Shear Difference: Flow Type Dictates Endothelial Flow‐Responsive Gene Programs in a 3‐Dimensional–Printed In Vitro Model

Abstract

Background

Endothelial cells are mechanosensitive and adopt distinct phenotypes in response to hemodynamic forces. These responses are difficult to study in conventional in vitro platforms, which rarely reproduce vessel‐scale geometry or clinically relevant flow. Our aim was to determine how flow impacts endothelial morphology and transcriptional activity in a 3‐dimensional macrofluidic model.

Methods

Idealized vessels were 3‐dimensional printed using a water‐soluble polyvinyl alcohol and cast in polydimethylsiloxane. After core dissolution, human microvascular endothelial cell line‐1 cells were grown on the lumen and perfused for 24 hours under static, continuous, or pulsatile flow. The pulsatile waveform was derived from arteriovenous fistula Doppler profiles and scaled to match mean volumetric flow (~100 mL/min) and time‐averaged wall shear stress (~1.5 dyn/cm²) to continuous flow. Morphology was assessed using immunofluorescence. Bulk RNA‐sequencing and gene set enrichment analysis were performed.

Results

Relative to static culture, continuous flow increased cell eccentricity (0.74 versus 0.48; P<0.0001) and reduced orientation variability (Δ=−47.1°, P<0.0001). Differential gene expression was extensive (continuous versus static: 2103 genes; pulsatile versus static: 2643 genes; pulsatile versus continuous: 384 genes). Continuous flow reduced interferon signaling and the Hallmark inflammatory response program relative to static, whereas tumor necrosis factor‐α/nuclear factor‐κ–light‐chain enhancer of activated B cells signaling was increased. Under matched mean shear, pulsatile flow enriched cell cycle/checkpoint programs. Conversely, continuous flow enriched oxidative phosphorylation and p53 pathways. Transforming growth factor‐β signaling was enriched in pulsatile flow.

Conclusions

Under matched mean shear, pulsatile and continuous flow were associated with distinct endothelial morphologic and transcriptional signatures. This macrofluidic platform provides a validated, waveform‐controlled testbed for mechanistic and translational studies.

Nonstandard Abbreviations and Acronyms

EC

endothelial cell

HMEC‐1

human microvascular endothelial cell line‐1

KLF

Krüppel‐like factor

NF‐κB

nuclear factor‐κ–light‐chain enhancer of activated B cells

NES

normalized enrichment score

NO

nitric oxide

PI

pulsatility index

PVA

polyvinyl alcohol

TGF‐β

transforming growth factor‐β

Womersley number α

dimensionless index of pulsatile‐flow unsteadiness

WSS

wall shear stress

Research Perspective.

What Is New?

• A 3‐dimensional–printed vessel‐scale macrofluidic model showed that, even under matched mean volumetric flow and time‐averaged wall shear stress, continuous and pulsatile flow produce distinct endothelial transcriptomic programs.

• Compared with continuous flow, pulsatile flow was associated with relative enrichment of cell cycle and checkpoint pathways, whereas continuous flow was associated with relatively greater oxidative phosphorylation and p53 pathway enrichment, indicating that waveform itself shapes endothelial state beyond mean shear alone.

What Question Should Be Addressed Next?

• Future studies should determine how patient‐specific vascular geometry and clinically derived waveforms interact with endothelial, smooth muscle, and inflammatory cell signaling to drive vascular remodeling and dysfunction.

Endothelial cells (ECs) form a continuous, metabolically active monolayer with ~10 trillion cells lining ~350 m² of the adult vasculature. ¹ , ² At the blood–tissue interface, ECs function as biochemical sensors and mechanotransducers, integrating circulating cues with hemodynamic forces to regulate anticoagulant, antithrombotic, and vasoactive outputs, including nitric oxide, prostacyclin, endothelial‐derived hyperpolarization, C‐type natriuretic peptide, and membrane‐bound constituents, such as heparinoids, thrombomodulin, and tissue plasminogen activator. ³ , ⁴

Within intact vessels, blood flow over complex geometries imposes spatially and temporally varying mechanical inputs on ECs, including radial forces linked to intravascular volume, tangential forces at cell–cell interfaces, and the frictional forces of flowing blood, termed wall shear stress (WSS). ⁵ In the arterial wall, WSS is the dominant endothelial stimulus, ⁶ whereas vascular smooth muscle principally senses intraluminal pressure and circumferential strain. ⁷ As WSS correlates with viscosity, volumetric flow, and luminal radius, even minor changes attributable to curvature, taper, or branch points can generate large local changes in WSS that have a significant impact on endothelial phenotypes. ⁸ Endothelial function is therefore critically dependent on mechanotransduction by coupling luminal flow to nitric oxide (NO) production, ion‐channel activity, cytoskeletal remodeling, gene expression, and ultimately tissue perfusion. ⁵ , ⁹ Although these core signaling pathways are conserved, venous ECs display distinct baseline set points and sensitivity to shear that alter transcriptional and functional responses that are distinct from those in arterial ECs. ¹⁰

Endothelial mechanosensing relies on receptors and macromolecule complexes to initiate downstream signaling; these include integrins, ¹¹ , ¹² , ¹³ receptor tyrosine kinases, ¹⁴ G‐protein–coupled receptors, ¹⁵ ion channels, ¹⁶ , ¹⁷ and intercellular junctional proteins. ¹⁸ Flow magnitude and pattern also impact on distinct biologic states. Steady laminar WSS engages lipid signaling and kinase cascades that converge on phosphoinositide 3‐kinase–protein kinase B serine/threonine kinase–dependent endothelial NO synthase activation, resulting in NO‐mediated vasodilation and suppression of inflammatory programs. ⁵ Oscillating WSS increases expression of nuclear factor‐κ–light‐chain enhancer of activated B cells (NF‐κB), and promotes an adhesive, procoagulant, and proinflammatory phenotype typical of atherosclerosis‐prone regions. ⁹ These shear‐responsive signaling cascades control cytoskeletal architecture, polarity, junctional integrity, cell cycle progression, and proliferation. ¹⁹ , ²⁰ Although the complete transcriptional control of mechanotransduction remains to be resolved, several key factors are involved, including Krüppel‐like factor‐2 (KLF2), NF‐κB, AP‐1 (activator protein‐1), and YAP (Yes‐associated protein). ²¹ , ²²

