ARTICLE IN PRESS

Journal of Theoretical Biology 248 (2007) 411–417

www.elsevier.com/locate/yjtbi

Aging and immortality in a cell proliferation model

T. Antala,b, K.B. Blagoevc, S.A. Trugmanc, S. Rednera,
a
Center for Polymer Studies and Department of Physics, Boston University, Boston, MA 02215, USA
b
Program for Evolutionary Dynamics, Harvard University, Cambridge, MA 02138, USA
c
Theory Division, Los Alamos National Laboratory, Los Alamos, NM 87545, USA
Received 6 October 2006; received in revised form 27 April 2007; accepted 6 June 2007
Available online 12 June 2007

Abstract

We investigate a model of cell division in which the length of telomeres within a cell regulates its proliferative potential. At each
division, telomeres undergo a systematic length decrease as well as a superimposed fluctuation due to exchange of telomere DNA
between the two daughter cells. A cell becomes senescent when one or more of its telomeres become shorter than a critical length. We
map this telomere dynamics onto a biased branching-diffusion process with an absorbing boundary condition whenever any telomere
reaches the critical length. Using first-passage ideas, we find a phase transition between finite lifetime and immortality (infinite
proliferation) of the cell population as a function of the influence of telomere shortening, fluctuations, and cell division.
r 2007 Elsevier Ltd. All rights reserved.

MSC: primary 02.50.r; 05.40.Jc; 87.17.Ee

Keywords: Cell division; Telomere dynamics; Senescence

1. Introduction tional post-replicative processing of the telomeric DNA is

necessary to protect the end of the chromosome from being
Aging is a complex and incompletely understood process recognized as a double strand break in need of repair. This
characterized by deteriorating cellular, organ, and system processing also contributes to the end replication problem.
function. Replicative senescence, the phenomenon whereby What remains unexplained, however, is why senescent
normal somatic cells show a finite proliferative capacity, is cells occur in cell cultures long before the expected number
thought to be a major contributor to this decline (Hayflick, of cell divisions estimated from the gradual basal loss. It
1965; Harley, 1991; Campisi, 1996, 1997, 2001). Telomeres has been shown recently that in addition to the basal loss of
are repetitive DNA sequences (ðTTAGGGÞn in human 100 base pairs per division, a complex set of events that
cells) at both ends of each linear chromosome and their leads to telomere exchange between sister chromatids can
role is to protect the coding part of the DNA. Normal occur. This telomere sister chromatid exchange (T-SCE),
human somatic cells become senescent after a finite number together with basal telomere loss and a number of observed
of doublings. or suggested telomere recombination events, collectively
As a general rule, cells for which telomerase activity is define telomere dynamics, and this dynamics leads to a
absent lose of the order of 100 base pairs of telomeric DNA wide distribution of telomere lengths in cell cultures
from chromosome ends in every cell division. This basal (Reddel, 2003). Telomere exchange can also occur between
loss has been attributed to the end replication problem the telomeres of different chromosomes. Currently avail-
(Watson, 1972; Olovnikov, 1971) in which the DNA- able data cannot distinguish between this process and
polymerase cannot replicate all the way to the end of the T-SCE. It is also believed that sister chromosome exchange
chromosome during DNA lagging strand synthesis. Addi- is induced by DNA damage. Because the sister chromatids
are in closer proximity compared to the distance between
Corresponding author. Tel.: +1 617 353 2618; fax: +1 617 353 9393. different chromatids, it can be hypothesized that the
E-mail address: redner@bu.edu (S. Redner). probability for telomere interchromatid exchange (T-ICE)

