Article Text
Abstract
Objectives To study the correlation between wall shear stress and endothelial cell expression in a patient-specific, three-dimensional (3D)-printed model of a cerebral aneurysm.
Materials and methods A 3D-printed model of a cerebral aneurysm was created from a patient’s angiogram. After populating the model with human endothelial cells, it was exposed to media under flow for 24 hours. Endothelial cell morphology was characterized in five regions of the 3D-printed model using confocal microscopy. Endothelial cells were then harvested from distinct regions of the 3D-printed model for mRNA collection and gene analysis via quantitative polymerase chain reaction (qPCR.) Cell morphology and mRNA measurement were correlated with computational fluid dynamics simulations.
Results The model was successfully populated with endothelial cells, which survived under flow for 24 hours. Endothelial morphology showed alignment with flow in the proximal and distal parent vessel and aneurysm neck, but disorganization in the aneurysm dome. Genetic analysis of endothelial mRNA expression in the aneurysm dome and distal parent vessel was compared with the proximal parent vessels. ADAMTS-1 and NOS3 were downregulated in the aneurysm dome, while GJA4 was upregulated in the distal parent vessel. Disorganized morphology and decreased ADAMTS-1 and NOS3 expression correlated with areas of substantially lower wall shear stress and wall shear stress gradient in computational fluid dynamics simulations.
Conclusions Creating 3D-printed models of patient-specific cerebral aneurysms populated with human endothelial cells is feasible. Analysis of these cells after exposure to flow demonstrates differences in both cell morphology and genetic expression, which correlate with areas of differential hemodynamic stress.
- aneurysm
- blood flow
- genetic
Statistics from Altmetric.com
Introduction
Work attempting to understand aneurysm formation, growth, rupture, and effects of treatment has hypothesized a connection between mechanical stresses on the aneurysmal wall, and endothelial dysfunction. These studies have frequently employed computational fluid dynamics (CFD) techniques to simulate the flow of blood inside intracranial arteries and aneurysms, calculating metrics from the hemodynamic environment of aneurysms before and after treatment.1 These CFD simulations have been successful in quantifying important hemodynamic factors such as wall shear stress (WSS) and WSS gradient (WSSG), both of which have been implicated in aneurysmal growth, rupture, and treatment outcome.2 3 Because endothelial cells contain mechanosensors that respond to changes in such hemodynamic stresses,4 these metrics have been studied in relation to their effects on endothelial dysfunction, abnormal vascular remodeling, and inflammation, which are fundamental components of cerebral aneurysm pathophysiology.3 5 For instance, VCAM-1 is an adhesion molecule upregulated in aneurysmal regions of low WSS,6 7 while the matrix metalloproteinase ADAMTS-1 is upregulated in areas of elevated WSS and WSSG.8 9
A major drawback of existing CFD studies in clinical applications (such as risk of aneurysm growth or rupture, or outcome of aneurysm treatment) is that they do not directly measure the pathological effects of such hemodynamic variables on biological tissue, since CFD uses computer models to evaluate blood flow and shear stress. These models influence, through poorly understood mechanisms, the vascular biology controlling aneurysmal evolution. Thus, the connection between aneurysmal hemodynamics observed in CFD studies and their pathobiological impact is inferred from a variety of sources, such as human studies of non-aneurysmal domains, animal studies, and highly simplified in vitro experiments, in which the response of endothelial cells to blood flow stresses is measured.3 6 8 10–14 The design of these experiments, such as studying the effects of laminar or disturbed flow on endothelial cells in a straight tube or idealized aneurysm shape at small scale,15 or ligating the carotid arteries of animals and measuring subsequent vascular endothelial remodeling,14 do not reflect the patient-specific, highly complex anatomy of human cerebral aneurysms and thus cannot provide sufficient quantification for predictive prognosis.
We have developed a patient-specific cerebral aneurysm model populated with human endothelial cells, and studied the endothelial cell response to hemodynamic stresses. Our method suggested that areas of pathological hemodynamic stresses in CFD simulations can be correlated with subsequent endothelial transcription and expression of important vascular factors in 3D-printed models, which might be used to quantify the relationship between hemodynamics and vascular biology.