Although traditional 2‐dimensional cell culture provides compositional control and throughput, it constrains ECs to rigid 2‐dimensional surfaces lacking physiological curvature and authentic hemodynamics. This results in a well‐documented divergence in cell morphology, cell–cell junctional organization, proliferation, differentiation, metabolism, and pharmacologic responses relative to in vivo states. ²³ , ²⁴ , ²⁵ Three‐dimensional (3D) culture, such as spheroids, organoids, and engineered tissues, better recapitulate the morphology, polarity, junctional architecture, and transcriptome of ECs in vivo, ²⁶ , ²⁷ , ²⁸ , ²⁹ , ³⁰ but often do not recapitulate physiological shear. ²⁶ , ²⁹ Advances in high‐resolution imaging and additive manufacturing now enable precise capture of patient‐specific geometries ³¹ , ³² and accurate fabrication ³³ , ³⁴ that reproduces vessel curvature, taper, and branching. Growing ECs on such printed constructs under controlled hemodynamic environments generate a macrofluidic testbed that mimics the experimental control of classic in vitro systems, as well as the geometric determinants of WSS that govern endothelial phenotype in vivo.

The aim of this study was to develop a 3D‐printed macrofluidic platform that supports EC growth, coupled with a pump system to enable perfusion of continuous laminar flow or programmable, physiologically relevant pulsatile flow. The morphologic and transcriptomic impacts of these properties on ECs were characterized to demonstrate the critical role of both WSS and flow profile in endothelial mechanotransduction.

METHODS

Ethics Statement

This study used only an established immortalized cell line and did not involve human participants, identifiable human data or biospecimens, or animals. Accordingly, institutional review board approval, informed consent, and animal ethics approval were not required.

Data Availability

Data supporting the findings of this study are available from the corresponding author on reasonable request. The high‐throughput data generated for this work have been deposited in the National Center for Biotechnology Information's Gene Expression Omnibus ³⁵ and are accessible through Gene Expression Omnibus Series accession number GSE328243 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE328243). Analytic methods (including code used for preprocessing and statistical analyses), study materials, and laboratory protocols are available from the corresponding author on request.

Additive Manufacturing and Polydimethylsiloxane Casting of Idealized Vessel Geometries

Idealized 6‐cm vessels were designed using commercial computer‐aided design software, SolidWorks (2023; Dassault Systèmes, Vélizy‐Villacoublay, France). Several soluble printing filaments were screened on a fused‐deposition manufacturing 3D printer (Prusa i3 MK3S; Prusa Research, Prague, Czech Republic). These included 3 polyvinyl alcohol (PVA) filaments (eSUN PVA, 1.75 mm, Shenzhen eSUN Industrial Co, Ltd, Shenzhen, China; MatterHacker Pro Series, 1.75 mm, MatterHackers Inc, Lake Forest, CA; FormFutura Aquasolve PVA, B.V., Nijmegen, the Netherlands), 1 butenediol vinyl alcohol co‐polymer (BVOH) filament (BASF Ultrafuse, 1.75 mm, BASF 3D Printing Solutions GmbH, Heidelberg, Germany), and 1 lye‐soluble copolymer (Xioneer VXL90, 1.75 mm, Xioneer Systems GmbH, Vienna, Austria). These printing filaments were assessed using a temperature‐tower calibration print (https://www.thingiverse.com/thing:2729076, Figure S1). Postprint surface smoothing methods tested included the following: sandpaper, wet wiping, static dip in room temperature water for 30 seconds, static dip in room temperature water for 60 seconds, and static dip in room temperature water for 180 seconds. Postmodification surface roughness was assessed using a scanning laser microscope (Keyence VK‐X200, Keyence Corporation, Osaka, Japan) (Figure S1). Polydimethylsiloxane (Sylgard 184, Dow Corning, supplied by CBC Bearings, Australia) prepolymer and curing agent were mixed in a 10:1 weight ratio and degassed for 60 minutes. Once smoothed, PVA cores were positioned in acrylic resin molds (designed in commercial computer‐aided design software, SolidWorks) with a 0.5‐mm offset and filled with polydimethylsiloxane. Polydimethylsiloxane was cured at room temperature for 48 hours and then immersed in agitated room‐temperature water for 72 hours to dissolve the PVA core geometries.

Extracellular Matrix Surface Coating and Endothelial Monolayer Formation

Surface modification was tested by incubating polydimethylsiloxane surfaces in 1% (w/v) gelatin, 10 μg/mL fibronectin (Sigma‐Aldrich, Merck; F0895‐5MG), 20 μg/mL fibronectin, 50 μg/mL fibronectin, and 1% (w/v) Pluronic F‐127 (Invitrogen, catalog No. P6866) for 1 hour at 37 °C (5% CO₂) or by using a 1% (w/v) polyethylene glycol–polydimethylsiloxane (Specific Polymers, catalog No. SP‐8P‐3‐001) additive to the polydimethylsiloxane prepolymer base agent (Figure S2). Coating solution was removed, and the surface was air dried for 60 minutes at room temperature. Human microvascular endothelial cell line‐1 (HMEC‐1, ATCC; CRL‐3243) were maintained in MCDB‐131 complete medium (Gibco; 10372019) supplemented with 10 ng/mL human epidermal growth factor (Sigma‐Aldrich, Merck; E9644), 1 μg/mL hydrocortisone (Sigma‐Aldrich, Merck; H‐4001), 10 nM L‐glutamine (Gibco; 25030‐081), and 10% (v/v) fetal bovine serum (Gibco; 10099‐141) at 37 °C and 5% CO₂. HMEC‐1 ECs were seeded onto the model surfaces at a density of 3.75×10⁴ cells/cm² in a sample spinner for 24 hours at 37 °C and 5% CO₂ (Figure S3).

