Relationship between fluctuating asymmetry and fitness within and between stressed and unstressed populations of the wolf spider Pirata piraticus


Frederik Hendrickx, Department Biology, Unit of Animal Ecology, Ghent University, K.L. Ledeganckstraat 35, 9000 Gent, Belgium.
Tel.: ++32/9/264.52.56; fax: ++32/9/264.87.94;


Although developmental instability, measured as fluctuating asymmetry (FA), is expected to be positively related to stress and negatively to fitness, empirical evidence is often lacking or contradictory when patterns are compared at the population level. We demonstrate that two important properties of stressed populations may mask such relationships: (i) a stronger relationship between FA and fitness, resulting in stronger selection against low quality (i.e. developmental unstable) individuals and (ii) the evolution of adaptive responses to environmental stress. In an earlier study, we found female wolf spiders Pirata piraticus from metal exposed populations to be characterized by both reduced clutch masses and increased egg sizes, the latter indicating an adaptive response to stress. By studying the relationship between these two fitness related traits and levels of FA at individual level, we here show a significant negative correlation between FA and clutch mass in metal stressed populations but not in unstressed reference populations. As a result, levels of population FA may be biased downward under stressful conditions because of the selective removal of developmentally unstable (low quality) individuals. We further show that females that produced larger eggs in stressed populations exhibited lower individual FA levels. Such interaction between individual FA and fitness with stress may confound the effect of metal stress on FA, resulting in an absence of relationships between FA, fitness and stress at the population level.


Fluctuating asymmetry (FA) (Van Valen, 1962), being random differences in the development of both sides of a bilateral symmetrical character (Palmer & Strobeck, 1986), has been proposed as an indicator of environmental as well as genetic stress (Leary & Allendorf, 1989; Clarke, 1992; Parsons, 1992; Markow, 1995). The concept underlying FA is that random right–left (R–L) differences reflect the level of developmental instability (DI) of an organism, i.e. the inability to buffer its development against random developmental perturbations (Palmer & Strobeck, 1992). Increased stress generally causes increased energy expenditure by organisms (Bengtsson et al., 1985; Sibly & Calow, 1989; Baillieul et al., 1996). As this reduces resources for homeostasis because of energetic constraints, stress may result in increased levels of developmental instability (DI) (Clarke, 1993; Sommer, 1996). FA is, therefore, believed to positively relate to stress and negatively to quality or fitness (reviewed by Leung & Forbes, 1997; Møller, 1997b, 1999). Other studies, however, failed to find such correlates (e.g. Clarke, 1998; Björksten et al., 2000, 2001; Chapman & Goulson, 2000; Vollestad & Hindar, 2001; Woods et al., 2002). Similar inconsistencies are found in relation to stress caused by exposure to ecotoxic substances, with some papers reporting higher levels of FA for pollution exposed populations (Valentine et al., 1972; Pankakoski et al., 1992; Clarke, 1993; Hardersen, 2000) while others fail to do so (e.g. Rabitsch, 1997; Polak et al., 2002). The deteriorating effect of toxicant exposure on fitness is, however, widely established (Walker et al., 2001).

As reviewed by Lens et al. (2002a), main factors that may be responsible for the confounding relationship between FA, stress and fitness are (i) problems in the estimation of DI through FA, (ii) problems in assessing what environmental factors cause stress and thus affect fitness and (iii) interactive effects of genotype and environment on FA. The latter might be especially important if populations differ in their level of adaptation to deteriorating effects of potential stressors, causing decreased levels of susceptibility and hence increased levels of developmental stability (DS). In addition to this confounding effect on the relationship between FA and stress, adaptation can further obscure the FA–fitness relationship if the allocation of fitness (life history) components is altered along trade-off curves. As fitness is usually measured indirectly, its estimate is based on values of specific life history traits which obsures the between-population relationship between FA and fitness even further. Although such adaptations might be present at population level, this does not necessarily hold if individuals are compared within a population. Individual-level and population-level relationships between stress and fitness may indeed be based on different processes (Campbell et al., 1998). Within a population, individuals may differ strongly in their degree of resource acquisition and adaptation (van Noordwijk & de Jong, 1986). If one assumes that FA is a fundamental property of an organism reflecting its level of adaptation, relationships between FA and life history traits can be expected to be more pronounced at individual level than at population level. Strength and direction of this relationship might thereby differ between populations, depending on the selective forces affecting these traits. However, stress in itself might additionally strengthen the relationship between FA and fitness if reduced resource acquisition because of stress defence decreases homeostasis and individual quality (Leung & Forbes, 1997; Lens et al., 2002a,b). Consequently, if the strength of relationships between DI and individual quality increases with increasing stress levels, low quality (i.e. developmental unstable) individuals can be expected to become eliminated in stressed populations, leading to counterintuitive results such as lower FA levels under higher stress (developmental selection hypothesis sensu Møller, 1997a).