Materials and methods
Creation of a 3D-printed model
A 3D rotational angiogram from a patient enrolled in a clinical study of cerebral aneurysm hemodynamics was used for this pilot study, as approved by the institutional review board (figure 1A). First, a ‘positive’ aneurysm model and the surrounding vessels were 3D printed in a stereolithography printer (Form 2, Formlabs, Somerville, Massachusetts, USA) using photoreactive resins (High Temp resin, Formlabs, printed at 25 µm layer height, figure 1B). The aneurysm model included the aneurysmal sac, and proximal and distal parent vessels, with precise anatomy and scale matching the patient’s 3D rotational angiogram. The ‘positive’ 3D-printed model was smoothed by sanding and polishing until transparent (Novus Plastic Polish, St Paul, Minnesota, USA). The model was embedded in polydimethylsiloxane (PDMS) using a customized acrylic housing to achieve thin walls for optimal imaging and endothelial cell sampling. After crosslinking the PDMS, the positive aneurysm model was removed from the PDMS structure to generate a negative aneurysm lumen (the final 3D-printed aneurysm model, figure 1C) for flow measurement and endothelialization. Any defects created during this process were corrected by resealing with PDMS to ensure a high burst pressure, and to prevent leakage under flow. This process was repeated to create identical models of the same cerebral aneurysm.
All PDMS models were sterilized through a high-pressure and high-temperature autoclave cycle, followed by plasma treatment in a Plasma Prep II Etcher (SPI Supplies, West Chester, Pennsylvania, USA) for 1 min to make the PDMS surface hydrophilic. For gene expression experiments a 5 µg/mL fibronectin solution (MilliporeSigma, Burlington, Massachusetts, USA) was perfused immediately after plasma treatment and incubated for 1 hour at 37°C to enhance cell adhesion during seeding. For morphologic characterization a single PDMS model was plasma treated for 1 min and incubated with 0.2% gelatin for 20 min at 37°C. Gene expression models were seeded with a suspension of human carotid artery endothelial cells (HCtAEC; Cell Applications, San Diego, California, USA) at 1 million cells/mL and incubated at 37°C for 15 min. Models were inverted once while seeding to ensure even cell coverage, and fresh medium was used to remove cells that had not adhered. The morphology model was seeded using the same protocol, but with human umbilical vein endothelial cells (HUVEC; Lonza, Morristown, New Jersey, USA). After 12 hours of attachment, each endothelialized model was connected to a peristaltic pump (Cole-Parmer, Vernon Hills, Illinois, USA) to drive the flow of the medium at a rate of 150 mL/min for 24 hours with a standardized pulsatile waveform reaching a peak pressure of ~40 mm Hg.16 The flow medium was endothelial cell growth medium supplemented with 3.5% dextran (Sigma-Aldrich, St Louis, Missouri, USA) to mimic the dynamic viscosity of blood.17
CFD simulation
The same 3D rotational angiogram, used for the flow phantom model creation above, was used for CFD simulation, which has been previously described.18 Briefly, image segmentation was performed using the Vascular Modeling Toolkit (www.vmtk.org), creating a computerized 3D reconstruction of the vascular lumen and aneurysm. Then, tetrahedral meshes were generated by StarCCM + software (CD-adapco, Melville, New York, USA). CFD simulations were executed using ANSYS Fluent (ANSYS Inc, Canonsburg, Pennsylvania, USA), a finite-volume flow solver with ample reported use for the simulation of intracranial aneurysm hemodynamics. Very high spatial and temporal resolution was used to capture the hemodynamics in the intra-aneurysmal and peri-aneurysmal vasculature—CFD mesh sizes ranged from 50 to 100 μm, amounting to nearly 1 million elements, and a time step size of 0.0005 s was used for the simulations. We set the convergence criteria for the continuity and the momentum equations (Navier-Stokes equations) to 10−6. Each simulation was run for a minimum of three cardiac cycles, and the first two cycles were discarded to ensure that the simulations were independent of the initial conditions. Velocity streamlines were created and hemodynamic variables (WSS and WSSG) were calculated over the entire third cardiac cycle to obtain the spatio-temporal variation of hemodynamic stimuli for the aneurysm and surrounding vasculature. Consistent with perfusion in the 3D-printed model, velocity inlet conditions were applied to both vertebral arteries.
