Gábor Herczeg, Department of Biological and Environmental Sciences, Ecological Genetic Research Unit, P.O. Box 65, FI-00014 University of Helsinki, Finland. Tel.: +358 9 191 57807; fax: +358 9 191 57694; e-mail: email@example.com
The proximate and ultimate explanations for behavioural syndromes (correlated behaviours – a population trait) are poorly understood, and the evolution of behavioural types (configuration of behaviours – an individual trait) has been rarely studied. We investigated population divergence in behavioural syndromes and types using individually reared, completely predator- or conspecific-naïve adult nine-spined sticklebacks (Pungitius pungitius) from two marine and two predatory fish free, isolated pond populations. We found little evidence for the existence of behavioural syndromes, but population divergence in behavioural types was profound: individuals from ponds were quicker in feeding, bolder and more aggressive than individuals from marine environments. Our data reject the hypothesis that behavioural syndromes exist as a result of genetic correlations between behavioural traits, and support the contention that different behavioural types can be predominant in populations differing in predation pressure, most probably as a result of repeated independent evolution of separate behavioural traits.
Even though correlations between different behaviours were described some time ago (Huntingford, 1976), it is only recently that the interest has shifted towards studying behaviours in different contexts, and/or different behaviours together (e.g. Verbeek et al., 1994; Dingemanse et al., 2003; Réale & Festa-Bianchet, 2003; Bell, 2005; Dingemanse et al., 2007). Results of these studies indicate that, besides the often considerable amounts of individual plasticity, certain behaviours are not totally context-dependent or independent from each other. In other words, an individual that is more aggressive in competitive situations than others in the population might also be more aggressive towards its own offspring, or bolder towards its predators (e.g. Gosling, 2001). Correlations across contexts and between behaviours within population are often referred to as behavioural syndromes (Sih et al., 2004a,b), temperament (Réale et al., 2007) or, as this phenomenon is in close resemblance to human personality, animal personality (Gosling, 2001). Several studies have supported the link between behavioural syndromes and fitness (for a recent meta-analysis see: Smith & Blumstein, 2008). There is also solid evidence for personality traits like exploratory or coping behaviour to be heritable (e.g. Drent et al., 2003; van Oers et al., 2004, 2005).
There are two popular hypotheses aiming to explain the evolution of behavioural syndromes known as ‘constraint’ and ‘adaptive’ hypotheses, respectively (Bell, 2005). First, as genetically correlated traits tend to evolve together (Price & Langen, 1992; Lynch & Walsh, 1998), behavioural syndromes might be characteristic to a species and present across different populations as a result of the constraints of the underlying genetic architecture. On the contrary, the ‘adaptive’ hypothesis predicts that interpopulation variation in the presence–absence of behavioural syndromes exists as a result of the different selective environments. Testing these hypotheses is not easy because – besides natural selection – stochastic processes (e.g. mutation, drift, founder effects, gene flow) can also cause differences in the correlation structure between populations, while population differences in genetic architecture – and thus in the presence/absence of genetic correlations – cannot be excluded either. Therefore, the ‘constraint’ hypothesis can be rejected if the studied populations lack behavioural correlations, while support for the ‘adaptive’ hypothesis would require detection of predictable environment-dependent patterns to exclude stochastic effects. In cases where population differences are detected in an unpredictable, environment-independent fashion, more detailed genetic studies are needed to distinguish between the hypotheses. Riechert & Hedrick (1993) found that boldness and aggression were correlated in two, markedly different Agelenopsis aperta (Araneae) populations. However, their results could be explained by both the ‘constraint’ or ‘adaptive’ hypotheses. Bell (2005) was able to reject the ‘constraint’ hypothesis by comparing wild-caught adult three-spined sticklebacks (Gasterosteus aculeatus) from two populations, where the population being under heavy predation showed strong behavioural correlations, whereas the other under low predation pressure did not. Similar divergence was found in group-reared first laboratory generation sticklebacks from the same populations (Bell & Stamps, 2004). Brydges et al. (2008) studied wild-caught adult three-spined sticklebacks from eight populations and found behavioural correlations only in one, a result contradicting the ‘constraint’ hypothesis. The only study which has statistically tested if the presence/absence of behavioural syndromes in a population was associated with ecological conditions is that of Dingemanse et al. (2007): they provided support for the ‘adaptive’ hypothesis by comparing wild-caught juvenile three-spined sticklebacks from six predator free and six predation exposed populations. Despite the limited support, different explanations for the adaptive value of behavioural syndromes have already been proposed (Stamps, 2007; Wolf et al., 2007).
