Tissue-engineered collagenous fibrous cap models to systematically elucidate atherosclerotic plaque rupture

A significant amount of vascular thrombotic events are associated with rupture of the fibrous cap that overlie atherosclerotic plaques. Cap rupture is however difficult to predict due to the heterogenous composition of the plaque, unknown material properties, and the stochastic nature of the event. Here, we aim to create tissue engineered human fibrous cap models with a variable but controllable collagen composition, suitable for mechanical testing, to scrutinize the reciprocal relationships between composition and mechanical properties. Myofibroblasts were cultured in 1 × 1.5 cm-sized fibrin-based constrained gels for 21 days according to established (dynamic) culture protocols (i.e. static, intermittent or continuous loading) to vary collagen composition (e.g. amount, type and organization). At day 7, a soft 2 mm ∅ fibrin inclusion was introduced in the centre of each tissue to mimic the soft lipid core, simulating the heterogeneity of a plaque. Results demonstrate reproducible collagenous tissues, that mimic the bulk mechanical properties of human caps and vary in collagen composition due to the presence of a successfully integrated soft inclusion and the culture protocol applied. The models can be deployed to assess tissue mechanics, evolution and failure of fibrous caps or complex heterogeneous tissues in general.

www.nature.com/scientificreports/ rest, to promote collagen synthesis and remodeling (i.e. evoking fiber anisotropy), or (2) a continuous cyclic straining protocol (C-strain) to shift this balance to collagen remodeling and maturation, stimulating alignment and crosslinking, improving tissue mechanical properties 19,21,[48][49][50] . As a third group, statically cultured samples were included to obtain more isotropically orientated collagenous tissues. Of note, a soft inclusion (i.e. lipid core analog) was integrated in each model type, as it is known that local variations in matrix stiffness create heterogeneous stress and strain distributions that dictate collagen formation and remodeling 47 .

Materials and methods
Experimental outline. Figure 2 provides a graphical representation of the in vitro fibrous cap models ( Fig. 2A), the (dynamic) cell culture protocols and the analyses performed (Fig. 2B). In brief, vascular derived myofibroblasts (i.e. human vena saphena cells (HVSCs)) were seeded in fibrin-based gels, longitudinally constrained between two Velcro strips (1 × 1.5 cm-sized fibrin gels). After 7 days of static culture a 2 mm ∅ core was punched in the center of each sample and filled with fibrin to form a soft inclusion as an analog for the lipid core. It was hypothesized that local variations in matrix stiffness, due to the integration of a soft inclusion, albeit in combination with dynamic loading would result in variations in the local collagen architecture 47 . The samples were statically cultured for another week to ensure integration of the soft inclusion within the tissue. Next, the samples were equally divided amongst three experimental groups and exposed to either dynamic loading (intermittent-1 h 4% strain followed by 3 h of rest-or continuous-4% straining) or no loading (static) up to day 21. At day 21, samples were exposed to uniaxial tensile tests and further analyzed with tissue assays, histological analyses and imaging to assess mechanical properties and collagen composition (i.e. amount, type, orientation and crosslinking), respectively. Two experiments were performed in serial to verify reproducibility.
Engineering of fibrin-based tissues. Mold   Graphical representation of the tissue engineered in vitro fibrous cap models generated. Collagenous tissues were generated by seeding HVSCs in fibrin gels between pares of Velcro strips and exposing them to three different (dynamic) cell culture protocols (A). More specifically, samples were first statically cultured for 14 days, whereafter they were exposed to another 7 days of static culture (group 1), 7 days of intermittent strain (group 2, i.e. alternating periods of straining ( www.nature.com/scientificreports/ parallel 10 mm apart to form the base of the mold. These Velcro strips allow cells to grab onto, while simultaneously providing a grip for mechanical testing. Silicon glue to attach the Velcro strips was dried overnight in an oven at 60 °C whereafter the strips were incubated for 30 min with 5 mL of 70% EtOH/well at room temperature (RT). After this incubation period, the wells were rinsed with sterile PBS (3 × 5 min) and placed under UV for 30 min to decontaminate. Rectangular polydimethylsiloxane (PDMS) bars (10 × 4 × 4 mm) were created by mixing a silicon elastomer base and a curing agent (10:1 ratio) and curing the mixture overnight at RT. The bars were sterilized using 70% EtOH, washed with sterile PBS (3 × 5 min) and placed on the sides in between the Velcro strips to create a rectangular mold (dimensions: 10 × 15 mm) for the fibrin-cell suspension to solidify during the initial phase of the culturing process. The bars were removed on day 5 of cell culture.
