Interacting phenotypes and the evolutionary process. III. Social evolution.
ABSTRACT Interactions among conspecifics influence social evolution through two distinct but intimately related paths. First, they provide the opportunity for indirect genetic effects (IGEs), where genes expressed in one individual influence the expression of traits in others. Second, interactions can generate social selection when traits expressed in one individual influence the fitness of others. Here, we present a quantitative genetic model of multivariate trait evolution that integrates the effects of both IGEs and social selection, which have previously been modeled independently. We show that social selection affects evolutionary change whenever the breeding value of one individual covaries with the phenotype of its social partners. This covariance can be created by both relatedness and IGEs, which are shown to have parallel roles in determining evolutionary response. We show that social selection is central to the estimation of inclusive fitness and derive a version of Hamilton's rule showing the symmetrical effects of relatedness and IGEs on the evolution of altruism. We illustrate the utility of our approach using altruism, greenbeards, aggression, and weapons as examples. Our model provides a general predictive equation for the evolution of social phenotypes that encompasses specific cases such as kin selection and reciprocity. The parameters can be measured empirically, and we emphasize the importance of considering both IGEs and social selection, in addition to relatedness, when testing hypotheses about social evolution.
[show abstract] [hide abstract]
ABSTRACT: Climate change is imposing intensified and novel selection pressures on organisms by altering abiotic and biotic environmental conditions on Earth, but studies demonstrating genetic adaptation to climate change mediated selection are still scarce. Evidence is accumulating to indicate that both genetic and ecological constrains may often limit populations' abilities to adapt to large scale effects of climate warming. These constraints may predispose many organisms to respond to climate change with range shifts and phenotypic plasticity, rather than through evolutionary adaptation. In general, broad conclusions about the role of evolutionary adaptation in mitigating climate change induced fitness loss in the wild are as yet difficult to make.BioEssays 07/2012; 34(9):811-8. · 4.95 Impact Factor
Article: Wax on, wax off: nest soil facilitates indirect transfer of recognition cues between ant nestmates.[show abstract] [hide abstract]
ABSTRACT: Social animals use recognition cues to discriminate between group members and non-members. These recognition cues may be conceptualized as a label, which is compared to a neural representation of acceptable cue combinations termed the template. In ants and other social insects, the label consists of a waxy layer of colony-specific hydrocarbons on the body surface. Genetic and environmental differences between colony members may confound recognition and social cohesion, so many species perform behaviors that homogenize the odor label, such as mouth-to-mouth feeding and allogrooming. Here, we test for another mechanism of cue exchange: indirect transfer of cuticular hydrocarbons via the nest material. Using a combination of chemical analysis and behavioral experiments with Camponotus aethiops ants, we show that nest soil indirectly transfers hydrocarbons between ants and affects recognition behavior. We also found evidence that olfactory cues on the nest soil influence nestmate recognition, but this effect was not observed in all colonies. These results demonstrate that cuticular hydrocarbons deposited on the nest soil are important in creating uniformity in the odor label and may also contribute to the template.PLoS ONE 01/2011; 6(4):e19435. · 4.09 Impact Factor
[show abstract] [hide abstract]
ABSTRACT: Many biologists are calling for an 'extended evolutionary synthesis' that would 'modernize the modern synthesis' of evolution. Biological information is typically considered as being transmitted across generations by the DNA sequence alone, but accumulating evidence indicates that both genetic and non-genetic inheritance, and the interactions between them, have important effects on evolutionary outcomes. We review the evidence for such effects of epigenetic, ecological and cultural inheritance and parental effects, and outline methods that quantify the relative contributions of genetic and non-genetic heritability to the transmission of phenotypic variation across generations. These issues have implications for diverse areas, from the question of missing heritability in human complex-trait genetics to the basis of major evolutionary transitions.Nature Reviews Genetics 01/2011; 12(7):475-86. · 38.08 Impact Factor
INTERACTING PHENOTYPES AND THE
EVOLUTIONARY PROCESS. III. SOCIAL
Joel W. McGlothlin,1,2Allen J. Moore,3,4Jason B. Wolf,5,6and Edmund D. Brodie III1,7
1Department of Biology, University of Virginia, Charlottesville Virginia 22904
3School of Biosciences, University of Exeter, Cornwall TR10 9EZ, United Kingdom
5Department of Biology and Biochemistry, University of Bath, Bath BA2 7AY, United Kingdom
Received November 6, 2009
Accepted March 11, 2010
Interactions among conspecifics influence social evolution through two distinct but intimately related paths. First, they provide
the opportunity for indirect genetic effects (IGEs), where genes expressed in one individual influence the expression of traits in
others. Second, interactions can generate social selection when traits expressed in one individual influence the fitness of others.
Here, we present a quantitative genetic model of multivariate trait evolution that integrates the effects of both IGEs and social
the breeding value of one individual covaries with the phenotype of its social partners. This covariance can be created by both
relatedness and IGEs, which are shown to have parallel roles in determining evolutionary response. We show that social selection is
central to the estimation of inclusive fitness and derive a version of Hamilton’s rule showing the symmetrical effects of relatedness
and IGEs on the evolution of altruism. We illustrate the utility of our approach using altruism, greenbeards, aggression, and
weapons as examples. Our model provides a general predictive equation for the evolution of social phenotypes that encompasses
specific cases such as kin selection and reciprocity. The parameters can be measured empirically, and we emphasize the importance
of considering both IGEs and social selection, in addition to relatedness, when testing hypotheses about social evolution.
KEY WORDS: Altruism, greenbeards, indirect genetic effects, kin selection, reciprocity, social selection.
Evolution by natural selection can be modeled as two sequen-
tial processes: phenotypic selection, which leads to changes in
the distribution of phenotypes within a generation, and inheri-
tance, which translates these changes across generations (Fisher
view, fitness differences among individuals reflect responses to
the physical environment, giving rise to selection favoring phe-
notypes appropriate to that environment. Evolutionary change
then occurs when alleles associated with these favored pheno-
types increase in frequency across generations. The relationships
between genotype and phenotype and between phenotype and fit-
ness are often more complex than this, however, and these intrica-
cies may influence the rate and direction of phenotypic evolution
(Dawkins 1982; Mousseau and Fox 1998; Wolf et al. 1998, 2000;
Pigliucci 2001; Odling-Smee et al. 2003; West-Eberhard 2003;
Traits involved in interactions among conspecifics are no-
C ?2010 The Author(s). Journal compilation C ?2010 The Society for the Study of Evolution.
Evolution 64-9: 2558–2574
INDIRECT GENETIC EFFECTS AND SOCIAL SELECTION
(Darwin 1859; West-Eberhard 1979, 1983; Lande 1981;
Frank 1998). In animals, such interactions typically involve
social behavior, the study of which makes up much of the field of
behavioral ecology (Sz´ ekely et al. 2010; Westneat and Fox 2010).
However, social interactions do not necessarily require an animal
nervous system and occur in virtually all taxa (Frank 2007; e.g.,
microorganisms, Crespi 2001; Foster 2010; plants, Dicke et al.
2003; Karban 2008). Behavior such as aggression and courtship
in animals are the most intuitive examples, but competitive inter-
actions such as relative growth rates of neighboring plants exhibit
the same kinds of phenotypic feedback (Mutic and Wolf 2007).
Social effects on phenotypic evolution may occur whenever “in-
teracting phenotypes” are present, that is, when the phenotype of
one individual affects the phenotype or fitness of a conspecific
(Moore et al. 1997; Wolf et al. 1998, 1999; Bleakley et al. 2010;
Wolf and Moore 2010).
Social interactions moderate inheritance by altering the re-
lationship between genotype and phenotype. For example, the
behavior of a focal individual may depend on the behavior of its
social partner(s) so that phenotypic expression is determined by
both the individual’s own genes and those of its social partner(s)
(Fig. 1). The phenotypic effects of genes in social partners are
known as “indirect genetic effects” (IGEs, Moore et al. 1997;
Wolf et al. 1998; Wolf and Moore 2010) or associative genetic
effects (Griffing 1967, 1969, 1976, 1981a; Bijma et al. 2007a;
Bijma and Wade 2008). Because they are genetic in origin, IGEs
represent an environmental source of variance that is heritable
and thus contributes to the evolutionary response to selection.
Figure 1. Path diagram depicting the effects of interacting phe-
individual are shown with no superscript, and variables associated
with its interactant are given a prime. Each phenotype is affected
by additive genetic (a), environmental (e), and indirect genetic ef-
fects. Relatedness between the two individuals is represented by
the path coefficient r. The path coefficient ψ translates the inter-
actant’s phenotype into an indirect genetic effect, which is shown
as a double arrow to indicate that the effect is reciprocal. The
fitness of the focal individual (w) is affected by both its own phe-
notype (nonsocial selection, βN) and that of its interactant (social
Consequently, IGEs can alter the evolutionary dynamics of traits
compared to those expected under traditional quantitative genetic
models (Moore et al. 1997).
