Dysfunctional mechanotransduction regulates the progression of PIK3CA-driven vascular malformations

Vascular malformations (VMs) are a class of rare genetic disorders associated with localized developmental abnormalities of venous, arterial, capillary, or lymphatic vessels.^1–4 Histologically, VMs are characterized by complexes of structural lesions with enlarged, irregular lumens that are lined with vascular endothelial cells (ECs) and surrounded by disorganized extracellular matrix.^3,5 These lesions are often congenital, progress in severity over time, are associated with vascular obstruction and impaired drainage, and can be life threatening. Treatment for VMs, especially for complex malformations that occur in multiple tissues, is difficult and rarely curative, and the standard of care relies on surgery, sclerotherapy, and a limited repertoire of drugs.

VMs are caused by somatic mutations in genes that are involved in vascular development, and these same genes are commonly implicated in malignant tumor angiogenesis.⁴ VMs are classified based on blood flow dynamics and are categorized as either slow-flow or fast-flow malformations.³ The majority of slow-flow malformations, which do not involve an arterial component, are caused by mutations that result in upregulated phosphoinositide 3-kinase (PI3K) and mammalian target of rapamycin (mTOR) signaling.⁴ In particular, somatic activating mutations of PIK3CA have been identified as the cause of the majority (∼80%) of cystic lymphatic malformations, and other pathologies, including generalized lymphatic anomalies, capillary malformations, and venous malformations.^2,6–9 Several studies have demonstrated that overstimulation of PI3K signaling, independent of genetic mutations in PIK3CA, is associated with the pathological development of slow-flow VMs, including cerebral cavernous malformation^10–13 and fast-flow arteriovenous malformations in mouse models of hereditary hemorrhagic telangiectasia.^14–16 Given the implication of PI3K overactivation in both slow- and fast-flow VMs, understanding how hemodynamic forces contribute to the pathogenesis of PIK3CA-driven VMs could shed light on novel therapeutic approaches for various vascular anomalies.

PIK3CA encodes the p110α catalytic subunit (PI3Kα) of phosphatidylinositol-3-kinases (PI3K), a lipid kinase that is ubiquitously expressed and essential for regulating effector signaling pathways involved in cell proliferation, metabolism, migration, and survival.¹⁷ At the cellular level, PI3K is activated by receptor tyrosine kinases, focal adhesion kinase, and G-protein coupled receptor signaling.¹⁷ Upon activation, PI3K catalyzes the production of phosphatidylinositol 3,4,5-bisphosphate [PtdIns(3,4,5)P₃ or PIP3] at the plasma membrane, leading to membrane translocation and downstream activation of AKT and mTOR signaling cascades that promote cell growth and survival.¹⁷ PI3K signaling is also involved in the remodeling of the actin cytoskeleton. In particular, activation of PI3Kα has been demonstrated to modulate actomyosin contractility and cell motility in endothelial cells by regulating the activity of myosin light chain phosphatase,¹⁸ Rac1-,¹⁹ and RhoA-GTPases.²⁰ In addition to being regulated by growth factors signaling, PI3K functions downstream of the fluid shear stress sensitive mechanosensory complex that consists of platelet endothelial cell adhesion molecule 1 (PECAM-1), vascular endothelial (VE-) cadherin, and vascular endothelial growth factor receptor 2 (VEGFR2).^21–23 While shear-stress induced activation of PI3K signaling is associated with endothelial cell elongation, alignment, and vasodilatory response, how overactivation of PI3K signaling impacts endothelial mechanotransduction remains unknown.

Multiple studies using mouse models have revealed that PI3Kα plays a crucial role in regulating venous and lymphatic vessel growth and development.^{6–9,24–26} Both the loss of function and inhibition of PI3Kα result in embryonic lethality due to deficient angiogenesis and venous fate specification.^24,27 Conversely, expression of gain-of-function PIK3CA^H1047R variant of PI3Kα in the endothelium leads to over-proliferation of venous and lymphatic endothelial cells (LECs).^25,26 Interestingly, expression of the same oncogenic PIK3CA^H1047R variant resulted in distinct vascular overgrowth and malformation defects in blood and lymphatic vessels.²⁵ Specifically, PIK3CA-driven vascular malformations manifest as locally restricted vessel overgrowth and dilation in veins and capillaries but as dense hypersprouting vascular networks in the lymphatics.²⁵ However, the mechanisms by which one PIK3CA activating mutation produces distinct pathological phenotypes in different vascular beds remain to be investigated. The lack of arterial involvement in slow-flow VMs, together with common biophysical signatures of tissues susceptible to lesion formation, including compliant extracellular matrix, low rates of perfusion, and drainage flow, suggest significant contributions of vascular hemodynamics and the biophysical microenvironment to PIK3CA-driven lesion progression.

Previously, we established a microphysiological model of VMs and identified that PIK3CA activating mutations result in elevated Rac1 activity in endothelial cells, which drives lesion formation through dysregulated cytoskeletal dynamics.¹⁹ In this study, we address the role of fluid shear stress and the biophysical microenvironment in the development of PIK3CA-driven slow-flow malformations, and we identify hemodynamic shear stress and defective cellular mechanotransduction as key regulators of malformation progression. We demonstrate that endothelial cells expressing PIK3CA^E542K activating mutation fail to elongate, align, and establish diffusive barrier function in response to laminar shear stress. We observe increased cellular and nuclear compliance, and we further identified deficient recruitment of vinculin, a mechanosensitive adaptor protein, to intercellular adherens junctions in human umbilical vein endothelial cells (HUVECs) expressing a PIK3CA activating mutation commonly associated with VMs. Additionally, using tissue-engineered microvessels, we demonstrate that mutant HUVECs form dilated microvessels that are insensitive to flow interruption and are hypersensitive to basal-to-apical transmural drainage, resulting in sprouting and invasion responses to transmural flow. These results demonstrate that altered endothelial cellular and nuclear mechanical properties, as well as dysfunctional mechanotransduction of hemodynamic stresses, contribute to pathological vascular overgrowth and hypersprouting in PIK3CA-driven VMs.

The development of PIK3CA-driven vascular lesions is restricted to slow-flow regions of the vasculature.²⁸ To determine whether VM causal mutations in PIK3CA impact mechanotransduction of hemodynamic shear stress, we generated human umbilical vein endothelial cells (HUVECs) expressing wild-type PIK3CA (PIK3CA^WT) or PIK3CA harboring a PI3K-activating mutation (PIK3CA^E542K), as described previously,¹⁹ and applied flow via orbital shaking (Fig. 1). This approach allowed us to apply a range of shear stresses, both laminar and oscillatory, to cultured PIK3CA^WT and PIK3CA^E542K monolayers. At a wall shear stress magnitude similar to that experienced by venous ECs (∼11 dyn/cm²)^29,30 near the edge of the well, PIK3CA^WT HUVECs became highly elongated and aligned with the local flow direction [Figs. 1(a)–1(c)]. In response to lower magnitude wall shear stress and oscillatory flow toward the center of the well, PIK3CA^WT HUVECs adopted a more isometric morphology [Figs. 1(a) and 1(c)]. In contrast, PIK3CA^E542K HUVECs remained isometric at all magnitudes of wall shear stress [Figs. 1(a)–1(c)]. We further observed increased cell density in PIK3CA^E542K HUVECs across all applied shear stress magnitudes, indicating impaired flow-mediated quiescence with excess PI3K activation [Fig. 1(d)]. Consistent with orbital shaking results, PIK3CA^E542K HUVECs remained misaligned and overcrowded in response to 10 dyn/cm² laminar shear stress applied in a Hele–Shaw flow cell [Fig. 1(f)].

