Sex differences in survival selection in the serin, Serinus serinus
M. BJOÈ RKLUND* & J. C. SENAR *Department of Animal Ecology, Evolutionary Biology Centre, Uppsala University, Uppsala, Sweden
Museu de Zoologia, Barcelona, Spain
Keywords:
Abstract
adaptive landscape;
con¯icting selection;
correlational selection;
®nches;
natural selection.
Natural selection is demonstrated in most natural populations which suggests
that populations are dispatched from their adaptive peaks as a result of
selection on correlated characters, or con¯icting selection between the sexes.
We analysed patterns of survival selection in a population of serins (Serinus
serinus) outside Barcelona over a period of 13 years. There was directional
selection for increased wing length in males and females accompanied by
strong disruptive selection on both tail and wing length in males and a
selection against a positive correlation between the two characters in males. In
females there was directional selection for increased bill width but decreased
bill depth, which should be contrasted to the stabilizing selection acting on bill
depth in males. There were con¯icting selection on the characters within a sex
and con¯icting selection of the same characters between sexes, which
constrain the rate of access to the nearest adaptive peak.
Introduction
Natural selection is the most powerful process to shape
evolutionary change in the majority of circumstances in
nature. Consequently, the details of the process of
natural selection have been studied in considerable detail
in many species (see review by Endler, 1986). A major
pattern is that, given appropriate sample sizes, whenever
selection is looked for, indications of selection in some
form, is actually found. This means that natural selection
is currently acting on local populations and that these
populations rarely are on the top of their adaptive peak
but are more or less displaced for some reason. There are
several, not mutually exclusive, reasons for this: (i) a
changing environment, i.e. the position of the peak is
moving regularly which means that the population is
tracking the moving peak; (ii) con¯icting selection
pressures by different parts of the phenotype. As selection acts on whole phenotypes, con¯icting selection on
correlated parts of the phenotype results in unexpected
selection forces (Lande, 1979; Lande & Arnold, 1983;
Correspondence: Mats BjoÈrklund, Department of Animal Ecology, Evolutionary Biology Centre, NorbyvaÈgen 18D, SE-752 36 Uppsala, Sweden.
Tel.: +46-18-471-26-66; fax: +46-18-471-64-84;
e-mail: [email protected]
J. EVOL. BIOL. 14 (2001) 841±849 ã 2001 BLACKWELL SCIENCE LTD
Zeng, 1988), which in turn may result in unexpected
responses given the genetic architecture of the population (BjoÈrklund, 1996); (iii) strong genetic correlation
between the sexes for certain traits (Lande, 1980; Reeve
& Fairbairn, 1996; MerilaÈ et al., 1998), may force the two
sexes to depart from their sex-speci®c optimum if there
are sex-related differences in selection; (iv) selection is
acting only on the environmental part of the variation
(Alatalo et al., 1990; Van Tienderen & de Jong, 1994).
One major pattern in the morphology of birds is that
species within a genus tend to differ almost entirely in
terms of size, but not in terms of shape (BjoÈrklund, 1991,
1994). At a short-term scale this can readily be explained
in terms of genetic correlations among characters. However, as these patterns are found even at higher
taxonomic units, like families, which have lasted for
millions of years, this explanation loses its power with
increasing time (BjoÈrklund, 1994; but see Schluter,
1996). This is because the correlations themselves may
be subject to selection and thus change over time, and
there are theoretical studies showing that even strong
genetic correlations are unlikely to last for a very long
time (Lande, 1979; Zeng, 1988; Gromko, 1995; Lascoux,
1997). This suggests that the patterns we see are to some
extent a result of correlational selection (selection on
841
842
M . B J OÈ R K L U N D A N D J . C . S E N A R
character combinations, Endler, 1986) in addition to
selection for changes in mean and variance (Hansen,
1997). This kind of selection has, however, rarely been
analysed in natural populations compared with studies of
selection on means and variances (Schluter & Smith,
1986; BjoÈrklund, 1992; Brodie, 1992; Swain, 1992;
Schluter & Nychka, 1994).
Selection can be analysed in many ways in natural
populations. The most common way is to measure shortterm, or episodic selection, i.e. the selection that acts at
one (or a few) selective events. This will give information
about the current, short-term, selective forces acting on a
population. Fluctuating selection pressures at different
episodes will then give a picture of the extent of
environmental stochasticity (Grant, 1986). An alternative approach is to look for long-term selection by pooling
data over a large number of years. This will then give
additional information about the selection that acts
consistently on a given population. Signi®cant long-term
selection may be the result of long-term environmental
changes that may be masked by short-term environmental ¯uctuations. By extending a selection analysis to
include correlational selection, a long-term study can
give valuable insights in the relationship between traits.
