Natural selection on testosterone production in a wild songbird population.

vol. 175, no. 6 the american naturalist june 2010 �
Natural Selection on Testosterone Production
in a Wild Songbird Population
Joel W. McGlothlin,1,* Danielle J. Whittaker,2 Sara E. Schrock,2 Nicole M. Gerlach,2
Jodie M. Jawor,3 Eric A. Snajdr,2 and Ellen D. Ketterson2
1. Department of Biology, University of Virginia, Charlottesville, Virginia 22904; 2. Department of Biology and Center for the
Integrative Study of Animal Behavior, Indiana University, Bloomington, Indiana 47405; 3. Department of Biological Sciences,
University of Southern Mississippi, Hattiesburg, Mississippi 39406
Submitted August 11, 2009; Accepted February 10, 2010; Electronically published April 15, 2010
Online enhancement: appendix.
abstract: Because of their role in mediating life-history trade-offs,
hormones are expected to be strongly associated with components
of fitness; however, few studies have examined how natural selection
acts on hormonal variation in the wild. In a songbird, the dark-eyed
junco (Junco hyemalis), field experiments have shown that exogenous
testosterone alters individuals’ resolution of the survival-reproduc-
tion trade-off, enhancing reproduction at the expense of survival.
Here we used standardized injections of gonadotropin-releasing hor-
mone (GnRH) to assay variation in the testosterone production of
males. Using measurements of annual survival and reproduction, we
found evidence of strong natural selection acting on GnRH-induced
increases in testosterone. Opposite to what would be predicted from
the survival-reproduction trade-off, patterns of selection via survival
and reproduction were remarkably similar. Males with GnRH-
induced testosterone production levels that were slightly above the
population mean were more likely to survive and also produced more
offspring, leading to strong stabilizing selection. Partitioning repro-
duction into separate components revealed positive directional se-
lection via within-pair siring success and stabilizing selection via
extrapair mating success. Our data represent the most complete dem-
onstration of natural selection on hormones via multiple fitness com-
ponents, and they complement previous experiments to illuminate
testosterone’s role in the evolution of life-history trade-offs.
Keywords: natural selection, sexual selection, testosterone, life-history
trade-offs, GnRH challenge.
Introduction
Hormones are often crucial for the translation of genotype
to phenotype, regulating key steps in development and
integrating the expression of suites of functionally im-
portant traits (Moore 1991; Finch and Rose 1995; Ketter-
son and Nolan 1999; Wingfield et al. 2000; Nijhout 2003;
* Corresponding author; e-mail: jmcgloth@virginia.edu.
Am. Nat. 2010. Vol. 175, pp. 687–701. � 2010 by The University of Chicago.
0003-0147/2010/17506-51514$15.00. All rights reserved.
DOI: 10.1086/652469
Adkins-Regan 2005). Hormones are also expected to be
intimately related to fitness, and they often mechanistically
underlie life-history trade-offs among fitness components
such as survival and reproduction (Ketterson and Nolan
1992; Stearns 1992; Sinervo and Svensson 1998; Zera and
Harshman 2001; Ricklefs and Wikelski 2002; Adkins-
Regan 2005; Breuner et al. 2008; Lessells 2008; Mills et al.
2008, 2009; Bonier et al. 2009a). Despite the expected
relationship between hormones and fitness, we have little
information about how selection acts on individual vari-
ation in hormone levels and hormone-mediated traits in
the wild (Kingsolver et al. 2001; Adkins-Regan 2005). Tes-
tosterone has been linked to mating success in a number
of species (Borgia and Wingfield 1991; Alatalo et al. 1996;
Mills et al. 2007), and several recent studies have shown
relationships between corticosterone, a glucocorticoid
stress hormone, and survival or reproductive success
(Brown et al. 2005; Blas et al. 2007; Bonier et al. 2007,
2009a, 2009b; Cabezas et al. 2007; Breuner et al. 2008;
Angelier et al. 2009; John-Alder et al. 2009; MacDougall-
Shackleton et al. 2009). However, there is very little evi-
dence indicating how selection on hormones varies across
multiple components of fitness. This information is crucial
to an understanding of the role of hormones in mediating
life-history trade-offs (Arnold and Wade 1984; Schluter et
al. 1991).