In this study, we measured levels of FA in six populations of the wolf spider Pirata piraticus (Clerck, 1757) that vary in degree of metal exposure. In a previous study, life history patterns indicative for low growth and/or low reproductive environments were observed in the most contaminated populations (Hendrickx et al., 2003a). This was reflected by a severe reduction in clutch mass and, hence, female fecundity. Simultaneously increases in egg size, expected to occur under restricted resource acquisition, suggested an adaptive stress response. Life history theory predicts that optimal egg size depends on a balance between producing large offspring, which in general survive better, mature earlier and develop faster (Einum & Fleming, 1999; Fox & Czesak, 2000) and the production of smaller eggs, which results in a higher fecundity (Stearns, 1992; Bernardo, 1996; Roff, 2002). If offspring fitness is reduced for a given offspring size, this will result in selection towards the production of fewer, but larger offspring (Smith & Fretwell, 1974; Parker & Begon, 1986; Lloyd, 1987; McGinley et al., 1987). Theoretical as well as empirical results provided evidence that fitness reduction, because of a reduction in reproductive output, can be compensated for by increasing offspring size (Sibly et al., 1988; Tamate & Maekawa, 2000).

In the present study, we relate levels of FA with levels of reproductive output and egg size, between and within populations that differ in degree of metal exposure. We thereby address the following questions:

  • (i) Is there a negative relationship between individual FA and resources devoted to reproduction and is this relationship more pronounced in stressed populations?
  • (ii) Is there a negative relationship between individual FA and egg size and is this relationship only present in stressed populations (i.e. populations selected to produce larger eggs)?
  • (iii) Are the relationships between FA and stress and both fitness related traits absent at the population level?

Studies comparing the relationship between DI, fitness and stress, both within and between populations are still largely lacking. However, we demonstrate that such an approach gives strong insight in the processes underlying FA and, as such may contribute to disentangle the confounding results so often observed in studies of FA.

Material and methods

Origin of the populations

Between 26 and 36 adult females of the wolf spider P. piraticus, carrying an egg cocoon at the end of their spinnerets, were sampled at six different localities that differed in degree of metal contamination. All populations were sampled at the third and fourth of June 1999 and only adult females were sampled, as sex, developmental stage or season might influence the degree of FA (Hardersen, 2000). Populations SA, GS, GW and KR were all located on the banks of the river Schelde, a tidal river known to be severely polluted by heavy metals during the last century (Zwolsman et al., 1996). Two inland populations DV and MO showed lower levels of metal contamination (Hendrickx et al., 2003a). All spiders were kept individually and killed and stored in a freezer at −10 °C in the laboratory.