Obviously, there are additional possible mechanisms explaining behavioural syndromes. If we accept the facts that (i) behavioural syndromes can be present in one but not in all populations, and (ii) that the pattern corresponds to differences in the environmental conditions among populations (e.g. Bell, 2005; Dingemanse et al., 2007), phenotypic plasticity alone, or genotype * environment interactions can just as well be responsible for the observed differences as genetic differences resulting from local adaptation. Indeed, in the experiment of Bell & Sih (2007), exposure of predator-naïve wild-caught sub-adult three-spine sticklebacks (previously lacking behavioural correlations) to predation resulted in the emergence of behavioural syndromes. This emergence was a result of both the selection imposed by predators as well as the phenotypically plastic response of the prey (Bell & Sih, 2007).
Irrespective of the presence–absence of behavioural correlations, geographic (interpopulation) variation in behaviour is common (Foster, 1999; Foster & Endler, 1999). Many behavioural traits are affected by predation, for instance Magurran & Seghers (1991, 1994) reported that predator inspection, shoaling and aggression all (co)vary with predation pressure in wild guppies (Poecilia reticulata). Single behavioural traits have been found to be heritable in many cases (e.g. Breden et al., 1987; Magurran, 1990; Brown et al., 2007). However, as predation (or the lack of it) is expected to impose complex effects on life history trade-offs (e.g. Blanckenhorn, 2000), it might influence a series of different behaviours at the same time. One way of describing individuals from several behavioural aspects is the adoption of the concept of behavioural type, which is ‘a particular configuration of behaviours that an individual expresses’ (Bell, 2007).
The aim of the present study was to investigate the presence–absence of behavioural syndromes within, and to compare the behavioural types among populations of nine-spined sticklebacks (Pungitius pungitius Linnaeus) differing markedly in predation pressure. The nine-spined stickleback is a perfect model for this purpose, as it occurs in a bewildering range of habitats from seas through large lake or river systems to the smallest creeks and ditches, being able to persist in small, isolated ponds as the only fish species (e.g. Bănărescu & Paepke, 2001; Östlund-Nilsson et al., 2007). We compared four geographically (Fig. 1) and genetically (Shikano, Herczeg & Merilä, unpublished work [Correction added on 23 February 2009, after first online publication: conflation of unpublished work author names corrected]) isolated populations (two marine and two pond, the latter lacking predatory fish) looking for answers to three questions. First, do behavioural syndromes exist when individuals lack prior experience with either predators or conspecifics? Second, if so, does the between population pattern of correlations support the ‘adaptive’ hypothesis? Third, irrespective of the syndromes, do individuals from different populations represent different behavioural types? In order to remove most of the environmental variation, we measured behaviour of laboratory born first generation individuals reared in the absence of any biological interactions (i.e. contact with predators or conspecifics) until they had reached adult size.
Materials and methods
Sampling, breeding and rearing
Adult P. pungitius were collected before or during the early phase of the reproduction period (late May–mid June) in 2007. Four populations were sampled (Fig. 1) with the aid of minnow traps and seine nets. Nine-spined sticklebacks in the two marine populations (Baltic Sea [Finland] and White Sea [Russia]) are sympatric to a large number of predatory and nonpredatory fish, while in the small isolated ponds (Bynästjärnen, Sweden and Pyöreälampi, Finland) it is the only fish species with the exception of a few recently introduced small-bodied whitefish (Coregonus lavaretus) in Pyöreälampi. At any rate, considering the potential prey of whitefish (e.g. Kahilainen et al., 2004), it can only be a competitor of P. pungitius, and as we never caught a single whitefish besides thousands of sticklebacks during our extensive sampling in Pyöreälampi, even that effect should be minimal. The marine sites were shallow coastal bays close to creek inlets, and thus, they represented low salinity sea habitats (Baltic Sea being a brackish water environment in general). The surface area of the ponds was less than 5 ha. Adult fish were transported to the aquaculture facilities of the University of Helsinki and kept under 24 h light photoperiod and fed with frozen bloodworms (Chironomidae sp.) until enough fish from each population had reached reproductive condition.