Social interactions affect selection by directly generating fit-
ness differences among individuals. This path is termed social
selection, which is broadly defined to include any case in which
conspecific interactions lead to variance in reproductive success,
including competition, cooperation, and sexual selection (Darwin
1859; Wynne-Edwards 1962; Crook 1972; West-Eberhard 1979,
1983; Frank 2006). Although this perspective captures the range
of relevant interactions, it does not provide a means to distinguish
social selection from nonsocial selection in a quantitative way.
Wolf et al. (1999) narrowed the definition by explicitly recogniz-
ing social selection as effects on the fitness of one individual that
directly result from traits of interacting individuals. There are at
least three advantages to adopting this definition of social selec-
tion. First, social selection can be measured using standard and
generateawealth of dataonindividual-level selection(Landeand
Arnold 1983; Kingsolver et al. 2001). This approach measures
social selection as the effect of social-partner or group traits on
that individual’s own traits (as in contextual analysis, Heisler and
Damuth 1987; Goodnight et al. 1992). Second, the accounting of
fitness is individual based and remains consistent across all indi-
viduals in the population. The individual fitness approach yields
equivalent results to inclusive and multilevel fitness formulation
when appropriate assumptions are made (Queller 1992a; Frank
1998; Taylor et al. 2007; Bijma and Wade 2008). Third, selection
gradients can be combined with quantitative genetic parameters
to predict evolutionary response to selection (Lande 1979; Lande
and Arnold 1983; Bijma and Wade 2008).
lation (Moore et al. 1997; Wolf et al. 1999). However, when traits
function as interacting phenotypes, their effects on both stages of
the evolutionary process are inexorably coupled. The same phe-
notypic interactions that affect the paths of inheritance within a
population also are expected to affect the modes and targets of
selection. We therefore seek to integrate previous work by deriv-
ing a general multivariate equation for the response to social and
cial selection is predicted whenever there is a nonzero covariance
variance may arise from relatedness (or other similar associations
between direct breeding values), IGEs, or both. Our results re-
veal a fundamental symmetry between coefficients of relatedness
and the strength of IGEs. We provide examples showing how the
symmetry between relatedness and IGEs alters the predictions of
Hamilton’s (1963, 1964a,b) rule for the evolution of altruism, and
EVOLUTION SEPTEMBER 2010
JOEL W. MCGLOTHLIN ET AL.
how altruism, greenbeards, aggression, and weapons can evolve
by social selection and IGEs.
Evolutionary Response to Social
A GENERAL EQUATION
In the most general sense, interacting phenotypes are either traits
that affect the phenotype or fitness of another individual or traits
tend to be subject to two types of selection. As with any trait,
individual, or nonsocial, selection arises when the fitness of the
of a focal individual is affected by the phenotype of a social
partner or partners, social selection occurs (Wolf et al. 1999). The
contribution of social selection to the total covariance between
phenotype and fitness (the selection differential, s) depends on
the covariance between the phenotypes of the focal individual
and its social partner. In matrix notation,
s = PβN+ CIβS,
where s is a column vector of selection differentials, P represents
the phenotypic variance–covariance matrix, CIis the matrix of
covariances among the phenotypes of interactants, βNis a column
vector of individual, or nonsocial, selection gradients, and βSis
a column vector of social selection gradients (cf. equation 11 of
Wolf et al. 1999). This approach to partitioning selection cor-
responds to the “neighbor” approach derived by Okasha (2006)
use matrix notation to achieve multivariate generality. The one-
trait and two-trait examples discussed in earlier papers [Moore
et al. 1997; Wolf et al. 1999] can be derived from the multi-
variate equations presented here). To facilitate following models,
we provide in Table 1 a summary of the notation used in this
Equation (1) shows that a nonzero CIis necessary for social
selection to contribute to overall phenotypic selection on a trait.
intothemultivariatebreeder’sequation,?¯ z = GP−1s(whereGis
the additive genetic (co)variance matrix, Lande 1979) to predict
the evolutionary response to selection (represented by the vec-
tor of changes in mean trait values, ?¯ z). This is because CIcan
reflect both heritable and environmental sources of covariance.
For example, a nonzero CImay arise when the phenotypes of
neighbors covary solely because they exist in more similar envi-
ronments relative to some other group of neighbors; that is, there
is environmentally determined population subdivision.
1970) to derive generalized equations for the response to simulta-
neous social and nonsocial selection (see also Frank 1997, 1998;
Rice 2004; Bijma et al. 2007a; Bijma and Wade 2008). We first
develop the equations for pairs of interacting individuals for sim-
plicity and later extend the case to larger groups. Price’s theorem
Table 1. Notation used in this article.
Vectors of additive genetic and residual values
Vectors of social (indirect) breeding and residual values (Bijma et al. 2007a; McGlothlin and Brodie 2009)
Vector of total breeding values (Moore et al. 1997; Bijma et al. 2007a; McGlothlin and Brodie 2009)
Vector of nonsocial (“natural”) selection gradients (Wolf et al. 1999)
Vector of social selection gradients (Wolf et al. 1999)
Covariance matrices of a focal individual’s total breeding values with its own phenotypes and those of an interactant
Matrix of phentoypic covariance between interacting individuals (Wolf et al. 1999)
Additive genetic variance–covariance matrix (Lande 1979)
Variance–covariance matrix of direct breeding values (McGlothlin and Brodie 2009)
Covariance matrix of direct and social breeding values and its transpose (McGlothlin and Brodie 2009)
Variance–covariance matrix of social breeding values (McGlothlin and Brodie 2009)
Phenotypic variance–covariance matrix (Lande 1979)
Matrix of indirect genetic effects coefficients ψij(Moore et al. 1997)
Univariate and multivariate relatedness
Multivariate selection differential, a vector (Lande 1979)
Relative fitness, a scalar
Vectors of focal individual and interactant phenotypes (Moore et al. 1997)
Multivariate evolutionary change in phenotypic mean, a vector (Lande 1979)
2560EVOLUTION SEPTEMBER 2010
INDIRECT GENETIC EFFECTS AND SOCIAL SELECTION
between total breeding values and fitness, or
?¯ z = cov(A,w),
where A is a vector of total breeding values for each phenotype
of interest. An individual’s total breeding value represents its
expected contribution to average offspring phenotype, and may
consist of both direct and indirect genetic effects (Falconer and
MacKay 1996; Moore et al. 1997; Bijma et al. 2007a; McGlothlin
and Brodie 2009). Following Wolf et al. (1999), we represent
relative fitness, w, as a regression equation including the vectors
of focal-individual and social-partner phenotypic values,
w = α + zTβN+ z?TβS+ ε,
where α is an intercept, ε is an error term, z is a vector of phe-
notypic values in focal individuals, z?is a vector of phenotypes
in social partners, and T denotes matrix transposition (Fig. 1).
Throughout our derivation, we use a prime to denote variables
that belong to a social partner rather than the focal individual.
Substituting equation (3) into equation (2), we find
?¯ z = cov?A,α + zTβN+ z?TβS+ ε?
= cov(A,zT)βN+ cov(A,z?T)βS,
which can be written in a compact form as
?¯ z = CAzβN+ CAz?βS.
Equation (4b) shows that the portion of evolutionary change
attributable to nonsocial selection is proportional to the matrix
of covariances between the focal individuals’ total breeding val-
ues and their own phenotypic values (CAz). Similarly, the por-
tion of evolutionary change due to social selection is propor-
tional to the matrix of covariances between the focal individuals’
breeding values and the phenotypic values of their social part-
ners (CAz?). In other words, in order for evolutionary change to
occur, a nonrandom association between the genes of individu-
als and the phenotypes of their social partners must accompany
social selection. The causes of a nonzero CAz? can be divided
into two general classes: either individuals are associated nonran-
domly based on their genotypes (e.g., relatedness) or phenotypic
expression is influenced by the association between individuals
Below, we explore the components of the covariance ma-
trices using a trait-based model of IGEs (Moore et al. 1997). In
approach adopted here and the phenotypic variance component
models of IGEs of Griffing (1967, 1969, 1976, 1981a) and Bijma
and colleagues (Bijma et al. 2007a,b; Bijma and Wade 2008).
RELATEDNESS AND IGEs
The phenotypic vectors of two interacting individuals are defined
z = a + e + ?z?
z?= a?+ e + ?z,
matrix of interaction coefficients ?, which consists of elements
ψijdescribing the effect of trait j in an interactant on trait i in
a focal individual (Moore et al. 1997). Again, throughout this
derivation, we use primes to denote that a variable belongs to the
focal individual’s social partner.
