Understanding microbial cooperation

James A. Damore, Jeff Gore

Department of Physics, Massachusetts Institute of Technology, 77 Massachusetts Ave., Cambridge, MA, United States
a r t i c l e i n f o
Keywords:
Evolutionary game theory
Kin selection
Hamiltons rule
Multilevel selection
Prices equation
a b s t r a c t
The eld of microbial cooperation has grown enormously over the last decade, leading to improved
experimental techniques and a growing awareness of collective behavior in microbes. Unfortunately,
many of our theoretical tools and concepts for understanding cooperation fail to take into account the
peculiarities of the microbial world, namely strong selection strengths, unique population structure,
and non-linear dynamics. Worse yet, common verbal arguments are often far removed from the math
involved, leading to confusion and mistakes. Here, we review the general mathematical forms of Prices
equation, Hamiltons rule, and multilevel selection as they are applied to microbes and provide some
intuition on these otherwise abstract formulas. However, these sometimes overly general equations can
lack specicity and predictive power, ultimately forcing us to advocate for more direct modeling
techniques.
& 2011 Elsevier Ltd. All rights reserved.
1. Introduction
Cooperation presents a fundamental challenge to customary
evolutionary thinking. If only the ttest organisms survive, why
would an individual ever pay a tness cost for another organism
to benet? Traditionally, kin selection and group selection have
been the most prominent explanations for the maintenance of
cooperation in nature. Kin selection refers to cooperative beha-
viors being favored when they are preferentially directed towards
relatives (Hamilton, 1964). In turn, group selection, also known as
multilevel selection, suggests that altruistic traits can be favored
because of their benecial effect on a group, despite the indivi-
dual cost of such behaviors (Wilson, 1975).
While the connection between kin and multilevel selection
was initially unclear, recent theoretical work has elucidated many
of the similarities and differences between the two concepts
(Nowak, 2006b; Page and Nowak, 2002; Fletcher and Zwick,
2006; Wenseleers et al., 2009). In particular, the underlying
theme behind all mechanisms for the evolution of altruism is
the assortment of similar individuals (Fletcher and Doebeli, 2009).
When similar individuals are assorted, cooperators are more
likely than average to interact with other cooperators and non-
cooperating defectors are, in turn, more likely to interact with
defectors. This assortment is ascribed to relatedness in kin
selection and between-group variance in multilevel selection
models. Both methods are equivalent when applied correctly and
under certain assumptions. Unfortunately, when these assumptions
do not hold, both methods resort to abstract generalities, making
application difcult and prone to error.
Microbes present a unique opportunity for scientists interested in
the evolution of cooperation because of their well-characterized and
simple genetics, fast generation times, and easily manipulated and
measured interactions. While these advantages are often well appre-
ciated, other differences between organisms of the microscopic and
macroscopic world are sometimes forgotten when transferring ideas
and methods from the study of animals to that of microbes.
Important differences include strong selection strengths, fast evolu-
tion times, high levels of diversity, and non-linear dynamics, all of
which invalidate many less general techniques derived using specic
assumptions, a fact too often ignored or simply unknown by non-
theorists.
In this review, we examine the standard techniques used to
understand cooperation, as they are applied to microbes. This
allows us to make several simplications, particularly in genetics,
but it also means that we will not cover any technique inap-
propriate to microbes. With that said, many of our points are
pertinent to any scientist in the eld of cooperation or microbial
biology, but some may only be valid when considering microbial
cooperation.
2. Four basic classes of cooperation
To begin, let us dene cooperation as any act that increases
the tness of others. Now, if we have two strains of a microbe,
one an obligate cooperator and the other a defector that never
cooperates, there are still four fundamentally different classes of
interactions that t within this denition of cooperation. For
Contents lists available at ScienceDirect
journal homepage: www.elsevier.com/locate/yjtbi
Journal of Theoretical Biology
0022-5193/$ - see front matter & 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jtbi.2011.03.008

Corresponding author.
E-mail address: [email protected] (J. Gore).
Please cite this article as: Damore, J.A., Gore, J., Understanding microbial cooperation. J. Theor. Biol. (2011), doi:10.1016/
j.jtbi.2011.03.008
Journal of Theoretical Biology ] (]]]]) ]]]]]]
example, imagine mixing the two strains at different relative
fractions in a test tube, either: one strain performs better than the
other at every frequency, or one strain grows faster than the other
at some fractions and slower at other fractions (Fig. 1). If one
strain always does better either the cooperator or defector
then we say that it dominates and natural selection will always
favor the increase in frequency of the dominating strain (Fig. 1A
and D). Thus, if we grow the cells in a turbidostat where the
conditions and total population size are kept constant by con-
tinuous dilution, then the dominated strain will eventually go
extinctbarring any stochastic effects which we will ignore here.
If, however, one strain does not always grow faster than the
other, predicting the nal result of such a experiment will be
slightly more complicated. For instance, if cooperators grow
better at low frequencies and worse at high frequencies, then
the two strains will coexist at some intermediate frequency,
independent of the starting frequency of the experiment
(Fig. 1B). On the other hand, if cooperators grow better at high
frequencies and worse at low frequencies, then cooperators will
x in the population if they start at a high enough frequency, but
will go extinct if they start at a lower frequency (Fig. 1C).
For most cooperative interactions there are thus only four
possible outcomes: either (A) defectors will always x in the
population, (B) the two strains will coexist at some intermediate
frequency, (C) one strain will x depending the initial frequency,
or (D) cooperators will always x (although there may be
complications, e.g. see MacLean et al., 2010). Each situation yields
drastically different dynamics and should be treated as a distinct
circumstance. When cooperators dominate defectors, natural
selection acts to increase the mean tness of the population, the
socially optimal outcome. On the other hand, if defectors dom-
inate cooperators, the mean tness decreases and we are often
interested in how this helpful cooperative behavior can be
maintained despite this dominance, a question kin and group
selection were developed to answer. Before discussing these
theories, we will rst examine in more detail the outcome of
the two other forms of cooperation, both of which are limited by
defectors, but not dominated by them.
While modelers often assume that any costly behavior dis-
played by bacteria that helps other cells is dominated by a
defector cell that does not display that behavior, this is actually
rarely the case. In fact, coexistence between cooperators and
defectors has been found both in nature and the lab for many
cooperative systems, including protease and siderophore
production, b-lactam antibiotic degradation, defective viruses,
fruiting body formation, and sucrose metabolism in yeast
(Diggle et al., 2007; Ross-Gillespie et al., 2007; Dugatkin et al.,
2005; Turner and Chao, 2003; smith et al., 2010; Gore et al.,
2009). These traits are said to have negative frequency-depen-
dence because the relative tness of a cooperator decreases as
their frequency increases. In particular, the stable mixed equili-
brium is where the relative tness equals one. A common theme
among these negative frequency-dependent traits is that a coop-
erator somehow receives a more than average share of the public
good that it personally produces. A prime example of this is the
cooperative sucrose metabolism system in budding yeast,
Saccharomyces cerevisiae (Greig and Travisano, 2004; Gore et al.,
2009; MacLean et al., 2010).
S. cerevisiae cannot import the sugar sucrose directly into the
cell very efciently. Instead, yeast is forced to break the dis-
accharide sucrose into glucose and fructose outside of the cell
with the extracellular enzyme invertase and then import the
monosaccharides glucose and fructose. Unfortunately for yeast,
because this happens in the periplasmic space, approximately
99% of the created glucose (and fructose) diffuses away to its
neighbors and the cell can only capture about 1% of the products
of sucrose hydrolysis (the capture efciency, e, is thus 0.01; Gore
et al., 2009). This would suggest that a cheater strain that does not
pay the cost of producing invertase could potentially take advan-
tage of the invertase producing cooperators if the cost is higher
than the benet of the extra 1% of glucose captured. In a simple
model, where growth is a linear function of the available glucose
(Fig. 2B), cooperators would either always do better or always do
worse than defectors depending on whether or not the normal-
ized cost, c, is greater than e. However, experimental results
suggest coexistence between cooperators and defectors (Fig. 2A);
therefore, one of the assumptions of the simple linear model is
incorrect.
