Simulation of Yeast Cooperation in 2D: Bulletin of Mathematical Biology March 2016
Simulation of Yeast Cooperation in 2D: Bulletin of Mathematical Biology March 2016
Simulation of Yeast Cooperation in 2D: Bulletin of Mathematical Biology March 2016
discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/298905486
CITATIONS READS
0 24
3 authors, including:
Yuanyuan Huang
RapidMiner
8 PUBLICATIONS 15 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Text Mining, Predictive Maintenance, Fraud Detection, Customer Prediction, Web Analytics View
project
All content following this page was uploaded by Yuanyuan Huang on 21 November 2017.
ISSN 0092-8240
1 23
Your article is protected by copyright and
all rights are held exclusively by Society
for Mathematical Biology. This e-offprint is
for personal use only and shall not be self-
archived in electronic repositories. If you
wish to self-archive your work, please use the
accepted author’s version for posting to your
own website or your institution’s repository.
You may further deposit the accepted author’s
version on a funder’s repository at a funder’s
request, provided it is not made publicly
available until 12 months after publication.
1 23
Author's personal copy
Bull Math Biol
DOI 10.1007/s11538-012-9721-5
O R I G I N A L A RT I C L E
Abstract Game theoretic models, along with replicator equations, have been applied
successfully to the study of evolution of populations of competing species, including
the growth of a population, the reaching of the population to an equilibrium state,
and the evolutionary stability of the state. In this paper, we analyze a game model
proposed by Gore et al. (Nature 456:253–256, 2009) in their recent study on the co-
development of two mixed yeast strains. We examine the mathematical properties
of this model with varying experimental parameters. We simulate the growths of the
yeast strains and compare them with the experimental results. We also compute and
analyze the equilibrium state of the system and prove that it is asymptotically and
evolutionarily stable.
1 Introduction
Game theory, originally developed in economics, has been applied to many other
scientific fields including ecology, psychology, sociology, and political sciences. In
1940s, von Neumann and Morgenstern (1944) investigated the zero-sum games and
proved the existence of the optimal strategies. In the 1950s, Nash (1950) considered
more general games and showed the existence of the equilibrium solution, called the
Y. Huang () · Z. Wu
Department of Mathematics, Program on Bioinformatics and Computational Biology, Iowa State
University, Ames, IA 50011, USA
e-mail: sunnyuan@iastate.edu
Z. Wu
e-mail: zhijun@iastate.edu
Author's personal copy
Y. Huang, Z. Wu
Nash equilibrium. Later, in the 1970s, Smith and Price (1973) introduced game theory
to biology and proposed the concept of evolutionarily stable state, which laid the
foundation for evolutionary game theory. The latter, along with replicator equations,
has been applied successfully to the study of evolution of populations of competing
species, including the growth of a population, the reaching of the population to an
equilibrium state, and the evolutionary stability of the state (Nowak 2006a).
The evolution of cooperation has been an area of intensive research in the past
several decades (West et al. 2006). General rules have been studied in terms of
kin selection, direct and indirect reciprocity, network selection, and group selection
(Nowak 2006b). Special strategies have been discovered such as the tit for tat strate-
gies in the repeated prisoner’s dilemma games (Axelrod and Hamilton 1981) and
the restrictions of dispersal, movement, and interaction for the co-existence of com-
peting species (Nowak and May 1992; Durrett and Levin 1997; Kerr et al. 2002;
Hauert and Doebeli 2004). The cooperative behaviors have been explained in game
theoretic models and verified in experiments with model bacteria and yeast sys-
tems (Rainey and Rainey 2003; Griffin et al. 2004; MacLean and Gudelj 2006;
Diggle et al. 2007; Santorelli et al. 2008; Gore et al. 2009).