Human vena saphena cell culture. HVSCs were isolated from the human vena saphena magna of a 63-year old women after coronary bypass surgery according to previously established protocols 51 conform the Dutch advice for secondary-use material. The cells were cultured in a standard incubator (37 °C, 5% CO 2 ) in complete growth medium consisting of advanced-DMEM supplemented with 10% Fetal Bovine Serum, 1% Glutamax and 1% Penicillin/Streptomycin. Cells were passaged at 80% confluency and used for the experiment at passage 7.
Preparation of collagenous tissues. To create the tissues, HVSCs were seeded in a fibrin matrix that served as a temporary matrix for cells to deposit their own extracellular matrix (ECM). A suspension of bovine fibrinogen (10 mg mL −1 , Sigma F8630), bovine thrombin (10 U mL −1 , Sigma T4648) and HVSCs (1.5 × 10 6 cells mL −1 ) were mixed and seeded in the mold between the Velcro strips and the PDMS bars, as well as covering half of each Velcro strip to allow cells to anchor to the Velcro brushes. Potential air bubbles were removed before solidification of the fibrin gel, using a sterile needle (BD Microlance, 24 G). The samples were placed in an incubator (37 °C, 5% CO 2 ) for 30 min to gelate, whereafter l-ascorbic acid 2-phosphate (vitamin C, 0.25 mg mL −1 ) and ɛ-Amino Caproic Acid (ɛ-ACA, 1 mg mL −1 ) enriched medium (5 mL well −1 ) was added. Vitamin C and ɛ-ACA were added to the medium to stimulate collagen synthesis and reduce fibrin degradation, respectively. Medium was refreshed every 3 days with ɛ-ACA being added during the first 7 days, to prevent fibrin break-down and allow cells to deposit their own ECM 47 .
Creating a soft inclusion to mimic the soft content of the lipid core and heterogeneity of a plaque. After 7 days of static cell culture, media were removed and a core (2 mm ∅) was punched in the middle of each tissue with a sterile biopsy punch (Kai Medical, Japan). The circular tissue was removed and filled with a suspension of bovine fibrinogen (10 mg mL −1 , Sigma F8630) and bovine thrombin (10 U mL −1 , Sigma T4648) to form a fibrinogenic soft inclusion (stiffness 0.5-1.0 kPa) 52 that represents the lipid core region. The tissues were placed in a standard incubator (37 °C, 5% CO 2 ) for 15 min to allow the suspension to solidify, whereafter fresh vitamin C enriched medium was added.
Dynamic cell culture to stimulate collagen formation and remodeling. The fibrin-based gels in the Bioflex plates were placed on the Flexcell FX-4000T (Flexcell Int, McKeesport, PA). The Flexcell system applies a vacuum onto the compartment under the flexible silicon membrane pulling the membrane over a post to induce a strain in the gel. The samples were exposed to either 7 days of I-strain (4% strain for 1 h (0.5 Hz) followed by 3 h of rest) or C-strain (4% strain, 0.5 Hz) to promote collagen synthesis and remodeling (i.e. evoking fiber anisotropy), or collagen remodeling and maturation, respectively 19,21,[48][49][50] .
Strain validation. In order to validate the amount of strain applied to the tissues during culture (set to 4% in the software), Digital Image Correlation (DIC) analysis was performed on the images of maximum compaction and extension in the direction of loading. Videos were captured at day 21 and converted to images at 30 Hz in MAT-LAB (Mathworks, Massachusetts, USA), whereafter the images of maximum compaction and extension were selected. Subsequently, the average strains in the y direction (ɛ yy ) per loading regime group were calculated (Supplementary Table S1) in the region of interest (ROI) (i.e. mid-cap section, Supplementary Fig. S1) using the open source 2D DIC software Ncorr (v1.2, www. ncorr. com) 53 (see paragraph tissue mechanical characterization).