Manufacturing of Flow System and Generation of Flow

Three‐dimensional components were designed using Solidworks computer‐aided design software and fabricated using heat‐resistant resin (TR‐300 resin, Phrozen, Hsinchu, Taiwan; catalog No. TR‐300) on a stereolithography 3D printer (Sonic Mighty 12K SLA printer, Phrozen, Hsinchu, Taiwan). The perfusion system consisted of a closed loop incorporating two 100‐mL MACS GMP Cell Differentiation Bags (Miltenyi Biotec, Bergisch Gladbach, Germany; catalog No. 170‐076‐400) as culture media reservoirs, custom‐printed connectors, and a peristaltic pump (silicone tubing LongerPump No. 25 tubing, LongerPump Co, Ltd, China). Flow was generated using a programmable peristaltic pump (NE‐9000; New Era Pump Systems Inc, Farmingdale, NY) and monitored using an ultrasonic flow meter (Sonoflow CO.55; Sonotec GmbH, Halle, Germany) positioned immediately upstream of the vessel segment. To generate pulsatile flow, patient‐derived Doppler ultrasound data from arteriovenous fistulas (n=13) were analyzed to extract magnitude of baseline flow, time at baseline flow, time to peak flow, magnitude of peak flow, time at peak flow, time to postpeak baseline, magnitude of postpeak baseline flow, and time at postpeak baseline flow over 1 cardiac cycle. Across the cohort, the mean volumetric flow was 595±281 mL/min, with an average minimum of 463±240 mL/min and an average maximum of 779±352 mL/min. The average cardiac cycle duration was 0.84 seconds, with mean time to peak 0.334 seconds, peak duration 0.20 seconds, and time from peak back to baseline 0.347 seconds. The flow profiles from these data were scaled to achieve an average flow of 100 mL/min (85–145 mL/min over 0.84 seconds) and programmed to generate a realistic pulsatile waveform. Flow conditions were maintained for 24 hours of perfusion following an initial 24‐hour static attachment phase. Except for the pump, the system was housed and operated within a humidified incubator at 37 °C (5% CO₂). Culture media temperature measured in the reservoir remained stable at incubator set point during operation.

Flow characteristics under continuous flow and pulsatile flow conditions were approximated under the assumptions of steady laminar flow in a straight, smooth, rigid, circular tube of 4.8‐mm internal diameter. The culture medium was modeled as an incompressible, Newtonian, isothermal fluid with constant viscosity and density. Medium density and viscosity were assumed (not directly measured) and set to ρ=1000 kg/m³ and μ=0.001 Pa s (ν=1.0×10⁻⁶ m²/s) at 37 °C. Gravitational and entrance effects, wall compliance, and fluid‐structure interaction were neglected. With these assumptions, hemodynamics follow the bulk‐flow law ³⁶ where flow pattern is characterized by the Reynolds number (Re)=ρVD/μ (ρ, density; μ, viscosity; V, bulk velocity; D, diameter). Mean flow refers to the time‐averaged volumetric flow rate (Qmean); for pulsatile waveforms, Qmean=(1/T)∫₀ᵀ Q(t)dt, and bulk velocity is defined as V=Q/A using the lumen cross‐sectional area (A). WSS (τw) for laminar tube flow is given by τw=4μQ/(πr³), or equivalently τw=8μV/D. Pulsatility index (PI) was defined as PI=(Qmax−Qmin)/Qmean, and for pulsatile flow, unsteadiness was quantified using the Womersley number (α), defined as α=R(ω/ν)½ (R, radius; ω, angular frequency; ν, kinematic viscosity). For pulsatile conditions, a quasi‐steady approximation was applied such that instantaneous wall shear stress scaled linearly with the instantaneous flow rate with no flow reversal. Continuous and pulsatile conditions were calibrated to match Qmean (≈100 mL/min), yielding matched time‐averaged τw and comparable mean Re, while differing in pulsatility amplitude (PI) and temporal shear gradients.

Morphologic Assessment by Fluorescence Microscopy

At the end of flow exposure, the constructs were fixed and stained with Phalloidin–Atto‐565 (94072; Merck, Darmstadt, Germany) and counterstained with 4',6'‐diamidino‐2‐phenylindole (DAPI, D1306; Thermo Fisher Scientific, Waltham, MA). Imaging was performed using a Zeiss LSM 880 confocal microscope (Carl Zeiss, Oberkochen, Germany) housed at the Katharina Gaus Light Microscopy Facility, Mark Wainright Analytical Centre, University of New South Wales, Sydney. Images were analyzed in Fiji (ImageJ) (Fiji distribution v2.16.0/ImageJ v1.54p; RRID:SCR_002285/RRID:SCR_003070) with default settings.

