Chapter 3: Simulating Cell Islands with the CPM
3.3 Application of the model to a single isolated cell
Before investigating cell islands, we consider a single cell case. Implementing the model with a single cell will include only two of the three mechanisms from the energy function, namely the area constraint and the perimeter contraction. The interaction between the two mechanisms in the single cell size and shape can be observed.
In the simulations a single cellβs size and shape is determined by the interaction between the perimeter contraction parameter, Ξ, area constraint parameter, ππ, and the preferred area, π΄π΄ππ. We
understand that parameter ππ maintains the cell area close to the preferred area and parameter Ξ
attempts to reduce the cellβs perimeter to zero. This means that increasing ππ will make it more
difficult for the cell area to differ from the preferred area and increasing Ξ will decrease the cells perimeter and size.
The equilibrium state of the single cell can be characterised by the cell area and perimeter. The simulation results are compared to an optimisation solution resulting from the optimisation problem described below.
We assume all cells have a shape categorised by the cell area and perimeter. The cell perimeter squared cannot fall below a particular area, i.e. π΄π΄ β€ πππΏπΏ2. The value of ππ is the ratio of area and perimeter squared, π΄π΄ πΏπΏβ , of a defined shape, referred to as the minimum shape in Chapter 2 2, and occurs when π΄π΄ = πππΏπΏ2. In an unrestrictive lattice the minimum shape is represented by a circle; however, in a restrictive lattice the cell is instead represented by a polygon. Due to the selection of hexagonal pixels the polygon is a βhexagonβ, see Figure 2.4, to replicate the simulations, and the ratio is given as ππ = 1 48β .
Only the interaction between the cell area constraint and cell perimeter contraction is expressed in the energy function for a single cell,
πΈπΈ(π΄π΄, πΏπΏ) =ππ
2 οΏ½π΄π΄ β π΄π΄πποΏ½
2
+Ξ
2 πΏπΏ2. ( 3.2 )
Additionally, simple conditions are applied, namely that the cell area and perimeter are non-
negative and the inequality discussed previously between the cell area and perimeter, i.e. π΄π΄ β€ πππΏπΏ2. An optimisation problem can be expressed with the energy Function (3.2) and conditions giving the Lagrangian function
πΈπΈβ(π΄π΄, πΏπΏ, ππ
1, ππ2, ππ3) =ππ2 οΏ½π΄π΄ β π΄π΄πποΏ½2+Ξ2 πΏπΏ2
+ππ1(π΄π΄ β πππΏπΏ2) + ππ2(βπ΄π΄) + ππ3(βπΏπΏ),
( 3.3 )
with the use of Karush-Kuhn-Tucker (KKT) conditions (Lange, 2004). The KKT conditions are a generalisation of the equality conditions used with Lagrangian multipliers to incorporate the
inequalities of the problem. The Lagrangian Function (3.3) with these conditions results in the simultaneous equations πππΈπΈβ(π΄π΄, πΏπΏ, ππ 1, ππ2, ππ3) πππ΄π΄ = πποΏ½π΄π΄ β π΄π΄πποΏ½ + ππ1β ππ2 = 0, πππΈπΈβ(π΄π΄, πΏπΏ, ππ 1, ππ2, ππ3) πππΏπΏ = π€π€πΏπΏ β 2ππ1πππΏπΏ β ππ3 = 0, ππ1(π΄π΄ β πππΏπΏ2) = 0, ππ2(βπ΄π΄) = 0 and ππ3(βπΏπΏ) = 0.
The results of these simulations equations need to satisfy the inequalities of ππ1 β₯ 0, ππ2 β₯ 0, ππ3 β₯ 0, (π΄π΄ β πππΏπΏ2) β€ 0, βπ΄π΄ β€ 0 and β πΏπΏ β€ 0.