Similarly, consistent long-term selection in one sex
suggest that there is opposing selection acting on another
correlated trait or on the same trait in the other sex.
Thus, by the analysis of long-term selection for different
traits and trait combinations, we can get a picture of the
relation between the phenotypes and their environments
that is over and beyond that from a short-term selection
analysis, in particular the extent of possible constraints.
This approach identi®es the pattern of selection, i.e. the
statistical relationship between phenotypes and their
environment. Given these patterns the ecological causes
of selection can then be identi®ed by proper observational and experimental data.
In this paper we analyse patterns of survival selection
in the serin (Serinus serinus), looking for selection not
only on means and variances, but also character combinations. By using data gathered over a time period of
13 years we are able to analyse long-term selection
patterns to obtain a picture of the long-term ®tness
surface. In particular, we are interested in the possibility
of con¯icting selection patterns both between characters,
but also between the sexes.
Materials and methods
Field data
The ®eld data has been collected by JCS, 1985±97 at an
orchard outside Barcelona, Spain. In total 4967 individuals were individually marked, with 6967 recaptures.
Captures were carried out about once every week all year
round (for details see Conroy et al., 1999). Birds were
sexed and aged according to Svensson (1992). In the
analysis birds were included only if ®rst captured before
1996 (4265 birds), because at present a certain proportion
of the latest captured birds are still alive and including
them will bias survival estimates downwards for these
birds. Birds were included only if they were recaptured
after 10 days or more to exclude migrants (Buckland &
Baillie, 1987; Peach et al., 1990). The population receives
an unknown number of migrants during winter. Most
migration of serins in Europe occurs in November
(Asensio, 1985). However, only about 4% (n ˆ 11 934)
of the total captures occur at this time in the study area
which strongly suggests that the migrants in this area are
few compared with the local population. In addition, the
migrants show a high degree of winter site ®delity, birds
returning the next winter to the same area (JCS, unpublished PhD thesis, Barcelona University), so that survival
is not masked by the effect of permanent emigration. We
only used data on birds in adult plumage and thus
excluded all data on juvenile birds (EURING age 3J) to
avoid any bias as a result of dispersal. We used several
trapping methods to reduce possible age bias in trapping
(DomeÁnech & Senar, 1997). Analyses are presented
separately for males and females to avoid bias because
of the absolute difference in survival between the sexes
(Conroy et al., 1999). Recapture rates did not differ
according to sex or age (Conroy et al., 1999).
The following traits were measured: wing length
(maximum chord), tail length, bill length (from the tip
to the anterior edge of nostrils), beak depth, beak width,
tarsus length, body mass. For details of the measurements
including information about measurement error see Senar
& Pascual (1997) and Borras et al. (2000). Body mass was
excluded from the analyses because of large withinindividual variance. All measurements were not taken all
years, which causes sample size to differ among traits.
Basic statistical analyses
All traits were checked for normality using ln-transformed data within each sex and individual values larger
than ‹3 SDs were deleted. We then checked for possible
outliers by calculating Hotelling's T 2-scores for all individuals (males and females separately) using ln-transformed data (Morrison, 1990). Individuals beyond ‹3
SDs were deleted. None of these procedures resulted in
any major reductions in the sample size as the `faulty'
values were generally very few (around 1%) and could
in most cases be interpreted as a result of simple typing
errors at data gathering. The means and standard deviations for each trait and sex are given in Table 1, and the
phenotypic correlations among the traits for each sex is
given in Table 2.
Analysis of selection ± the Lande±Arnold approach
All traits were then standardized (zero mean, unit variance) before selection analyses. Fitness was de®ned as the
J. EVOL. BIOL. 14 (2001) 841±849 ã 2001 BLACKWELL SCIENCE LTD
Selection in serins
Table 1 Basic descriptive data (in mm) for the traits analysed.
Females
Trait
Males
Mean SD
n
Mean SD
n
F
P
Wing length 68.74 1.67 1227 71.49 1.74 1439 1708.3 <0.0001
Tail length
46.85 1.66 453 48.92 1.64 589 406.1 <0.0001
Tarsus length 14.14 0.54 435 14.15 0.53 564
0.11 0.74
Bill length
5.84 0.28 258 5.80 0.24 342
2.7
0.10
Bill depth
5.77 0.40 383 5.72 0.38 490
2.50 0.11
Bill width
5.71 0.24 380 5.70 0.26 488
0.72 0.72
Table 2 Phenotypic correlations among the traits analysed for each
sex. Males above diagonal, females below. Figures in boldface are
signi®cantly different from zero at a table-wise a-level of 0.05.
n (males) = 293, n (females) = 214.
