Phenotypic Evolutionary Models in Stem Cell Biology:

Replacement, Quiescence, and Variability

Marc Mangel1*, Michael B. Bonsall2
1 Center for Biomolecular Science and Engineering, Department of Applied Mathematics and Statistics, University of California Santa Cruz, Santa Cruz, California, United
States or America, 2 Mathematical Ecology Research Group, Department of Zoology, University of Oxford, Oxford, United Kingdom

Abstract
Phenotypic evolutionary models have been used with great success in many areas of biology, but thus far have not been
applied to the study of stem cells except for investigations of cancer. We develop a framework that allows such modeling
techniques to be applied to stem cells more generally. The fundamental modeling structure is the stochastic kinetics of
stem cells in their niche and of transit amplifying and fully differentiated cells elsewhere in the organism, with positive and
negative feedback. This formulation allows graded signals to be turned into all or nothing responses, and shows the
importance of looking beyond the niche for understanding how stem cells behave. Using the deterministic version of this
framework, we show how competition between different stem cell lines can be analyzed, and under what circumstances
stem cells in a niche will be replaced by other stem cells with different phenotypic characteristics. Using the stochastic
version of our framework and state dependent life history theory, we show that the optimal behavior of a focal stem cell will
involve long periods of quiescence and that a population of identical stem cells will show great variability in the times at
which activity occurs; we compare our results with classic ones on quiescence and variability in the hematopoietic system.
Citation: Mangel M, Bonsall MB (2008) Phenotypic Evolutionary Models in Stem Cell Biology: Replacement, Quiescence, and Variability. PLoS ONE 3(2): e1591.
doi:10.1371/journal.pone.0001591
Editor: Mark Rees, University of Sheffield, United Kingdom
Received January 9, 2008; Accepted January 16, 2008; Published February 13, 2008
Copyright: 2008 Mangel, Bonsall. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits
unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: This work was supported by the NSF, Royal Society, and the Astor Travel Fund, University of Oxford.
Competing Interests: The authors have declared that no competing interests exist.
*E-mail: msmangel@ucsc.edu

It is generally agreed that a focal stem cell in its niche is

influenced by signals (both positive and negative feedback) from
the other stem cells in the niche, the support cells in the niche, and
from more differentiated cells throughout the body. How exactly
to model this is unclear, with the consequence that most models of
stem cells focus on a single stem cell (or a population of identically
behaving stem cells) in a single niche. We show how to make
operational the recommendation in [9] that the next advances will
combine probability-based events and mechanistic parameters as
the best approximation to actual phenomena. Our models are
rooted in evolution by natural selection, which is ultimately the
mechanism for all of biology.
Theoretical approaches to stem cell biology have usually involved
typological thinking [10], although there are examples of population
thinking [11]. Typically, one assumes that all stem cells behave in a
similar manner and writes either a set of deterministic differential
equations or a probabilistic master equation [12] to characterize the
dynamics of stem cells, but ignores natural selection. What is now
required is a kinetic theory of stem cell dynamics that accounts for
variation and natural selection [13]. We offer the first such theory
here. We also show how the biological version of the Heisenberg
uncertainty principle [14,15] can be made operational to show how
manipulations may alter the behavior and progeny of stem cells. To
achieve our goals, we use a somewhat stylized model, motivated by
the hematopoietic system. There is much to be learned from such
stylized models, in systems biology [16], ecology [17,18], and
evolution [1921].
September 2007 was the 50th anniversary of the publication of
the first successful demonstration of the intravenous infusion of

Introduction
The current enthusiasm for stem cells (both adult and embryonic)
[1], and their role in regenerative medicine is based on the
assumption that we can remove stem cells from their natural habitat,
propagate them in culture, transplant them into a foreign
environment and assume that the transplanted cells will do as we
wish or that we can manipulate them in vivo with desired results [2].
For example, stem cells can differentiate into cell lineages that are
different from the tissue in which they originate and adult stem cells
can migrate to sites of injury [3]. Harold Varmus has said that It is
not unrealistic to say that [stem cell] research has the potential to
revolutionize the practice of medicine and improve the quality and
length of life (pg. 513 in [4]). But long before that, Theodosius
Dobzhansky said that us nothing makes sense in biology except in
the light of evolution [5, pg 449].
There may be enormous differences between what stem cells do
in their original tissue and what they can and will do when put into
culture or when transplanted to a new location [6]. Raff [7] noted
that perhaps the greatest challenge in stem cell biology is to
uncover themechanisms that determine whether a daughter of a
stem cell division self-renews or commits to a particular pathway of
differentiation. Cracking this problem for the adult mammalian
stem cells of interest will be a crucial step for both developmental
biology and cell therapy. Solutions in this area of stem-cell
biology will require all sorts of different biology, and innovative
ideas. Elsewhere [8] we have argued that the time is right for the
development of evolutionary theory about stem cells and the
microenvironment (their niche) that maintains them.
PLoS ONE | www.plosone.org

February 2008 | Volume 3 | Issue 2 | e1591

Models of Stem Cells

bone marrow in patients who had experienced radiation or

chemotherapy that destroyed bone marrow [22,23]. In a
subsequent classic study of the hematopoietic stem cell system
[24], stem cell colony forming units showed enormous epigenetic
variation, leading to highly overdispersed numbers of descendant
cells in individual colonies in spleens of experimental animals. We
will show how to interpret this result in terms of variability in the
cell cycle [25]; this begs the question of why such variation has not
been eliminated, or alternatively how natural selection maintains
such variation.
In its niche, a stem cell responds to its local environment (signals
received from the support cells of and the other stem cells in the
niche) and its global environment (signals received from more
differentiated cells) but we assume that the behavior of the focal
stem cell does not affect the behavior of the other stem or more
fully differentiated cells. We are thus able to avoid, at least initially,
game theoretical aspects of stem cell biology; ultimately they must
be considered. Even with this simplified framework, we are able to
understand how natural selection will act on competition between
stem cells with different phenotypic characteristics (explained
below) and how natural selection can lead to both quiescence and
great variation in the behavior of stem cells.