0022-5193/$ - see front matter r 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jtbi.2007.06.009
 ARTICLE IN PRESS
412 T. Antal et al. / Journal of Theoretical Biology 248 (2007) 411–417

is smaller than the probability for sister chromatid telomere dynamics provides the possibility for a telomere
exchange between sister chromatids. to occasionally grow when a cell divides. This subpopula-
Recently, one of the present authors proposed a theory tion of cells with long telomeres and thus higher
(Goodwin and Blagoev, 2004; Blagoev and Goodwin, proliferative potential can become even more so at the
2005), based on telomere dynamics including T-SCE and next cell division, a mechanism that allows a long-lived
T-ICE, that is capable of explaining Werner’s syndrome, subpopulation to thrive.
an inherited disease characterized by premature aging and We are interested in basic statistical properties of the cell
death, and a subset of cancers that seem to use a proliferative potential. Some fundamental questions that
recombination mechanism to maintain telomere length. we will study include:
This latter mechanism, known as alternative lengthening of
telomeres (ALT), as well as additional telomerase activity 1. Can a cell population divide indefinitely?
in some cases, is thought to contribute to the large 2. How long does it take for a cell population to become
proliferative potential of these cells (Bryan et al., 1997; senescent?
Henson et al., 2002). Here, proliferative potential denotes 3. How many dividing cells exist after a given number of
the time over which cells can continue to divide. In divisions?
Goodwin and Blagoev (2004) and Blagoev and Goodwin
(2005), we showed that both Werner’s syndrome and ALT Within an idealized model of telomere kinetics described by
can be described within the general framework of telomere Eq. (1) for a single telomere per cell, it is possible to answer
dynamics in which there is an elevated rate of exchange of these questions analytically by mapping the telomere
telomere DNA between two daughter cells, as observed in dynamics to a first-passage process. Using this approach,
many experiments (Bailey et al., 2004). One of the main we find a phase transition between a finite-lifetime cell
numerical results from Goodwin and Blagoev (2004) and population and immortality as a function of three basic
Blagoev and Goodwin (2005) is a transition from finite to control parameters—the magnitude of the systematic part
infinite proliferative potential in a cell culture as the of the telomere evolution, determined by Dx, the effective
parameters controlling the telomere recombination rates diffusion coefficient associated with the stochastic part of
are varied. the telomere evolution, determined by hdxðtÞ2 i, and the cell
In this work, we analytically investigate an idealized division rate.
model of cell proliferation in which the number of cell
divisions before senescence occurs is controlled by the 2. Telomere replication model
dynamics of telomeres during cell division. Each cell
contains a certain number of telomeres. When a cell In our telomere replication model, we assume that the
divides, the telomeres in each daughter cell ostensibly initial length of each telomere in a cell is x0 . In any cell
shorten by a fixed amount Dx. In addition to this division event, the length of a telomere changes by two
systematic telomere shortening, the effect of T-SCE distinct processes:
processes during cell division leads to a superimposed
stochastic component to the telomere length dynamics by (i) a systematic shortening of each telomere by Dx;
an amount dx (Fig. 1). Thus in each cell division event, the (ii) an additional stochastic component of the length
length of an individual telomere evolves by the combined change of magnitude dx.
effects of these systematic and stochastic mechanisms.
When the length x of any of the telomeres within a cell Thus, the length of a telomere changes according to
reaches zero, the cell stops dividing and becomes senescent.
On the other hand, the stochastic component of the x ! x  Dx þ dx. (1)
Here, the stochastic variable dx accounts for the T-SCE
processes that we assume to have mean value equal to zero,
x-Δx-δx
hdxi ¼ 0, and no correlations at different times,
x hdxðtÞdxðt0 Þi / dðt  t0 Þ. The justification for the absence
of correlation is that T-SCE events have been linked to
x-Δx+δx DNA damage (Bailey et al., 2001), which occurs randomly
in the cell.
Because of the systematic and stochastic contributions to
the change in telomere length in each division event, the
length of a telomere undergoes a biased random walk, with
a bias toward shrinking. It is instructive to estimate the
relative importance of the systematic and stochastic
Fig. 1. Schematic illustration of telomere evolution. The telomeres in the
initial cell contain x ¼ 15 units. Upon division, each telomere ostensibly
components of this length evolution. For this purpose,
shortens by Dx ¼ 5 units, but additional exchange of dx ¼ 2 units we define the time unit as the physical time between cell
between daughter telomeres (dashed) leads to final lengths of 8 and 12. divisions dt. Thus, in the absence of stochasticity a
 ARTICLE IN PRESS
T. Antal et al. / Journal of Theoretical Biology 248 (2007) 411–417 413