Between 25 June and 1 July, five full-sib crosses were made artificially (by gently squeezing the eggs out from the ripe females, and pouring the sperm solution obtained by mincing the testicles of over-anaesthetized males on the eggs) from each population. Clutches were transported to 1.4 L tanks of two Allentown Zebrafish Rack Systems (hereafter rack, Aquaneering Inc., San Diego, USA). Racks had a closed water circulation system, with multi-level filtering including physical, chemical, biological and UV filters. We regularly checked the clutches under dissecting microscope, and removed the dead or unfertilized eggs. After the fish started to swim freely (four days after hatching) 10 fish from each family (200 individuals) were housed in the two racks individually (the extra fish were used in other experiments). Visual contact between the tanks was blocked by white plastic panels. Individual rearing was important for two main reasons. Firstly, to provide uniform environment for the individuals within population by removing behavioural variation related to previous experience from predatory encounters or aggressive encounters with conspecifics. Secondly, to provide uniform environment for all populations considering the fact that constraints of group living differ markedly between pond and marine P. pungitius (Herczeg et al., 2008), most probably as a result of habitat-dependent differences in aggression (present study). While total isolation might be ‘unnatural’ for P. pungitius and lead to ‘abnormal’ behaviour, we judged the benefits of measuring behaviour unbiased by previous interactions with predators and conspecifics to be a more important consideration. This in particular from the point of view of genetics: individuals have social ranks in groups, and thus, every individual in a group experience a different social environment, which is bound to confound genetic and environmental influences on behaviours. Likewise, keeping fish without the choice of leaving the group could also be argued to be ‘unnatural’. While our setup was efficient in removing most of the possible environmental variation and bias by previous experience, we could not statistically control for possible maternal effects – a problem common to all studies utilizing F1-generation offspring.
Fish were fed in excess first with live brine shrimp (Artemia sp.) nauplii, and then with frozen copepods (Cyclops sp.) and bloodworms. Twenty-four hour light photoperiod, representative for summer conditions at high latitudes, was used until the fish were 12-weeks-old after which the photoperiod was switched gradually (over 1 week) to a 12 : 12 h light : dark periodism. Because of latitudinal differences between the source populations (Fig. 1), we did not aim to mimic the natural light regimes. Water temperature was set to 17 °C throughout the experiment. Apart from some sparse size measurements (10 measurements of each individual, started after hatching) we did for other purposes, fish remained in isolation for approximately 8 months, after which the behavioural experiments started. Chemical cues of conspecifics, however, could not be eliminated in the closed water circulation in the racks. Use of adult fish was important because the population differences in predation-related selective pressures are most prominent at adult stage in our populations; juveniles are predated by insects, insect larvae and adult conspecifics in both environments. Further, Bell & Stamps (2004) reported ontogenetic changes not only in single behaviours but also in behavioural syndromes. Thus, behaviours measured at one stage of development do not necessary predict those in another. As a result of other scientific purposes and some mortality, we could use 82 fish, 18 from the Baltic Sea (family representations: 6, 3, 3, 2, 4), 22 from the White Sea (family representations: 5, 3, 7, 2, 5), 20 from Bynästjärnen (family representations: 3, 3, 4, 3, 7) and 22 from Pyöreälampi (family representations: 1, 6, 5, 5, 5). We assumed that our samples are representative of the source populations. All fish were in a nonreproductive state (i.e. their behaviour was not affected by reproduction), meaning that their sex could not be determined by eye. We assumed that the sexes were represented equally. All behavioural tests were conducted between 11.00 and 17.00 h. Fish were tested for the given trait in random order. Experimental fish were not fed the day before the given experiment.