Endothelial cell morphometric analysis
To qualitatively characterize endothelial cell morphology, one HUVEC seeded model (n=1) was fixed after 24 hours and washed with phosphate-buffered saline. Immunofluorescence staining was performed for junctional protein CD31. Images of this model were obtained through wide-field and fluorescent confocal microscopy to obtain large-scale z-stack images for five distinct aneurysm regions: the proximal parent vessel, the proximal aneurysm neck, the aneurysm dome, the distal aneurysm neck and the distal parent vessel (figure 2). These characteristics were qualitatively compared with hemodynamic variables WSS and WSSG in CFD simulations of the same aneurysm.
Endothelial cell genetic analysis
In HCtAEC endothelialized 3D-printed aneurysm models (n=3), three zones were defined for endothelial cell analysis: (1) the proximal parent vessel, (2) the aneurysm dome, and (3) the distal parent vessel. Each of the zones was mechanically separated at the end of each experiment and independently perfused with RLT lysis buffer (Qiagen, Hilden, Germany), and the lysate was collected. Total RNA from each zone was purified using RNAeasy Mini Kit (Qiagen) and RT-PCR was performed using the real-time PCR system (Applied Biosystems, Foster City, California, USA) with Fast SYBR Green Master Mix (Applied Biosystems). The abundance of several key vascular factors (VCAM-1, ADAMTS-1, MMP-9, GJA4, and NOS3) was determined relative to an internal control using glyceraldehyde-3-phosphate dehydrogenase RNA. These factors were selected based on previous studies relating their expression to either hemodynamic stress or aneurysm pathophysiology.7–10 14 15 19–22 The three regions were compared with hemodynamics metrics from CFD simulations of the same aneurysmal anatomy to assess the effect of intra-aneurysmal WSS and WSSG on endothelial expression of each of the key vascular factors.
Statistical analysis
Cell morphology and qPCR data were evaluated using R and Minitab 18 (State College, Pennsylvania, USA). Data were analyzed in a repeated measures analysis of variance design and pairwise comparisons were performed using Tukey’s method. P values <0.05 were considered statistically significant.
Results
All aneurysmal models were successfully populated with human endothelial cells, which survived flow for at least 24 hours and formed a robust monolayer in the majority of the lumen area (figure 2). Endothelial cells showed alignment to flow direction in the proximal and distal parent vessel, and in the proximal and distal aneurysm neck, but not in the aneurysm dome, where flow is not unidirectional. This difference in endothelial morphology also correlated with differences in hemodynamics between the aneurysm dome and the parent vessel: velocity, WSS, and WSSG were substantially lower in the aneurysm dome in those regions in which endothelial alignment was not consistent (figure 3).
Endothelial cell harvest for genetic expression analysis yielded consistent total RNA levels between models in the proximal parent vessel (1.90±0.08 µg), aneurysm dome (1.07±0.07 µg), and distal parent vessel (1.54 ±. 01 µg) regions, sufficient for large-scale screening and qPCR studies of over 100 genes. Genetic analysis of five key vascular factors (ADAMTS-1, GJA4, MMP-9, NOS3, and VCAM-1) is shown in figure 4. Compared with the proximal parent vessel and the aneurysm dome, GJA4 was significantly lower in the distal parent vessel (p<0. 05). The expression of ADAMTS-1 and NOS3 suggested a lower concentration in the aneurysm dome when compared with both parent vessel regions in all three model runs, but this change was not statistically significant. Neither MMP-9 nor VCAM-1 showed a consistent trend of expression in the three vessel regions. Decreased levels of ADAMTS-1 and NOS3 correlated with areas of substantially low velocity, WSS, and WSSG in the aneurysm dome in CFD simulations (figure 3).