telomere shrinks to zero length in x0 =Dx cell division determine the evolution of the telomere length distribution.
events. Let the initial number of cells be N 0 . The solution of (2) is
It is now helpful to recall some basic numbers about simply
human telomeres to connect our mechanistic telomere
model and real cell division: nðx; tÞ ¼ N 0 ekt cðx; tÞ, (3)
where cðx; tÞ, given in (A.2), is the solution to the
Parameter Definition Numerical value convection–diffusion equation with an absorbing bound-
ary condition at x ¼ 0 and the initial condition that all cells
dt Time between cell 20 min to several have initial telomere length x0 .
divisions hours We may also treat the situation in which each cell
x0 Initial telomere  104 base pairs contains M independent telomeres as a zeroth-order
length description for cells that contain many telomeres whose
Dx Systematic length  102 base pairs dynamics is coupled by the T-SCE exchange process
decrease per division (schematically illustrated in Fig. 1). This assumption of
dx Stochastic length  102 base pairs independence of different telomeres allows us to apply the
decrease per division single telomere per cell first-passage description with only
minor modifications. For cells containing M independent
The quantity dx is known only very roughly. It is possible, telomeres of lengths x ¼ ðx1 ; . . . ; xM Þ, the density of cells
with low probability, that even whole telomeres can be lost nðx; tÞ satisfies the M-dimensional convection–diffusion–
in TCE processes (Hasenmaile et al., 2003). growth equation
With the above numbers, a telomere shrinks to zero in
x0 =Dx  102  N cell division events with purely determi- @nðx; tÞ
¼ knðx; tÞ þ vrnðx; tÞ þ Dr2 nðx; tÞ, (4)
nistic shrinking. Now consider the role of stochasticity: in @t
N cell divisions, the root-mean-square length change due to with absorbing boundaries when any telomere length xi
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
stochastic events is ‘rms ¼ NðdxÞ2  103 . Thus, in the reaches zero. The solution simply factorizes as a product of
one-dimensional solutions:
time for a telomere to systematically shrink from 104 to
zero, stochasticity gives a length uncertainty of 103 —a 10% Y
M
nðx; tÞ ¼ N 0 ekt cðxi ; tÞ. (5)
correction to the bias.
i¼1
Finally, we need to include the role of cell division on
this biased random walk description to arrive at a theory If each telomere has initial length x0 , all the cðxi ; tÞ are the
for telomere dynamics. That is, we need to allow a random same and are given by (A.2). We can also straightforwardly
walk to replicate as it undergoes biased hopping. While study the case where each telomere has a different initial
there are many ways to parameterize the effects of bias, length by merely using the unique initial length of each
stochasticity, and replication in the continuum limit, all telomere in Eq. (5). It is worth mentioning that in addition
such models lead to the following convection-diffusion to its application to cell division statistics, Eq. (4) is also
equation with multiplicative growth: related to the Fleming–Viot (FV) process (Fleming and
Viot, 1979; Ferrari and Marić, 2006), in which a population
@nðx; tÞ @nðx; tÞ @2 nðx; tÞ of diffusing particles can get absorbed at boundary point
¼ knðx; tÞ þ v þD , (2)
@t @x @x2 and then be re-injected into the system at a rate that is
proportional to local particle density.
in which v represents the bias for telomere shrinking, D
From our description of telomere dynamics as a biased
accounts for the stochastic part of the telomere length
branching-diffusion process, we now determine basic
evolution, and k accounts for cell division. While there is
features about the time dependence of the cell population
only an indirect connection between the model parameters
and the statistics of telomere lengths.
v; D; k, and the parameters Dx; dx that account for what
happens to a telomere in a single cell division, this
continuum description has the advantage of capturing the 2.1. Number of dividing and senescent cells
physical essence of telomere dynamics while being analy-
tically tractable. Cells in which each telomere has positive length can
The basic question that we seek to understand is how divide. The number of such active cells is given by the
long it takes for a cell to become senescent, an event that integral of the number density of cells over the positive 2M -
occurs when the length of one of its telomeres reaches zero. tant (x1 ; x2 ; . . . ; xN 40) of length space:
This condition translates to an absorbing boundary Z
condition at x ¼ 0 for the biased diffusion process that N active ðtÞ ¼ nðx; tÞ dx ¼ N 0 ekt S M ðtÞ, (6)
described the telomere length distribution. We now exploit x40

some classic results about the first-passage probability of where SðtÞ is the survival probability of a biased random
biased diffusion (see the Appendix and Redner, 2001) to walk (given by (A.4)). From (A.6), the long-time behavior
 ARTICLE IN PRESS
414 T. Antal et al. / Journal of Theoretical Biology 248 (2007) 411–417