Figure 1. The feedback network affecting a stem cell in its

niche. Here stem cells are clear circles, transit amplifying cells colored
circles, and fully differentiated cells in black. A focal stem cell has
positive feedback (+) on itself but shares inhibition and negative
feedback (2) with other stem cells. Transit amplifying cells receive
positive feedback from stem cells and exert positive feedback on fully
differentiated cells. Both transit amplifying and fully differentiated cells
exhibit negative feedback on stem cells. The symbols that characterize
these processes are explained in the Materials and Methods.
doi:10.1371/journal.pone.0001591.g001

Overview of the Models

Stem cells have the ability to both self-renew and to differentiate.
The niche of the stem cell is the local microenvironment that
supports the maintenance, renewal, and differentiation of stem cells.
In the case of the villi of the small intestine or bone marrow it is easy
to conceive of the niche as a defined physical environment. In other
cases, such as the epidermis, the niche will be determined by the
spatial extent of signaling molecules. Signals and feedback controls
(both negative and positive) affect the pattern of behavior of the stem
cells in their niche. These signals include cell-autonomous promotors
(such as piwi); self-feedback, autocrine and paracrine influences;
hormones, growth factors, cytokines, and short-range cell to cell
signalling pathways (e.g. Notch, Hedgehog, Wnt); hormones from
remote sources; positional cues from chemical gradients as in
classical Turing systems [26]; and adhesion [27]. Thus, stem cells are
not simply single input-output systems, but are complex, nonlinear,
multiple input-output systems [28].
The simplest population biology that one can imagine is to focus
on the number of stem S(t), transit amplifying progenitor A(t), and
fully differentiated D(t) cells at time t associated with the stem cells
in the niche being modeled. In doing so, for simplicity we
compress the differentiation stages of a transit amplifying cell, but
they could be unpacked should one wish to do so. In addition,
there will be other transit amplifying and fully differentiated cells,
produced by other stem cells, that create an additional signaling
background. We denote those cells by B. For simplicity we
consider only one kind of progenitor cell and only one kind of
differentiated cell, but extensions are clear and obvious. The
positive and negative feedback links affecting stem cell fate are
illustrated in Figure 1.

In fact, there are many more possible transitions of stem cells and
their progeny. We use the following indexing system for transitions of
stem, transit amplifying and fully differentiated cells: 1) a stem cell
divides symmetrically, 2) a stem cell divides asymmetrically, 3) a stem
cell divides and differentiates symmetrically producing two transit
amplifying cells, 4) a stem cell under goes apoptosis or migrates out of
the niche, 5) a transit amplifying cell divides, 6) a transit amplifying
cell undergoes apoptosis, 7) a transit amplifying cell fully
differentiates, 8) a fully differentiated cell undergoes apoptosis. To
fully understand the evolutionary ecology of stem cells we need to
consider the entire range of possibilities of transitions and both
positive and negative feedback. This is implicit in [32]. In
Materials and Methods we describe how to model these
transitions, implement the model and associate a system of ordinary
differential equations with the conditional means of the stochastic
numbers of stem, transit amplifying, and fully differentiated cells.

Trade offs
The transitions of stem, transit amplifying, and differentiated
cells are characterized by a number of parameters. Biological
realities will constrain some choices for parameter values. For
example, we may anticipate that the probability that all stem cells
in the niche either undergo apoptosis or migrate out of the niche is
not too high since if it were the system could not persist. That is, if
the niche empties (either through apoptosis or migration) it has
essentially gone extinct. Similarly, in most situations, the
population of transit amplifying cells is much more active than
the population of stem cells (with consequence that the number of
amplifying cells will thus be larger than the number of stem cells)
and we require parameter values that make this so. We also
require that when there is just one stem cell in the niche, the
numbers of stem cells increase.
In addition, we may expect there to exist links and trade-offs
between the different parameters. For example, since symmetric
renewal, asymmetric renewal and symmetric differentiation
involve many of the same signals and processes, we may expect
that their rates are correlated. Cells that undergo high rates of
replication are more likely to have errors in them and thus are

The transitions of stem, transit amplifying, and fully

differentiated cells
Most simply, stem cells divide symmetrically doubling and
producing two stem cells, divide asymmetrically doubling and
producing one stem cell and one cell that will be a progenitor of a
fully differentiated cell, undergo apoptosis (to maintain an errorfree genome [29,30]) or migrate out of the niche [31]. Almost all of
the literature focuses on whether stem cells will divide symmetrically or asymmetrically.
PLoS ONE | www.plosone.org

February 2008 | Volume 3 | Issue 2 | e1591

Models of Stem Cells

likely to undergo apoptosis. When transit amplifying cells

disappear at a higher rate because of replication error, fewer of
them will become fully differentiated cells and this too can be
captured by a trade-off.
In summary, one may conceive of the parameter set describing
the transitions of the stochastic kinetics as the phenotype of a stem
cell and its descendants, or if all of the stem cells in a niche are
identical descendants of a single cell, then these parameters are the
phenotype of the stem cell clone. These trade-offs are fully
described in Materials and Methods.

Table 1. Parameters, Their Interpretations, and Values.

Parameter

Interpretation

Value

Parameter of feedback inhibition of stem cells on

each other (Larger K implies less inhibition)

10
0.10

l1

Cell cycle rate for symmetric renewal

l2

Cell cycle rate for asymmetric renewal

0.20

l3

Cell cycle rate for symmetric differentiation

0.01

m40

l1 independent per capita rate of stem cell migration

and apoptosis

0.01

m41

l1 dependent per capita rate of stem cell migration

and apoptosis

0.1

The focal stem cell in the niche

To understand fully how natural selection acts on a stem cell, we
need to consider the fitness of a focal stem cell. Stem cells (and
transit amplifying cells) do not by themselves achieve fitness.
Rather, they support the organism which can be viewed as an
organized collection of fully differentiated cells. Fitness of the focal
stem cell depends upon what it does, what the other stem cells in
the niche do, and how many transit amplifying and differentiated
cells are present. We thus define a fitness function F(y, s, a, d) to be
the maximum expected accumulated fitness through differentiated
cells, given that the focal stem cell has accumulated y resources
towards the next division, that there are s stem cells in the niche
and that there are a and d transit amplifying and differentiated
cells sending signals back to the focal stem cell. State dependent
life history theory, as implemented through stochastic dynamic
programming [3336], allows us to compute the fitness-maximizing responses of the focal stem cell.

Results
A variety of parameters are used in this section; they are
explained in Table 1 and constitute the phenotype of the stem cell
and its descendants. For modeling, it is generally easier to work
with the rates of cell cycles (generically, l) rather than cell cycle
times (generically T).