Bulk RNA Sequencing

Total RNA was isolated using a RNeasy Plus Mini Kit (74134, QIAGEN, Hilden, Germany) according to manufacturer's instructions. RNA integrity was verified using an Agilent TapeStation system (Agilent Technologies, Santa Clara, CA), with all samples showing excellent quality (RNA Integrity Number equivalent ‐ RINe ≥9.9) (Figure S4). Libraries were constructed from total RNA using the Illumina stranded mRNA workflow, dual indexed, PCR amplified within recommended cycle limits, pooled equimolarly, and sequenced on an Illumina NovaSeq X by the Ramaciotti Centre for Genomics (UNSW Sydney, Australia) (Figure S5). Raw paired‐end RNA‐sequencing reads were staged from archives and assessed with FastQC for base quality, Q30 rate, GC content, adapter/contaminant sequences, and duplication. Reads were adapter/quality trimmed with fastp where possible, and FastQC/fastp outputs were aggregated with MultiQC. Trimmed reads were aligned to the human genome using Rsubread, BAMs were quality control (QC) checked, and gene‐level counts were generated with featureCounts. Differential expression was performed in differential expression analysis for sequence count data 2, using a likelihood ratio test for the overall effect of condition and Wald tests for pairwise contrasts. Raw P values were adjusted for multiple testing using the Benjamini–Hochberg procedure to control the false discovery rate, and log₂ fold changes for pairwise contrasts were shrunken using apeglm. ³⁷ Genes were considered differentially expressed at false discovery rate–adjusted P<0.05 and |log₂ fold change| ≥1. Hallmark pathway enrichment was assessed by overrepresentation (clusterProfiler) and preranked gene set enrichment analysis (fgsea). Results were ranked by false discovery rate and normalized enrichment score (NES). The top 30 enriched pathways (false discovery rate <0.05) are reported. The full pipeline is summarized in Figure 1.

Figure 1. Fabrication and validation of a 3D vessel‐mimetic flow platform.

A, 3D‐printed idealized vessel used as the sacrificial core/master for casting. B, Surface smoothing optimization was assessed by laser profilometry. A 30‐second water dip was found to be the most effective method. The numeric labels above each panel indicate the measured mean roughness (μm) for the imaged field. C, PDMS casting around the printed core and subsequent dissolution of the core to yield a perfusable channel. D, HMEC‐1 viability on fibronectin‐coated PDMS was confirmed by Live/Dead staining. E, Flow cytometry validation of endothelial responsiveness on PDMS: 6‐hour TNF‐α stimulation increases ICAM‐1 (CD54) and VCAM‐1 (CD106) expression compared with unstimulated controls. F, Schematic of the closed‐loop perfusion rig showing bubble trap/media reservoir and a programmable peristaltic pump. The second bubble trap (immediately downstream of peristaltic pump) can be removed when modeling physiological flow. G, This setup delivers either steady continuous flow (purple) or a pulsatile (blue) waveform as demonstrated by a representative trace. Scale bars: B=213.325 μm, D=400 μm. 3D indicates 3 dimensional; HMEC‐1, human microvascular endothelial cell line‐1; ICAM‐1, intercellular adhesion molecule 1; PDMS, polydimethylsiloxane; TNF‐α, tumor necrosis factor‐α; and VCAM‐1, vascular cell adhesion molecule 1.

Statistical Analysis

Descriptive statistics are reported as frequencies (percentages) for categorical variables, and as mean±SD or median (interquartile range) for continuous variables, as appropriate. For image‐derived morphology outcomes, individual cell measurements were aggregated to the biological replicate before inference by calculating a sample‐level mean for each outcome (static, n=5; continuous, n=5; pulsatile, n=4). Group differences were assessed using exact permutation 1‐way tests. Where the global test was significant, pairwise permutation tests were performed with Bonferroni correction for multiple comparisons. Exact 2‐sided P values are reported, and P<0.05 was considered statistically significant. All statistical analyses were performed in R (R Core Team, 2025) using RStudio (Posit team, 2025).

RESULTS

Construction and Characterization of a 3D Macrofluidic System

Soluble printing filaments were screened using temperature‐tower calibration prints. The most consistent extrusion and dimensional fidelity was achieved with eSUN PVA (Figure S1A). Trials of surface smoothing methods (Figure S1B–S1G) identified a 30‐second static water dip reduced topographic variability (47.4±14.8 to 14.03±0.9 μm; P<0.001) without introducing deforming features (Figure 2B). PVA prints were cast in polydimethylsiloxane, then washed away once the polydimethylsiloxane was cured, leaving behind a lumen within the polydimethylsiloxane of the printed geometry (Figure 2C). Among multiple surface coatings tested (Figure S2), a single‐component fibronectin coating at a concentration of 10 μg/mL for between 1 and 24 hours provided the most robust cellular adhesion. Seeding HMEC‐1 cells at a starting density of 3.75×10⁴ cells/cm² consistently produced a confluent layer of endothelial on the polydimethylsiloxane lumen after 24 hours (Figure S3). The HMEC‐1 monolayer protocol derived from these data included fibronectin surface modification between 1 and 24 hours, HMEC‐1 seeding density of 3.75×10⁴ cells/cm², and culture in static conditions for ≥24 hours to obtain a robust endothelial monolayer. Live/Dead staining confirmed viability of all adherent cells under these conditions (Figure 2D). Preservation of EC functional responsiveness was confirmed, with tumor necrosis factor (TNF)‐α stimulation resulting in significant increases in intercellular adhesion molecule 1 (ICAM‐1) (272.3±23.0%, P<0.001) and vascular cell adhesion molecule 1 (VCAM‐1) (648.2±66.6%, P<0.01) expression compared with controls (Figure 2E). Continuous flow was generated with a mean flow of 98.85±0.8 mL/min (minimum, 98 mL/min; maximum, 101 mL/min) (Figure 2G). Pulsatile flow was modeled from patient arteriovenous fistula Doppler ultrasounds to generate a mean flow of 99.32±5.66 mL/min (minimum, 85 mL/min; maximum, 145 mL/min) over a 0.84‐second pulse cycle (Figure 2F).