102 immortality is controlled by details of telomere evolution

as encoded by the coefficients a and b.
100 We also obtain the total number of cells that become
senescent at any given time during the evolution as the
10-2 diffusive flux to xi ¼ 0 for any i. For the symmetric initial
condition, the number of cells reaching any boundary
10-4 xi ¼ 0, i ¼ 1; 2; . . . ; M, are the same. Hence we may
Nactive

consider a single boundary, say i ¼ 1. The number of

10-6 dying cells at this boundary can be written as the integral
over the x1 ¼ 0 surface
Z
10-8 @nðx; tÞ
J 1 ðtÞ ¼ D dx? , (8)
x40
x ¼0
@x1
10-10
1

where the integral is over all the M  1 coordinates

10-12 perpendicular to x1 . Note again the absence of a convective
0 20 40 60 80 100 term in this expression because there is no convective flux
t when the concentration is zero. Using the product form (5)
of nðx; tÞ for the total number of dying cells we obtain
Fig. 2. Plot of the number of active cells versus time from the second line
of Eq. (7) for the case of M ¼ 1 telomere per cell, with 1 ¼ 0:8 (solid), 0.9 JðtÞ ¼ MJ 1 ðtÞ ¼ N 0 Mekt F ðtÞS M1 ðtÞ, (9)
(dashed), and 0.99 (dotted).
where SðtÞ and F ðtÞ are given by (A.4) and (A.3),
respectively. The total number of senescent cells that are
of N active ðtÞ is given by produced during the course of the evolution is
rffiffiffiffi !M Z 1 Z 1
@S M ðtÞ
D 2x0 2 N sen ¼ JðtÞ dt ¼ N 0 dt ekt
N active ðtÞ eMvx0 =2D t3M=2 eðkMv =4DÞt , 0 0 @t
p v2  Z 1 
/ t3M=2 ektðM 1Þ=M ð7Þ ¼ N0 1  k dt ekt SM ðtÞ , ð10Þ
0
2 where we again use the fact that F ðtÞ ¼ dSðtÞ=dt to
with M  4Dk=Mv (Fig. 2). Thus, the fundamental
parameters of the system are the Péclet number Pe  perform the integration by parts.
vx0 =2D (Probstein, 1994), a dimensionless measure of the For the special case of one telomere per cell (M ¼ 1), we
relative importance of the bias and the stochasticity, and, have
more importantly, the dimensionless growth rate, M . Note N 0 x0 2

that both the diffusion coefficient D and the bias velocity v JðtÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffi ekt eðx0 vtÞ =4Dt . (11)
3
4pDt
are proportional to k, since the telomere length changes
occur only when a cell divides. As a result, the growth rate In this case the total number of senescent cells is
Z
M is actually independent of k. For M o1, cell division is N 0 x0 1 1 kt ðx0 vtÞ2 =4Dt
insufficient to overcome the effect of inexorable death due N sen ¼ pffiffiffiffiffiffiffiffiffi e e dt. (12)
4pD 0 t3=2
to the systematic component of the telomere shortening
We now make the substitution z ¼ t1=2 to recast the above
and the population of dividing cells decays exponentially in
integral into the form
time. Z
To get a feeling for the interplay between telomere N 0 x0 vx0 =2D 1 2 2 2 2
N sen ¼ p ffiffiffiffiffiffi
ffi e dz ex0 z =4Dðv =4DkÞ=z . (13)
shortening and cell division, let us employ the numerical pD 0
parameters given at the beginning of this section in Eq. (7) R 1 az2 b=z2 pffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffi ffi
Using 0 e dz ¼ ðp=4aÞe 4ab from 3.325 of
to obtain the total number of dividing cells. Since cells
(Gradshteyn and Ryzhik, 1965), we thus obtain
double at each time step, k ¼ ln 2 when we express t in " sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
units of cell division times. Furthermore, we define the  ffi#
vx0 =2D v2 x20 4Dk
coefficients a and b by v ¼ a  102 and D ¼ b  104 . Then N sen ¼ N 0 e exp  1 2
4D v
the time-dependent exponential factor in (7) for the case pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
M ¼ 1 becomes ¼ N 0 ePe exp½ Pe2 ð1  1 Þ, ð14Þ

