A total of 10 patients undergoing proximal aortic surgery (either aortic root, ascending aorta, proximal arch replacements, or a combination of these) were recruited into this cohort study. The main exclusion criteria were: connective tissue disease (i.e. Marfans, Ehler-Danlos, Loeys-Dietz), bicuspid aortic valve and redo-aortic surgery. Patient characteristics were ascertained from clinic letters and pre-operative echocardiography. Recruited patients provided informed consent for participation in the study, which involved the additional MR imaging and provision of tissue sample from surgery.

Patients were scheduled to undergo cardiac gated magnetic resonance imaging (MRI) scanning at a time point prior to surgery (range 1–90 days) using a 3T MRI scanner (Siemens Healthcare, Erlangen, Germany). The addition of a time-resolved 3-dimensional sequence (4D-flow) visualized and measured temporal changes and flow-patterns throughout the whole volume of the thoracic aorta.

Patient-matched aneurysm specimens were obtained en-bloc and acquired immediately after surgical excision in the operating theatre. Rectangular or dog-bone shaped subsections aligned in either the longitudinal or circumferential direction were punched out at specific locations (Figure 1A) and assessed for tensile mechanical properties, peel strength and strain inhomogeneities (Olchanyi et al., 2020). In particular, for each patient, the aneurysm region was unfolded at the outer wall and subdivided into up to six (depending on the size of the collected sample) circumferential anatomical regions to the right (R1, R2, R3) and left (L1, L2, L3) of the midline (Figure 1B). These were further divided into upper and lower regions, giving 6 to 12 segments in total per patient (Figure 1B). These segments were matched to the 6–12 segmental regions undergoing computational WSS mapping (Section 2.4; Figure 1C), using the anterior midline (widest region of the aneurysm region in the vertical plane) as the common frame of reference. Whole-aneurysm tissue specimens underwent thickness mapping (∼0.01 µm resolution using a bench top device - Litematic VL-50-B Mitutoyo Ltd.).

Sharp stencils were used to create 20 mm × 5 mm dogbone samples (for uniaxial testing). The dogbone shaped tissue provided larger shoulders for easier gripping, whereas the narrower gauge section in the center provided a smaller cross-sectional area for a focused region of deformation and failure. The samples for peel testing required a different shape: the width of the peeling sample needed to be kept constant throughout the peeling arc. As such rectangular stencils were used (5 mm × 20 mm) to obtain eight further samples from both halves of the aneurysm specimen (both circumferential and longitudinal) –these would be used for peel testing.

Uniaxial tensile tests were performed on all dog-bone shaped subsections using a Test Resources R-Series Controller with a 44 N load cell. All tests were conducted in an environmental chamber containing phosphate buffered saline (PBS), maintained at 37°C. Tensile force-elongation data were used to provide estimations of region specific ultimate tensile strength (UTS) and maximum tangential stiffness (MTS) as described by previous groups (Vorp et al., 2003). Peel testing allowed estimation of peel force (F_peel) and dissection energy function (DEF) for rectangular subsections.

Image-based CFD modelling was used to evaluate pre-operative aneurysmal WSS distribution in a patient-specific and region-specific fashion. To this aim, the 3D thoracic aortic geometry, including the arch branches, of each patient was reconstructed from MRI data (bright blood). Reconstructed geometries (Figures 1C, D) started upstream of the aneurysm, in the aortic root between mid-sinus and STJ levels, and ended distally, in the descending thoracic aorta at the level of the pulmonary artery.

Patient-specific geometries were discretized using unstructured meshes (Supplementary Figure S1) with a tetrahedral core and 10 prismatic layers at the walls. Local mesh refinements were prescribed at the aneurysm wall and arch branches. Mesh refinement in the region of interest (i.e. aneurysm wall) was guided by flow features observed from 4D flow MRI data (i.e. a finer mesh was designed where higher velocity gradients were observed). Sensitivity analyses were conducted to ensure mesh-independent results. Final meshes consisted of 5.5–16.6 million elements (Pirola et al., 2019).