In this problem there are eight different cases including (ππ1 = 0, ππ2 = 0, ππ3 = 0), (ππ1 > 0, ππ2 = 0, ππ3 = 0), (ππ1 = 0, ππ2 β₯ 0, ππ3 = 0), (ππ1 = 0, ππ2 = 0, ππ3 β₯ 0), (ππ1 β₯ 0, ππ2 β₯ 0, ππ3 = 0),
(ππ1= 0, ππ2 β₯ 0, ππ3 β₯ 0), (ππ1 β₯ 0, ππ2 = 0, ππ3 β₯ 0) and (ππ1β₯ 0, ππ2 β₯ 0, ππ3 β₯ 0). The only solution
that satisfies the equations and inequalities is from the case (ππ1 β₯ 0, ππ2 = 0, ππ3 = 0) and gives the solution of the cell area,
π΄π΄1 = π΄π΄ππβ2ππππΞ , ( 3.4 )
and the solution of the cell perimeter,
πΏπΏ1 = οΏ½π΄π΄ππ =οΏ½π΄π΄ππ βππ 2ππΞ2ππ . ( 3.5 )
These solutions show that the area of a single cell, π΄π΄1, and perimeter of a single cell, πΏπΏ1,
decrease with the perimeter contraction parameter Ξ, see Figure 3.2a and 3.2c, respectively, and
increase with the area constraint ππ, see Figures 3.2b and 3.2d, respectively. The cell can disappear
when the area and perimeter equal zero for both the simulation and optimisation solutions, shown in Figure 3.3, when π΄π΄ππ β€ Ξ (2ππππ)β .
The simulation results and the mathematical optimisation solutions above are qualitatively similar. However, a few differences between them are a result of the restrictions of the discretised lattice structure of the simulations. The main differences are observed in Figure 3.2c and 3.2d,
showing the cell perimeter against the parameters ππ and Ξ, respectively. They show differences
between the simulations and optimisation solution occur at comparatively small values of the
contraction parameter of those values tested, such as Ξ = 1. With βlowβ values of contraction the
deviation of the cellβs perimeter requires less energy, and therefore the noise in the system distorts the cellβs shape more readily than a cell with a larger contraction parameter. This trend is also
replicated in the ratio ππ = π΄π΄ πΏπΏβ against ππ and Ξ, shown in Figure 3.2e and 3.2f, respectively. The 2
equilibrium results of the cell simulations and ππ = 1 48β , representing the βhexagonalβ shape of the
cell on the lattice (shown on the figure as a black line), display a large difference between them for
βsmallβ values of the contraction, such as Ξ = 1. The values of ππ from the simulation are smaller
than 1 48β , representing distorted shapes. These differences are similar to the differences in the perimeters because the ratio is calculated with the perimeter, ππ = π΄π΄ πΏπΏβ . 2
Furthermore, there is a difference between the simulation and the optimisation solution cell area, observed in Figure 3.3a, which is caused by the lattice and smaller preferred area. For
simulations with an area constraint parameter ππ = 3 and perimeter contraction parameter greater
than 5, the simulation cell area results are greater than the optimisation solutions. The optimisation
solutions tend towards zero while the simulation cell area increases as Ξ increases. As the cell
becomes smaller in the simulations there are situations where the decrease in the cell perimeter requires the cell to initially increase its perimeter. This means the cell can become fixed at a larger perimeter with larger perimeter contraction coefficient values and no longer decreases its perimeter to achieve the calculated equilibrium. This artefact can be eliminated by increasing the amount of noise, or for larger preferred area values as shown in Figure 3.2. However, the artefact of the cell perimeter, discussed above, is still observable in both situations.
(a) (d)
(b) (e)
(c) (f)
Figure 3.2. Plots, with the preferred area π΄π΄ππ = 500, for a single cell simulated on a
hexagonal pixel grid with the CPM showing the (a) mean cell area with the geometric solution from Equation (3.5), (b) mean cell perimeter with the geometric solution from Equation (3.4) and (c) ratio ππ = π΄π΄ πΏπΏβ against the perimeter contraction, Ξ, for different area 2
constraint parameters, ππ; and the (d) mean cell area with the geometric solution from
Equation (3.5), (e) mean cell perimeter with the geometric solution from Equation (3.4) and (f) ratio ππ = π΄π΄ πΏπΏβ against the area constraint parameter, ππ, with different perimeter 2
(a) (d)
(b) (e)
(c) (f)
Figure 3.3. Plots, with the preferred area π΄π΄ππ = 100, for a single cell simulated on a
hexagonal pixel grid with the CPM showing the (a) mean cell area with the geometric solution from Equation (3.5), (b) mean cell perimeter with the geometric solution from Equation (3.4) and (c) ratio ππ = π΄π΄ πΏπΏβ against the perimeter contraction, Ξ, for different area 2
constraint parameters, ππ; and the (d) mean cell area with the geometric solution from
Equation (3.5), (e) mean cell perimeter with the geometric solution from Equation (3.4) and (f) ratio ππ = π΄π΄ πΏπΏβ against the area constraint parameter, ππ, with different perimeter 2
contractions, Ξ. Each point is averaged over the last 50 of 1000 iterations and the