0.05

m61

l1 dependent rate of disappearance of transit

amplifying cells

0.5

w0

l1 independent probability that a transit amplifying

cell differentiates

0.9

w1

l1 dependent reduction in w

1.0

m8

Mortality rate of fully differentiated cells

.04

Parameter for feedback inhibition from amplifying

and differentiated cells

0.001

Parameter for feedback inhibition from amplifying

and differentiated cells

0.01

Background level of transit amplifying and fully

differentiated cells

100

dt

Time step for iterating the equations

0.01

yd

Threshold level of resources needed for differentiation 5.5

cc

Sensitivity of error correction to resources



exp {b0 A ~b1 A

A special case of the differential equations

3.0


m
m {l5 l1 expm4 =l1
where b0 ~e 1zw 6 and b1 ~ 6
.
m8
l2 m4 K
This special case also serves as a check of our numerical
methods.

If the rate of symmetric differentiation is 0, we can solve for the

steady state levels of stem and differentiated cells explicitly and
easily obtain that of the amplifying cells numerically. In particular,
the steady state number of stem cells is



m
S~K :exp { 4
1
l1

Qualitative properties of the stochastic system

As described in the Materials and Methods, under very
general conditions, S = 0 is dynamically unstable and has
maximum value K (determined by feedback inhibition). Thus,
the solution of the set of differential equations will move from S = 1
(the minimum non-zero value of the number of stem cells in the
niche) towards K, determined by the balance of m4 and l1. In the
full stochastic simulation, all the same properties hold, as can be
seen from the equations characterizing the transition rates. One
important consequence of S = 0 being dynamically unstable is that
when a niche goes extinct, it will do so quickly [25]. Thus, we
expect extant clones of stem cells to have much longer lifetimes
than those of clones heading for extinction. Similarly, the number
of transit amplifying cells is determined by the balance of m62l5

which shows that the steady state number of stem cells increases
with increasing values of l1 and with decreasing values of m4
(Table 1), both of which accord with intuition. If the steady state
number of transit amplifying cells were known, the steady state
number of differentiated cells is
2

Thus, by adjustment of any of w, m6 or m8 (Table 1), natural

selection can lead to different numbers of fully differentiated cells
without changing the fundamental structure of the system of stem,
PLoS ONE | www.plosone.org

0.05

l1 independent rate of disappearance of transit

amplifying cells

transit amplifying and differentiated cells. That is, we can create a

ratio of fully differentiated to stem cell that ranges from the
observed 3:1 to 25,000:1 through the evolution of phenotypic
characteristics.
We obtain a single nonlinear equation for the steady state level
of transit amplifying cells

We begin with another look at Till et al [24]. Their data lead to

a frequency distribution for cell cycle times that is shown in
Figure 2. We are again led to ask how natural selection maintains
such a broad and slowly declining distribution.

wm6
A:
m8

Per capita rate of transit cell amplification

doi:10.1371/journal.pone.0001591.t001

Another look at Till et al (1964)

D~

l5
m60

February 2008 | Volume 3 | Issue 2 | e1591

Models of Stem Cells

Figure 2. The distribution of cell cycle times inferred from the

overdispersed data of Till et al. This should be compared to Figure 6.
doi:10.1371/journal.pone.0001591.g002

and w. The number of fully differentiated cells is determined by m8,

w, and m6 thus a variety of simple mechanisms can create tissues
with different structures.

Quantitative results
In Figure 3, we show the trajectories for transit amplifying
(Figure 3, upper panel) and fully differentiated (Figure 3, lower
panel) cells and the associated solution of the differential equation
system. We see that the set of differential equations is a good proxy
for the mean trajectory of the stochastic system and that the
number of fully differentiated cells has a trajectory that mimics
that of the transit amplifying cells, as it must. In Figure 4, we show
the realizations of the stochastic system as phase plane plots of the
density of stem cells and transit amplifying cells (upper panel) or
stem cells and fully differentiated cells (lower panel). We see that a
wide-range of cell numbers is expected due to the inherent
stochasticity in the system.
In Figure 5, we show two results of the invasion/replacement
analysis. First (upper panel) we consider the invasion of a resident
population of stem cells by a new phenotype
 with a different value
of cell cycle rate for symmetric renewal l
1 . Here we see that in
general if l1 exceeds l
1 then the residents will resist the invasion by
the new stem cell type and in general if the converse is true then
the residents are excluded by the invading stem cell. However,
there is a small region in which both kinds of stem cells coexist.
Second (lower panel) we consider the replacement analysis when
the value of the rate of the cell cycle for symmetric differentiation
l3 varies between the resident and invading stem cells. Here the
region of coexistence is greater and the boundary curves, while still
symmetrical are more complicated.
In Figure 6 we show the frequency distributions of transitions by
the focal stem cell following the behavior determined from the
stochastic dynamic programming model. These data refer to
22,200 simulations in which there are about 3507 transitions of the
simulated stem cells, of which only 6.42 per cent were symmetric
renewal; the remaining transitions were asymmetric renewal. The
minimum value of resources needed for a transition is yd = 5.5, but
PLoS ONE | www.plosone.org

Figure 3. The stochastic trajectories for 100 simulations of the

stem-transit amplifying-differentiated system described in the
text and the solution of the differential equations 2527 (red
line).
doi:10.1371/journal.pone.0001591.g003

in the simulated stem cell population the average value of

resources at the time of transition is 7.1, and the variance of those
resources is 0.84. The average number of transit amplifying and
background cells at the time of transition is 11.1, and the variance
in those numbers is 53.5.

