Anisotropic Collagen Fibrillogenesis Within Microfabricated Scaffolds: Implications For Biomimetic Tissue Engineering


  • Aurélie Jean,

    1. Department of Bioengineering, The Pennsylvania State University, 223 Hallowell Building, University Park, PA 16802, USA
    Search for more papers by this author
  • George C. Engelmayr Jr.

    Corresponding author
    1. Department of Bioengineering, The Pennsylvania State University, 223 Hallowell Building, University Park, PA 16802, USA
    • Department of Bioengineering, The Pennsylvania State University, 223 Hallowell Building, University Park, PA 16802, USA.

    Search for more papers by this author


original image

Anisotropic collagen fibrillogenesis is demonstrated within the pores of an accordion-like honeycomb poly(glycerol sebacate) tissue engineering scaffold. Confocal reflectance microscopy and image analysis demonstrate increased fibril distribution order, fibril density, and alignment in accordion-like honeycomb pores compared with collagen gelled unconstrained. Finite element modeling predicts how collagen gel and scaffold mechanics couple in matching native heart muscle stiffness and anisotropy.

Collagen gels find widespread application as three-dimensional substrates in cell culture assays,1 drug delivery,2 and tissue engineering.3 At the macroscale, boundary constraints influence cell-laden collagen gel anisotropy;4 at shorter length scales, composites of collagen gels with microfabricated materials5–7 raise questions concerning how fibrillogenesis itself may be influenced by the geometry of such microstructures. In particular, collagen gel morphology imparted by compartmentalization within microfabricated materials could impact functional performance parameters (e.g., cell mobility, shape, or alignment;8–14 drug diffusion;2 hierarchical engineered tissue mechanics15–17) of such composite devices.

In biomimetic tissue engineering,18 collagen gels have been used for generating functional myocardium from heart cells.3 Collagen gels are capable of promoting cell alignment under boundary constraint4, 10 or cyclic loading,19 however they tend to be mechanically inferior to myocardium.8 We generated tissue engineered myocardium by cultivating heart cells on an accordion-like honeycomb (ALH) scaffold rendered by laser microablation of poly(glycerol sebacate) (PGS).20 The ALH scaffold provided cardio-mimetic anisotropic elastic properties and a capacity to guide preferential cell alignment. Toward enhancing heart cell-mediated contractility, we developed periodic finite element simulations for investigating changes in ALH scaffold geometry21 and investigated improving heart cell seeding efficiency via Matrigel.22 Matrigel, however, did not promote cardiomyocyte elongation.

Based on observations of directional collagen fibrillogenesis in collagen-doped microfluidic devices23 and that elongated scaffold pores can promote cell-secreted collagen alignment,9 we speculated that directional collagen fibrillogenesis might manifest within ALH pores.18 Anisotropic collagen fibrillogenesis could potentially overcome the limited heart cell elongation observed in Matrigel-ALH composites.22

To elucidate if ALH scaffolds can induce anisotropic collagen fibrillogenesis, the three-dimensional fibril organizations within ALH (Figure 1A) and square diamond (Figure 1B) pores were imaged by confocal reflectance microscopy (Figure 1C,D) and compared to collagen gelled unconstrained on glass slides (Figure S1, Supporting Information). ALH scaffolds preferentially oriented fibrils along the long axis of the pore (Figure 1C,E), with an orientation index OIALH = 17.75 ± 6.55% significantly higher than that measured for glass slides (OIglass = 1.45 ± 0.40%; p < 0.05). Indeed, neither glass slides nor square diamond scaffolds (Figure 1D,E; OIsquare = 0.26 ± 0.16%) induced preferential fibril orientation along the pore long axis. Providing a measure of both the density and homogeneity of the collagen gel, the inter-fibril distance distributions were quantified (Figure 1F). The mean interfibril distance was 5.06 ± 0.05 μm for collagen gelled on glass slides; values for the ALH (4.075 ± 0.5 μm) and square diamond (4.08 ± 0.3 μm) scaffolds were lower (p < 0.05). Hence, the organization of fibrils tended to be denser upon compartmentalization within the pores of ALH and square diamond scaffolds than on glass slides. The entropy value calculated for ALH pores (ϵALH = 3.8 ± 0.23) was significantly lower than that calculated for glass slides (ϵglass = 4.6 ± 0.10; p < 0.05), demonstrating that the organization of fibrils was more ordered in ALH pores. Square diamond pores exhibited an intermediate value ϵsquare = 4.18 ± 0.12.