In particular, Gore et al. (2009) investigated the evolution of cooperation using
a deliberately designed system of two competing yeast strains. What special about
this work is the following: First, the designed system depends on several experimen-
tal parameters. Therefore, by varying the parameters, the system’s behavior can be
changed from a mutual benefit game to a snowdrift game and from a snowdrift game
to a prisoner’s dilemma game. Then, cooperation can be investigated in different
game settings. Second, the authors showed that if a simple matrix game model was
used, there might be only two types of game, the mutual benefit game or prisoner’s
dilemma game. The snowdrift game is not implied in this model, while in the exper-
iments the system behaved most time like a snowdrift game. Therefore, the authors
derived a nonlinear payoff function for the game, i.e., a nonlinear game model, with
a more accurate description of the system’s behaviors.
While Gore et al. (2009) focused on experimental results, this paper is a further
theoretical stretch of their work. In this paper, we examine the mathematical proper-
ties of the proposed game model with varying experimental parameters. We simulate
the growths of the yeast strains and compare them with the experimental results. We
also compute the equilibrium states of the system and analyze their stabilities. We
show that the equilibrium state of the snowdrift game is evolutionarily stable, which
is not so obvious when the model is given in a nonlinear form.
In what follows in this section, we introduce some basic concepts in game theory.
In Sect. 2, we describe the yeast experiments in Gore et al. (2009). We describe
their model and show some of their results. In Sect. 3, we describe the methods
for computing the growths and equilibrium states of the yeast system. We present
our computational results and compare them with the experimental ones. In Sect. 4,
we analyze the model and prove the asymptotic and evolutionary stabilities of the
equilibrium states. We conclude the paper in Sect. 5 and make some additional re-
marks.
For introduction, let’s consider a simple two-player two-strategy game. Then, the
outcomes of the game can be measured by a pair of payoff functions that depend
Author's personal copy
Game Dynamic Model for Yeast Development
on the strategies of the players. Let π1 and π2 be the payoff functions for player 1
and 2, respectively. Let x = (x1 , x2 )T be the player 1’s choice of strategies, playing
strategy 1 with probability x1 and strategy 2 with probability x2 , and x1 + x2 = 1.
Let y = (y1 , y2 )T be the player 2’s choice of strategies, playing strategy 1 with prob-
ability y1 and strategy 2 with probability y2 , and y1 + y2 = 1. Then, the payoffs
for player 1 and 2 will be π1 (x, y) and π2 (y, x), respectively. If π1 is equal to π2 ,
the game is called symmetric, i.e., the payoff will be the same for a pair of compet-
ing strategies no matter which player chooses which strategy. For a biological game,
a player could be a certain species; a strategy could be a certain genotypic or pheno-
typic feature; and the payoff function could be the fitness function defined in terms
of resource gain or reproduction success.
In any case, each player competes to maximize his payoff, but can only achieve it
when the other player can no longer reduce his gain with a counter strategy. If both
players reach such a point, the game is said to reach equilibrium or more specifically
Nash equilibrium, and the strategies of the players are called the optimal strategies.
Consider a symmetric game. Define Sx = {x ≥ 0 : i xi = 1}. Then, a game reaches
Nash equilibrium if there are optimal strategies x ∗ in Sx for player 1 and y ∗ in Sy for
player 2 such that
π x ∗ , y ∗ ≥ π x, y ∗ , ∀x in Sx and
(1)
π y ∗ , x ∗ ≥ π y, x ∗ , ∀y in Sy .
Usually, the payoffs for all the pairs of competing strategies can be given in a matrix
U with Uij being the payoff for strategy i against strategy j . The payoff function
π can then be defined as the average payoff of the player whenhe chooses strategy
x against the other player’s strategy y, i.e., π(x, y) = x T Uy = i (xi j Uij yj ). In
this case, the game is called a matrix game. Otherwise, the payoff function can be in
a general nonlinear form.
A game is called a prisoner’s dilemma game (PD) if each player chooses to be
a cooperator or defector to counter the other: A cooperator gains substantially (with
a payoff, say 4) if the opponent also cooperates, but loses the most (with a payoff,
say −1) if the opponent defects. A defector gains the most (with a payoff, say 5)
if the opponent cooperates, and does not lose that much as the cooperator (with a
payoff, say 1) if the opponent defects. So a safe guard is to always play a defector,
for nobody wants to take a risk to cooperate but become a loser when the opponent
defects.
