betaFeedback

WorksAgricultural and Biological Sciences

A discrete-to-continuum model for the human cornea with application to keratoconus

Abstract

We introduce a discrete mathematical model for the mechanical behaviour of a planar slice of human corneal tissue, in equilibrium under the action of physiological intraocular pressure (IOP). The model considers a regular (two-dimensional) network of structural elements mimicking a discrete number of parallel collagen lamellae connected by proteoglycan-based chemical bonds (crosslinks). Since the thickness of each collagen lamella is small compared to the overall corneal thickness, we upscale the discrete force balance into a continuum system of partial differential equations and deduce the corresponding macroscopic stress tensor and strain energy function for the micro-structured corneal tissue. We demonstrate that, for physiological values of the IOP, the predictions of the discrete model converge to those of the continuum model. We use the continuum model to simulate the progression of the degenerative disease known as keratoconus, characterized by a localized bulging of the corneal shell. We assign a spatial distribution of damage (i. e., reduction of the stiffness) to the mechanical properties of the structural elements and predict the resulting macroscopic shape of the cornea, showing that a large reduction in the element stiffness results in substantial corneal thinning and a significant increase in the curvature of both the anterior and posterior surfaces.

Introduction

The cornea is the external lens of the eye, with specific mechanical and optical functions. The cornea confines and protects the anterior chamber, and it refracts the light rays supplying about two thirds of the total refractive power of the eye. Structurally, the cornea is a layered shell, where each layer is approximately uniformly curved, being pressurized on the posterior surface by the intraocular pressure (IOP) due to the presence of ocular fluids. The cornea comprises five main layers; from the anterior surface, the layers are the epithelium, the Bowman membrane, the stroma, the Descemet membrane, and the endothelium. The stroma is the thickest layer, playing the main structural role, and it is composed of a network of approximately equidistant-equidiameter collagen fibrils immersed in a matrix of proteoglycans, responsible for the formation of chemical bonds (crosslinks). The stroma is the core of our modelling study.

Layers of collagen fibrils, organized into ribbon-like lamellae, are interwoven in a complicated pattern to give the cornea structural integrity [18, 17]. Average mechanical properties of human corneal tissue can be estimated ex vivo using simple experimental tests [6] [36], but more recent work has highlighted how material properties vary through the tissue thickness [27, 40, 38]. Aiming to understand how the organization of the constituents of the cornea affects the overall mechanical response, theoretical and numerical approaches have been used to introduce this information in constitutive models [7, 22, 34, 26, 39]. In general, the human cornea is modelled as nearly-incompressible hyperelastic material with highly nonlinear behaviour. The strain energy function is decomposed into an isotropic contribution, modelling the proteoglycan ground matrix (e.g. neo-Hookean or Mooney-Rivlin materials), and an anisotropic contribution, incorporating nonlinear strain-stiffening effects, describing preferential orientations and spatial dispersion of the collagen fibrils [11, 20, 16]. However, continuum models are rather limited in their ability to incorporate local changes in the material properties and are thus unable to fully capture the localized degeneration of the various stromal components [23]. Moreover, as all non-collagenous components of the cornea are accounted for as a single (isotropic) contribution to the strain energy, the mechanical role of the crosslinked network of proteoglycans is not explicitly modelled.