Much of our understanding of the role of hormones in
the evolution of natural populations derives from “phe-
notypic engineering” studies, in which hormone levels are
experimentally altered in order to test for their effects on
phenotype and fitness (Ketterson and Nolan 1992, 1999;
Ketterson et al. 1996; Sinervo and Svensson 1998; Zera
and Harshman 2001; Adkins-Regan 2005; Reed et al.
2006). Despite their power to demonstrate causality, such
experiments provide only limited information about evo-
lutionary processes because the act of hormonal manip-

688 The American Naturalist
ulation alters the phenotypic variation on which selection
operates. For this reason, manipulative experiments should
be complemented by studies that focus on the causes and
consequences of individual hormonal variation (Kempen-
aers et al. 2008; McGlothlin and Ketterson 2008; Williams
2008; Ketterson et al. 2009). Hormonal manipulations can
probe the mechanistic bases of trade-offs, but studying
individual variation is necessary to reveal how these trade-
offs translate into selection in the wild and to predict
evolutionary change. Furthermore, manipulative studies
that measure components of fitness do not generate
straightforward predictions about natural selection be-
cause selection acting via fitness components that trade
off with one another does not always resemble the indi-
vidual trade-off function (Roff and Fairbairn 2007).
Rather, the pattern of natural selection via multiple com-
ponents of fitness is expected to depend on how individ-
uals vary in allocation (i.e., the resolution of trade-offs)
versus individual quality or resource acquisition (i.e., the
amount of currency, such as energy, available to allocate
between fitness components; van Noordwijk and de Jong
1986; Roff 2002; Roff and Fairbairn 2007). Depending on
how allocation and quality vary, patterns of selection on
the hormonal basis of trade-offs may range from mirroring
the individual trade-off function to having no resemblance
to it.
One of the reasons measuring selection on hormones
and hormone-mediated traits has proved to be difficult is
the ubiquity of within-individual variation (Williams
2008). For example, in most male songbirds, levels of tes-
tosterone (the most common androgen in birds) vary
markedly over the breeding season, usually decreasing after
an early breeding season peak (Wingfield et al. 1990; Goy-
mann et al. 2007). Although such seasonal patterns are
often of intrinsic interest, they may obscure the ability to
detect variation at the individual level. Short-term varia-
tion in testosterone levels is also common. In many species,
males transiently increase testosterone levels in response
to social stimuli such as male competitors or potential
mates (Harding 1981; Moore 1983; Wingfield 1985; Wing-
field et al. 1990; Hirschenhauser et al. 2003; Hirschen-
hauser and Oliveira 2006; Goymann et al. 2007; Landys
et al. 2007; Pinxten et al. 2007). This socially modulated
elevation in testosterone (or “androgen responsiveness”)
is likely to be particularly important when considering the
evolution of testosterone-mediated traits, including their
roles in the evolution of mating systems and life histories
(Wingfield et al. 1990; Hirschenhauser et al. 2003;
Hirschenhauser and Oliveira 2006; Goymann et al. 2007;
Landys et al. 2007; Goymann 2009). Although the func-
tions of short-term testosterone changes have not been
completely described, these changes have long been as-
sociated with territoriality (Wingfield 1985; Wingfield et
al. 1987) and response to females (Moore 1983; Pinxten
et al. 2007) and have recently been associated with parental
behavior as well (McGlothlin et al. 2007). Our previous
work suggests that measuring short-term testosterone in-
creases in the same individuals multiple times across the
breeding season is an effective way to assess individual
variation in hormone profiles and should generate suitable
measurements for measuring selection (Jawor et al. 2006).
In this study, we examine the relationships between in-
dividual variation in short-term testosterone elevation and
fitness components of adult males in a breeding population
of dark-eyed juncos (Junco hyemalis). A long-term study
of this population found that experimentally elevated tes-
tosterone levels decreased survival (Reed et al. 2006).
