redirected to the area, resulting in enhanced shearstress from both luminal and transmural flows, which in turn trig-gers endothelial cells to sprout, create collaterals, and reduce theshear stress. Our results also predict that if flow to a certain regionis reduced, then the sprouting vessels would prune and retract, alsoconsistent with what occurs in vivo (36). Although this mechanismprovides the basic structure for regulating vessel densities in tissue,the presence of other cell types, matrices, and soluble factors indifferent settings likely can modulate these effects, as exempli-fied by the effects of low serum on the shear stress thresholdreported here. Taken together, these results demonstrate thedynamic and complex interplay between mechanical, geometric,and biochemical factors that endothelial cells use to regulate theremodeling of vascular architecture.

Materials and MethodsThe device used in this study consists of a bilayer polydimethylsiloxane (PDMS)mold adhered to a glass coverslip and perfused with a linear syringe pump.Bovine collagen is polymerized within the device, and 400-μm needles areused to create voids in the hydrogel where necessary. Endothelial cells areseeded on the surface of the gel or within the voids, allowed to attach for30 min or 36 h, and then perfused with flow rates determined from thefinite-element simulation used to model the system. Detailed explanationsof the materials and methods can be found in SI Materials and Methods.

ACKNOWLEDGMENTS. We thank Lin Gao, Dewight Williams, and RayMeade for technical assistance and Christopher Yu, Evangelia Bellas, MikeYang, Sarah Stapleton, Jeroen Eyckmans, and Anant Chopra for helpfuldiscussions. This work was supported in part by National Institutes of HealthGrants EB00262, EB08396, and EB017103 and the Penn Center for Engineer-ing Cells and Regeneration. P.A.G. acknowledges postdoctoral fellowshipsupport from Grant HL007954.

1. Tuder RM, Flook BE, Voelkel NF (1995) Increased gene expression for VEGF and theVEGF receptors KDR/Flk and Flt in lungs exposed to acute or to chronic hypoxia.Modulation of gene expression by nitric oxide. J Clin Invest 95(4):1798–1807.

2. Potente M, Gerhardt H, Carmeliet P (2011) Basic and therapeutic aspects of angio-genesis. Cell 146(6):873–887.

3. Kuo L, Davis MJ, Chilian WM (1990) Endothelium-dependent, flow-induced dilation ofisolated coronary arterioles. Am J Physiol 259(4 Pt 2):H1063–H1070.

4. Dawson JM, Hudlická O (1989) The effects of long term administration of prazosin onthe microcirculation in skeletal muscles. Cardiovasc Res 23(11):913–920.

5. Egginton S, Zhou AL, Brown MD, Hudlická O (2001) Unorthodox angiogenesis inskeletal muscle. Cardiovasc Res 49(3):634–646.

6. Colgan OC, et al. (2007) Regulation of bovine brain microvascular endothelial tightjunction assembly and barrier function by laminar shear stress. Am J Physiol Heart CircPhysiol 292(6):H3190–H3197.

7. Tarbell JM (2010) Shear stress and the endothelial transport barrier. Cardiovasc Res87(2):320–330.

8. Price GM, et al. (2010) Effect of mechanical factors on the function of engineeredhuman blood microvessels in microfluidic collagen gels. Biomaterials 31(24):6182–6189.

9. Song JW, Munn LL (2011) Fluid forces control endothelial sprouting. Proc Natl AcadSci USA 108(37):15342–15347.

10. 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(1):175–185.

11. Semino CE, Kamm RD, Lauffenburger DA (2006) Autocrine EGF receptor activationmediates endothelial cell migration and vascular morphogenesis induced by VEGFunder interstitial flow. Exp Cell Res 312(3):289–298.

12. Tarbell JM, Weinbaum S, Kamm RD (2005) Cellular fluid mechanics and mechano-transduction. Ann Biomed Eng 33(12):1719–1723.

13. Tarbell JM, Demaio L, Zaw MM (1999) Effect of pressure on hydraulic conductivity ofendothelial monolayers: Role of endothelial cleft shear stress. J Appl Physiol (1985)87(1):261–268.

14. Parker JC, Stevens T, Randall J, Weber DS, King JA (2006) Hydraulic conductanceof pulmonary microvascular and macrovascular endothelial cell monolayers. Am JPhysiol Lung Cell Mol Physiol 291(1):L30–L37.

15. Galie PA, Spilker RL, Stegemann JP (2011) A linear, biphasic model incorporatinga brinkman term to describe the mechanics of cell-seeded collagen hydrogels. AnnBiomed Eng 39(11):2767–2779.

7972 | www.pnas.org/cgi/doi/10.1073/pnas.1310842111 Galie et al.