Wing
Tail
Tarsus
Bill length
Bill depth
Bill width
Wing
Tail
±
0.17
0.024
0.17 ±
)0.14 0.033
0.067 0.095
0.053 0.034
0.69
0.26
)0.089
0.013
0.11
0.64
±
Tarsus
Bill length Bill depth
Bill width
0.043
0.15
0.068
±
0.15
0.20
0.13
0.048
0.27
0.12
0.33
±
0.21
0.22
0.30
0.13
±
0.43
number of days between ®rst and last capture. Thus, the
®tness component studied in this paper is survival only.
Relative ®tness was calculated as absolute ®tness divided
by mean ®tness. All analyses were performed on the sexes
separately. The univariate directional selection gradients,
b, following the terminology by Arnold & Wade (1984),
were calculated as the regression of relative ®tness on the
standardized trait values. The univariate quadratic selection gradient C was calculated as the regression of relative
®tness on the squared trait values. This will generate a
negative slope if there is stabilizing selection and a positive
slope if there is disruptive selection. Quadratic selection
affects variance (increase or decrease) but so also does
directional selection. Thus, when calculating directional
and quadratic selection coef®cients they were both
entered into the regression, thus the quadratic selection
gradient is the amount of quadratic selection gradient
taking directional selection into account and vice versa.
The opportunity for selection (I) is de®ned as the relative
change in relative ®tness caused by selection and is
estimated as the variance in relative ®tness. Thus, the
opportunity for selection measures how much scope there
is for selection to act in general, rather than the actual
selection itself, i.e. a theoretical upper limit to selection
(Lande & Arnold, 1983).
Multivariate directional (bi) and stabilizing selection
gradients (cii) could not be calculated using all traits as
the matrix became singular. Therefore, we had to search
for character complexes to reduce the number of
parameters in the model. This was performed by means
of factor analysis for each sex separately. Based on these
J. EVOL. BIOL. 14 (2001) 841±849 ã 2001 BLACKWELL SCIENCE LTD
843
results, we used bill depth and bill width as one
compound character and wing length and tail length as
another, whereas the remaining traits seems to have little
connection with each other or the compound traits. In
addition to the directional and stabilizing components we
also estimated the correlational selection coef®cients (cij)
as the crossproducts of trait i and trait j. Thus, the full
regression model within each trait group involved two
directional selection, two quadratic selection and two
correlational selection coef®cients, at most six coef®cients. Signi®cance of all the coef®cients (univariate and
multivariate) was assessed by a weighted delete-one
jackknife procedure (Wu, 1986; Mitchell-Olds & Shaw,
1987) to avoid bias as a result of heteroscedastic
variances. To account for multiple tests, a simultaneous
100(1 ± a) con®dence interval was calculated as
bi ‹ Ö(r + 1) Fr+1, n)r+1(a), where r is the number of
variables and n is the number of observations (Johnson &
Wichern, 1988). Note that in the univariate case r was
chosen to be 12 (six traits ´ two coef®cients) to account
for the number of tests on the same data set. As we could
not remove the heteroscedasticity of the variances
signi®cance tests of the difference in selection coef®cients
between the sexes had to be in terms of analysing the
con®dence intervals of unbiased regressions (see above).
Non-overlapping con®dence intervals were taken as
evidence that the coef®cients differed between the sexes.
This procedure is likely to be conservative.
The values obtained using the multivariate approach
were then used to calculate individual ®tness values
using the ®tness function
X
X
XX
2
1
^ ˆa‡
w
bi zi ‡
cij zi zj
…1†
2cii zi ‡
where the summation is over all characters. b is the
standardized selection gradient, cii is the quadratic selection coef®cient where a negative value indicates stabilizing selection and a positive value indicates disruptive
selection, and cij is the selection coef®cient on character
correlations. Equation 1 can be written in matrix form as
follows:
^ ˆ a ‡ bT z ‡ 12zT cz:
w
…2†
This latter form can be used to get a visual interpretation
of the selection surface as shown by (Phillips & Arnold,
1989) whose procedure will be followed below. Although
the selection gradient vector, b, describes the direction
and length to the nearest adaptive peak, the c-matrix
contains the information about the shape (orientation
and curvature) of the ®tness surface. In particular, the
signs of eigenvalues of the c-matrix contain the information about the shape of the ®tness surface, whereas
the magnitude describes the curvature of the surface. By
taking the derivative of eqn 2 and setting the result equal
to zero the stationary point of the ®tness surface can be
calculated as
z0 ˆ ÿcÿ1 b
…3†
844
M . B J OÈ R K L U N D A N D J . C . S E N A R
(Phillips & Arnold, 1989). If all eigenvalues are of the
same sign then z0 is at a ®tness maximum (all eigenvalues negative), or ®tness minimum (all eigenvalues
positive), and if the signs differ then the ®tness surface
has a more complicated shape and the equilibrium is
unstable. Furthermore, if the eigenvalues are close to
zero then the surface is relatively ¯at, whereas high
values indicate a highly curved ®tness surface. The
con®dence interval for the eigenvalues was estimated
by the delete-one jackknife procedure.