As it turns out, the growth rate is actually a concave function
of the available glucose, which leads to coexistence of cooperators
and defectors because there are diminishing returns to increased
glucose (Fig. 2C). Thus, when the local glucose concentration is
low, the increased glucose due to invertase production outweighs
the cost of cooperation, but as there is more glucose in the
environment from other cooperators, the additional growth rate
per glucose molecule reduces below the cost of invertase produc-
tion. This qualitative model has proven to be robust to experi-
mental manipulation of the costs and benets of cooperation and
Fig. 1. (Color online) There are four basic, distinct outcomes of a cooperative interaction. (A) Defectors (red) may always have a higher tness than cooperators (blue),
which would lead to the extinction of cooperators. (B) Alternatively, cooperators may have a higher tness than defectors when they make up a small fraction of the
population and a lower tness at high fractions, leading to coexistence of cooperators and defectors. (C) Also, cooperators may have a higher tness only when they make
up a large fraction of the population and a lower tness otherwise. This creates an unstable xed point, above which cooperators x in the population and below which
defectors x, characteristic of bistability. (D) Finally, cooperators may always have a higher tness than defectors (or a complicated combination of (B) and (C)). In game
theory, the popular names for these interactions (when linear) are (A) prisoners dilemma, (B) snowdrift, (C) stag hunt, and (D) mutually benecial or harmony game.
J.A. Damore, J. Gore / Journal of Theoretical Biology ] (]]]]) ]]]]]] 2
Please cite this article as: Damore, J.A., Gore, J., Understanding microbial cooperation. J. Theor. Biol. (2011), doi:10.1016/
j.jtbi.2011.03.008
provides insight into how non-linear dynamics can lead to
coexistence (see Gore et al., 2009, for details).
In contrast to deterministic coexistence, some systems may
instead show bistability, which is characterized by mutual non-
invasability: cooperators cannot invade a pure defector culture,
but also defectors cannot invade a pure cooperator culture.
Bistability is predicted whenever cooperators preferentially inter-
act with themselves. Perhaps the most well-known example of
this is the hypothetical green beard gene (Hamilton, 1964;
Dawkins, 1976). Imagine a gene that simultaneously codes for a
green beard or any other arbitrary tag and a cooperative
behavior directed towards individuals with green beards. Such a
green beard gene could have an advantage and spread in the
population. However, if the cost of this gene is too high, then it
may still be disfavored at low frequencies because it does not
receive enough benet. Alternatively, two different versions of a
green beard gene may compete and could also show bistability.
Despite the theoretical appeal, green beard genes are rare in
nature, mostly because it is difcult for a single gene to code for a
beard, a recognition mechanism, and a cooperative action. With
that said, the o1 occulation gene in yeast has recently been
identied as a green beard gene by causing bearers of the gene to
stick together, protecting the inside of the clump from harm, like
the build up of ethanol in the environment (Smukalla et al., 2008).
Also, the csaA self adhesion gene in the slime mold Dictyostelium
discoideum works by a similar mechanism to exclude non-bearers
from fruiting bodies (Queller et al., 2003). Although it is out of the
scope of this paper, mutually antagonistic relationships can also
show bistability (Majeed et al., 2011).
If, however, cooperators always grow slower than the defec-
tors that they are mixed with, then an additional mechanism is
required to maintain cooperation. This situation corresponds to
cooperators paying some cost, not necessarily a xed amount, for
others to benet in some way. While in the well-mixed case,
cooperators are destined for extinction, the dynamics become
more complicated when structure is added to the population.
Now, if cooperators on average gain a high enough proportion of
the benets from other cooperators by interacting with them
more often than defectors do, then the increased benets may
outweigh the costs of cooperation and they can spread in the
population. Thus, the simplest and most fundamental explanation
for the maintenance of costly cooperation is that cooperators
must interact more often with cooperators than defectors do. This
assortment of similar individuals (i.e. cooperators with coopera-
tors and defectors with defectors) is the ultimate outcome of
every mechanism for the evolution of otherwise dominated
cooperation, including kin discrimination, group selection, direct
and indirect reciprocity, spatial structure and many others
(Axelrod and Hamilton, 1981; Nowak and Sigmund, 2005;
Ohtsuki et al., 2006; Nowak, 2006b; Fletcher and Zwick, 2006,
2007; Taylor and Nowak, 2007).
In the next section we introduce Prices equation, a powerful
and visually appealing tool that connects many kin and multilevel
selection models, particularly when usual simplications are
impossible. After introducing Prices equation, we provide some
intuition and discuss some of its limitations; any reader already
very comfortable with Prices equation can skip to Section 4 on
Hamiltons rule.
3. Prices equation
The Price equation provides a general and exact mathematical
description of evolution, with applications in kin and multilevel
selection models (Price, 1970; Page and Nowak, 2002). First
assume each individual, i, has a genotype, G
i
, that can be
quantitatively described. For example, G
i
is often arbitrarily set
to 1 if the (haploid) individual has the cooperative gene and 0 if it
does not, a convention we will repeat in this paper. Now, if we
want to know how the genotype frequency changes in the
population over time, we rst need to dene an individuals
tness, W
i
, to be the number of i individuals at the next time point
divided by the number present now. For example, the two time
points could be in generation times, in which case W
i
would be a
measure of how many offspring an individual has. The change in
average genotype in the population, DG, is then given by the Price
equation:
WDG CovW,G W
i
DG
i
1
where CovW,G is the covariance between tness and genotype
in the population and W
i
DG
i
accounts for the average change in
genotype between parent and offspring (e.g. mutation). For
asexual microbes, mutation and other effects that might change
an offsprings genotype are often assumed to be small, in which
Fig. 2. Non-linear dynamics can lead to coexistence of cooperators and defectors. (A) Cooperators and defectors coexist in the cooperative sucrose metabolism system,
which can be seen by their mutual invasability, where cooperators can invade pure defector cultures (bottom) and defectors can invade pure cooperator cultures (top).
(B) In a simple model of the cooperative yeast sucrose metabolism system, if the normalized growth rate, g, is a linear function of the available glucose, then cooperators or
defectors always dominate depending on the parameters. Here, we assume that cooperators pay a cost, c, and capture a small fraction, e, of the glucose made personally,
with the rest, 1e, diffusing away to be used by other cells. The amount of available glucose to a cell is then approximately given by ep1e, where p is the proportion of
cooperators in the population. Thus cooperators always grow faster than defectors if coe and grow slower otherwise. (C) If, however, the growth rate is a non-linear
function of the available glucose, then there is a large region where coexistence between cooperators and defectors is measurable (between dark lines). The exponent, a,
was experimentally measured to be 0.1570.01 (adapted from Gore et al. (2009)).
J.A. Damore, J. Gore / Journal of Theoretical Biology ] (]]]]) ]]]]]] 3
Please cite this article as: Damore, J.A., Gore, J., Understanding microbial cooperation. J. Theor. Biol. (2011), doi:10.1016/
j.jtbi.2011.03.008
case the Price equation becomes
WDG CovW,G b
WG
VarG 2
where b
WG
is the slope of the least-squares linear regression of
tness on genotype and VarG is the variance of genotypes in the
populationthe last equality being justied by the denition of
covariance (Fig. 3). Increasing the variance or the average effect of
genotype on tness, b
WG
, increases the speed of evolution. In fact,
when tness is perfectly heritable, Prices equation reduces to
Fishers Fundamental Theorem of Natural Selection which states
DW
VarW
W
3
which is valid when G is replaced with W (b
WW
1) and the last
term in (1) is neglected. Prices equation is a simple and visually
appealing formula that is widely used in the kin and multilevel
selection literature; however, some of its limitations should be
noted. First, practitioners often confuse probability theory and
statistics when applying Prices equation which has consequences
for its interpretation in derivations and models. Because this point
is subtle and out of the scope of this review, we refer the
interested reader to van Veelen (2005, 2009, 2010). On another
technical note, the Price equation lacks dynamical sufciency by
only telling the current speed and direction of selection and not
the path of evolution (Traulsen, 2010). Finally, with generality
comes a loss of specicity, and Prices equation can mask many of
the important intricacies of the system. This is a point we will
elaborate on later because the same criticism holds for many kin
selection and multilevel selection models that use Prices equa-
tion. While powerful and general, the Price equation does not
usually ease empirical calculations and can only aid in under-
standing a social behavior when it is coupled with an explicit
model. We will now move on to a generalized form of Hamiltons
rule, a simple two variable model, that uses Prices equation in its
derivation.
4. Hamiltons rule
In 1964, Hamilton stated his famous rule: a cooperative act
will be favored by natural selection if the cost, c, of performing the
cooperative act is less than the benet, b, given to the other
individual times the relatedness, r, between the two individuals,
or equivalently if
rbc40: 4
While pleasingly simple, the assumptions that went into
deriving this rule almost never hold even in simple models, much
less microbial systems (Cavalli-Sforza and Feldman, 1978; Karlin
and Matessi, 1983; Nowak et al., 2010; smith et al., 2010).