Tissue analyses. At day 21, tissues were processed for further analysis. Samples were either (1) washed in PBS (3 × 5 min), fixated for 1 h in 3.7% formaldehyde, followed by a PBS washing step (3 × 5 min) and stored at 4 °C, or (2) directly used for uniaxial tensile testing. After the uniaxial tensile test, left-over material was washed in PBS, snap frozen in liquid nitrogen and stored at − 80 °C. At a later timepoint, these samples were used to determine DNA and hydroxyproline (HYP) content. Fixated samples were used for histological analysis to determine collagen orientation, collagen type and the presence of LOX.
Tissue composition. The tissues that were reserved for the DNA and HYP determination were consecutively lyophilized, weighted and digested in papain enriched digestion buffer (100 mM phosphate buffer (pH = 6.5), 5 mM l-cysteine (C-1276), 5 mM ethylene-di-amine-tetra-acetic acid (EDTA, ED2SS), and 140 μg mL −1 papain (P4762), all from Sigma-Aldrich). Each sample was cut in small pieces, mixed with 300 µL of digestion buffer in a new Eppendorf tube and placed at 60 °C (16 h, Thermormixer Compact F 1.6 A,VWR) for digestion. After digestion, samples were stored at 4 °C until analyses. Prior to DNA quantification, samples were vortexed and centrifuged (12,000 rpm, 10 min on 4 °C) to mix and precipitate remnants, respectively. DNA content was quantified using 10 μL of the supernatant, the Qubit dsDNA BR assay kit (Life Technologies, Carlsbad, California, www.nature.com/scientificreports/ USA) and the Qubit fluorometer (Life Technologies) following manufacturer's instructions and normalized to tissue dry mass. To quantify HYP content digested samples were first hydrolyzed using 16 M sodium hydroxide (Merck; B1438798). Subsequently, HYP content was quantified with a Chloramin-T assay, including trans-4-hydroxyproline as a reference (Sigma; H5534) 54 . HYP values were normalized to total dry mass and DNA content to assess overall collagen formation and collagen production per cell.

Soft inclusion geometry and integrity.
Brightfield microscopy images were taken of each soft inclusion after one and two weeks of culture to assess cellular infiltration and soft inclusion geometry and integrity (i.e. no visible defects within the soft inclusion or at the tissue-soft inclusion interface). Soft inclusion geometry was quantified using the Image J software (version 1.52 k, imagej.nih.gov).
Matrix content and organization. Whole-mount samples or 7 µm thick cryosections were analyzed to evaluate matrix content and organization using either fluorescence stainings, histology or immunohistochemistry (IHC). As a positive control, 7 µm thick cryosections of a human atherosclerotic plaque of a 68 year old male were included. This carotid endarterectomy specimen was retrieved in line with the ethical guidelines of the 1975 Declaration of Helsinki, after written informed consent from the patient and approval of the Medical Ethics Committee of the Erasmus Medical Center. After endarterectomy, the plaque specimen was snap frozen, stored at − 80 °C and embedded, similar as the collagenous tissue engineered fibrous cap models, in Tissue Tek, to allow sectioning.
Collagen organization. Whole-mounts samples were incubated with an Oregon Green labeled CNA35 probe (CNA35-OG488) and the nuclear stain DAPI to assess collagen and cellular organization. The in-house developed CNA35 probe binds to a variety of fibril forming collagens, including type I and III, with little to no specificity for other extracellular matrix proteins 55  Tissue mechanical characterization. Uniaxial tensile tests were performed at day 21 to assess the mechanical properties and rupture behavior of the samples of the different experimental groups using a test set-up equipped with a 5 N load cell (CellScale Biomaterial Testing, Waterloo, Canada). Before testing, samples were removed from the well-plate, washed in PBS and photographed on a grid background to determine the compaction rate in the horizontal (x, sample width) and vertical (y, sample length) direction. The compaction in the z direction (thickness) was measured using a digital microscope (Keyence VHX-500FE, Itasca, IL). The Velcro's on each side of the sample were placed in the clamps of the tensile tester. Each sample was mounted in the set-up and exposed to uniaxial loading in PBS at 37 °C to simulate physiological conditions. After preconditioning (10 cycles of 10% strain 56 ), the samples were uniaxially strained at a 100%/min strain rate until failure. A camera and lens were located directly above the tissues, capturing images during testing at a speed of 15 Hz for digital image analyses.