eðkv
2
=4DÞt 2
eðln 2a =4bÞt . which is plotted in Fig. 3 for the case N 0 ¼ 1 and Pe ¼ 10.
This result for N sen holds only for 1 o1; this is the
The exponent can be either positive or negative exponent regime where the cell population eventually becomes
depending on a and b, which, in turn, depend on details of senescent so that the total number of senescent cells
the telomere evolution of a single cell division event. Thus, produced during the evolution is finite. When 1 51, the
using the numbers appropriate for humans, senescence or leading behaviors of the two exponential factors in Eq. (14)
 ARTICLE IN PRESS
T. Antal et al. / Journal of Theoretical Biology 248 (2007) 411–417 415

105 approaches a constant

8D2
hx2 i  hxi2 ! . (17)
v2
104
A more pedantic way to arrive at this same result is to
compute the exact time-dependent mean telomere length
103
hxðtÞi using Eq. (A.2) for cðx; tÞ and then taking the t ! 1
limit. In summary, the mean telomere length is time
Nsen

independent, as predicted by the asymptotic form of the

102 telomere length distribution in Eq. (15). We emphasize that
this result pertains to the fraction of cells that are active. If
all cells—senescent and active—are included in the average,
101 then hxi would asymptotically decay with time.

2.3. Mean proliferative potential and immortality

100
0 0.2 0.4 0.6 0.8 1 As discussed above, for M o1, all cells eventually
ε1 become senescent. However, for M 41, a subpopulation
of infinitely dividing cells arise so that the average cell
Fig. 3. Plot of the total number of senescent cells that are produced during population becomes immortal. We make this statement
the evolution of the cell population as a function of the dimensionless
more precise by computing the average lifetime of the
growth rate 1 ¼ 4Dk=v2 .
population for M o1. This lifetime is defined as the
average age of each cell when it becomes senescent and is
cancel and N sen ! ð1 þ 1 =2PeÞ. Thus, as 1 ! 0 (no thus given by
division), the initial cell immediately becomes senescent R1 R 1 kt
0 tJðtÞ dt te F ðtÞSM1 ðtÞ dt
so that the total number of senescent cells that are hti ¼ R 1 ¼ R01 kt M1
, (18)
produced equals one. This unrealistic result arises because 0 JðtÞ dt 0 e F ðtÞS ðtÞ dt
in the continuum description telomeres can shrink to zero where we use (9) for JðtÞ, the number of cells that become
length before any cell division can occur. On the other senescent at time t. More simply, the mean proliferative
hand, for 1 41, the population is immortal and an infinite potential (lifetime) can also be obtained from the total
number of senescent cells are produced during the number of senescent cells via hti ¼ ð@=@kÞ ln N sen .
evolution. Using the general definition of Eq. (18), it is straightfor-
ward to compute higher moments of the lifetime for the
2.2. Telomere length distribution case M ¼ 1. Here, the integrals in (18) can be written in
terms of the modified Bessel functions of the second kind
A curious aspect of the population of dividing cells is the (see Gradshteyn and Ryzhik, 1965 #12 in 3.471) to give
dependence of the cell density on telomere length as  g=2 pffiffiffiffiffiffiffiffiffiffiffiffiffi
t ! 1. For any number of telomeres per cell M, the g x20 K g1=2 ð 1  1 Pe=2Þ
ht i ¼ 2 pffiffiffiffiffiffiffiffiffiffiffiffiffi , (19)
telomere length distribution is independent of this number, v ð1  1 Þ K 1=2 ð 1  1 Pe=2Þ
since the distribution is simply proportional to cðx; tÞ in where g does not necessarily have to be an integer. Using
Eq. (A.2). In the t ! 1 limit, we then obtain this general formula the variance of the lifetime is
xx0 2
cðx; t ! 1Þ / qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi eðv =4DÞtvðxx0 Þ=2D 2Dx0
ht2 i  hti2 ¼ . (20)
4pðDtÞ3 v3 ð1  1 Þ3=2
/ xevx=2D . ð15Þ In fact, the nth cumulant can be obtained simply from the
Thus, apart from an overall time-dependent factor, the general formula (vanKampen, 1997)
density of dividing cells in which the constituent telomere @n ln N sen
Cn ¼ .
has length x is linear in x for small lengths and has an @kn
exponential cutoff for large lengths. From the distribution Therefore, for the higher moments of the lifetime, the
given in Eq. (15), the mean telomere length goes to a diffusion coefficient does play an essential role.
constant for large times: For M ¼ 1, we immediately obtain from Eq. (14)
R1
xcðx; tÞ dx 4D x0 1
hxi ¼ R0 1 ! , (16) hti ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffi . (21)
0 cðx; tÞ dx
v v 1  1
independent of whether the total cell population is growing Thus, the mean cell proliferative potential diverges to
or decaying. The variance of the telomere length also infinity as 1 ! 1 from below. This result for the number of
 ARTICLE IN PRESS
416 T. Antal et al. / Journal of Theoretical Biology 248 (2007) 411–417

