ARTICLE IN PRESS

Journal of Theoretical Biology 243 (2006) 386–392

www.elsevier.com/locate/yjtbi

Assortative mating through a mechanism of sexual selection

Laureano Castro, Miguel A. Toro!
Departamento de Mejora Genética Animal, Instituto de Investigacion y Tecnologia Agraria y Alimentaria, SGIT-INIA,
Carretera de La Coruña km. 7, 28040 Madrid, Spain
Received 6 February 2006; received in revised form 7 July 2006; accepted 7 July 2006
Available online 15 July 2006

Abstract

We propose that assortative mating can arise through a mechanism of sexual selection by active female choice of partners based on a
‘self-seeking like’ decision rule. Using a mathematical model, we show that a gene can be selected that make females to choose mates that
are similar to themselves with respect to an arbitrary tag, even if two independent and unlinked genes determine the preference and the
tag. The necessary requisite for this process to apply is an asymmetry between partners, such that the female can choose the male but this
one must always accept to mate. The fitness advantage is due to linkage disequilibrium built up between both genes. Simulations have
been run to check the algebraic results and to analyse the influence of several factors on the evolution of the system. Any factor that
favors linkage disequilibrium also favors the evolution of the preference allele. Moreover, in a large population subdivided in small
subpopulations connected by migration, the assortative mating homogenizes the population genotypic structure for the tags in contrast
to random mating that maintains most of the variation.
r 2006 Elsevier Ltd. All rights reserved.

Keywords: Sexual selection; Mate choice; Assortative mating; Phenotype matching

1. Introduction selection theory proposes Fisherian self-reinforcing selection,

Darwin’s concept of sexual selection is typically defined
as the tendency of members of one sex, usually females, to
preferentially choose as mates certain members of the
opposite sex, and provides one of the main explanations for
the evolution of non-adaptive traits such as the peacock
tail. Females could prefer to mate with males carrying a
population-specific trait, a preference model, or to mate
with males that share their own trait phenotype, an
assortative mating model (Servedio, 2000).
Assortative mating could occur if females asses their own
phenotype and use it as a basis for mate choice, a type of self-
referent phenotype matching (Verzijden et al., 2005), or if they
asses and use the phenotype of known relatives—i.e. through
sexual imprinting (Lorenz, 1935; ten Cate and Vos, 1999; Aoki
et al., 2001). It is not clear how common self-referent
phenotype matching is in nature. A review of the evidence
for this mechanism suggests that it may be more frequent than
previously thought (Hauber and Sherman, 2001). Sexual
species recognition, genetic indicator mechanisms, and direct
benefits as possible mechanisms driving the evolution of non-
random mating (Andersson, 1994).
In this paper we show that assortative mating can arise
through a mechanism of sexual selection by an active
female choice of partners based on a ‘self seeking like’
decision rule—i.e. a type of self-referent phenotype
matching. First, using a simple mathematical model, we
show that a gene can be selected that makes females to
choose as mates those males that are similar to themselves
with respect to an arbitrary tag even if two independent
and unlinked genes determine the preference and the tag.
Second, simulations have been carried out along a number
of generations to check the algebraic results and to analyse
the influence of several factors: multiple tags, intensity of
preference, and population size and structure.
2. Theoretical model

!Corresponding author. Tel. +34 913476807; fax: +34 913572293. The population consists of sexual haploid individuals
E-mail address: toro@inia.es (M.A. Toro). AB, Ab, aB and ab at frequencies r, s, t and u, respectively,

0022-5193/$ - see front matter r 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jtbi.2006.07.003
 ARTICLE IN PRESS
L. Castro, M.A. Toro / Journal of Theoretical Biology 243 (2006) 386–392 387