Analysis of selection ± the nonparametric
approach
m
X
aj zj ˆ aT z
There was highly signi®cant directional selection for
increased size on feather characters (wing length, tail
length) in males (Table 3), but not in females (Table 3).
In addition, there was signi®cant stabilizing selection
on bill length in females and signi®cant disruptive
selection on wing length in males. There were also
indications of stabilizing selection on both bill depth and
width in males, as the upper 2.5% level for both traits
were almost zero (Table 3).
Multivariate selection ± Lande±Arnold approach
The Lande±Arnold approach has been criticized for its
dependence on a particular form of the ®tness surface,
i.e. the quadratic form (Schluter & Nychka, 1994).
Instead, Schluter and Nychka proposed a nonparametric
analysis which differs in two major ways from the
Lande±Arnold approach: it is ¯exible as no speci®c
mathematical form of the ®tness curve is assumed and it
is designed to handle non-normal ®tness components,
like survival. This method uses projection pursuit
regression (see Schluter & Nychka, 1994 for details)
which is based on a number of univariate cubic splines
(Schluter, 1988). In particular, the method tries to ®nd
the ®tness surface by extracting the cross-sections of the
surface that accounts for most of the variance in ®tness.
Thus, this cross-section de®nes a new compound trait, x,
which is a linear combination of the original traits, z.
This procedure is akin to a standard principal components analysis where the parameter space is reduced to
only a few new compound characters. This can be
formalized as
xj ˆ
Univariate selection
…4†
In females, there was a signi®cant directional selection
only for increased wing length when selection on tail
length was accounted for (Table 4). In males, there were
signi®cant directional selection for both wing and tail
length, but also signi®cant disruptive selection acting on
wing length, and a negative correlational selection
between wing and tail length (Table 4).
The selection gradient for tail was signi®cantly larger
for males as evident from the nonoverlapping con®dence
interval. Likewise, the disruptive selection differed signi®cantly from that of females, whereas correlational
selection coef®cient is not.
The sexes differed considerably with regard to bill
traits. In females, there was signi®cant directional selection for both bill traits but in opposite directions: increase
in bill width and decrease in bill depth (Table 5). In
males, on the other hand, there was signi®cant stabilizing
selection acting on bill depth, but no directional selection
(Table 5).
These differences can be visualized in terms of ®tness
surfaces. In the feather traits (wing and tail), there was
only one signi®cant eigenvalue of the c-matrix (Table 6)
iˆ1
where a is the constant that relates the trait value to the
new compound trait. Thus, a large a for a trait indicates
that this trait has a strong relation to the ®tness measure.
The a vector and its standard error was estimated by the
program PP by Schluter & Nychka (1994).
Results
The mean number of days surviving was 308.4
(SD ˆ 327.1, range 11±1729 days, median ˆ 195) in
males and 261.1 (SD ˆ 321.9, range 11±2688 days,
median ˆ 135) in females. The opportunity for selection
(I) was 1.12 for males (95% CI ˆ 0.93±1.34, based on
5000 bootstraps) and 1.52 (1.07±2.03) for females. This
means that the maximum change was 1.06 and 1.23
standard deviation units for males and females, respectively. The difference between the sexes was not signi®cant as judged from their overlapping con®dence
intervals.
Table 3 Univariate selection gradients and quadratic selection
coef®cients. Figures in boldface are signi®cantly different from zero
as indicated by the 95% con®dence interval.