For each patient, 3D subject-specific velocity profiles of the aortic valve were incorporated at the computational model inlet from 4D flow MR images (Armour et al., 2021). This therefore took into account the patient-to-patient variation in the ascending aorta hemodynamics which arise due to differences in left ventricular outflow tract and aortic valve anatomy (Figure 1D). Model outlets, located distal to the aneurysm, consisted of the descending aorta outlet and the arch branches. 3-element Windkessel model was applied at all outlets; model parameters were tuned using patient-specific central mean aortic pressure and blood flow rates evaluated using 4D flow MRI data (Pirola et al., 2017; 2019). Briefly, the total resistance (R_T) of the 3-element Windkessel model was calculated as R_T = P_m/Q_m (Les et al., 2010). P_m is the mean central pressure of the patient, which was measured ∼30 min before the MR scan using a BP Plus device (BP Plus, Uscom, Australia). Q_m is the mean flow through the outlet. This was evaluated using 4D flow MR images. The proximal resistance was evaluated as R₁ = ρc/A (Xiao et al., 2014, where ρ is the blood density, c is the pulse wave speed and A is the outlet cross-sectional area. The pulse wave speed was evaluated as c = 13.3/(2r)^0.3 Reymond et al. (2009)), where r is the outlet radius. The distal resistance was evaluated as R₂ = R_T-R₁ (LaDisa et al., 2011). Total vasculature compliance was calculated as C = τ/R_T, where τ is the time-constant of the exponential diastolic pressure-fall (Xiao et al. (2014)). This was assumed equal to 1.79 and 1.92 s for normotensive and hypertensive subjects, respectively (Simon et al., 1979).

Simulations were run in ANSYS CFX (v 15.0). The aortic wall was assumed to be rigid with a no-slip condition. Blood was modelled as an incompressible Newtonian fluid, with a density of 1,060 kg m⁻³ and viscosity of 4·10⁻³ Pa s. High-resolution advection scheme and second-order Backward Euler scheme were employed for spatial and temporal discretization of the Navier-Stokes equations, respectively. A fixed timestep of 10⁻³ s was used, and the maximum RMS residual was set to 10⁻⁵ as a convergence criterion. The shear stress transport transitional (SST-Tran) model was used to account for the transitional nature of the aortic flow (Tan et al., 2009), with a 1% turbulence level prescribed as the inlet boundary condition. The patient-specific central diastolic pressure was used as the initial condition. Simulations were run for the number of cycles necessary to achieve periodicity. The period of the cardiac cycle of each patient was recorded during 4D flow MR acquisition and stored in the image header. Periodicity was considered as achieved when differences in pulse pressure and pressure maxima between two consecutive heartbeats were less than 3 and 1%, respectively, at each model outlet. Results from the last cycle only were analyzed. Before proceeding with result analysis, for each patient, predicted blood flow features were checked against 4D flow MR acquired features.

Computational result postprocessing workflow was designed to coincide with tissue processing for mechanical and histological testing: for each patient, the aneurysm region was unfolded at the outer wall and subdivided into segments corresponding to the aortic aneurysm tissue segments used to characterize the material properties (Figures 1B,C). Results were analyzed in ANSYS EnSight.

The primary derived measurement from CFD analysis was the magnitude of the wall shear stress (WSS) vector. For each anatomical region, several WSS-derived parameters were evaluated. These are summarized in Table 1. Briefly, WSS values are reported as maxima in time (WSS_max, WSS_mean) and time-averaged (TAWSS_max, TAWSS_mean, WSS^Tmean _Max). Mean and max subscripts refer to spatial mean and maxima, respectively. Tmean superscript refers to temporal mean. The spatial mean value was calculated over each segment (R1, R2, R3, L1, L2, L3, upper and lower).

Full circumferential rings of tissue were obtained from the inferior border of the ATAA specimen (Figure 2A). Whole slide imaging was performed using a high-resolution digital optical system (Hamamatsu TM) and uploaded onto the digital processing software QuPath (https://qupath.github.io). Cross-sectional images of each ATAA ring were divided into six circumferential regions (Figure 2B), matching the right-to-left segments used in the CFD and mechanical analyses (Figures 1B, C). Using a pre-defined workflow (Figure 2C), microstructural density calculation of medial structural proteins was conducted, namely of elastin (from slides stained with Elastin Van Gieson) and collagen (slides stained with picrosirius red). Using the H&E stained slides (Figure 2D), the thresholding function was used to highlight the stained cells of the medial layer. Setting the range of particle size from 20 to 100 µm the total cell count was extracted and divided by the total area to obtain the density.