To determine if the defective mechanotransduction we observed in blood endothelial cells could play a role in lymphatic malformation progression, we generated lymphatic endothelial cells (LECs) expressing either PIK3CA^WT or PIK3CA^E542K (Fig. S1). Similar to PIK3CA^E542K HUVECs, we observed that PIK3CA^E542K LECs were larger in cell size and more isometric compared to PIK3CA^WT LECs in response to orbital shaking (Fig. S1), suggesting that excess PI3K activation and effector signaling perturb mechanotransduction of hemodynamic shear stress in both the blood and the lymphatic endothelium. To determine the functional consequences of cellular misalignment to flow, we measured the permeability of PIK3CA^WT and PIK3CA^E542K HUVEC monolayers in response to flow. We cultured PIK3CA^WT and PIK3CA^E542K HUVEC under orbital shaking conditions on biotinylated-fibronectin substrates, and permeability was quantified by fluorescence intensity of Cy3-conjugated streptavidin that bound to the biotinylated surface [Figs. 1(a) and 1(e)]. In agreement with the previous reports,³¹ we observed improved barrier function in PIK3CA^WT HUVECs at regions corresponding to high laminar shear stress and a gradual increase in permeability with decreasing wall shear stress magnitude [Figs. 1(a) and 1(e)]. Interestingly, PIK3CA^E542K HUVECs demonstrated a dichotomous relationship between shear stress magnitude and barrier function, with increased permeability as compared to wild-type cells at high magnitudes of shear stress and reduced permeability at lower shear stress magnitudes [Figs. 1(a) and 1(e)]. Collectively, these findings demonstrated that constitutive PI3K activation hinders flow-mediated endothelial cell alignment and barrier function.

To establish the temporal relationship between elevated PI3K signaling, defective endothelial cell remodeling, and loss of vascular barrier function, we generated HUVECs expressing either tet-ON-PIK3CA^WT or tet-ON-PIK3CA^E542K in which the expression of WT or mutant variant of PIK3CA can be induced by the addition of doxycycline [Fig. S2(a)]. Treatment with doxycycline to induce PIK3CA^WT or PIK3CA^E542K expression induced similar levels of p110α expression as measured by Western blot, indicating a comparable level of lentivirus integration and protein overexpression in both cell lines [Fig. S2(a)]. Upon addition of doxycycline, we observed increased levels of phospo-AKT1(Ser473) in tet-ON-PIK3CA^E542K but not in tet-ON-PIK3CA^WT endothelial cells, demonstrating activation of PI3K effector signaling in response to induction [Fig. S2(a)]. To verify the functional relevance of this construct, we generated 3D human microvascular networks with tet-ON-PIK3CA^E542K HUVECs [Figs. S2(b) and S2(c)]. We observed that the addition of doxycycline during seeding of tet-ON-PIK3CA^E542K HUVECs accelerated lumenogenesis and resulted in the formation of microvascular networks with enlarged and irregular lumens [Fig. S2(b)], in agreement with our previous study demonstrating that constitutive activation of PIK3CA^E542K resulted in pathologic network formation.¹⁹ 

To establish the temporal relationships between dysregulated PI3K activation and mechanotransduction of hemodynamic shear stress, we induced expression of PIK3CA^E542K either prior to or after the application of flow [Fig. 2(a)]. When PIK3CA^E542K expression was induced before the application of flow, ECs remained unresponsive to shear stress, as measured by cell orientation and elongation [Figs. 2(a)–2(c)]. However, when PIK3CA^E542K expression was triggered 1 day after flow, flow-conditioned HUVECs remained elongated and aligned parallel to the flow direction [Figs. 2(a)–2(c)]. Interestingly, although flow preconditioning rescued cell morphology and alignment, PI3K activation significantly reduced vascular barrier function compared with uninduced (no doxycycline) control [Fig. 2(d)]. Collectively, these results indicate a mechanistic separation between flow-mediated cell alignment and barrier function, whereby excessive PI3K signaling may have a more direct and immediate impact on EC adherens junction stability and resultant permeability.

We previously reported the presence of discontinuous VE-cadherin-containing adheren junctions, decreased cortical actin localization, and increased stress fibers formation in HUVECs expressing PIK3CA activating mutations.¹⁹ We observed similar decreased in the distribution of cortical actin, lack of stress fibers alignment, and reduced VE-cadherin signal in cell–cell contacts of flow-conditioned PIK3CA^E542K HUVECs [Figs. 3(a)–3(d)]. Additionally, we observed increased microtubule density and similar loss of microtubule and vimentin alignment in PIK3CA^E542K HUVECs (Figs. S3 and S4). These findings suggest that the mechanical coupling between the cytoskeleton and adhesion complexes and stability of VE-cadherin adhesion complexes may be hampered with constitutive PI3K activation. To quantify the stability of VE-cadherin intercellular adherens junctions, a VE-cadherin pulse-chase experiment was performed [Figs. 4(a)–4(c)]. Surface-pool of VE-cadherin were pulse-labeled with nonfunctional blocking antibody and turnover of labeled VE-cadherin complexes were quantified after a 2-h chase period. We observed a reduced level of labeled VE-cadherin at cell-cell contacts and a significant increase in the number of internalized labeled VE-cadherin positive vesicles within the cytoplasm of PIK3CA^E542K HUVECs [Figs. 4(a)–4(c)], indicating that adheren junctions are indeed less stable in mutant endothelium. Given that endothelial mechanotransduction of shear stress is partially mediated by cortical Src kinase activation driven by the formation of junctional VE-cadherin/PECAM-1/VEGFR2 mechanosensory complex,²³ we investigated the dynamics of Src kinase activation in PIK3CA mutant endothelium [Figs. 4(d) and 4(e)]. Interestingly, we found that the temporal dynamics of Src activation measured by the level of phospho-Src (Tyr416) was similar both in magnitude and temporal activation among untransduced HUVEC, PIK3CA^WT, and PIK3CA^E542K HUVECs [Figs. 4(d) and 4(e)], suggesting that while assembly and turnover of adheren junctions are disrupted in mutant endothelial cells, some elements of downstream effector signaling remain functional.