cell divisions before senescence is one of our primary essential features of telomere dynamics in cell cultures.
results. Notice that in the case of no cell division (k ¼ 0, Because of the competing influences of cell division, which
1 ¼ 0) the average lifetime hti ¼ x0 =v. That is, hti coincides obviously increases the number of proliferative cells, and
with the time for a biased diffusing particle to be convected the general trend of telomere shortening, we showed that
to the origin. It is surprising at first sight that diffusion there is a phase transition between a normal state where a
plays no role in determining the average number of cell cell culture becomes senescent to a new state where a cell
divisions. Exactly the same type of result arises for the culture can become immortal.
discrete random walk with a bias v (Gikhman and From our theory, we were able to answer the basic
Skorokhod, 1969; Karlin and Taylor, 1975; Nisbet and questions posed in the Introduction. Specifically:
Gurney, 1982).

2.4. Senescence plateau (i) We determined the condition for whether a cell
population ultimately becomes senescent or whether
A useful characterization of the number of cell divisions it continues to divide ad infinitum. The transition
distribution function of a population is the senescence rate between these two regimes is given by the condition
mðtÞ. The senescence (or mortality) rate is the ratio of the M ¼ 1, where M ¼ 4Dk=Mv2 is a dimensionless
number of cells that become senescent at time t to the total measure of the relative effect of cell division, random
number of cells that are still dividing at this time. fluctuations, and basal loss in the length evolution of a
Equivalently, the senescence rate is the probability that a telomere.
randomly chosen dividing cell becomes senescent at the (ii) We also found that for the case of M ¼ 1 telomere per
next moment. The senescence rate is thus given by cell, the mean time for a cell population to become
JðtÞ senescent is
mðtÞ ¼ , (22)
NðtÞ x0 1
hti ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffi
where the number of dying cells JðtÞ and dividing cells NðtÞ v 1  1
are given by Eqs. (9) and (6), respectively. Substituting for 1 o1. Here, x0 is the initial length of the telomere,
these expressions into (22), we find that the senescence rate v is the amount by which the telomere shrinks by basal
is independent of the number of telomeres per cell M, i.e. loss in each division, and M  4Dk=Mv2 is a
F ðtÞ @ ln SðtÞ dimensionless measure of the relative importance of
mðtÞ ¼ ¼ . (23) cell division to basal loss.
SðtÞ @t
(iii) Finally, we found that the total number of cells
Using the asymptotic form of SðtÞ given in (A.6), the produced before the entire cell culture becomes
senescence rate approaches a time-independent value in the senescent becomes extremely large as  approaches its
long-time limit and is given by critical value from below, as presented in Eq. (14).
 
