SEARCH

SEARCH BY CITATION

Keywords:

  • compensatory growth;
  • condition dependence;
  • developmental stress;
  • genic capture;
  • nutritional stress;
  • sexual ornament;
  • Taeniopygia guttata

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. References

The developmental stress hypothesis offers a mechanism to maintain honesty of sexually selected ornaments, because only high quality individuals will be able to develop full ornamentation in the face of stress during early development. Experimental tests of this hypothesis have traditionally involved the manipulation of one aspect of the rearing conditions and an examination of effects on adult traits. Here, we instead use a statistically powerful quantitative genetic approach to detect condition dependence. We use animal models to estimate environmental correlations between a measure of early growth and adult traits. This way, we could make use of the sometimes dramatic differences in early growth of more than 800 individually cross-fostered birds and measure the effect on a total of 23 different traits after birds reached maturity. We find strong effects of environmental growth conditions on adult body size, body mass and fat deposition, moderate effects on beak colour in both sexes, but no effect on song and plumage characters. Rather surprisingly, there was no effect on male attractiveness, both measured in mate choice trials and under socially complex conditions in aviaries. There was a trend for a positive effect of good growth conditions on the success at fertilizing eggs in males breeding in aviaries whereas longevity was not affected in either sex. We conclude that zebra finches are remarkably resilient to food shortage during growth and can compensate for poor growth conditions without much apparent life-history trade-offs. Our results do not support the hypothesis that sexually selected traits show heightened condition dependence compared to nonsexually selected traits.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. References

Current models of sexual selection assume that ornamental traits should be costly to develop or maintain to serve as honest indicators of quality (Zahavi, 1975; Grafen, 1990; Getty, 2006). This has lead to the wide-held assumption that condition dependence is a common feature of sexual traits (e.g. Johnstone, 1995; Rowe & Houle, 1996). Quality indicators could signal either (1) genetic and/or (2) environmental quality variation among males (Iwasa & Pomiankowski, 1999; Cotton et al., 2004). Thus, the condition dependence of sexual traits might evolve to signal male genetic or environmental quality, or both, and females might benefit from choosing elaborated males either via good genes (if 1) or direct benefits (if 1 and/or 2). Because most traits will exhibit some degree of condition dependence, a convincing demonstration of condition dependence of sexual ornaments requires that the relative condition dependence of sexual traits are compared to that of naturally selected traits (Hamilton & Zuk, 1982; Zuk et al., 1990; Cotton et al., 2004).

Condition dependence could come about through developmental or maintenance costs of ornament elaboration. Environmental conditions during early development might be especially influential, because stress during this crucial period might have strong effects on the development of sexual ornaments, either because the traits are fixed during early ontogeny (like song learning in close-ended learners, Nowicki et al., 1998) or because early conditions have knock-on effects later on in life. Thus, if individuals that fared better in the face of early environmental stress have more elaborate ornaments, these ornaments can signal past condition. This was termed the developmental stress hypothesis by Nowicki et al. (1998).

From this outline, it becomes clear that the measurement and definition of condition is not a straightforward issue. Here, we will mainly focus on the effects of the early environment on the expression of secondary sexual traits in adulthood (i.e. the potential to signal past condition in the sense of the developmental stress hypothesis). However, we will also devote some attention to the genetic basis of condition dependence (the genic capture hypothesis, Rowe & Houle, 1996; Tomkins et al., 2004).

If ornaments reflect past condition, a female choosing a male with an elaborate ornament might receive more direct benefits or indirect benefits, or both. However, there also exists a line of evidence showing that individuals that grow up under poor conditions are often able to compensate for this when conditions improve (Metcalfe & Monaghan, 2001). To be adaptive, this compensatory growth should bring potential fitness benefits that outweigh the potential long-term costs (Metcalfe & Monaghan, 2001). For example, it might be beneficial to trade-off investment in favour of a sexual ornament if this leads to short-term fitness benefits (reproductive success) that outweigh long-term costs because of the reduced somatic maintenance (e.g. shortened life-span, Lindström et al., 2005). Therefore, more studies are needed that look at life-time effects of early condition, not only on subsequent growth and development of secondary sexual characters, but also on life-history consequences.

The ease with which environmental quality can be manipulated in the laboratory has lead to a surge of studies that have tested the developmental stress hypothesis in a range of taxa and found effects on sexual ornaments (e.g. David et al., 2000; Nowicki et al., 2002; Ohlsson et al., 2002; Scheuber et al., 2003). The zebra finch, Taeniopygia guttata, has become a model species to study the effects of developmental stress. Typically, two treatment groups are created by manipulating the early environment (e.g. brood size or food access or quality), and then traits of interest are compared between the two groups in adulthood (see Table 1). However, because of the ethical considerations, relatively mild treatments have been applied, such that, typically the mean mass at the age of 8 days was about 10–15% less in the food stress treatment than in the control group (e.g. Spencer et al., 2003; Arnold et al., 2007). Results, particularly those regarding sexually selected traits, have not always been consistent among studies (Table 1), and it might be argued that this is because the applied food restriction treatments might be ineffective under ad libitum food supply in the laboratory.