Therefore, a more general inequality is required for which we
will use Prices equation to derive (see Appendix A for full
derivation or Queller, 1992). Similar to how we regress tness,
W, against genotype, G, in Prices equation, we will rst regress
tness against an individuals genotype, G
X
, and the genotype of
its interactants, G
Y
, which results in
W W
0
b
WGX
G
X
b
WGY
G
Y
e 5
where W
0
is an individuals base tness and the bs are the partial
regression coefcients. b
WGX
is the average effect an individuals,
genotype has on its own tness, ignoring the effect of the
environment. For example, if the only difference between a
cooperator, G
X
1, and a defector, G
X
0, is that a cooperator pays
a cost, c, for others to benet, then b
WGX
would be c. Similarly,
b
WG
Y
is the average tness effect the environment, G
Y
, has on an
individuals tness, ignoring the effect of the individuals geno-
type. Again, when G
Y
measures the number of cooperators in an
individuals environment, and each cooperator gives its interac-
tant a benet, b, then b
WGY
is b. The last term, e, is the residual,
which may be different for every individual and describes the
difference between an individuals actual tness and the tness
predicted by the regression. After substitution into (2) and
simplifying, we get that a cooperative act is favored when
b
GY GX
b
WGY
b
WGX
40 6
where b
WGY
and b
WGX
are the same as above and now b
GY GX
is a
measure of how an individuals social environment covaries with
the individuals genotype and is the linear regression denition of
relatedness, r (Fig. 4). Thus, in the simplest case where every
cooperator pays a xed cost c for other individuals to gain a xed
benet b all measured in number of offspring cooperators are
favored when rbc40 (Fig. 5A).
Before going on to more complicated cases, a more thorough
discussion of relatedness is warranted. Contrary to the popular
use of the word, relatedness describes a population of interact-
ing individuals, where r refers to how assorted similar individuals
are in the population. r is 1 if every individual only interacts with
genetically identical individuals (G
X
G
Y
) and r is 0 if interactions
are random (G
Y
does not change with G
X
). Thus, if we have ten
clonal colonies, all from different strains of bacteria, and we put
each one in a separate test tube to grow, this would be full
assortment and r would be 1. Any otherwise dominated coopera-
tive strains with b4c40 would end up growing better than the
defector strains and thus would increase in frequency in the ten
test tube population because 1 bc40. On the other hand, if we
take an equal number of cells from each colony and put a little bit
in each of the ten test tubes, then there would be no assortment
and r would be 0. Similarly, when two strains are competing in a
well-mixed environment like a single test tube, then relatedness
is again 0 because each individual is no more likely than average
Fig. 3. Understanding how the Price equation works can aid in understanding evolution. (A) In its simplest form, Prices equation states that the change in genotype in a
population is DG CovW,G b
WG
VarG, where W is relative tness here. (B) Doubling the slope of the linear regression, b
WG
, while keeping everything else constant,
doubles DG. (C) Also note that increasing the spread of genotypes by a factor of 2 reduces b
WG
by a factor of
1
2
, but it multiplies the variance by 2
2
4 so DG still doubles.
(D) Effects multiply and combining (B) and (C) leads to a quadrupling of DG. For the purpose of illustration, the data points all sit perfectly on the regression, but that is not
necessary for Prices equation to work.
J.A. Damore, J. Gore / Journal of Theoretical Biology ] (]]]]) ]]]]]] 4
Please cite this article as: Damore, J.A., Gore, J., Understanding microbial cooperation. J. Theor. Biol. (2011), doi:10.1016/
j.jtbi.2011.03.008
to interact with its own type. In both these cases where r0, any
cooperators dominated by defectors would decrease in frequency,
again consistent with Hamiltons rule because 0 bco0.
There is, however, a measure of how related two individuals in
a population are. Importantly, Hamiltons rule asks if a trait will
increase in frequency and not necessarily number, so even if we
are only interested in the relatedness between two individuals,
we still have to take into account the population. The inclusive
tness denition of relatedness describes how similar an actor of
an action is to its recipient, relative to the population, and is given
by
r
P
A-R
P
A-P
P
A-A
P
A-P
7
where P
A-R
is, formally, the probability that a gene drawn at
random from the actor at the focal loci (e.g. the gene coding for
the cooperative behavior) is identical in state to a gene drawn at
random from the recipient at the focal loci (West et al., 2006).
Similarly, P
A-P
compares the actor to a member of the population
drawn at random and P
A-A
compares the actor to itself (P
A-A
1
for haploids). In the case of a rare gene in a large population,
P
A-P
-0, and r-P
A-R
=P
A-A
which gives the classic result that r
is one half for siblings and one eighth for cousins (in the absence
of inbreeding). Eq. (7) also gives r1 for clonemates, and r0 for
two individuals drawn at random from the population. Note that
the population relatedness is the average relatedness of inter-
acting individuals, but for the purposes of this paper, we will
stick to the linear regression denition of relatedness in Eq. (6)
because it is easier to visualize (Fig. 4). For other denitions and
methods of conceptualizing relatedness and assortment, see the
supplemental text.
Returning to our linear regression formula for Hamiltons rule
(6), it is important to understand how this is applied to more
complicated (i.e. realistic) situations. The tness of a microbe will
very rarely be a linear function of its environment, if only because
linearity is just one of the innite number of possibilities. When
an individuals tness is a non-linear function of its genotype
and environment, then the linear regression denitions of the
benets and costs depend on the data points sampled in
the population (Fig. 5B and C, data points are dark dots). Now the
measured b and c are no longer inherent properties of the
interaction, but rather functions of the population structure. In
fact, b and c are completely dependent on the sampled data points
and change every time they are measured.
The measured b and c can also give nonsensical results
depending upon the population structure being sampled. For
example, Hamiltons rule can predict negative tness values for
some cooperator frequencies (Fig. 5B). Alternatively, the mea-
sured b could be less than c for some population structures, which
would incorrectly suggest that cooperation is in general impos-
sible to evolve (Fig. 5C). Also, if at some point b is measured to be
2 and c to be 1, then as long as the relatedness, r, is greater than
c=b
1
2
at that moment, cooperation will increase in frequency;
however, this does not mean that r 4
1
2
is the general condition for
cooperation to be favored because b and c change with the
population structure and thus also with r. Note that while a
linear regression for the costs and benets can always be done, it
often is uninformative. Rather than being a powerful predictive
tool, this method can become a cumbersome way to add up
tnesses.
Fig. 4. Relatedness as a regression coefcient. To apply the general form of
Hamiltons rule, we need to assign a number to each genotype, and if we are
working with two strains (e.g. a cooperator and defector) we can arbitrarily set the
cooperator genotype to 1 and the defector genotype to 0. The relatedness, r, is then
the linear regression coefcient connecting an individuals genotype with the
genotype of its interactants. Note that when everyone interacts with the same
number of individuals, the regression passes through the average of eachs
interactants (black dots). This allows us to redene relatedness as
r PCjCPCjD: the probability for a cooperator to interact with a cooperator
minus the probability for a defector to interact with a cooperator. Also note that
scaling the x axis would also scale the y axis by the same amount, leaving the slope
the same, so arbitrarily setting the genotypes to 0 and 1 does not lose generality.
Fig. 5. Hamiltons rule is a linear regression to generally non-linear data. (A) In the simplest case, where everything is linear, the slope of the tness function is b and the
difference between a defector and cooperators tness with the same frequency of cooperator interactants is c. For a cooperator to have a higher tness than a defector, it
must interact with r more cooperators than defectors do (see gure). Because of the geometry, this results in a critical relatedness, rc c=b, above which cooperators are
favored (i.e. when rbc40). Unfortunately, life, and particularly microbes, is rarely this simple. (B and C) When the tness is a non-linear function of the fraction of
cooperators (light lines), b and c become linear regression coefcients based on the sample points (gray dots). Unfortunately this masks many of the interesting (non-
linear) qualities of the system and provides no prediction power because b and c change with the population structure. Also note that this method can lead to negative
tness values (B) and situations where boc (C), which would incorrectly suggest that cooperation can never evolve. Note that b and c determine the necessary r for
cooperation to evolve, but b and c are in turn affected by the population structure and thus r, which makes disentangling tness effects from population structure often
impossible. It is important to note that the cooperator and defector tnesses are not independently regressed, rather only one regression is done with the slope (b), vertical
difference between defectors and cooperators at a xed cooperator fraction (c), and base tness (y-intercept) as the three parameters (adapted from smith et al. (2010)).