VE-cadherin containing junctions are mechanically connected to the actin cytoskeleton through the interaction of VE-cadherin with β- and α-catenin.^32,33 Under mechanical load, tensile stress unfolds α-catenin, allowing for vinculin binding to α-catenin and subsequent force-dependent strengthening of cell–cell junctions.^34–36 To address whether the mechanical coupling between actin cytoskeleton and adhesion complexes could underlie the increased permeability in mutant endothelium, we investigated the spatial localization of vinculin in PIK3CA^WT and PIK3CA^E542K HUVECs [Figs. 5(a)–5(c)]. In PIK3CA^WT HUVECs, vinculin localized at the cell cortex and co-localized with VE-cadherin containing cell-cell adhesions and along actin stress fibers [Figs. 5(a) and 5(b)]. In contrast, PIK3CA^E542K HUVECs demonstrated reduced recruitment of vinculin to intercellular adheren junctions as compared to PIK3CA^WT HUVECs [Figs. 5(a) and 5(b)]. We further observed that while vinculin primarily localizes in a straight and linear pattern in focal adhesion plaques of PIK3CA^WT HUVECs, vinculin appears as larger, dot-like structures in flow-conditioned PIK3CA^E542K HUVECs [Fig. 5(c)]. Despite significant changes in vinculin localization, western blot analysis showed no measurable differences in the amounts of VE-cadherin, β-catenin, and vinculin between control and mutant HUVECs [Figs. 5(d) and 5(e)]. Interestingly, in probing for changes for overall actomyosin contractility, we observed reduced level of phosphorylated myosin light chain II (pMLC2) in mutant ECs compared with control ECs [Figs. 5(d) and 5(e)]. These data demonstrate defective transmission of mechanical forces and stabilization of cell–cell and cell–matrix adhesion complexes in PIK3CA^E542K HUVECs.

To relate our findings in 2D cell culture to the pathophysiological phenotype of vascular malformations, we generated and characterized the effect of flow on PIK3CA^WT or PIK3CA^E542K tissue engineered human microvessels (Fig. 6). Engineered microvessels were molded using an acupuncture needle to generate a hollow 160-μm cylindrical channel lined with HUVECs and fully embedded within 2.5 mg/ml type I collagen hydrogel.^31,37 Cyclic flow of approximately 1–3 dyn/cm² was imparted by culturing microvessels on a laboratory rocker.³⁷ After 2 days of culture under a slow flow environment, PIK3CA^WT HUVECs formed uniform microvessels with a limited number of sprouts and no apparent HUVEC invasion into the subluminal matrix [Figs. 6(a) and 6(b)]. In contrast, PIK3CA^E542K microvessels were significantly dilated and developed numerous sprouts along the vessel length [Figs. 6(a) and 6(b)]. The expanded mutant vessel diameter coincides with an increased number of PIK3CA^E542K HUVECs, implicating defective regulation of flow-mediated quiescence with excessive PI3K signaling. Treatment with rapamycin, an mTOR inhibitor, or alpelisib, a PIK3CA-specific kinase inhibitor, reduced vessel dilation [Figs. 6(a) and 6(b)]. Importantly, when compared to rapamycin, we observed a more effective reduction in number of sprouts with alpelisib treatment [Figs. 6(a) and 6(b)]. This observation is consistent with our previous finding and the notion that in addition to contributing to uncontrolled growth through downstream mTORC1/2 and growth signaling activation, PI3K hyperactivation alters cytoskeletal dynamics and mechanical properties of endothelial cells.¹⁹ 

Given the defective recruitment of vinculin to adherens junctions, increased vascular permeability, loss of cortical actin organization, and prominent stress fiber formation in PIK3CA^E542K HUVECs, we reasoned that the dilated microvessel phenotype may reflect reduced cell-cell adhesion strength and altered cell–matrix traction forces in mutant microvessels. To assess changes in cell-generated forces, we visualized the collagen matrix surrounding the microvessels through reflectance imaging and compared the displacement of matrix before and after lysing ECs with detergent [Figs. 6(c) and 6(d)]. Notably, we observed significantly lower matrix displacement in mutant microvessels when compared to control [Figs. 6(b)–6(d)], indicating reduced traction forces in PIK3CA^E542K microvessels.

To determine whether the slow-flow environment of native lesions potentiates VM progression, we next investigated the effect of transitioning from flow to static culture of PIK3CA^WT and PIK3CA^E542K microvessels by preconditioning microvessels with flow for 24 h then transitioning to static conditions for the subsequent 24 h. After transition to static culture, PIK3CA^WT HUVECs delaminated and detached into the apical lumen, resulting in gaps in the endothelium [Fig. 7(a)]. In contrast, PIK3CA^E542K HUVECs were insensitive to flow interruption, with no observable delamination and the formation of tortuous, sprouted, and dilated microvessels that are akin to mutant vessels cultured under flow conditions [Figs. 7(b) and 7(c)]. In all culture conditions, PIK3CA^E542K microvessels contained approximately twice the number of cells and were approximately twofold wider than PIK3CA^WT microvessels [Fig. 7(c)]. While PIK3CA^E542K microvessels showed minimal effects from flow interruption, flow interruption in PIK3CA^WT microvessels resulted in gaps covering approximately 6.5% of the vessel area [Fig. 7(c)]. Similar lack of cell delamination and cellular invasion were observed in PIK3CA^E542K microvessels cultured under fully static condition [Figs. S5(a)–S5(c)]. Collectively, these findings indicate that constitutive PI3K activation renders ECs insensitive to vascular regression upon flow interruption and may lead to improved survival and favorable expansion of mutant ECs in the no flow and slow flow characteristics of PIK3CA-driven VMs.

In addition to luminal shear stress, venous and lymphatic ECs are subjected to draining transmural flow. Hydrostatic and oncotic pressure differences across the vessel wall causes fluid to extravasate from blood capillaries, resulting in the accumulation of interstitial fluid in the surrounding matrix which is subsequently reabsorbed and collected by venous and lymphatic vessels.³⁸ Therefore, in addition to experiencing low level of wall shear stress, venous and lymphatic ECs are also subjected to fluid stresses arising from transmural drainage. To investigate the effect of transmural flow on control HUVECs, PIK3CA^WT, and PIK3CA^E542K microvessels, we cultured control or mutant HUVECs microfluidic devices that contain two parallel microvessels in a shared collagen I hydrogel.³⁹ This two-channel configuration allows us to establish a pressure gradient between the two microchannels to drive transmural flow into (draining vessel, basal-to-apical flow) or out of (source vessel, apical-to-basal flow) a microvessel [Figs. 8(a) and S6(a)]. We observed increased HUVEC invasion, reminiscent of angiogenic sprouting,⁴⁰ into the subluminal collagen I hydrogel upon application of transmural flow in the basal-to-apical direction in PIK3CA^E542K microvessels as compared to control HUVECs or PIK3CA^WT microvessels [Figs. 8(a)–8(c)]. Consistent with previous reports,^41–43 this increase in cell invasion primarily occurred with the application of basal-to-apical but not apical-to-basal transmural flow [Figs. 8(a) and 8(b)].