Discussion
We have demonstrated that creating and maintaining a monolayer of living human endothelial cells in a patient-specific, 3D-printed model of a cerebral aneurysm is feasible, and that exposure to fluid flow results in changes in both endothelial morphology and expression of key vascular factors. Both endothelial morphology and genetic expression can be qualitatively associated with hemodynamic stress metrics from CFD simulations, with differential expression of ADAMTS-1 and NOS3 related to predominantly low WSS in the aneurysm dome, and reduced GJA4 in the distal parent vessel. This study, while exploratory, provides proof of the concept that specific areas of differential hemodynamic stress characterized by CFD simulations of a cerebral aneurysm and the surrounding vasculature can be directly related to endothelial responses in an in vitro model of the same aneurysm, and that this relationship could subsequently be quantified.
Our work seeks to go beyond existing studies aimed at understanding the pathobiology of cerebral aneurysm endothelial cells. Animal models of the pathobiological effects of hemodynamic stress have provided a solid background in identifying key factors of endothelial dysfunction driving aneurysm formation, growth, and rupture as related to hemodynamic stresses, such as ADAMTS-1, VCAM-1, MCP-1, and PDGF-B.7 8 10 19 23 However, animal models of cerebral aneurysms require artificial aneurysm creation, the anatomic result of which is unlike patient-specific human cerebral aneurysm morphology. This hinders the animal model approach, as it is cannot relate specific patterns of hemodynamic stresses as they appear in clinically relevant CFD simulations with disease-related human endothelial expression.24
Direct study of human vascular endothelial responses to pathological levels of hemodynamic stress in cerebral aneurysms, on the other hand, has not been extensively or regularly performed in parallel with CFD studies of aneurysm hemodynamics, since in vivo studies of cerebral aneurysms observed during surgery have examined only gross external features, such as wall thinning25 or atherosclerosis,26 27 and related them to areas of low WSS on corresponding CFD simulations of patient anatomy.
Ex vivo study of aneurysm tissue, while mechanistically possible, requires surgical excision of the aneurysm sac, with several potential drawbacks. First, the majority of cerebral aneurysms are preferentially treated using minimally invasive endovascular devices rather than surgery, reducing the number of potential aneurysms that are surgically harvested and studied ex vivo.28 More importantly, because the aneurysm sac must be removed from the parent vessel in an ex vivo study, its precise geometry is difficult to re-create in a laboratory setting, and direct correlation of aneurysm tissue with areas of hemodynamic stress defined by CFD simulations is difficult, impractical, and carries significant uncertainty. Existing ex vivo studies of human aneurysm tissue have thus primarily focused on structural phenomena, such as inflammatory cell infiltrate and differential wall thicknesses.29 Only a few studies30 31 have directly related CFD-identified areas of hemodynamic stresses to structural changes in human cerebral aneurysms harvested ex vivo. They found that wall stiffness, inflammation, and abnormal collagen organization were generally associated with regions of elevated WSS, while mural thrombus and atherosclerosis were associated with regions of low WSS. These observations are consistent with the downstream effects of key aneurysm-related vascular responses, though specific quantification of the endothelial transcriptional profile was not performed.