for the four genotypes at the two loci considered. The gene
frequencies of allele A in the first locus and of allele B in the
second locus are p1 ¼ r þ s and p2 ¼ r þ t. Let us assume
that there is an asymmetry in the role played by males and
females in mating such that the female is in position to
choose a mate whereas the male can only accept to mate
(otherwise he will be at an obvious disadvantage). Females
carrying the a allele mate randomly, but those carrying the
A allele mate randomly with probability 1#m and, if
partners are available, mate with probability m with a male
of the population which is genetically similar to her but at a
second locus (B,b), that acts as tag. Thus, AB females tend
to be assorted with AB and aB males, whereas females Ab
tend to be assorted with Ab and ab males.
The frequencies for the different types of matings are
shown in Table 1. Recombination between gametes follows
and the next generation of zygotes is formed. It is a
question of simple algebra to show that the change in gene
frequencies will be
mD2
Dp1 ¼ , (1)
2p2 ð1 # p2 Þ
mD
Dp2 ¼ , (2)
2 populations and to test the conclusions of the algebraic
and
! "
mD Dð1 # 2p2 Þ # 0:5D2
DD ¼ # 2c þ mc , (3)
2 p2 ð1 # p2 Þ
where D is the linkage disequilibrium between both loci
and c the rate of recombination.
Several consequences from the above formulae must be
emphasized. The system is at equilibrium when the linkage
disequilibrium is null (D ¼ 0) irrespective of the values of
the gene frequencies of A(p1) and B(p2), because if D ¼ 0
both loci behave neutrally, as intuitively expected. There-
fore, if the system is perturbed but maintaining D equal to
0 the new frequencies of A and B will be again at
equilibrium. However, if the perturbation affects linkage
disequilibrium (D6¼0) the gene frequencies of A(p1) y B(p2)
will also be affected. Thus, starting with some arbitrary
values of p1, p2 and D, the system evolves dissipating the
linkage disequilibrium until a value of D ¼ 0 is reached,
and at this moment an equilibrium point is attained. At the
same time that D goes to 0, the behavior of the two loci is
different. In the first locus (A,a), the increase of A allele
depends on the square value of the linkage disequilibrium
parameter D (Eq. (1)). Consequently, the A allele will
always increase in frequency if D6¼0, irrespective of the sign
of D. The increase will be greater as m increases. In the
second locus (B,b), however, the change in gene frequency
depends directly on the sign and magnitude of the linkage
disequilibrium parameter D (Eq. (2)) and, therefore, the
frequency of the allele B can increase (D40) or decrease
(Do0). The recombination fraction c between the two loci
does not affect directly the change in gene frequencies, but
it has influence on D.
It is well known that genetic drift arising in finite
populations is one of the main sources of linkage
disequilibrium even between unlinked genes. Eqs. (1)–(3)
cannot give the evolution of gene frequencies in finite
populations because the effect of drift has not been
formally included. Therefore, we should rely in simulations
that will allow to study the evolution of the system in finite
analysis. The main conclusion is that the linkage disequili-
brium generated by drift in each generation would be
enough to allow that the A allele increases on average in
frequency until fixation as long as the variability in both
locus A/a and B/b are not lost by drift. Any factor that
favors the generation of disequilibrium D (e.g. a greater
number of tag loci) or that contributes to the maintenance
of genetic variability for the tags (e.g. intermediate
frequencies of the tag alleles or a greater number of tags)
could facilitate the increase of the A allele. On the other
hand, population size has opposite effects on the evolution
of A: for example, small populations facilitate the
generation of linkage disequilibrium but also the loss of
genetic variability. Finally, the algebraic analysis also
predicts that random fluctuations of the D value around

