Collagen network strengthening following cyclic tensile loading

Abstract

The bulk mechanical properties of tissues are highly tuned to the physiological loads they experience and reflect the hierarchical structure and mechanical properties of their constituent parts. A thorough understanding of the processes involved in tissue adaptation is required to develop multi-scale computational models of tissue remodelling. While extracellular matrix (ECM) remodelling is partly due to the changing cellular metabolic activity, there may also be mechanically directed changes in ECM nano/microscale organization which lead to mechanical tuning. The thermal and enzymatic stability of collagen, which is the principal load-bearing biopolymer in vertebrates, have been shown to be enhanced by force suggesting that collagen has an active role in ECM mechanical properties. Here, we ask how changes in the mechanical properties of a collagen-based material are reflected by alterations in the micro/nanoscale collagen network following cyclic loading. Surprisingly, we observed significantly higher tensile stiffness and ultimate tensile strength, roughly analogous to the effect of work hardening, in the absence of network realignment and alterations to the fibril area fraction. The data suggest that mechanical loading induces stabilizing changes internal to the fibrils themselves or in the fibril–fibril interactions. If such a cell-independent strengthening effect is operational in vivo, then it would be an important consideration in any multiscale computational approach to ECM growth and remodelling.

1. Introduction

The application of mechanical loading to soft connective tissues in vivo can alter the tissue's properties in the long term. Athletic training, for instance, can trigger tendon tissue adaptation [1–3], whereas repetitive overloading and overuse can induce progressive microstructural damage which causes mechanical fatigue [4,5]. An absence of physiological mechanical stimulation leads to atrophy and degradation of tissue mechanical properties [6]. Other than tissue maintenance, mechanical loading is also essential in development. For instance, the removal of intraocular pressure in chick eyes during embryonic development severely reduces eye growth [7]. In an in vitro study, the presence of anchors that allow mechanical contraction by tendon fibroblasts promotes the development of an embryonic-like collagenous matrix [8]. In vivo, the changes in tissue structure and mechanics over time can partly be attributed to cellular metabolic activity in response to dynamic mechanical stimulation, such as synthesis of extracellular matrix (ECM) protein molecules (e.g. collagen [9,10], proteoglycans [11]) and proteinases, such as the MMP family [12] which modify, replace or remove components of the ECM. However, even without the active involvement of cells, the intrinsic mechanical properties of the tissue itself can change with time and history of the applied loads [13].

Cyclic loading can cause dramatic changes in the stiffness and nonlinearity of the stress–strain behaviour of tendon, skin and other soft collagenous tissues [13–18]. The stress response is altered between the loading and unloading step, and also with each subsequent cycle [13]. Repeated cyclic loading causes the stress–strain curves to shift to the right [14,15,19], and the peak force to decrease [14–16]. The extent that preconditioning changes the mechanical properties of soft tissues depends on the tissue system and the applied stress/deformation state [20]. This preconditioning effect stabilizes with each subsequent cycle to a repeatable reference state [13]. The application of cyclic loading may also alter the failure properties of tissue. Remarkably, an increase in the ultimate stress was observed after cyclic loading in tendons [17,18,21], ligaments [17,18] and collagen gels [22]. A study by Legerlotz et al. [23], however, showed a decrease in tendon ultimate stress as a result of cyclic loading. This inconsistency might have been caused by the application of cyclic loading with a high strain magnitude (more than half of the failure strain). The underlying mechanism that drives the tissue strengthening in response to cyclic loading is still not well understood. One possible source is the change in microstructure. Quinn et al. [16] showed small modifications in the fibre direction (approx. 11.4° rotation) after the application of cyclic loading, although this investigation did not report the ultimate stress. In another study by Tower et al. [24], cyclic loading of collagen tissue equivalent was shown to alter the collagen orientation only when the cyclic loading was force-controlled (the specimen was returned to preload in each cycle), whereas collagen orientation was recoverable when cyclic loading was displacement-controlled (the specimen was returned to zero displacement). The dependence of recoverable fibril orientation on whether the specimen was returned to preload or initial length was also observed in tendons in a study by Miller et al. [25], although both papers did not report the effect of preconditioning on the tissue failure properties. Both force-controlled and displacement-controlled cyclic loading showed a preconditioning-induced strengthening phenomenon [17,18,21,22], though, none of the studies measured collagen fibril orientation distribution post-cyclic loading. Therefore, it is unclear whether the tissue strengthening was as a result of fibril realignment. Cheema et al. [22] investigated collagen microstructure and found that tissue strengthening was accompanied by an increase in fibril diameter with increasing numbers of cycles.