Discussion
Phenotypic models of evolutionary processes have made
important and long-lasting contributions in ecology and evolutionary biology [3739] and in the study of cancer [20,40]. They
offer promise to help us understand great swaths of the life history
and dynamics of stem cells and their descendants. Most of our
work in this paper is in developing the population biology of stem,
transit amplifying and fully differentiated cells in their niche to
provide a fundamental structure for the analyses shown in
Figures 5 and 6. This structure provides the framework for
answering a wide range of questions about the evolutionary
biology of stem cells. The results of Figure 5 show that it is possible
to use evolutionary replacement analyses to understand conse4

February 2008 | Volume 3 | Issue 2 | e1591

Models of Stem Cells

Figure 4. The results of the stochastic simulation can be shown

as phase plots of stem cells and transit amplifying cells (uppere
panel) or stem cells and fully differentiated cells (lower panel)
doi:10.1371/journal.pone.0001591.g004

Figure 5. Deterministic replacement analysis allow us to

determine when a mutation in one of the phenotypic
parameters or the introduction of stem cells with different
phenotypic parameters into a niche will lead to replacement of
one stem cell clone by another, or coexistence of clones in the
same niche. Upper panel) Stem cells differ in the cell cycle rate for
symmetric renewal l1. Lower panel) Stem cells differ in the cell cycle
rate for symmetric differentiation l3.
doi:10.1371/journal.pone.0001591.g005

quences of manipulation of stem cell life history parameters in vivo

and show, for example, that not all manipulations will be
successful (the invading stem cell line cannot persist) or perhaps
may be too successful (the invading stem cells fully replace the
resident ones). We have not allowed mutations in the phenotypic
parameters of the stem cell clones, but this is possible too with our
methods [38] and would allow one to investigate how the clone
changes over time due to mutation.
Figure 6 shows that it is possible to understand the key
properties of stem cells quiescence and great variability in
activity in terms of natural selection acting on the decisions of
stem cells in response to the signals from the local microenvironment (the other stem cells) and from the more differentiated cells in
the rest of the organism. Although it is comforting to see the
similarity of Figure 2 and 6, one should remember that pattern
does not imply process. However, the mechanistic model that
underlies Figure 6 will allow us (elsewhere) to explore the key
components of the process that lead to the pattern.
An empirical goal suggested by our work would be to more
carefully identify these trade-offs and the feedback functions
PLoS ONE | www.plosone.org

(described in detail in Materials and Methods) that characterize asymmetric renewal and symmetric differentiation. We have
used simple exponential functions, but others in which the all-ornothing response is built in are clearly possible [16]. How the
strength of these signals will evolve also remains an open question.
Our work also provides a number of qualitative insights. For
example, in healthy organisms, we should not expect the stem
populations to be at their maximum sizes. Rather, the stochastic
processes of birth and death will lead to population sizes smaller
than the maximum. This, of course, makes identifying pathological
situations more difficult, but should also help us avoid leaping to
inappropriate conclusions. Furthermore, one cannot conceive of
the transitions of stem cells (quiescence, symmetric renewal,
asymmetric renewal, symmetric differentiation, apoptosis, or
5

February 2008 | Volume 3 | Issue 2 | e1591

600
400
0

200

Frequency

800

1000

Models of Stem Cells

0.42

2.11

3.8

6.94

10.55

19.97

46.78

73.58

113.79

Cell cyle time midpoint

Figure 6. The forward simulation of the decisions by the focal stem cell allows us to generate the frequency distribution of the rates
of transition, comparable to Figure 2.
doi:10.1371/journal.pone.0001591.g006

migration) in the absence of feedback from the amplifying transit

cells and the fully differentiated cells. Indeed, they are crucial to
what happens in the niche. We have also shown that a single
framework can account for the observed ratio of fully differentiated to stem cells [41] by varying the phenotypic characteristics.
Many questions remain and there is much work to be done,
ranging from the development of systems biology models for the
phenotypic parameters, to the evolution of the signalling functions
and their parameters, to the development of more complicated
game theoretical aspects. Each will provide new insight into the
evolutionary ecology of stem cells and their niches.

to be small. As noted before, o(Dt) denotes terms that are higher

powers of Dt (typically Dt2, Dt3 etc.).
Symmetric Renewal.

PrDS~1,DA~0,DD~0jSt~s,At~a,Dt~d
~r1 s,a,d DtzoDt
r1 s,a,d ~l1 :s:w1 s

In this equation, l1 = 1/T1 is the cell cycle time for stem cells
that renew symmetrically. We assume no variation in this rate.
The function W1(s) characterizes the inhibition of stem cells upon
each other [43]. A variety of function forms are possible [16] but
the key is that W1(s) decreases as s increases (which captures the
inhibition); for computations we use W1(s) = ln(K)2ln(s).

Materials and Methods

Another look at Till et al 1964
Till et al. [24] recognized that overdispersed data (in which the
variance is much larger than the mean) such as theirs could not be
described by a Poisson distribution. Instead, they fit a gamma
distribution [25] to their data. The Poisson process is equivalent to
the assumption that the probability of a stem cell being active in the
next dt is lDt+o(Dt), where o(Dt) indicates times that are higher powers
of Dt. An improvement upon their approach is the following. It is
known that there is variation in cell cycle times [42]. Thus, rather
than fitting the counts with a gamma density, we assign a gamma
density to l, leading to a negative binomial distribution for the
counts [25]. We use the method of moments to infer the parameters
of this gamma density. Since the cell cycle time T = 1/l, it is then an
elementary calculation to determine the frequency distribution of cell
cycle time, given the inferred gamma density.

Asymmetric Renewal.

PrDS~0,DA~1,DD~0jSt~s,At~a,Dt~d ~
r2 s,a,d DtzoDt
r2 s,a,d ~l2 :s:W1 s:W2 a,d,b

6
7

In analogy to Eqn (5), l2 is the cell cycle rate when a stem cell renews
asymmetrically, with associated cell cycle time T2. Inhibition of
symmetric renewal occurs for two reasons. First, negative feedback
from stem cells on stem cells decreases overall activity. Second,
negative feedback from existing transit amplifying and fully
differentiated cells inhibits the production of transit amplifying
cells. Thus W2(a,d,b) characterizes the signal from the transit
amplifying and fully differentiated cells to the stem cells. Once
again, this function will decrease as the number of transit amplifying
and fully differentiated cells increases. For computations or analysis
we use W2(a,d,b) = e2s(a+d+b) which is equivalent to a Hill function

The stochastic kinetics of transitions

We use s, a, d for particular values of the number of stem, transit
amplifying, or fully differentiated cells. We characterize the
feedback network in Figure 1 as follows. We let DS, DA and DD
denote the change in the number of stem, amplifying, or
differentiated cells in the interval of time Dt, which is assumed
PLoS ONE | www.plosone.org

February 2008 | Volume 3 | Issue 2 | e1591

Models of Stem Cells

when the argument is small. e measures the sensitivity of the stem cell
population to what is happening in the population of transit
amplifying and fully differentiated cells. Stem cells convert graded
stimuli into all or nothing responses [44] and we have specifically
chosen a graded rather than all or nothing response [16] because
that allows the all or nothing response to emerge, rather than to be
built into the model.

as in
PrDS~0,DA~{1,DD~1jS t~s,At~a,Dt~d ~
r7 s,a,d DtzoDt
r7 s,a,d ~w:m6 :a

16
17

Symmetric Differentiation.