Heavy metal determination

Between six and eight adult females from each population were separated from their egg cocoon and, after determining female size, clutch mass and egg size, cadmium, lead, copper and zinc content were determined for each individual spider. Spiders were first rinsed in a 1% HNO3 solution to remove metals attached to the hairy surface of the spiders. Before digestion, spiders were dried for 48 h at 70 °C and weighted afterwards on a Mettler Toledo AT 21 Comparator analytical balance (Namicon, Switzerland) to the nearest 0.01 mg. Spiders were digested according to method four described in (Tack et al., 2000). Animals were immersed in 5 mL ultra-pure 65% HNO3 solution at 130 °C. After 1 h, 2 mL of 20% H2O2 was added to the solution. After 30 min another 2 mL of 20% H2O2 was added to complete digestion. Metal concentrations were determined by means of flame atomic absorption spectrometry (AAS) (SpectrAA-10; Varian, Palo-Alto, CA, USA) for Cu and Zn and on a graphite furnace AAS (SpectrAA-100; Varian), equipped with Zeeman background correction for Cd. All metal concentrations are expressed as μg metal g−1 dry weight spider. Quality control of the metal analysis was carried out by analysing reference materials Bovine muscle (CRM 184) and Mussel tissue (CRM 278R).

FA and life history measurements

Three metric traits, i.e. distance from the distal edge of the femur to the proximal dorsal spine (SpI), distance from the distal edge of the femur to the distal dorsal spine (SpII) and tibia length (Ti), were measured at both sides on the first pair of legs (Fig. 1). As argued by (Whitlock, 1996; Whitlock, 1998), each value of FA represents only a single sample of the DI of an individual, leading to only weak correlations between FA values within individuals or even populations. Taking more than one trait into account is therefore strongly recommended to increase the reliability of the population FA estimates (Leary & Allendorf, 1989; Leung et al., 2000).

Figure 1.

Correlation between population FA estimates of the two traits ( VFA SpI and VFA SpII = variance of signed FA of distance from the distal edge of the femur to the proximal and distal dorsal spine, respectively).

Measurements were carried out with a Wild M3 stereomicroscope (Heerburg, Switzerland) combined with a digitalising table to the nearest 0.01 mm. Because differences between the left and right side of a bilateral character are often very small, they might be strongly biased with measurement error. Repeated measures are therefore needed to obtain unbiased estimates of FA (Palmer & Strobeck, 1986; Palmer, 1994). Repeated measurements were taken several days apart. For the measurement of the life history traits clutch mass and egg size, we refer to Hendrickx et al. (2003a).

Estimating metal contamination of the populations

Estimates of the degree of metal contamination were obtained by subjecting individual Cd, Zn, Cu and Pb concentrations to a principal component analysis (PCA). The average score of the individuals of each population along PCA-axis 1, which was significantly correlated with Cd, Zn and Cu, was used to estimate the degree of metal contamination (see Hendrickx et al., 2003a for details). Differences in metal concentration between reference and contaminated populations were tested by mixed model analysis (SAS v. 6.12) with PCA 1 score as the dependent variable, factor contamination (reference or contaminated) as fixed effect and population as random effect, nested within contamination level. Significance of random effects was tested by likelihood ratio (LR) tests.

FA analysis

Before interpreting FA population estimates, several statistical assumptions have to be met that can be divided into three groups, being (i) tests for the detection of directional asymmetry (DA) and antisymmetry (AS), (ii) tests for the independence of the three measured traits as they were all measured on the same pair of legs and (iii) tests of measurement error relative to FA.

True FA is characterized by a normal distribution with zero mean of the signed R–L values. In the case of DA, the distribution is skewed or has a mean different from zero, whereas AS is characterized by a bimodal or platykurtic distribution with zero mean (Palmer & Strobeck, 1992). The latter two asymmetry types are presumed to have a heritable component and are therefore considered not to reflect DI (Palmer & Strobeck, 1992). However, recent empirical as well as theoretical studies suggest that developmental perturbations might cause transitions from FA to other asymmetry types (Graham et al., 1998; Lens & Van Dongen, 2000). Additionally, among-individual heterogeneity in DI might lead to a leptokurtic distribution, making standard tests for normality (e.g. Shapiro-Wilks W) inappropriate (Leung & Forbes, 1997; Van Dongen et al., 1999). Recently the use of mixture analysis has been introduced to decompose a frequency distribution into its k normal distributions (N) with means μi, variances σi2 and proportions qi (see Van Dongen et al., 1999 for details). Estimates of these parameters can be obtained by maximum likelihood, such that the observed frequency distribution is approximated by a weighted sum of normal distributions:


Following (Palmer & Strobeck, 1992), FA was modelled as a normal distribution with zero mean, DA as a normal distribution with a mean different from zero, and AS as two normal distributions with equal density, equal variance and equal absolute nonzero means with opposite sign. Mixture analysis was performed on the frequency distribution of each trait separately, pooled over all populations. To select the model that fitted best to the observed distribution, we first determined the number of components that described the asymmetry data. By use of LR tests, each k component model was tested against a k − 1 component model, until adding an additional component did not result in a significant increase of the maximum likelihood. After determining the number of components, the same procedure was used to test which components had a mean different from zero or a distribution typical for AS. Mixture analysis and model selection were performed with Mixture 1.0 (Van Dongen et al., 1999).

As mentioned before, all three traits were measured on the same pair of legs, and might therefore be highly redundant. To test for correlated responses in signed FA between traits, a Pearson correlation was performed on the signed R–L values of all individuals for all trait combinations.

To test for the significance of FA relative to measurement error (ME), a two-way mixed model anova (side × individual) was performed for each population and trait combination with factor individual treated as a random effect. A significant side × individual interaction reveals the significance of FA relative to ME (Palmer & Strobeck, 1986).

Individual FA values were calculated by summing the individual unsigned R–L values of the traits that met the aforementioned assumptions. Prior to this, R–L values were divided by the average trait FA value to correct for differences in magnitude of trait FA (Leung et al., 2000). This composite FA index (CFA) was used as an individual estimate of DI.

Unbiased population FA values for each trait were obtained by computing variance components of the side × individual interaction (VFA = MSinteraction–MSerror/number of repeated measurements) of the above mentioned two-way mixed model anova. The consistency of the population FA estimates between traits was tested with a Pearson product correlation. Afterwards, a population CFA estimate was calculated as the average of the individual CFA values for each population.

Life history analysis

To obtain the relative clutch mass (RCM), we corrected for maternal size effects by calculating residual from a regression of clutch mass on female cephalothorax width3. Cephalothorax width was cubed to obtain a volumetric measure of female size. Both traits were log10 transformed to adjust for allometric relationships (Roff, 1992). As the slopes of the regression lines relating clutch mass and female size did not statistically differ between the populations (F5,164 = 1.06; P = 0.38), data were pooled. As egg size is independent of female size in this species (Hendrickx & Maelfait, in press), untransformed values were used. For details on the analysis of life history traits clutch mass and egg size at population level, we refer to (Hendrickx et al., 2003a).

Tests for the relationship between FA and life history traits

Relationships between individual CFA, clutch mass and egg size, in contaminated populations (KR, GW, GS and SA) and reference populations (DA and MO) were examined with mixed models in Proc Mixed (SAS v6.12). Factor ‘population’, and all relevant interaction terms were treated as random effects nested within the fixed factor ‘contamination’. Clutch mass and egg size were used as covariates. A Spearman Rank Order correlation was carried out to test for the relationship between population CFA levels and (i) internal metal load and (ii) average life history traits between the populations.


Metal contamination

The contaminated populations KR, GW, GS and SA, located along the river Schelde, had significantly higher metal body burdens (PCA1 values) compared with the reference populations DA and MO (F1,4 = 10.44; P < 0.05) (Table 1). A significant amount of variance could, however, be attributed to differences within each level of contamination (σ2 = 0.52; LR test; P < 0.05).

Table 1.  Metal concentrations (mean concentration ± SD) of adult female Pirata piraticus sampled at six different localities.
PopulationnCd (μg g−1)Pb (μg g−1)Cu (μg g−1)Zn (μg g−1)PCA 1 score
  1. Metal concentrations are expressed in μg g−1 dry weight.
    n = sample size, PCA 1 score = the average score of the individuals scores on the first axis of a PCA analysis conducted on the four metal concentrations of all individuals.