A game is called a mutual benefit game (MB) if each player chooses to be a coop-
erator or defector, but the payoff outcomes are very different from PD: A cooperator
gains (with a payoff, say 5) no matter what the opponent chooses to be. A defector
always loses (with a payoff, say 3) no matter what the opponent is. As a result, both
players will choose to play cooperators.
A snowdrift game is played when two players act as two drivers whose cars are
blocked by the snow on the road. Each of them needs to decide if he should go
out to remove the snow or stay in the car to wait until the other driver clears the
road. In this game, a defector (who stays in the car) takes advantage (with a payoff,
say 5) if the opponent cooperates (removing the snow), but the worst scenario is
Author's personal copy
Y. Huang, Z. Wu
when the opponent also defects and then both get stuck in the car (with a payoff,
say 0). A cooperator benefits (with a payoff, say 3) if the opponent also cooperates,
but would benefit more if he were a defector. Besides, he may get worse (with a
payoff, say 1) if the opponent defects. In such cases, the cooperators are attempted to
be defectors, and the defectors have to cooperate sometimes if there is no cooperator.
As a result, a good scenario would be for the player to sometimes cooperate and
sometimes defect, to obtain an optimal amount of benefit.
Gore et al. (2009) designed an experiment with two mixed yeast strains in sucrose
culture. The wild type cells, considered as cooperators, hydrolyze sucrose into glu-
cose and fructose with protein invertase. They consume only a small portion (about
1 %, called the capture rate) of the products. The rest of the glucose and fructose dif-
fuses into the culture. The mutant cells, considered as defectors, do not have the gene
SUC2 that produces protein invertase and therefore, cannot make glucose or fructose.
The wild type cells are also made histidine auxotroph (with defected HIS3 genes) so
they have an additional fitness cost due to histidine anabolic deficiency. The cost is
high when the histidine concentration in culture is low (see Fig. 1).
In general, the growth of the mutant strain depends on the wild type strain, but
the mutant strain will outgrow the wild type if the culture is maintained with high
concentration glucose and low concentration histidine, because the wild type cells
incur a fitness cost due to invertase production as well as histidine anabolic deficiency.
In this case, the interaction between the wild type cell and the mutant cell is more like
a prisoner’s dilemma game (PD), and the mutant cells, the defectors, eventually take
over the whole population, assuming the high glucose concentration of the culture
is maintained. On the other hand, if the glucose concentration in culture is low and
Fig. 1 A mixed culture of two yeast strains. One is the wild type and the other mutated. The wild type
inverts sucrose into glucose and fructose (producing food). The mutated only takes glucose and fructose
(without producing them). The wild types are called cooperators, while the mutated cheaters. The wild
type strain has a gene SUC2 for producing sucrose invertase, while the mutated strain lacks SUC2 and
therefore, cannot invert sucrose. The wild type has a defective HIS3 gene and can only take histidine from
outside to grow and thereby incurs an additional fitness cost. (From Gore et al. 2009)
Author's personal copy
Game Dynamic Model for Yeast Development
Fig. 2 The co-existence of two yeast strains. (a) If started with a high population, the cooperator fraction
decreases. If started with a low portion, the cooperator fraction increases. Somewhere in between, there
must be an equilibrium state of nonzero cooperator fraction. (b) If the cost is high (in low histidine concen-
tration), the equilibrium fraction tends to be low (red). If the cost is low (in high histidine concentration),
the equilibrium fraction tends to be high (blue). (From Gore et al. 2009)
the histidine concentration is high, the wild type cells will eventually win, because
they have the advantage of consuming their own glucose or fructose directly while
the “cost of living” due to histidine anabolic deficiency is not so high. This makes the
interaction between the two types of cell a mutual benefit game (MB).