Thus far, only a few studies have attempted to relate endothelial response to hemodynamic stress in a rigorous manner. One study used a simple, idealized, and non-patient aneurysm geometry of a sidewall aneurysm to demonstrate upregulation of VCAM-1 in areas of low WSS in the aneurysm dome.15 This study is an important proof of concept and demonstrates the important link between low WSS and VCAM-1 expression, though the geometric scale (0.5 mm parent vessel and 0.5 mm aneurysm dome) was much smaller than that of actual arteries of the circle of Willis. Another study directly observed the endothelial morphological organization, but not the gene expression response, to hemodynamic stresses by populating a 3D-printed model of a human cerebral aneurysm with endothelial cells.13 Using a bovine endothelial cell line, their work showed alignment of endothelial cells with the direction of flow within the parent vessel, but did not have a well-defined elongation shape or direction in areas of complex flow within the aneurysm sac. While demonstrating that 3D-printed models can be populated with endothelial cells, this work only considered a single aneurysm under steady (non-physiologic and non-pulsatile) flow, which reduces its clinical relevance.2 In addition, their study had no molecular insights for the endothelial cell response in a human cerebral aneurysm. An abstract by the same group describes such analysis taking place, but the results of this analysis have not been peer-reviewed or published.32
Our study takes additional steps towards a more realistic simulation of the effect of aneurysm hemodynamics on endothelial cells. Our method builds anatomically accurate, 3D-printed, patient-specific intracranial aneurysm flow phantoms, cultures human endothelial cells in them under pulsatile flow, and characterizes the mRNA expression of these endothelial cells in different regions of the aneurysm, in addition to cell morphology changes demonstrated in previous studies. More importantly, we are able to make direct correlations with CFD simulations of hemodynamics of the same aneurysm anatomy. The endothelial cell morphology in our model had similar trends to those previously reported, including alignment in areas of organized flow, and lack of defined shape or direction in areas of complex, non-unidirectional flow.11 13 Importantly, we found that ADAMTS-1 and NOS3 levels were reduced (figure 4) in the endothelial cells harvested from the aneurysmal dome, compared with the proximal parent vessel, while GJA4 was reduced in the distal parent vessel. The CFD simulations of the same aneurysm found predominantly very low WSS and WSSG throughout the aneurysm dome (figure 1C). Thus, our work suggests a relationship between ADAMTS-1, NOS3, and aneurysmal endothelial mechanotransduction (including downregulation in regions of low WSS),8 9 22 while reduced GJA4 is a marker for oscillatory flow in the distal parent vessel.33
This exploratory, pilot study has several limitations. First, flow through the 3D-printed model, while pulsatile, did not use a patient-specific physiological waveform. Future studies will incorporate patient-specific pulsatile flow into such experiments by introducing a programmable pump in the incubator.
Second, our study used human umbilical vein endothelial cells and human carotid endothelial cells and did not include smooth muscle cells. Future studies will incorporate vascular smooth muscle cells, which play an important role in cerebrovascular biology.34
Third, we performed genetic analysis on endothelial cells after only 24 hours under flow. This cut-off point was used because vascular biology factors related to hemodynamic stress have demonstrated maximal change in response to such flow before 24 hours,35 and cell detachment can become common at longer flow exposures (data not shown).
Fourth, the aneurysm’s anatomy, while a precise replica of an actual patient’s aneurysm, does not contain the degree of complex geometry typically seen in many aneurysms of the anterior circulation. In addition, the photoreactive resin used to create this model is difficult to dissolve, precluding its use in more complex geometries, and these models require manual smoothing to ensure that no microscopic ridges have developed. We have recently developed a more robust technique using wax,36 which will permit more geometrically varied patient-specific 3D models, permitting smoothing of the wax cast and resulting in dimensional accuracy of ~100 μm. More complex aneurysm geometries may also require additional plasma treatment optimization to ensure complete hydrophilization of the luminal surface. The diffusion constant and very low etch rate of oxygen plasma ions on PDMS suggest that this strategy can be applied to complex aneurysms without compromising model integrity.37 38
Finally, we found no consistently significant change in gene expression in this particular patient-specific aneurysm for most of the genes studied, and no quantitative morphological changes. A larger cohort of 3D-printed, patient-specific aneurysm models using this method will enable experiments over a wide variety of cerebral aneurysms with differential areas of high and low WSS and WSSG, necessary to relate specific areas of hemodynamic stress with endothelial expression across a wide range of patterns and maximum/minimum values in a statistically significant manner.
Conclusion
Creating 3D-printed cerebral aneurysms populated with living human endothelial cells is feasible, and morphologic and genetic analysis of the cells harvested after 24 hours of exposure to physiological pulsatile flow can quantify the endothelial response to hemodynamic stress.
References
Footnotes
Contributors MRL, CMK, YZ, AAl, and LJK contributed to the planning of this research. MRL, CM, AAb, CMK, VKC, YZ, AAl, and SL contributed to the conduct of the research. MRL, CM, AAb, CMK, SL, VKC, YZ, AA, and LJK contributed to the reporting of the research results.
Funding Thiswork was supported by the National Institutes of Health/National Institute of Neurological Disorders and Stroke grants R01NS088072 and R01NS105692.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement No unpublished data available.
Correction notice Since this article was first published online, figure 4 has been replaced.
Patient consent for publication Not required.