Reference populations
 DV79.9 ± 5.2271.8 ± 46.1176.3 ± 9.80595.5 ± 186.22−1.64 ± 0.102
 MO74.8 ± 2.25178.1 ± 98.12146.4 ± 39.85654.4 ± 284.53−1.34 ± 0.231
Contaminated populations
 KR821.9 ± 15.2581.3 ± 36.59196.5 ± 97.75764.2 ± 188.40−0.34 ± 0.378
 GW831.7 ± 6.33218.7 ± 148.10258.6 ± 79.451128.8 ± 323.990.81 ± 0.322
 GS657.6 ± 26.4685.3 ± 38.39375.8 ± 109.01855.7 ± 157.231.91 ± 0.608
 SA922.7 ± 4.98109.4 ± 63.94281.7 ± 39.831070.6 ± 369.310.62 ± 0.223

FA analysis

Signed R–L values for the three traits were distributed according to a one- or two-component normal distribution with zero mean (Table 2). For SpI and SpII, adding a second component did not result in a significant increase in model fit (P = n.s.). For Ti, adding a second component resulted in a significant increase in model fit (P < 0.05), while adding a third component did not (P = n.s.). For none of the aforementioned components, including a mean different from zero (i.e. the average sample mean) resulted in a significant increase in model fit (P = n.s.). Finally, adding a bimodal distribution also did not lead to a significant increase in model fit (P = n.s.). Asymmetry measures for all three traits were, therefore, interpreted as true FA.

Table 2.  Results of mixture analysis for the traits SpI, SpII and Ti.
 nProportion of individualsSelected multicomponent model
  1. n  = sample size; FA, DA and AS = proportion of individuals with fluctuating asymmetry (FA), directional asymmetry (DA) and antisymmetry (AS); selected multicomponent model =  inline image , see text for details.

SpI1641001.00 × (0, 0.00137)
SpII1641001.00 × (0, 0.00140)
Ti1641000.95 × (0, 0.00016) + 0.05 × (0, 0.00348)

The level of ME varied strongly between the different traits (Table 3). For SpI and SpII, ME was relatively low and ranged from 2.6 to 21.8%. As a result, highly significant FA estimates were obtained for each population. For Ti, measurement error was much higher (14.2–69.5%), resulting in nonsignificant FA estimates for the populations DA and MO. Because the population FA estimates are therefore unreliable for Ti, this trait was not considered for further analysis. The correlation between the individual signed FA values between SpI and SpII was not significant (r = 0.009; P = n.s.), suggesting that the development of both traits was independent, despite their close location on the same segment of the same pair of legs.

Table 3.  Results of FA analysis for all population-trait combinations.
TraitPopulationnVME  × 10 3MSinter × 103% ErrorPinterVFA  × 10 3
  1. n , sample size; VME , variance in measurement error; MS inter , mean squares side × individual interaction; % error, % of variance in the side × individual interaction because of measurement error; Pinter , significance of the side × individual interaction; VFA , variance in signed FA.


FA levels for SpI varied from 0.310 in the most contaminated population GS to more than twice this value (0.705) in the moderately contaminated population KR (Table 3). Both reference populations DV and MO had relatively high levels of FA. Unsigned population FA values of both traits were significantly correlated (r = 0.84; P < 0.05; Fig. 1), indicating consistent population FA patterns for both traits. Differences in CFA were significant among populations (Bartlett's test χ2 = 11.47; P < 0.05).

Relationships within populations

Within each level of contamination, slopes of the regression lines relating CFA and RCM did not significantly differ between populations (Table 4; Fig. 2). Removing the respective nonsignificant random two-way interactions from the model revealed significance of the fixed RCM × contamination interaction (Table 4): levels of RCM and CFA were inversely related at contaminated sites (r = −0.26; P < 0.05) while no such relationship occurred at the reference sites (r = 0.23; P = 0.1).

Table 4.  Results of mixed model analysis to test for fixed and random effects on individual estimates of CFA.
CovariateSourceF / s2P -value
Relative clutch mass (RCM)Fixed effects
ContaminationF  = 1.23 0.33
RCMF  = 0.37 0.54
RCM × contaminationF  = 7.75 0.006
Random effects
Populations2  = 0.00 n.s.
RCM × populations2  = 828 n.s.
Egg sizeFixed effects
ContaminationF  = 3.71 0.13
Egg sizeF  = 0.00 0.97
Egg size × contaminationF  = 3.51 0.06
Random effects
Populations2  = 5.91 n.s.
Egg size × populations2  = 0.00 n.s.
Figure 2.