Trait
b (95% SE)
C (95% SE)
n
Males
Wing
Tail
Tarsus
Bill length
Bill depth
Bill width
0.37 (0.25, 0.49)
0.29 (0.14, 0.44)
)0.079 ()0.25, 0.089)
0.059 ()0.14, 0.26)
0.024 ()0.19, 0.24)
)0.013 ()0.17, 0.14)
0.092 (0.046, 0.18)
)0.0010 ()0.097, 0.095)
0.0091 ()0.13, 0.15)
0.016 ()0.11, 0.14)
)0.11 ()0.21, )0.0080)
)0.086 ()0.18, 0.0031)
496
281
285
202
258
257
Females
Wing
Tail
Tarsus
Bill length
Bill depth
Bill width
0.15 ()0.013, 0.31)
)0.023 ()0.31, 0.26)
)0.0085 ()0.22, 0.20)
0.12 ()0.15, 0.39)
)0.12 ()0.44, 0.20)
0.11 ()0.14, 0.36)
)0.016 ()0.12, 0.089)
0.069 ()0.16, 0.30)
0.029 ()0.14, 0.20)
)0.21 ()0.44, )0.021)
)0.038 ()0.21, 0.13)
)0.10 ()0.26, 0.049)
352
172
182
131
172
171
J. EVOL. BIOL. 14 (2001) 841±849 ã 2001 BLACKWELL SCIENCE LTD
Selection in serins
845
Table 4 Multivariate selection gradients (b), quadratic selection (diagonal of c) and correlational selection coef®cients (off-diagonal of c) using
bill characters only. Figures in boldface are signi®cantly different from zero as indicated by the 95% con®dence interval.
c (95% SE)
Trait
b (95% SE)
Bill depth
Bill width
Females (n = 169)
Bill depth
Bill width
)0.20 ()0.40, ±0.012)
0.21 (0.050, 0.38)
)0.0018 ()0.092, 0.089)
)0.073 ()0.22, 0.076)
)0.063 ()0.17, 0.042)
)0.11 ()0.19, ±0.031)
0.013 (±0.085, 0.11)
)0.051 ()0.12, 0.017)
Males (n = 254)
Bill depth
Bill width
0.014 (±0.11, 0.15)
0.058 (±0.042, 0.16)
Table 5 Multivariate selection gradients (b), quadratic selection (diagonal of c) and correlational selection coef®cients (off-diagonal of c) of
body traits. Figures in boldface are signi®cantly different from zero as indicated by the 95% con®dence interval.
c (95% SE)
Trait
b (95% SE)
Wing
Tail
Females (n = 169)
Wing
Tail
0.30 (0.14, 0.35)
)0.15 ()0.34, 0.029)
)0.12 ()0.28, 0.028)
0.10 ()0.13, 0.33)
)0.02 ()0.32, 0.29)
0.22 (0.086, 0.33)
0.16 (0.050, 0.28)
0.19 (0.033, 0.32)
)0.22 (±0.38, ±0.026)
Males (n = 275)
Wing
Tail
Table 6 Eigenvalues (95% CI) of the c-matrix and values at the
stationary point (z0) for each sex. Figures in boldface are
signi®cantly different from zero as indicated by the 95% con®dence
interval.
Eigenvalues
Females
Males
)0.073 ()0.24, 0.09)
)0.069 ()0.19, 0.048)
0.034 ()0.099, 0.17)
)0.20 ()0.25, )0.14)
0.11 ()0.24, 0.46)
)0.13 ()0.23, )0.028)
0.35 (0.039, 0.66)
)0.086 ()0.20, 0.029)
Bill depth and width
Wing and tail length
in each sex. This means that the surface is curved in only
one direction.
This can be seen for males in Fig. 1a where the surface
has a valley in the direction of the associated eigenvector
(0.79, ±0.61), i.e. opposite directions in wing and tail
length, whereas the other dimension is (almost) ¯at.
Most individuals are found in the direction of no
curvature, i.e. in the ®tness valley. In females, the
curvature is less pronounced as indicated by the eigenvalue, which is about one-third of the eigenvalue in
males (Fig. 1b). The ®tness surface has in females a
curvature that is different from that in males ()0.23,
)0.97), such that the major direction of the curvature is
in terms of increasing wing and tail size.
J. EVOL. BIOL. 14 (2001) 841±849 ã 2001 BLACKWELL SCIENCE LTD
0.072 (±0.029, 0.17)
In the two bill traits the difference between the sexes is
considerable. In males, there was one signi®cant eigenvalue (Table 6), with the direction of the curvature being
in¯uenced by the stabilizing selection on bill depth,
whereas the surface is ¯at in the other possible direction.
In females, there was no signi®cant eigenvalue, and the
®tness surface is ¯at and pointing in the direction of the
b-vector (increased bill width and decreased bill depth).
Multivariate selection ± nonparametric approach
The nonparametric approach revealed a similar picture as
above. In males, the direction that accounted for most of
the variation was related to wing length (Table 7) and to
a lesser, nonsigni®cant, degree wing length in relation to
tail length. This means that the diverging selection on
wing and tail length was also apparent by using this
approach.
The univariate spline function for wing length shows
an almost linear function of survival and wing length
(Fig. 2a).