Slow-flow VMs arise from sporadic somatic mutations, resulting in vascular lesions consisting of endothelial cells with and without the driving mutations. The mosaic nature of vascular lesions implies a role of non-cell autonomous signaling where mutated endothelial cells influence the behavior of neighboring non-mutated cells to amplify lesion growth and progression. To determine how PIK3CA^E542K HUVECs influence non-mutated HUVECs, we generated mosaic microvessels containing HUVECs expressing histone H2B-mCherry and inducible tet-ON-PIK3CA^E542K. We then applied basal-to-apical transmural flow and investigated whether doxycycline induced expression of PIK3CA^E542K influenced sprouting and invasion behavior of control H2B-mCherry HUVECs [Fig. 9(a)]. We found that doxycycline induction increased both the number of H2B-mCherry and PIK3CA^E542K HUVECs located within sprouts [Fig. 9(b)], indicating a non-cell-autonomous effect of PI3K activation on sprouting angiogenesis. We further quantified the tip cell distribution in mosaic sprouts and observed that tet-ON-PIK3CA^E542K and H2B-mCherry HUVECs contributed to 40% and 60% of tip cells, respectively, in untreated mosaic microvessels [Figs. 9(c) and 9(d)]. Doxycycline treatment drastically changed the proportion of control and mutant tip cells, where PIK3CA^E542K HUVECs occupied 87% of the tips and H2B-mCherry HUVECs contributed mostly as stalk cells in mosaic sprouts of doxycycline treated microvessels [Figs. 9(c) and 9(d)]. Together, these results suggest that PIK3CA^E542K endothelial cells are more migratory and have a propensity to lead the sprouting and invasion of non-mutated endothelial cells.

We next determined whether the non-autonomous effect on sprouting is a result of matrix remodeling by PIK3CA^E542K HUVECs. Proteolytic degradation of the subluminal collagen matrix is known to be required for angiogenic sprout initiation and extension.⁴⁴ Previously, we described an increase in protease secretion and ECM degradation by PIK3CA-mutant HUVECs,¹⁹ and we, thus, hypothesized that ECM degradation and remodeling by PIK3CA^E542K HUVECs may prime PIK3CA^WT HUVECs for angiogenic sprouting. To test this hypothesis, we formed PIK3CA^WT and PIK3CA^E542K microvessels and lysed cells after 24 h, then reseeded the resulting channels with PIK3CA^WT HUVECs [Fig. 9(e)]. We then applied basal-to-apical transmural flow and quantified the number of sprouts formed in primed ECM [Fig. 9(f)]. We found that PIK3CA^WT HUVECs cultured in microvessels primed by PIK3CA^E542K HUVECs failed to reproduce the hypersprouting phenotype [Fig. 9(f)]. This finding suggests that changes in matrix architecture and mechanics caused by PIK3CA^E542K HUVECs are not sufficient to account for the etiopathology of VMs.

Increased cell and nuclear compliance has been shown to promote 3D migration in various cell types,⁴⁵ and we, thus, hypothesized that the hypersprouting observed in PIK3CA^E542K microvessels was driven by changes in cell and/or nuclear mechanical properties of PIK3CA^E542K HUVECs. To distinguish changes in cell intrinsic mechanics from tissue mechanics, we quantified cellular and nuclear deformation on a single cell basis using a microfluidic micropipette aspiration assay^46,47 (Figs. 10 and S7). The microfluidic device was designed to have two inlets allowing for the perfusion and loading of cells through the main channel. The main channel contains an array of 18 pockets with narrow constriction micropipette channels and are connected to an outlet port which is open to the atmosphere (Fig. S7). A pressure of 7.0 kPa was applied to inlet 1 and 1.4 kPa to inlet 2, and both outlets are drained through constricted channels to atmospheric pressure [Fig. S7(b)]. Cells are initially trapped in larger pockets before deforming into 5 × 5 μm² constricted channels under a fixed pressure gradient, and deformation was imaged via optical microscope for 2 min [Figs. 10(b) and S7]. Interestingly, while untransduced HUVECs maintained cellular integrity for the duration of the experiment, PIK3CA^WT and PIK3CA^E542K endothelial cells were prone to fracture as the cells passed through the narrow constriction, with PIK3CA^E542K cells having the lowest probability of survival throughout the experiment [Fig. 10(c)]. The cytoplasm and nuclear protrusion length were measured as a function of time, and while the cytoplasm of PIK3CA^WT and PIK3CA^E542K HUVECs deformed similarly in response to the applied pressure gradient, both cell types demonstrated greater cytoplasmic deformation and cytoplasm fracture than untransduced HUVECs [Figs. 10(c) and 10(d)]. The nuclei of the three cell types deformed at different rates, with untransduced HUVEC nuclei experiencing the least deformation over time and PIK3CA^E542K expressing nuclei experiencing the greatest deformation [Fig. 10(e)]. Nuclear and cytoplasm viscoelastic properties were computed from the protrusion curves, as described previously.⁴⁷ Interestingly, PIK3CA^E542K and PIK3CA^WT HUVECs share similar cytoplasmic viscoelastic properties and are more compliant and less viscous than the cytoplasm of untransduced HUVECs [Fig. 10(f)]. The nuclei of PIK3CA^E542K HUVECs showed reduced elasticity as compared to the nuclei of PIK3CA^WT and untransduced HUVECs but exhibited the same viscous properties as PIK3CA^WT nuclei [Fig. 10(g)]. Thus, PIK3CA^E542K HUVECs were distinguished only by their reduced nuclear elastic modulus compared to PIK3CA^WT HUVECs. Together, these results demonstrated increased cellular and nuclear compliance in PIK3CA-mutant ECs which may underlie dilation and overgrowth of venous and lymphatic vessels.

PIK3CA-driven VMs stereotypically present in soft tissues with low rates of perfusion.^4,48 Here, using a series of engineering and microfluidic tools, we investigated the role of the biophysical microenvironment in the development of slow-flow vascular malformations to test the hypothesis that low hemodynamic shear stress potentiates lesion formation. Alignment and elongation of the cytoskeleton in the direction of flow is a hallmark of endothelial mechanotransduction,⁴⁹ and laminar shear stress has been shown to regulate vascular barrier function locally and acutely.^31,50 Furthermore, a lack of alignment and barrier function in response to hemodynamic shear stress is indicative of endothelial dysfunction.⁵¹ Here, using both an orbital flow system to impart a range of shear stress magnitudes and a Hele–Shaw flow cell for unidirectional, laminar flow, we observed that HUVECs and LECs expressing a constitutively active PIK3CA-activating mutation fail to elongate and align in response to laminar shear stress (Figs. 1 and 2). The lack of cell alignment was also reflected in a more isotropic distribution of microtubules and intermediate filaments in mutant cells (Figs. S3 and S4). Furthermore, the permeability of PIK3CA-mutant endothelial cells remained elevated even in the presence of physiologic shear stress (Figs. 1 and 2). Together, these results suggest that dysfunctional mechanotransduction and resultant insensitivity to flow could be a key mechanism that underlies the dysregulated vascular networks characteristic of VMs (Fig. S8).