v2 3 1
mðtÞ ’ þ þO 2 . (24)
4D 2t t Our results may be helpful for understanding how
ALT cells maintain their telomeres. In these cells, increased
Amazingly, as the cell population ages, the senescence
T-SCE rate, wide telomere size distribution, and increased
rate of the cells that remain dividing ultimately tends
cell lifetimes have been observed (Reddel, 2003). These
to a constant value for large times. This phenomenon is
observations are all natural outcomes from our model.
known as the mortality plateau. Namely, the probability
How these results depend on the number of short telomeres
that the somatic cells of an organism become senescent
and on the details of the T-SCE process are very important
becomes independent of its age in the long-time limit. This
questions that are under investigation.
surprising fact was observed experimentally in human
populations (Vaupel, 1997) and for fruit flies (Carey
et al., 1992). It was also observed numerically in a model Acknowledgements
of aging that is similar to ours (Weitz and Fraser, 2001).
The existence of such a senescence plateau is actually Much of this work was performed when S.R. was on
typical of a wide range of Markov processes (Steinsaltz and leave at the CNLS at Los Alamos National Laboratory.
Evans, 2007). He thanks the CNLS for its hospitality and P. Ferrari for
encouragement. K.B.B. thanks E.H. Goodwin for the
3. Summary useful discussions. We also gratefully acknowledge finan-
cial support from NIH Grant R01GM078986 (T.A.) and
We studied an idealized model for the dynamics of Jeffrey Epstein for support of the Program for Evolu-
telomere lengths during cell division that is based on a tionary Dynamics at Harvard University, DOE Grant DE-
systematic basal loss and a stochastic component to the AC52-06NA25396 (K.B.B. and S.A.T.), and NSF Grant
evolution that arises from T-SEE. This model captures DMR0535503 and DOE Grant W-7405-ENG-36 (S.R.).
 ARTICLE IN PRESS
T. Antal et al. / Journal of Theoretical Biology 248 (2007) 411–417 417