Figure 2. Bulk RNA‐seq workflow overview.

BAM indicates binary alignment/map; DE, differential expression; FC, fold change; GC%, guanine–cytosine content; GSEA, gene set enrichment analysis; HGNC, HUGO Gene Nomenclature Committee; LRT, likelihood‐ratio test; NB GLM, negative‐binomial generalized linear model; ORA, overrepresentation analysis; PCA, principal component analysis; QC, quality control; Q20/Q30, Phred quality thresholds; RNA‐seq, RNA sequencing; and VT, variance‐stabilizing transform.

With diameter D=4.8 mm, the continuous flow profile yielded τw of 0.150 Pa (1.50 dyn/cm²) at trough, 0.152 Pa (1.52 dyn/cm²) at mean, and 0.155 Pa (1.55 dyn/cm²) at peak, with corresponding Re of 433, 437, and 447, respectively. The pulsatile flow profile spanned τw 0.130 to 0.169 Pa (1.30–1.69 dyn/cm²) from trough to peak (mean, 0.153 Pa, 1.52 dyn/cm²) and Re 376 to 486 (mean, 439). In both cases, wall shear stress remained positive along the conduit, with no inlet or localized flow reversal throughout the cycle owing to the idealized vessel geometry, yielding an oscillatory shear index of 0. For the continuous flow condition, Q_mean=98.85 mL/min, Q_min=98 mL/min, and Q_max=101 mL/min, making PI_cont ≈0.03. For the pulsatile condition, Qmean=99.32 mL/min, Qmin=85 mL/min, and Qmax=110 mL/min, giving PI_pulse≈0.25. As wall shear stress scales linearly with volumetric flow under these laminar conditions, the shear‐based PI was effectively identical to the corresponding flow‐based PI in each condition. Womersley number (α) was calculated as α=R(ω/ν)½, where R=D/2=2.4×10⁻³ m, the cardiac cycle T=0.84 seconds gave a frequency f=1/T≈1.19 Hz, ω=2πf≈7.5 rad/s, and the kinematic viscosity of the medium was ν≈1.0×10⁻⁶ m²/s, yielding α≈6.6.

Flow Impacts EC Morphology

Endothelial morphology and cytoskeletal organization varied as a function of flow as measured by immunofluorescent staining (Figure 3). Under static conditions, cells displayed a cobblestone morphology with no preferred orientation (Figure 3A and 3B). After 24 hours of continuous or pulsatile flow, cells elongated and F‐actin bundles aligned with the direction of flow (Figure 3C and 3D). Using sample‐level aggregation with exact permutation testing, both cell eccentricity and orientation dispersion differed across conditions. Mean eccentricity differed overall (global P=3.6×10⁻⁴), being lowest in static culture (0.79±0.06), intermediate under pulsatile flow (0.91±0.03), and highest under continuous flow (0.96±0.01). In Bonferroni‐adjusted pairwise exact permutation tests, eccentricity was higher under continuous flow than in static culture (P=0.024) and pulsatile flow (P=0.024), whereas the difference between static and pulsatile flow was not statistically significant (P=0.12). Orientation dispersion likewise differed across conditions (global P=2.9×10⁻⁴), with the greatest dispersion in static culture (52.9°±17.7°) and markedly lower values under continuous flow (9.5°±3.5°) and pulsatile flow (11.9°±3.1°). In Bonferroni‐adjusted pairwise comparisons, both continuous and pulsatile flow were associated with lower orientation dispersion than static culture (P=0.024 for both), whereas continuous and pulsatile flow did not differ from one another (P=0.86) (Figure 3H).

Figure 3. Flow‐dependent remodeling of cell morphology and cytoskeleton.

HMEC‐1 cells were grown on fibronectin‐coated PDMS macrofluidic models and then exposed to flow or control conditions for 24 hours. A and B, Static culture for 24 hours. C and D, Continuous flow for 24 hours. E and F, Pulsatile flow for 24 hours. Cyan, nuclei (DAPI); magenta, F‐actin (phalloidin‐Atto565). G, Quantification of cell eccentricity. H, Variation in orientation angle. Scale bars: 100 μm (A, C, and E) or 50 μm (B, D, and F). Each point marks a cell. Statistical testing was performed on sample‐level aggregated values using exact permutation tests, with Bonferroni‐adjusted pairwise comparisons. DAPI, 4',6'‐diamidino‐2‐phenylindole; HMEC‐1 indicates human microvascular endothelial cell line‐1; and PDMS, polydimethylsiloxane.

Flow Alters Transcription of Shear‐Responsive Genes

Bulk RNA sequencing was performed on 3 independent biological replicates per condition. Principal component analysis demonstrated clustering of replicates within condition (Figure 4A), with no separation suggestive of a dominant batch effect.

Figure 4. Altering flow profile results in changes in transcription.

A, PC analysis of variance‐stabilized counts (n=3 biological replicates per condition) demonstrated separation of cells maintained under static conditions (purple) from flow‐exposed samples and a secondary separation between continuous (pink) and pulsatile (blue) flow. B–D, Volcano plots depict differential expression (DESeq2; colored points denote Padj <0.05 with |log2FC| ≥1; dashed lines mark ±1 log2FC). B, Continuous flow vs static. C, Pulsatile flow vs static. D, Pulsatile vs continuous flow. DESeq2 indicates differential expression analysis for sequence count data 2; FC, fold change; FDR, false discovery rate; Padj, adjusted P; and PC, principal component.