Table 1.   Overview of zebra finch studies relating early growth conditions to the same adult traits that are investigated in the present study. When possible, effect sizes (r-values) are given. Effect sizes were either obtained directly or, in most cases, estimated from figures in the original publications. Significant effects are marked in bold and the direction of the effect is indicated with + and − (for studies where an estimation of effect sizes was not possible, the effect is indicated only with + and −, cases where the direction of the effect was not indicated in the original study is denoted by ?). To calculate effect sizes, we first calculated Cohen’s d and then transformed it to r as a standardized measure of the strength of the relationship. We calculated d based on means and standard deviations (σ) for the two treatment groups as: = M1 − M2pooled, where inline image (Cohen, 1988). We then used the following formula to convert d to r: r = d/√(d2 + 4) (Cooper et al., 2009).
TraitDirection and strength of effects
  1. *Wing: effect in males only.

  2. †All effects averaged for wild and domesticated birds. A first principal component of body size used instead of tarsus.

  3. ‡Beak colour: measured in males only. Longevity: effect only in females.

  4. §Wing: effect in females only.

  5. ¶Longevity: measured in males only.

Mass+ + ?+0.24+0.44+0.45+++0.43+0.54
Tarsus  ++0.42 +0.24+0.57+0.45++0.37 +0.55
Wing  + +0.66 +0.47 ++0.58 +0.30
Condition  +    +0.12   +0.41
Beak colour     ? +0.33++0.33
Attractiveness+ +    ++  0.06
Song rate? ?  −0.32 +0.32  +0.11
Syllable rate +0.4        −0.10
Repertoire size+??+?      +0.05
Motif duration+0.24–0.36?+?      0.01
Longevity       +  +−0.03
Type of stressFood accessFood accessBrood sizeBrood sizeFood accessBrood sizeFood qualityBrood sizeFood qualityFood qualityFood qualityNatural variation
Approximate durationD 5–30D 0–30D 2–35D 0–35D 5–30D 0–35D 0–100D 0–50D 0–15D 0–15D 0–35D 0–35
StudySpencer et al. (2003, 2005)Zann & Cash (2007)Naguib et al. (2004); Gil et al. (2006); Naguib et al. (2008)Holveck & Riebel (2007)Brumm et al. (2009)*Tschirren et al. (2009)Boag (1987)de Kogel & Prijs (1996); de Kogel (1997)Blount et al. (2003)Arnold et al. (2007)§Birkhead et al. (1999)This study

However, even under ad libitum food supply, a substantial number of nestlings actually experience very harsh rearing conditions, for instance because of their late hatching within a larger brood or because their parents are very poor provisioners. In our population, this variation in early growth is quite dramatic (Schielzeth et al., 2008), with offspring mass at day 8 varying between 1.9 and 12.2 g (7.5 ± 1.7, mean ± SD, N = 974). Although this enormous variation in rearing conditions is difficult to accommodate within the classical fixed-effect treatment approach (where it represents noise around treatment means), it is readily accessible for analysis of condition dependence with a quantitative genetic approach. This approach requires that birds are randomly cross-fostered among broods, because this allows us to separate the environmental effects from the genetic differences in early growth.

Here, we test the effects of this full range of early growth conditions on a range of traits in a large sample. We employ full individual cross-fostering of the offspring at the egg stage within 1 day of laying, so that environmental conditions are independent of genetic or parental qualities. The cross-fostering ensures that each egg is randomly assigned with regard to clutch size, hatch order and foster parents, and siblings are spread randomly across foster pairs. We use a quantitative genetic approach employing an ‘animal model’ to account for the effect of the individual’s own genotype on its early growth and adult traits (Kruuk, 2004). Our main focus is on environmental correlations (i.e. the extent to which the experienced environmental growth conditions have correlated effects on early growth and adult traits). However, the use of multi-trait animal models allows us to also look at genetic correlations (i.e. the extent to which the genotypes of individuals have correlated effects on early growth and adult traits). With our approach, experimental manipulation of early growth conditions takes a different value for each cross-fostered individual (rather than the classical two-level fixed effect treatment), and the strength of this treatment is estimated from the growth phenotype by statistically accounting for heritable variation in early growth. Importantly, this analysis captures the entire environmental component of early growth, which represents about 87% of the total phenotypic variation in growth (see Methods), instead of focusing on only one aspect of the environment (e.g. brood size).

We measure both presumably naturally and presumably sexually selected traits in both sexes, so that the condition dependence of sexually selected traits can be compared to that of nonsexually selected traits, or to homologous, but presumably not sexually selected traits in the opposite sex (Cotton et al., 2004). To identify any life-history trade-offs, we also look at reproductive success and longevity.