J.A. Damore, J. Gore / Journal of Theoretical Biology ] (]]]]) ]]]]]] 5
Please cite this article as: Damore, J.A., Gore, J., Understanding microbial cooperation. J. Theor. Biol. (2011), doi:10.1016/
j.jtbi.2011.03.008
Also, on the practical side, G
Y
is often hard to dene. In the
simplest case, where every individual interacts equally with the
same number of individuals (e.g. in well-mixed test tubes seeded
with the same number of cells), then G
Y
is just the average
genotype of the interactants. However, when different interac-
tions are stronger than others, like on an agar plate where
interactions weaken with distance, or when individuals interact
with different numbers of individuals, then G
Y
becomes more
complicated (Le Gac and Doebeli, 2010). While Hamiltons rule
can still be applied, the relatedness and benet terms, b
GY GX
and
b
WGY
, become less intuitive and potentially arbitrary. It should
also be noted that when multiple alleles are competing, G
X
, and
thus also b
GY GX
, can become arbitrary. In fact, DG loses much of its
meaning in this case. Therefore, in cases of multiple competing
alleles or nontrivial population structure, Hamiltons rule is often
replaced with explicit modeling (e.g. see Xavier and Foster, 2007).
More complicated forms of Hamiltons rule are also possible.
For example, one can take a linear regression of cooperator and
defector tnesses separately, resulting in two different slopes
(Queller, 1985, see supplemental text for details). Importantly,
this method can predict bistability and coexistence, a distinction
that is traditionally out of reach for Hamiltons rule. While this
can be a more accurate view of the interaction and should be
favored in certain cases, it still cannot capture the full picture and
has some of the same pitfalls of traditional Hamiltons rule
without the advantages that come from its simplicity.
Alternatively, a highly modied form of Hamiltons rule has
recently been suggested for the study of microbes (smith et al.,
2010). To apply this unique formula, one rst ts two (non-linear)
regressions to the cooperator and defector tnesses. A Taylor
series expansion is then taken of these two functions and the
coefcients of these Taylor series are used in vector forms of b and
a new synergy term d. r is now a vector containing the traditional
denition of relatedness, but also additional terms describing
higher order moments of assortment (Godfrey-Smith and Kerr,
2009). In this formulation, the population structure is completely
separated from the tness effectsthat is, b, c, and the synergy
term d are now constants and do not change with each iteration
of the calculation. We see smith et al.s reformulation as a major
step forward for the application of Hamiltons rule to microbes,
but it is not without its limitations. First, although it may be a
more accurate view of microbial interactions, higher order relat-
edness and tness terms can be much less intuitive than Hamil-
tons simple rule. Also, it cannot accommodate multiple strains
competing or dependence on density or time, at least as it is
currently written. Finally, the reformulation still is not a replace-
ment for a mechanistic model, which can identify biologically
relevant parameters and aid in predictions.
5. Weak selection
Before moving on to alternative methods of conceptualizing
and analyzing cooperation, let us discuss why the simple form of
Hamiltons rule without any regressions is usually inappropri-
ate for the study of microbes. Hamilton thought of evolution as a
very slow and gradual process where each mutant varied only
slightly from the wild type and this can be seen in many of the
assumptions used in deriving the simple form of Hamiltons rule,
namely the assumption of weak selection (Hamilton, 1964;
Wenseleers, 2006). In the kin selection literature, (d) weak
selection refers to the assumption that everyone in the population
plays approximately the same strategy and any mutant can only
play a slightly different strategy (wild type strategy 7d as d-0;
Wild and Traulsen, 2007). This is analogous to every microbe in a
population producing the same amount of a public good like
invertase and the only mutations allowed are those that produce
slightly more or slightly less invertase. Any mutants, therefore,
also only have a vanishingly small difference in tness, leading to
dynamics dominated by random drift.
In evolutionary game theory, a similar concept called w weak
selection is often used (Nowak et al., 2004). In contrast to d weak
selection, individuals in the population can play distinct, discrete
strategies, but the tness difference between any two strategies is
always very small (mixed strategies are also allowed depending
on the model). This would correspond to two yeast cells that
produce the maximum and minimum amount of invertase having
only an innitesimal difference in tness. While weak selection is
often useful for gaining insight by simplifying the dynamics in
analytical models, it is not the general case and is often invalid for
cooperative microbial behaviors because of large differences in
tness.
The intuition behind weak selection partly comes from the
idea that any one interaction between two animals often has only
a small effect on their tness; for example, whether or not an
animal shares a berry with its group on a particular day is unlikely
to have a large impact on anyones survival. However, many forms
of microbial interactions are literally do or die. If sucrose is the
only available sugar, then a microbe will starve if it cannot break
it down. The difference in generation times is also important
because while an animal may have many small interactions in its
much longer lifetime, a microbe is likely to have only a few
interactions, many of which will be important in determining if it
will divide in the next hour. Additionally, while the complexity of
animal behavior makes evolution by small changes more likely in
animals, mutations of large effect are common in microbes.
So the concept of weak selection is inappropriate for microbes,
why is this important? First of all, while the linear regression
form of Hamiltons rule still works, inclusive tness theory and
many of its results are provably correct only in the limit of weak
selection (Nowak et al., 2010). For example, the commonly cited
rule that natural selection acts to maximize inclusive tness has
only been proven using stringent assumptions, including weak
selection (Grafen, 2006; van Veelen, 2007). Also, assuming weak
selection gives a quantitatively different answer than more
accurate models (Queller, 1984; Wenseleers, 2006, see supple-
mental text for details). This is particularly relevant when
empiricists apply results from a weak selection model to actual
data. In addition, when empiricists cannot measure the coopera-
tive allele of interest, they often use neutral genetic markers as a
proxy to estimate relatedness. Unfortunately, measuring related-
ness this way is only valid when the markers and the allele of
interest are under weak selection (Queller and Goodnight, 1989;
Hardy, 2003). Because genes like invertase and other cooperative
alleles are often not under weak selection, using a genetic marker
approach to measure relatedness in microbes will likely give a
wrong result. In fact, despite their seemingly simpler genetics,
relatedness is harder to estimate for microbes than for animals
(West et al., 2006).
Also, even if the costs and benets of an interaction are linear
functions of ones genotype and environment, the measured b
and c will not be linear unless selection is weak (Ross-Gillespie
et al., 2007). This results from growth being an exponential
process, so the total growth difference between a cooperator
and defector grows exponentially with the costs and benets
of the interaction. This will be a linear function only when
b and c51, which is the limit of weak selection (see supplemental
text for details). Another way to arrive at the same conclusion is
by imagining that, in the limit of d weak selection, every
individuals phenotype would be almost the same, so all the gray
dots in Fig. 5B and C would be right next to each other. This close-
ness allows the linear regressions to be replaced by derivatives,
J.A. Damore, J. Gore / Journal of Theoretical Biology ] (]]]]) ]]]]]] 6
Please cite this article as: Damore, J.A., Gore, J., Understanding microbial cooperation. J. Theor. Biol. (2011), doi:10.1016/
j.jtbi.2011.03.008
turning the otherwise hard frequency-dependent situation into
an easy frequency-independent maximization problem (Frank,
1998; Wenseleers et al., 2009). While this may be alluring to the
theorist, this technique is often ill-suited for the empirical study
of microbial cooperation.