In the continuous flow versus static condition comparison, a total of 2103 genes met the significance criteria, with 856 increased and 1247 decreased under continuous flow. Relative to static culture, pulsatile flow similarly induced a canonical shear response. Under pulsatile flow, a total of 2643 genes were significantly differentially expressed, with 1017 increased and 1626 decreased (Figure 4B). A heat map of the top 50 differentially expressed genes comparing continuous flow with static culture recapitulated the principal component analysis, with obvious clustering of flow conditions compared with static culture (Figure 5A).

Figure 5. Differentially expressed genes and pathway analysis in continuous flow vs. static conditions.

A, Heat map of the top 50 differentially expressed genes in continuous flow vs. static conditions comparison across all samples (columns). Values are variance‐stabilized counts (DESeq2) and z‐scored by gene. Rows and columns were hierarchically clustered using 1–Pearson correlation distance and complete linkage. B, NES for pulsatile vs continuous flow (positive=enriched in pulsatile; negative=enriched in continuous). CONT. indicates continuous; DESeq2, differential expression analysis for sequence count data 2; DN, down; IL, interleukin; KRAS, Kirsten rat sarcoma virus oncogene homologue; MTORC, mammalian target of rapamycin complex; NES, normalized enrichment score; and TGF, transforming growth factor; TNFA, tumor necrosis factor‐alpha; WNT, wingless‐related integration site.

Flow Modulates Growth, Stress, and Immune Signaling Programs

Preranked gene set enrichment analysis of Hallmark pathways for continuous flow versus static showed positive enrichment in continuous flow for myelocytoma (MYC) target V1 (NES=2.66, q=5.11×10⁻¹⁸), transforming growth factor (TGF)‐β signaling (NES=2.84, q=1.78×10⁻¹³), MYC target V2 (NES=2.73, q=1.10×10⁻¹¹), unfolded protein response (NES=2.46, q=1.02×10⁻¹⁰), TNF‐α signaling via NF‐κB (NES=2.21, q=6.58×10⁻¹⁰), mammalian target of rapamycin complex 1 signaling (NES=2.15, q=4.89×10⁻⁹), epithelial–mesenchymal transition (NES=2.04, q=3.50×10⁻⁸), and UV response up (NES=2.01, q=8.71×10⁻⁷). Negative enrichment included interferon‐γ response (NES=−2.34, q=1.10×10⁻¹¹), interferon‐α response (NES=−2.65, q=2.46×10⁻¹³), UV response down (NES=−1.65, q=1.92×10⁻³), bile‐acid metabolism (NES=−2.12, q=1.78×10⁻⁶), KRAS signaling DN (NES=−1.79, q=7.90×10⁻⁴), and peroxisome (NES=−1.69, q=2.84×10⁻³) (Figure 5).

For pulsatile flow versus static conditions, positively enriched pathways in pulsatile flow included MYC target V1 (NES=2.87, q=8.20×10⁻²⁴), MYC target V2 (NES=3.01, q=7.08×10⁻¹⁸), TGF‐β signaling (NES=2.89, q=1.21×10⁻¹⁵), unfolded protein response (NES=2.36, q=1.42×10⁻⁹), TNF‐α signaling via NF‐κB (NES=2.17, q=9.35×10⁻¹⁰), epithelial–mesenchymal transition (NES=2.18, q=7.23×10⁻¹⁰), mammalian target of rapamycin complex 1 signaling (NES=2.10, q=2.81×10⁻⁹), and cell cycle programs, including G2M checkpoint (NES=1.79, q=2.69×10⁻⁵) and mitotic spindle (NES=1.74, q=7.04×10⁻⁵). Negative enrichment included interferon‐α response (NES=−2.63, q=1.05×10⁻¹⁴), interferon‐γ response (NES=−2.24, q=2.84×10⁻¹⁰), bile‐acid metabolism (NES=−2.24, q=5.52×10⁻⁸), estrogen response late (NES=−1.71, q=1.29×10⁻⁴), fatty acid metabolism (NES=−1.51, q=8.76×10⁻³), estrogen response early (NES=−1.46, q=1.06×10⁻²), adipogenesis (NES=−1.44, q=1.77×10⁻²), inflammatory response (NES=−1.45, q=2.29×10⁻²), and apical junction (NES=−1.33, q=2.68×10⁻²) (Figure S6).

Flow Pattern Alters Transcription of Shear‐Responsive Genes

In a direct comparison of pulsatile versus continuous flow, 149 genes were upregulated and 235 genes were downregulated during pulsatile flow compared with continuous flow.

Positive enrichment in pulsatile flow included mitotic spindle (NES=2.91, q=2.09×10⁻²²), UV response down (NES=2.82, q=7.86×10⁻¹⁷), G2M checkpoint (NES=2.25, q=9.45×10⁻¹⁰), MYC target V2 (NES=2.56, q=3.91×10⁻⁸), E2F target (NES=1.83, q=1.29×10⁻⁵), TGF‐β signaling (NES=2.16, q=5.07×10⁻⁵), epithelial–mesenchymal transition (NES=1.65, q=4.43×10⁻⁴), apical junction (NES=1.47, q=1.64×10⁻²), and MYC target V1 (NES=1.45, q=1.49×10⁻²). Negative enrichment included estrogen response late (NES=−1.93, q=2.39×10⁻⁶), P53 pathway (NES=−1.85, q=2.78×10⁻⁵), UV response up (NES=−1.82, q=6.42×10⁻⁵), oxidative phosphorylation (NES=−1.67, q=3.55×10⁻⁴), and coagulation (NES=−1.51, q=1.64×10⁻²) (Figure S7).

A heat map of the top 50 differentially expressed genes comparing pulsatile flow with continuous flow reinforced these differences and recapitulated the principal component analysis structure, clearly showing the separation under the differing flow conditions (Figure S7).