In females, the only signi®cant trait loading was for bill
width, which relates positively to survival (Table 7). The
function relating survival to the main projection show a
tendency of a peak at intermediate values (not shown),
whereas the univariate spline of bill width on survival
show an almost linear relationship with increasing
survival with increasing bill width below the mean
value, but levels off for larger individuals (Fig. 2b).
M . B J OÈ R K L U N D A N D J . C . S E N A R
846
(a)
(b)
Females
Females
Fitness
Fitness
il l
en
gth
gt
h
dth
i
Bill w
Ta
en
gl
in
W
Bill depth
Males
Males
Fitness
Fitness
Bill w
il l
idth
Ta
en
gth
gt
h
en
gl
in
W
Bill depth
Fig. 1 Fitness surfaces for (a) wing and tail length in both sexes (in SD units), and (b) for bill depth and width in both sexes (in SD units).
Fitness values for each individual were estimated using eqn 1. The dots represent individual values. The surface was calculated using a
quadratic smooth function.
Discussion
The selection analysis show that in both sexes there was
strong directional selection for increased wing length. In
addition, males experience strong directional selection
for tail length and selection for a negative correlation
between wing length and tail length. In females there
was strong directional selection for increased bill width
and decreased bill depth, contrasting to the pattern found
in males with stabilizing selection acting on bill depth.
J. EVOL. BIOL. 14 (2001) 841±849 ã 2001 BLACKWELL SCIENCE LTD
Selection in serins
Table 7 The direction of selection in males and females estimated
by the projection pursuit regression. Boldface ®gures are signi®cantly different from zero.
Direction a1
Trait
Females (SE)
Males (SE)
Wing length
Tail length
Tarsus length
Bill length
Bill depth
Bill width
ln (k)
0.42 (0.30)
)0.43 (0.35)
0.36 (0.41)
0.18 (0.34)
0.095 (0.39)
0.69 (0.32)
)13.57
0.77 (0.27)
)0.52 (0.30)
0.061 (0.36)
0.20 (0.35)
0.046 (0.38)
0.16 (0.31)
)13.57
Note. SE determined by 100 bootstraps.
2.00
1.75
(a)
Females
Relative fitness
1.50
1.25
1.00
0.75
0.50
0.25
0.00
–4
–3
–2
–1
0
1
3
2
4
Bill depth
2.6
2.2
(b)
Males
Fitness
1.8
1.4
1.0
0.6
0.2
–0.2
–5
–4
–3
–2
–1
0
Wing length
1
2
3
4
Fig. 2 The univariate spline function for (a) wing length (in SD
units) in males, and (b) bill width (in SD units) in females.
This means that there is con¯icting selection pressures
acting on the phenotype in both males and females, in
addition to the difference in selection acting on each sex.
These ®ndings have several causes and implications.
The ®nding that in both sexes long wings are bene®cial
can be related to ¯ight agility (Norberg, 1990) as this
species is a ground feeder in open areas and thus very
vulnerable to predation. In dominance experiments in
this species, tail length seems to be the most important
J. EVOL. BIOL. 14 (2001) 841±849 ã 2001 BLACKWELL SCIENCE LTD
847
factor determining dominance in males, whereas wing
length was not important at all (Senar, unpublished).
This suggests that wing length is indeed related to agility
rather than dominance, as tail length is selected only in
males. Interestingly, a negative correlational selection
between wing length and tail length was found. In this
population, the phenotypic correlation between these
two traits is around 0.60 in both sexes (Table 2). The
pattern of selection in bill traits also reveals con¯icting
selection for bill depth and width in females. There was
also con¯icting selection between the sexes as males
experience stabilizing selection on a trait there is directional selection for in females, i.e. bill depth. The
phenotypic correlation between bill depth and bill width
is positive and high in females (Table 2). If the genetic
correlation approximates that of the phenotypic (Roff,
1995) we can expect the rate of evolution in divergent
directions for these traits to be constrained by this
correlation (Zeng, 1988; BjoÈrklund, 1996). The rate of
change in the selected direction in the bill is even more
constrained by the stabilizing selection on bill depth in
males. If there is a high genetic correlation among the
sexes, which is reasonable to assume, given the available
evidence (Lande, 1980; Reeve & Fairbairn, 1996; MerilaÈ
et al., 1998) males will be faced with stabilizing selection
as long as they are displaced from the optimum by the
directional selection acting on females, whereas females
will experience directional selection as long as they are
displaced from their optimum by the stabilizing selection
acting on males.