Keratoconus is a degenerative disease of the eye characterized by corneal thinning and uneven protrusion of the corneal tissue, which can lead to a loss of vision. Despite the availability of several clinical treatments (e.g. corneal transplant [29], lamellar keratoplasty [19] and crosslinking with riboflavin [3] [31]), the aetiology of keratoconus is not yet fully understood. Multiple irreversible changes in the organization of the collagen architecture and in the chemical composition have been shown to accompany the progression of the disease [2] [12, 30, 24]. Firstly, keratoconus is associated with a reduction in the number of collagen lamellae through the corneal thickness [25, 28]. Secondly, in keratoconus the collagen fibril structure has been observed to become disordered, contrary to the high level of fibril organization typical of a healthy cornea [8]. Thirdly, these changes in the collagen composition within the cornea are also mirrored by changes in the interconnecting proteoglycan bonds. For example, keratoconus is associated with a reduction in the density of keratan sulfate proteoglycans compared to a healthy cornea [41, 32] and a corresponding increase in dermatan sulfate molecules which are comparatively softer [10]. Several distinct perturbations (including chemical, genetic, and mechanical) have been investigated (alone or in combination) as potential causes of the disorder [24], but the exact causal relationships between these changes and their impact on mechanical, geometrical, and optical properties of the cornea remain unclear.

Theoretical models of corneal mechanics have been adapted to model the progression of keratoconus, mimicking the underlying weakening of the tissue via a localized reduction of the material stiffness. While existing continuum models have successfully predicted small modifications of the initial (nearly spherical) shape of the cornea, models have so far been unable to predict large scale conical deformations because the underlying models cannot systematically capture microstructural changes in the disease [23, 33]. Recent discrete models have managed to overcome this barrier, but are so far limited to two collagen layers through the corneal thickness and so cannot reliably represent the stromal microstructure [21, 24].

In this study we propose an extension of these discrete models which includes a large number of layers of collagen lamellae, from which we can systematically derive a continuum model capable of predicting the large scale cornea deformations evident in keratoconus. Specifically, we propose a discrete, two-dimensional, model of the human cornea which allows us to describe progressive changes in the mechanical properties, leading to a modification of the corneal shape. The model includes an explicit representation of the collagen fibrils and of the crosslink microstructure. For simplicity, here we restrict our attention to a meridian slice of the corneal stroma of fixed depth. The strip is fixed at both ends to the limbus, as a segment of a cylindrical annulus in plane-strain configuration, loaded with the IOP on the posterior surface. Aiming at a rational continuum description of the stroma, we use discrete-to-continuum upscaling to obtain the corresponding strain-energy density [4] [13], which will provide the stress and the local stiffness of the tissue. We use this framework to model the onset and progression of keratoconus, by imposing a reduction in the stiffness of the stromal components localized at the central portion of the cornea, and compute the resulting displacement field and the corresponding stress and strain distributions.

The paper is organized as follows. In § 2, we propose a new discrete model for the mechanical behaviour of a slice of human corneal tissue. In § 3 we use discrete-to-continuum analysis to derive a corresponding macroscale continuum description of this tissue slice. In § 4 we compare the outcomes of discrete and continuum models, and then apply the continuum model to investigate response of the tissue to a prescribed reduction in stiffness of both the collagen lamellae and the proteoglycan matrix, as a simple model for the formation of keratoconus. The model is critically discussed in § 5.

Discrete model

We consider a two-dimensional model of the human cornea, consisting of a thin meridian slice of fixed depth D~\widetilde{D} and uniform thickness T~\widetilde{T}. We assume that the unloaded anterior and posterior surfaces of the cornea are concentric circular arcs, spanned by an angle 2Φ∗2\Phi^{*}. We introduce a planar coordinate system with origin at the centre of the circular arcs, parameterising the domain with two-dimensional Cartesian coordinates (X,Y)(X,Y), such that the XX-axis cuts the midpoint of the two circular arcs, see Figure 1. In the following, dimensional variables are denoted with the tilde, while dimensionless variables are plain.

The unloaded geometry

Figure 1. The unloaded geometry. Zoom onto the macroscale geometry of an idealized two-dimensional corneal slice (created with BioRender.com). The unloaded-configuration cornea is shown in blue, the key parameters are indicated in red and the green dotted lines (circles) depict the curves of constant Φ\Phi (R~\widetilde{R}). The unloaded configuration is then discretized (N=4N=4, M=80M=80, γ=20\gamma=20). Panel on the right presents a zoom onto the unit cell with the dimensional lengths of the elements (denoted with tilde) dependent on the radial position (index ii). Terms in red represent corresponding quantities in the continuum limit (N→∞N\to\infty) of the dimensionless model. Note that while the unit cell for a finite NN and MM forms an isosceles trapezoid, in the continuum limit this becomes a rectangle.