Appendix A. First-passage for biased diffusion behavior of SðtÞ is given by

rffiffiffiffiffiffi
Dt 2x0 2
We recall some basic results about first-passage for SðtÞ ekt eðvtx0 Þ =4Dt
biased diffusion on the positive half line x40 (Redner, p ðvtÞ2  x20
rffiffiffiffi
2001) that will be used to describe cell proliferation D 2x0 vx0 =2D 3=2 ðv2 =4DÞt
statistics. According to Eq. (1), the length of each telomere  e t e . ðA:6Þ
p v2
undergoes biased diffusion, in the continuum limit, with
positive bias (v40) defined to be directed towards smaller As expected intuitively, the survival probability asympto-
telomere length x. An absorbing boundary condition at the tically decays exponentially with time because the bias
origin imposes the constraint that when x reaches zero the drives the particle towards the absorbing point. This result
cell effectively becomes senescent and is thus removed from is used in Eqs. (6) and (7) to determine the number of active
the population of dividing cells. cells as a function of time.
Let cðx; tÞ be the concentration of diffusing particles at x
at time t. The concentration evolves by the convection– References
diffusion equation
Abramowitz, M., Stegun, I.A., 1970. Handbook of Mathematical
2 Functions. Dover, New York.
@cðx; tÞ @cðx; tÞ @ cðx; tÞ
¼v þD . (A.1) Bailey, S.M., Cornforth, M.N., Kurimasa, A., Chen, D.J., Goodwin,
@t @x @x2 E.H., 2001. Science 293, 2462.
Bailey, S.M., Brenneman, M.A., Goodwin, E.H., 2004. Nucleic Acids Res.
For the initial condition cðx; t ¼ 0Þ ¼ dðx  x0 Þ, corre-
32, 3743.
sponding to a single particle starting at x0 , the concentra- Blagoev, K.B., Goodwin, E.H., 2005. In: Proceedings of the Telomeres
tion at any later time is (Redner, 2001) and Telomerase, Cold Spring Harbor Laboratory.
Bryan, T.M., Englezou, A., Dalla-Pozza, L., Dunham, M.A., Reddel,
1 2
cðx; tÞ ¼ pffiffiffiffiffiffiffiffiffiffiffi ½eðxx0 þvtÞ =4Dt R.R., 1997. Nat. Med. 3, 1271.
4pDt Campisi, J., 1996. Cell 94, 497.
2 Campisi, J., 1997. Eur. J. Cancer 33, 703.
 eþvx0 =D eðxþx0 þvtÞ =4Dt
. ðA:2Þ Campisi, J., 2001. Trends. Cell Biol. 11, S27.
Carey, J.R., Liedo, P., Orozco, D., Vaupel, J.W., 1992. Science 258, 457.
The second term represents the ‘‘image’’ contribution; Ferrari, P.A., Marić, N., 2006. math.PR/0605665.
notice that the bias velocity of the image is in the same Fleming, W.H., Viot, M., 1979. Indiana Univ. Math. J. 28, 817.
direction as that of the initial particle. The exponential Gikhman, I.I., Skorokhod, A.V., 1969. Introduction to the Theory of
prefactor in the image term ensures that the absorbing Random Processes. Dover, New York.
Goodwin, E.H., Blagoev, K.B., 2004. In: Proceedings of the Second Cold
boundary condition cðx ¼ 0; tÞ ¼ 0 is always fulfilled.
Spring Harbor Laboratory Meeting on Molecular Genetics of Aging.
From this concentration profile, the first-passage prob- Gradshteyn, I.S., Ryzhik, I.M., 1965. Tables of Integrals, Series, and
ability, namely, the probability for a diffusing particle to Products. Academic Press, New York.
hit the origin for the first time at time t, is Harley, C.B., 1991. Mutat. Res. 256, 271.
 Hasenmaile, S., Pawelec, G., Wagner, W., 2003. Biogerontology 4, 253.
@c  x0 2 Hayflick, L., 1965. Exp. Cell Res. 37, 614.
F ðtÞ ¼ D  vc ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffi eðx0 vtÞ =4Dt . (A.3) Henson, J.D., Neumann, A.A., Yeager, T.R., Reddel, R.R., 2002.
@x x¼0 4pDt 3
Oncogene 21, 598.
The convective contribution to the flux, vc, gives no van Kampen, N.G., 1997. Stochastic Processes in Physics and Chemistry,
contribution because c ¼ 0 at x ¼ 0. Until the particle hits second ed. North-Holland, Amsterdam.
Karlin, S., Taylor, H., 1975. A First Course in Stochastic Processes.
the boundary it stays in the system; hence, its survival Academic Press, New York.
probability is simply Nisbet, R.M., Gurney, W.S.C., 1982. Modelling Fluctuating Populations.
Z 1    Wiley, New York.
1 vt  x0 Olovnikov, A.M., 1971. Dokl. Akad. Nauk SSSR 201, 1496.
SðtÞ ¼ cðx; tÞ dx ¼ erfc pffiffiffiffiffiffiffiffi
0 2 4Dt Probstein, R.F., 1994. Physicochemical Hydrodynamics, second ed. Wiley,
  New York (to be referred for a general discussion on the meaning of
vx0 =D vt þ x0
e erfc pffiffiffiffiffiffiffiffi , ðA:4Þ the Péclet number).
4Dt Reddel, R.R., 2003. Cancer. Lett. 194, 155.
Redner, S., 2001. A Guide to First-Passage Processes. Cambridge
where erfcðzÞ is the complementary error function. Since University Press, New York.
there is only one absorbing point in the system, the survival Steinsaltz, D., Evans, S.N., 2007. Theor. Pop. Biol. 71, 473.
probability and the first-passage probability are related by Vaupel, J.W., 1997. In: Wachter, K.W., Finch, C.E. (Eds.), Between Zeus
and the Salmon, The Biodemography of Longevity. National
dSðtÞ Academies Press, Washington, DC, 1737pp (Chapter 2).
F ðtÞ ¼  . (A.5) Watson, J.D., 1972. Nat. New Biol. 239, 197.
dt
Weitz, J.S., Fraser, H.B., 2001. Proc. Natl Acad. Sci. USA 98, 15383 In
From the asymptotics of the error function (Abramowitz their notation the plateau value, namely, the asymptotic hazard rate, is
2 pffiffiffi
and Stegun, 1970), erfcðzÞez = p z, the long-time 2 =2s2 .