In this way, we obtain a relatively complete picture of the potential for traits to honestly indicate quality by reflecting past condition in the zebra finch. Although we might expect most traits to exhibit some degree of condition dependence, the developmental stress hypothesis predicts that sexually selected traits (e.g. song, beak colour, plumage ornaments) should show heightened condition dependence in comparison with naturally selected traits. Further, if compensation occurs and is costly, we would expect effects on attractiveness, reproductive success and/or longevity. We expect the strongest effects on traits that develop during the actual period of stress (e.g. tarsus length), whereas we expect weaker effects on traits that develop after the period of stress has ended. In contrast to the environmental correlations, the predictions regarding the genetic correlation between early growth and adult traits are unclear. If growing to a large size reflects ‘good genes’, genes for a high mass at day 8 may be generally ‘good genes’. Alternatively, a high mass at day 8 may reflect a fast growth strategy (e.g. to avoid nest predation by minimizing the time spent in the nest). In this case, we may rather expect trade-offs between early growth and the development of adult traits. Our aim here is therefore to conduct an initial exploration of the genetic correlations between mass at day 8 and adult traits.

Methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. References

Subjects and housing

We based our analyses on data from three generations of zebra finches (termed F1–F3). The F1 generation was held at the Max Planck Institute for Ornithology in Seewiesen, Germany, since October 2004, whereas the F2 and F3 were bred in Seewiesen during 2005–2007 (see the aviary experiments described in the following text). For details of rearing conditions of the F1 generation, see Forstmeier (2005). For details of general housing conditions, see Bolund et al. (2007). Birds of the F1 (N = 309) and F2 generation (N = 512) were randomly and individually cross-fostered between nests within 24 h of egg-laying, ensuring that foster-siblings were unrelated to each other and to their foster parents. Whenever possible, the size of the foster clutch was not enlarged or reduced during this procedure. During the hatching period, nests were checked three times daily for new hatchlings to ensure accurate age assignment of chicks. Birds of the F3 generation (N = 153) were not cross-fostered, but their phenotypic data were still included in most of the animal models, because the main focus was not on elucidating different sources of environmental variation.

Growth measure and quantitative genetic analyses

The mass of all chicks was measured at day 8 after hatching to the nearest 0.1 g (massD8). This measure formed the basis for our estimation of early growth conditions. To account for the genetic contribution to massD8, we employ an ‘animal model’ (using REML VCE 6.0.2), which uses pedigree information to estimate the genetic component of massD8 (Lynch & Walsh, 1998; Kruuk, 2004). This approach allows us to separate out the additive genetic contribution to early growth and focus on the remaining environmental component (this will include any dominance and epistatic effects). The heritability (ratio of additive genetic variance to total phenotypic variance) of massD8 was 0.13 ± 0.048, and hence the environmental component represents 87% of the total phenotypic variation in massD8. The environmental component can be further decomposed into maternal effects, brood of rearing effects, hatch order within brood effects and residual variance representing the otherwise unaccounted individually specific rearing environment. To do this, we excluded the phenotypic data of 153 birds that had not been cross-fostered (the F3 generation) and ran the animal model including maternal clutch of origin (Vmaternal/Vphenotypic = 0.096 ± 0.036, N = 355 clutches), brood of rearing (Vfoster/Vphenotypic = 0.17 ± 0.037, N = 350 broods), hatch order (Vhatch order = 0.17 ± 0.081, range = 1–6) and residual variance (Vresidual = 0.45 ± 0.067). This model yielded a slightly lower heritability estimate of mass at day 8 (Vadditive/Vphenotypic = 0.11 ± 0.045). We estimated the heritabilities of all adult traits from one-trait models (Table 2).

Table 2.   Correlations between mass at day 8 and a range of traits in zebra finches. Shown are the heritabilities of all traits (h2), environmental correlations (rE), the genetic correlations (rG) and the phenotypic correlations (rP). The environmental correlation reflects the effect of the entire environmental component (87% of the total variation) of mass at day 8 on the adult trait. Note that standard errors are generally larger for genetic than environmental correlations, because of the relatively low heritability of mass at day 8. This is especially pronounced when also the adult trait has a low heritability and thus genetic correlations are not shown (marked with ‘X’) for traits with low heritabilities or limited sample sizes (for n = 99 given). Bold print highlights effects that are more than 1.96 standard errors from zero. For details of traits investigated, see text. Note that sample sizes vary for the different traits.
Traith2rE ± SErG ± SErPN
MalesFemalesMalesFemalesMalesFemalesMalesFemales
  1. FA, fluctuating asymmetry; Ch., choice; pref., preference; PC1, first principal component; PC2, second principal component.