The geometry of the model is described by three parameters: the curvature radius of the anterior surface R~A=7.8\widetilde{R}_{A}=7.8 mm, the uniform thickness T~=0.62\widetilde{T}=0.62 mm, and the anterior in-plane diameter D~A=11.46\widetilde{D}_{A}=11.46 mm [23]. The in-plane diameter is related to the aperture angle 2Φ∗2\Phi^{*} as D~A=2R~Asin⁡Φ∗\widetilde{D}_{A}=2\widetilde{R}_{A}\sin\Phi^{*}, and it follows that Φ∗≈0.83\Phi^{*}\approx0.83 rad.

We note that in plane strain configuration the depth is set to one unit of length, in this case D~=1\widetilde{D}=1 mm.

Corneal microstructure

In the discrete approach we model the cornea as a set of N+1N+1 concentric arcs, each representing an individual collagen lamella, separated by the distance L~(2)\widetilde{L}^{(2)}

L~(2)=T~/N.(1)\widetilde{L}^{(2)}=\widetilde{T}/N. \tag*{(1)}

The radius of the ithi^{\text{th}} circular arc is

R~i=R~A−T~+iL~(2),(i=0,…,N),(2)\widetilde{R}_{i}=\widetilde{R}_{A}-\widetilde{T}+i\widetilde{L}^{(2)},\qquad(i=0,\ldots,N), \tag*{(2)}

thus, the radius of the posterior surface is R~0=R~A−T~\tilde{R}_0=\tilde{R}_A-\tilde{T} and the radius of the anterior surface is R~N=R~A\tilde{R}_N=\tilde{R}_A, see Figure 1. Each arc is subdivided into MM (even) equally spaced arcs, each sweeping an angle ϕ∗=2Φ∗/M\phi^*=2\Phi^*/M, so that

Φj=(j−M/2)ϕ∗,(j=0,…,M),(3)\Phi_j=(j-M/2)\phi^*,\qquad(j=0,\ldots,M), \tag*{(3)}

and Φ0=−Φ∗\Phi_0=-\Phi^* and ΦM=Φ∗\Phi_M=\Phi^* identify the boundaries. The intersections between adjacent segments are defined by a set of planar points of coordinates

X~i,j=R~icos⁡Φj,Y~i,j=R~isin⁡Φj(i=0,…,N,j=0,…,M).(4)\tilde{X}_{i,j}=\tilde{R}_i\cos\Phi_j,\qquad\tilde{Y}_{i,j}=\tilde{R}_i\sin\Phi_j\qquad(i=0,\ldots,N,\quad j=0,\ldots,M). \tag*{(4)}

The resulting discretization size is defined by NN and MM, and the mesh aspect ratio γ=M/N\gamma=M/N is taken as a fixed parameter. In the spirit of truss models, two adjacent nodes are connected by a straight segment, thus each collagen lamella is a piecewise linear approximation to a circular arc. We denote the lamellar segments with the superscript (1){}^{(1)}. The geometric/mechanical properties of the lamellar segment between the angles Φj\Phi_j and Φj+1\Phi_{j+1} are denoted by the subscript i,j+1/2i,j+1/2. Hence, the current length of each segment is l~i,j+1/2(1)\tilde{l}^{(1)}_{i,j+1/2}, while the initial length L~i(1)\tilde{L}^{(1)}_i is independent of jj and proportional to the arc radius through the relation

L~i(1)=2R~isin⁡(Φ∗M).(5)\tilde{L}^{(1)}_i=2\tilde{R}_i\sin\left(\frac{\Phi^*}{M}\right). \tag*{(5)}

In the human cornea, the collagen lamellae have an average thickness of about 2μm2\mu\mathrm{m}, and the number of lamellae across the corneal thickness varies between 200 and 500 [17]. In our model, we assume that each circular arc is representative of the behavior of several parallel lamellae. For computational reasons we limit our model to a maximum number of arcs N=64N=64. The lamellae are connected to one another via a dense network of chemical bonds (crosslinks) originated by the proteoglycans.