Figure 1.

Representative scanning electron microscopy images of ALH (A) and square diamond (B) scaffolds (100× original magnification; scale bars = 1 mm). Confocal reflectance micrographs of collagen fibrils within a representative pore of an ALH (C) and square diamond (D) scaffold (400× original magnification; scale bars = 100 μm). Collagen fibril angular orientation distributions (E) and inter-fibril distance distributions (F) measured by image analysis of collagen-filled ALH (top; red) and square diamond pores (bottom; red) compared with collagen gelled unconstrained on a glass slide (blue).

To predict the mechanical stiffnesses and anisotropy of ALH scaffold-collagen composites, compare these with native heart muscle, and to investigate the ramifications of anisotropic collagen fibrillogenesis in the ALH scaffold, periodic FE simulations21 were conducted. FE predicted effective stiffnesses EPD and EXD, and anisotropy ratio r = EPD/EXD, and Voigt (Equation 4) and Reuss (Equation 5) elastic bounds of the composite initially assumed the collagen was isotropic using upper (24.3 kPa15) and lower (5 kPa24) bounds of collagen stiffness. The Reuss bound (range 7.3–35.1 kPa) was dictated by the most compliant component of the composite (i.e., the collagen); the Voigt bound (aka the “rule-of-mixtures”; range 265–278 kPa) was dictated by the stiffer component, and therefore varied only slightly with the stiffness of the collagen. By contrast, FE predicted effective stiffnesses EPD (range 108–215 kPa) and EXD(range 62.2–173 kPa) depended strongly on the stiffness of the collagen and were comparable to values measured by uniaxial tensile testing (Figure S3, Supporting Information). As expected, FE predicted and measured values of EPD and EXD with 3 mg mL−1 collagen gelled within the pores were higher than those reported for the ALH scaffold itself (EPD = 83 kPa and EXD = 31 kPa20). The FE predicted anisotropy ratios r = 1.2 and r = 1.7 associated with the upper and lower collagen stiffness bounds, respectively, were significantly lower than that predicted for the ALH scaffold without collagen (2.5)21. Recognizing that the collagen within the ALH scaffold was itself anisotropic (Figure 1), we simulated collagen anisotropy via an orthotropic material model targeting left ventricular stiffnesses in the circumferential (157 kPa) and longitudinal (84 kPa) directions (r = 1.87)20 and solved for the requisite collagen gel stiffnesses in the PD and XD directions. FE simulations predicted collagen gel stiffnesses equation image = 47.0 kPa and equation image = 26.5 kPa (rcoll = 1.77). Finally, comparing the spatial distribution of equivalent von Mises strain within the collagen matrix for isotropic (Figure 2A) and orthotropic (Figure 2B) assumptions, orthotropic yielded a more homogeneous strain distribution (range 0.04–0.05) along the central PD axis versus isotropic collagen gel (range 0.03–0.07). Hence, FE simulations can be used to predict collagen gel properties required to match ALH scaffold-collagen composite stiffnesses and anisotropy to native heart muscle.

Figure 2.

Predicted spatial distribution of strain (equivalent von Mises) within the ALH pore simulating (A) isotropic versus (B) orthotropic collagen gel filling the pore. For each simulation a macroscopic strain of 0.1 was prescribed along the PD direction. (C) The orthotropic collagen gel was predicted to yield a more homogeneous strain distribution (range 0.04–0.05) along the central PD axis of the ALH pore compared with an isotropic collagen gel (range 0.03–0.07). Of note, in the isotropic case (Ecoll = 10 kPa) the spatial strain distribution pattern was dictated solely by the shape of the ALH pore; by contrast, in the orthotropic case (equation image) the pattern was further influenced by simulation of PD-aligned collagen. Color bar and associated numeric values indicate equivalent von Mises strain and apply to all figure panels A–C.