General morphology
 Adult mass0.60 ± 0.050.52 ± 0.050.55 ± 0.060.50 ± 0.150.49 ± 0.130.470.47485459
 Adult tarsus0.51 ± 0.060.59 ± 0.070.50 ± 0.060.37 ± 0.140.52 ± 0.140.420.46489462
 Adult wing0.72 ± 0.050.35 ± 0.090.26 ± 0.070.38 ± 0.130.29 ± 0.140.270.23489462
 Adult condition0.55 ± 0.060.43 ± 0.060.41 ± 0.060.21 ± 0.160.28 ± 0.160.310.32485459
 Adult fat score0.32 ± 0.050.20 ± 0.060.25 ± 0.060.27 ± 0.190.54 ± 0.150.210.30485458
Presumably sexually selected traits
 FA tarsusX0.16 ± 0.060.12 ± 0.05  X  X0.0670.081383370
 Beak colour, PC10.41 ± 0.080.22 ± 0.070.40 ± 0.07−0.32 ± 0.17−0.33 ± 0.170.0890.14480454
 Beak colour, red  reflectance0.43 ± 0.060.11 ± 0.060.37 ± 0.08−0.26 ± 0.15−0.32 ± 0.170.0520.14480454
 Beak colour, Munsell  score0.36 ± 0.020.22 ± 0.080.30 ± 0.080.20 ± 0.22−0.21 ± 0.220.210.16485459
 Cheek colour, PC10.42 ± 0.15−0.031 ± 0.1 0.37 ± 0.21 0.07 241 
 Cheek colour, PC20.018 ± 0.10.045 ± 0.07   X 0.010 241 
 Breast band size0.38 ± 0.10−0.056 ± 0.07 0.095 ± 0.22 −0.020 489 
 Song rate0.33 ± 0.060.11 ± 0.07 0.11 ± 0.23 0.11 474 
 Repertoire size0.21 ± 0.090.048 ± 0.06 −0.047 ± 0.30 0.033 431 
 Motif duration0.14 ± 0.080.0081 ± 0.06 −0.044 ± 0.37 0.0013 431 
 Syllable rate0.25 ± 0.10−0.098 ± 0.07 0.18 ± 0.27 −0.047 434 
Life-history traits
 Extra-pair attractiveness0.27 ± 0.23−0.027 ± 0.12 −0.65 ± 0.49 −0.13 99 
 Ch. chamber  attractiveness0.05 ± 0.050.070 ± 0.05   X 0.11 456 
 Ch. chamber hops0.29 ± 0.09 −0.016 ± 0.07 −0.089 ± 0.24 0.030 439
 Ch. chamber pref.  strength0.097 ± 0.07 −0.020 ± 0.06 −0.087 ± 0.35 0.027 439
 Responsiveness to  males0.5 ± 0.08 −0.075 ± 0.08 −0.093 ± 0.22 −0.073 456
 Longevity0.068 ± 0.16−0.14 ± 0.160.077 ± 0.16  XX0.075−0.0676779
 Fitness males0.28 ± 0.220.21 ± 0.12 0.04 ± 0.33 0.16 103 
 Fitness females0.33 ± 0.21 0.13 ± 0.13 0.74 ± 0.29 −0.055 107

To look at how early growth relates to adult traits, we follow Hadfield et al. (in press) and estimate genetic and environmental correlations between traits directly in one step from a series of two- and three-trait animal models. Thus, adult traits are also decomposed into their genetic and environmental components, and so are the correlations between massD8 and adult trait. Hence, for each trait, we obtain three different correlations: a phenotypic, a genetic and an environmental correlation. The genetic correlation estimates the extent to which both traits (massD8 and the respective adult trait) are affected by the same set of genes, whereas the environmental correlation measures the extent to which the sum of all environmental influences on massD8 has correlated effects on the adult trait.

We use VCE to estimate the genetic and environmental covariances between massD8 and adult traits. We aimed to make animal models as simple as possible, and included only massD8 and the adult trait of interest (i.e. no fixed effects were added). For traits that occur in both sexes, we obtain separate covariances for the two sexes by running three-trait models including massD8 of both sexes, the adult trait in females and the adult trait in males. There was a small degree of sexual dimorphism in massD8 (males: 7.36 ± 1.68 g, females: 7.65 ± 1.65 g, mean ± SD, effect size = 0.17). To incorporate this, we ran models with massD8 treated as two traits (e.g. a four-trait model of massD8 and beak colour, with both traits treated separately in the two sexes). This did not qualitatively alter the results compared to models where massD8 was treated as one trait. We therefore opted for the more simple models, where massD8 is treated as one trait. With this procedure, VCE returns the genetic and environmental covariances between massD8 and the adult traits separately for the two sexes. To transform the covariance into a correlation, the covariance between the traits is divided by the geometric mean of the variances in both traits. For quantitative genetic analyses, we used a pedigree including all genetic relationships of our focal birds and their direct ancestors; = 1374 individuals from five generations. The pedigree contained 253 mothers and 220 fathers in 375 pair combinations. Phenotypic data of mass at day 8 was available for = 974 individuals from generation F1–F3 (474 females and 500 males). Table 2 shows the sample sizes for adult traits, sample sizes vary because not all traits were measured in all generations.