Drive to feed
Drive to feed was measured in two ways, first as time to feed in the normal environment and second as time to feed in an altered one. The 1.4 L plastic tanks had two holes at their top, one for feeding (closer to the front) and one for the water inlet (closer to the tank’s centre). However, there was enough space to feed the fish with a pipette through both holes. First, on February 22 2008, we provided bloodworms (as on normal days, we provided food in excess) with a pipette through the feeding hole at the top of the tanks to the focal fish (in a similar way to the ordinary feedings) and measured the time until the first biting attempt with a stopwatch in seconds. Second, on February 24 2008, just before the feeding, we placed a shiny steel screw (18 cm long, 1 cm in diameter) through the feeding hole in a way that the screw was reaching the bottom of the tank. Considering the dimensions (ca. 15 × 5 × 25 cm, height, width, length, respectively, the shape was not completely regular) of the tanks, the sudden appearance of the screw was envisioned to represent an alteration of the fish’s environment. We provided the bloodworms through the hole where the water inlet entered the tank, and measured the time until the first biting. The whole process (i.e. placing the screw and giving the food) took a few seconds. If a fish did not feed during the first 5 min, the experiment was terminated, and the fish was assigned to a time of 300 s as feeding time.
Alteration of the environment significantly increased the feeding time of the individual fish in all populations (Wilcoxon Matched Pairs Tests: Z > 2.52, P < 0.012) except in Pyöreälampi (Wilcoxon Matched Pairs Test: Z = 0.92, P = 0.357), so we assumed that the alteration was successful. Originally, we planned to use the feeding time in an altered environment as a measure of boldness, but it did not correlate with the other measures of boldness (see below) in any of the populations (Spearman rank correlations:−0.07 < rs(18–22) < 0.26, P > 0.299). However, there was a trend for correlation between feeding time in the normal and altered environments in all populations (Spearman rank correlations; Baltic Sea: rs(18) = 0.58, P = 0.011; White Sea: rs(22) = 0.69, P < 0.001; Bynästjärnen: rs(20) = 0.41, P = 0.075; Pyöreälampi: rs(22) = 0.36, P = 0.099), so finally we treated both measurements as describing ‘drive to feed’.
Tanks were organised in five rows (20 tanks per row) on both racks. The measured 82 fish were placed in a way that only every second tank held a fish so as that every fish had empty tanks on both sides. As we kept the extra fish produced from each population, we could present stimulus fish from the same age and population to the focal fish. Stimulus fish were placed into a tank at a random side of the focal fish’s tank and allowed to settle for 15 min. After this, the plastic panels blocking visual contact at both sides of the focal fish were removed. This was important for separating aggression from the reaction to a novel object (novel empty tank vs. novel tank with a conspecific; we did not observe any interest or attacks towards the novel empty tank). We waited until the first orientation (being head-first towards the stimulus fish with eyes fixed on it) of the focal fish towards the stimulus fish, and then measured the time the focal fish spent with orientation and the number of attacks (sudden bursts often coupled with biting attempts) it made during the next 5 min. Fish that did not orient until 5 min received zero scores for both measures. Aggression measurements were conducted between 26 February and 3 March 2008.
For the boldness measurements, focal fish was gently netted out from their holding tanks and placed tail-first into a grey PVC pipe (28 cm long, 3 cm in diameter) filled up with water (one end of the pipe was permanently sealed). The pipe was immediately submerged into an opaque plastic tank (38 × 36 × 62 cm, height, width, length, respectively), filled up with water to the level of 10 cm. There was a 26 × 13 cm white plastic panel (1 mm thick) glued permanently at the bottom of the tank, positioned so that it covered the area right in front of the unsealed end of the PVC pipe. The unsealed end of the pipe was then blocked with a 6 × 6 × 3 cm grey plastic block. Netting the fish, putting it into the pipe, placing the pipe into the plastic tank, and blocking the pipe’s unsealed end with the plastic block took ca. 10 s. We let the fish settle for 3 min, after which the plastic block was lifted with the aid of a string. We measured the time until (i) the fish’s head and then its (ii) full body left the pipe. If the fish’s head did not appear within 10 min we gave a score of 600 s for both measurements. Water in the plastic tank and the PVC pipe was taken directly from the racks’ reservoirs. Boldness measurements were done between 9 and 16 March 2008.
Within populations tests
To collapse the two variables for every behavioural trait we aimed to estimate (see above) to a single variable (drive to feed, aggression and boldness), we run three Principal Component Analyses (PCAs) across populations. In all cases, only the first principal component (PC) had an eigenvector > 1, showing a strong positive relationship with the original variables, and describing the main proportion of variation (see Results). The individual PC scores were used in the subsequent analyses. To see if the PC scores obtained this way were useful for the within population tests (see below), we also run PCAs in a similar fashion, but separately for the different populations. The correlation structure was very similar to that found in the across population PCAs: in all cases the first PC strongly and positively correlated with the two original variables, describing the main proportion of the variation (data not shown). Thus, use of the PC scores calculated across the populations was adequate for our purposes.