Substituting equation (5b) into (5a) and vice versa, the phe-
notypic vectors can be defined explicitly as
z = (I − ??)−1(a + e + ?a?+ ?e?) (6a)
z?= (I − ??)−1(a?+ e?+ ?a + ?e),
where a is a column vector of additive genetic values, e is a
column vector of environmental effects, I represents the identity
matrix (Fig. 1). As in Moore et al. (1997), we assume that the
total effects of interacting phenotypes can be decomposed into
additive genetic (?a?) and environmental effects (?e?). That is,
we assume the effects of a social partner’s genes and environment
on the expression of a focal individual’s phenotypes occur solely
via effects of the social partner’s phenotype (Fig. 1). The factor
(I − ??)−1represents the potential for feedback, which arises
because traits in the two individuals may simultaneously affect
one another (see Moore et al. 1997 for derivation). The vector
of total breeding values can be determined by solving for an
individual’s additive genetic contribution to the expectation of
A = (I − ??)−1(a + ?a),
A = (I − ?)−1a
(see Moore et al. 1997 for derivation). Substituting equations (6)
and (7) into equation (4a), we find
?¯ z = cov[(I−?)−1a,[(I−??)−1(a+e+?a?+?e?)]T]βN
+ cov[(I − ?)−1a,[(I − ??)−1(a?+ e?+ ?a + ?e)]T]βS.
To simplify this equation, we make the standard quantitative ge-
netic assumption that all covariances between additive genetic
values and residual (environmental) values are zero, giving
EVOLUTION SEPTEMBER 2010
JOEL W. MCGLOTHLIN ET AL.
?¯ z = (I − ?)−1[G + cov(a,a?T)?T](I − ?T?T)−1βN
+ (I − ?)−1[cov(a,a?T) + G?T](I − ?T?T)−1βS.
or due to population genetic structure, cov(a, a?T) = rG, where r
is the coefficient of relatedness, and equation (9) simplifies to
?¯ z = (I − ?)−1G(I + r?T)(I − ?T?T)−1βN
+ (I − ?)−1G(rI + ?T)(I − ?T?T)−1βS.
The formulation in equation (10) uses a regression-based
definition of relatedness, which can incorporate both pedigree re-
latedness and population structure (Wright 1965; Hamilton 1972;
Michod and Hamilton 1980; Uyenoyama and Feldman 1981;
Uyenoyama 1984; Grafen 1985; Queller 1992b; Bijma et al.
2007a). Hamilton’s regression definition was stated in its most
general form, r =cov(a, a?)
will be zero or positive, but may be negative if individuals avoid
(Hamilton 1972; Rousset 2002; Konovalov and Heg 2008).
Equation (10) assumes that relatedness is uniform across
all traits. This assumption is likely to be violated whenever in-
dividuals assort nonrandomly based on specific phenotypes, as
Wilson 1975; Dawkins 1976; Wilson and Dugatkin 1997; Wolf
et al. 1999; Pepper 2000; Jansen and van Baalen 2006; Grafen
2009; Gardner and West 2010). To encompass such situations
and achieve greater generality, we define a matrix of association
among breeding values as
var(a), by Queller (1992b). In most cases, r
R = cov(a,a?T)G−1,
which is a simple multivariate extension of Queller’s (1992b)
expression of additive genetic relatedness. R is a square matrix
of partial regression coefficients rijthat describe the translation
between the additive genetic values of focal individuals and their
interactants. In general, cov(a, a?T) will tend to be symmetrical,
but R will not. Incorporating R into equation (9) and simplifying,
?¯ z = (I − ?)−1G(I + RT?T)(I − ?T?T)−1βN
+ (I − ?)−1G(RT+ ?T)(I − ?T?T)−1βS.
When relatedness is uniform across traits, R = rI, and equa-
tion (12) reduces to equation (10). Otherwise, it is difficult to
determine the structure of R based on a priori expectations, al-
though specific structures may be postulated to explore specific
models of social evolution. For example, in a true greenbeard
situation, individuals assort nonrandomly based on a single rec-
ognizible trait that signals (zi) the propensity to behave altruisti-
cally (Dawkins 1976; Gardner and West 2010). Here, R could
be modeled as zero except for a single diagonal element rii.
Off-diagonal elements of R would be nonzero when one trait
in a focal individual predicts an association with a second trait
in its partner, as in the coevolution of male ornaments and fe-
male preference in sexual selection (Lande 1981; Kirkpatrick
1982). R is therefore a parameter that can be experimentally mea-
sured to test predictions associated with specific models of social
Equation (12) is general and can be applied to many specific
situations. In the absence of IGEs and social selection, equa-
tion (12) simplifies to the standard multivariate breeder’s equa-
tion (Lande 1979; Lande and Arnold 1983). In the Discussion,
we present worked examples for two traits showing how IGEs
and social selection combine to lead to evolution of traits such as
altruism, greenbeards, aggression, and weapons.
EVOLUTIONARY RESPONSES TO SOCIAL VERSUS
To understand how the response to social selection differs from
in detail. Comparing equations (12) and (4b), we see that the
portion of the evolutionary response due to nonsocial selection is
determined by CAzand that this matrix can be defined as
CAz= (I − ?)−1G(I + RT?T)(I − ?T?T)−1.
Similarly, the response to social selection is determined by CAz?,
which is defined as
CAz? = (I − ?)−1G(RT+ ?T)(I − ?T?T)−1.
Common to both responses are the factors (I − ?)−1and (I −
?T?T)−1, which represent the linear and feedback influences,
Bleakley et al. 2010). Thus, the portions of evolutionary change
due to nonsocial and social selection are influenced equally by
two of the main effects of IGEs: alteration of the relationship
between genotype and phenotype and the acceleratory effects of
feedback between interacting individuals.
(I + RT?T) and (RT+ ?T) (and, when relatedness is assumed to
be uniform across traits, their analogs in eq. 10). The first factor
that the response to nonsocial selection is generally increased by
IGEs that occur among related individuals (Griffing 1976, 1981b;
Bijma et al. 2007a; Ellen et al. 2007; Bijma and Wade 2008). The
second factor shows that both relatedness and IGEs contribute to
the response to social selection (Griffing 1976; Muir 1996; Bijma
et al. 2007a; Bijma and Wade 2008), and do so independently and
2562EVOLUTION SEPTEMBER 2010
INDIRECT GENETIC EFFECTS AND SOCIAL SELECTION
EFFECTS OF GROUP SIZE
Although our treatment so far has been limited to interactions
between two individuals to illustrate basic concepts, social inter-
actions often involve multiple individuals. With multiple social
partners, fitness can be expressed as
w = α + zTβN+ (n − 1)¯ z?TβS+ ε,
where the social selection gradient is defined as a single inter-
actant’s contribution to the fitness of a focal individual and n
represents group size. The mean phenotype of a focal individual’s
social partners is represented by ¯ z?. Equation (15) assumes that
groups are of equal size, but may be extended to include variable
group sizes by replacing n with the arithmetic mean group size, ¯ n.
When group-level selection is of interest, fitness can be defined
w = α + zTβN+ ¯ z?Tβgroup+ ε,
(?¯¯ z) by substituting equation (15), obtaining
?¯¯ z = cov(A,zT)βN+ (n − 1)cov(A, ¯ z?T)βS.
Equation (17) assumes that group size is either invariant, or (re-
placingnwith ¯ n)isvariablebuthasnogeneticbasis,butotherwise,
equation (17) is general and applicable to populations with any
group structure. To quantify the influences of relatedness versus
IGEs, the group structure must be specified explicitly. In Ap-
pendix B, we derive the case for a population with equally sized,
nonoverlapping groups of n individuals, which yields
?¯¯ z = [I − (n − 1)?]−1G[I − (n − 2)?T+ r(n − 1)?T]
× [I − (n − 2)?T− (n − 1)?T?T]−1βN
+ (n − 1)[I − (n − 1)?]−1G(rI + ?T)
× [I − (n − 2)?T− (n − 1)?T?T]−1βS.
We also derive an equivalent equation using the variance-
components framework in Appendix B.
Examination of equation (18) reveals that group size affects
both the linear and feedback effects of ? for both nonsocial and
social selection. The effect of group size on the response to selec-
tion when IGEs are present will depend on the direction and mag-
nitude of the elements of ?. The single-trait case is particularly
instructive, as an IGE between the same trait in two interactants
can cause a strong feedback effect that drastically increases the
response to selection (see Moore et al. 1997, Fig. 3B). Figure 2
shows the change in the evolutionary response to nonsocial selec-
relative to a model that includes only direct genetic effects. Using
Figure 2. Relative change in the portion of evolutionary response
of a single trait due to (A) nonsocial and (B) social selection as
a function of the strength of the indirect genetic effect (ψ) and
group size (n), when compared to a model that does not include
indirect genetic effects (see equations 19 and 20 to derive A and
B, respectively). Note that in both graphs, the range of is limited
to −1 < ψ < 1/(n − 1).
a single-trait version of equation (18) with r = 0, these can be
shown to be
1 − (n − 2)ψ
[1 − (n − 1)ψ][1 − (n − 2)ψ − (n − 1)ψ2]
for the response to nonsocial selection and
(n − 1)ψ
[1 − (n − 1)ψ][1 − (n − 2)ψ − (n − 1)ψ2]
evolutionary rate rapidly approaches infinity as ψ increases, but
does so for smaller values of ψ as group size increases. However,
EVOLUTION SEPTEMBER 2010
JOEL W. MCGLOTHLIN ET AL.
when ψ < 0, group size has very little effect. In other words, it
takes a very large negative ψ to create evolutionary acceleration
large effects if the group size is large. The ratios in equations (19)
and (20) suggest that for a single-trait system, −1 < ψ < 1/(n −
zero. Ultimately, determining the sign and magnitude of ψ is an
empirical question and provides a test of our model’s predictions
regarding social evolution.
the denominator in equations (19) and (20) more quickly than do
negative values of ψ. This occurs due to two separate effects.