Adult measurements

General morphology

At nutritional independence (day 35), young were weighed and transferred to peer groups of approximately same-aged birds. Peer groups differed in group size (2–75 birds) and sexes were either separated (F1 and F2) or mixed (F2 and F3). At approximately 100 days of age (106.5 ± 12.1 mean ± SD), we measured body mass (to the nearest 0.1 g, mass D100), tarsus length (to the nearest 0.1 mm), wing length (to the nearest 0.5 mm) and fat score (clavicular and abdominal fat, scored on a scale from 0 to 5, with 0.5 point increments) and width of the breast band (in males, on a scale from 0 to 5, with 0.5 point increments, following Burley & Bartels, 1990). We estimated body condition as the residuals from a multiple regression of linearized adult mass (cube-root transformed) over tarsus and wing length.

Presumably sexually selected traits and behaviours

We measured beak colour of birds around day 100 as follows: (1) we scored the colour according to the Munsell colour chip system (see Forstmeier & Birkhead, 2004), (2) we used spectrophotometry to measure the reflectance of the beak, including the UV-part of the spectrum. From the beak reflectance spectrograms, we extracted two variables that were used for further analyses: (i) the first principal component (‘PC1’) of six spectral characteristics (as described in Bolund et al., 2007) and (ii) the point where the reflectance curve reaches half its maximum height in the red part of the spectrum (between 511 and 600 nm, ‘redness’). This measure reflects the ‘redness’ of the beak and is highly correlated with the Munsell scores (measures from the day 100 measuring sessions: r = 0.97, P < 0.00001, N = 654, across the two sexes). All beak colour measures range from orange female-like (low values) to red male-like (high values).

Around day 100, we also measured the colour of the cheek patches with spectrophotometry (in males of the F2 generation only). To capture the main axes of variation in cheek colour, we used the per cent reflectance at every 20 nm over the relevant part of the spectrum (310–690 nm) and entered these into a principal component analysis. PC1 reflects mainly brightness and explained 78% of the total variation. All entered points had strong positive loadings, and higher values of PC1 correspond to cheeks that were visually scored as brighter. Second principal component (PC2) explained a further 19% of the variation. Points from the lower half of the spectrum loaded negatively on PC2, whereas points from the upper half of the spectrum loaded positively. Higher values of PC2 correspond to a relatively higher contribution of reflectance at long wavelengths.

To measure the fluctuating asymmetry (FA) of the tarsus, both tarsi were measured twice, and the FA was estimated according to the mixed regression method described in Van Dongen & Molenberghs (1999).

We measured courtship song rate (also termed ‘directed’ song rate, Zann, 1996) under standardized conditions as described in Forstmeier (2004). During the same trials, we also measured ‘female responsiveness’ (Forstmeier, 2004). The song of all males was recorded in a soundproof chamber (the male was presented with a female and the directed song recorded). With the aid of Sound Analysis Pro version 2.063 (Tchernichovski et al., 2004), we extracted three parameters that have frequently been used in zebra finch song research: the length of the motif, the number of different elements in the motif and the syllable rate (for details, see Forstmeier et al., 2009).

Life-history traits

Male attractiveness and female choice behaviour were measured in a choice chamber set-up as described in Forstmeier & Birkhead (2004). We measured two aspects of female choice behaviour in the choice chamber trials: (1) the total number of hops (square root transformed to approach normality) and (2) the deviation from a random time allocation among males, as a measure of the strength of the preference for a particular male (for details of these measures, see Forstmeier, 2005).

To study fitness consequences, groups of zebra finches were allowed to breed in aviaries. A subset of birds from the F1 generation were allowed to breed for periods of 3 months, in nine aviaries during two breeding rounds (2005 and 2006). Aviaries contained six males and six females, but three aviaries held an additional three females (sex ratio 0.4), and another three held an additional three males (sex ratio 0.6). Pairs were exchanged among aviaries and sex-ratio treatments between years. We replaced dead birds so that in total 139 birds from the F1 generation were used. Birds from the F2 generation were allowed to breed in six aviaries in 2007 (for details, see Martin, 2008). These aviaries had an even sex ratio of six males and six females and 74 birds were used in total. The paternity of eggs or offspring in both studies was determined using ten microsatellite loci (Forstmeier et al., 2007) and assigned to parents as described elsewhere (H. Schielzeth & E. Bolund, submitted). Calculations are based on 2087 eggs. We measured male success as the total number of eggs fertilized and female success as the total number of eggs laid and referred to these measures as ‘fitness’. For analyses, we used relative success within the aviary (this controls for the effect of the sex ratio treatment because the success of each individual is adjusted to the average success of birds of that sex in the aviary). For birds that participated in two breeding seasons (2005 and 2006), the average success was used. Male attractiveness to females in the aviaries was measured by scoring female responsiveness to males in courtships (N = 100 males, 6537 courtships). This was based on 280 h of early morning observation (using video surveillance) from the 2006 breeding season and 1827 h of observation (video coverage of the complete first 3 weeks) from the 2007 season.

In the F1 generation, we measured survival over the first 1600 days of life (at this point, 146 of 307 birds had died). In the group that had died, we looked at the relationship between mass at day 8 and life span.