The hydrated ground matrix of proteoglycans compressed by the IOP transmits the load to the collagen structure [1]. We model the function of the matrix by introducing a set of radial struts, denoted with the superscript (2){}^{(2)}. The current length of the radial struts spanning between the nodes ii and i+1i+1 at the polar angle Φj\Phi_j is denoted as l~i+1/2,j(2)\tilde{l}^{(2)}_{i+1/2,j}. The corresponding initial length, L~(2)\tilde{L}^{(2)}, Eq. (1), is uniform.

Tensile stresses resulting from the action of IOP are primarily carried by collagen lamellae, whose mechanical stability is partially provided by extracellular matrix components. The sliding between lamellae is contrasted by the presence of proteoglycans which create crosslinks between collagen fibrils [10]. Elastin fibres, covered with a sheath of microfibrils, may also contribute to contrast the lamellar sliding [1] [37, 38].

We mimic the combined action of these elements in providing shear stiffness to the structure by introducing diagonal struts, with geometrical and mechanical properties denoted with the superscript (3){}^{(3)}. The node (i,j)(i,j) is connected diagonally to the nodes of the two adjacent layers, i.e., points i+1,j±1i+1,j\pm1 and i−1,j±1i-1,j\pm1. The current length of the diagonal struts is denoted with l~i±1/2,j±1/2(3)\tilde{l}^{(3)}_{i\pm1/2,j\pm1/2}. The corresponding initial lengths

L~i±1/2(3)=L~i(1)L~i±1(1)+(L~(2))2,(6)\tilde{L}^{(3)}_{i\pm1/2}=\sqrt{\tilde{L}^{(1)}_i\tilde{L}^{(1)}_{i\pm1}+\left(\tilde{L}^{(2)}\right)^2}, \tag*{(6)}

are independent of jj and increase from the posterior to the anterior layer, proportionally to the radius R~i\tilde{R}_i of the arc.

We choose an aspect ratio γ=M/N\gamma=M/N that guarantees a marked inclination of the diagonal struts, so that the corresponding axial forces contribute to contrast the sliding between the lamellae and avoid zero energy deformation modes.

The discrete balance equations are stated under the assumption of finite kinematics, therefore the problem can lead to multiple solutions. The unknowns of the problem, thus, are either the current coordinates x~,y~\tilde{x},\tilde{y} of the nodes or the node displacements u~X,u~Y\tilde{u}_X,\tilde{u}_Y.

Constitutive models

We assume that all the elements are stress-free in the unloaded configuration, disregarding the presence of pre-stresses, typical in arteries [5]. For simplicity, we assume that the force in each structural element is modelled as a linear function of its elongation, i.e., the difference between the current length l~(m)\tilde{l}^{(m)} and the initial length

ComponentElastic modulus [kPa]Stiffness [N/mm]
Lamellar segmentsE~(1)\tilde{E}^{(1)}7656K~(1)\tilde{K}^{(1)}7.2
Radial crosslinksE~(2)\tilde{E}^{(2)}7.656K~(2)\tilde{K}^{(2)}0.0072
Diagonal crosslinksE~(3)\tilde{E}^{(3)}535.92K~(3)\tilde{K}^{(3)}0.504

caption: tbl:2: Material parameters used in simulations.