The nominal engineering stress-strain curves were derived from the force and displacement measurements utilizing the average initial thickness and width measurements in the ROI ( Supplementary Fig. S1), assuming incompressibility and plane-stress conditions.
Data averaging was performed on the stress-strain data and the stiffness values were computed as the linear slopes at 2%, 5% and 10% strain (i.e. ɛ 2% , ɛ 5% , ɛ 10%, ) based on the DIC analyses described below.
Before mounting the sample in the uniaxial tensile set-up, graphite particles were applied onto the surface of each sample with a cotton swab to facilitate optical strain analysis of the region of interest via digital image correlation. 30% of the sides of each tissue were excluded from analyses to exclude boundary artifacts caused by the clamping and compaction.
DIC analyses were performed as a normalization method to exclude potential clamping and compaction artifacts and allow a consistent comparison between the different samples. The DIC analyses were performed in MATLAB on the ROI using the open source 2D DIC software Ncorr (v1.2, www. ncorr. com) 53 .
A relative error comparison technique was used to select the DIC parameters (subset size: 35 pixels, spacing size: 3 pixels, and strain radius size: 5 pixels). Local strains were assessed in the ROI to determine at which image www.nature.com/scientificreports/ there was an average of 12% strain in the direction of loading (ɛ yy ). The regions on each side of the soft inclusion were treated separately as local variation may occur. The stress-strain data was scaled according to this image and the averages were calculated (i.e. average between 0 and 2% strain resulted in an average data point at 1% strain). The linear slope was calculated between 1 and 3% strain, 4% and 6%, and 9% and 11%, in order to extract the stiffness values at 2%, 5%, and 10%, respectively. The average stress-strain curves for each group (static control, I-strain, and C-strain) were extracted, including the standard deviations, to provide a representative plot of the stress-strain behavior for all tissues.
Statistical analysis. All data (DNA, HYP, compaction and stiffness values) are presented as the mean ± standard deviation unless stated otherwise. Statistical analyses were performed using Prism (GraphPad, La Jolla, CA, USA). A Shapiro-Wilk test and Levene's test were performed for normality and equal-variances. Statistical significant differences between the experimental groups, in case of proven normality, were detected using an one-way ANOVA with subsequent Tukey's post-hoc test (assuming equal variances) or Dunnett's posthoc test (unequal variances). In case of not normally distributed data, a nonparametric Kruskal-Wallis test with a Dunn's multiple comparison test was performed. Differences were considered statistical significant for p values < 0.05 (visualized as *p < 0.05; **p < 0.01; ***p < 0.001).

Results
Three reproducible fibrous cap models with integrated soft inclusions. After 21 days of cell culture, consistent tissues were formed between the Velcro strip constraints for each (dynamic) cell culture protocol applied (Fig. 3A). Exposure to cyclic strain during the last week of culture resulted in enhanced compaction and consequent dog bone shaped tissues for both the I-strain and C-strain group. Whereas the statically cultured  www.nature.com/scientificreports/ controls exhibited relative low percentages (− 18 ± 6%) of compaction in width, higher rates of compaction were seen when loading during cell culture was applied (I-strain: − 44 ± 4% and C-strain: − 42 ± 12%). Smaller differences in compaction in the z direction (i.e. tissue thickness) were observed between the experimental groups, with an overall average thickness of approximately 537 ± 169 µm (Fig. 3B).
To quantify collagen content in the tissues, the amount of HYP, a matrix component for stabilization of the collagen molecule, was quantified (Fig. 3C). In general, negligible differences in HYP presence per tissue dry mass were found. However, a statistically significant increase in cell number upon exposure to continuous loading in comparison to the static control (p = 0.008) was observed. This increased cell number was accompanied by a significant decrease in amount of HYP presence per cell between the static and continuous strain group (p = 0.013).