First, the linear effect of IGEs, [1 − (n − 1)ψ]−1, decreases when
ψ < 0 but increases when ψ > 0. This effect compensates for
the effect of group size on the numerator in equation (19), which
decreases with (n − 2). Second, the feedback effect of IGEs, [1 −
(n − 2)ψ − (n − 1)ψ2]−1, increases with increasing group size
when ψ > 0 but decreases with increasing group size when ψ <
0. This is caused by the term (n − 2)ψ.
These considerations suggest that positive feedback is more
effective at producing evolutionary acceleration than is negative
feedback. This asymmetry may be visualized by thinking of pos-
itive values of ψ as attractive forces that act to make pairs of
interacting individuals more similar. With positive values of ψ,
even small effects can act to make all individuals in a group
nearly identical. Negative values of ψ may be visualized as repel-
lent forces, pushing phenotypes of pairs of individuals in a group
away from each other. As groups become larger, any given pair
of phenotypes cannot be spread apart maximally because these
phenotypes are simultaneously affected by the phenotypes of all
the other group members.
The results above decompose IGEs caused by group interac-
tions into cumulative effects of individuals. However, in certain
systems, particularly when groups are large, it may be simpler
to consider total effects of a group on phenotypic expression.
Such group-level IGEs can be encompassed in the matrix ?group
(McGlothlin and Brodie 2009).
Altruism, Social Selection, and IGEs
HAMILTON’S RULE EXTENDED
We analyze the evolution of altruism using our model to illus-
trate the similar but independent roles of relatedness and IGEs in
determining the response to social selection. As summarized by
Hamilton’s rule, altruism should increase in frequency if the cost
to the performer (C) is outweighed by the benefit to the recip-
ient (B) multiplied by the relatedness of the two individuals (r)
C < rB.
In the notation of phenotypic selection, −βN and βS naturally
correspond to C and B in Hamilton’s formulation, because they
measure the fitness cost of possessing a given trait value z and
the fitness benefit of interacting with an individual possessing
trait value z?(Wolf et al. 1999). This notation emphasizes that
the problem of altruism arises when two levels of selection—
Hamilton 1975; Wade 1985; Taylor and Frank 1996; Wilson and
Wolf et al. (1999; Wolf and Moore 2010) derive a pheno-
typic version of Hamilton’s rule based on equation (1), which
can be used as a heuristic tool when genetic data are unavail-
able. Similarly, Goodnight et al. (1992; Goodnight 2005) derives
Hamilton’s rule in terms of individual- and group-level selection.
However, the phenotypic versions of Hamilton’s rule cannot be
used to quantitatively predict the evolution of altruistic traits, be-
cause they consider only selection and not the inheritance that
mediates the response across generations. Here, we derive a gen-
eral version of Hamilton’s rule for interacting phenotypes. For
simplicity, we consider only a single-trait case for a pair of in-
teracting individuals, but our derivation can be easily extended to
describe multivariate suites of traits or larger groups. Using a uni-
variate version of equation (4b) and asking when the phenotype z
will increase (i.e., when ?¯ z > 0):
0 < CAzβN+ CAz?βS,
which can be simplified to
This inequality is identical to the general version of Hamilton’s
rule derived by Queller (1985, 1992b). The ratioCAz?
of the similarity of interacting individuals. It will tend to be pro-
portional to the phenotypic correlation between interactants (the
defining parameter of the phenotypic Hamilton’s rule; Wolf et al.
1999), but measures only genetic effects.
In a one-trait model with relatedness and IGEs, CAzand CAz?
are defined as
CAzis a measure
1 + rψ
(1 − ψ)(1 − ψ2)
CAz? = G
r + ψ
(1 − ψ)(1 − ψ2).
Substituting these definitions into inequality (22b), we find
r + ψ
1 + rψβS.
2564EVOLUTION SEPTEMBER 2010
INDIRECT GENETIC EFFECTS AND SOCIAL SELECTION
Figure 3. Contribution of the indirect genetic effect (ψ) and re-
latedness (r) to the evolution of a single altruistic trait (inequality
25). The balance between social and nonsocial selection in the ex-
tended version of Hamilton’s rule is determined by the quantity
plotted on the vertical axis. See text for discussion.
This formulation shows that r and ψ have symmetrical effects
on the evolution of altruism (Fig. 3). Indeed, when there are no
IGEs, ψ = 0 and our result reduces to the typical formulation of
Hamilton’s rule (albeit expressed with different notation):
Similarly, when r = 0,
This result shows the similarity of r and ψ. Both r and ψ can
be expressed as regression coefficients: r is the regression of an
individual’s direct genetic effect on its partner’s direct genetic
effect (Hamilton 1972), whereas ψ is the regression of an individ-
ual’s direct genetic effect on its partner’s indirect genetic effect
(McGlothlin and Brodie 2009). In this sense, ψ can be thought of
nents of relatedness are encompassed by Queller’s (1985, 1992b)
generalized definition of relatedness, which in our notation is ex-
relatedness from indirect genetic relatedness is the potential to
parse evolutionary consequences of interactions among relatives
from those due to phenotypic influences that may occur between
Inequality (27) demonstrates that altruism should be able to
evolve without direct genetic relatedness, as long as IGEs provide
form of relatedness (West et al. 2007, 2008) but helps explain
CAz. The advantage of distinguishing direct genetic
some of the controversy as the genetic effects need not be direct.
For the single-trait case, ψ must be positive in order for IGEs
to drive the evolution of altruism. When ψ > 0, the phenotypes
of two interacting individuals become more similar (Moore et al.
1997). Thus, ψ may be considered to represent the strength of
reciprocity (Bleakley and Brodie 2009; Bleakley et al. 2010), and
to the extent that altruism depends upon ψ, it may be considered
addition to reciprocity, ψ may be used to model other behavioral
mechanisms such as manipualtion or punishment, that may lead
to cooperation between individuals (Clutton-Brock 2009).
In addition to deriving a general rule for the evolution of altru-
ism, Hamilton (1964a) defined the concept of inclusive fitness,
a measure of fitness that takes into account how one’s actions
affect one’s relatives. According to Hamilton, an individual’s in-
clusive fitness can be defined as personal fitness when social
effects of neighbors are removed but social effects on neighbors,
weighted by relatedness, are added. Hamilton (1964a) and oth-
ers (Grafen 2006, 2007; Gardner and Grafen 2009; Bijma 2010)
have argued that selection tends to maximize or optimize inclu-
sive fitness. However, measuring inclusive fitness is not always
straightforward, particularly for phenotypes (Grafen 1982; Creel
that inclusive fitness may differ across traits, making the notion
of measuring an individual’s inclusive fitness impossible (Queller
Using our model, it can be shown that inclusive fitness can
be defined as a quantity belonging to an individual, even when
our multivariate notation, we can define relative inclusive fitness
winc= w − z?TβS+ zTC−1
This definition holds when individuals interact in pairs. For larger
groups, the last two terms on the right-hand side should be
weighted by group size. Substituting our definition of relative
fitness (eq. 3), we see that
winc= α + zT?βN+ C−1
This definition of inclusive fitness assumes that all pertinent traits
(i.e., all interacting phenotypes) have been measured. It is easy to
check that evolutionary change is reliably predicted using either
relative fitness or relative inclusive fitness, or
EVOLUTION SEPTEMBER 2010
JOEL W. MCGLOTHLIN ET AL.
?¯ z = cov(A,w) = cov(A,winc).
We can explore how relatedness and IGEs affect relative
inclusive fitness by substituting equations (13–14) into equa-
tion (29) and simplifying,
winc= α + zT?βN+ (I − ?T?T)(I + RT?T)−1
× (RT+ ?T)(I − ?T?T)−1βS
Thus, if we are interested in estimating inclusive fitness, we must
consider both relatedness and IGEs in addition to nonsocial and
social selection. When relatedness is uniform across traits, this
winc= α + zT?βN+ (I − ?T?T)(I + r?T)−1
× (rI + ?T)(I − ?T?T)−1βS
In the absence of IGEs this simplifies to
winc= α + zT(βN+ rβS) + ε.
This multivariate formulation reveals that, even in the absence of
IGEs, social selection is central to the estimation of Hamilton’s
original conception of inclusive fitness.