Amongst the 10 recruited patients, a total of 371 ATAA anatomical samples had matched CFD, material properties and microstructural features, and were thus included in the analysis, being treated as separate data points. Of these samples, thickness was measured in 102 samples, 63 samples (29 longitudinal, 34 circumferential) were used for tensile testing, 63 samples (30 longitudinal, 33 circumferential) for peel testing, 143 for histological analyses (60, 35 and 48 for elastin, collagen and SMCs, respectively). All haemodynamic parameters were acquired as continuous variables, as were the histological and mechanical data. Where relevant, data were reported as means and standard deviations, being averaged across the whole patient cohort. All statistical analysis were conducted using STATA 13.0 (Stata Corp. College Station, TX, United States).

Univariable linear regression analysis was used to explore the effect of WSS on acquired material parameters (including mechanical and histological). These models tested the hypothesis of a flow-mediated degenerative process in the aortic wall giving rise to altered material properties. Results of regression analysis were reported as standardized beta-coefficients with 95% confidence intervals. The significance level was set at α = 0.05. Values of longitudinal F_peel, longitudinal DEF, UTS and MTS were positively skewed; values of elastin abundance were negatively skewed. Therefore, logarithmic transformation with skewness correction was applied in STATA to these datasets.

Scatter plots and trendlines have been generated using Microsoft Excel, using log₁₀ transformation for longitudinal DEF and UTS, and log₁₀ (90-x) transformation for elastin abundance.

The patient-specific TAWSS distribution maps for all 10 patients are shown in Figure 4. The R1 segment (outer right wall) was the region yielding the highest values of WSS (Table 3), including temporal maximum WSS values (WSS_max and WSS_mean) as well as time-averaged values (TAWSS_max and TAWSS_mean). The highest value for WSS_max (24.98 ± 7.79 Pa) in the R1 region contrasted to the lowest value yielded from the inner curve (10.18 ± 4.14 Pa). When measuring the (peak in time) WSS averaged over space (WSS_mean), the R1 region still yielded the highest value (11.68 ± 6.42 Pa). For TAWSS_max, R1, also the region of highest WSS values, was 4.85Pa ± 2.07, compared to the inner curve (2.67 ± 0.81 Pa). This indicated that as well as the asymmetric peak flow patterns reached in systole acting on the potentially fragile intima in ATAA, the sustained stress over the cardiac cycle, as displayed by the TAWSS, was also asymmetrical and affecting the outer curve more intensely.

In order to test the hypothesis of flow mediated wall degeneration, WSS was compared with material properties of the aortic wall. Statistically significant linear regression analysis results are summarized in Table 4. Scatter plots and trendlines showing the relationship between WSS and aortic wall material properties are shown in Figure 5.

Significant results of multilevel mixed-effects linear models are reported in Tables 5, 6.

The WSS distribution in the ascending aorta is dominated by the curvature of the aortic arch, which forces blood emanating from the heart to change its direction. Fluid at the centre of the vessel is more difficult to displace as it is travelling at a higher velocity compared to that closer to the wall, so it is displaced to a greater degree (Caballero et al., 2013). Thus, blood is skewed towards the outer curvature of the bend.

Our findings agree with several previous studies assessing flow in ATAA, which also noted the highest WSS on the outer curve (Bieging et al., 2011; Bürk et al., 2012; Van Ooij et al., 2015).

Owing to the low temporo-spatial resolution of cross-sectional imaging and difficulty in tracking the motion of the aortic wall (<2.5 mm thickness), estimations of stress distribution within the aneurysm wall are reliant on important assumptions of aortic wall thickness (commonly assumed to be constant) and material properties (commonly not patient- or even disease-specific) (Martin et al., 2015; Gomez et al., 2020). Our results have identified a potential link between areas of high WSS and aortic wall thinning. Reduced wall thickness in areas of high WSS has been previously shown in femoral artery bifurcation by Kornet et al. (1999), who explained this finding with the link between low WSS, increased influx of substances into the aortic wall (through an increase in blood residence time) and increased release of vasoactive molecules potentially causing thickness increase of the intima-media layer.

