A Hierarchical Probability Model of Colon Cancer

We consider a model of fixed size $N = 2^l$ in which there are $l$ generations of daughter cells and a stem cell. In each generation $i$ there are $2^{i-1}$ daughter cells. At each integral time unit the cells split so that the stem cell splits into a stem cell and generation 1 daughter cell and the generation $i$ daughter cells become two cells of generation $i+1$. The last generation is removed from the population. The stem cell gets first and second mutations at rates $u_1$ and $u_2$ and the daughter cells get first and second mutations at rates $v_1$ and $v_2$. We find the distribution for the time it takes to get two mutations as $N$ goes to infinity and the mutation rates go to 0. We also find the distribution for the location of the mutations. Several outcomes are possible depending on how fast the rates go to 0. The model considered has been proposed by Komarova (2007) as a model for colon cancer.

💡 Research Summary

The paper presents a rigorous probabilistic analysis of a hierarchical cell‑division model that was originally proposed by Komarova (2007) as a framework for understanding the initiation of colon cancer. The model consists of a fixed total population of size (N=2^{\ell}) cells, organized into (\ell) generations of differentiated daughter cells plus a single stem cell. Generation (i) contains (2^{i-1}) cells; at each discrete time step the stem cell divides into a new stem cell and a generation‑1 daughter, while every daughter cell of generation (i) splits into two cells of generation (i+1). Cells of the last generation are removed from the population, mimicking the turnover of the colonic epithelium.

Cancer is assumed to arise after the accumulation of two distinct somatic mutations. The stem cell acquires the first and second mutations at rates (u_{1}) and (u_{2}), respectively, while each daughter cell acquires the first and second mutations at rates (v_{1}) and (v_{2}). Once a mutation occurs it is permanent and is inherited by all progeny of the mutated cell. The central object of study is the random time (T) required for a cell to acquire both mutations, as well as the “location” of the mutations, i.e., the pair of generations ((i,j)) where the first and second mutations first appear.

The authors consider the joint asymptotic regime where the population size tends to infinity ((N\to\infty), equivalently (\ell\to\infty)) and all mutation rates tend to zero. Under these conditions each mutation event becomes a rare event and can be approximated by a Poisson process. By exploiting this approximation together with martingale central limit theorems, the authors derive explicit limiting distributions for (T) and for the mutation location.

A key insight is that the limiting behavior of (T) depends critically on how fast the products (N u_{1}) and (N v_{1}) converge. Four qualitatively distinct regimes are identified:

1. Stem‑cell‑dominant regime ((N u_{1}\to0) and (N v_{1}\to0)). The first mutation almost surely occurs in the stem cell; the second mutation also occurs in the stem cell. In this case (T) converges to an exponential distribution with rate (u_{1}u_{2}).

2. Daughter‑cell‑dominant regime ((N v_{1}\to c>0) while (N u_{1}\to0)). The first mutation typically arises in a high‑generation daughter cell because the sheer number of such cells outweighs the lower per‑cell mutation rate. The second mutation then occurs either in the same generation or in the next generation, leading to a limiting distribution for (T) that is a mixture of gamma and exponential components; the shape parameter reflects the expected generation index of the first mutation.

3. Mixed regime ((N u_{1}\to c_{1}>0) and (N v_{1}\to c_{2}>0)). Both stem and daughter cells contribute appreciably to the first mutation. The time to acquire both mutations is the minimum of two independent exponential clocks, giving a cumulative distribution function (1-\exp(-c_{1}t)\exp(-c_{2}t)).

4. Fast‑mutation regime ((N u_{1}, N v_{1}\to\infty)). Mutations are so frequent that the two‑hit event occurs essentially instantaneously; (T) collapses to zero in the limit.

For each regime the authors provide closed‑form expressions for the probability density function (pdf) and cumulative distribution function (cdf) of (T). They also verify the analytical results with extensive stochastic simulations, demonstrating excellent agreement.

The spatial aspect of the model is treated by deriving the joint probability mass function for the pair ((i,j)) where the first and second mutations appear. Conditioning on the generation of the first mutation, the probability that the second mutation occurs in generation (j\ge i) is proportional to the number of cells in generation (j) ((2^{j-1})) and the second‑mutation rate (v_{2}). The resulting formula
\