DISCUSSION

We engineered a vessel‐mimetic, 3D‐printed macrofluidic system that reproduces continuous and physiological pulsatile waveforms, providing a controllable testbed to separate and identify how the presence and temporal structure of flow is associated with endothelial phenotype and transcriptional state in this model system. Static culture, continuous flow, and pulsatile flow generated distinct endothelial transcriptomes. Culture under static conditions was associated with lower eccentricity and reduced alignment, elevated adhesion molecule expression, and enrichment of interferon and inflammatory signaling relative to cells exposed to flow. Both continuous and pulsatile shear showed relative enrichment of stress‐adaptive and growth‐associated programs compared with static culture (including unfolded protein response, mammalian target of rapamycin complex 1 signaling, epithelial–mesenchymal transition, and MYC target gene sets). Pulsatile shear with the same mean load produced an intermediate morphologic phenotype. Compared with continuous flow, pulsatile flow was associated with lower eccentricity (0.91±0.03 versus 0.96±0.01; Bonferroni‐adjusted P=0.024), but no difference in absolute variation of orientation was detected (11.9°±3.1° versus 9.5°±3.5°; Bonferroni‐adjusted P=0.86). Collectively, these observations validate the macrofluidic model as a physiologically informative platform and highlight that ECs are exquisitely sensitive to the temporal structure of flow under matched mean shear conditions.

In this system, continuous and pulsatile profiles were closely matched for mean flow, Reynolds number, and time‐averaged WSS (oscillatory shear index≈0; no flow reversal). Nevertheless, waveform shape remained associated with distinct endothelial transcriptional states under constant mean WSS, whereas morphology‐based differences between the 2 flow modalities were feature specific, with a significant difference in eccentricity but not in absolute variation of orientation. This divergence despite matched means indicates sensitivity to features such as the rate of change of shear. The pulsatile flow conditions showed a markedly higher PI (PI pulse ≈0.25), and a Womersley number α≈6.6, consistent with physiologically relevant, large‐artery–like pulsatility. At the pathway level, comparisons to static culture showed that both continuous and pulsatile flow induced a shared shear‐responsive transcriptional signature, including enrichment of unfolded protein response, mammalian target of rapamycin complex 1 signaling, and epithelial–mesenchymal transition, alongside growth‐associated gene sets (eg, MYC targets). Notably, proliferative programs diverged between flow modalities: continuous flow showed negative enrichment of mitotic spindle and E2F targets relative to static, whereas pulsatile flow positively enriched cell cycle/checkpoint sets (including G2M checkpoint and mitotic spindle), consistent with a pulsatility‐linked shift toward cycling/checkpoint activity. Compared with static culture, both flow conditions showed reduced interferon programs (interferon‐α/γ response) and a modest reduction in the Hallmark “inflammatory response” set, whereas TNF‐α signaling via NF‐κB was positively enriched, suggesting a shift in immune signaling composition rather than uniform suppression.

In the direct pulsatile versus continuous comparison, pulsatile flow showed relative enrichment of cell cycle/checkpoint programs (mitotic spindle, G2M checkpoint, MYC target V2, and E2F targets), whereas continuous flow showed relative enrichment of oxidative phosphorylation along with higher p53 pathway enrichment. By contrast, TGF‐β signaling (as well as epithelial–mesenchymal transition and apical junction gene sets) was relatively enriched in pulsatile flow. Together, these results suggest that under matched mean shear, waveform shape is associated with differential emphasis on cell cycle/checkpoint versus oxidative‐metabolic signaling in HMEC‐1.

These differences align with established mechanobiology in which waveform features govern endothelial state, rather than mean load alone. In landmark experiments that reconstructed the human carotid artery bifurcation hemodynamics and replayed region‐specific waveforms in vitro, irregular oscillatory profiles activated NF‐κB, increased expression of interleukin‐8, and heightened cytokine‐inducible adhesion molecules, whereas steady laminar profiles upregulated KLF2, increased NO signaling, and stabilized junctions. ³⁸ , ³⁹ , ⁴⁰ In the human carotid sinus, high oscillatory shear index at low mean shear arises from cardiac‐cycle separation‐boundary motion, yielding proinflammatory mechanical inputs. ⁴¹ Importantly, our pulsatile condition remained laminar and unidirectional (oscillatory shear index=0) and therefore differs fundamentally from classic oscillatory/disturbed shear models that involve flow reversal. Pathologies dominated by disturbed shear, including the epicardial coronaries, carotid bifurcation/sinus, and peripheral conduit arteries, are classic settings in which waveform shape governs endothelial state and disease expression. In the present platform, waveform‐dependent signatures under matched mean shear are most plausibly attributable to pulsatility amplitude and temporal shear gradients rather than oscillatory shear per se. That said, the platform is readily configurable to impose higher‐shear pulsatile waveforms, including reversal‐containing profiles characteristic of the aorta, carotid bifurcation, and selected peripheral arterial beds. Given the low mean WSS applied (~1.5 dyn/cm²) and the original intent to model arteriovenous fistula–relevant hemodynamics, these data are best interpreted in the context of low‐shear, high‐flow conduit environments; extension to higher arterial shear regimens and additional geometries will be important in future work.