Of the ECM proteins, collagen is the most abundant component in connective soft tissues, such as skin [26,27], cornea [28], sclera [29] and tendon [27]. It not only serves as the primary load-bearing structure in the tissue, but it also appears to be mechano-sensitive. The presence of mechanical load can stabilize collagen against enzymatic [30–36] and thermal [37,38] degradation. We hypothesize that the change of tissue mechanical properties from cyclic loading in the absence of cells is caused by direct structural alterations of collagen. Collagen originates in the cell as a nanoscale string that assembles into hierarchical structures which are orders of magnitude longer than the molecules from which they are derived. Thus, the structural organization at different hierarchical scales can affect the bulk mechanics of a network of collagen. We speculate that the structural alteration following mechanical loading includes not only the re-orientation of collagen fibrils in the collagen network towards the direction of load, but also a shift in collagen molecules towards the loaded fibrils.

We approach this investigation by employing a surrogate for natural ECM: a dense, disorganized collagen substrate (DDCS), which consists of a dense isotropic network of nearly pure type I collagen fibrils and permits the isolation of the response of collagen to the applied mechanical stimulation, without the influence of living cells, permanent cross-links and other extracellular components. The single component nature of this substrate is essential in this investigation, because this type of network will allow the direct comparison of the mechanical properties to the microstructural characteristics. Previous studies [39,40] have established that a non-woven fibrous network of non-interacting material, such as polyester felt or polypropylene fabric, possesses a stress–strain curve with similar characteristics to soft tissues: an initial low stiffness caused by the entropic rearrangement of the fibrils, followed by a higher stiffness at higher strain levels caused by the enthalpic deformation (bending and elongation) of the fibrillar material [41]. This type of network also generates a high Poisson's ratio [39] which has also been observed in connective tissue [42] and low concentration collagen gels under uniaxial loading [43]. The mechanical properties of such fibrous networks are regulated by the material properties of each fibril, the nature of the interfibrillar connections and the network architecture [40]. Because molecular mobility is not inhibited within the uncross-linked collagen network in DDCS, unlike the non-interacting nature of polyester or polypropylene, alteration in the mechanical properties of DDCS may be attributed to modification of any of these three factors.

Towards understanding the microstructural source of the adaptive mechanical properties of soft tissue, we first ask whether cyclic mechanical stimulation of a collagen network would induce network remodelling. The aim of this study was to characterize the changes in mechanical properties and network microstructure of the DDCS as a result of cyclic loading. The microstructure after cyclic loading was captured using transmission electron microscopy (TEM) in planes parallel and perpendicular to the loading direction, allowing direct visualization and analysis of the collagen network. Microstructural parameters such as collagen fibril orientation distribution, diameter distribution, area fraction and fibril D-periodicity were measured from the images.