PrDS~{1,DA~2,DD~0jS t~s,At~a,Dt~d ~
r3 s,a,d DtzoDt
r3 s,a,d ~l3 :s:W1 s:W3 a,d,b

Note that r6+r7 = m6a, the total rate at which transit amplifying
cells disappear.
Mortality of Differentiated Cells. Finally, fully differentiated cells die, so that

8
9

PrDS~0,DA~0,DD~{1jS t~s,At~a,Dt~d ~
r8 s,a,d DtzoDt

Here l3 and W3(a,d,b) are analogous to those above but can be

understood as a more desperate signal. For computations or
analysis we use W3(a,d,b) = e2c(a+d+b) where c is the sensitivity
parameter for symmetric differentiation.
Migration/Apoptosis. From the viewpoint of the focal stem
cell, whether another stem cell undergoes apoptosis or migrates out
of the niche (which is surely important for the state of the organism) is
immaterial the consequence is the same, a reduction in the number
of stem cells in this niche. Thus we combine them in
PrDS~{1,DA~0,DD~0jSt~s,At~a,Dt~d ~
r4 s,a,d DtzoDt
r4 s,a,d ~m4 :s

r8 s,a,d ~m8 :d

r5 s,a,d DtzoDt
r5 s,a,d ~l5 :a

State dynamics using Gillespies t method

We use the Gillespie t-method [4750] to convert from rates of
transitions in a small interval of time Dt to the dynamics of the
populations of cells. We assume that the time to the next transition
of the system, t, is a random variable with exponential distribution
and mean time determined by the rate of all processes. The overall
rate is

10
11

Rs,a,d ~

r6 s,a,d DtzoDt

8
X

ri s,a,d

20

i~1

so that
Prtvt~1{exp{Rs,a,d t

21

Given the current values of S(t), A(t) and D(t) we first compute the
time of the next transition by comparing the right hand side of
Eqn 21 with a uniformly distributed random variable.
Next, given that a transition has occurred we compute

12
13

fi s,a,d ~Prtransition i occured, given a transition

Here l5 is the cell cycle rate for transit amplifying cells. As mentioned
above, we have compressed the n stages of development from transit
amplifying to fully differentiated cell; unpacking this assumption can
be done through a linear chain [38,45,46].
Transit Cell Mortality. Transit amplifying cells will
disappear because of apoptosis (e.g. because they have too many
DNA errors) and because they fully differentiate. We let m6 denote
the total per capita rate at which transit amplifying cells disappear;
a fraction w of them are converted to differentiated cells and the
remainder experience apoptosis. Thus the mortality of transit
amplifying cells is
PrDS~0,DA~{1,DD~0jS t~s,At~a,Dt~d ~

19

Here m8 accounts for all sources of mortality of fully differentiated

cells.

Here m4 measures the rate of migration of stem cells out of the niche
plus the rate of apoptosis, both on a per stem cell basis.
Transit Cell Amplification. Transit cells amplify and since
they send a signal back to the niche, their dynamics must be
included. Hence
PrDS~0,DA~1,DD~0jS t~s,At~a,Dt~d ~

18

has occurred

22

Since when time t has elapsed one of the transitions has occurred,
we have
fi s,a,d ~

ri s,a,d
Rs,a,d

23

Thus, the fi are the probability density function for the transitions
of the system. It is also useful in numerical implementation to deal
with the cumulative distribution function Fi(s,a,d) defined by
Fi s,a,d ~

14

i
X

fj s,a,d

24

j~1

r6 s,a,d ~1{w:m6 :a

15

Associated differential equations

The numbers of stem, transit amplifying, and fully differentiated
cells are random variables. The rates of transitions of stem cells are
nonlinear functions of the states, thus it is a nontrivial matter to

Transit amplifying cells that

disappear but do not die are converted into fully differentiated cells
Transit Cell Differentiation.

PLoS ONE | www.plosone.org

February 2008 | Volume 3 | Issue 2 | e1591

Models of Stem Cells

For example, the disappearance of stem cells from the niche is

decomposed into terms characterizing migration to other niches
(m40) and mortality due to apoptosis. The second equation above
allows us to separate the disappearance of transit amplifying cells
into those potentially converting to differentiated cells and those
suffering apoptosis.
When transit amplifying cells disappear at a higher rate because
of replication error, fewer of them will become fully differentiated
cells. This can be captured by a trade-off between w and l1 as in

connect the transition rates to ordinary differential equations for

the states [5156]. It is perhaps easiest to associate differential
equations with the transitions if we interpret the differential
equations as the mean of conditioned kinetics [56]. Thus, taking
expectations and following the method of [56] we associate the
following differential equations with the transitions described in
the previous section
dS
~S:W1 S:l1 {l3 :W3 A,D,B{m4 :S
dt

25

w~w0 {w1 :l1

dA
~S :W1 S :l2 :W2 A,D,Bz2:l3 :W3 A,D,B
dt
zl5 {m :A

26

dD
~w:m6 :A{m8 :D
dt

27

In summary, we may think of the parameter set

[l1,l2,l3,m40,m41,l5,m60,m61,w0,w1,m8] as the phenotype of a stem
cell and its descendants, or if all of the stem cells in a niche are
identical descendants of a single cell, then these parameters are the
phenotype of the stem cell clone. But, of course, all the stem cells
in a niche need not be conspecifics. We thus turn to the
competition between different stem cell phenotypes.

This set of ordinary differential equations will allow us to develop a

number of novel insights about stem cells, their niche, and
replacement dynamics of stem cells in their niche.