6. Multilevel selection
Multilevel selection, also known as group selection, has gone
through its ups and downs, from once being completely discre-
dited to now being acknowledged as an important organizing
principle that can aid in the evolution of cooperation (Traulsen
and Nowak, 2006; Nowak, 2006b). Multilevel selection occurs
whenever selection acts on multiple levels, such as when indivi-
duals are partitioned into groups and the groups themselves
compete either directly or indirectly in addition to the
individuals inside each group competing against themselves. In
this case, traits that are costly to the individual but benecial to
the group can be selected for if the between-group selection is
stronger than the within-group selection. However, without any
mathematical modeling, this statement is empty and is even
dangerous because it can easily be used to justify any altruistic
behavior. Thus, let us quickly go through the derivation of a
simple multilevel selection model to better understand the math
behind selection at two levels. We start by assuming a population
is made up of several segregated groups, each with a total tness,
W
g
, and genotype G
g
. Now we use the full version of Prices
equation:
WDG CovW
g
,G
g
W
g
DG
g
: 8
However, unlike before, W
g
DG
g
cannot be disregarded because
the average genotype of the group changes, while before we only
assumed that individual genotypes stayed the same. Fortunately,
W
g
DG
g
is similar to the left side of the equation and another
iteration of Prices equation is possible:
WDG CovW
g
,G
g
Cov
g
W
i
,G
i
W
i
DG
i
9
WDG CovW
g
,G
g
Cov
g
W
i
,G
i
10
where W
i
DG
i
is now disregarded because i is the individual level
here. If, however, we wanted to model more than two levels, we
could just keep iterating Prices equation. Now CovW
g
,G
g
, the
relationship between group tness and group genotype, repre-
sents the between-group selective force and Cov
g
W
i
,G
i
is the
average within-group selective disadvantage of being a coopera-
tor and will likely be negative. If the between-group selection
advantage to being a cooperator is greater than the within-group
disadvantage, then DG40 and cooperation spreads.
Note that because CovW
g
,G
g
b
Wg Gg
VarG
g
(by the deni-
tion of covariance), the between-group selection pressure is
proportional to the variance between groups. In fact, connecting
this with the idea of assortment, VarG
g
is maximized when each
group is either all cooperators or all defectors (G
g
1 or 0 for each
group), which would correspond to full assortment of genotypes
and an r of 1. Similarly, when every group has the same
composition of genotypes, VarG
g
0, and there is no assortment
of genotypes, r0.
These connections should come as no surprise because when
the population structure is simple and everything is linear,
assortment is the fundamental explanation for the evolution of
altruism (Fletcher and Doebeli, 2009). How this assortment
happens in nature, and not necessarily how one measures it, then
becomes the most important question (Nowak, 2006b; Taylor and
Nowak, 2007). One possible mechanism is active kin discrimina-
tion, where individuals show a preference towards relatives
when segregating into interaction groups. This has been shown
to be a major mechanism for the maintenance of cooperative
spore formation in the social amoeba Dictyostelium purpureum
(Mehdiabadi et al., 2006). Also, spatial structure such as the
simple partitioning into groups can favor cooperation because
whenever smaller subpopulations are initiated by a random
sample of M number of cells, the level of assortment in the total
population is 1=M, which can be signicant for small M (Fletcher
and Zwick, 2004; Ackermann et al., 2008; Chuang et al., 2009).
The assortment is even higher if there is additional spatial
structure in each subpopulation such that each cell line interacts
more often with itself than others in the subpopulation. Care
should be taken with this argument, however, because this may
also mean that each cell line competes more intensely with itself
for resources and reproduction, potentially canceling out any
benet of assortment (Queller, 1994). This is why simple popula-
tion structure or global competition is often assumed beforehand
and is part of why relatedness measurements alone are
inconclusive.
To contrast multilevel selection and kin selection, emergent
properties are often better understood using multilevel selection
math and arguments than by Hamiltons rule. By emergent properties
we mean any property of the system that cannot be easily described
as a sum of smaller effects; standard examples of emergent proper-
ties include multicellularity and collective behavior where groups
themselves often compete directly. Because emergent properties are
more than just a sum of their parts, they are often characterized by
non-linearities, which is perhaps the main reason why Hamiltons
rule is less predictive in these situations.
By applying the same model to every situation, Hamiltons rule
and multilevel selection theory can mask many of the important
qualities of a system. True understanding comes only from explicit
modeling and not from simply examining statistical quantities,
which the Price equation, and any rule derived from it, coaxes us
all to do. This point is best made by examining Anscombes quartet,
four datasets with identical summary statistics, but strikingly
different graphs (Fig. 6; Anscombe, 1973). Each of these datasets
should be modeled and understood in distinct ways; too much
information is lost when we resort to plugging numbers into
covariances without conceptualizing the underlying data-generating
process. Note how ill-suited a linear regression is for some of these
datasets; this same situation happens when Hamiltons rule is
applied to potentially non-linear microbial data.
7. Common misconceptions
Perhaps the biggest misconception in the study of the evolu-
tion of cooperation is that there is something special, or even
magical, about kin selection that allows it to explain phenomena
that other theories cannot. Inclusive tness, the basis of kin
selection, is the effect of an individuals action on everyones
tness weighted by the relatedness between the individuals; this
is simply an alternative accounting scheme and, therefore, cannot
produce novel predictions outside of the scope of standard
natural selection theory (Wenseleers et al., 2009; Nowak et al.,
2010). It is also under contention whether or not inclusive tness
is even dened if certain assumptions are not met, such as: weak
selection, additivity, and simple population structure (Grafen,
2006). Inclusive tness also cannot describe dynamics, which
immediately makes it less general than standard techniques like
evolutionary game theory (Traulsen, 2010). It should also be
mentioned that evolutionary game theory has become an extre-
mely general and powerful tool for understanding evolutionary
dynamics (Nowak, 2006a). In particular, game theory is not
restricted to the use of game matrices which are often
J.A. Damore, J. Gore / Journal of Theoretical Biology ] (]]]]) ]]]]]] 7
Please cite this article as: Damore, J.A., Gore, J., Understanding microbial cooperation. J. Theor. Biol. (2011), doi:10.1016/
j.jtbi.2011.03.008
inappropriate for interactions that are not pairwise or linear (Frey,
2010).
Simply measuring the personal tness of each individual is
sufcient to explain altruism, without directly taking relatedness
into account because a gene coding for an altruistic behavior will
only spread if its bearers have a higher tness on average. This
point is often misconstrued, especially when considering inclu-
sive tness; for example, Hamilton mentions in his original 1964
paper that a gene may receive positive selection even though
disadvantageous to its bearers if it causes them to confer
sufciently large advantages on relatives. Of course a gene
cannot spread if the end effect is disadvantageous to its bearers;
the bearers have to also on average receive a sufciently large
advantage from others to outweigh the costs. The problem with
this quote, and other common informal inclusive tness argu-
ments, is the creation of a false dichotomy between bearers and
relatives, when in reality the only relatives that matter are
those that are also bearers (Fletcher and Doebeli, 2009).
There is also the misconception that inclusive tness calcula-
tions are generally easier than other approaches. While there are
cases under certain assumptions where this has been found to be
true (Taylor et al., 2007), inclusive tness measurements need all
of the same information as normal techniques and are often more
indirect and easier to mess up. In fact, even experts are not
immune to errors when applying Hamiltons rule; for example,
K ummerli et al. (2010) use the simple form of Hamiltons rule to
predict changes in cooperator frequency in an experimental
bacterial system with non-linearities. Unfortunately, because
the data did not t a linear model, the more complicated linear
regression or Quellers synergy reformulation was necessary (see
supplementary text for details). In a different paper, two of the
same authors admit that a na ve application of Hamiltons rule
may lead to mistakes. For this reason, it is easier to use standard
population genetics, game theory, or other methodologies to
derive a condition for when the social trait of interest is favored
by selection but go on to say that Hamiltons rule should be used
as an aid for conceptualizing this result. While we agree that
using as many perspectives as possible is often the best way to
understand a situation, and that Hamiltons rule can be a useful
intuitive tool if used correctly, the additive nature implicitly
assumed by Hamiltons rule often makes it inappropriate for
conceptualizing non-linear microbial models. Also, because it
requires transformations of variables, the end result in the form
of Hamiltons rule is often visually unconnected from the original
mechanistic model, making it difcult to judge how changes in
the original model will affect the transformed inequality. Note
that Hamiltons rule has been used correctly in the study of
microbes (e.g. see Gilbert et al., 2007), but the chance for error is
high, especially if certain assumptions do not apply like a simple
population structure (note also that the cited paper assumed
weak selection when measuring relatedness, which is likely to
give an overestimate).
A game theoretical viewpoint can be crucial and an important
insight could be nearly impossible to see without it. For example,
in a recent theoretical paper, we used evolutionary game theory
to study, both analytically and via simulations, host-symbiont
interactions (Damore and Gore, 2011). We showed that in a
generic population structure, when a symbiont evolves much
faster than its host which is often the case the equilibrium
distribution of strategies played by the host and its symbionts is
the same as the sequential game equilibrium where the host
moves rst and the symbiont second. This fundamental insight
inspired by game theory applies to parasites and mutualists alike,
explaining both the Red Queen hypothesis and the apparent
enslavement of endosymbionts.