2. Methods

2.1. Fabrication of the dense, disorganized collagen substrate

Pepsin-solubilized bovine collagen type I solution (PureCol, Advanced BioMatrix, Inc., Carlsbad, CA) was mixed with 10X phosphate-buffered saline (PBS) and 0.1 M sodium hydroxide in 8 : 1 : 1 ratio. The collagen concentration in this solution is 2.48 mg ml⁻¹. For optical deformation measurements, 7.5 µl of a 10 µm diameter polystyrene bead suspension (Polysciences, Inc. Warrington, PA) was mixed into the collagen solution. Using a syringe, 9 ml of the solution was transferred into a dialysis cassette (3500 MWCO slide-A-lyzer 3–12 ml model, Thermo Fisher Scientific, Inc., Rockford, IL). The cassette was incubated in a humid 37°C incubator for 3 h to allow for collagen self-assembly, and then immersed in 40% (w/v) 35,000 poly(ethylene) glycol (PEG) for 24 h to dehydrate and compress the collagen material into a thin sheet of randomly oriented fibrils in the en face plane (figure 1). After rinsing the cassette with 1X PBS to remove residual PEG solution, the centre portion of the DDCS sheet was cut into four approximately 5 × 15 mm strips using surgical scissors and sterilized by submerging in 70% ethanol for 30 min, followed by three washes of sterile 1X PBS. For consistency, DDCS specimens were tested within 6 h after sterilization. Long-term storage of in vitro assembled collagen at room temperature and pressure has been shown to significantly increase the ultimate tensile strength [44]. The collagen concentration was estimated from the dimensions of the dialysis cassette and the dehydrated thickness (see method section on Transmission electron microscopy) to be 293 ± 89 mg ml⁻¹.

Figure 1.

Dense, disorganized collagen substrate is a soft-tissue ersatz which was fabricated by dehydration against polyethylene glycol. The resulting product is a dense collagen sheet which is isotropic along the plane parallel to the surface. Scale bar, 1 µm. (Online version in colour.)

2.2. Mechanical testing and sample preparation

Mechanical testing was performed using a previously developed custom mechano-bioreactor [45]. The mechano-bioreactor allowed mechanical testing to be performed under controlled environmental conditions suitable for in vitro cell culture: at 37°C in 1X Dulbecco's modified Eagle's medium (GE Healthcare Life Sciences, Logan, UT) with 10% fetal bovine serum (Atlanta Biologicals, Flowery Branch, GA), 1% penicillin–streptomycin (GE Healthcare Life Sciences) and 0.1% amphotericin-B (Corning Inc., Corning, NY) conditioned with 5% CO₂. The medium was perfused using a syringe pump at a flow rate of 8 µl min⁻¹. Findings from previous studies have shown that testing temperature and solution medium can influence the measured mechanical behaviour. Meghezi et al. [46] measured a lower tangent modulus of collagen gel when the experiment was conducted at 37°C compared with room temperature. The force was measured using a 500 g load cell (Honeywell, Morristown, NJ) at a sampling rate of 1 Hz. The DDCS was mounted between two spring-loaded grips [45], and preloaded to 0.002 N. The reference length and width of the sample were measured, and the specimen was subjected to uniaxial tension cyclic loading. The length and width of the specimen was measured again following cyclic loading, and the specimens were subjected to a final loading to failure.

2.3. Cyclic strain protocol

The DDCS specimen was subjected to a displacement-controlled cyclic loading where each cycle consisted of a ramp loading for 20 min, a hold at maximum displacement for 5 min, a ramp unloading for 20 min and another hold at zero displacement for 5 min (figure 2). We examined two deformation levels: 1 and 2 mm, which corresponded to 13.0 ± 1.1% and 26.7 ± 2.2% strain. The lower strain level was well below the failure strain (at least three times smaller than the failure strain), and it corresponded to the linear region of the stress–strain curve. In the higher strain level group, the DDCS was likely to be stretched to a point where damage is likely occurring. At each deformation level, two groups were tested: one-cycle (control) and 50-cycle groups. Table 1 shows the experimental parameters for each group. The deformation rate for the two strain levels was different in order to obtain a consistent cycle period (50 min). The total length of cyclic loading was 41 h and 40 min. In the case of the one-cycle control groups, the DDCS was stored under the same environmental conditions as the 50-cycle group before starting the single cycle. We chose a low strain rate to achieve an equilibrium response. Following cyclic loading, the specimen was either subjected to a ramp load to failure protocol to determine the mechanical properties or fixed for TEM (table 1). The bioreactor was degassed to remove air bubbles, as needed, by flushing the chamber with a high flow rate. The force reading during the degassing cycle was excluded from the mechanical analysis.