L~(m)\tilde{L}^{(m)}, m=1,2,3m = 1, 2, 3. Denoting the stretch of a structural element in the mm family as λ(m)=l~(m)/L~(m)\lambda^{(m)} = \tilde{l}^{(m)}/\tilde{L}^{(m)}, the force in each element is given by

f~(m)(l~(m);L~(m))=K~(m)(l~(m)−L~(m))=K~(m)L~(m)(λ(m)−1),K~(m)=E~(m)A~(m)L~(m),(7)\tilde{f}^{(m)}\left(\tilde{l}^{(m)};\tilde{L}^{(m)}\right)=\tilde{K}^{(m)}\left(\tilde{l}^{(m)}-\tilde{L}^{(m)}\right)=\tilde{K}^{(m)}\tilde{L}^{(m)}\left(\lambda^{(m)}-1\right),\qquad\tilde{K}^{(m)}=\frac{\tilde{E}^{(m)}\tilde{A}^{(m)}}{\tilde{L}^{(m)}}, \tag*{(7)}

where K~(m)\tilde{K}^{(m)} is the strut axial stiffness (a force per unit length). The axial stiffness is expressed as the product of the Young’s modulus E~(m)\tilde{E}^{(m)} of the material and of the cross-sectional area A~(m)\tilde{A}^{(m)} [9]. In the Supplementary Material (Sec. S1), we estimate from experimental data and from previous numerical studies the axial stiffness for each structural element. The adopted values are listed in tbl:2.

Since the mesh aspect ratio γ\gamma is constant, the length of all structural elements must scale as O(1/N)O(1/N) in the limit as N→∞N \to\infty, cf. Eqs. (1)-(5)-(6). Also the cross-sectional area A~(m)\tilde{A}^{(m)} of all structural elements must scale as O(1/N)O(1/N) to ensure that the total volume of the discretized tissue remains constant. It follows that the elemental axial stiffness K~(m)\tilde{K}^{(m)} is independent of NN.

Finally, we note that both L~(1)\tilde{L}^{(1)} and L~(3)\tilde{L}^{(3)} increase slightly with the radius. However, this variation is small, because the thickness of cornea is small compared to the radius of curvature of the anterior surface. Therefore, in Eq. (7) we use R~=R~A\tilde{R} = \tilde{R}_A to estimate the stiffness for all the structural elements in the network.

Boundary conditions

References

  1. [1]T Christian Gasser, Ray W Ogden, and Gerhard A Holzapfel. Hyperelastic modelling of arterial layers with distributed collagen fibre orientations. Journal of the royal society interface, 3(6):15–35, 2006.pubmed.ncbi.nlm.nih.gov/16849214
  2. [2]Richard G Haverkamp, Martin AK Williams, and John E Scott. Stretching single molecules of connective tissue glycans to characterize their shape-maintaining elasticity. Biomacromolecules, 6(3):1816–1818, 2005.pubmed.ncbi.nlm.nih.gov/15877410
  3. [3]A Pandolfi, A Gizzi, and M Vasta. A microstructural model of cross-link interaction between collagen fibrils in the human cornea. Philosophical Transactions of the Royal Society A, 377(2144):20180079, 2019.DOI
  4. [4]Anna Pandolfi, Maria Laura De Bellis, Alessio Gizzi, and Marcello Vasta. Modeling the degeneration of the collagen architecture in a microstructural model of the human cornea. Mathematics and Mechanics of Solids, 28(1):196–207, 2023.DOI
  5. [5]S. J. Petsche and P. M. Pinsky. The role of 3-d collagen organization in stromal elasticity: a model based on x-ray diffraction data and second harmonic-generated images. Biomechanics and modeling in mechanobiology, 12(6):1101–1113, 2013.DOI
  6. [6]Jeremy Trageser and Pablo Seleson. Anisotropic two-dimensional, plane strain, and plane stress models in classical linear elasticity and bond-based peridynamics. arXiv preprint arXiv:1905.12761, 2019.arxiv.org/abs/1905.12761

Paper details

vlnc:W_2wd48pfx
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Agricultural and Biological Sciences

Not recorded for this paper

View statement
Looking for related papers…
Contents