The evolutionary consequences of these different
selection pressures on the sexes are determined by the
genetic variances and covariances between the characters
and the sexes. If the genetic correlations between the
sexes are high, then the pattern of selection can persist
for a long time as the resulting phenotypes will be a
compromise between the selection on each sex and thus
not be optimal. The same is also true for the correlations
between characters within each sex. Here, the magnitude
can be assumed to be lower than between the sexes, but
the short-term rate at which the adaptive peak can be
reached will be hampered by the presence of these
correlations. Unfortunately, nothing is known about
genetic correlations in this species, so further discussion
will be speculative.
The ®ndings in this study have implications for the
understanding of the adaptive landscape as estimated by
the quadratic approximation to the individual ®tness
surface. It is clear that different parts of the phenotype
are more or less displaced from their optima as a result of
con¯icting selection acting on different parts and the
sexes. This displacement can persist over a considerable
ecological time scale, affected mainly by selection and the
interrelationship between the traits. As it is the whole
phenotype that die or survive at a given time, the result
will be a compromise between different selection acting
on different parts of the phenotype. This study highlights
848
M . B J OÈ R K L U N D A N D J . C . S E N A R
the importance of long-term studies of selection for
understanding the curvature of the adaptive landscapes
found in nature which can act as a basis for further
theoretical studies.
This study also highlights the importance of not only
analysing the directional or stabilizing patterns of selection, but also the correlational selection. Unfortunately,
this requires considerable sample sizes and even in this
large study this may be a problem. The two methods used,
the parametric Lande±Arnold approach and the nonparametric approach of Schluter and Nyscha gave qualitatively
the same result, even if the result was clearer using the
former approach than the latter. Whether this is a result
of sample sizes, or the methods for inferring standard
error is not known, but it suggests that for medium levels
of selection, sample size has to be large. The ®nding of
directional selection for wing length has also been found
when survival rates were estimated by capture±recapture
data (Conroy et al., in press). This gives strong support for
the use of the methods used in this study.
In conclusion, this study illustrates that within a
population and given habitat the selection may differ
considerably between sexes. This in turn may result in a
slow response to selection, but details about that has to
await information about genetic correlations between
characters and sexes. However, the patterns of selection
give rise to several testable predictions about the ecological causes of selection and the relationship between
morphology and behaviour which could not have been
reached without this kind of analysis.
Acknowledgments
We thank Jon Stone and the reviewers for valuable
comments on this manuscript. MB was supported by
grants from the Swedish Natural Science Research
Council. JCS was supported by grants from DGICYT
research project BOS 2000-0141. JCS thanks Hermanitas
de la AsuncioÂn for allowing us to trap the birds in their
orchards, to J. DomeÁnech, D. BoneÂ, L.A. BoneÂ, J.L.
Copete, L.M. Copete, A. Degollada, J. Figuerola,
J. Cascales and D. Valera for ®eld assistance and
M.L. Arroyo for computer assistance.
References
Alatalo, R.V., Gustafsson, L. & Lundberg, A. 1990. Phenotypic
selection on heritable size traits: environmental variance and
genetic response. Am. Nat. 135: 464±471.
Arnold, S.J. & Wade, M.J. 1984. On the measurement of natural
and sexual selection: theory. Evolution 38: 709±719.
Asensio, B. 1985. Migracion en Espana del Verdecillo (Serinus
serinus) segun los resultados del anillamiento. Ardeola 32:
173±178.
BjoÈrklund, M. 1991. Patterns of morphological variation among
Cardueline ®nches (Fringillidae: Carduelinae). Biol. J. Linn.
Soc. 43: 239±248.
BjoÈrklund, M. 1992. Selection of bill size proportions in the
Common rose®nch Carpodacus erythrinus. Auk 109: 637±642.
BjoÈrklund, M. 1994. Processes generating macroevolutionary
patterns of morphological variation in birds: a simulation
study. J. Evol. Biol. 7: 727±742.
BjoÈrklund, M. 1996. The importance of evolutionary constraints
in ecological time scales. Evol. Ecol. 10: 423±431.
Borras, A., Pascual, J. & Senar, J.C. 2000. What do different bill
measures measure and what is the best method to use in
granivorous birds? J. Field Ornithol. 71: 606±611.
Brodie, E.D.I. 1992. Correlational selection for color pattern and
antipredator behaviour in the garter snake Thamnophis ordinoides. Evolution 46: 1284±1298.
Buckland, S.T. & Baillie, S.R. 1987. Estimating bird survival rates
from organized mist-netting programs. Acta Ornithol. 23:
89±100.
Conroy, M.J., Senar, J.C., Hines, J.E. & DomeÁnech, J. 1999.