Figure 2.

(a) In one-cycle period, the DDCS was loaded to a set maximum displacement magnitude (1 mm in this case), held at the maximum displacement, unloaded to zero displacement, and held at zero displacement. (b) After 50 cycles, the sample was loaded to failure. (c) In the one-cycle experimental groups, the cycle starts during the last cycle of the 50-cycle group before the sample was loaded to failure.

Table 1.

Experimental parameters for each of the four groups tested.

displacement magnitude	1 mm		2 mm
no. cycles	one	50	one	50
no. samples loaded to failure	6	6	6	6
no. samples for TEM	3	3	3	3
deformation rate during loading/unloading cycle (mm min−1)	0.05		0.10
deformation rate during ramp to failure (mm min−1)	0.05		0.05
cycle period (min)	50		50

2.4. Mechanics data post-processing

During the cyclic loading experiment, a slow force drift up to 0.005 N can appear within 1 day. The force drift was removed by subtracting the force response of the current cycle by the force at zero displacement during the corresponding cycle, where the sample was visibly slack from the plastic deformation due to preconditioning. The engineering stress, σ, was defined as the measured force normalized by the initial cross-sectional area before cyclic loading and strain, ɛ, was defined as the grip displacement divided by the reference length before cyclic loading. The initial sample length was established by applying a 0.002 N preload, which corresponded to a 6 kPa pre-stress. Sample width at the centre point of the sample was measured using the calibrated microscope stage. Average initial sample thickness was measured from cross-sectional sections of three additional samples embedded in resin (see methods section on transmission electron microscopy). The stress-free displacement at each cycle was defined when the force reading equalled the preload of 0.002 N. The slope of the stress–stretch curve after cyclic loading was computed from the stress–strain data using finite difference. Data post-processing was performed in Matlab (MathWorks, Natick, MA).

2.5. Transmission electron microscopy

Three specimens for each experimental group were used for TEM characterization of the collagen fibril structure after cyclic loading. The collagen substrates were cut at the grips and immediately fixed in Karnovsky fixative at 4°C for at least 24 h and then washed with 0.1 M sodium cacodylate buffer (pH 7.2). The central 2 × 3 mm section of the specimen was cut to 0.5 × 1.5 mm strips. The strips were post-fixed with 1% osmium tetroxide in 0.1 M sodium cacodylate buffer for 2 h and washed again with 0.1 M sodium cacodylate buffer before undergoing the ethanol dehydration process. The strips were immersed in solutions with gradually increasing concentrations of ethanol from 35% to 95% for 15 min each and then immersed in 100% ethanol for 1 h with two changes of solution. The dehydrated strips were embedded in squetol resin and cut into 60–90 nm sections (silver–gold interference colour) using an ultramicrotome (LKB 8802A Ultrotome III, Sweden) in two directions, transverse and en face (figure 1). The sections were stained with 5% uranyl acetate and Reynold's lead citrate and imaged with a transmission electron microscope (JEOL JEM 1010 Electron Microscope, Peabody, MA).

The loading direction was tracked carefully when preparing the en face sections. The long axis of the strip always corresponded to the loading direction. Artefacts such as local compression and knife marks can develop during sectioning, and we used these artefacts to identify the loading direction. Polystyrene beads scattered within the DDCS were elliptical in shape due to the local compression during sectioning. The minor axis of the compressed beads aligned with the loading direction [47]. At least five areas were selected in each section for quantitative image analysis.