Additionally, confusion about the denition of relatedness is
common among empiricists. Because Hamiltons rule is concerned
about an increase in frequency of a gene and not necessarily an
overall increase in number, every denition of relatedness must take
into account the population. Therefore, relatedness is not the
percent of genome shared, genetic distance, or any extent of
similarity between two isolated individuals in a larger population.
Also, because horizontal gene transfer is commonplace between
microbes and selection is strong, phylogenetic distance or any other
indirect genetic measure is likely to be inaccurate. Many of these
false denitions live on partly because ambiguous heuristics like
1
2
for brothers,
1
8
for cousins, which require very specic assumptions,
are repeated in the primary literature. Also, most non-theoretical
papers simply dene relatedness as a measure of genetic similar-
ity and do not elaborate or instead leave the precise denition to
the supplemental information (West et al., 2006). Unfortunately,
scientists can easily misinterpret this measure of genetic similarity
to be anything that is empirically convenient such as genetic
distance or percent of genome shared. Largely because of this
confusion, we support the more widespread use of the term
assortment, which is harder to misinterpret (Ackermann et al.,
2008). For similar reasons of reader understanding, we also encou-
rage authors to make calculations more explicit, either in the main
or supplemental text, and to avoid repeating previous results with-
out giving the assumptions that went into deriving them.
There is also an overemphasis on genotypic relatedness, which is
not strictly required for the evolution of altruism. To illustrate our
point, we will repeat an example from Fletcher and Doebeli (2009),
which we found particularly illuminating. Consider a haploid
microbial species that can produce a public good coded by two
completely different mechanisms, each encoded at a different locus.
At the rst locus, allele A produces the public good, while allele a
does not. Similarly, at the second locus, allele B produces the public
Fig. 6. Anscombes famous quartet illustrates the need of visualizing ones data
before statistically analyzing it. All four datasets have identical summary statistics
(e.g., mean, variance, regression coefcients, and covariance), but vastly different
graphs, which are more important in interpreting the data. Researchers should
graph and interpret their data before blindly applying Prices equation or
Hamiltons rule without a model. Importantly, if x was the genotype and y the
tness, Prices equation would predict the same change in average genotype in all
four populations, but we would still know nothing about the underlying dynamics
if we fail to visualize the data and model the situation. (adapted from Anscombe
(1973)).
J.A. Damore, J. Gore / Journal of Theoretical Biology ] (]]]]) ]]]]]] 8
Please cite this article as: Damore, J.A., Gore, J., Understanding microbial cooperation. J. Theor. Biol. (2011), doi:10.1016/
j.jtbi.2011.03.008
good and allele b does not. Now if Ab individuals only interact with
aB individuals and ab individuals interact with other ab individuals,
cooperation will spread, but it is not because of a genotypic
similarity between individuals that interact. Cooperation evolves
because individuals with an altruistic genotype are positively
assorted with other individuals that have cooperative phenotypes
(see also Appendix A). This may seem to be somewhat of a semantic
issue, but this distinction is ultimately important for evaluating
different models and deciding the necessary requirements for the
evolution of cooperation. Mutualisms between species where the
concept of relatedness is clearly out of place also exemplify the
usefulness of the term assortment (Foster and Wenseleers, 2006).
Nevertheless, even assortment is unnecessary when there are non-
linear effects (Hauert et al., 2006).
8. Discussion
For researchers studying microbial interactions, the most impor-
tant steps towards understanding the system are experimental
measurements and modeling (Br annstr om and Dieckmann, 2005;
MacLean and Gudelj, 2006; Xavier and Foster, 2007). Without even a
rough model of the interaction, no predictions or explanations of
phenomena can be made. Unfortunately, the simplicity of Hamil-
tons rule often discourages researchers from doing any modeling
beyond the simple idea that cooperators pay a xed cost, c, for
others to gain a xed benet, b. Importantly, mechanistic models are
also often required to guide the researcher to measure the relevant
biological parameters. While simplicity can aid initial comprehen-
sion, it is exactly the details and non-linearities that give rise to
many of the interesting qualities in microbes like graded responses,
collective behavior, and coexistence between multiple strains and
species (West et al., 2006; Mehta and Gregor, 2010; Reichenbach
et al., 2007; Gore et al., 2009).
On a similar note, researchers are often too quick to conclude
that a behavior can be explained by kin or group selection
without specifying any of the details or how assortment might
happen in nature. Identifying a behavior as cooperative is just the
beginning; in fact, many techniques cannot even distinguish all
four types of interactions mentioned in Section and Fig. 1, which
we see as a fundamental rst step in understanding the interac-
tion. For example, observing that pure cooperator cultures grow
faster than pure defector cultures tells you nothing except how to
label the strains because the interaction could still be any of the
four basic types (or even a combination of these four types). This
is a major problem because these situations should be studied
and thought of in distinct ways. Unfortunately, social interactions
are often classied as one of four types: altruistic, mutually
benecial, selsh, or spiteful; these four categories simply corre-
spond to the sign of b and c and cannot allow for coexistence or
bisability because more than two numbers are required to
describe them.
Authors should also be more explicit about what they mean
when they say tness. While we have only used tness to
mean the total number of offspring here because that is what is
used in the Price equation and Hamiltons rule, tness has also
been used to mean growth rate which is often more intuitive to
model and conceptualize (Kerr et al., 2002, 2006; MacLean and
Gudelj, 2006; Gore et al., 2009). In particular, growth rate is the
relevant quantity when using differential equations, a powerful
modeling tool. We support both uses, as long as the authors are
especially clear because semantic differences are common
between game theory and kin selection, which has created
confusion in the past (see supplemental text for details). With
this said, no matter what denition of tness one uses, absolute
tness is often more informative than relative tness because
absolute tness is the quantity used in calculations. Also, even if
the relative tness prole is the same for two populations they
could still have drastically different potentials for cooperation and
similarly even if the relative tness proles are signicantly
different, they have the same potential for the evolution of
cooperation (see supplemental text for details).
Biology thrives on the study of model organisms. Model
systems in microbial cooperation include: slime mold and bacter-
ial fruiting bodies, biolm formation, quorum sensing systems,
sucrose metabolism, restrained fermentation, bacteriocin and
siderophore production, persistence, programmed cell death,
virulence, cooperative antibiotic resistance, swarming, and many
others (Strassmann et al., 2000; smith et al., 2010; Xavier and
Foster, 2007; Diggle et al., 2007; Gore et al., 2009; MacLean and
Gudelj, 2006; Kerr et al., 2002; K ummerli et al., 2010; Lewis,
2006; Kerr et al., 2006; Turner and Chao, 1999; Dugatkin et al.,
2005; Chuang et al., 2010; Lee et al., 2010; Velicer and Yuen-tsu,
2003). In fact, collective behavior is beginning to be seen as the
norm in microbes, rather than the exception (for reviews of social
behavior and cooperation in microbes see Crespi, 2001; Velicer,
2003; Travisano and Velicer, 2004; West et al., 2006, 2007; Brown
and Buckling, 2008; Frey, 2010; Hibbing et al., 2010; Mehta and
Gregor, 2010). However, the necessary rst steps for studying a
cooperative behavior, explicit modeling and measurement of
biologically relevant parameters, have yet to be done for many
of these systems, a fruitful avenue for future research. Also, each
discovery of a new cooperative system advances our knowledge
of how cooperation evolves and is maintained in nature.
Because science is fruitless if it lacks questions to answer, here
we summarize some of the many important questions in the eld
of microbial cooperation. First, what produces the necessary
assortment and non-linearities for cooperation to evolve in
natural and experimental microbial systems? Similarly, what
effects and characteristics are lost in the lab? Many features that
may favor cooperation, like clumping, are selected against by
standard lab techniques and are, therefore, missing in some
domesticated strains (Aguilar et al., 2007). What is the speed
and strength of selection in natural systems? How do green beard
genes evolve? How relevant is horizontal gene transfer to the
maintenance and spread of cooperation (Nogueira et al., 2009)?
How are interspecic mutualisms formed and maintained
(Doebeli and Knowlton, 1998)? What is the correct spatial scale
for microbial interactions (Whitaker, 2009)? And, nally, how can
these results be applied to ght disease and promote cooperation
in synthetic systems (Axelrod et al., 2006; Wintermute and Silver,
2010)?