Table 1
Frequencies of mating

Female Male
P
AB Ab aB ab

AB mr2 rs(1#m) mrt ru(1#m) r

r2 ð1 # mÞ þ rtð1 # mÞ þ
rþt rþt

Ab sr(1#m) ms2 st(1#m) msu s

s2 ð1 # mÞ þ suð1 # mÞ þ
sþu sþu

aB tr ts T2 tu t
aB ur us ut u2 u
P mrt
mr2 s#ms(r+s) t # mtðr þ sÞ þ u#mu(r+s)
r # mrðr þ sÞ þ rþt
rþt
 ARTICLE IN PRESS
388 L. Castro, M.A. Toro / Journal of Theoretical Biology 243 (2006) 386–392
0 will not modify on average frequencies in the second
locus B/b.
Notice the difference between our model and the
classical models of sexual selection where A females
preferentially choose males with a fixed variant of the
tag, for example, B males (Kirkpatrick, 1982). If the A
females prefer B males, the equations of change in gene
frequency will be for m ¼ 1 and assuming the assumptions
of our model (i.e., that the B alleles are neutral and that the
female can choose but the male must always accept to
mate)
Dp1 ¼ Dp1 =p2 ,
and
Dp2 ¼ p1 ð1 # p2 Þ=2.
Obviously, the B allele is always favored even if D ¼ 0.
This increase in frequency of B induces positive linkage
disequilibrium that, in turns, favors the increase of the A
allele frequency until B attains fixation. However, classic
models of sexual selection (see Lande, 1981; Kirkpatrick,
1982) also assume that the B males have viability (1#s)
times that of b males and that this locus B/b is not
expressed in females. Both authors find that under a wide
range of parameter values, their models exhibit continuous
equilibrium lines. In our model, the tags are neutral and the
equilibrium is only attained when D is 0. As a consequence,
the A allele always increases in frequency in finite
populations.
It could be argued that in relatively large populations
having a preference for a mate entails some cost for
females, in either evaluating tags, or seeking out similar
individuals. Models of Fisher’s runaway process have
shown that when such costs exist, the reinforcing evolution
of trait and preference cannot usually occur (Kirkpatrick,
1985; Pomiankowski et al., 1991; Day, 2000), although the
outcomes of models depend on the details of several
assumptions (Mead and Anold, 2004). This objection can
be mitigated if, in order to avoid costs, we make use of
Seger’s (1985) model of behavioral preference (‘the best-of-
n males rule of female mating preferences’) where the
female chooses the male that is more similar to her but only
among a subset of n available males. In fact, for m ¼ 1, the
behavioral preference of our model is the Seger’s one when
n includes all males in the population. We have run some
simulations to study the effect of decreasing the number of
males evaluated.
If we introduce a cost e for the A females associated to
the activity of choosing, this cost can annul the advantage
of the A allele derived from Eq. (1). Thus, whereas the A
males have always an advantage over the a males if D6¼0,
because they are chosen more frequently than a males as
partners, it could be overcompensated by the disadvantage
of the A females due to the cost of choosing and, as a
consequence of this the A allele would be selected against.
Finally, it would seem that the A allele will increase in
frequency if A females show a tendency (measured by m) to
choose males that are genetically different to them for the
second locus (B,b). Thus, AB females tend to be assorted
with Ab and ab males, whereas Ab females tend to be
assorted with AB and aB males. In this case, however, the
allele A will only increase in frequency if
mD2 mDp1 ð1 # 2p2 Þ
Dp1 ¼ # þ . (6)
2p2 ð1 # p2 Þ 2p2 ð1 # p2 Þ
The first component will always be negative whereas the
sign of the second depends on the values of D and (1#2p2).
Therefore, the frequency of A can either increase or
decrease.
(4)
3. Simulation model
(5)
A two-locus (A,a; B,b) sexual haploid population was
studied. The a female mates randomly but the A female
prefers to mate with a male that is similar to her for the
neutral tag (B, b). We also consider the situation of
multiple tags where each tag has two selectively neutral
alternative variants (i.e. for i ¼ 5 there are B/b, C/c, D/d,
E/e, F/f). Thus, ABCDEF females tend to mate with
ABCDEF and aBCDEF males, whereas Abcdef female
tends to assort with Abcdef and abcdef males. When no
individuals matching for all tags are present, the most
similar ones are chosen—i.e., each female choose at
random a male among all that share with her the maximum
number of tags. We assume that all females mate only once
but the males can be chosen as mates several times. The
number of males equals the number of females and remains
constant along generations. From each mating, one son
and one daughter are obtained to form the next generation.
Brother and sister matings are avoided except in the case
expressly indicated.
First, we study the evolution of the A (and B) allele in
single populations of different sizes (50, 500 or 5000
individuals). The base population is generated at random
assuming an initial frequency of 0.1 for the A allele and 0.5
for the B allele. This last value implies maximum variability
for the tags. The average over runs of the initial linkage
disequilibrium is 0. Moreover, the assumed rate of
recombination is always 0.5. We also consider a large
population of 5000 individuals subdivided in 100 sub-
populations of 50 individuals each, half of them of each
sex. In all generations and before matings occur each
population interchanges k individuals with other popula-
tions. Matings take place within populations that remain of
the same size along generations. Selection between groups
is avoided because processes of extinction and re-coloniza-
tion are not considered. Each situation shows the evolution
of the average frequency of 10,000, 1000 or 10 runs in
single populations of sizes 50, 500 or 5000, respectively.
When we studied large subdivided population, the number
of runs was 100. Besides population structure (single or
subdivided) and population size, other parameters are
modified in the simulations: the number of tags i, the
assortment parameter m, the number of migrants k, the
 ARTICLE IN PRESS
L. Castro, M.A. Toro / Journal of Theoretical Biology 243 (2006) 386–392 389