Prior work reproducing patient‐specific waveforms derived from magnetic resonance imaging and ultrasound‐informed computational fluid dynamics with computer‐controlled devices has focused on emphasizing accurate magnitude, frequency, and shape. ⁴² Our data extend this work by showing that even when mean WSS, Reynolds number, and time‐averaged metrics are held constant, waveform shape remains associated with differential pathway enrichment across proliferation, metabolism, junctional composition, and inflammatory tone, together with more modest, feature‐specific differences in morphology. This suggests that ECs are sensitive to subtleties in flow, such as the rate of change of shear stress over time. This observation captures the complex interplay between force magnitude, directional stability, rise and decay kinetics, and duty cycle. These waveform features are sensed and integrated by the endothelial glycocalyx, integrins, ion channels, junctional complexes, and Yes‐associated protein (YAP)–transcriptional co‐activator with PDZ‐binding motif (TAZ) mechanotransducers, which together regulate NO bioavailability, oxidation‐reduction balance, and transcriptional control. ⁵ , ¹⁸ , ²⁰ , ²¹ The findings described in this study demonstrate that even in a controlled in vitro system, small changes to a single factor can alter gene expression and phenotype in ways that can alter EC function. These results highlight the importance of explicit reporting of waveform descriptors, including mean wall shear stress, oscillatory shear index, pulsatility amplitude and frequency, and rise and decay times, to ensure reproducibility, experimental fidelity, and external validity in studies of flow–function relationships.

Endothelial vascular bed of origin may also modulate the thresholds at which flow alters phenotype. Because this study used a single EC line, our data do not test vascular‐bed–specific decoding directly; however, extensive prior work supports endothelial origin as a major determinant of shear‐responsive programs. Although core shear‐responsive factors, such as KLF2, KLF4, and NO synthase 3, are induced in both human umbilical vein ECs and human aortic ECs, pathway outputs spanning cell cycle, oxidative stress, TGF‐β signaling, angiogenesis, and sentinel genes, including VCAM1, ICAM1, SELE, NOX4, and SOD2, display bed‐specific changes or opposite trajectories, indicating lineage‐encoded thresholds. ¹⁰ Maurya et al showed that human aortic ECs exposed to shear stress mount a faster and broader early response with more differentially expressed genes at 1 to 4 hours and show distinct regulation of KLF4, E2F1, and inflammatory mediators relative to human umbilical vein ECs. ¹⁰ In another study, exposure of human umbilical vein ECs, human pulmonary artery ECs, and human microvascular ECs to matched laminar shear increased canonical mechanotransduction in all cell types. ⁴² Despite this, only 44.2% of the differentially expressed genes were shared, and human microvascular ECs remained transcriptionally distinct by principal component analysis, indicating lineage‐gated decoding of shear signals. ⁴³ Accordingly, endothelial subtype and waveform should be treated as codeterminants of state rather than interchangeable experimental details.

In the present study, we used the HMEC‐1 EC line, acknowledging that immortalized microvascular ECs can exhibit distinct baseline metabolic and proliferative set points compared with primary arterial ECs. Accordingly, conclusions are limited to HMEC‐1 under the specific low‐shear, nonreversing laminar conditions tested here. In addition, mechanistic interpretation in this article is based on bulk RNA‐sequencing and Hallmark pathway enrichment; these results describe transcriptional signatures consistent with pathway engagement rather than confirming pathway activation at the protein or functional level.

The hemodynamic regimen was intentionally low (time‐averaged WSS~1.5 dyn/cm²), consistent with the original engineering intent to support future studies of arteriovenous fistula–relevant hemodynamics. Although the present work validates waveform control and biological responsiveness in a vessel‐scale 3D construct, future studies should extend experiments to primary human arterial and venous ECs (including patient‐derived cells) and to additional shear regimens (including higher arterial laminar shear) and geometries (eg, carotid‐type configurations) to test generalizability across vascular beds and flow contexts.

With further refinement, vessel‐mimetic 3D macrofluidic platforms that deliver validated flow waveforms will provide an opportunity to improve mechanistic fidelity, reproducibility, and translational relevance. By matching device geometry and material compliance to patient‐specific morphologies and by enforcing verified waveforms, such systems can better preserve hemodynamic conditions required for endothelial polarization, glycocalyx integrity, junctional architecture, and NO signaling than conventional 2‐dimensional culture. Translational relevance would further increase by coupling these platforms to computational fluid dynamics derived from clinical imaging, permitting imposition of patient‐level waveforms, such as those encountered in arteriovenous fistulas or prestenotic/poststenotic segments, and by aligning readouts with clinical end points (vasodilator reserve, permeability, leukocyte adhesion, thrombogenicity, and antirestenotic responses). The approach is inherently adaptable to vessel class and patient morphology. Channel geometry, compliance, branch angle, curvature, and extracellular matrix can be engineered to model arterial versus venous beds, high‐flow states, or disturbed‐flow niches. Integration of the platform described here with patient‐derived cells (primary ECs/smooth muscle cells or induced pluripotent stem cell derivatives) would enable a pipeline in which devices configured from an individual's computed tomography angiography or duplex ultrasound could be used to test medical devices, drug‐elution strategies, or flow‐reduction procedures under patient‐specific hemodynamic conditions. A practical workflow for this would comprise imaging and segmentation, computational fluid dynamics, device fabrication, waveform calibration with inline pressure and flow sensors, patient‐derived endothelial seeding, and exposure to the corresponding wall shear profiles, with orthogonal transcriptomic and functional readouts to verify mechanistic targeting. Although a fully integrated pipeline of patient‐specific geometry, patient‐derived cells, and closed‐loop waveform control remains to be realized, the present study establishes a reproducible framework for endothelial modeling.

Sources of Funding

This research received support from the University of New South Wales Cardiac, Vascular, and Metabolic Medicine Collaborative Grant Scheme, the University of New South Wales 3Rs Research Grant, and the Prince of Wales Hospital Foundation Annual Research Grant. N.S. was additionally supported by an Australian Government Research Training Program Scholarship, a Royal Australasian College of Physicians Jacquot Research Entry Scholarship, and a Baxter Family Postgraduate Scholarship.

Disclosures

None.

Supporting information

Figures S1–S7