We used Spearman rank correlation coefficients to test for the existence of behavioural syndromes within populations with sequential Bonferroni correction (Rice, 1989). Statistical comparisons of the correlation coefficients of the same pairs of behaviours across populations were done with χ2-tests according to Zar (1999, p. 390). Lack of significance in statistical tests could be argued to be due to low statistical power to detect them. This seems unlikely in our case because a priori power calculations estimated with a hypothetic N = 20 (our population sample sizes varied between 18 and 22), suggest that our chance to detect biologically significant correlations was reasonable (power [β] = 0.6 for rs = 0.5; β = 0.8 for rs = 0.6; β = 0.9 for rs = 0.7). We did not conduct a posteriori power calculations as they are a 1 : 1 function of the P-values, and thus, do not provide any extra information about the reliability of nonsignificant tests (e.g. Hoenig & Heisey, 2001). In cases where individuals did not respond to a given treatment (i.e. did not feed, attack the conspecific, leave the refuge, etc.) the maximal time (time to feed, time to leave refuge) or zero score (time spent with orientation, number of attacks) was assigned. In some behavioural variables (Table 1), the ratio of nonresponders exceeded 50% of the tested individuals from a given population, making correlations somewhat problematic. However, as we believe that not responding is actually a type of response (strengthened by the strong habitat specific pattern of the ratio of responders/nonresponders, see Results) we left these individuals in the analyses. Leaving individuals that did not react to the treatment or scored zero in both measures contributing to the estimate of a given behaviour (drive to feed, aggression or boldness; all based on two original variables) out from the correlations did not change the overall picture (the significant negative drive to feed – aggression correlation in the Baltic Sea became nonsignificant, while the marginally significant negative drive to feed – aggression correlation in the Bynästjärnen population became significant [see Results for the correlations]; data not shown).
Table 1. Population differences in the ratio of responders/nonresponders (in parentheses) observed in the original behavioural measurements. Population pairs that did not differ significantly in the χ2-tests following the loglinear analyses are also shown (all other pairs differed).
Population pairs not differing
BAS, Baltic Sea; WHS, White Sea; BYN, Bynastjärnen; PYÖ, Pyöreälampi.
Note that in the case of ‘time to head out’ and ‘time to body out’ the responders and nonresponders were exactly the same individuals in all populations (i.e. they either left the refuge or not).
*0.1 > P > 0.05; **P > 0.249; ***formal test impossible due to repeated zero counts in the same categories, populations treated as not differing
Time to feed in normal environment
Time to feed in altered environment
Time of orientation
BAS-WHS*; WHS-BYN*; WHS-PYÖ*; BYN-PYÖ*
Number of attacks
Time to head out
Time to body out
Among population tests
To compare behavioural types between populations, we first ran a PCA based on all six behavioural variables on data pooled across populations. This was important, as simply comparing the PC scores of the previous separate PCAs (see ‘Within population tests’ above) would not compare the individual configuration of different behaviours (= behavioural type, Bell, 2007) but the population means of the different behaviours separately. The first two PCs had eigenvectors > 1 and accounted for over 70% of the total variance together (see Results). To compare PC scores between populations, we constructed two separate General Linear Mixed Models (GLMMs) with the PC scores as dependent variables, population as a fixed factor, and family nested within population as a random factor. For pairwise population comparisons, we used Scheffe post hoc tests. The problem imposed by individuals not responding to the given treatments during the observation period being assigned to maximal time (see above) made these analyses only more conservative considering that in reality the response time should have been even larger than the assigned value. To test if the ratio of responders/nonresponders was population specific, we applied log-linear models separately for every original behavioural variable coupled with a posteriori pairwise χ2-tests. All statistical analyses were done with the statistica 6.1 for Windows (StatSoft Inc., Tulsa, USA) and sas 9.1 for Windows (SAS Institute Inc., Cary, USA) software packages.