The integration of selection and inheritance into a complete de-
scription of evolutionary change is usually simple and direct, as
epitomized by the well-known breeder’s equation (Lande 1979;
Falconer and MacKay 1996). Social interactions substantially
complicate this picture by generating covariances between geno-
types of individuals and the phenotypes and fitness of their inter-
actants. The results of our model demonstrate that to predict the
evolution of interacting phenotypes we must know (1) how the
genotypes of interacting individuals are related (whether through
pedigree relatedness, population subdivision, or trait-specific as-
sociations), (2) how the phenotypes of interacting individuals af-
acting individuals socially affect fitness. These factors are above
and beyond the additive genetic (co)variance and the strength of
nonsocial selection, which must be measured to predict the evo-
lution of any trait, social or otherwise (Lande 1979; Lande and
Arnold 1983; Roff 1997; Kingsolver et al. 2001).
The equations presented here emphasize that social selection
generates evolutionary change through a genetic pathway distinct
from that responding to nonsocial selection (eq. 4b). Evolution-
ary response to social selection results only when the breeding
value of a focal individual covaries with the phenotype of a social
partner or social group. This is consistent with Queller’s (1992b)
general formulation of kin selection; however, our results demon-
strate that the critical covariance might emerge via two biologi-
cally disparate paths—through relatedness or through phenotypic
modification arising from IGEs. Although relatedness in this con-
text is most likely to result from pedigree relationships or popula-
tion subdivision, it is also conceivable that it reflects trait-specific
associations among individuals. To encompass this multivariate
possibility, we introduced the potentially asymmetric matrix R
that describes trait-specific relatedness. This matrix allows situa-
such as in trait-group, greenbeard, and mate choice models to be
incorporated into a multivariate predictive equation (Hamilton
1964b; Wilson 1975; Dawkins 1976; Lande 1981; Wilson and
Dugatkin 1997; Pepper 2000; Grafen 2009; Gardner and West
Relatedness and IGEs have both independent and interactive
effects on the response to selection. Both social and nonsocial
selection is influenced by interacting phenotypes through the ?
terms that account for linear and feedback influences on pheno-
types (eqs. 10 and 12). These feedback effects are particularly
exacerbated by group size and lead to dramatic increases in the
rate of evolutionary response (Fig. 2). IGEs provide an additional
contribution to the response to nonsocial selection when they
occur among related individuals. Nevertheless, IGEs and related-
ness have surprisingly independent and parallel contributions to
the response to social selection. Our derivation of an extended
Hamilton’s rule emphasizes this symmetry, and suggests that re-
latedness and reciprocity should have nearly equal influences on
the evolution of altruism (cf. Fletcher and Zwick 2006). Further-
more, because IGEs generate a covariance between interacting
individuals via a genetic path, ? may be interpreted as a measure
sults that show how maternal effects should alter the predictions
of Hamilton’s rule to interactions occurring within a generation
(Cheverud 1984; see Wolf 2003 for a similar treatment of this
problem using variance components).
It is essential to recognize that although both phenomena
originate through interactions among conspecifics, IGEs and so-
cial selection represent distinct mechanisms by which interacting
phenotypes influence evolutionary change. These effects occur
via different pathways (Fig. 1): social selection represents a di-
rect relationship between the phenotype of social interactants and
fitness, whereas IGEs influence fitness indirectly by influencing
phenotypic expression. Both processes can create feedback that
accelerates the rate of evolution, but on different time scales.
IGEs create feedback by affecting phenotypic expression within
a generation, thereby inflating the response to selection (Moore
et al. 1997). Social selection creates feedback across generations
because interacting phenotypes act simultaneously as targets and
the next generation experiences a different social environment
than the previous one. This social environment provides the basis
2566 EVOLUTION SEPTEMBER 2010
INDIRECT GENETIC EFFECTS AND SOCIAL SELECTION
for selection, creating feedback in the evolution of the interact-
1983; Moore et al. 1997; Wolf et al. 1999).
This concept of social selection (Wolf et al. 1999) can be
considered as a subset of the broader contextual models of mul-
tilevel selection (Goodnight et al. 1992; Queller 1992a; Bijma
et al. 2007a; Bijma and Wade 2008). Hamilton’s (1963, 1964a,b)
inclusive fitness models have often been thought of as an al-
ternative to multilevel selection as an explanation for the evo-
lution of social behavior. Hamilton (1975) recognized that in-
clusive fitness models may be expressed as multilevel selection
models, and subsequent analyses have demonstrated the mathe-
matical equivalence between the two frameworks (Wade 1980;
Queller 1992a; Bijma and Wade 2008). As demonstrated by
our treatment of Hamilton’s rule, considering IGEs and social
selection provides additional insights into the inclusive-fitness
framework. In addition, our model is composed of parameters
that can be measured empirically to test hypotheses about social
Of the components identified here as critical to understanding
evolutionary response to social selection, substantial empirical
progress has been made only for relatedness or population sub-
structure. There are a number of well-established methods for
measuring relatedness (Queller and Goodnight 1989; Ritland
1996; Lynch and Ritland 1999; Van de Casteele et al. 2001; Weir
and Hill 2002; Csill´ ery et al. 2006; Weir et al. 2006; Konovalov
and Heg 2008), and a wealth of data exists regarding relatedness
and genetic structure in natural populations (Strassmann et al.
et al. 2005; Leinonen et al. 2008).
There has been limited empirical investigation of IGEs,
particularly nonmaternal IGEs (McGlothlin and Brodie 2009;
Bleakley et al. 2010). Several recent studies have quantified IGEs
as components of variance (e.g., Wolf 2003; Muir 2005; Petfield
et al. 2005; Linksvayer 2006; Bergsma et al. 2008; Brommer and
Rattiste 2008; Ellen et al. 2008; Danielson-Francois et al. 2009;
Wilson et al. 2009), as changes in gene expression (Wang et al.
2008), as response to different genotypes (Linksvayer 2007; Kent
et al. 2008; Linksvayer et al. 2009), or indirectly from selection
lines (Moore et al. 2002; Chenoweth et al. 2010). Very few have
is necessary to identify the paths of influence between specific
traits in interacting partners (Mutic and Wolf 2007; Bleakley and
Brodie 2009; Galloway et al. 2009). New methodology for esti-
mating ? (or its analog for maternal effects, M; Kirkpatrick and
Lande 1989) through components of quantitative genetic vari-
ance may address this paucity of data in the future (Galloway
et al. 2009; McGlothlin and Brodie 2009).
Social selection gradients can be measured in a straightfor-
ward way using multiple regression techniques similar to contex-
tual analysis (Heisler and Damuth 1987; Goodnight et al. 1992;
Okasha 2006). As a purely phenotypic process, estimation of so-
cial selection does not require knowledge of the covariance of
breeding values among interactants or other genetic information.
Simultaneous social and nonsocial selection can be measured us-
ing a simple extension of Lande and Arnold’s (1983) multiple
regression method to include the traits of social interactants or
groups in addition to those of the focal individual (Wolf et al.
1999).Dependingonthe populationstructure, investigators could
perform such analyses using equations (3), (15), or (16) as re-
gression models. Such an analysis differs subtly from traditional
the focal individual) rather than the mean of the social interac-
have used contextual analysis in plants (Stevens et al. 1995; Aspi
et al. 2003; Donohue 2003, 2004; Weinig et al. 2007) and ani-
mals (Tsuji 1995; Banschbach and Herbers 1996; McAdam and
Boutin 2003) but none have adopted the social selection approach
we describe here.
In addition to the relatively large sample sizes and reliable
estimates of individual fitness that are required for any study of
tion requires information on the phenotypes of social partners or
groups. However, large long-term selection datasets are becom-
ing increasingly common in evolutionary biology (Kruuk et al.
2001, 2002; Grant and Grant 2002). It is likely that many existing
datasets also contain information on social grouping that could
be used to measure social selection. We encourage empiricists
studying interacting phenotypes in natural populations to quan-
tify nonsocial and social selection so that the importance of each
may be determined.
ALTRUISM AND GREENBEARDS
To illustrate the applications of this theory, we present specific
worked examples derived from the multivariate equation pre-
sented in the text (eq. 12). We focus on situations in which nonso-
cial and social selection are in conflict, that is, where a trait is
beneficial to one individual in a pair but harmful to the other. We
social selection can lead to the coordinated evolution of interact-
First, suppose an altruistic behavior, z1, benefits a recipient
at the expense of the actor, so that it is under negative nonsocial
selection (βN1< 0) and positive social selection (βS1> 0). A
second trait, z2, is a visible morphological trait, such as a badge,
with no direct fitness consequences of its own (βN2= βS2= 0).
However, individuals base their level of altruistic behavior on the
morphological trait of their partner based on the coefficient ψ12.
EVOLUTION SEPTEMBER 2010
JOEL W. MCGLOTHLIN ET AL.
This morphological badge is determined by direct genetic and
environmental effects, and is static within an individual.
We first assume that these two traits are genetically uncor-
related (G12= 0) and that individuals assort at random (R = 0).
Using equation (12), the predicted evolutionary change in the two
traits can be expressed as
?¯ z1= G11βN1+ ψ122G22βS1
?¯ z2= ψ12G22βS1.