In Gore et al. (2009), the authors were able to change several experimental pa-
rameters including the glucose and histidine concentrations in culture. The authors
found that for a large range of parameter values, the interaction between the two
types of yeast cell maintained a snowdrift game (SG) instead, and the two strains
co-existed at equilibrium. More specifically, with a reasonable concentration of glu-
cose and histidine in culture, the authors observed the increase of the wild type
strain when it started from a small fraction and the decrease from a large frac-
tion, and it reached an equilibrium fraction no matter what fraction it started (see
Fig. 2).
Note that by the equilibrium of the yeast system in experiment we mean a rela-
tive stable state of the system after a sufficiently long time. It is an approximately
estimated equilibrium state, and in the yeast experiment, was roughly the state after
6 days.
To build a game model for a biological system, the population is usually assumed
well mixed so that each individual can interact with any others. An individual is
designated as one of the players, say player 1, while the population as another, say
player 2. The types of species in the population can be considered as the types of
strategies, and each individual can choose to be a certain type of species (i.e., a certain
strategy) or to be a mixed type of species (i.e., a mixed strategy). Such a game is called
a population game or an evolutionary game or a game against the field (Hofbauer and
Sigmund 1998; Webb 2007). For the system of two yeast strains, assume that all
yeast cells are well mixed. Then, the system can be considered to have two strategies:
f fraction of wild types (cooperators) and (1 − f ) fraction of mutants (defectors).
Each individual cell can be considered to play a mixed strategy, with probability f o
Author's personal copy
Y. Huang, Z. Wu
3 Computer Simulation
Suppose that the growth rates of cooperators and defectors are given as the formulas
in (5). Then, the average payoff function for a cell of strategy x = (f o , ho )T against
a population of strategy y = (f, h)T can be written as
This theorem is called the equal payoff theorem in evolutionary game theory.
A proof can be found in standard textbooks such as Webb (2007). The theorem
implies that if y ∗ is an optimal strategy, then π(y, y ∗ ) = π(y , y ∗ ), ∀y, y in Sy∗ .
For a two-strategy game, if y ∗ = (f ∗ , h∗ )T is an optimal strategy and f ∗ , h∗ > 0,
then Sy∗ = Sy . Let yC = (1, 0)T and yD = (0, 1)T . Then, yC , yD are in Sy∗ and
π(yC , y ∗ ) = π(yD , y ∗ ). It follows that for the payoff function π defined in (7),
α α
g + (1 − g) ε + f ∗ (1 − ε) − c = g + (1 − g)f ∗ (1 − ε) . (9)
This is equivalent to the condition when PC (f ∗ , h∗ ) = PD (f ∗ , h∗ ).
For computer simulation, the equation in (9) gives an important condition for the
computation of the equilibrium fraction f ∗ of the yeast game if the parameters g, ε,
and c are given. Indeed, in Gore et al. (2009), ε was estimated to be around 0.01.
Then, if glucose and histidine concentrations were set to certain values, the corre-
sponding equilibrium fraction could be observed after the system settled down for
Author's personal copy
Y. Huang, Z. Wu
Fig. 4 Cooperator fractions at equilibrium. The experimentally observed cooperator fractions at equilib-
rium with varying glucose and histidine concentrations are shown in a color-coded graph. From this graph,
we see that in low histidine and high glucose, the cooperator fractions are almost zero and the defectors
win, implying that the game is PD. On the other hand, in high histidine and low glucose, the cooperator
fractions are almost one and the cooperators win, implying that the game is MB. In between, the two
strains co-exist, and the game is SG. (From Gore et al. 2009)
several days (Rigorously speaking, the observed equilibrium fraction in finite time
is only an approximated equilibrium fraction). Figure 4 shows the experimentally
observed cooperator fractions f ∗ at equilibrium with varying glucose and histidine
concentrations in a color-coded graph. From this graph, we see that in low histidine
and high glucose, the cooperator fractions f ∗ are almost zero and the defectors win,
implying that the game is PD. On the other hand, in high histidine and low glucose,
the cooperator fractions f ∗ are almost one and the cooperators win, implying that the
game is MB. In between, the two strains co-exist, and the game is SG.
