Mechanical regulation of fibroblast migration and collagen remodelling in healing myocardial infarcts

Abstract

Effective management of healing and remodelling after myocardial infarction is an important problem in modern cardiology practice. We have recently shown that the level of infarct anisotropy is a critical determinant of heart function following a large anterior infarction, which suggests that therapeutic gains may be realized by controlling infarct anisotropy. However, factors regulating infarct anisotropy are not well understood. Mechanical, structural and chemical guidance cues have all been shown to regulate alignment of fibroblasts and collagen in vitro, and prior studies have proposed that each of these cues could regulate anisotropy of infarct scar tissue, but understanding of fibroblast behaviour in the complex environment of a healing infarct is lacking. We developed an agent-based model of infarct healing that accounted for the combined influence of these cues on fibroblast alignment, collagen deposition and collagen remodelling. We pooled published experimental data from several sources in order to determine parameter values, then used the model to test the importance of each cue for predicting collagen alignment measurements from a set of recent cryoinfarction experiments. We found that although chemokine gradients and pre-existing matrix structures had important effects on collagen organization, a response of fibroblasts to mechanical cues was critical for correctly predicting collagen alignment in infarct scar. Many proposed therapies for myocardial infarction, such as injection of cells or polymers, alter the mechanics of the infarct region. Our modelling results suggest that such therapies could change the anisotropy of the healing infarct, which could have important functional consequences. This model is therefore a potentially important tool for predicting how such interventions change healing outcomes.

Figure 1. Overview of the agent-based model of infarct healing

A, histological section of a cryoinfarct scar after 3 weeks of healing, showing picrosirius red stained collagen fibres (dark tissue). B, example of a model result simulating the cryoinfarction experiment. Greyscale indicates collagen density and dashes indicate collagen fibre orientation. C, example of fibroblast agents that have infiltrated the infarct region by migrating and proliferating. D, example of local fibre orientation histogram. Each dash depicted in B indicates the mean angle and mean vector length (strength of alignment) of the local fibre orientation distribution.

Figure 2. Fibroblast decision making flowchart

At every time step, every fibroblast executes these operations and decisions.

Figure 3. Method of determining fibroblast orientation

A, each guidance cue was represented as a vector. The vectors were normalized, weighted and averaged in order to compute a resultant vector representing the combined influence of all of the cues on the fibroblast orientation. B, the magnitude and orientation of the resultant vector defined the shape of a wrapped normal distribution. The new fibroblast orientation (indicated by the grey bar in the histogram) was chosen by taking a random sample from a set of angles representing this distribution.

Figure 4. Time course of infarct healing

Chemokine generated within the infarct stimulated fibroblast migration and proliferation (left), and upregulated collagen deposition (right). The illustrated time points are 4 (top), 7 (middle), and 10 (bottom) days after infarction.

Figure 5. Fit of model to published fibroblast migration behaviour in collagen gels

Dickinson et al. (1994) measured random migration coefficients for x-directed (Expt Ux) and y-directed (Expt Uy) motion of fibroblasts in collagen gels with unaligned or aligned collagen fibres. We obtained estimates of the model parameters W_p, W_s, and α by fitting simulated migration behaviour (Model Ux, Model Uy) to the published data.

Figure 6. Fit of model to published infarct collagen accumulation time course

Fomovsky & Holmes (2010) measured collagen area fraction at several time points after creating ligation infarcts in rats (Experiment). We obtained estimates of the model parameters k_cf,gen and k_cf,deg by fitting a reaction ODE to the published data. Using the fitted parameter values, the agent-based model was able to reproduce the data (Model).

Figure 7. Simulations of circular and elliptical infarcts (2:1 aspect ratio) experiencing uniaxial strain with initial matrix alignment as found in native myocardium

A, the model predicted little difference in the strength of collagen alignment (quantified as the mean vector length of the average collagen fibre distribution) of a longitudinally oriented elliptical infarct (LE), circular infarct (C), and a circumferentially oriented elliptical infarct (CE). This was consistent with experimental data from our cryoinfarction study (Fomovsky et al. 2012). B, by increasing the aspect ratio of the elliptical infarcts to 10:1 or increasing the chemical cue weight factor W_c fourfold (Chem High), the model predicted a measurable effect of the orientation of the elliptical infarcts on the strength of collagen alignment.

Figure 8. Simulations of circular infarcts experiencing uniaxial strain with different initial matrix alignment or density

A, the model predicted that infarcts with initial matrix structure that was either unaligned or perfectly aligned would have different collagen structures at early times, but gradually converge to the same degree of collagen alignment (Aligned vs Unaligned). Similarly, infarcts with initial matrix density that was either equal to or 1/10th the density of normal myocardium had different collagen alignment in the first week and then gradually converged thereafter (Native vs. Low). B, by 3 weeks, regardless of the strength of alignment of the initial matrix structure, the predicted collagen fibre orientation histograms fell within 1 standard deviation of the experimental mean for circular, mid-ventricular cryoinfarcts (Fomovsky et al. 2012).

Figure 9. Simulations of circular infarcts experiencing uniaxial strain with mean fibre orientations of the initial matrix varying from −60 deg at the epicardial surface (0% depth) to +60 deg at the endocardial surface (100% depth) to simulate a series of transmural layers in a healing infarct

A, the model predicted that mean collagen fibre angles gradually converged on the circumferential direction. C, the model (with collagen turnover rates derived from rat ligation data) therefore predicted a flatter transmural fibre angle profile in 3-week-old infarcts than in healthy myocardium (Model Infarct Rat vs. Model Native Rat). E, a similar trend was present in transmural fibre angle profiles measured before and 3 weeks after infarction in pigs (Expt Native Pig vs. Expt Infarct Pig) (Holmes, 1995; Holmes et al. 1997). B, the model suggested that mean vector length would converge more slowly than mean fibre angle. D, after 3 weeks of simulated healing, the model predicted strongest alignment (highest mean vector length) at mid-depth, where the direction of stretch and the initial matrix orientation were identical, and weakest alignment at the surfaces, where initial matrix orientation differed most from the direction of stretch (Model Infarct Rat). F, a similar transmural trend in mean vector length was reported in pig infarcts 3 weeks after ligation (Expt Infarct Pig).

Figure 10. Simulations of circular infarcts with uniaxial or equibiaxial strains and with initial matrix alignment as found in native myocardium

A, the model predicted that long term alignment of collagen in a healing infarct required a uniaxial mechanical cue, which was consistent with experimental data from our cryoinfarction study (Expt Uniax vs. Expt Biax). B and C, at 3 weeks, with mechanosensing turned on (Mech On), the model predicted collagen fibre orientation histograms similar to those measured in cryoinfarcts experiencing biaxial or uniaxial strain. With mechanosensing turned off (Mech Off), the aligned collagen structure for the case of uniaxial strain could not be predicted.