Relationship between individual CFA and relative clutch mass (RCM) in reference (DA and MO) and contaminated (KR, GS, GW and SA) populations. Bold line represents the average regression line if all individuals are pooled within the level of contamination.

Likewise, within each level of contamination, slopes of the regression lines relating CFA and egg size did not significantly differ between the populations (Table 4; Fig. 3). Removing the relevant interaction terms resulted in a nearly significant egg size × contamination interaction (Table 4) with r values of 0.23 (P = n.s.) and −0.23 (P = 0.051) for reference and contaminated populations, respectively.

Figure 3.

Relationship between individual CFA and egg size in reference (DA and MO) and contaminated (KR, GS, GW and SA) populations. Bold line represents the average regression line if all individuals are pooled within the level of contamination.

Relationships between populations

At population level, CFA estimates and average metal concentration were not significantly correlated (Table 5). Likewise, relationships between population CFA and levels of clutch mass, fecundity, female mass and egg size, were not significant.

Table 5.  Correlations between average population CFA and the degree of metal contamination and the life history traits relative clutch mass (RCM) and egg size ( rs  = Spearman rank order correlation coefficient and associated probability P ).
 rsP -value
PCA 1−0.710.11
Egg size−0.770.07


Our results confirm that the magnitude and direction of the relationship between individual DI and fitness related traits can depend on the amount of stress experienced at population level. Such pattern may confound DI–stress relationships at the population level.

Individual level patterns

FA and resources devoted to reproduction (i.e. clutch mass) were inversely related in populations subjected to high levels of metal pollution, but not so in the reference populations. In a previous study, we showed a significant reduction of reproductive output in exposed populations (Hendrickx et al., 2003a). As the study species has a semelparous life cycle (Hendrickx & Maelfait, in press), such reduced reproductive output suggests a limitation of the resources that can be devoted to reproduction (Stearns, 1992). If energy is limiting, because a large amount is allocated to stress defence (e.g. heavy metal detoxification by storing in midgut diverticulae), it will be limiting for both reproduction and DS and, as a consequence, both measures become correlated. Under such stressful conditions, it is plausible that individuals that have a reduced ability to cope with environmental stress must defend themselves against these harmful influences, which reduces the amount of energy that can be devoted to growth and reproduction (Sibly & Calow, 1989) and, hence, DS. In an environment without such a stress, energetic constraints on reproduction and DS will be absent or of another kind and are therefore unrelated to each other. Under such circumstances, reduced reproduction does not necessarily imply a reduction in DS and vice versa.

Alternative explanations have been put forward for an increase in magnitude in FA–fitness relationships in the presence of environmental stress (Palmer, 1994; Leung & Forbes, 1997; Van Dongen & Lens, 2000; Lens et al., 2002a). First, it is believed that FA more accurately reflects DS under stress. As two opposing processes generate FA, i.e. DS and developmental noise, developmentally unstable individuals may show low levels of FA if they develop under low levels of developmental noise. Second, elevated stress might strengthen the relationship between individual quality and fitness as low quality individuals are more likely to succumb under these circumstances. In a benign environment, observed levels of fitness may partly reflect chance effects and therefore be less accurate estimates of individual quality. For example, Lens et al. (2002b) showed that individual survival and individual FA were negatively correlated in a highly disturbed forest fragment, while such a relationship was absent in an undisturbed fragment.