To produce sections transverse to the loading direction, the strips were carefully positioned on top of a layer of semi-polymerized resin. More resin was added to fully encapsulate the strips, and the assembly was sectioned perpendicular to the loading direction. During sectioning, the sample was positioned, so that the cutting direction aligned with the sample width. For sample thickness measurement, 0.5 µm sections of all samples were collected and then dried onto microscope coverslips. It should be noted that the fixation, dehydration and embedding process can cause sample shrinkage [48,49]. In this study, we did not adjust the thickness measurements for this shrinkage, because the extent of shrinkage for this particular material has not been measured.

2.6. Transmission electron microscope image post-processing

2.6.1. Area fraction

The greyscale TEM image of the transverse section was filtered using the adaptive Wiener two-dimensional noise removal filter and then converted into a binary image using Otsu's thresholding method (figure 3). Area fraction of the binary image was computed by normalizing the pixels occupied by the fibrils by the image size. For each specimen, we identified three image areas (4.4 × 4.4 µm²) that were free of beads and edges for image analysis.

Figure 3.

The sequence of image processing to obtain a binary image.

2.6.2. Fibril diameter distribution

The binary images of the transverse section were used to determine the fibril diameter distribution. For each fibril, the area as well as minor axis and major axis diameters were determined (MATLAB function regionprops). The minor axis of each fibril was used to represent the actual fibril diameter [50,51]. Fibrils oriented within the plane of the section were excluded from the analysis. Other sources of error such as noise (minor diameter less than 25 nm) and large objects owing to unresolvable fibril boundaries (minor diameter larger than 500 nm, and compactness lower than 0.8) were also eliminated from the analysis. At least 500 fibril diameters were measured for each specimen.

2.6.3. Fibril orientation distribution

The Fourier transform method detailed in Sander & Barocas [52] was used to estimate the collagen fibril orientation distribution from the en face TEM images of the substrate. Briefly, the greyscale image was transformed into the frequency domain using a two-dimensional discrete Fourier transform. After applying a bandpass filter to remove high and low frequencies corresponding to potential image noise and uneven illumination, respectively, the frequency domain was transformed into polar coordinates to give the orientation distribution as a function of orientation angle. For each specimen prepared for TEM, the fibril orientation distributions were obtained from at least five images (each image corresponding to a DDCS area measuring 4.4 × 4.4 µm²). These distributions were pooled together to provide an estimate of the probability density function for the entire specimen. The mean, standard deviation and variance of the orientation angle was then estimated using circular statistics for bimodal distributions [52,53]. The mean angle of the distribution was defined as, where, and Here, f(θ_i) is the normalized frequency of the distribution. The standard deviation was defined using the equation, where is the dispersion [53]. The dispersion is related to circular variance, V = 1 − R. In circular statistics, V ranges from 0 to 1. When V is close to 0, the distribution is purely isotropic, and when V is close to 1, the distribution is highly anisotropic.

2.6.4. Fibril periodicity

For each specimen, five fibrils with at least 500 nm of straight section where the banding period was visible were chosen in TEM images of the en face section. The greyscale pixel values on the fibril length were averaged over 20 rows. The one-dimensional averaged spatial signal was converted into the frequency domain using discrete Fourier transform, and the power spectral density plot was obtained. The fibril banding periodicity was identified by choosing the highest peak between the 60 and 80 nm period in the power spectral density. All image processing was performed in MATLAB.

2.7. Statistical analysis

Tukey–Kramer post hoc analysis of variance with a significance level of 5% was used to determine statistical significance. All statistical analyses were performed in MATLAB.