Data preparation

For several of the traits, multiple measurements per individual were available. There are several approaches to deal with multiple measurements during the statistical analyses. To illustrate, consider our measurement of male attractiveness in aviaries. Each male participated in a number of courtships (range 2–495 courtships per male) and each courtship was directed at one of the 6–9 females in the aviary. To convert this information into an estimate of male attractiveness (e.g. proportion of successful courtships), four options present themselves. We can (1) standardize the sample size (and use a set number of courtships per male). This would necessitate the rejection of individuals with very few courtships and would discard a large amount of data for individuals with many courtships. Alternatively, we can (2) take the overall average courtship success. This introduces the problem of unequal sample sizes across individuals, which leads to decreasing variance of the mean with increasing sample size. This means that low sample data points will have a greatly inflated variance. The third option (3) is to use best linear unbiased predictors (BLUPs) from a linear mixed effect model as a means of averaging across unequal sample sizes. In contrast to simple averaging, this method will lead to increasing variance with increasing sample size, because low sample data points are treated more conservatively and are pulled towards the population mean. BLUPs have an overall slightly lower variance than averages, and the change in variance with sample size is more dramatic for averages than for BLUPs (E. Bolund, W. Forstmeier, H. Schielzeth, unpublished). BLUPs allow fixed effects to be controlled for. In the current example, female identity can be added as a fixed effect to control for the differences in responsiveness between females. A fourth option bypasses the need to obtain one measure per individual, by entering all available data into the animal model. A permanent environment effect is added to control for repeated measures on the same individual [this estimates a permanent environment matrix in addition to the G (genetic) and R (residual) matrices], and fixed effects can be added. Although this seems the most preferable method, it brings a computational complexity of the animal model beyond that of the other options (with option 1–3, only the G and R matrices need to be estimated and fixed effects are only rarely necessary). In our case, method 4 frequently leads to problems of model convergence in VCE, and we therefore decided to use options 1–3 for traits with multiple measurements.

We used the first measure of the life time for female choice behaviour in the choice chamber and for the three beak colour measures. Because birds were between 74 and 140 days old at the day 100 measures and beak colour continues to change during this period, we accounted for the precise age at measure for the day 100 measures by including age as a fixed effect in the animal model. Remaining traits measured at day 100 showed no influence of the precise age at measure, thus no age correction was applied.

We used a lifetime average of male attractiveness in the choice chamber (measured 4.1 ± 0.8 times, mean ± SD). Male attractiveness was transformed according to arcsine (y1/3) to approach normality. Adult mass was also averaged over the life time (measured 2.5 ± 2.3 times, mean ± SD). Using only the mass at approximately day 100 (and body condition calculated based on this mass) did not alter the results.

Finally, we used BLUPs of male song rate (5.8 ± 2.0 measures, mean ± SD) and female responsiveness (5.3 ± 2.6 measures, mean ± SD). BLUPs were extracted from an lmer controlling for measuring batch and mixed sex rearing (birds were reared from day 35 until around day 100 either in same sex groups or in groups with both sexes present) as fixed effects. We also use this method for male extra-pair attractiveness in aviaries and accounted for the following fixed effects: observer identity, day of the experiment, time of the day, type of courtship (within- or extra-pair), fertility status of the female, male identity and female identity.

Statistics

We used SPSS (SPSS for Windows, Rel. 15.0.1.; SPSS Inc., Chicago, IL, USA), R 2.7.0 (R Development Core Team; R Foundation for Statistical Computing, Vienna, Austria) and REML VCE 6.0.2 (Groeneveld et al., 2008) for statistical analyses. All statistical tests are two-tailed. To obtain random effect estimates of traits for individuals, we used the lmer function from the lme4-package in R 2.7.0 (Bates et al., 2008).

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. References

Environmental correlations

The very pronounced environmental differences in early growth had moderate effects on tarsus length (Fig. 1a, note that the phenotypic values of traits are used for illustration), wing length, body mass and fat storage over the life-time, in both sexes, whereas FA of the tarsus was only weakly affected (Table 2). Beak colour was affected by early growth in both sexes, more strongly in females than in males. Individuals that were environmentally heavier as nestlings had more male-coloured beaks as adults (Fig. 1b). Effects on the remaining traits were extremely weak and nonsignificant, with the exception of fitness (eggs fertilized and eggs laid, Table 2). There was a trend for the environmental component of massD8 to be positively correlated with our measure of fitness (total fertilization success of eggs) under aviary conditions in males, whereas the trend in females was very weak (Fig. 2a). There was no effect on longevity in either sex (Table 2, Fig. 2b).

image

Figure 1.  (a) Phenotypic correlation between mass at day 8 and adult tarsus length in males (filled triangles, solid regression line) and females (open squares, dashed regression line). (b) The relationship between mass at day 8 and adult beak colour in males and females. Beak colour was measured with photospectrometry and is expressed on an axis from female to male like beak colour.