We applaud Hamilton for pointing out the important insight
that genes care about not only themselves, but also about
identical copies in other organisms. If well understood, his crucial
observation can be a powerful intuitive tool, a rst step to
understanding how or why a behavior might evolve. Unfortu-
nately, the limitations and indirect nature of inclusive tness
make it nearly impossible to correctly apply quantitatively and,
similar to group selection, can lead to false intuitions if misused.
Rather than resorting to very general forms of Hamiltons rule and
multilevel selection that often lack specicity and prediction
power, we conclude by advocating for experimentally inspired
and tested models to better understand the evolution of microbial
cooperation.
Acknowledgements
We would like to thank j. smith, M. van Veelen, E. Yurtsev, S.
Serene, A. Velenich, K. Korolev, and the rest of the laboratory for
stimulating discussions and constructive criticisms. We are also
J.A. Damore, J. Gore / Journal of Theoretical Biology ] (]]]]) ]]]]]] 9
Please cite this article as: Damore, J.A., Gore, J., Understanding microbial cooperation. J. Theor. Biol. (2011), doi:10.1016/
j.jtbi.2011.03.008
thankful to the two anonymous reviewers for their useful com-
ments. The laboratory acknowledges nancial support through an
NIH K99 Pathways to Independence Award.
Appendix A. Derivation of Hamiltons rule
We will derive Hamiltons rule similarly to Queller (1992).
First, let us perform a linear regression on the tness of an
organism, W, with respect to the individuals genotype, G
X
, and
the average genotype of its interactants, G
Y
, as independent
variables. The tness of any one organism is then
W W
0
b
WGX
G
X
b
WGY
G
Y
e A:1
where W
0
is the organisms base tness and the bs are partial
regression coefcients. The last term, e, is the residual, which may
be different for every individual and describes the difference
between an individuals actual tness and the tness predicted by
the regression. Substitution into (2) gives
WDG CovW
0
,G
X
b
WGX
CovG
X
,G
X
b
WGY
CovG
Y
,G
X
Cove,G
X

A:2
The bs drop out of the covariances because they are constants.
Similarly, CovW
0
,G
X
is zero because W
0
is a constant. Impor-
tantly, Cove,G
X
is also zero because the residuals are necessarily
uncorrelated with the independent variables of a linear regres-
sion. Now, because W is positive, the necessary condition for G to
increase is
WDG b
WGX
CovG
X
,G
X
b
WGY
CovG
Y
,G
X
40 A:3
Now, because CovG
X
,G
X
VarG
X
is necessarily non-negative,
we can divide both sides by VarG
X
and maintain the sign of the
inequality:
b
WGX
b
WGY
CovG
Y
,G
X

VarG
X

b
WGX
b
WGY
b
GY GX
40 A:4
which is the equation used in the main text. We should also
mention that we personally prefer the formula
b
WGX
b
WPY
CovP
Y
,G
X

VarG
X

40 A:5
where G
Y
is replaced with P
Y
. This emphasizes the important
point that altruistic genes spread when they are assorted with the
altruistic phenotypes of others. The original form is presented in
the text because it is more widely used and simpler.
Appendix B. Supplementary data
Supplementary data associated with this article can be found
in the online version at doi:10.1016/j.jtbi.2011.03.008.
References
Ackermann, M., Stecher, B., Freed, N., Songhet, P., Hardt, W., Doebeli, M., 2008.
Self-destructive cooperation mediated by phenotypic noise. Nature 454,
987990.
Aguilar, C., Vlamakis, H., Losick, R., Kolter, R., 2007. Thinking about Bacillus subtilis
as a multicellular organism. Current Opinion in Microbiology 10, 638643.
Anscombe, F., 1973. Graphs in statistical analysis. American Statistician 27, 1721.
Axelrod, R., Axelrod, D., Pienta, K., 2006. Evolution of cooperation among tumor
cells. Proceedings of the National Academy of Sciences 103, 1347413479.
Axelrod, R., Hamilton, W., 1981. The evolution of cooperation. Science 211,
13901396.
Br annstr om,

A., Dieckmann, U., 2005. Evolutionary dynamics of altruism and
cheating among social amoebas. Proceedings of the Royal Society B: Biological
Sciences 272, 1609.
Brown, S., Buckling, A., 2008. A social life for discerning microbes. Cell 135,
600603.
Cavalli-Sforza, L., Feldman, M., 1978. Darwinian selection and altruism. Theore-
tical Population Biology 14, 268280.
Chuang, J., Rivoire, O., Leibler, S., 2009. Simpsons paradox in a synthetic microbial
system. Science 323, 272275.
Chuang, J., Rivoire, O., Leibler, S., 2010. Cooperation and Hamiltons rule in a
simple synthetic microbial system. Molecular Systems Biology 6, 398.
Crespi, B., 2001. The evolution of social behavior in microorganisms. Trends in
Ecology & Evolution 16, 178183.
Damore, J.A., Gore, J. A slowly evolving host moves rst in symbiotic interactions.
Evolution, in press, doi:10.1111/j.1558-5646.2011.01299.x.
Dawkins, R., 1976. The Selsh Gene. Oxford University Press, New York City.
Diggle, S., Grifn, A., Campbell, G., West, S., 2007. Cooperation and conict in
quorum-sensing bacterial populations. Nature 450, 411414.
Doebeli, M., Knowlton, N., 1998. The evolution of interspecic mutualisms.
Proceedings of the National Academy of Sciences of the United States of
America 95, 86768680.
Dugatkin, L., Perlin, M., Lucas, J., Atlas, R., 2005. Group-benecial traits, frequency-
dependent selection and genotypic diversity: an antibiotic resistance para-
digm. Proceedings of the Royal Society B: Biological Sciences 272, 79.
Fletcher, J., Doebeli, M., 2009. A simple and general explanation for the evolution
of altruism. Proceedings of the Royal Society B: Biological Sciences 276, 1319.
Fletcher, J., Zwick, M., 2004. Strong altruism can evolve in randomly formed
groups. Journal of Theoretical Biology 228, 303313.
Fletcher, J., Zwick, M., 2006. Unifying the theories of inclusive tness and
reciprocal altruism. American Naturalist 168, 252262.
Fletcher, J., Zwick, M., 2007. The evolution of altruism: game theory in multilevel
selection and inclusive tness. Journal of Theoretical Biology 245, 2636.
Foster, K.R., Wenseleers, T., 2006. A general model for the evolution of mutualisms.
Journal of Evolutionary Biology 19, 12831293.
Frank, S., 1998. Foundations of Social Evolution. Princeton University Press,
Princeton, NJ.
Frey, E., 2010. Evolutionary game theory: theoretical concepts and applications to
microbial communities. Physica A: Statistical Mechanics and its Applications
389, 42654298.
Gilbert, O., Foster, K., Mehdiabadi, N., Strassmann, J., Queller, D., 2007. High
relatedness maintains multicellular cooperation in a social amoeba by con-
trolling cheater mutants. Proceedings of the National Academy of Sciences
104, 8913.
Godfrey-Smith, P., Kerr, B., 2009. Selection in ephemeral networks. The American
Naturalist 174, 906911.
Gore, J., Youk, H., Van Oudenaarden, A., 2009. Snowdrift game dynamics and
facultative cheating in yeast. Nature 459, 253256.
Grafen, A., 2006. Optimization of inclusive tness. Journal of Theoretical Biology
238, 541563.
Greig, D., Travisano, M., 2004. The Prisoners Dilemma and polymorphism in yeast
SUC genes. Proceedings of the Royal Society of London. Series B: Biological
Sciences 271, S25.
Hamilton, W., 1964. The genetical evolution of social behaviour I and II. Journal of
Theoretical Biology 7, 152.
Hardy, O., 2003. Estimation of pairwise relatedness between individuals and
characterization of isolation-by-distance processes using dominant genetic
markers. Molecular Ecology 12, 15771588.
Hauert, C., Michor, F., Nowak, M., Doebeli, M., 2006. Synergy and discounting
of cooperation in social dilemmas. Journal of Theoretical Biology 239,
195202.
Hibbing, M., Fuqua, C., Parsek, M., Peterson, S., 2010. Bacterial competition:
surviving and thriving in the microbial jungle. Nature Reviews Microbiology
8, 1525.
Karlin, S., Matessi, C., 1983. Kin selection and altruism. Proceedings of the Royal
Society of London B 219, 327353.
Kerr, B., Neuhauser, C., Bohannan, B., Dean, A., 2006. Local migration promotes
competitive restraint in a host-pathogen tragedy of the commons. Nature
442, 7578.