At day 7 of culture, soft inclusions were created to mimic the soft lipid core, overlying fibrous cap and heterogeneity of a plaque. With time, the soft inclusions were integrated into the tissues, as characterized by cell and collagen ingrowth into the inclusions (Fig. 3D,E). At day 21, infiltrated cells covered the complete soft inclusion area for all groups assessed, though negligible amounts of collagen were deposited throughout this soft inclusion area in comparison to the rest of the sample (Fig. 3E).
Collagenous tissues with (dynamic) culture protocol dependent collagen organizations. The local collagen fiber dispersion was assessed and quantified (Fig. 4). Overall, a more condensed ± 5-50 µm thick collagen layer (1-10% of the tissue thickness) was detected at the upper surface of each tissue. This densely packed layer demonstrated a clear anisotropic collagen organization, in the direction of the Velcro constraints, irrespective of the type of tissue or locus assessed (Supplementary Fig. S5). Throughout the other ≥ 90% of the tissue thickness the collagen fiber dispersion was consistent though dependent on the locus analyzed and culture protocol applied. Symmetry in collagen fiber distributions were observed between the top and base and the left and right with respect to the soft inclusion of all tissues within one experimental group. As such, only the results of one quartile per tissue type are presented (Fig. 4).
All top regions analyzed exhibited isotropic collagen organizations irrespective of the culture protocol applied. The predominant fiber angles ranged between 74° degrees for the statically cultured samples up till 61° and 45° degrees when exposed to intermittent and continuous loading, respectively. Moving towards the shoulder and mid cap regions, the fibers appear more anisotropic in the direction of the Velcro constraints for the dynamically loaded samples. Those samples showed predominant fiber angles of 4° and 3° in the shoulder region and 4° and 5° in the mid cap region for the I-strain and C-strain respectively. The static control demonstrated a more isotropic collagen fiber distribution with a predominant fiber angle of 77° in the shoulder and 35° in the mid cap region. Figure 4 demonstrates the created organized collagenous tissues with increased collagen alignment in the direction of the Velcro constraints when moving from the top to the shoulder and ultimately the mid-cap region, especially when loading was applied during culture.
LOX positive collagenous tissues that encompass collagen type I and III. Histological assessment of the tissue engineered fibrous cap models tissue content, demonstrated a similar collagenous architecture as observed in the fibrous cap of human atherosclerotic plaques, especially when loading was applied during culture (Fig. 5A,B). H&E and PSR staining shows a similar morphology, although the intensity of the H&E stain was less intense for the tissue engineered fibrous cap models in comparison to the human cap control (Fig. 5A,B). Collagen specific stainings (Fig. 5C) further revealed the presence of collagen type III and I of which the latter appeared to have a predominantly increased expression in the C-strain group when compared to the I-strain and predominantly the static group. An increased number of cells are present at the luminal side of the fibrous cap, which is attributable to the presence of an endothelial layer (Fig. 5C,D) that is deliberately left-out in our models.
In correspondence to the human cap, LOX was detected in all engineered tissues demonstrating a homogeneous presence of LOX throughout the cross-section (Fig. 5D).

Discussion
Within this study we present three tissue engineered fibrous cap models, suitable for mechanical testing, that can each serve as a tunable human in vitro platform to elucidate the reciprocal relationships between atherosclerotic cap composition and mechanics. In contrast to real plaques, the created reductionist models comprise the main attributes that determine cap collagenous matrix mechanical behavior and strength in a controlled manner. All three variants presented show an integrated soft inclusion (i.e. lipid core analog) sealed by a collagen type I and III-positive, organized and crosslinked collagenous ´cap´ that mimics the microscopic architecture and mechanical behavior of human caps. Importantly, by changing the cell culture conditions applied collagenous matrix content and organization can be tuned. The models allow to properly define global and local tissue properties and can be seen as a first step to elucidate cap failure in vitro.