We observed that PIK3CA^E542K microvessels generate lower net traction forces than PIK3CA^WT microvessels (Fig. 6). Previous work has demonstrated that in multicellular systems, traction forces measured at cell–ECM interfaces are impacted by changes in cell–cell and cell–matrix adhesion⁵² and that maturation of adherens junction complexes requires a balance of MLC-mediated cell-generated forces and Rac1 activity.⁵³ Previously, we identified that PIK3CA activating mutations result in excess Rac1 activity.¹⁹ Here, we observed that expression of PIK3CA^E542K reduced MLC phosphorylation and recruitment of vinculin to adherens junction complexes, resulting in the discontinuity of junctional VE-cadherin and increased permeability (Figs. 3 and 4). Together, these results suggest that pathophysiologic PI3K activation results in adheren junction instability through mechanical imbalances at cell–cell junctions, consistent with the previous work demonstrating that PIK3CA signaling inhibits NUAK1-dependent phosphorylation of myosin phosphatase targeting-1 in ECs, leading to reduced overall actomyosin contractility.¹⁸ Collectively, these results suggest that adherens junction instability and lack of maturation in response to hemodynamic signals render PIK3CA mutant cells prone to angiogenesis and could underlie the overconnected topologies characteristic of VMs.

Clinically, VMs are described as soft and compressible, and histologically they appear as a single layer of endothelial cells in enlarged vein-like channels with disorganized pericellular matrix.⁴ To recapitulate the compliant ECM characteristic of tissues susceptible to VMs, we developed a 3D microfluidic model of vascular malformations and observed cellular signatures consistent with the clinical presentation of VMs, including irregular and dilated microvessels (Fig. 6). In the previous work, we performed a broad characterization of MMPs and fibrinolytic enzymes in conditioned media from wild type and mutant ECs, and we observed upregulation of MMP-1, which degrades collagen type I.¹⁹ These results are consistent with the increased diameter of mutant microvessels [Figs. 6(a) and 6(b)] and could contribute to the increased sprouting in response to flow, as perfusion has been shown previously to regulate angiogenesis in an MMP-1-dependent manner.⁵⁴ 

Previously, we have shown that reducing hemodynamic shear stress leads to rapid adherens junction dissolution,³¹ and experiments conducted in developmental animal models have demonstrated that reduction or interruption of flow induces vessel regression and network pruning,⁵⁵ which is thought to be essential for optimization of efficient delivery of oxygen and nutrients.^56,57 To recapitulate the hemodynamic signals associated with pruning, we exposed microvessels to flow for 24 h, then interrupted flow and cultured microvessels in static conditions for 24 h. Interestingly, PIK3CA mutant microvessels did not respond to flow interruption, and mutant ECs remain attached and continue to grow under static culture conditions, whereas wild-type ECs delaminate from the vessel wall (Figs. 7 and S5). These results suggest that cells expressing activating PIK3CA mutations are primed for survival in the unique hemodynamic environment of slow-flow lesions and that this survival results in a lack of pruning in response to reduced or interrupted perfusion to allow VM progression in the slow-flow microenvironment.

Computational modeling and a limited number of experiments have demonstrated that changes in vascular network topology modulate not only local wall shear stress but also luminal pressure.⁵⁸ To determine whether gradients in fluid pressure could play a role in lesion formation, we fabricated microfluidic devices to expose wild-type and mutant microvessels to transmural pressure gradients. We observed that transmural pressure gradients and resultant flow in the basal-to-apical direction results in increased sprouting and invasion of PIK3CA-mutant ECs as compared to control microvessels (Fig. 8). Previously, immature endothelial monolayers, characterized by discontinuous junctional VE-cadherin, have been shown to sprout in response to lower basal-to-apical pressure drops than mature monolayers with continuous pericellular VE-cadherin,⁵⁴ suggesting that the increased sprouting in PIK3CA-mutant ECs could be driven by immature adheren junctions. Furthermore, basal-to-apical transmural flow has been shown to synergize with VEGF gradients to increase angiogenic sprouting,⁴² and a recent study has demonstrated increased paracrine VEGF-C signaling in PIK3CA-driven mosaic LM mouse model.²⁶ Together, these results suggest that transmural pressure gradients and flow are a pro-angiogenic stimulus that drives lesion elaboration and growth.

Permeability measurements and images of pericellular VE-cadherin (Figs. 1 and 3) suggest that adherens junctions in mutant cells are immature and discontinuous. To relate the junctional phenotypes to the dysfunctional response to shear stress, we investigated whether downstream effector signaling modulated by key mechanosensory complexes that contain VE-cadherin^22,23,31 were differentially regulated in mutant cells. We first quantified the stability of VE-cadherin intercellular adherens junctions using a pulse-chase approach and observed a significantly reduced level of labeled VE-cadherin at cell–cell contacts (Fig. 4). We then tested the hypothesis that Src kinase, which is activated by the VE-cadherin/PECAM/VEGFR2 mechanosensory complex,²³ is differentially phosphorylated by mutant cells. Interestingly, we found that Src phosphorylation was similar in magnitude and temporal activation among untransduced HUVEC, PIK3CA^WT, and PIK3CA^E542K cells (Fig. 4 ). Furthermore, closer inspection of the pulse-chase staining of VE-cadherin reveals an increase in the intracellular puncta of AF647-VE-cadherin, suggesting increased VE-cadherin internalization and recycling. Previous work has established mature adheren junctions as sites of cytoskeletal organization,⁵⁹ that VE-cadherin is necessary for proper alignment of the cytoskeleton in response to shear stress,⁶⁰ and that pericellular VE-cadherin localization balances cell-generated forces associated with cortical actin elaboration.^53,61 Thus, our data suggest that hyperactive PIK3CA activity, driven by mutant expression, reduces adheren junction stability through increased VE-cadherin turnover, which results in defective coordination of cytoskeletal dynamics and resultant alignment in response to flow, but that sites of assembled VE-cadherin are capable of transducing flow responses as measured by Src phosphorylation. Future work will seek to corroborate these results and provide greater mechanistic insight through dynamic characterization of gene expression profiles and direct testing of mechanosensory complex assembly and stability through immunoprecipitation.