Download figure to PowerPoint

image

Figure 2.  Phenotypic correlation between mass at day 8 and ‘fitness’ (a) and longevity (b). Fitness under aviary conditions was significantly related to mass at day 8 in males (number of eggs fertilized, filled triangles, solid regression line), but not in females (number of eggs laid, open squares, dashed regression line). Raw counts of egg numbers are shown. Counts for individuals that participated in two breeding rounds (F1 generation) are represented by one average value. Longevity during the first 1600 days after hatch was not related to mass at day 8 in either sex.

Download figure to PowerPoint

Genetic correlations

The genetic correlations between massD8 and adult traits showed largely the same pattern as the environmental correlations (Table 2). However, because of the low heritability of massD8 and similarly low heritabilities of several of the adult traits, these estimates are rather imprecise, as is evident from the large standard errors. Thus, the genetic correlations between massD8 and adult traits should be interpreted with caution. Unexpectedly, the genetic component of massD8 was significantly negatively related to fitness in females (Table 2).

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. References

We made use of the considerable variation in early growth present in our population, where some of the birds are severely undernourished (Fig. 1), but still survive to adulthood. Remarkably, these extreme differences in early growth had very modest lasting effects on a broad range of traits and behaviours in adulthood. We also found no evidence of life-history trade-offs. It seems that zebra finches can largely compensate for stressful early growth conditions without carrying much long-term costs.

Effects of developmental stress

We found a relatively strong effect of early growth conditions on size, mass and fat deposition with a mean r of 0.40. This is in line with findings from other studies (Table 1). Zebra finches, like most altricial bird species, grow to full structural size in the nest before fledging (Ricklefs, 1969), which could explain these lasting effects, whereas traits that are developed mainly after independence (e.g. song structure, plumage ornaments) can develop normally when environmental conditions are more favourable.

There was a weak effect of early growth conditions on fertilization success under free-flying aviary conditions, primarily in males. It seems possible that this effect is partly mediated by the size differences between birds, because we have indications that larger males are more successful under these conditions (Bolund et al., 2007; E. Bolund, H. Schielzeth & W. Forstmeier, in preparation).

Interestingly, early growth conditions showed lasting effects on beak colour in both sexes. The effect tended to be larger in females than in males. Thus, beak colour did not show heightened condition dependence in males when compared to females. The beak colour is a dynamic trait that is dependent on continuous new input of carotenoids into the beak. Hence, our results indicate that the carotenoid metabolism may be primed by early growth experiences. Carotenoids are transported in the plasma exclusively by lipoproteins (Parker, 1996). Thus, it seems possible that the decreased beak colour is connected to the effects on fat deposition that we found. Studies on zebra finches have shown that testosterone is involved in the regulation of lipoprotein levels in the bloodstream (McGraw et al., 2006), thus providing a possible mechanism that could regulate both fat and carotenoid metabolism. These effects may be indicative of metabolic programming and are in line with studies that have found life-long effects of early nutrition regardless of adult diet (the so-called ‘thrifty phenotype’, e.g. Monaghan, 2008; Barrett et al., 2009).

We found no effect of developmental stress either on other traits that are commonly considered as ornaments, namely plumage traits, song structure and song rate, or on male attractiveness to females and on female mate choice behaviours. Male attractiveness was measured both in a choice chamber set-up and in a socially complex setting in aviaries. Contrary to previous claims (de Kogel & Prijs, 1996; Spencer et al., 2005), this strongly suggests that stressful early growth conditions do not affect traits that are important for female choice.

A survey of previous studies on zebra finches shows that effects on morphology usually persist into adulthood, whereas effects on other traits are rarely consistent among studies (Table 1). Note that only traits that were also investigated in the current study are included in the table. Several of the studies looked at additional traits not shown here. Still, only few studies have looked simultaneously at a broad range of traits. Part of the reason why our results contrast with those of earlier studies might be a publication bias in favour of positive effects. We note that comparisons between studies would be greatly facilitated if effect sizes would be routinely reported (see Nakagawa & Cuthill, 2007) because this was carried out in only one of the 15 studies compiled in Table 1.

The genetic basis of condition dependence

In this study, we looked separately at the genetic and the environmental correlation between massD8 and adult traits. This allows us to look at our findings in the light of the genic capture hypothesis (Rowe & Houle, 1996; Tomkins et al., 2004). This hypothesis rests on two premises: that the ability of a male to develop exaggerated secondary sexual traits is dependent on his overall condition, and that condition has a high genetic variance. If this is so, female choice for honest indicators of quality will result in the evolution of a genetic covariance between condition and trait. Thus, the trait ‘captures’ the genetic variance in condition, which is because of the cumulative good-gene effects of alleles at many loci. However, how to define and measure condition is not straightforward. If we use our measure of early development (massD8) to estimate condition, the genic capture hypothesis predicts a positive genetic correlation between massD8 and both ornaments and fitness. Contrary to this expectation, on average, both of these correlations tended to be negative rather than positive in the present study. Hence, ornaments depended little on genes that enable fast early growth, and those genes had little positive effects on fitness. Thus, genes for a high massD8 were selectively neutral with regard to fitness consequences. This may not be surprising, because there is no strong a priori reason to assume that genes for fast early growth would be ‘good genes’. If anything, genes for fast growth resulted in lower female fecundity (although this may represent a type I error). In general, the genetic basis of condition dependence remains poorly understood. A promising venue for future studies would be to focus on gene by environment interactions on early growth (see Cotton et al., 2004).