Conversely, using condition (9), if g and ε are given and f ∗ is known, the cost c,
dependent of the histidine concentration, may be determined. We call this process
an inverse game: determine the parameters that define the payoff function, given the
optimal strategies at equilibrium. It is called the inverse game, because it corresponds
to an inverse problem in mathematical terms (Aster et al. 2005). Such a problem can
be as important as a direct problem, for often in biological games, we can observe or
measure the equilibrium states at least approximately, but we do not know anything
about the payoff rules, which in fact are what we want to learn and discover. From
condition (9), the cost c can be determined given ε, g, and equilibrium fraction f ∗
using the following formula:
α α
c = g + (1 − g) ε + f ∗ (1 − ε) − g + (1 − g)f ∗ (1 − ε) . (10)
Figure 5 shows the values of c in a color-coded graph calculated using formula (10)
with the values of f ∗ shown in Fig. 4. Note again that these equilibrium fractions
are observed ones in experiment and are approximately estimated values. From this
figure, we see that the values of c are high for low histidine and high glucose (red
regions) and are low for high histidine and low glucose (blue regions). With high
cost, the system certainly tends to be like a PD game, while with low cost, it is more
Author's personal copy
Game Dynamic Model for Yeast Development
Fig. 5 Color-coded cost values. The values for cost c are calculated using formula (10). From this graph,
we see that the c values are high for low histidine and high glucose (red regions) and are low for high
histidine and low glucose (blue regions). With high cost, the system certainly tends to be like a PD game,
while with low cost, it is more of a MB game. In between, the system behaves as a SG game with (the wild
type) cooperators co-existing with (the mutant) defectors
of a MB game. In between, the system behaves as a SG game with (the wild type)
cooperators co-existing with (the mutant) defectors.
Once all the parameters in the payoff functions are determined, the equilibrium
fraction f ∗ and hence h∗ can be computed based on Theorem 3.1 or more specifically,
on condition (9) for the yeast problem. However, in order to learn how the yeast
strains have developed from some initial state to an equilibrium state, we have to
solve a system of replicator equations that describes the dynamic properties of the
population changes of the species.
A general principle for constructing a system of replicator equations for a given
population of species is that the growth rate for each species relative to its current
population is proportional to the payoff advantage of this species over the average
payoff of the whole population. Let yi be the population fraction of species i, i =
1, . . . , n, where n is the total number of species. Then, the replicator equations can
be written in the following form:
yi /yi = π(ei , y) − π(y, y), i = 1, . . . , n, yi = 1. (11)
i
Fig. 6 Dynamic changes of cooperator fractions. Shown in (b) are the simulated trajectories of cooperator
fractions with varying values of glucose concentration, where the value for ε = 0.01 and histidine concen-
tration is fixed to 0.05×. The simulations are all started with an initial cooperator fraction f0 = 10−1 .
Then, if glucose concentration is relatively low (0.001 %), the defector fraction grows and the cooperator
fraction decreases to an equilibrium fraction. When the glucose concentration is increased, the cooperator
fraction at equilibrium decreases (not as demanded yet costly for co-existence). When it is increased to
0.03 %, the cooperator fraction at equilibrium is almost zero (below 10−4 after 5 days). All these results
agree well with the experimental observations as displayed in (a) (from Gore et al. 2009)
where
α α
P (f ) = g + (1 − g) ε + f (1 − ε) − c − g + (1 − g)f (1 − ε) (16)
depends on the parameters ε, g, and c. By fixing the values for these parameters,
we can solve a corresponding equation (15) for any initial fraction f0 , to obtain a
trajectory for f changing from f0 in certain time period. In particular, we can also
find possible fixed points (or limiting points) for f from (15) when the growth rate
of f is zero, i.e., either f = 0, or f = 1, or f is such that P (f ) = 0. It follows from
the following theorem that a Nash-equilibrium point must be a fixed point, but a fixed
point may not necessarily be a Nash-equilibrium point.