Composite devices comprised of microfabricated materials and collagen gels offer the prospect of controlled bridging between the micro-to-nanometer length scales, potentially yielding novel in vitro cell culture assays,1 drug delivery systems,2 and engineered tissues.3 Hierarchically, engineered tissues formed by seeding cells onto microfabricated scaffolds evolve through the structural–mechanical interplay between scaffold, cells, and extracellular matrix. Without accounting for extracellular matrix, we demonstrated that heart cell-seeded ALH scaffolds could mimic aspects of cardiac anisotropy.20 Extracellular collagen structures, however, play important roles in myocardium.18 We demonstrated three key findings regarding collagen gelled within the ALH pore versus unconstrained on a glass slide: 1) increased order of the fibril distribution (i.e., decreased entropy; ϵALH = 3.8 ± 0.23 versus ϵglass = 4.6 ± 0.10; p < 0.05), 2) increased fibril density (i.e., decreased mean inter-fibril distance; dALH = 4.075 ± 0.5 μm versus dglass = 5.06 ± 0.05 μm; p < 0.05), and 3) increased fibril alignment along the reference angle defined by the ALH pore long axis (OIALH = 17.75 ± 6.55% versus OIglass = 1.45 ± 0.40%; p < 0.05). For comparison, Bayan et al. reported entropy values ranging from 6.37–6.5 and OI values ranging from 9.45–13.46% in similar acellular collagen gels.25 Of note, Bayan et al. did not detect significant differences in OI when comparing 1, 2, and 3 mg mL−1 collagen gels. Further, when gelled on a glass slide and compared with the 3 mg mL−1 gel, we did not detect any difference in the collagen orientation distribution in a 6 mg mL−1 collagen gel (Figure S4, Supporting Information). Of note, the degree of collagen fibril alignment mediated by the ALH pore geometry alone (OI = 17.75 ± 6.55%) was less than that observed by Bayan et al. in a 3 mg mL−1 cell laden gel cultivated for 12 days (OI = 30.86 ± 14.76%).25

A combination of mechanisms may have contributed to the anisotropic collagen fibrillogenesis observed herein. For example, when 3 mg mL−1 collagen solution was flowed into and gelled within the channels of a collagen-doped alginate microfluidic device, Gillette et al. observed that a number of collagen fibrils appeared to bridge contiguously, in straight lines, from the collagen-doped alginate (i.e., the channel walls) into the collagen gelled within the channel.23 Coupled with the preference for collagen fibril tip growth predicted by diffusion limited aggregation models by Parkinson et al.,26 the results from Gillette et al. suggest that collagen fibrils can grow in a straight line from the tips of collagen fibrils exposed at a surface into the bulk of a collagen solution. In the present study collagen solution was gelled in direct contact with the PGS structural elements of the ALH scaffold. In a previous study, Sales et al. demonstrated that type I collagen can adsorb to a PGS foam scaffold from dilute solutions, reaching a maximum surface concentration from solutions as dilute 20 μL mL−1 collagen.27 We thus expect that the surfaces of the PGS struts were saturated with adsorbed collagen under the conditions tested herein, and that upon exhausting the available PGS strut surface area, the growing tips of the collagen fibrils would tend to progress outward from the struts into the bulk collagen solution filling the pore. Indeed, proximal to the collagen gel-PGS strut interfaces, confocal reflectance micrographs qualitatively revealed that collagen fibrils were arranged not in parallel, but rather at finite angles or roughly perpendicular to the PGS struts (Figure 1C,D). In the case of the square diamond pore, in which the PGS struts were oriented at opposing angles of ±45° and at equal distances from each other, essentially equal fractions of the collagen fibrils were oriented at ±45°, yielding no single preferential angle of alignment (Figure 1E). By contrast, while the PGS struts were likewise oriented at ±45° in the ALH scaffold, the distances between opposing struts were longer along the PD versus XD direction, thereby offering a longer path for extension of collagen fibrils along the PD direction of the ALH pore.