The most obvious result is that the two traits are evolutionarily
coupled, even in the absence of direct genetic covariance. This
occurs via an interaction between the IGE of the cue on the
The evolution of altruism is not influenced by the sign of the IGE,
and will evolve whenever
For altruism to evolve, ψ12, will have to be quite strong unless
the relative cost of performing the behavior (βN1) is quite low. In
contrast, the evolution of the badge is determined by the sign of
the IGE. The badge that elicits altruism will increase when
0 < ψ12G22βS1.
the elicitation of altruistic behavior is positive (ψ12> 0), then the
two traits will tend to runaway together. However, if the badge’s
effect is negative (ψ12< 0), the badge will evolve in the direction
opposite of the behavior it elicits.
Another result of the inequalities above is that the badge will
tend to evolve more quickly than the altruistic behavior it elicits.
This occurs because the cue is not subject to nonsocial selection
and because of the squared IGE coefficient in equation (34a).
We now allow individuals to assort nonrandomly based on
the morphological trait. This is similar to a greenbeard model
that we assume no genetic correlation between the behavior and
the badge. If individuals assort nonrandomly based solely on the
morphological trait, the matrix R is no longer zero. Rather, it
contains a single nonzero element, r22?, that measures the strength
of the relationship between the breeding values of the badge of
the two interacting individuals. The evolution of the badge and
altruism is now predicted by
?¯ z1=?G11+ r22?ψ122G22
?¯ z2= r22?ψ12G22βN1+ ψ12G22βS1.
Nonrandom assortment adds an effect of nonsocial selection on
and the badge increases when
appears in the denominator, strong positive assortment slows the
ism is unlikely without genetic covariance between the behavior
and the badge.
When a genetic correlation exists between the behavior and
the badge, the evolution of the two traits is predicted by
?¯ z1=?G11+ ψ12G12+ r22?ψ12G12+ r22?ψ122G22
?¯ z2= (G12+ r22?ψ12G22)βN1+ ψ12G22βS1.
The evolution of altruism is now predicted when
G11+ ψ12G12+ r22?ψ12G12+ r22?ψ122G22βS1, (42)
and the badge increases when
Here, the evolution of both traits is dominated by the strength
of the IGE (ψ12). In practice, as the genetic covariance between
traits becomes stronger, it will be difficult to distinguish a situa-
tion such as the one above from a true greenbeard, where altruism
and the visible signal are associated with the same gene (Gardner
and West 2010). In the case in which the two traits become in-
distinguishable, their evolution will be predicted by the extended
Hamilton’s rule (inequality 25).
AGRESSION AND WEAPONS
As a second example, consider a two-trait system in which z1
is aggressive behavior. When individuals interact, their level of
0). In fights, success is determined both by the aggression of the
focal individual (βN1 > 0) and that of its partner (βS1 < 0).
2568EVOLUTION SEPTEMBER 2010
INDIRECT GENETIC EFFECTS AND SOCIAL SELECTION
Success is also influenced by a morphological weapon (z2; βN2>
0, βS2< 0). We assume aggressive behavior is not adjusted based
on the partner’s weapon (ψ12= 0). (See Moore et al. 1997 eq. 23
for a similar example in the absence of social selection.) When
interactions occur at random (R = 0), the evolution of the two
traits is predicted by
(1 − ψ11)?1 − ψ112?(βN1+ ψ11βS1) +
(1 − ψ11)βN2
1 − ψ112(βN1+ ψ11βS1) + G22βN2.
As is the case in the absence of social selection, the feedback
effect of IGE on aggression magnifies the correlated evolution-
ary response of the weapon (Moore et al. 1997). However, the
IGE has an additional effect when social selection is present. As-
suming βS1< 0, a negative ψ11tends to lead to the evolution
of greater aggression via the social selection pathway, whereas
a positive ψ11tends to lead to the evolution of less aggression.
Biologically, this means that a tendency to act submissively to
aggressive individuals leads to lead to an evolutionary increase in
mean aggression, whereas a tendency to escalate fights leads to
an evolutionary decrease in aggression. The evolutionary trajec-
tory of weapons, of course, will depend on the sign of the genetic
covariance as well.
Again, nonrandom assortment complicates matters further.
One biologically plausible scenario is that individuals choose
fighting partners based on a linear response to the size of their
weapons. This would lead to a correlation between breeding val-
ues for weaponry, r22?, in which case the response to selection
would be predicted by
(1 − ψ11)?1 − ψ112?(βN1+ ψ11βS1)
1 − ψ11(βN2+ r22?βS2)
1 − ψ112(βN1+ ψ11βS1) + G22(βN2+ r22?βS2).
This is identical to the previous case, except now social selection
on weapon size contributes to evolutionary change. Its effect will
depend upon whether individuals assort positively (partners have
similar weapons) or negatively (partners have oppositely sized
The results presented here, together with related theoretical work
on the genetic ramifications of social interactions (Griffing 1967,
1969, 1976, 1981a; Moore et al. 1997; Wolf et al. 1999; Bijma
et al. 2007a; Bijma and Wade 2008), demonstrate the importance
of IGEs and social selection in the evolution of social behavior
been viewed as having a central role in evolution, and various au-
thors have considered it as both a facilitator and an inhibitor of
2004; Duckworth 2009). The interacting phenotype models pro-
vide a quantitative framework that demonstrates how this conjec-
ture can be true. IGEs and social selection provide mechanisms
by which behavior and other interacting phenotypes influence the
rate and direction of their own evolution. Both mechanisms can
greatly accelerate the rate of evolutionary change and, in some
cases, reverse the direction of evolution from that predicted in
their absence. Here, we have identified the critical parameters to
phenotypes in driving evolutionary change.
We thank B. Bleakley, P. Fields, V. Formica, J. Hunt, E. Liebgold, W.
Muir, and C. Wood for helpful discussions and comments. C. Goodnight
and two anonymous reviewers made suggestions that improved an earlier
version of the manuscript. This research was supported by NSF grant
DEB-0650078 to EDB and EU framework 6 grant GEBACO to AJM.
Agrawal, A. F., E. D. Brodie, III, and M. J. Wade. 2001. On indirect genetic
effects in structured populations. Am. Nat. 158:308–323.
Aspi, J., A. Jakalaniemi, J. Tuomi, and P. Siikamaki. 2003. Multilevel phe-
notypic selection on morphological characters in a metapopulation of
Silene tatarica. Evolution 57:509–517.
Banschbach, V. S., and J. M. Herbers. 1996. Complex colony structure in
social insects: II. Reproduction, queen-worker conflict, and levels of
selection. Evolution 50:298–307.
Bergsma, R., E. Kanis, E. F. Knol, and P. Bijma. 2008. The contribution of
social effects to heritable variation in finishing traits of domestic pigs
(Sus scrofa). Genetics 178:1559–1570.
Bijma, P. 2010. Fisher’s fundamental theorem of inclusive fitness and the
change in fitness due to natural selection when conspecifics interact. J.
Evol. Biol. 23:194–206.
Bijma, P., and M. J. Wade. 2008. The joint effects of kin, multilevel selection
and indirect genetic effects on response to genetic selection. J. Evol.
Bijma, P., W. M. Muir, E. D. Ellen, J. B. Wolf, and J. A. M. Van Arendonk.
2007b. Multilevel selection 2: estimating the genetic parameters deter-
mining inheritance and response to selection. Genetics 175:289–299.
Bleakley, B. H., and E. D. Brodie, III. 2009. Indirect genetic effects influ-
ence antipredator behavior in guppies: estimates of the coefficient of
interaction psi and the inheritance of reciprocity. Evolution 63:1796–
EVOLUTION SEPTEMBER 2010
JOEL W. MCGLOTHLIN ET AL.
Bleakley, B. H., J. B. Wolf, and A. J. Moore. 2010. Evolutionary quantitative
eds. Social behaviour: genes, ecology and evolution. Cambridge Univ.
mates in a wild bird population. Evolution 62:2326–2333.
Chenoweth, S. F., H. D. Rundle, and M. W. Blows. 2010. Experimental evi-
dence for the evolution of indirect genetic effects: changes in the inter-
action effect coefficient, psi (?), due to sexual selection. Evolution 20
January [Epub ahead of print].
illustrated by maternal performance in mice. Evolution 38:766–777.
Clutton-Brock, T. 2009. Cooperation between non-kin in animal societies.
Cole, C. T. 2003. Genetic variation in rare and common plants. Annu. Rev.
Ecol. Evol. S. 34:213–237.
Creel, S. 1990. How to measure inclusive fitness. Proc. R. Soc. Lond. B
Ecol. Evol. 16:178–183.
Crook, J. H. 1972. Sexual selection, dimorphism, and social organization.
Pp. 251–281 in B. Campbell, ed. Sexual selection and the descent of
man. Aldine, Chicago.
Csill´ ery, K., T. Johnson, D. Beraldi, T. Clutton-Brock, D. Coltman, B.
based relatedness estimators in natural populations of outbred verte-
brates. Genetics 173:2091–2101.
Danielson-Francois, A. M., Y. H. Zhou, and M. D. Greenfield. 2009. Indirect
genetic effects and the lek paradox: inter-genotypic competition may
strengthen genotype x environment interactions and conserve genetic
variance. Genetica 136:27–36.