3. Results

3.1. Tissue geometry

An increase in the stress-free length was observed after cyclic loading in all samples. The extent of lengthening was significantly higher in the 50-cycle groups compared with the corresponding one-cycle controls (p < 0.001; table 2). The decrease in width after 50 cycles was significantly different from the one-cycle control for both displacement magnitudes (p < 0.001), but no difference was observed in the one-cycle control groups between the two cyclic strain levels. Sample thickness for all groups increased significantly compared with the initial sample thickness from 49.53 ± 11.65 (n = 3) to 88.64 ± 12.95 µm (n = 12), but it was not significantly different between groups. There was no statistically significant difference in sample thickness after cyclic loading between the 50-cycle and control groups (table 3). However, the thickness of all groups was significantly higher compared with the initial sample thickness (49.53 ± 11.65 µm, n = 3). This swelling was not caused by the application of loading, because the thickness of unloaded DDCS specimens stored under the same environmental conditions as the experimental groups for the duration of the cyclic loading (84.94 ± 7.06 µm, n = 3) was not statistically different compared with the cyclic loading groups.

Table 2.

DDCS length and width with the corresponding normalized experimental parameters.

normalized experimental parameters
displacement magnitude (mm)	1		2
strain magnitude (%)	13.0 ± 1.1		26.7 ± 2.2
strain rate during loading/unloading cycle	0.7 ± 0.1% per min		1.3 ± 0.1% per min
strain rate during ramp to failure	0.7 ± 0.1% per min		0.7 ± 0.1% per min
no. cycles	one (control)	50	one (control)	50
no. samples	6	6	6	6
DDCS length
initial length (mm)	7.53 ± 0.66	7.78 ± 0.65	7.46 ± 0.60	7.62 ± 0.64
stress-free length after cyclic loading (mm)	7.80 ± 0.65	8.51 ± 0.66	8.29 ± 0.63	9.19 ± 0.64
% increase in length after cyclic loading	3.6 ± 0.6	9.4 ± 0.8	11. 2 ± 1.1	20.6 ± 1.8
DDCS width
initial width (mm)	4.83 ± 0.23	4.99 ± 0.19	4.94 ± 0.58	4.93 ± 0.31
width after cyclic loading (mm)	4.81 ± 0.24	4.60 ± 0.20	4.76 ± 0.57	4.05 ± 0.38
% decrease in width after cyclic loading	0.5 ± 0.3	7.7 ± 3.2	3.8 ± 1.7	18.0 ± 3.9

Table 3.

Collagen fibril periodic banding (average ± s.d.). No statistically significant difference was observed.

strain magnitude	13.0% (nm)	26.7% (nm)
after one cycle (control)	65.7 ± 2.4	67.2 ± 1.4
after 50 cycles	68.8 ± 4.3	66.6 ± 1.0

3.2. Cyclic loading response

During cyclic loading, the stress–strain curve shifted to the right with each cycle, as expected. The most substantial decrease in the stress amplitude of the cyclic response (37.13% and 51.2%, for the low and high strain levels, respectively) occurred during the first five cycles (figure 4). The stress amplitude decreased to a similar steady-state value, 0.065 ± 0.012 and 0.072 ± 0.011 MPa, for the two different applied strain amplitudes (figure 4). The degree of stress relaxation during the hold at the maximum displacement also decreased with each cycle and the stress-free length of the sample increased gradually during cyclic loading, and the largest increase in length occurred during the first few cycles (figure 5). These observations were consistent with preconditioning effects observed for uniaxial tension loading in the literature [13,14,54].

Figure 4.

(a,b) The peak stress in each cycle is the value of maximum stress achieved in the loading step of a cycle, which decreased substantially within the first few cycles, and plateaued to a constant stress value for both strain levels (solid lines indicate average values and the corresponding shading indicates the standard deviation). (Online version in colour.)

Figure 5.

The strain at preload substantially increased within the first few cycles (solid lines indicate average values and the corresponding shading indicates the standard deviation). (Online version in colour.)

3.3. Mechanical properties of the dense, disorganized collagen substrate after cyclic loading

The mechanical response of DDCS was significantly altered by cyclic loading but less so by the amplitude of the imposed deformation (figure 6). Significant plastic lengthening in the 50-cycle groups drove the stress–stretch curve shift to the right. Remarkably, the 50-cycle groups demonstrated significantly higher ultimate stresses and steeper slopes compared with the single cycle control groups (figures 6 and 7). However, the ultimate stress was reached at a lower stretch value (figure 6).