We undertook FE simulations to predict what stiffnesses the collagen matrix would need to manifest in order for the effective stiffnesses and anisotropy of the ALH-collagen composite to match those of native left ventricular myocardium. FE simulations predicted the collagen would need to exhibit equation image = 47.0 kPa and equation image = 26.5 kPa (rcoll = 1.77) in order for the composite to reach 157 kPa and 84 kPa (r = 1.87).20 Simulations suggested two potential routes toward matching ALH-based constructs to left ventricular mechanical properties. In the context of heart cell-seeding,20, 22 the stiffness of the collagen gel would be expected to increase as the gel is contracted by the seeded cells; a potential limitation, however, could be debonding of the collagen from the PGS scaffold upon cell-mediated gel contraction. Toward such approaches, we have demonstrated that cells and collagen can be retained within the ALH pore upon stretching the ALH scaffold (Figure S5, Supporting Information). A broad range of cell-seeded collagen gel stiffnesses have been reported ranging from ∼37 kPa (estimated from Figure 4 of Feng et al.8) to 5.33 ± 1.33 MPa.28 These studies suggest simulation predicted collagen stiffnesses of 26.5–47.0 kPa could be achieved by an appropriate combination of collagen gel concentration, cell seeding density, and cultivation time. An alternative approach could involve co-varying the ALH scaffold structure (e.g., strut width) and PGS curing conditions (i.e., PGS modulus).21 We demonstrated by FE simulations that two distinct values of strut width (w) are capable of yielding an anisotropy ratio equal to that of left ventricular myocardium (i.e., r = 1.87): w = 20 μm or w = 140 μm.21 The 20 μm strut width would be both feasible to microfabricate and provide allowance for increased collagen matrix stiffnesses associated with heart cell-mediated contraction. As collagen fiber alignment alone is not sufficient to explain the high degree of anisotropy observed in fibroblast-seeded collagen gels,29 we speculate that the anisotropic collagen fibrillogenesis demonstrated herein, while significant, represents only a starting point in understanding the interplay between pore geometry, collagen morphology, and cell morphology. In future studies, Voronoi tessellation-based models could be useful in coupling collagen gel morphology to mechanical behavior.30 Further, the evolution of collagen anisotropy demonstrated in the present study may be extendable to other hydrogels, such as fibrin and Matrigel. Of particular note, Bian et al. demonstrated that muscle cell-laden fibrin-based hydrogels can be spatially patterned into anisotropic tissue bundles by casting within microfabricated poly(dimethysiloxane) molds.31 More broadly, collagen gel-based cell culture assays and drug delivery systems may manifest and potentially exploit anisotropic collagen fibrillogenesis, in particular in miniaturized composites of collagen gel and microfabricated or microscale materials. For example, in a miniaturized aortic ring assay introduced by Reed et al., 30 μL of collagen solution was gelled within and supported by a nylon mesh ring (3 mm inside diameter) comprised of ∼50 μm diameter fibers arranged in a square lattice (∼125 × 125 μm inside pore dimensions).1 As such systems are further miniaturized for high throughput screening, collagen morphology induced by the system boundaries could potentially impact the directionality of capillary sprouting; similar phenomena could potentially be exploited in microfluidic collagen gels mimicking human microvascular networks.32 In the context of drug delivery, De Paoli et al. have reported on the effects of oscillating magnetic fields on drug release from magnetic collagen gels (i.e., collagen gels containing iron oxide particles of up to 3 μm diameter).33 In magnetic collagen gels, structural changes within the gel associated with oscillating magnetic fields were demonstrated to impact drug release kinetics; similar effects could potentially be mediated by collagen gelation within the compartments of microfabricated drug delivery devices.34 In concert with auxiliary biophysical and biochemical regulators, compartmentalizing collagen gels within microfabricated materials represents a promising strategy for controlling collagen fibril anisotropy and associated functional performance parameters in advanced cell culture assay, drug delivery, and tissue engineering applications.

Experimental Section

PGS Scaffold–Collagen Gel Composite Preparation: The ALH and square diamond scaffolds (5 mm × 5 mm) were fabricated by 213 nm laser microablation (LSX-213; CETAC Technologies, Inc., Omaha, NE) of 250 μm thick PGS cured under vacuum (<50 mTorr) for 7.5 h at 160 °C.22, 35 The ALH and square diamond scaffolds each comprised a periodic tessellation of unit cells with a strut width of 50 μm and a strut length (inside the pore) of 200 μm (Figure 1A,B). The long axis of the ALH pore was defined as the preferred direction (PD) and the short axis the cross-preferred direction (XD). Collagen gel was prepared on ice by adding 100 μL of 10X phosphate buffered saline (PBS) to 800 μL of 3 mg mL−1 bovine dermal collagen solution (Sigma, St. Louis, MO). The solution was neutralized to pH 7.2–7.6 by 75 μL of 0.1 M NaOH and 300 μL was pipetted onto a glass microscope slide (VWR, West Chester, PA) or the scaffold, coverslipped, and then incubated for 1 h in a CO2 incubator at 37 °C. Specimens were then imaged immediately without fixation.