Darwin, C. 1859. The origin of species. John Murray, London.
Dawkins, R. 1976. The selfish gene. Oxford Univ. Press, New York.
———. 1982. The extended phenotype. Oxford Univ. Press, Oxford.
Dicke, M., A. A. Agrawal, and J. Bruin. 2003. Plants talk, but are they deaf?
Trends Plant Sci. 8:403–405.
Donohue, K. 2003. The influence of neighbor relatedness on multilevel selec-
tion in the Great Lakes sea rocket. Am. Nat. 162:77–92.
———. 2004. Density-dependent multilevel selection in the Great Lakes sea
rocket. Ecology 85:180–191.
Duckworth, R. A. 2009. The role of behavior in evolution: a search for mech-
anism. Evol. Ecol. 23:513–531.
Ellen, E. D., W. M. Muir, F. Teuscher, and P. Bijma. 2007. Genetic improve-
ment of traits affected by interactions among individuals: sib selection
schemes. Genetics 176:489–499.
Ellen, E. D., J. Visscher, J. A. M. van Arendonk, and P. Bijma. 2008. Survival
of laying hens: genetic parameters for direct and associative effects in
three purebred layer lines. Poult. Sci. 87:233–239.
Falconer, D. S., and T. F. C. MacKay. 1996. Introduction to quantitative
genetics. Prentice Hall, Harlow, England.
Fisher, R.A. 1930.Thegenetical theory of naturalselection. ClarendonPress,
Fletcher, J. A., and M. Zwick. 2006. Unifying the theories of inclusive fitness
and reciprocal altruism. Am. Nat. 168:252–262.
Foster, K. 2010. Social behaviour in microorganisms. In T. Sz´ ekely, J.
Komdeur, and A. J. Moore, eds. Social behaviour: genes, ecology and
evolution. Cambridge Univ. Press, Cambridge.
Frank, S. A. 1997. The Price Equation, Fisher’s fundamental theorem, kin
selection, and causal analysis. Evolution 51:1712–1729.
———. 1998. Foundations of social evolution. Princeton Univ. Press,
genetics: concepts and case studies. Oxford Univ. Press, Oxford.
———. 2007. All of life is social. Curr. Biol. 17:R648–R650.
Galloway, L. F., J. R. Etterson, and J. W. McGlothlin. 2009. Contribution of
direct and maternal genetic effects to life-history evolution. New Phytol.
Gardner, A., and A. Grafen. 2009. Capturing the superorganism: a formal
theory of group adaptation. J. Evol. Biol. 22:659–671.
Gardner, A., and S. A. West. 2010. Greenbeards. Evolution 64:25–38.
Goodnight, C. J. 2005. Multilevel selection: the evolution of cooperation in
non-kin groups. Popul. Ecol. 47:3–12.
Goodnight, C. J., J. M. Schwartz, and L. Stevens. 1992. Contextual analysis
of models of group selection, soft selection, hard selection, and the
evolution of altruism. Am. Nat. 140:743–761.
Grafen, A. 1982. How not to measure inclusive fitness. Nature 298:425–426.
———. 1985. A geometric view of relatedness. Oxford Surv. Evol. Biol.
———. 2006. Optimization of inclusive fitness. J. Theor. Biol. 238:541–563.
———. 2009. Formalizing Darwinism and inclusive fitness theory. Philos.
Trans. R. Soc. Lond. B 364:3135–3141.
of Darwin’s finches. Science 296:707–711.
Griffing, B. 1967. Selection in reference to biological groups. I. Individual
and group selection applied to populations of unordered groups. Aust.
J. Biol. Sci. 20:127–139.
———. 1969. Selection in reference to biological groups. III. Generalized
results of individual and group selection in terms of parent-offspring
covariances. Aust. J. Biol. Sci. 21:1171–1178.
———. 1976. Selection in reference to biological groups. V. Analysis of
full-sib groups. Genetics 82:703–722.
———. 1981a. A theory of natural selection incorporating interaction among
individuals. I. The modeling process. J. Theor. Biol. 89:635–658.
———. 1981b. A theory of natural selection incorporating interaction among
individuals. II. Use of related groups. J. Theor. Biol. 89:659–677.
Hamilton, W. D. 1963. The evolution of altruistic behavior. Am. Nat. 97:354–
———. 1964a. The genetical evolution of social behavior. I. J. Theor. Biol.
———. 1964b. The genetical evolution of social behavior. II. J. Theor. Biol.
Rev. Ecol. Syst. 3:193–232.
———. 1975. Innate social aptitudes of man: an approach from evolution-
ary genetics. Pp. 133–153 in R. Fox, ed. ASA Studies 4: biological
anthropology. Malaby Press, London.
Heisler, I. L., and J. Damuth. 1987. A method for analyzing selection in
hierarchically structured populations. Am. Nat. 130:582–602.
Hughes, C. 1998. Integrating molecular techniques with field methods in
studies of social behavior: a revolution results. Ecology 79:383–399.
Jansen, V. A. A., and M. van Baalen. 2006. Altruism through beard chromo-
dynamics. Nature 440:663–666.
Kent, C., R. Azanchi, B. Smith, A. Formosa, and J. D. Levine. 2008. Social
context influences chemical communications in D. melanogaster males.
Curr. Biol. 18:1384–1389.
Kingsolver, J. G., H. E. Hoekstra, J. M. Hoekstra, D. Berrigan, S. N. Vignieri,
C. E. Hill, A. Hoang, P. Gilbert, and P. Beerli. 2001. The strength
of phenotypic selection in natural populations. Am. Nat. 157:245–
2570EVOLUTION SEPTEMBER 2010
INDIRECT GENETIC EFFECTS AND SOCIAL SELECTION
Kirkpatrick, M. 1982. Sexual selection and the evolution of female mate
choice. Evolution 36:1–12.
Kirkpatrick, M., and R. Lande. 1989. The evolution of maternal characters.
Konovalov, D. A., and D. Heg. 2008. A maximum-likelihood relatedness
estimator allowing for negative relatedness values. Mol. Ecol. Resour.
Kruuk, L. E. B. 2004. Estimating genetic parameters in natural populations
using the ‘animal model’. Philos. Trans. R. Soc. Lond. B 359:873–
Kruuk, L. E. B., J. Meril¨ a, and B. C. Sheldon. 2001. Phenotypic selection on
a heritable size trait revisited. Am. Nat. 158:557–571.
Clutton-Brock. 2002. Antler size in red deer: heritability and selection
but no evolution. Evolution 56:1683–1695.
Lande, R. 1979. Quantitative genetic analysis of multivariate evolution, ap-
plied to brain: body size allometry. Evolution 33:402–416.
———. 1981. Models of speciation by sexual selection on polygenic traits.
Proc. Natl. Acad. Sci. USA 78:3721–3725.
———. 2009. Adaptation to an extraordinary environment by evolution of
phenotypic plasticity and genetic assimilation. J. Evol. Biol. 22:1435–
characters. Evolution 37:1210–1226.
ies of quantitative trait and neutral marker divergence: a meta-analysis.
J. Evol. Biol. 21:1–17.
Li, C. C. 1967. Fundamental theorem of natural selection. Nature 214:505–
Linksvayer, T. A. 2006. Direct, maternal, and sibsocial genetic effects on
individual and colony traits in an ant. Evolution 60:2552–2561.
———. 2007. Ant species differences determined by epistasis between brood
and worker genomes. PLoS ONE 2:e994.
Linksvayer, T. A., O. Rueppell, A. Siegel, O. Kaftanoglu, R. E. Page, Jr., and
G. V. Amdam. 2009. The genetic basis of transgressive ovary size in
honeybee workers. Genetics 183:693–707, 691SI–613SI.
Lucas, J. R., S. R. Creel, and P. M. Waser. 1996. How to measure inclusive
fitness, revisited. Anim. Behav. 51:225–228.
Lukas, D., V. Reynolds,C. Boesch, and L. Vigilant. 2005.To what extentdoes
living in a group mean living with kin? Mol. Ecol. 14:2181–2196.
Lynch, M., and K. Ritland. 1999. Estimation of pairwise relatedness with
molecular markers. Genetics 152:1753–1766.
Mayr, E. 1963. Animal species and evolution. Belknap Press, Cambridge.
McAdam, A. G., and S. Boutin. 2003. Variation in viability selection among
cohorts of juvenile red squirrels (Tamiasciurus hudsonicus). Evolution
McGlothlin, J. W., and E. D. Brodie, III. 2009. How to measure indirect
genetic effects: the congruence of trait-based and variance-partitioning
approaches. Evolution 63:1785–1795.
Meril¨ a, J., and P. Crnokrak. 2001. Comparison of genetic differentiation at
marker loci and quantitative traits. J. Evol. Biol. 14:892–903.
Michod, R. E., and W. D. Hamilton. 1980. Coefficients of relatedness in
sociobiology. Nature 288:694–697.