Coexistence, replacement, and invasion

We can use evolutionary replacement analysis [38,5761] with
the system of ordinary differential equations to explore the
competition between and modification of stem cell clones and
when manipulation of stem cell transition parameters will be
successful. We now use i and j to index the stem cell clone with
different phenotypic parameters, let cji denote the influence of
transit amplifying and fully differentiated cells of type j on stem cell
clone i so that

Biological constraints on parameter values

The transitions of stem, transit amplifying, and differentiated cells
are characterized by a number of parameters. Biological realities will
constrain some choices for parameter values. For example, we may
anticipate that the probability that all stem cells in the niche either
undergo apoptosis or migrate out of the niche is not too high since if
it were the system could not persist. That is, if the niche empties
(either through apoptosis or migration), it has essentially gone
extinct. In computation, we choose parameters so that the
probability of extinction of the niche is less than 20 percent.
Similarly, in most situations, the population of transit amplifying
cells is much more active than the population of stem cells (with
consequence that the number of amplifying cells will thus be much
larger than the number of stem cells). For purposes of computation
we use the condition r5.4r1.
We require that the number of support cells is sufficient such
that when there is just one stem cell in the niche, the numbers of
stem cells increase. In terms of the associated ordinary differential
dS
equations, this is equivalent to
w0 when S = 1. Since W3(A,D,B)
dt
will be a decreasing function of A, D and B, it follows from Eqn 25
that we require
m4
lnK w
l1 {l3 W3 0,0,0

!
!!
X
X
X
1 dSi
~W1
Sj : l1i {l3i :W3
cji Aj ,
cji Dj ,B
Si dt
32
j
j
j
{m4i
!
!
X
X
X
dAi
:
:
~Si W1
Sj
cji Aj ,
cji Dj ,B
l2i W2
dt
j
j
j
!!
X
X
:
:
z2 l3i W3
c Aj ,
c Dj ,B
zl5i {m :Ai
ji

m6 ~m60 zm61 :l1

30

PLoS ONE | www.plosone.org

6i

34

In these equations, we assume that all stem cells share the

functional forms for the influence of transit amplifying and fully
differentiated cells on asymmetric renewal and symmetric
differentiation. After specifying the set of phenotypic parameters,
we can integrate the coupled differential equations to determine
the outcome of competition between any number of stem cell
clones that differ in their phenotypic parameters.
However, to illustrate the ideas and explain them in as simple a
manner as possible, consider the situation in which we wish to
manipulate some of the phenotypic parameters in a niche by
inserting a new stem cell with different parameters into an existing
clone. In that case, we need to consider the stem cells present (the
resident stem cells) and the new one (the invading stem cell). For
definiteness, let us assume that the rate of symmetric differentiation of the residents is l3 and of the invader is l
3 . We first solve

In addition, we may expect there to exist links and trade-offs

between the different parameters. For example, since symmetric
renewal, asymmetric renewal and symmetric differentiation
involve many of the same signals and processes, we may expect
that the cell cycle rates l1, l2, and l3 are correlated. Thus, for
computations, we use l2 = 2?l1 and l3 = 0.05?l2. Similarly, cells
that undergo high rates of replication are more likely to have
errors in them and thus be more likely to undergo apoptosis. The
simplest version of that trade-off is a linear one in which
29

ji

33

dDi
~wi :m6i :Ai {m8i :Di
dt

28

m4 ~m40 zm41 :l1

31

February 2008 | Volume 3 | Issue 2 | e1591

Models of Stem Cells

variable describing the time to the next transition, and the vector
D = (DS,DA,DD) that describes the changes in stem, transit
amplifying and differentiated cells when a transition occurs.
When this transition occurs, the focal stem cell has level of
resources y. If y is less than the resources available for
differentiation, then the focal stem cell cannot do anything other
than remain quiescent. In that case

Eqns 2527 for the steady state numbers of stem, transit

amplifying, and fully differentiated
cells with the resident

phenotype which we denote by Sl3 ,Al3 ,Dl3 to focus
our attention on these steady state values as they depend upon the
value of l3 (clearly, they depend upon the other phenotypic
parameters too, but we only consider the manipulation of l3 here).
To make the notation simpler let us return to Eqn 25 and
rewrite it as

F y,s,a,d ~Et ED Rd :tzF yzW1 s:t,szDS,az

1 dS
35
~gS,A,Djl3
S dt


Then g S l3 ,Al3 ,Dl3 jl3 ~0 since those values correspond
to a steady state. Now imagine that we introduce
 a stem cell with
different phenotypic parameters. We let S
l
3 denote the
number of stem cells with symmetric differentiation rate l
3 . A
number of outcomes are possible.
 

 
First, it is possible that g S l3 zS
l
3 ,Al3 ,Dl3 l
3 v0



when S l3 ~1 (initially there are no transit amplifying or fully
differentiated cells descended from the invading stem cells). In this
case, the number of stem cells with symmetric differentiation rate
l
3 declines and we say that the resident stem cells resist invasion or
replacement.

 
Second,
it is possible
that g Sl3 zS
l
3 ,Al3 ,






Dl3 l3 w0 when S l3 ~1. In this case, the number of
invading stem cells will increase, and lead to transit amplifying and
fully differentiated descendants. One effect will be that the resident
stem cells will be disturbed from their steady state. Then their
numbers may decline, in which case we say that the invading stem
cell has replaced or excluded the resident stem cells. Alternatively,
after the perturbation by the invading them cell (and the production
of transit amplifying and fully differentiated cells from the invader)
the resident stem cells may remain steady or even increase. In that
case, we say that the two kinds of stem cells coexist.

DA,dzDD

which we define as V0, the fitness value of remaining quiescent.

The right hand side of this equation represents the future
accumulation of fitness and is averaged over the time of the next
transition and the change in the cell population when that
transition occurs.
If y exceeds yd, then in the simplest case the focal stem cell may
remain quiescent (with value V0 given above), renew symmetrically
(with value V1 to be determined) or renew asymmetrically (with
value V2 also to be determined). Allowing for full differentiation of
the focal stem cell, or migration out of the niche requires certain
complications which we will postpone for a later paper. After a
division resources y2yd remain for correcting errors in the
daughter cells. We thus let Y(y2yd) denote the probability that a
daughter cell is error free (i.e. does not undergo apoptosis), given
that y2yd resources are used for correcting errors. For computations we use Y(z) = 12exp(2ccz) where cc is a parameter.
If the stem cell renews symmetrically, we assume that the focal
stem cell remains error-free and a daughter stem cell joins the
other stem cells in the niche. Assuming segregation of errors upon
division [29,31], all resources remaining after division are used to
ensure that the daughter stem cell is error free. We thus have
V1 ~Et ED Rd :tzY y{yd F W1 sz1:t,szDSz1,az