Theorem 3.2 A strategy y ∗ is a fixed point of the system of replicator equations (11)
if it is a Nash-equilibrium strategy for the game defined by the payoff function π .
Fig. 7 Dynamic changes of cooperator fractions. Shown in (b) are the simulated trajectories of cooperator
fractions with varying values of histidine concentration, where the value for ε = 0.01 and glucose concen-
tration is fixed to 0.003 %. The simulations are all started with an initial cooperator fraction f0 = 10−1 .
Then, if histidine concentration is relatively high (>0.2×), the cooperators maintain their population to
equilibrium. When the histidine concentration is decreased, the cooperator fraction decreases from its
initial fraction to equilibrium, and in general, the lower the histidine concentration, the smaller the coop-
erator fraction at equilibrium (costly to maintain). When it is decreased to 0.005×, the cooperator fraction
at equilibrium is close to 10−5 in 5 days, while the experimentally observed one is close to 10−4 . Overall,
these results agree well with the experimental observations as displayed in (a) (from Gore et al. 2009)
concentration is fixed to 0.05×, where × means times the normal histidine concen-
tration, which is 20 µg/mL. The simulations are all started with an initial cooperator
fraction f0 = 10−1 . Then, if glucose concentration is relatively low (0.001 %), the
defector fraction grows and the cooperator fraction f decreases to an equilibrium
fraction. When the glucose concentration is increased, the cooperator fraction at equi-
librium decreases (not as demanded yet costly for co-existence). When it is increased
to 0.03 %, the cooperator fraction at equilibrium is almost zero (below 10−4 after
5 days). All these results agree well with the experimental observations as displayed
in Fig. 6(a).
Figure 7(b) shows the solution curves for the equation in (15) with varying values
of histidine concentration instead, while the value for ε = 0.01 and glucose concen-
tration is fixed to 0.003 %. The simulations are all started with an initial coopera-
tor fraction f0 = 10−1 . Then, if histidine concentration is relatively high (>0.2×),
the cooperators maintain their population to equilibrium. When the histidine concen-
tration is decreased, the cooperator fraction f decreases from its initial fraction to
equilibrium, and in general, the lower the histidine concentration, the smaller the co-
operator fraction at equilibrium (costly to maintain). When it is decreased to 0.005×,
the cooperator fraction at equilibrium is close to 10−5 in 5 days, while the exper-
imentally observed one is close to 10−4 . Overall, these results agree well with the
experimental observations as displayed in Fig. 7(a).
Note that in Figs. 6 and 7, the computed trajectories are shown only in 6 days, in
order to compare with the experimental ones. They also end slightly higher than the
experimental curves, because they are supposed to converge to the observed equilib-
rium fractions as the time goes to infinity.
Figure 8(b) shows the equilibrium fractions f ∗ obtained using condition (9) with
varying values for glucose and histidine concentrations, while the value for ε = 0.01.
Author's personal copy
Y. Huang, Z. Wu
Fig. 8 Equilibrium fractions. Shown in (b) are the computed cooperator fractions at equilibrium with
varying values for glucose and histidine concentrations, while the value for ε = 0.01. The curves for
histidine concentrations 1.0× and 0.2× match with experimentally observed equilibrium fractions well,
decreasing relatively slowly as the glucose concentration increases. When the histidine concentration fur-
ther decreases, the equilibrium fraction is supposed to approach to zero quickly as glucose concentration
increases. Indeed, for example, when the histidine concentration is decreased to below 0.02×, the equilib-
rium fractions become zero for all glucose concentrations higher than 0.005 %. Otherwise, a true fractional
cooperator equilibrium strain is found. All these results agree well with the experimentally observed equi-
librium fractions as displayed in (a) (from Gore et al. 2009)
The condition (9) holds when 0 < f ∗ < 1 and two yeast strains co-exist. Therefore,
the curves for histidine concentrations 1.0× and 0.2× match with experimentally
observed equilibrium fractions well, decreasing relatively slowly as the glucose con-
centration increases. When the histidine concentration further decreases, the equilib-
rium fraction is supposed to approach to zero quickly as glucose concentration in-
creases. Indeed, for example, when the histidine concentration is decreased to below
0.02×, the equilibrium fractions become zero for all glucose concentrations higher
than 0.005 %. Otherwise, a true fractional cooperator equilibrium strain is found. All
these results agree well with the experimentally observed equilibrium fractions as
displayed in Fig. 8(a).