Testosterone-treated males more than compensated for re-
duced survival by siring more offspring via extrapair fer-
tilizations than did controls, and as a result, they had
higher fitness as measured by l, the projected relative rate
of population growth (Raouf et al. 1997; Reed et al. 2006).
These results strongly suggest a role for testosterone in
mediating the trade-off between survival and reproduc-
tion. However, the demonstration of this trade-off does
not lead directly to predictions for how selection should
act on testosterone in the wild. For example, if males differ
primarily in the allocation of resources to survival and
reproduction, we would expect to find that testosterone
decreases survival but increases reproductive success, sim-
ilar to that which has been demonstrated by experimental
studies. Alternatively, if males differ primarily in quality,
selection may act similarly via survival and reproduction
because high-quality males are able to expend more effort
on mating without diminishing their survival.
To test how natural variation in testosterone is related
to survival and reproduction, we measured selection acting
on both circulating testosterone levels and the ability of
males to produce short-term testosterone increases. Tes-
tosterone was measured at multiple points in the breeding
season in order to assess a male’s average testosterone
production. Short-term testosterone increases were mea-
sured using standardized injections of gonadotropin-
releasing hormone (GnRH challenges; Wingfield and Far-
ner 1993; Meddle et al. 2002; Millesi et al. 2002; Moore
et al. 2002; Jawor et al. 2006). In vivo, GnRH is produced
by the hypothalamus and regulates testosterone produc-
tion by stimulating the hypothalamo-pituitary-gonadal
(HPG) axis. Our GnRH challenge protocol is designed to
measure a male’s ability to elevate testosterone levels (i.e.,
the responsiveness of a male’s HPG axis), and it generates
repeatable short-term testosterone increases ( ; Ja-r p 0.36
wor et al. 2006). Levels of testosterone produced in re-
sponse to exogenous GnRH are strongly correlated with
those produced in response to social stimuli (male terri-
torial intruders) in the wild ( ; McGlothlin et al.r p 0.68

Natural Selection on Testosterone 689
2008). In addition, GnRH-induced increases are positively
correlated with attractive plumage and negatively corre-
lated with parental behavior, and absolute levels of post-
GnRH-challenge testosterone are positively correlated with
aggressive behavior (McGlothlin et al. 2007, 2008).
Over two breeding seasons, we measured selection act-
ing through annual survival, which was measured by re-
capture in the following breeding season, and annual off-
spring production, which was quantified using DNA
paternity analysis. These selective episodes were added to
estimate total annual selection (Arnold and Wade 1984;
Wade and Kalisz 1989; McGlothlin 2010). We also
partitioned reproductive selection in two different ways in
order to examine potential trade-offs between components
of reproduction. Juncos are socially monogamous breed-
ers, meaning that they may achieve reproductive success
either within or outside of a social pair (Ketterson et al.
1997; Nolan et al. 2002). Therefore, we asked whether
selection differed when acting via mating success versus
number of offspring per mate and via within-pair versus
extrapair reproduction.
Material and Methods
Study Species and General Methods
We studied a population of the Carolina subspecies of the
dark-eyed junco (Junco hyemalis carolinensis) that breeds
at and around Mountain Lake Biological Station in Giles
County, Virginia (37�22�N, 80�32�W), during the breeding
seasons of 2003 and 2004. Males in this population had
last been implanted with testosterone in 2000. In March
and April, male juncos establish breeding territories that
they defend throughout the season (Nolan et al. 2002).
Typically, a single female nests on a male’s territory. Fe-
males build nests, incubate clutches (usually of four eggs),
and brood nestlings alone. Both parents feed the offspring
after hatching. Mating often occurs outside the pair, and
experimentally elevated testosterone has been shown to
increase extrapair mating success (Ketterson et al. 1997;
Raouf et al. 1997; Nolan et al. 2002; Reed et al. 2006).
In April–August 2003–2004, males ( ) were cap-n p 90
tured using mist nets or Potter traps. On capture, birds
were transported to a central laboratory in a holding bag.