The focal stem cell in its niche

DA,dzDDz1{Y y{yd F W1 s:t,szDS,azDA,dzDD

We now describe how one may characterize the behavior of a

focal stem cell in its niche. To be sure, thinking of a single stem cell
having behavior while others cells in the niche (and the transit
amplifying and fully differentiated cells) have a fixed behavior is a
simplification. Ultimately, one should view stem cells in their niche
as playing a dynamic cooperative game, but we reserve that for
subsequent work. For the time being, we focus on the state
dependent life history of a focal stem cell, determining the optimal
life history through stochastic dynamic programming [3336].
Like all cells, the focal stem cell must accumulate resources for
division. We denote by Y(t) the resources available for cell division
at time t, assuming that a minimum resource level yd. We assume
that the inhibition of the focal stem cell by other stem cells also
affects the flow of resources to the focal stem cell and has the same
functional form as used in the inhibition of cell cycling. Thus
dY
~q:W1 s:
dt

38

On the other hand, if the stem cell renews asymmetrically, a new

transit amplifying cell is produced and the resources remaining
after division are used to correct errors in the new transit
amplifying cell. In this case
V2 ~Et ED Rd :tzY y{yd F W1 s:t,szDS,azDAz

39
1,dzDDz1{Y y{yd F W1 s:t,szDS,azDA,dzDD
Thus, when y exceeds yd we have
F (y,s,a,d~maxV0 y,s,a,d ,V1 y,s,a,d ,V2 y,s,a,d 

40

Note that there is no explicit time in any of these equations. Thus,

they cannot be solved by the usual method of backward iteration
[3336]. Rather, we determine F(y,s,a,d) by value iteration (see [62]
for a general discussion and [63] for an application in behavioral
biology). When the value iteration has stabilized, we have found in
addition to fitness, the optimal decisions i*(y,s,a,d) of the focal stem
cell when its accumulated resources for division are y, there are s
other stem cells in the niche and the number of transit amplifying
and differentiated cells sending signals to the niche are a and d
respectively. We then use the Markov simulation described
previously to create the environment of the focal stem cell and
follow its subsequent behavior (quiescence, symmetric renewal, or
asymmetrical renewal) as the number of stem cells in the niche and

36

Here q can be scaled out by interpreting t as multiples of 1/q, so we

set it equal to 1.
We assume that the organism gains fitness from these fully
differentiated cells at rate RF(d) when there are d fully differentiated
cells sending signals to the focal stem cell. We assume that the
decision of the focal stem cell is made at the time of the next
transition of the stem, amplifying or differentiated cells. Using the
methods describe above, we compute t(s,a,d), which is the random
PLoS ONE | www.plosone.org

37

February 2008 | Volume 3 | Issue 2 | e1591

Models of Stem Cells

of the manuscript. Kate Richerson and Laith Yakob provided excellent

and timely help in the preparation of the final version.

the number of amplifying and differentiated cells change through the

stochastic processes described previously. In this way, we are able to
construct the times between transitions of the focal stem cell, and
thus the analogue of the distribution of Till et al. [24].

Author Contributions
Wrote the paper: MM MB. Other: Conceived and designed the models:
MM. Analyzed the models: MM. First proposed the problem: MM.
Worked extensively on the modeling and the paper: MB MM.

Acknowledgments
We thank Richard Gardner, Chris Graham, and Arthur Lander for very
useful discussions and Arthur Lander for comments on a previous version

References
33. Mangel M, Clark CW (1988) Dynamic modeling in behavioral ecology.
Princeton: Princeton University Press. 308 p.
34. Mangel M, Ludwig D (1992) Definition and evaluation of behavioral and
developmental programs. Annu Rev Ecol Syst 23: 507536.
35. Houston AI, McNamara JM (1999) Models of adaptive behavior. Cambridge:
Cambridge University Press. 392 p.
36. Clark CW, Mangel M (2000) Dynamic state variable models in ecology:
Methods and applications. New York: Oxford University Press. 289 p.
37. Kingsland SE (1985) Modeling nature: Episodes in the history of population
ecology. Chicago: University of Chicago Press. 324 p.
38. Bonsall MB, Mangel M (2004) Life-history trade-offs and ecological dynamics in
the evolution of longevity. Proc R Soc Lond B 271: 11431150.
39. Nowak MA (2006) Evolutionary dynamics. Cambridge: Harvard University
Press. 384 p.
40. Wodarz D, Komarova N (2005) Computational biology of cancer: Lecture notes
and mathematical modeling. Singapore: World Scientific. 268 p.
41. Potten CS, Wilson JW (2006) The development of epithelial stem cell concepts.
In: Lanza R, Gearhart J, Hogan B, Melton D, Pedersen R, et al, eds. Essentials
of stem cell biology. Amsterdam: Elsevier. pp 1121.
42. Painter PR, Marr AG (1968) Mathematics of microbial populations. Ann Rev
Microbiol 22: 519548.
43. Madhavi A, Davey RE, Bhola P, Yin T, Zandstra PW (2007) Sensitivity analysis
of intracelluar signaling pathway kinetics predicts targets of stem cell fate control.
PLoS Comput Biol 3: 111.
44. Davey RE, Onishi K, Mahdavi A, Zandstra PW (2007) LIF-mediated control of
embryonic stem cell self-renewal emerges due to an autoregulatory loop.
FASEB J 21: 20202032.
45. MacDonald N (1989) Biological delay systems: Linear stability theory.
Cambridge: Cambridge University Press. 256 p.
46. Mangel M, Bonsall MB (2004) The shape of things to come: using models with
physiological structure to predict mortality trajectories. Theor Popul Biol 65:
353359.
47. Gibson MA, Bruck J (2000) Efficient exact stochastic simulation of chemical
systems with many species and many channels. J Chem Phys 104: 1876.
48. Gillespie DT (2000) The chemical Langevin equation. J Chem Phys 113: 297.
49. Gillespie DT (2001) Approximate accelerated stochastic simulation of chemically
reacting systems. J Chem Phys 115: 1716.
50. Gillespie DT, Petzold LR (2003) Improved leap-size selection for accelerated
stochastic simulation. J Chem Phys 119: 8229.
51. Barbour AD (1974) On a functional central limit theorem for Markov
population processes. Adv Appl Probab 6: 2139.
52. Barbour AD (1980) Equilibrium distributions for Markov population processes.
Adv App Probab 12: 591614.
53. Kurtz TG (1970) Solutions of ordinary differential equations as limits of pure
jump Markov processes. J Appl Probab 7: 4958.
54. Kurtz TG (1971) Limit theorems for sequences of jump Markov processes
approximating ordinary differential processes. J Appl Probab 8: 344356.
55. Kurtz TG (1972) The relationship between stochastic and deterministic models
for chemical reactions. J Chem Phys 57: 29762978.
56. Mangel M (1994) Barrier transitions driven by fluctuations, with applications to
ecology and evolution. Theor Popul Biol 45: 1640.
57. Dieckmann U (1997) Can adaptive dynamics invade? Trends Ecol Evol 12:
128131.
58. Metz JAJ, Nisbet RM, Geritz SAH (1992) How should we define fitness for
general ecological scenarios? Trends Ecol Evol 7: 198202.
59. Vincent TL, Brown JS (2005) Evolutionary game theory, natural selection and
Darwinian dynamics. Cambridge: Cambridge University Press. 400 p.
60. Pielou EC (1977) Mathematical ecology. London: John Wiley and Sons. pp 385
p.
61. Mangel M, Kindsvater HK, Bonsall MB (2007) Evolutionary analysis of life
span, competition, and adaptive radiation, motivated by the Pacific rockfishes
(Sebastes). Evolution 61: 12081224.
62. Puterman ML (1994) Markov decision processes: Discrete dynamic stochastic
programming. New York: Wiley Interscience. 672 p.
63. Venner S, Chads I, Bel-Venner M-C, Pasquet A, Charpillet F, et al. (2006)
Dynamic optimization over infinite-time horizon: Web-building strategy in an
orb-weaving spider as a case study. J Theor Biol 241: 725733.