We focus on the stability of the equilibrium states of the snowdrift game for the yeast
problem. To be self-contained, we first introduce some stability terms and theorems
without proofs and refer the readers to Hofbauer and Sigmund (1998) and Webb
(2007) for more details.
Definition 4.2 A fixed point y ∗ for a system of replicator equations (11) is said to be
asymptotically stable if for any ε > 0, there is δ > 0 such that y(t) − y ∗ < ε for all
t > 0 whenever y(0) − y ∗ < δ and if y(t) → y ∗ as t → ∞.
Author's personal copy
Game Dynamic Model for Yeast Development
Note that by this definition, asymptotically stable for a fixed point y ∗ simply
means that a small deviation or perturbation y(t) from y ∗ will remain arbitrarily
close to y ∗ for all t and converge to y ∗ as t goes to infinity. However, it is hard to ver-
ify the asymptotic stability of a fixed point directly from its definition. The following
theorem, called the Lyapunov Theorem, is often used in stability analysis.
Theorem 4.1 A fixed point y ∗ for a system of replicator equations (11) is said to
be asymptotically stable if there is a function V (y) ≥ 0 with V (y) = 0 if and only if
y = y ∗ such that Vt (y) ≤ 0 with Vt (y) = 0 if and only if y = y ∗ .
Note that evolutionary stability is a stronger stability term than asymptotic sta-
bility. It was first proposed by Smith and Price (1973) for measuring the stability
of equilibrium states of a biological system and later justified as a general stability
measure. However, the definition does not provide a simple way for verifying evolu-
tionary stability. The following theorem is one among many others that can be used
more directly for formal proofs.
Theorem 4.3 An asymptotically stable fixed point for a system of replicator equa-
tions is a Nash-equilibrium point for the corresponding population game.
Theorem 4.5 Let F be the set of fixed points, N the set of Nash-equilibrium points,
A the set of asymptotically stable fixed points, and E the set of evolutionarily stable
Author's personal copy
Y. Huang, Z. Wu
Fig. 9 Stability of equilibrium state. There are three fixed points: f = 0, f = 1, f = f ∗ such that
P (f ) = 0 for (15). Assume that 0 < f ∗ < 1. Then, f is positive and f increases to f ∗ for 0 < f < f ∗
and f is negative and f decreases to f ∗ for f ∗ < f < 1. The point f ∗ must be a stable point
Nash-equilibrium points for a given population game and its corresponding system
of replicator equations. Then, F ⊃ N ⊃ A ⊃ E.
We now consider the game model for the yeast problem. Consider the replicator
equation (15) for the model. By the definition of fixed points, this equation has three
fixed points: f = 0, f = 1, f = f ∗ such that P (f ) = 0. It is easy to verify that
P (f ) < 0 and therefore, P (f ) is a monotonically decreasing function. Assume that
0 < f ∗ < 1. Then, f is positive and f increases to f ∗ for 0 < f < f ∗ and f is
negative and f decreases to f ∗ for f ∗ < f < 1, as can be depicted in Fig. 9. The
point f ∗ must be a stable point. We state this fact more formally in the following
theorem.
Theorem 4.6 The fractional fixed point f ∗ for the replicator equation in (15) of the
yeast game is asymptotically stable.
Theorem 4.7 The optimal mixed strategy y ∗ of the yeast game defined by the non-
linear payoff function in (7) is an evolutionarily stable strategy.
5 Concluding Remarks
against any mutant invasion, but the proof for such a property often requires some
mathematical techniques. On the other hand, the proof is only based on the theoret-
ical model, and the true stability of the system is yet to be verified by experiment,
which may not be easy to do as well.