Table 2
Evolution of A allele frequency along generations in single populations for the different simulated situations

Population size N Assortment m Tag number i Generation

0 25 100 500

50 1 1 0.100 0.109 0.122 0.12070.003

50a 1 1 0.100 0.132 0.166 0.16970.004
50 1 5 0.100 0.203 0.265 0.26670.004
500 1 1 0.100 0.100 0.111 0.11270.008
500 1 5 0.100 0.119 0.232 0.53070.015
5000 1 5 0.100 0.104 0.104 0.23170.049
50 0 1 0.100 0.100 0.100 0.10070.003
50 0.5 1 0.100 0.103 0.106 0.10670.003
50 0.5 5 0.100 0.118 0.130 0.12870.003
50 Negative –1 1 0.100 0.106 0.118 0.11870.003

The initial average frequency (0.5) of B tags remains constant along all generations. Common situation: no brother–sister matings.
a
With brother–sister matings.

number of evaluated males n, and exceptionally the system Table 3

of mating. Evolution of A allele frequency along generations in a large subdivided
population (100 groups of 50 individuals) with k migrants
4. Simulation results Assortment Situations Generations
m
Tag k 0 25 100
The main question is if the A allele can increase in
number migrants
frequency under different scenarios. Table 2 shows the i
evolution of the frequency of the A allele along 500
generations in single populations. The values have been 1 5 1 0.100 0.216 0.89770.003
1a 5a 1a 0.100a 0.653a 1.00070.000a
averaged over runs. In all case the average frequency of the 1 5 0.5 0.100 0.213 0.68070.006
B allele remains constant (p2 ¼ 0.5) along generations 1 5 2 0.100 0.209 0.94470.004
although, due to drift, in some populations either B or b 1 5 50 0.100 0.102 0.11670.005
become fixed. In small populations (50 individuals) and 1 1 1 0.100 0.112 0.15470.005
1 10 1 0.100 0.364 0.99970.000
only one tag, the initial frequency of the A allele (p1 ¼ 0.1) 0.5 5 1 0.100 0.117 0.20570.006
increases significantly (p1 ¼ 0.120). This increase is small Negative -1 5 1 0.100 0.109 0.17070.006
because the probability of loss of genetic variability by
The initial average frequency (0.5) of B tags remains constant along all
genetic drift is high and fixation of A or a is attained in
generations. Common situation: no brother–sister matings.
each population quickly before 100 generations. Thus, the a
With brother–sister matings.
frequency of the A allele is practically the same after 100 or
500 generations. If mating between brother and sister is not
avoided, the response for the A allele is greater frequency of the A allele is slight but significant. This
(p1 ¼ 0.169), even though only 4.4% of mating are result is due to a positive correlation between D and
incestuous. (1#2p2) with negative assortment and p2E0.5 and, there-
When linkage disequilibrium increases (five tags), the fore, the product of these two factors is positive in Eq. (6).
increase in frequency of the A allele is substantial Table 3 shows the evolution of the A allele in a large
(p1 ¼ 0.266). As population size increases, the genetic population (N ¼ 5000) subdivided into 100 groups of 50
variability stays longer but the magnitude of linkage individuals with different rates of migration k. In the initial
disequilibrium decreases. For N ¼ 500 individuals and 1 situation (i ¼ 5 tags and k ¼ 1 migrant), the results show a
tag, the increase in frequency is small even if the process striking increase in frequency of A (p1 ¼ 0.897). The reason
continues up to generation 500 (p1 ¼ 0.112). If the number is that migration allows the preservation of the genetic
of tags is five, 100 generations are enough to obtain a variability within populations and, for 5 tags, the
significant increase (p1 ¼ 0.232). On the contrary, for disequilibrium is high. The allowance of mating between
N ¼ 5000, 5 tags are not enough to produce an increase sibs facilitate still more the increase (p1 ¼ 1). For k ¼ 0.5,
of A in 100 generations but it occurs in 500 generations the increase is lower (p1 ¼ 0.680) because there is less
(p1 ¼ 0.231). genetic variability whereas for k ¼ 2 it is greater
The preference parameter also plays an important role. (p1 ¼ 0.944). However, there is an optimum value for the
For m ¼ 0, the A allele does not change in frequency and, number of migrants as, if k is very large, subdivision tends
with m ¼ 0.5, five tags are required to detect an increase in to dissipate linkage disequilibrium. In fact, when migration
frequency. For negative assortment, the increase in is complete (k ¼ 50), the situation reduces to a single large
 ARTICLE IN PRESS
390 L. Castro, M.A. Toro / Journal of Theoretical Biology 243 (2006) 386–392