16. Carmeliet P, et al. (1999) Targeted deficiency or cytosolic truncation of the VE-cadherin gene in mice impairs VEGF-mediated endothelial survival and angiogenesis.Cell 98(2):147–157.

17. Tzima E, et al. (2005) A mechanosensory complex that mediates the endothelial cellresponse to fluid shear stress. Nature 437(7057):426–431.

18. Jinga VV, Bogdan I, Fruchter J (1986) Experimental model for the quantitative estimation oftransendothelial transport in vitro; a two-compartment system. Physiologie 23(4):227–236.

19. Chrobak KM, Potter DR, Tien J (2006) Formation of perfused, functional microvasculartubes in vitro. Microvasc Res 71(3):185–196.

20. Partridge CR, Hawker JR, Jr., Forough R (2000) Overexpression of a secretory form of FGF-1promotes MMP-1-mediated endothelial cell migration. J Cell Biochem 78(3):487–499.

21. Nguyen DH, et al. (2013) Biomimetic model to reconstitute angiogenic sproutingmorphogenesis in vitro. Proc Natl Acad Sci USA 110(17):6712–6717.

22. 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 Physiol295(5):H2087–H2097.

23. Kang H, Kwak HI, Kaunas R, Bayless KJ (2011) Fluid shear stress and sphingosine1-phosphate activate calpain to promote membrane type 1 matrix metalloproteinase(MT1-MMP) membrane translocation and endothelial invasion into three-dimensionalcollagen matrices. J Biol Chem 286(49):42017–42026.

24. Chouinard-Pelletier G, Jahnsen ED, Jones EA (2013) Increased shear stress inhibitsangiogenesis in veins and not arteries during vascular development. Angiogenesis16(1):71–83.

25. Yamamoto K, et al. (2005) Fluid shear stress induces differentiation of Flk-1-positiveembryonic stem cells into vascular endothelial cells in vitro. Am J Physiol Heart CircPhysiol 288(4):H1915–H1924.

26. Wang DM, Tarbell JM (1995) Modeling interstitial flow in an artery wall allowsestimation of wall shear stress on smooth muscle cells. J Biomech Eng 117(3):358–363.

27. Galie PA, Russell MW, Westfall MV, Stegemann JP (2012) Interstitial fluid flow andcyclic strain differentially regulate cardiac fibroblast activation via AT1R and TGF-β1.Exp Cell Res 318(1):75–84.

28. Ng CP, Hinz B, Swartz MA (2005) Interstitial fluid flow induces myofibroblastdifferentiation and collagen alignment in vitro. J Cell Sci 118(Pt 20):4731–4739.

29. Polacheck WJ, Charest JL, Kamm RD (2011) Interstitial flow influences direction oftumor cell migration through competing mechanisms. Proc Natl Acad Sci USA 108(27):11115–11120.

30. Shi ZD, Ji XY, Qazi H, Tarbell JM (2009) Interstitial flow promotes vascular fibroblast,myofibroblast, and smooth muscle cell motility in 3-D collagen I via upregulation ofMMP-1. Am J Physiol Heart Circ Physiol 297(4):H1225–H1234.

31. Shi ZD, Ji XY, Berardi DE, Qazi H, Tarbell JM (2010) Interstitial flow induces MMP-1expression and vascular SMC migration in collagen I gels via an ERK1/2-dependentand c-Jun-mediated mechanism. Am J Physiol Heart Circ Physiol 298(1):H127–H135.

32. Milkiewicz M, Kelland C, Colgan S, Haas TL (2006) Nitric oxide and p38 MAP kinasemediate shear stress-dependent inhibition of MMP-2 production in microvascularendothelial cells. J Cell Physiol 208(1):229–237.

33. Rivilis I, et al. (2002) Differential involvement of MMP-2 and VEGF during musclestretch- versus shear stress-induced angiogenesis. Am J Physiol Heart Circ Physiol283(4):H1430–H1438.

34. Sho E, et al. (2002) Arterial enlargement in response to high flow requires early ex-pression of matrix metalloproteinases to degrade extracellular matrix. Exp Mol Pathol73(2):142–153.

35. Yun S, et al. (2002) Transcription factor Sp1 phosphorylation induced by shear stressinhibits membrane type 1-matrix metalloproteinase expression in endothelium. J BiolChem 277(38):34808–34814.

36. Isogai S, Lawson ND, Torrealday S, Horiguchi M, Weinstein BM (2003) Angiogenicnetwork formation in the developing vertebrate trunk. Development 130:5281–5290.

Galie et al. PNAS | June 3, 2014 | vol. 111 | no. 22 | 7973

ENGINEE

RING

APP

LIED

BIOLO

GICAL

SCIENCE

S