Previous studies have demonstrated that nuclear stiffness and deformability play a role in 3D cellular migration through dense collagen matrices and spatially constraining micropores.^45,62 Using a microfluidic micropipette aspiration assay, we found that PIK3CA^WT and PIK3CA^E542K ECs exhibit increased cellular compliance compared to untransduced HUVECs (Fig. 10) and that PIK3CA^E542K ECs have significantly more compliant nuclei than PIK3CA^WT ECs. The increased cellular and nuclear compliance in PIK3CA-mutant ECs may facilitate EC migration and sprout formation, potentially contributing to hypervascularization in VM lesions. Previous work has demonstrated a role of Akt in phosphorylating Lamin A/C, resulting in degradation in keratinocytes⁶³ and C2C12 myocytes.⁶⁴ Furthermore, Lmna^-/- MEFs demonstrate increased nuclear deformation in a microfluidic micropipette aspiration assay similar to the one employed here.⁴⁷ These results are consistent with our observations that PIK3CA^E542K cells, which we have previously shown have increased p-AKT (Ser⁴⁷³) as compared to GFP transduction control cells,¹⁹ are characterized by increased nuclear deformation (Fig. 10). While the specific mechanisms that relate PIK3CA activation to nuclear compliance remain unclear, increased invasion of colorectal cancer cells expressing PIK3CA activating through 8 μm pores and through Matrigel has been observed.⁶⁵ Conversely, knockdown of PIK3CA in glioblastoma cells reduces invasion into Matrigel,⁶⁶ collectively suggesting that PIK3CA-mediated changes in nuclear mechanics could underlie a wide variety of pathogenic processes.

In summary, our findings demonstrate a role of altered mechanotransduction and cell mechanics in PIK3CA-driven vascular malformations (Fig. 8). These findings could potentially be leveraged for genetic diagnosis and treatment of vascular anomalies. Molecular genetic testing for VMs remains challenging as the lesions are mosaic, with only a small population of cells expressing the driving mutations. Therefore, these reported changes in cellular mechanics may potentially be used as a biomarker for the sorting and enrichment of mutated ECs from patient biopsies. Furthermore, targeting the defective contractility and recalibrating the mechanotransduction in mutant endothelial cells could represent an avenue to reduce or regress lesion progression.

HUVECs (Lonza) and juvenile HDLECs (Promocell) were cultured in endothelial growth medium (EGM)-2 growth medium (Lonza) in a humidified incubator at 37 °C and 5% CO₂. HUVECs were used between passages 6 and 12. HDLECs were used between passage 3 and 7. Lentivirus generation and transduction of HUVECs with pHAGE-PIK3CAWT, pHAGE-PIK3CAE542K, and pLenti6-H2B-mCherry (Addgene, plasmid #89766, gift from Torsten Wittmann) were performed as previously described.¹⁹ PIK3CAE542K cDNA was amplified from pHAGE-PIK3CAE542K (Addgene, plasmid # 116479, gift from G. Mills and K. Scott) and was subcloned into pCW57.1, a third-generation inducible lentiviral expression vector (Addgene, plasmid # 41393, gift from David Root). Q5-site directed mutagenesis (New England Biolabs) was performed on pHAGE-PIK3CAE542K to generate pHAGE-PIK3CAWT plasmid. All plasmids were expressed through lentiviral transduction. For lentiviruses generation, HEK-293T cells were transfected with pHAGE or pCW57.1 constructs, pSPAX2 plasmid, and pMD2.g plasmid (Adggen plasmid # 12260 and #12259, gifts from D. Trono) using calcium phosphate transfection method. Lentiviral supernatants were collected between 48 h and 72 h post transfection. Supernatants containing lentiviruses were filtered purified using 0.45 μm PES filters. Cytoplasmic GFP fluorescence signal through the expression of pHAGE-PIK3CAE542K-ires-GFP or pHAGE-PIK3CAWT-ires-GFP was used to measure transduction efficiency and lentivirus tittering assays were performed to determine the lowest concentration required to achieve >90% transduction efficiency in HUVECs. HUVECs expressing pCW57.1-tetON-PIK3CAE542K were selected with 2 μg/ml of puromycin (Corning) and 5 μg/ml of blasticidin (Fisher) in EGM-2, respectively, for at least two cell passages. Doxycycline (Sigma, 10 μg/ml in EGM-2) was used to induce the expression of tet-ON-PIK3CA^WT or tet-ON-PIK3CA^E542K.

HUVECs were seeded at 0.15 × 10⁶ cells per well in a 6-well plate (Corning Costar 6-well Clear TC-treated, Cat #3516) coated with 50 μg/ml of rat-tail derived type I collagen (Corning 354326) in 0.02 M acetic acid solution. Cells were allowed to adhere overnight prior to the application of orbital shaking. Fresh EGM-2 (2 ml) was added into each well prior to the application of orbital flow. Orbital flow was applied by culturing ECs for 3 days on an orbital shaker set to 200 rpm, which corresponds to a maximum wall shear stress of 15 Dyn/cm².⁶⁷ Computational fluid dynamics (CFD) models verified with particle imaging velocimetry (PIV) demonstrate that a well of this geometry, fill volume, and rotational speed will generate periodic wall shear stresses with mean values spanning from 0.8 to 13 Dyn/cm² from near the center of the well to the outer edge, respectively.⁶⁷ For measurements of local changes in vascular permeability, 0.3 × 10⁶ cells HUVECs were seeded in each well of a 6-well plate coated with 50 μg/ml of biotinylated-human fibronectin (Corning 354008) in PBS. Fibronectin biotinylation was performed with EZ-link-Sulfo-NHS-LC-Biotin (Thermo Fisher A39257) as described previously.⁶⁸ Briefly, 0.1 mg/ml of fibronectin in PBS was incubated with 0.5 mM of EZ-link-Sulfo-NHS-LC-Biotin for 30 min at room temperature. At the end of orbital flow culture, cells were incubated with 25 mg/ml Cy3-Streptavidin (Jackson Immuno Research Labs 016160084) in PBS containing calcium and magnesium (PBS++) for 2 min. After streptavidin labeling of exposed matrix, cells were rinsed once with PBS++ and were fixed with 4% paraformaldehyde (Electron Microscopy Sciences) in PBS++ for 15 min at room temperature. Permeability was measured by quantifying the streptavidin positive area using the default auto thresholding method in ImageJ.