If previously uncaptured, birds were fitted with a num-
bered aluminum leg band and a unique combination of
colored plastic leg bands for identification. We determined
age (yearling or older adult [≥2 years]) by examining first
the color of the primary wing coverts and second the iris,
which are both lighter in yearlings (Nolan et al. 2002).
Age in years was determined by appearances in our capture
records from previous years. If a bird was first captured
and banded as an older adult, it was conservatively as-
sumed to be 2 years old in that year. Mass (g) was mea-
sured using a spring balance.
GnRH Challenges and Testosterone Assays
Each time a bird was captured, a blood sample was ob-
tained from the wing vein (initial or prechallenge sample).
Sampling details are reported in the appendix in the online
edition of the American Naturalist. Handling time was
recorded as the time in minutes from capture to collection
of this blood sample, averaging 48 min (range 2–217 min;
Jawor et al. 2006). Previous analyses have shown that in-
creased handling time weakly affects our measurements of
testosterone (Jawor et al. 2006). However, we were able
to control for handling time statistically, and analyzing a
reduced data set did not affect our results (see appendix).
A solution of 1.25 mg chicken GnRH-I (Sigma L0637) in
50 mL of 0.1 M phosphate-buffered saline (PBS) was in-
jected into an individual’s pectoral muscle. The bird was
returned to its holding bag, and after exactly 30 min, a
second blood sample was collected (postchallenge sample).
After collection of this sample, the bird was released at
the site of capture. Plasma was separated and frozen
(�20�C) for later hormone analysis. Red blood cells were
saved and stored in lysis buffer (Longmire et al. 1992) for
paternity analysis. Males were exposed to up to four GnRH
challenges over the course of a breeding season.
Our GnRH challenge method stimulates a large increase
in testosterone levels after 30 min, and levels return to
baseline in less than 2 h (Jawor et al. 2006). Individual
male juncos show repeatable differences in the magnitude
of testosterone increases above initial levels, despite sig-
nificant seasonal variation (Jawor et al. 2006).
We determined testosterone concentrations using an en-
zyme immunoassay kit (Assay Designs 901–065; Clotfelter
et al. 2004) and summarized measurements to generate a
single measurement for each male in each year. Details of
the testosterone assay and a summary of measurements
are reported in the appendix.
Paternity Analysis
We attempted to find all nests of pairs that nested on our
study site. When nests were found during egg laying, one
egg was collected as a part of another study (Jawor et al.
2007). Putative mothers and fathers of each nest were
identified by behavior at the nest. We used genetic pater-
nity analysis in order to assign genetic sires to offspring.
For adults, we extracted DNA from red blood cells col-
lected during GnRH challenges on males and from blood
samples from adult females that had been collected con-
currently. Blood samples were obtained from nestlings 6
days after hatching and stored in lysis buffer. We genotyped

690 The American Naturalist
a total of 265 6-day-old nestlings from the 2 years, as well
as 57 dams and 108 potential sires, using five microsatellite
loci, and we assigned paternity using Cervus 3.0 (Marshall
et al. 1998; Kalinowski et al. 2007). Details are reported
in the appendix.
Selection Analysis
To test how selection may act differently through different
components of fitness, we measured selection acting via
annual survival and via offspring production. We also par-
titioned annual reproduction in two different ways in or-
der to ask about trade-offs between different paths to re-
productive success. First, offspring production was par-
titioned into number of mates and number of offspring
per mate. Second, offspring production was partitioned
into within-pair and extrapair offspring. Details of fitness
component assignment and correlations between pairs of
fitness components are reported in the appendix.
Selection gradients for initial testosterone levels and
GnRH-induced testosterone increases were estimated as
regression slopes (Lande and Arnold 1983; Brodie et al.