population and the A allele increases very slightly in random mating (the females does not choose). The increase
frequency after 100 generations, but there is a significant in frequency is greater when n is equal to the number of
increase after 500 (p1 ¼ 0.156) or 1000 generations males in the populations but it also occurs, although more
(p1 ¼ 0.181). Notice that the last case is not equivalent to slowly, when n is smaller.
that of a single population of 5000 individuals, because in In some simulations we have assumed that, as a
the subdivided populations each female can only choose consequence of the choosing activity, the A females have
among the 25 males of her subpopulation. a probability e of remaining unmated. The results show
Table 3 also shows how the evolution of the A allele that, as expected, if e is great enough the A allele loses its
changes when the values of some parameters are modified. advantage and decreases in frequency. For example, in a
The results confirm those obtained in the isolated popula- large subdivided population (100 groups of 50 individuals)
tions. The increase in frequency of the A allele is greater as with 5 tags, m ¼ 1, and 1 migrant, if the probability
the number of tags increases (i ¼ 10 vs. i ¼ 1) but it is e ¼ 0.05, the frequency of the A allele increases from 0.1 to
retarded if the preference is incomplete (m ¼ 0.5). Negative 0.39 after 100 generations, but if the probability e ¼ 0.1,
assortment will also produce a significant but less then the frequency of A allele decreases to 0.068.
important increase. The simulation results indicate that the initial average
Table 4 shows the evolution of the A allele when the frequency of the B allele (p2 ¼ 0.5) remains equal along
female chooses the more similar male to her but only generations in all cases. However, the genotypic variability
among a subset of n males. The A allele does not increase in for the tags changes during the process. The change has
frequency if n ¼ 1, because this case corresponds to been studied comparing situations where either the A
(p1 ¼ 1) or a (p1 ¼ 0) allele is fixed. We consider a large
Table 4 population (N ¼ 5000) subdivided in 100 small subpopula-
Evolution of A allele frequency along generations in single populations for tions (N ¼ 50) connected by migration, because in this
different number of males assessed by females situation the probability that the A allele become fixed is
Population The best of Generation high. Table 5 shows the evolution along generations of the
size N n males average number of different genotypes for the 5 tags. In
0 25 100 500 absence of allele A (p1 ¼ 0 or what is equivalent m ¼ 0), the
50 1 0.100 0.098 0.096 0.09570.003
migration is able to maintain not only almost all genotypes
50 5 0.100 0.131 0.146 0.14770.004 (31.4–32) in the whole population after 1000 generations,
50 10 0.100 0.154 0.178 0.17870.004 but also a high number of genotypes within subpopula-
50 15 0.100 0.173 0.206 0.20870.004 tions—e.g., 14.4 genotypes for k ¼ 1 and 18.0 genotypes
50 25 0.100 0.203 0.265 0.26670.004 for k ¼ 10.
500 1 0.100 0.102 0.103 0.09970.021
500 25 0.100 0.096 0.149 0.20470.034
When the A allele is fixed and the preference parameter is
500 50 0.100 0.117 0.192 0.28770.040 maximum (p1 ¼ 1 and m ¼ 1), there is a quick loss of
500 100 0.100 0.123 0.210 0.36670.043 genotypic variability within subpopulations and in the
500 250 0.100 0.122 0.234 0.51170.049 population as a whole. Females choose males of the same
The initial average frequency (0.5) of B tags remains constant along all
genotype to mate with them and, consequently, there are
generations. Common situation: no brother–sister matings; assortment males that do not obtain mates in each subpopulation and,
m ¼ 1; tag number i ¼ 5. therefore, the variability decreases in the short term.