Confocal Reflectance Imaging: The collagen fibrils were imaged using a confocal microscope (FluoView 1000; Olympus America, Center Valley, PA) in reflectance mode (488 nm) (Figure 1C,D).36 The interval between z-stack images corresponded to the (x,y) plane resolution (i.e, 0.621 μm per pixel for 40X objective and 512 × 512 pixel image). Images were first segmented batch-wise via a custom automated algorithm implemented in Matlab37 (Figure S1, Supporting Information). The algorithm comprised: RGB to grayscale conversion, a median filter, local adaptive thresholding to yield binary images, morphological opening with a 1 pixel sized structural element.

Image Analysis of Collagen Fibril Organization: The morphology of the collagen fibrils was analyzed from confocal micrographs based on set theory for image analysis using custom code written in Matlab and Python.38 Measures were averaged over the entire z-stack (i.e., ∼50–70 images) and over ∼10 pores (or locations for the glass slide). The size (i.e., interfibril distance) distribution of the inter-fibril spaces was computed by opening granulometry with a disk D of diameter d on complementary binary images. Opening granulometry consisted of morphologically opening the set of pixels in each image corresponding to the inter-fibril spaces using a sequence of increasing larger structural elements (i.e., disks D of diameter d). At each value of d, the probability P of one point (i.e., one pixel in the discretized image) belonging to the opened set was calculated (O(d)) (Equation 1) (Figure S2, Supporting Information).39 The opening operation included an erosion (Θ) with D(d) following by a dilation (⊕) with an identical D(d), removing all disconnected groups of pixels with a size less than d.

equation image((1))

The fibril angular orientation distribution was measured using the fast Fourier transform (FFT),40 which yields the intensity angular distribution I(θ). The angle θm for which the intensity was maximum (m) was identified on I(θ) and defined the predominant fibril direction. The FFT-based image analysis used a Matlab program previously used to quantify cellular F-actin filament orientation.20 An orientation index OI(θm) was calculated from I(θ) at θm to yield the percentage of fibrils strictly oriented in the reference angle direction (herein designated as 0° to correspond with the PD direction of the ALH pore) (Equation 2)25.

equation image(2)

Of note, the orientation index used herein25 was distinct from that used in our previous study.20 The entropy (i.e., a measure of disorder) was measured by processing the Hough transform41 using a custom Matlab program. The Shannon entropy ϵ of the probability distribution p(θ) of predominant directions was calculated (Equation 3).25

equation image((3))

Statistical differences were determined by two-tailed student t-tests (Matlab) with a p-value < 0.05 considered significant. Values reported represent mean ± standard deviation for n = 6 samples.

Periodic Finite Element Simulations: A multiscale periodic FE method previously applied to the ALH scaffold itself21 was used herein to predict the elastic stiffnesses of the ALH scaffold-collagen gel composite (Z-set software42). The Young's modulus and Poisson ratio of the PGS forming the scaffold struts were specified as 825 kPa and 0.45, respectively.21 Upper and lower bounds on Young's modulus (linear elastic isotropic assumptions) of a 3 mg mL−1 collagen were specified as 24.3 kPa15 and 5 kPa24 and a Poisson ratio of 0.45 was assumed. The elastic stiffnesses of the ALH scaffold-collagen gel composite in the PD (EPD) and XD (EXD) directions were predicted. An anisotropy ratio r = EPD/EXD was calculated.21 In addition, the Voigt (Equation 4) and Reuss (Equation 5) bounds known as the upper (i.e., the rule-of-mixtures) and lower elastic bounds were determined:

equation image((4))
equation image(5)

Simulations were conducted allowing for anisotropic elastic behavior of the collagen gelled within the ALH pores via an orthotropic material model in which elastic constants were optimized (augmented Lagrangian method) based on the criteria that EPD and EXD equaled values measured previously for adult rat left ventricular myocardium in the respective circumferential and longitudinal directions.20 In the orthotropic material model the collagen gel stiffness in the z (i.e., thickness) direction was assumed to take an intermediate value (i.e., 10 kPa) between upper15 and lower24 bounds.

Supporting Information

Supporting Information is available from the Wiley Online Library or from the author.


Funding for this work was provided by National Institutes of Health (NIH) Grant 1-R01-HL086521-01A2 (PI Dr. Lisa E. Freed, Subaward PI GCE). This funding was made possible by the American Recovery and Reinvestment Act. The authors gratefully acknowledge Dr. Lisa E. Freed (The Charles Stark Draper Laboratories and Massachusetts Institute of Technology, Cambridge, MA) for helpful discussions in preparing the manuscript and Sarah R. Bass, John L. Dzikiy, and Harshal Sawant for their assistance conducting experiments.