These results represents a step towards the incorporation of wall thickness inferences from baseline imaging that can be incorporated into fluid-structure interaction models and estimations of wall stress distribution. These results also help to strengthen the flow-mediated degeneration hypothesis, whereby persistently elevated WSS throughout the cardiac cycle exposes mechanocytes within the aortic wall to prolonged stimuli and downstream potentially maladaptive remodeling (Humphrey et al., 2015). However, the conducted hierarchical analysis also suggests that data variance at the patient level remains significant.

Our analyses on the effect of WSS on wall mechanical properties produced two main findings. Firstly, higher WSS was associated with a reduced DEF in the longitudinal direction. This contrasted with circumferential DEF, which showed an insignificant increasing trend. This further exemplifies the anisotropic nature of the aortic wall and suggests that it extends to influence both tear direction and location in aneurysm dissection (Manopoulos et al., 2018). Secondly, the aortic wall had significantly higher values for UTS with higher WSS, indicating a reduced likelihood for wall rupture in response to elevated WSS. In addition, the aorta might be stiffer in response to high WSS, as suggested by elevated values for MTS. This result, however, was not statistically significant.

Higher WSS levels were found to be associated with a reduction in both elastin abundance and smooth muscle cell (SMC) count. Loss of elastin integrity and relative increase in collagen describes the reduced compliance and increased stiffness of the aorta seen in ageing (Wagenseil et al., 2009). Studies have shown that global increases in vascular structural stiffness reflect increased central pulse pressures and pulse wave velocities (Lacolley et al., 2009; Safar et al., 2010) that pathologically increase proximal aortic loading. Collagen deposition in many cases of ATAA has been observed to be higher, reflecting the compensatory fibrotic changes as a result of the disease process (Wågsäter et al., 2013; Meng et al., 2014). The microstructural response to elevated WSS therefore explains these changes (Humphrey et al., 2015), and describes the resulting aortic wall stiffness and delamination potential, thus increasing the likelihood of dissection (Phillippi et al., 2011).

In addition to endothelial cells and fibroblasts, SMCs are part of the repertoire of crucial mechano-sensing and -regulating cells in the aortic wall. These cells display adaptive remodeling in response to shear stress encountered by endothelial cells, which they detect with integrins, glyclocalyx, membrane microdomains, cytoskeleton, receptor tyrosine kinases and others. The direct link between WSS level and SMC found in the present study is likely to result from mechanodysregulation in response to elevated WSS, involving a disruption of cell matrix connections which are vital to aortic wall integrity (Michel et al., 2018). Dysfunctional mechanosensing can lead to cellular apoptosis and/or an atrophic remodeling response, thus disturbing the structural integrity of the aortic wall (Leung et al., 1976). Whilst there are a number of possible intracellular and matrix signaling pathways associated with the process that we have not tested for, they are likely to culminate in a final common pathway, leading to cellular loss and matrix degeneration.

This study benefits from a robust method of ATAA flow-to-tissue patient-specific association, which arises from segmental aneurysm analysis. Conducting the tissue characterization and CFD portions of work separately has helped reduce the risk of bias, that could result from basing aortic tissue acquisition on findings from flow analysis retrospectively (Guzzardi et al., 2015). Our statistical methods have aimed to appreciate the spread of data and made use of multilevel regression models, which have not been utilized in similar studies. Altogether this aims to improve the validity of the findings.

The study is limited by its small sample size. This may perhaps explain the lack of significance in some relationships. Larger studies would also help deal with potential confounders such patient covariates and valve function. In addition, CFD simulations were conducted under rigid wall assumption (i.e. aortic wall compliance was not taken into account). Including aortic wall compliance could further improve WSS estimation. However, this would significantly increase the computational time required to conduct patient-specific simulation further decreasing the likelihood of adoption of this technique into the clinic. Rigid wall is therefore a common assumption is several computational studies (Morbiducci et al., 2013) with a translational goal. In addition, we do not expect WSS results to be significantly affected by this assumption. Firstly, because our mechanical test results suggest increased stiffness in regions of enhanced WSS. Secondly, previous imaging studies have shown a reduction in aneurysm wall compliance when compared with healthy tissues. Larger sample sizes and the incorporation of computational methods to couple flow and wall material properties will form the basis of future work.