A life-history perspective

Poor early environmental conditions can be compensated for by accelerated growth when conditions improve (reviewed in Arendt, 1997). In our study, as in most laboratory studies, food was available ad libitum after independence from the parents at day 35 (Table 1). Compensatory growth could be costly, leading to life history trade-offs (Metcalfe & Monaghan, 2001; Lindström et al., 2005). We would then expect to find effects on longevity. However, longevity was not affected in this study. Considered together with the very moderate effect on fitness, this indicates that individuals seem to be able to compensate for poor growth conditions without paying much long-term costs. Thus, it seems possible that the lab environment might offer ample opportunities to fully compensate in the development of most traits. We can only speculate about the effects of longer periods of stress. It seems likely that extended nutritional stress during the entire period until sexual maturity would have more pronounced effects on development.

Surprisingly high levels of developmental plasticity have also been found in several precocial species and species with unpredictable food supply. For example, young chicken (Gallus gallus) can be maintained on a low protein diet at a body mass corresponding to a 10-day-old chick for periods of several weeks, and return to normal growth when food quality is normalized (McRoberts, 1965). Similar results have been obtained in other galliform and anseriform birds (reviewed in Starck & Ricklefs, 1998). However, these studies rarely looked at effects on adult traits related to fitness.

Of the two previous studies that have looked at effects on longevity in zebra finches (Table 2), one study found a strong effect on survival until day 500 in males (Birkhead et al., 1999) whereas the other found a weak effect on survival during the first year of life in females, but not in males (de Kogel, 1997). Both these studies also found a significant effect on tarsus in adulthood. Thus, it is possible that the effect on longevity might be mediated by differences in size, if size affects dominance relationships (Bolund et al., 2007) or food access. This would especially apply under relatively crowed aviary conditions, such as in the Birkhead et al. (1999) study. In the wild situation, where crowding is not a concern, this effect might be absent. In general, predictions about the costs and benefits of relative size in the wild are difficult to make and would be an interesting venue for future studies.

Exploring the alternatives

Our results fit well with a recent review showing that there is surprisingly little evidence for heightened condition dependence of sexually selected traits (Cotton et al., 2004). This review emphasized the importance of comparing sexually selected traits to nonsexually selected traits (or the analogous trait in the opposite sex), because sexually selected traits are expected to show heightened condition dependence. Although we found that the environmental conditions during growth had effects on traits commonly used as indices of condition (body size, mass, fat deposition), we found a weak effect on only one presumably sexually selected trait (beak colour). However, in contrast to the expectation, the effect was weaker in males than in females. Other presumably sexually selected traits, such as song and plumage ornaments, were not affected by developmental stress. These results are corroborated by the absence of an effect of growth conditions on male overall attractiveness to females. Thus, our study discounts condition dependence with respect to the early development as a mechanism to maintain the honesty of these traits as signals. However, the possibility remains that honesty could be maintained by other mechanisms, such as current condition dependence or a badge of status system (where honesty is maintained by social costs). However, in our population, beak colour and song rate are not important in female choice (Forstmeier & Birkhead, 2004; Forstmeier, 2007) and might thus not function as honest indicators of male quality at all.

Conclusions

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. References

In conclusion, our results show that zebra finches are remarkably resilient to early developmental stress. We found lasting effects mainly on traits that develop during the actual period of stress, whereas traits developing later were not affected, with the exception of beak colour. Thus, it seems that zebra finches are largely able to compensate for early environmental stress, with very little or no costs in terms of fitness consequences or longevity. That presumably sexually selected traits are not dependent on past condition indicates that their signalling value is maintained through other mechanisms if they function as signals of quality at all. Alternatively, they might serve other signalling functions.

Acknowledgments

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. References

We thank Bart Kempenaers for providing facilities and various other support. Xavier Baecke and Stefan van Dongen contributed the FA data (research project G.0025.07 from the FWO Flanders). James Dale and two anonymous referees provided helpful suggestions that improved the manuscript. We thank Melanie Schneider for performing molecular work. Our gratitude also goes to our animal care takers: Sonja Bauer, Edith Bodendörfer, Annemarie Grötsch, Johann Hacker, Markus Lehr, Jenny Minshull, Petra Neubauer, Frances Preiniger, Magnus Ruhdorfer and Agnes Türk. Funding was provided by the German Science Foundation (DFG) through an Emmy Noether Fellowship to W.F. (FO 340/1-2 and FO 340/1-3).

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. References