To create the models dynamic cell culture protocols were selected and compared to static control strategies to create matrices in relative short culture periods with variable collagen content (i.e. amount, type and crosslinking) and organization. Confirming expectations, we observed that cyclic strain, predominantly the continuous straining protocol, induced HVSC proliferation [57][58][59][60][61] . Although cyclic strain is often associated with accelerated www.nature.com/scientificreports/ matrix deposition, as specifically shown for intermittently strained HVSCs 21 , no difference in total collagen content was detected between the different experimental groups in our study. This might be attributed to the duration of load exposure (i.e. 7 days) as it was previously reported that dynamically loaded HVSCs initially, during the early phases of cell culture, favour a proliferative over a synthetic response 21,50 .
We obtained more compacted dog-bone shape matrices upon dynamic loading, with no preservation of the initial fibrin gel volume, suggesting enhanced matrix remodeling. Indeed, when evaluating the intrinsic collagenous architecture, clear anisotropic collagen configurations, in the direction of loading, were observed throughout the complete tissue thickness for the dynamically loaded samples as well as in the top layer of the static control. These matrix (re-)orientation responses are known to be a direct consequence of cellular (re-) orientations due to tissue constraints, contact guidance of the endogenous matrix and the magnitude and type of applied loads 45,47,49,62,63 . Within this study, the Velcro constraints were the main directors of observed collagen anisotropy. Confirming previous studies, it was observed that cells oriented and deposited collagen parallel to   www.nature.com/scientificreports/ the direction of the Velcro constraints 47 . Moreover, remodeling processes were significantly faster upon dynamic loading but respectively slower in deeper tissue layers in comparison to the superficial top layer due to the contact guidance provided by the endogenous matrix 47 . As the Velcro constraints were only placed in two positions, in the direction of loading during cell culture, the physiological strain shielding response of HVSCs (i.e. orientating themselves perpendicular to the direction of loading to shield themselves from excessive straining) was prevented. Interestingly, a predominantly isotropic collagen organization was detected in the top and base regions of the dynamically stimulated tissues (i.e. the area between the Velcro constraint and the soft inclusion). These results illustrate that cellular (re-)orientation processes and consequent collagen fiber (re)distributions depend on the cumulative presence of constraints, matrix stiffness and local strain patterns. Histological comparisons between native fibrous caps and the engineered fibrous cap regions demonstrate that both collagen type I and III, the main load bearing fibrillar collagen types of the fibrous cap [23][24][25] , were present in the engineered tissues with qualitative increasing rates of collagen type I upon continuous cyclic loading during cell culture. Engineered fibrous cap collagen type III expression was limited when compared to the human cap control. However, both decreases 64 as well as increases 65 in the proportions of type III to I have been observed in atherosclerotic lesions before. Variations in collagen type have been correlated with tissue mechanics 66,67 , though so far no correlations were found between collagen type proportion and fracture stress in fibrous caps 68 . Changing cell culture loading protocol offers the opportunity to adapt this ratio, embracing the variance observed in vivo.
Besides collagen content, organization and type we assessed LOX expression because of its catalyzing role in collagen crosslinking. Previous observations revealed a dominant role for collagen crosslinks over collagen content with respect to tissue mechanics in comparable tissue-engineered matrices 19 , bone 69 , scar tissue of collateral ligaments 70 , patellar tendon 71 as well as cardiovascular tissues 19 . Within this study, we found similar levels of LOX expression between the different tissue engineered fibrous cap models and the human cap control.
It is possible that subtle differences in collagenous matrix organization and collagen type are reflected in tissue mechanics. Yet, in line with the similar rate of LOX expression, no statistically significant differences in the global mechanical properties (i.e. elastic moduli) were detected between the dynamically and statically cultured tissues. Although it is difficult to make a direct comparison to human caps, because of their complex composition and the variety of mechanical analyses performed in literature 6 , all engineered tissues analyzed demonstrated physiological strain stiffening behavior and calculated stiffness values fell within the range reported for human carotid fibrous caps (i.e. 0.5-5 MPa) 6 . We chose to exclude ultimate tensile strengths values, as many samples, especially the statically cultured, ruptured outside the locus of interest due to local stress accumulations at the Velcro side, underlining the necessity to develop new clamps that improve load transmission. Although a thorough evaluation of local strain patterns is beyond the scope of this study, DIC analyses before rupture demonstrate regions of elevated extension left and right of the soft inclusion. Those regions coincided with the locus of failure for those samples that failed in the region of interest (i.e. the fibrous cap region) for all experimental groups assessed (Supplementary Fig. S6).