Kerr, B., Riley, M., Feldman, M., Bohannan, B., 2002. Local dispersal promotes
biodiversity in a real-life game of rock-paper-scissors. Nature 418, 171174.
K ummerli, R., Van Den Berg, P., Grifn, A., West, S., Gardner, A., 2010. Repression of
competition favours cooperation: experimental evidence from bacteria. Jour-
nal of Evolutionary Biology 23, 699706.
Le Gac, M., Doebeli, M., 2010. Environmental viscosity does not affect the
evolution of cooperation during experimental evolution of colicigenic bacteria.
Evolution 64, 522533.
Lee, H., Molla, M., Cantor, C., Collins, J., 2010. Bacterial charity work leads to
population-wide resistance. Nature 467, 8285.
Lewis, K., 2006. Persister cells, dormancy and infectious disease. Nature Reviews
Microbiology 5, 4856.
MacLean, R., Fuentes-Hernandez, A., Greig, D., Hurst, L., Gudelj, I., 2010. A mixture
of cheats and co-operators can enable maximal group benet. PLoS
Biology 8, 3135.
MacLean, R., Gudelj, I., 2006. Resource competition and social conict in experi-
mental populations of yeast. Nature 441, 498501.
Majeed, H., Gillor, O., Kerr, B., Riley, M., 2011. Competitive interactions in
Escherichia coli populations: the role of bacteriocins. The ISME Journal 5,
7181.
Mehdiabadi, N., Jack, C., Farnham, T., Platt, T., Kalla, S., Shaulsky, G., Queller, D.,
Strassmann, J., 2006. Social evolution: kin preference in a social microbe.
Nature 442, 881882.
J.A. Damore, J. Gore / Journal of Theoretical Biology ] (]]]]) ]]]]]] 10
Please cite this article as: Damore, J.A., Gore, J., Understanding microbial cooperation. J. Theor. Biol. (2011), doi:10.1016/
j.jtbi.2011.03.008
Mehta, P., Gregor, T., 2010. Approaching the molecular origins of collective
dynamics in oscillating cell populations. Current Opinion in Genetics &
Development, 574580.
Nogueira, T., Rankin, D., Touchon, M., Taddei, F., Brown, S., Rocha, E., 2009.
Horizontal gene transfer of the secretome drives the evolution of bacterial
cooperation and virulence. Current Biology 19, 16831691.
Nowak, M., 2006a. Evolutionary Dynamics: Exploring the Equations of Life.
Harvard University Press, Cambridge, MA.
Nowak, M., 2006b. Five rules for the evolution of cooperation. Science 314, 1560.
Nowak, M., Sasaki, A., Taylor, C., Fudenberg, D., 2004. Emergence of cooperation
and evolutionary stability in nite populations. Nature 428, 646650.
Nowak, M., Sigmund, K., 2005. Evolution of indirect reciprocity. Nature 437,
12911298.
Nowak, M., Tarnita, C., Wilson, E.O., 2010. The evolution of eusociality. Nature 466,
10571062.
Ohtsuki, H., Hauert, C., Lieberman, E., Nowak, M., 2006. A simple rule for the
evolution of cooperation on graphs and social networks. Nature 441, 502505.
Page, K., Nowak, M., 2002. Unifying evolutionary dynamics. Journal of Theoretical
Biology 219, 9398.
Price, G., 1970. Selection and covariance. Nature 227, 520521.
Queller, D., 1984. Kin selection and frequency dependence: a game theoretic
approach. Biological Journal of the Linnean Society 23, 133143.
Queller, D., 1985. Kinship, reciprocity and synergism in the evolution of social
behaviour. Nature 318, 366367.
Queller, D., 1992. A general model for kin selection. Evolution 46, 376380.
Queller, D., 1994. Genetic relatedness in viscous populations. Evolutionary Ecology
8, 7073.
Queller, D., Goodnight, K., 1989. Estimating relatedness using genetic markers.
Evolution 43, 258275.
Queller, D., Ponte, E., Bozzaro, S., Strassmann, J., 2003. Single-gene greenbeard
effects in the social amoeba Dictyostelium discoideum. Science 299, 105.
Reichenbach, T., Mobilia, M., Frey, E., 2007. Mobility promotes and jeopardizes
biodiversity in rock-paper-scissors games. Nature 448, 10461049.
Ross-Gillespie, A., Gardner, A., West, S., Grifn, A., 2007. Frequency dependence
and cooperation: theory and a test with bacteria. American Naturalist 170,
331342.
smith, j., Van Dyken, D., Zee, P., 2010. A generalization of Hamiltons rule for the
evolution of microbial cooperation. Science 328, 17001703.
Smukalla, S., Caldara, M., Pochet, N., Beauvais, A., Guadagnini, S., Yan, C., Vinces, M.,
Jansen, A., Prevost, M., Latge , J., et al., 2008. FLO1 is a variable green beard gene
that drives biolm-like cooperation in budding yeast. Cell 135, 726737.
Strassmann, J., Zhu, Y., Queller, D., 2000. Altruism and social cheating in the social
amoeba Dictyostelium discoideum. Nature 408, 965967.
Taylor, C., Nowak, M., 2007. Transforming the dilemma. Evolution 61, 22812292.
Taylor, P., Day, T., Wild, G., 2007. Evolution of cooperation in a nite homogeneous
graph. Nature 447, 469472.
Traulsen, A., 2010. Mathematics of kin- and group-selection: formally equivalent?
Evolution 64, 316323.
Traulsen, A., Nowak, M., 2006. Evolution of cooperation by multilevel selection.
Proceedings of the National Academy of Sciences 103, 1095210955.
Travisano, M., Velicer, G., 2004. Strategies of microbial cheater control. TRENDS in
Microbiology 12, 7278.
Turner, P., Chao, L., 1999. Prisoners dilemma in an RNA virus. Nature 398,
441443.
Turner, P., Chao, L., 2003. Escape from prisoners dilemma in RNA phage F6.
American Naturalist 161, 497505.
van Veelen, M., 2005. On the use of the Price equation. Journal of Theoretical
Biology 237, 412426.
van Veelen, M., 2007. Hamiltons missing link. Journal of Theoretical Biology 246,
551554.
van Veelen, M., 2009. Group selection, kin selection, altruism and cooperation:
when inclusive tness is right and when it can be wrong. Journal of Theoretical
Biology 259, 589600.
van Veelen, M., 2010. A rule is not a rule if it changes from case to case (a reply to
Marshalls comment). Journal of Theoretical Biology, 189195.
Velicer, G., 2003. Social strife in the microbial world. TRENDS in Microbiology 11,
330337.
Velicer, G., Yuen-tsu, N., 2003. Evolution of novel cooperative swarming in the
bacterium Myxococcus xanthus. Nature 425, 7578.
Wenseleers, T., 2006. Modelling social evolution: the relative merits and limita-
tions of a Hamiltons rule-based approach. Journal of Evolutionary Biology 19,
14191422.
Wenseleers, T., Gardner, A., Foster, K., 2009. Social evolution theory: a review of
methods and approaches. In: Szekely, T., Komdeur, J., Moore, A. (Eds.), Social
Behaviour. Cambridge University Press, Cambridge, UK.
West, S., Diggle, S., Buckling, A., Gardner, A., Grifn, A., 2007. The social lives of
microbes. Annual Review of Ecology, Evolution, and Systematics 38, 5376.
West, S., Grifn, A., Gardner, A., Diggle, S., 2006. Social evolution theory for
microorganisms. Nature Reviews Microbiology 4, 597607.
Whitaker, R., 2009. Evolution: spatial scaling of microbial interactions. Current
Biology 19, R954R956.
Wild, G., Traulsen, A., 2007. The different limits of weak selection and the
evolutionary dynamics of nite populations. Journal of Theoretical Biology
247, 382390.
Wilson, D., 1975. A theory of group selection. Proceedings of the National
Academy of Sciences of the United States of America 72, 143.
Wintermute, E., Silver, P., 2010. Emergent cooperation in microbial metabolism.
Molecular Systems Biology 6.
Xavier, J., Foster, K., 2007. Cooperation and conict in microbial biolms.
Proceedings of the National Academy of Sciences 104, 876881.
J.A. Damore, J. Gore / Journal of Theoretical Biology ] (]]]]) ]]]]]] 11
Please cite this article as: Damore, J.A., Gore, J., Understanding microbial cooperation. J. Theor. Biol. (2011), doi:10.1016/
j.jtbi.2011.03.008