Microfluidic devices were fabricated as described in the supplementary material. To improve hydrogel adhesion to PDMS devices prepared by photolithography, 2 mg/ml dopamine hydrochloride in 10 mM of pH 8.5 Tris HCl buffer (BioWorld) was pipetted into the hydrogel chamber, incubated for 1 h at room temperature, rinsed twice with DI-H₂O, left dry at room temperature, and sterilized via UV for 20 min. Sterile 160 μm stainless steel acupuncture needles (Llahsa Oms) were coated with 0.01% bovine serum albumin (BSA) in PBS for at least 30 min prior to hydrogel injection. BSA-coated needles were inserted into each device, and 2.5 mg/ml of rat tail-derived collagen I hydrogel was prepared and pipetted into the hydrogel chamber as described previously,³⁷ and allowed to polymerize at 37 °C for 30 min prior to hydrating with EGM-2. After 4 h of further incubation at 37 °C, acupuncture needles were gently removed from devices to form a cylindrical channel within the collagen I hydrogel. Needle inlets were sealed with vacuum grease (Millipore Sigma) and fresh EGM-2 was added to the media ports of the devices. Devices were kept on a rocker at 5 cycles/min at 37 °C and 5% CO₂ for at least 12 h prior to cell seeding to rinse and equilibrate the collagen I hydrogel.³⁷ To form microvessels, HUVECs were resuspended at 0.3 × 10⁶ cells/ml in EGM-2. Media was removed from both media ports and gel filling pipette tips were used to aspirate and empty the hydrogel ports prior to seeding. To promote pressure driven flow into the microchannel, 50 and 40 μL of cell suspension were added to the inlet and outlet media ports, respectively. Devices were inverted every minute for 8 min at room temperature, followed by incubating upside-down in a tissue culture incubator for 6 min. Devices were subsequently inverted and incubated for an additional 6 min in the incubator. After 20 min of seeding, devices were placed on a rocker for at least 2 h before changing EGM-2 media, which was subsequently changed daily.

Two-channel devices were used to apply transmural flow into or out of microvessels (Fig. S6). For the application of transmural flow, microvessels were first cultured on rocker (5 cycles per minute at ±30° angle) for 2 days. Media reservoirs were inserted into all four media ports and any gaps between reservoirs and PDMS were sealed with vacuum grease. Pre-warmed EGM-2 was added either to the empty or microvessel channel to apply a 20 mm H₂O hydrostatic pressure driven transmural flow into or out of the microvessels, respectively. Microvessels were fixed 24 h after the application of transmural flow.

Microfluidic devices were fabricated as described in the supplementary material. The micropipette aspiration device was used to deform cells under an applied pressure gradient. Untransduced HUVECs, PIK3CA^WT, and PIK3CA^E542K cells were cultured as described previously before being stained with Hoechst 3342 (10 mg/ml, Thermo Scientific) and CellMask Deep Red plasma membrane stain (5 mg/ml, 1:2000 v/v, Invitrogen) for 10 min to stain the nucleus and plasma membrane, respectively. Stained cells were washed with PBS and allowed to recover in EGM-2 for 30 min. Cells were then trypsinized, counted, and resuspended at a concentration of 5 000 000 cells/ml in BSA (20 mg/ml in PBS) before being put on ice. Constant-pressure microfluidic flow controllers (“Flow EZ,” Fluigent) were used to apply flow to the opposing inlets at constant pressures, creating flow from one inlet to the other using a known pressure gradient. Two flow controllers were each connected to a dry air source and a power outlet. One controller was used to drive stained and suspended cells from a 15 ml conical tube into one inlet of the microfluidic device at a pressure of 7.0 kPa. A second controller was used to drive cell-free PBS from a 15 ml conical tube into the second inlet of the microfluidic device at a pressure of 1.4 kPa. The outlet port of the device was open to the atmosphere (Fig. S7).

Fresh sterile decellularization solution (0.5% Triton with 50 mM NH₄OH in PBS) was prepared on the day of decellularization. Microvessels were rinsed once with sterile PBS prior to decellularization. After PBS wash, 100 μL of decellularization solution was added to the media ports of each device. Devices were kept on a rocker at 37 °C and 5% CO₂ for 15 min. Decellularized devices were washed three times with PBS and residual DNA was digested via treatment with 10 μg/ml of DNAseI (Sigma Aldrich, D4527) in PBS++ for 45 min at 37 °C. DNAseI treated devices were washed three times with PBS and were incubated overnight on a rocker in fresh EGM-2 media. Devices were reseeded with PIK3CA^WT cells the next day as described in the “Hydrogel formation and cell seeding” section. Basal-to-apical pressure driven transmural flow was applied a day after reseeding as described above.

Mouse Anti–VE-cadherin (F-8) was from Santa Cruz Biotechnology. Rabbit Anti-VE-cadherin (ab33168) was from Abcam. Mouse Anti-Vinculin (hVIN-1) was from Sigma Aldrich. 4′,6-Diamidino-2-phenylindole (DAPI), rhodamine phalloidin (1 mg/ml), Alexa Fluor 488 phalloidin (1 mg/ml), and Alexa Fluor–conjugated secondary antibodies were from Life Technologies. Unless otherwise mentioned, for end point immunostaining, cells or microvessels were fixed in 4% PFA in PBS++ for 20 min at room-temperature. For microtubule staining, cells were fixed in 100% Methanol for 3 min at −20 °C. Fixed samples were subsequently washed and rehydrated with PBS. For visualization of vinculin at focal adhesion, cells were subjected to mild permeabilization using 0.1% Triton X-100 in 4% PFA. Cells were partially perm-fixed for 5 min prior to switching to complete fixation in 4% PFA for 15 min at room-temperature. All samples were permeabilized with 0.3% Triton X-100 (Millipore Sigma) for 30 min at room temperature and nonspecific antibody binding was blocked by treatment with 2% (w/v) BSA in PBS for an hour at room temperature. Primary antibodies against VE-cadherin (1:200, v/v, F-8, Santa Cruz Biotechnology; 1:250, v/v, ab33168, Abcam), vinculin (1: 400, v/v, hVIN-1, Sigma Aldrich), α-tubulin (1:1000, v/v, DM1A, Cell Signaling), and vimentin (1:200, v/v, D21H3, Cell Signaling) were diluted in 2% BSA blocking solution and applied overnight on a laboratory rocker at 4 °C. Cells and devices were subsequently rinsed three times over 15 min with PBS, and were incubated in secondary antibodies (1:1000, v/v, goat anti-mouse or goat-anti rabbit immunoglobulin G conjugated to Cy3 or Alexa Fluor 647, Thermo Fisher Scientific), Alexa Fluor 488 or rhodamine phalloidin (1:200, v/v, Thermo Fisher Scientific), and DAPI (1:1000, v/v, Thermo Fisher Scientific) diluted in blocking solution for an hour at room temperature. Labeled cells and devices were rinsed three times over 15 min with PBS and were stored in 0.01% Sodium Azide solution diluted in PBS until imaging. Orientation of actin fibers was quantitatively determined from images of phalloidin using the FibrilTool plugin for ImageJ.⁶⁹ 