Figure 6.

(a) DDCS stress–strain curve (average ± s.d.) after cyclic loading and the corresponding (b) derivative of the curve. Stress and strain were normalized against initial DDCS geometry before cyclic loading. (Online version in colour.)

Figure 7.

Ultimate tensile strength of the DDCS after cyclic loading. *p < 0.01.

3.4. Tissue microstructure after cyclic loading

Circular variance, which provides a measure of fibril alignment, was not significantly different between the 50-cycle and the one-cycle groups for 13.02% strain magnitude (figure 8). In contrast, 26.68% strain magnitude resulted in a significantly lower variance in the distribution after 50 cycles (figure 8). A mean angle of 15.8 ± 11.1° from the loading direction was calculated for the 50-cycle group, which indicates that collagen fibrils reoriented towards the direction of loading. Despite the significant differences in sample geometry after the application of cyclic loading in different groups, there was no statistically significant difference in the area fraction of the cross section (figure 9), or in the periodic banding between groups (table 3). There was no significant difference in the mean fibril diameter between the single cycle and 50-cycle groups (table 4).

Figure 8.

Representative TEM images of the en face section (red arrow represents the loading direction and scale bar, 5 µm) with the fibril orientation distribution polar plots and summary of circular variance of the orientation distribution after cyclic loading. *p < 0.02. (Online version in colour.)

Figure 9.

Representative TEM images of the transverse section (scale bar, 1 µm) and summary of area fraction of the DDCS cross section after cyclic loading. No statistically significant difference was observed.

Table 4.

Mean fibril diameter (average ± s.d., n = 3 for each group). No statistically significant difference was observed.

strain magnitude	13.0% (nm)	26.7% (nm)
after one cycle (control)	78.1 ± 3.4	77.0 ± 7.4
after 50 cycles	82.9 ± 3.3	75.6 ± 4.5

4. Discussion

Previous studies have shown that the mechanical properties of connective soft tissues are time and loading history-dependent. However, the mechanism that generates this behaviour is still not well understood. This study measured the effect of cyclic loading at two different strain amplitudes on the mechanical properties, as well as the microstructure, of a dense collagenous substrate. In agreement with the preconditioning effects widely observed in biological tissues and engineered collagen constructs [13–16], the DDCS also showed the shift of the stress–strain curve and decrease in peak force during cyclic loading. In addition, the DDCS also showed the preconditioning-induced strengthening through a significant increase in ultimate strength, similar to that observed in tendons [17,18,21], ligaments [17,18] and collagen gel [22]. The increase in ultimate strength occurred at both strain levels. As discussed in the Methods section, while the lower strain level was selected because it resides within the linear region and well below failure, the higher strain level was chosen because it represents the point where damage is likely to start occurring. Our results are inconsistent with the finding in [23], where the ultimate strength of tendon tissue decreased as a result of preconditioning with a strain magnitude close to the failure strain. This difference might be due to the uncross-linked nature of the DDCS, where molecular mobility is less constrained compared with the cross-linked collagen in native tendon.

Cyclic loading also altered the specimens' geometry: the stress-free length was increased and the width was decreased. Increase in stress-free length from cyclic loading has been observed in other studies involving native soft tissue, but the extent of lengthening can be exacerbated by the removal of proteoglycan [14] or the addition of an isometric hold during cyclic loading [55]. Despite this substantial change in sample geometry, cyclic loading did not significantly change the cross-sectional area fraction and fibril orientation only changed significantly in the higher strain level group. These observations suggest that the increase in the tangent modulus and strength in both cycled samples was not attributed to fibril crowding or realignment with the loading direction. The lack of reorientation observed at the lower strain level is likely, because each cycle in our experiment allowed the specimen to return to zero displacement. Tendon and tissue equivalents have been shown to recover their fibril orientation distribution after preconditioning and stress relaxation if the tissue was unloaded to zero displacement, instead of preload [24,25] with a recovery time as short as 60 s [25].