Moore, A. J., E. D. Brodie, III, and J. B. Wolf. 1997. Interacting phenotypes
and the evolutionary process: I. Direct and indirect genetic effects of
social interactions. Evolution 51:1352–1362.
of interacting phenotypes: genetics and evolution of social dominance.
Am. Nat. 160:S186–S197.
Mousseau, T. A., and C. W. Fox, eds. 1998. Maternal effects as adaptations.
Oxford Univ. Press, New York.
Muir, W. M. 1996. Group selection for adaptation to multiple-hen cages:
selection program and direct responses. Poult. Sci. 75:447–458.
———. 2005. Incorporation of competitive effects in forest tree or animal
breeding programs. Genetics 170:1247–1259.
Mutic, J. J., and J. B. Wolf. 2007. Indirect genetic effects from ecological
interactions in Arabidopsis thaliana. Mol. Ecol. 16:2371–2381.
Nunney, L. 1985. Group selection, altruism, and structured-deme,odels. Am.
Odling-Smee, F. J., K. N. Laland, and M. W. Feldman. 2003. Niche construc-
Okasha, S. 2006. Evolution and the levels of selection. Oxford Univ. Press,
Pepper, J. W. 2000. Relatedness in trait group models of social evolution. J.
Theor. Biol. 206:355–368.
Petfield, D., S. F. Chenoweth, H. D. Rundle, and M. W. Blows. 2005. Genetic
variance in female condition predicts indirect genetic variance in male
sexual display traits. Proc. Natl. Acad. Sci. USA 102:6045–6050.
Pigliucci, M. 2001. Phenotypic plasticity: beyond nature and nurture. Johns
Hopkins Univ. Press, Baltimore, MD.
Price, G. R. 1970. Selection and covariance. Nature 227:520–521.
———. 1972. Extension of covariance selection mathematics. Ann. Hum.
Queller, D. C. 1985. Kinship, reciprocity and synergism in the evolution of
social behaviour. Nature 318:366–367.
———. 1992a. Quantitative genetics, inclusive fitness, and group selection.
Am. Nat. 139:540–558.
———. 1992b. A general model for kin selection. Evolution 46:376–380.
———. 1996. The measurement and meaning of inclusive fitness. Anim.
Queller, D. C., and K. F. Goodnight. 1989. Estimating relatedness using
genetic markers. Evolution 43:258–275.
Rice, S. H. 2004. Evolutionary theory: mathematical and conceptual founda-
tions. Sinauer Associates, Sunderland, MA.
Ritland, K. 1996. Estimators for pairwise relatedness and individual inbreed-
ing coefficients. Genet. Res. 67:175–185.
Anim Prod 8:95–108.
Roe, A., and G. G. Simpson, eds. 1958. Behavior and evolution. Yale Univ.
Press, New Haven.
Roff, D. A. 1997. Evolutionary quantitative genetics. Chapman & Hall, New
Rousset, F. 2002. Inbreeding and relatedness coefficients: what do they mea-
sure? Heredity 88:371–380.
Stevens, L., C. J. Goodnight, and S. Kalisz. 1995. Multilevel selection
in natural populations of Impatiens capensis. Am. Nat. 145:513–
Strassmann, J. E., C. R. Hughes, D. C. Queller, S. Turillazzi, R. Cervo, S.
K. Davis, and K. F. Goodnight. 1989. Genetic relatedness in primitively
eusocial wasps. Nature 342:268–270.
Sz´ ekely, T., J. Komdeur, and A. J. Moore, eds. 2010. Social behaviour: genes,
ecology and evolution. Cambridge Univ. Press, Cambridge.
Taylor, P. D., and S. A. Frank. 1996. How to make a kin selection model. J.
Theor. Biol. 180:27–37.
how shall we model kin selection? J. Evol. Biol. 20:301–309.
Trivers, R. L. 1971. Evolution of reciprocal altruism. Q. Rev. Biol. 46:35–
Tsuji, K. 1995. Reproductive conflicts and levels of selection in the ant Pris-
tomyrmex pungens: contextual analysis and partitioning of covariance.
Am. Nat. 146:586–607.
EVOLUTION SEPTEMBER 2010
JOEL W. MCGLOTHLIN ET AL.
Uyenoyama, M. K. 1984. Inbreeding and the evolution of altruism under kin
aelection: effects on relatedness and group structure. Evolution 38:778–
Uyenoyama, M. K., and M. W. Feldman. 1981. On relatedness and adaptive
topography in kin selection. Theor. Popul. Biol. 19:87–123.
Van de Casteele, T., P. Galbusera, and E. Matthysen. 2001. A compari-
son of microsatellite-based pairwise relatedness estimators. Mol. Ecol.
Wade, M. J. 1980. Kin selection: its components. Science 210:665–667.
Am. Nat. 125:61–73.
Wang, J., K. G. Ross, and L. Keller. 2008. Genome-wide expression patterns
and the genetic architecture of a fundamental social trait. PLoS Genet.
Weinig, C., J. A. Johnston, C. G. Willis, and J. N. Maloof. 2007. Antagonistic
multilevel selection on size and architecture in variable density settings.
Weir, B. S., and W. G. Hill. 2002. Estimating F-statistics. Ann. Rev. Genet.
Weir, B. S., A. D. Anderson, and A. B. Hepler. 2006. Genetic relatedness
analysis: modern data and new challenges. Nat. Rev. Genet. 7:771–
West, S. A., A. S. Griffin, and A. Gardner. 2007. Social semantics: altruism,
cooperation, mutualism, strong reciprocity and group selection. J. Evol.
West-Eberhard, M. J. 1979. Sexual selection, social competition, and evolu-
tion. Proc. Am. Phil. Soc. 123:222–234.
———. 1983. Sexual selection, social competition, and speciation. Q. Rev.
———. 2003. Developmental plasticity and evolution. Oxford Univ. Press,
Westneat, D. F., and C. G. Fox, eds. 2010. Evolutionary behavioral ecology.
Oxford Univ. Press, New York.
Wilson, D. S. 1975. A theory of group selection. Proc. Natl. Acad. Sci. USA
Wilson, D. S., and L. A. Dugatkin. 1997. Group selection and assortative
interactions. Am. Nat. 149:336–351.
Wilson, D. S., and E. O. Wilson. 2007. Rethinking the theoretical foundation
of sociobiology. Q. Rev. Biol. 82:327–348.
Wilson, A. J., U. Gelin, M.-C. Perron, and D. R´ eale. 2009. Indirect genetic
effects and the evolution of aggression in a vertebrate system. Proc. R.
Soc. Lond. B 276:533–541.
Wolf, J. B. 2003. Genetic architecture and evolutionary constraint when the
environment contains genes. Proc. Natl. Acad. Sci. USA 100:4655–
Westneat and C. W. Fox, eds. Evolutionary behavioral ecology. Oxford
Univ. Press, New York.
Wolf, J. B., E. D. Brodie, III, J. M. Cheverud, A. J. Moore, and M. J. Wade.
1998. Evolutionary consequences of indirect genetic effects. Trends
Ecol. Evol. 13:64–69.
the evolutionary process. II. Selection resulting from social interactions.
Am. Nat. 153:254–266.
Wolf, J. B., E. D. Brodie, III, and M. J. Wade, eds. 2000. Epistasis and the
evolutionary process. Oxford Univ. Press, New York.
special regard to systems of mating. Evolution 19:395–420.
Wynne-Edwards, V. C. 1962. Animal dispersion in relation to social behavior.
Oliver and Boyd, Edinburgh.
Associate Editor: C. Goodnight
EQUIVALENCE OF VARIANCE-COMPONENTS MODELS
The framework of Griffing (1967, 1969, 1976, 1981a) and Bijma
and colleagues (Bijma et al. 2007a,b; Bijma and Wade 2008) ex-
presses IGEs as components of phenotypic variance rather than
strengths of IGEs (McGlothlin and Brodie 2009). Here, we derive
a multivariate version of the equation for evolutionary response
derived by Bijma and Wade (2008) and show its equivalence to
the trait-based equation.
In the variance-components framework, the total breeding
value is equal to the sum of the direct and social (i.e., indirect
genetic) breeding values. Expressed multivariately,
A = aD+ aS,
where aDand aSare column vectors of direct and social breeding
additive genetic value when IGEs display feedback (McGlothlin
and Brodie 2009). The latter can be thought of as a genetic value
for “social performance,” encompassing the social effects of all
individuals can be expressed as
z = aD+ eD+ a?
D+ aS+ eS,
where eDand eSterms are environmental (or residual effects). As
individual. Note that the focal individual is affected by the social
breeding value and environmental effect of its social partner, and
Substituting these definitions into equation (4a), and making
the standard quantitative genetic assumption of zero covariance
between breeding values and residuals, we find
?¯ z = cov?(aD+ aS),?aT
+ cov?(aD+ aS),?a?T
or in compact form,
?¯ z = (GD+ GDS? + GSD+ GSS?)βN
+ (GDD? + GDS+ GSD? + GS)βS,
2572EVOLUTION SEPTEMBER 2010