Table 5
Evolution of average number of genotypes within subpopulations and on the whole population for different migration rates k

Situations Number of different genotypes

Generation 0 Generation 100 Generation 500 Generation 1000

Subpopulat Populat. Subpopulat. Populat. Subpopulat. Populat. Subpopulat. Populat.

k ¼ 0.5 p1 ¼ 1 m ¼ 1 25.5 32.0 6.3 32.0 4.0 30.1 1.6 6.6

k ¼ 0.5 p1 ¼ 1 ¼ 0.9 6.6 2.0 13.3 1.2 1.5
k ¼ 0.5 p1 ¼ 0 m ¼ 0 13.1 11.9 32.0 11.0 32.0

k ¼ 1 p1 ¼ 1 m ¼ 1 25.5 32.0 7.6 32.0 1.7 2.3 1.5 1.7

k ¼ 1 p1 ¼ 1 m ¼ 0.9 8.71 1.4 1.5 1.3 1.3
k ¼ 1 p1 ¼ 0 m ¼ 0 17.7 16.2 32.0 14.4 31.7
k ¼ 10 p1 ¼ 1 m ¼ 1 25.5 32.0 3.3 5.9 2.3 2.8 2.2 2.5
k ¼ 10 p1 ¼ 1 m ¼ 0.9 4.4 9.4 1.7 1.7 1.63 1.63
k ¼ 10 p1 ¼ 1 m ¼ 0 23.7 32.0 21.0 32.0 18.0 31.4