Development and application of a mark-recapture model
incorporating predicted sex and transitory behaviour. Bird
Study 46: S62±S73.
Conroy, M.J., DomeÁnech, J. & Senar, J.C. in press. Analysis of
individual and time-speci®c covariate effects on survival of
Serinus serinus in Northeastern Spain. J. Appl. Stat. (in press).
DomeÁnech, F. & Senar, J.C. 1997. Trapping methods can bias age
ratio in samples of passerine populations. Bird Study 44:
348±354.
Endler, J.A. 1986. Natural Selection in the Wild. Princeton
University Press, Princeton, NJ.
Grant, P.R. 1986. Ecology and Evolution of Darwin's Finches.
Princeton University Press, Princeton, NJ.
Gromko, M.H. 1995. Unpredictability of correlated response to
selection: pleiotropy and sampling interact. Evolution 49:
685±693.
Hansen, T.F. 1997. Stabilizing selection and the comparative
analysis of adaptation. Evolution 51: 1341±1351.
Johnson, R.A. & Wichern, D.W. 1988. Applied Multivariate
Statistical Analysis. Prentice Hall, London.
Lande, R. 1979. Quantitative genetic analysis of multivariate
evolution, as applied to brain : body size allometry. Evolution
33: 402±416.
Lande, R. 1980. Sexual dimorphism, sexual selection and
adaptation in polygenic characters. Evolution 34: 292±307.
Lande, R. & Arnold, S.J. 1983. The measurement of selection on
correlated characters. Evolution 37: 1210±1226.
Lascoux, M. 1997. Unpredictability of correlated response to
selection: linkage and initial frequency also matter. Evolution
51: 1394±1400.
MerilaÈ, J., Sheldon, B.C. & Ellegren, H. 1998. Quantitative
genetics of sexual size dimorphism in the collared ¯ycatcher,
Ficedula albicollis. Evolution 52: 870±876.
Mitchell-Olds, T. & Shaw, R.G. 1987. Regression analysis of
natural selection: statistical inference and biological interpretations. Evolution 41: 1149±1161.
Morrison, D.F. 1990. Multivariate Statistical Methods. McGrawHill, New York, NY.
Norberg, U.M. 1990. Vertebrate Flight. Springer-Verlag, Berlin.
Peach, W.J., Buckland, S.T. & Baillie, S.R. 1990. Estimating
survival rates using mark±recapture data from multiple
ringing sites. The Ring 13: 87±102.
Phillips, P.C. & Arnold, S.J. 1989. Visualizing multivariate
selection. Evolution 43: 1209±1222.
J. EVOL. BIOL. 14 (2001) 841±849 ã 2001 BLACKWELL SCIENCE LTD
Selection in serins
Reeve, J.P. & Fairbairn, D.J. 1996. Sexual size dimorphism as
a correlated response to selection on body size: an empirical
test of the quantitative genetic model. Evolution 50:
1927±1938.
Roff, D.A. 1995. The estimation of genetic correlations from
phenotypic correlations: a test of Cheverud's conjecture.
Heredity 74: 481±490.
Schluter, D. 1988. Estimating the form of natural selection on a
quantitative trait. Evolution 42: 849±861.
Schluter, D. 1996. Adaptive radiation along genetic lines of least
resistance. Evolution 50: 1766±1774.
Schluter, D. & Nychka, D. 1994. Exploring ®tness surfaces. Am.
Nat. 143: 597±616.
Schluter, D. & Smith, J.N.M. 1986. Natural selection on beak
and body size in the song sparrow. Evolution 40: 221±231.
Senar, J.C. & Pascual, J. 1997. Keel and tarsus length may provide
a good predictor of avian body size. Ardea 85: 269±274.
J. EVOL. BIOL. 14 (2001) 841±849 ã 2001 BLACKWELL SCIENCE LTD
849
Svensson, L. 1992. Identi®cation Guide to European Passerines. Lars
Svensson, Stockholm.
Swain, D.P. 1992. Selective predation for vertebral number in
Gasterosteus aculeatus: reversals in the direction of selection at
different larval sizes. Evolution 46: 998±1013.
Van Tienderen, P.H. & de Jong, G. 1994. A general model of the
relation between phenotypic selection and genetic response.
J. Evol. Biol. 7: 1±12.
Wu, C.F.J. 1986. Jackknife, bootstrap and other resampling methods in regression analysis. Ann. Statistics 14:
1261±1295.
Zeng, Z.-B. 1988. Long-term correlated response, interpopulation covariance, and interspec®c allometry. Evolution 42:
363±374.
Received 4 April 2001; revised 23 May 2001; accepted: 2 June 2001