1995). We conducted selection analyses for both years
combined. To correct for repeated measures of a single
male across the 2 years, we used PASW Statistics 17.0
(SPSS) to fit linear mixed models, with individual as a
random factor in each. To generate measurements of rel-
ative fitness, before analysis each fitness component was
divided by its average (survival, 0.553; offspring, 3.45;
number of mates, 1.29; offspring per mate, 2.84; within-
pair offspring, 2.23; extrapair offspring, 1.15). We mea-
sured both linear selection gradients (b), which indicate
positive or negative directional selection, and nonlinear
(quadratic) selection gradients (g), which arise from the
curvature of the relationship between fitness and pheno-
type and indicate either stabilizing/disruptive selection (gii)
or correlational selection (gij ; Lande and Arnold 1983;
Brodie et al. 1995). Directional selection gradients were
estimated in a model that included the two hormone mea-
surements (least squares individual means [see appendix],
standardized to zero mean and unit variance) as well as a
fixed effect of year and age in years as a covariate. Non-
linear terms were estimated by adding the squared terms
(initial testosterone2 and GnRH-induced testosterone in-
crease2) and the cross product (initial testoster-
one # GnRH-induced testosterone increase) to this
model. The squared-term regression coefficients (and their
standard errors) were doubled to generate the stabilizing/
disruptive selection gradients (Brodie et al. 1995; Stinch-
combe et al. 2008). Standardization of traits was performed
separately for each fitness component.
Statistical significance of selection gradients was tested
in ASReml 2.0 (Gilmour et al. 2006) using generalized
linear mixed models with appropriate error structures (bi-
nomial for annual survival, normal for offspring per mate,
and Poisson for all other fitness components). These mod-
els used absolute (not relative) fitness components as de-
pendent variables and a structure of independent variables
that was identical to the linear mixed models described
above. We report conditional values for F and P (Type III
sums of squares). Standard errors were calculated using
regression slopes from the linear mixed models and F val-
ues from generalized linear mixed models using the for-
mula . To visualize the form of selec-1/2SE p b (or g)/(F)
tion, we fitted univariate cubic splines using glms version
4.0/glmsWIN 1.0 (Schluter 1988). The smoothing param-
eter (l) was chosen by minimizing generalized cross-
validation scores. We also added selection gradients for
survival and reproduction to estimate the strength of total
annual selection. Details of this calculation are reported
in the appendix.
Results
We found statistically significant relationships between fit-
ness components and the testosterone increase induced by
GnRH, but there were no significant relationships between
fitness components and initial testosterone levels (table 1).
There was no significant directional selection on GnRH-
induced testosterone increases via annual survival or total
offspring production. The relationships with both fitness
components were curvilinear, as evidenced by the negative
quadratic selection gradients that indicate stabilizing se-
lection (table 1). Examination of the fitness function in-
dicated a pattern of stabilizing selection acting via both
survival and reproduction (fig. 1). In both cases, the fitness
optimum was shifted slightly to the right, reflecting the
weakly positive directional selection gradient for each fit-
ness component. Stated another way, males with slightly
higher than average GnRH-induced testosterone increases
were most successful in both survival and reproduction,
but those with very high increases were less successful
along both dimensions.
When total offspring production was split into number
of genetic mates and offspring per mate, we found no
significant selection acting via mating success but relatively
strong directional selection acting via offspring per mate
(table 1). The fitness function for mating success was rel-
atively flat, with a weak stabilizing component (fig. 1).
Males that produced intermediate GnRH-induced testos-
terone increases tended to have the highest mating success.
In contrast, the fitness function for offspring per mate was
strongly directional: males with higher GnRH-induced tes-
tosterone increases produced more offspring with each
mate (fig. 1).
A similar pattern was observed when total offspring

Natural Selection on Testosterone 691
production was split into within-pair and extrapair com-
ponents. We found significant positive directional selection
via within-pair success (table 1), and this remained sig-
nificant when we controlled for apparent within-pair off-
spring production (the total number of 6-day-old nestlings
produced by a male’s social partner, including the off-
spring he did not sire; table 1). This effect indicates that
males with higher GnRH-induced testosterone increases
were more successful primarily because they sired a higher
percentage of the offspring of their social mate that sur-
vived to be nestlings, and not because they were able to
attract a more fecund social mate. The fitness function for
within-pair offspring was mostly linear with a weak sta-
bilizing component that was probably driven by a single
male that produced a very high testosterone increase but
sired no offspring (fig. 1). Controlling for apparent off-
spring production removed this stabilizing component
(fig. 1). In contrast, selection acting via extrapair offspring
production was strongly stabilizing, with no significant
evidence of directional selection (table 1). The fitness func-
tion confirmed that the fitness optimum was situated near
the population mean (fig. 1).