The initial average frequency (0.5) of B tags remains constant along all generations. Common situation: no brother–sister matings, number of tags i ¼ 5.
 ARTICLE IN PRESS
L. Castro, M.A. Toro / Journal of Theoretical Biology 243 (2006) 386–392
Besides, migration impedes the formation of stable groups
of similar males and females mating between themselves.
When an individual migrates to other subpopulation, this
can destabilize the group of origin and, if he is a male,
probably cannot find mates and his genotype disappears.
In the long term, each subpopulation tends to be fixed for
one or two genotypes. Moreover, if the rate of migration is
large enough (e.g., kX1), the whole population is homo-
genized to about two or three genotypes. The result is
dramatic: the number of different genotypes in the whole
population goes, after 1000 generations, from 32 to only
1.7–2.5, but all the genotypes (31.4–32) would be main-
tained in the absence of this type of assortment. This
homogenizing effect is still strong if a more realist rate of
assortment is considered (m ¼ 0.9) because stable groups
of males and females with the same genotype for the tags
can never be consolidated.
5. Discussion
In this paper, we propose that assortative mating can
arise through sexual selection based on the female
preference for males who are similar to them for a number
of arbitrary characteristics or tags, even when both the
preference and the tags depend on independent and
unlinked genes. The necessary requisite for this process to
act is an asymmetry between partners in such a way that
the female can choose the male whereas the later must
always accept to mate. Under these circumstances it is
possible that a mechanism with a genetic basis (allele A)
arises, favoring the active female choice of partners using a
‘self seeking like’ decision rule. The fitness advantage is due
to the built up of linkage disequilibrium between locus A/a
and the tags. The disequilibrium can favor either one or the
other tag alleles but, in both cases, A males mate more
frequently.
As the number of tags increases the evolution of the A
allele is favored and in the limit (for example with more
than 1000 tags) the behavior of the A allele will be
equivalent to that of an allele promoting that the female
mates with her brothers. It is well known that such allele
would increase in frequency until its advantage is counter-
acted by inbreeding depression. This situation can be
interpreted using kin selection arguments (Bengtsson, 1978;
Maynard Smith, 1978), as a female who mated with her
brother will increase her inclusive fitness by increasing the
number of her brother’s genes passed to future generations.
However, this argument could not explain the increase in
frequency of the A allele observed in our model, which is
due to linkage disequilibrium between the loci involved as
shown by Eq. (1).
Any factor that favors linkage disequilibrium also favors
the evolution of the A allele. Thus, the advantage of A is
amplified if the population is not too large and the
preference relies on several tags and, for enough number
of tags (e.g. i ¼ 5), the A allele can increase in frequency
even in populations of considerable size (N ¼ 5000).
391
Notably, a subdivided population with a reasonable rate
of migration is very efficient to facilitate the increase in
frequency of the A allele and, in the simulation results, 100
generations are enough to bring the A allele to fixation.
To mitigate the cost that for females could arise as a
consequence of having preferences, we also test a situation
where females choose only among a limited subset of males
(e.g. females choose the best of n males) showing that the
A allele also increases in frequency although more slowly.
Moreover, when we assume a fixed cost such that the A
females have a probability e of remaining unmated, the
A allele still increases in frequency unless the e value is
quite substantial.
In classical models of sexual selection, the increase in
frequency of the preference allele goes together with an
increase in the trait object of preference but, in our model,
the average frequency of the tags does not change. This is
simply because assortative mating does not change gene
frequencies. Despite of this, the female tending to mate
with a male that is alike to her for some arbitrary markers
produces a dramatic effect on the population genotypic
structure of the tags. In a large population subdivided in
small subpopulations connected by migration, the assorta-
tive mating homogenizes the population as a whole in
contrast to random mating that maintains much of the
variation. In each subpopulation, the assortment decreases
the initial number of genotypes present because there are
males that are not chosen as mates. Besides, migration
impedes the attainment of a stable equilibrium between
groups of similar males and females mating between
themselves. As consequence, the whole population became
homogeneous with two or three genotypes coexisting
during many generations. This result is obtained for
m ¼ 1 or, in a more realistic situation, when some errors
are allowed (m ¼ 0.9).
Acknowledgments
We thank Mark Kirkpatrick and two anonymous
reviewers for their comments on an earlier version of the
manuscript.
References
Andersson, M., 1994. Sexual Selection. Princeton University Press,
Princeton, NJ.
Aoki, K., Feldman, M.W., Kerr, B., 2001. Models of sexual selection on a
quantitative genetic trait when preference is acquired by sexual
imprinting. Evolution 55, 25–32.
Bengtsson, B.O., 1978. Avoiding inbreeding: at what cost? J. Theor. Biol.
73, 439–444.
Day, T., 2000. Sexual selection and the evolution of costly female
preferences: spatial effects. Evolution 54, 715–730.
Hauber, M.E., Sherman, P.W., 2001. Self-referent phenotype matching:
theoretical considerations and empirical evidence. Trends Neurosc. 24,
609–616.
Kirkpatrick, M., 1982. Sexual selection and the evolution of female choice.
Evolution 36, 1–12.
 ARTICLE IN PRESS
392 L. Castro, M.A. Toro / Journal of Theoretical Biology 243 (2006) 386–392

Kirkpatrick, M., 1985. Evolution of female choice and male parental
investment in polygynous species: the demise of the ‘‘sexy son’’. Am.
Nat. 125, 788–810.
Lande, R., 1981. Models of speciation by sexual selection on polygenic
traits. Proc. Natl. Acad. Sci. USA 78, 3721–3725.
Lorenz, K., 1935. Der Kumpan in der Umwelt des Vogels. J. Ornithol. 83,
137–215, 289–413.
Maynard Smith, J., 1978. The Evolution of Sex. Cambridge University
Press, Cambridge.
Mead, L.S., Anold, S.J., 2004. Quantitative geneticmodels of sexual
selection. Trends Ecol. Evol. 19, 264–271.
Pomiankowski, A., Iwasa, Y., Nee, S., 1991. The evolution of costly
mate preferences. I. Fisher and biased mutation. Evolution 45,
1422–1430.
Seger, J., 1985. Unifying genetic models for the evolution of female choice.
Evolution 39, 1185–1193.
Servedio, M.R., 2000. Reinforcement and the genetics of non-random
mating. Evolution 54, 21–29.
ten Cate, C., Vos, D.R., 1999. Sexual imprinting and evolutionary
processes in birds: a reassessment. Adv. Study Behav. 28, 1–31.
Verzijden, M.N., Lachlan, R.F., Servedio, M.R., 2005. Female mate-
choice behavior and sympatric speciation. Evolution 59, 2097–2108.