In contrast to Cheema et al., we found no effect of cyclic loading on the fibril diameter distribution. Cheema et al. [22] reports fibril diameter increases, possibly due to fusion of multiple smaller fibrils. The DDCS used in our investigation has undergone tremendous compression during fabrication when it was dehydrated by dialysing against 40% polyethylene glycol for 24 h, possibly pre-fusing the fibrils.

Other than the fibril spatial distribution and diameter, two other factors that determine the mechanical properties of entangled fibrillar network include the degree of interfibrillar bonding or friction and the mechanical properties of the fibrils [40]. The interfibrillar contact rupture and reconnection was proposed as a possible mechanism for the increase in ultimate shear stress following cyclic shear loading observed in the actin network with cross-linking proteins [56]. Even though this study did not include a cross-linking agent in the medium during cyclic loading, it is unclear whether the strength or quantity of interfibrillar connections in the DDCS was altered by cyclic loading. The change in fibril mechanical properties into a more brittle-like material with a higher fracture stress was demonstrated in a molecular model proposed by Buehler [57] by an increase in adhesion between collagen molecules. Future investigation of the effect of cyclic loading on interfibrillar and intrafibrillar bonding can provide a more definite theory regarding the mechanism involved in changing the mechanical properties of soft connective tissues.

Biological tissues are viscoelastic materials. The rate of strain in soft tissues has been found to increase the tangent modulus when the strain rate was increased [54,58–66], which can generate a more severe stress relaxation response as a higher peak stress is reached at the beginning of the stress relaxation period [55]. The low displacement rates for this investigation were deliberately selected to reduce the viscoelastic effects of the DDCS. In addition, because our principal findings compare material properties measured on different samples using the same method, we do not expect small viscoelastic effects to influence the results appreciably. Failure to permit an adequate viscoelastic recovery time between testing events can lead to apparent increases in the zero-load strain and apparent increases in the tangent modulus. Previous investigations in the literature have used a range of recovery times, from the order of hundreds of seconds [14,67] to hours [23] to days [68]. In this investigation, a 5-min hold at zero displacement was prescribed between events. However, because the specimens unloaded to a stress-free state (force is lower than preload) well before they reached zero displacement, the recovery time was extended. In the next cycle, the specimen also did not reach the preload immediately. As a result, the effective period where the specimen was stress-free was much longer than the 5-min period of zero displacement. Between the 49th and 50th cycle, the stress-free period can be as long as 34 ± 3 and 38 ± 1 min, for the lower and higher strain levels, respectively. As a comparison, the time constant of the 50th cycle's stress relaxation was 1.8 ± 0.3 min, when the data were curve fitted to the Maxwell–Weichert model [69,70]. Because the recovery time was approximately 20 times the stress relaxation time constant, the material had adequate time to recover prior to the next event. Given our experimental approach, we do not expect viscoelastic material behaviour to significantly affect our results.

This study presents the mechanical and microstructural analysis of dense collagenous substrates after cyclic loading. TEM imaging provides direct visualization of the substrate's microstructure. Results of this study suggest that the mechanical strengthening from cyclic loading of uncross-linked collagenous substrates did not arise only from collagen fibril alignment towards the direction of load. This suggests that mechanical loading induces stabilizing changes internal to the fibrils themselves or in the fibril–fibril interactions. Resolving the source of this phenomenon is an important step towards understanding how biological tissues are maintained in vivo.

Data accessibility

Data available from the Dryad Digital Repository: doi:10.5061/dryad.tr2f1.

Competing interests

We declare we have no competing interests.

Funding

T.D.N. acknowledges funding from NSF Career Award CMMI: 1253453. E.A.S. acknowledges funding from NSF Career Award CMMI: 1452728.