Finally, we make the note that we made some well-considered choices. First, we chose for a cellular approach because of the inherent advantage that we generate more physiological-like cell made and crosslinked collagenous matrices. Theretofore, cells of the human vena saphena magna were selected as consistent myofibroblastic cell source. They were selected based on their proven role in collagen deposition and remodeling 51 , our previously obtained experience using these cells for cardiovascular tissue engineering purposes 19,21,44,47,63,72 , and the anticipated large number of caps that can be consistently created in vitro with cells of a single donor. The human vena saphena magna is often used for coronary bypass surgeries 73 and adventitial myofibroblast as well as phenotypically altered synthetic VSMCs are deemed responsible for fibrous cap formation 74,75 . Consistent results were obtained between the two serial experiments, confirming the repeatability of the methodology. Although comparable cell responses have been observed in VSMCs and between various HVSC donors, potential cell type and donor to donor related variations in matrix deposition, possibly as a result of age, sex and co-morbidity [76][77][78] , should be taken into consideration. Creating more personalized models could further enhance risk of rupture stratification.
Second, as human caps develop and adapt over decades and collagen half-life in disease might decrease up to 10-fold 79 , it is unrealistic to expect (and as such not our purpose to reproduce) an exact mimic. The purpose of our in vitro model is to create a physiological-like tunable platform that will be iteratively used, together with more simplified in vitro and more complex ex vivo models (e.g. human plaque lesions), to test, validate and revise hypotheses on matrix mechanics and failure. Importantly, the model is suitable for high frequency and resolution real-time imaging to study collagen evolution and failure instantaneously. By elongating load duration, changing load type or changing soft inclusion geometry, it is expected that the engineered collagenous tissue properties as demonstrated in this study can be further manipulated 19,21,44,47,72 . To stretch the boundaries of our current approach, additives like TGF β 80 , ribose 81,82 , copper 83 or beta-aminopropionitrile 84 might be added, to further stimulate collagen deposition, induce AGE related crosslinking and enhance or inhibit LOX catalyzed crosslinking respectively.
Lastly, plaque and cap rupture cannot be attributed to solely suboptimal collagenous properties. It is a cumulative consequence of these properties, local inflammatory cell driven collagen degradation, and steep gradients in stiffnesses and loading resulting in local stress accumulation [85][86][87][88][89] . The current models can be used as a first step covering cap complexity. Stiff calcifications, intraplaque hemorrhages, inflammatory cells or substitutes can be added to further embrace physiological variance and plaque failure modes. Overall, the models will aid in understanding material properties, local strain fingerprints, stress accumulations and tissue failure modes in collagenous tissues. Information that can be implemented in in silico models to improve biomechanical stress analyses and the assessment of plaque vulnerability specifically or tissue damage in complex heterogeneous tissues in general 6 .

Conclusion
To conclude, we present three tissue-engineered fibrous cap models that aid in the controlled and systematic investigation of the reciprocal relationships between cap composition and mechanical properties. By applying different (dynamic) cell culture protocols we were able to create reproducible naturally crosslinked tissues, with a tunable collagen content and organization, suitable for mechanical testing. All models contained a successfully integrated soft inclusion (i.e. lipid core analog) that was covered by a collagen type I and III-positive organized 'cap' , to simulate plaque tissue heterogeneity. These caps exhibited physiological strain stiffening responses and calculated stiffness values that fell within the ranges reported for human caps in literature. Each model can be seen as a tunable physiological-like 3D platform, that can be iteratively used, together with more simplified in vitro models and more complex ex vivo material, to ultimately study and elucidate cap rupture. However, the models can also be of value to understand tissue mechanics and damage in complex heterogeneous tissues in general. Moreover, as this model is suitable for high-resolution, high-frequency real-time imaging, it can be used to investigate collagen remodeling and failure at the microlevel. Future efforts should focus on integrating stiff inclusions, potentially calcifications, or collagen degrading factors, like macrophages, to further aid in the identification of cap rupture risk.

Data availability
The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.