When we combined survival and reproductive selection,
total annual selection on GnRH-induced testosterone in-
crease was strongly stabilizing (table 2). Directional selec-
tion via the two fitness components was reinforcing, lead-
ing to net positive directional selection via annual fitness
(table 2).
Discussion
Natural selection acting on the magnitude of GnRH-
induced testosterone production in our population was
primarily stabilizing, with some evidence of a directional
component. Stabilizing selection was strong, with a mag-
nitude that was well above the median quadratic selection
( ) reported by Kingsolver et al. (2001). PatternsFgFp 0.10
of selection via annual survival and offspring production
were very similar, and these two episodes of selection re-
inforced each other to generate stronger total annual se-
lection. The location of the fitness optimum did not differ
substantially for survival and reproduction, suggesting that
selection on natural variation in GnRH-induced testos-
terone increases does not mirror the within-individual
trade-off between survival and reproduction that is indi-
cated by experimental studies. Some evidence for different
trade-offs was found between components of annual off-
spring production; directional selection favored greater
GnRH-induced testosterone increases via offspring per
mate and within-pair siring success, but selection was
mostly stabilizing (with a nonsignificant negative direc-
tional selection component) via mating success and ex-
trapair offspring production.
Targets of Selection
Traditionally, measurements of selection on natural vari-
ation in hormones and other physiological traits have been
much less common than those on morphological traits
(Kingsolver et al. 2001; Adkins-Regan 2005). A number
of studies, most of them very recent, have demonstrated
selection acting on natural variation in animal hormone
levels (Borgia and Wingfield 1991; Alatalo et al. 1996;
Brown et al. 2005; Blas et al. 2007; Bonier et al. 2007,
2009a, 2009b; Cabezas et al. 2007; Mills et al. 2007; Breuner
et al. 2008; Angelier et al. 2009; John-Alder et al. 2009;
MacDougall-Shackleton et al. 2009). By far, most of this
work has been conducted on glucocorticoid stress hor-
mones. Significant directional selection on baseline glu-
cocorticoid levels has been found in a number of studies,
with some studies finding a negative relationship and oth-
ers finding a positive one (Bonier et al. 2009b; John-Alder
et al. 2009). Relationships between the acute glucocorti-
coid stress response and fitness also seem to be quite var-
iable (Breuner et al. 2008; Angelier et al. 2009;
MacDougall-Shackleton et al. 2009). Fewer studies have
reported selection on testosterone. Three studies found
evidence of strong positive relationships between circu-
lating testosterone and mating success (Borgia and Wing-
field 1991; Alatalo et al. 1996; Mills et al. 2007), whereas
other studies found no evidence of selection acting on
circulating testosterone (Brown et al. 2005; John-Alder et
al. 2009). Our study adds to the growing body of literature
demonstrating selection acting on natural variation in
hormones.
We found strong evidence of selection acting on GnRH-
induced testosterone increases but no evidence of selection
acting on pre-GnRH challenge levels. This difference is
perhaps unsurprising because GnRH-induced testosterone
increases are fairly repeatable within individuals, whereas
initial testosterone levels do not show similar individual
consistency (Jawor et al. 2006). Our measurements of ini-
tial testosterone are not necessarily identical to measure-
ments of breeding baseline testosterone because of the
handling time involved in obtaining the blood sample (see
appendix); therefore, additional study is required to ex-
amine potential relationships between baseline testoster-
one and fitness.
Testosterone levels produced in response to GnRH in-
jections have been shown to predict those produced in
response to territorial intruders, and responses to GnRH
challenges are also associated with variation in behavior
and plumage (McGlothlin et al. 2007, 2008). Specifically,
GnRH-induced testosterone increases were positively cor-
related with tail white, an attractive plumage trait, and
were negatively correlated with nestling feeding rate, while
absolute levels of GnRH-induced testosterone were posi-