Cell lysates were collected from cells cultured in complete media for 24 h. Briefly, cells were washed once with ice cold PBS and lysed on ice with radioimmunoprecipitation assay buffer (RIPA buffer, Thermo Fisher Scientific) containing 1X HALT protease and phosphatase inhibitor (Thermo Fisher Scientific). Lysates were homogenized by pulse-vortexing and clarified by centrifugation at 14 000 × g for 5 min at 4 °C. The concentration of clarified lysates was quantified using Pierce BCA Protein Assay (Thermo Fisher Scientific). Lysates were then reduced with NuPAGE LDS reducing agent and dithiothreitol and denatured for 5 min at 100 °C. Denatured proteins were resolved in NuPAGE 4 to 12% bis-tris gradient gels (Thermo Fisher Scientific) and transferred to iBlot2 PVDF ministack membranes. Membranes were blocked with Super-Block Blocking Buffer (Thermo Fisher Scientific) for an hour with rocking at room temperature and were incubated with the following primary antibodies overnight with rocking at 4 °C: anti-phosphoSRC(Tyr416) (1:1000, Cell Signaling Technology, #2101), anti-Src (1:1000, Cell Signaling Technology, #2108), anti-p110α (1:1000, Cell Signaling Technology, #9272), anti-vinculin (1:10 000, Sigma-Aldrich, V9264), anti-VE-cadherin (1:1000, Santa Cruz, SC9989), anti-Beta-catenin (1:1000, Cell Signaling Technology, #8480), anti-glyceraldehyde-3-phosphate dehydrogenase (1:5000, Cell Signaling Technology, #2118), anti-phospho-myosin light chain 2 (Thr18/Ser19) (1:1000, Cell Signaling Technology, #3674), and anti-myosin light chain 2 (1:1000, Cell Signaling Technology, #8505). After overnight incubation with primary antibodies, membranes were washed three times with TBS-T and were subsequently incubated with horseradish peroxidase conjugated antibodies (1:10 000, v/v, diluted in 5% milk (w/v)) for 1 h with rocking at room temperature. Membranes were then washed three times over a period of 15 min with TBS-T and were developed via chemiluminescence using either Super-Signal West Femto Maximum Sensitivity Substrate reagent (Thermo Fisher Scientific) or Clarity Western ECL reagent (BioRad). Images of western blots were quantified using ImageJ.

For pulse-chase experiment, HUVECs were seeded and subjected to orbital shaking as described in the Methods section. After conditioning with orbital flow for 3 days, cells were incubated with AF647-anti-VE-cadherin antibody (1:200, v/v, diluted in pre-warmed EGM-2 media, BD Biosciences, #561567) for 30 min orbital shaking. Unbound antibody was removed with two washes with pre-warmed EGM-2 media, and cells were cultured for additional 2 h under orbital shaking prior to fixing.

Confocal z-stacks of microvessels (4 μm step size, captured from top to bottom of the microvessels) were acquired with an Olympus FV3000 laser scanning confocal microscope with a 10× U Plan S-Apo NA 0.4 air objective. Confocal z-stack images of cells (0.8 μm step size, captured from top to bottom of the samples) stained for VE-cadherin, vinculin, and cytoskeletal proteins were captured with a 30× U Plan S-Apo NA 1.05 silicon oil objective. Fluorescence images of cells after orbital flow were captured at 10× magnification (10× U Plan Fluor NA 0.3 air objective, Olympus) using an automated scanning stage on an Olympus wide-field microscope. Briefly, the edge of a well of a 6-well plate was positioned over the objective, and a uniform rectangular area from the edge to the middle of the well was selected. A focus map was generated to include at least 10 focus points along the length of the rectangular area. The autofocus feature was used to cycle through the focus map and set the z-focus for each imaging point used throughout the area scan. Computational image processing is described in the supplementary material.

Cell deformation into the micropipette channels was imaged using a timelapse acquisition on an inverted widefield fluorescence microscope (Olympus IX83) with a white light source 20X/NA 0.45 objective. DAPI and Cy5 filters were used to image the nucleus and cytoplasm, respectively, and images were acquired every 2.935 s for a total acquisition time of 2.5 min. At the end of each timelapse acquisition, deformed cells were cleared from the cell pockets by applying pressure at the outlet port by pipette. Cleared pockets were then filled again by flowing cells during the next acquisition period.

Graphs were generated and statistical analyses were performed in Prism 10 (GraphPad). Unless otherwise mentioned, data were plotted as mean ± standard deviation of ≥3 experiments and data points denote values from each cell or microvessel. Dataset with normal distribution were compared with two-tailed Student's test and one-way ANOVA followed by Tukey's post hoc test. Kruskal–Wallis test followed by Dunn's multiple-comparison was used for comparison of non-Gaussian distributed data. Two-way ANOVA followed by Šidák's test was used for comparison of more than two variables.

See the supplementary material for detailed microfabrication methods for the engineered microvessel and micropipette aspiration microfluidic devices, as well as methods for quantitative image processing. Also included in the supplementary material are supplemental results figures referenced in the Results.

This work was funded by the National Institutes of Health (No. R35GM142944), by the American Heart Association (No. CDA857738), and by research grants from the CLOVES Syndrome Community and the Lymphatic Malformation Institute. W.Y.A. was supported by grant from the CLOVES Syndrome Community. C.P.W. and E.L.D. were supported by a training fellowship from the Integrative Vascular Biology Training Program (No. T32HL69768), C.P.W. was supported by an American Heart Association predoctoral fellowship (No. 24PRE1192185) and E.L.D. was supported by a Ruth L. Kirchstein predoctoral individual fellowship (No. F31HL162462). E.M. acknowledges support from an NIH training fellowship (No. T32GM135128). We thank Bob Geil for advice and help with microfabrication. The fabrication of microfluidic devices was performed in the Chapel Hill Analytical and Nanofabrication Laboratory (CHANL), a member of the North Carolina Research Triangle Nanotechnology Network (RTNN), which was supported by the National Science Foundation (No. ECCS-2025064), as part of the National Nanotechnology Coordinated Infrastructure (NNCI).

The authors have no conflicts to disclose.

Ethics approval is not required.

Wen Yih Aw: Conceptualization (equal); Data curation (lead); Formal analysis (lead); Funding acquisition (equal); Investigation (lead); Methodology (lead); Writing – original draft (lead); Writing – review & editing (equal). Aanya Sawhney: Formal analysis (equal); Investigation (equal); Methodology (equal); Writing – review & editing (equal). Mitesh Rathod: Formal analysis (equal); Investigation (equal); Methodology (equal); Writing – review & editing (equal). Chloe P. Whitworth: Formal analysis (equal); Investigation (equal). Elizabeth L. Doherty: Investigation (equal); Methodology (equal); Writing – review & editing (equal). Ethan Madden: Investigation (equal); Writing – review & editing (equal). Jingming Lu: Investigation (equal). Kaden Westphal: Investigation (equal). Ryan Stack: Visualization (equal); Writing – original draft (equal). William J. Polacheck: Conceptualization (lead); Funding acquisition (lead); Methodology (equal); Writing – original draft (equal); Writing – review & editing (lead).

The data that support the findings of this study are available from the corresponding author upon reasonable request.