Concerning the relationship between FA and egg size, nearly significant differences were present between metal stressed and unstressed populations. Individuals from contaminated populations that produced larger eggs tended to show decreased levels of FA while no such trend was observed in reference populations. Such a pattern is in agreement with the prediction that under unfavourable circumstances, fitness advantages of larger offspring will be higher than under favourable conditions (Einum & Fleming, 1999). If so, fitness advantages of producing larger offspring can be expected to outweight a reduction in fecundity, resulting in a shift in the balance between the production of larger, but fewer offspring in a low growth environment (Smith & Fretwell, 1974; McGinley et al., 1987; Sibly et al., 1988; Einum & Fleming, 1999; Tamate & Maekawa, 2000). The weakness of the relationship between FA and egg size observed in this study might be because of several reasons. In addition to the low predictive power of DS through FA, we used female egg size as an estimate of the size of the eggs from which the females originated. Under laboratory conditions, however, egg size variation in this species showed a highly significant heritable component (average h2 for populations DV and GS = 0.49) (F. Hendrickx & J.-P. Maelfait, unpublished data). It is important to note that for our study species, this egg size variation is independent of female size (Hendrickx & Maelfait, in press).

Population-level patterns

No relationship between FA and heavy metal pollution was detected at the level of the population. Given the equivocal relationship between DS and stress reported in the literature (Clarke, 1998), it is important to unravel which factors may explain such a relationship (Lens et al., 2002a). First, a general property of FA is that it only weakly reflects the amount of DI (Whitlock, 1998). Involving multiple traits therefore increases the reliability of population FA estimates (Leung et al., 2000). FA levels of the two traits used in our study were significantly correlated, and it can therefore be assumed that our FA estimates reliably reflect the amount of DI in these six study populations. Second, the level of metal concentration might have been unrelated to the level of stress experienced by the population. However, as spiders are known to accumulate the metals analysed in our study to a great extent (Breymeyer & Odum, 1969; Van Hook & Yates, 1975; Hopkin, 1989; Hendrickx et al., 2003b), internal concentrations can be assumed to accurately reflect the amount of metals populations are exposed to. Indeed, field as well as laboratory observations corroborate the significant negative effect of these metal body burdens on fecundity as well as growth rate (Hendrickx et al., 2003a,b; F. Hendrickx & J.-P. Maelfait, unpublished data).

Third, relationships observed within populations may offer the basis for understanding the apparent lack of association between DI, stress and fitness at population level. The phenotypic correlation between individual FA and RCM in metal stressed populations confirms a stronger association between FA and fitness or individual quality in stressed populations. Such a pattern has previously been reported for bird populations exposed to different levels of habitat disturbance (Lens et al., 2002b). Consequently, low quality individuals (i.e. characterized by reduced stability during development), are more likely to be eliminated from the population. This will result in a downward bias of the population FA estimates of stressed populations. Likewise, Polak et al. (2002) showed lower average FA for Drosophila melanogaster reared in arsenite treated vials with increasing arsenite concentrations. Moreover, FA was positively related to the percentage of emerging flies in arsenite treated vials, but not in control vials. Our study thus provides additional support for the developmental selective hypothesis.

Alternatively, the lack of correlation between FA and stress may be because of adaptive life history patterns evolved to compensate for the reduction in fitness. In the case of metal pollution, such adaptation has been observed through life-history as well as physiological mechanisms (Posthuma et al., 1993; Posthuma & van Straalen, 1993; Shaw, 1999; Shirley & Sibly, 1999). Although marginally significant, the positive relationship between DS and egg size in the metal stressed populations was in the direction expected from such a hypothesis. Increased egg size in stressed populations may reduce stress susceptibility, hence resulting in increased population DS. Laboratory observations revealed that egg size differences between populations are genetically determined (F. Hendrickx, unpublished results). Such genotype-environment interaction additionally underestimates the association between FA and metal exposure when locally adapted populations are compared.

In conclusion, our results showed that the effect of stress on the magnitude and strength of the relationship between FA and fitness related traits within populations reveals the potential difficulty to detect a relationship between FA, stress and fitness when different populations are compared.


The Research Fund of Ghent University as a part of research programme BOF 01110498 supported this work. The authors would like to thank F. Tack and M. Verloo of the Laboratory of Analytical Chemistry and Applied Ecochemistry (Ghent University) for allowing us to conduct metal analysis at their laboratory and B. Sinervo and an anonymous referee for their comments on an earlier draft of this manuscript.