Finally, the paper is written in a more expository form, with both biological and
mathematical accounts, while certain level of mathematical rigorousness is main-
tained. The paper invites interest from biological and mathematical communities in
evolutionary game theory and its applications in biological modeling. Further ex-
tension of this work can be interesting as well such as applying the studied game
model to the yeast strains in a 2D plane. The outcomes can be different from what
we have so far observed either experimentally or computationally, because of the
effect of the scale of the local movement and interaction (Nowak and May 1992;
Durrett and Levin 1997; Kerr et al. 2002; Hauert and Doebeli 2004).
Acknowledgements We would like to thank Prof. Jeff Gore for kindly giving us the Matlab code for
generating some of the graphical results in their paper and for providing helpful suggestions and advices
for how to proceed in our computation. This work is partially supported by the NIH/ /NIGMS grants
R01GM072014 and R01GM081680 and by the NSF/ /DMS grant DMS0914354.
References
Aster, R., Borchers, B., & Thurber, C. (2005). Parameter estimation and inverse problems. Amsterdam/San
Diego: Elsevier/Academic Press.
Axelrod, R., & Hamilton, W. D. (1981). The evolution of cooperation. Science, 211, 1390–1396.
Diggle, S. P., Griffin, A. S., Campbell, G. S., & West, S. A. (2007). Cooperation and conflict in quorum-
sensing bacterial populations. Nature, 450, 411–414.
Durrett, R., & Levin, S. (1997). Allelopathy in spatially distributed populations. J. Theor. Biol., 185, 165–
171.
Gore, J., Youk, H., & van Oudenaarden, A. (2009). Snowdrift game dynamics and facultative cheating in
yeast. Nature, 456, 253–256.
Griffin, A. S., West, S. A., & Buckling, A. (2004). Cooperation and competition in pathogenic bacteria.
Nature, 430, 1024–1027.
Hauert, C., & Doebeli, M. (2004). Spatial structure often inhibits the evolution of cooperation in the snow-
drift game. Nature, 428, 643–646.
Hofbauer, J., & Sigmund, K. (1998). Evolutionary games and population dynamics. Cambridge: Cam-
bridge University Press.
Kerr, B., Riley, M. A., Feldman, M. W., & Bohannan, B. J. M. (2002). Local dispersal promotes biodiver-
sity in a real-life game of rock–paper–scissors. Nature, 418, 171–174.
MacLean, R. C., & Gudelj, I. (2006). Resource competition and social conflict in experimental populations
of yeast. Nature, 441, 498–501.
Nash, J. (1950). Equilibrium points in n-person games. Proc. Natl. Acad. Sci. USA, 36, 48–49.
Nowak, M. A. (2006a). Evolutionary dynamics: exploring the equations of life. Cambridge: Belknap Press.
Nowak, M. A. (2006b). Five rules for the evolution of cooperation. Science, 314, 1560–1563.
Nowak, M. A., & May, R. M. (1992). Evolutionary games and spatial chaos. Nature, 359, 826–829.
Rainey, P. B., & Rainey, K. (2003). Evolution of cooperation and conflict in experimental bacterial popu-
lations. Nature, 425, 72–74.
Santorelli, L. A., et al. (2008). Facultative cheater mutants reveal the genetic complexity of cooperation in
social amoebae. Nature, 451, 1107–1110.
Smith, J. M., & Price, G. R. (1973). The logic of animal conflict. Nature, 246, 15–18.
Von Neumann, J., & Morgenstern, O. (1944). Theory of games and economic behavior. Princeton: Prince-
ton University Press.
Webb, J. N. (2007). Springer undergraduate mathematics series. Game theory: decisions, interaction and
evolution. Berlin: Springer.
West, S. A., Griffin, A. S., Gardner, A., & Diggle, S. P. (2006). Social evolution theory for microorganisms.
Nat. Rev., Microbiol., 4, 597–607.