1. Robert JS (2004) Model systems in stem cell biology. BioEssays 26: 10051012.
2. Fuchs E, Tumbar T, Guasch G (2004) Socializing with their neighbors: stem
cells and their niche. Cell 116: 769778.
3. Anversa P, Leri A, Kajstura J (2006) Stem cells and heart disease. In: Lanza R,
Gearhart J, Hogan B, Melton D, Pedersen R, et al, eds. Essentials of stem cell
biology. Amsterdam: Elsevier. pp 425430.
4. Lanza R, Gearhart J, Hogan B, Melton D, Pedersen R, et al. (2006) Essentials of
stem cell biology. Amsterdam: Elsevier. 548 p.
5. Dobzhansky T (1964) Biology, molecular and oganismic. Integ Compar Biol 4:
443452.
6. Anderson DJ (2001) Stem cells and pattern formation in the nervous system: The
possible versus the actual. Neuron 30: 1935.
7. Raff M (2003) Adult stem cell plasticity: fact or artifact? Annu Rev Cell and
Devel Biol 19: 122.
8. Mangel M, Bonsall MB (2007) The evolutionary ecology of stem cells and their
niches-the time is now. Oikos 116: 17791781.
9. Viswanathan S, Davey R, Cheng D, Raghu RC, Lauffenburger DA, et al. (2005)
Clonal evolution of stem and differentiated cells can be predicted by integrating
cell-intrinsic and extrinsic parameters. Biotechnol Appl Bioc 42: 119131.
10. Mayr E (1982) The growth of biological thought: Diversity, evolution, and
inheritance. Cambridge: Harvard University Press. 1048 p.
11. Watt FM, Hogan BLM (2000) Out of Eden: Stem cells and their niches. Science
287: 14271430.
12. Vogel H, Niewisch H, Matioli G (1969) Stochastic development of stem cells.
J Theor Biol 22: 249270.
13. Neff T, Beard BC, Kiem H-P (2006) Survival of the fittest: in vivo selection and
stem cell gene therapy. Blood 107: 17511760.
14. Potten CS, Loeffler M (1990) Stem cells: attributes, cycles, spirals, pitfalls and
uncertainties lessons for and from the crypt. Development 110: 10011020.
15. Potten CS, Booth C (2002) Keratinocyte stem cells: A commentary. J Invest
Dermatol 119: 88.
16. Alon U (2007) An Introduction to Systems Biology. Design Principles of
Biological Circuits. Boca Raton: Chapman and Hall/CRC. 320 p.
17. May RM (2004) Uses and abuses of mathematics in biology. Science 303:
790793.
18. May RM, McLean A (2007) Theoretical ecology. Principles and applications.
Third Edition. Oxford: Oxford University Press. 272 p.
19. Fisher RA (1930) The genetical theory of natural selection. Oxford: Oxford
University Press. 360 p.
20. Frank SA (2007) Dynamics of cancer: Incidence, inheritance, and evolution.
Princeton: Princeton University Press. 400 p.
21. Grant PR, Grant BR (2007) How and why species multiply: The radiation of
Darwins finches. Princeton: Princeton University Press. 272 p.
22. Thomas ED, Lochte HL, Lu WC, Ferrebee JW (1957) Intravenous infusion of
bone marrow in patients receiving radiation and chemotherapy. New Engl J Med
257: 491496.
23. Appelbaum FR (2007) Hematopoietic-cell transplantation at 50. New Engl J Med
357: 14721475.
24. Till JE, McCulloch EA, Siminovitch L (1964) A stochastic model of stem cell
proliferation based on the growth of spleen colony-forming cells. Proc Natl Acad
Sci U S A 51: 2936.
25. Mangel M (2006) The theoretical biologists toolbox: Quantitative methods for
ecology and evolutionary biology. Cambridge: Cambridge University Press.
376 p.
26. Murray JD (2003) Mathematical Biology, Volume 2. Heidelberg: Springer.
736 p.
27. Shostak S (2006) (Re)defining stem cells. BioEssays 28: 301308.
28. Kirouac DC, Zandstra PW (2006) Understanding cellular networks to improve
hematopoietic stem cell expansion cultures. Curr Opin Biotech 17: 538547.
29. Cairns J (1975) Mutation selection and the natural history of cancer. Nature 255:
197200.
30. LeGrand EK (1997) An adaptationist view of apoptosis. Q Rev Biol 72:
135147.
31. Potten CS, Owen G, Booth D (2002) Intestinal stem cells protect their genome
by segregation of template DNA strands. J Cell Sci 115: 23812388.
32. Zandstra PW, Nagy A (2001) Stem cell bioengineering. Annu Rev of Biomed
Eng 3: 275305.

PLoS ONE | www.plosone.org

10

February 2008 | Volume 3 | Issue 2 | e1591