In all three PCAs of the behaviours across populations, the first PC described the given behaviour (drive to feed, aggression and boldness; Table 2). In all cases, the loadings of the original variables were high and positive, and the PCs explained a high proportion of the variance of the original variables (Table 2).
Table 2. Results of three principal component analyses of the different behavioural traits conducted by pooling the data across all populations. Loadings, eigenvalues and the percentage of explained variance is provided for each principal component (PC). Only PCs with eigenvalues > 1 are shown. See text for the description of the original variables.
Drive to feed
Time to feed in normal environment
Time to feed in altered environment
Time of orientation
Number of attacks
Time to head out
Time to body out
Correlations between the three PCs were tested separately in every population applying the sequential Bonferroni correction. In the Baltic Sea population we found a negative correlation between drive to feed and aggression (rs(18) = −0.62, P = 0.006), meaning that fish that fed quicker were also more aggressive. Drive to feed and boldness (rs(18) = 0.21, P = 0.408) or aggression and boldness were not correlated (rs(18) = −0.11, P = 0.651). In the other populations none of the correlations were significant (White Sea: drive to feed–aggression, rs(22) = − 0.34, P = 0.125; drive to feed–boldness, rs(22) = 0.06, P = 0.804; aggression and boldness, rs(22) = −0.19, P = 0.401; Bynästjärnen: drive to feed–aggression, rs(20) = −0.43, P = 0.056; drive to feed–boldness, rs(20) = 0.04, P = 0.879; aggression and boldness, rs(20) = −0.08, P = 0.734; Pyöreälampi: drive to feed–aggression, rs(22) = −0.13, P = 0.563; drive to feed–boldness, rs(22) = 0.09, P = 0.706; aggression and boldness, rs(22) = −0.18, P = 0.414). Correlation coefficients did not differ among the different populations (drive to feed–aggression, χ23 = 3.05, P > 0.250; drive to feed–boldness, χ23 = 0.29, P > 0.950; aggression and boldness, χ23 = 0.15, P > 0.975) indicating a lack of correlation structure across populations.
The PCA run on all behavioural variables across populations resulted in two PCs with eigenvalues > 1 (Table 3). The first PC explained more than half of the variation in behaviours, and gave a biologically interpretable complex description of behaviour: it represented a gradient from quickly feeding, aggressive and bold fish towards slowly feeding, peaceful and shy fish (Table 3). The second PC explained 22% of the variation, and described a somewhat unexpected component of complex behaviour: a gradient from bold and peaceful towards shy and aggressive fish (Table 3).
Table 3. Results of the principal component analysis using all six behavioural variables, based on data pooled across populations. Loadings, eigenvalues and the percentage of explained variance for each principal component (PC) are given. Only PCs with eigenvalues > 1 are shown. See text for the description of the original variables.
Time to feed in normal environment
Time to feed in altered environment
Time of orientation
Number of attacks
Time to head out
Time to body out
The GLMM run on the scores of the first PC revealed that families did not differ within populations (Z = 0.53, P = 0.299), but populations corrected for the family effects differed significantly in behavioural types (F3,12.5 = 24.47, P < 0.001, Fig. 2a). Furthermore, post hoc tests revealed a habitat specific systematic pattern: neither marine (P = 0.510), nor small pond populations (P = 0.896) differed from each other, while both marine populations differed from both small pond populations (Baltic Sea – Bynästjärnen: P = 0.005; Baltic Sea – Pyöreälampi: P = 0.001; White Sea – Bynästjärnen: P < 0.001; White Sea – Pyöreälampi: P < 0.001). In all cases, small pond fish were quicker to feed, more aggressive and bolder than marine fish (Fig. 2a). The GLMM on the second PC revealed that neither families within population (Z = 1.54, P = 0.062), nor populations corrected for the family effects differed significantly from each other (F3,15 = 1.60, P = 0.231, Fig. 2b).
The ratio of responders/nonresponders differed significantly between populations in all original behavioural variables as revealed by log-linear analyses (time to feed in normal environment, time to feed in altered environment, time of orientation, number of attacks, time to head out from refuge, time to full body out from refuge; all χ2 > 15.34, all P < 0.0015). Pairwise population comparisons separately for the original variables revealed a strong habitat specific pattern; populations tended to be similar within but different between habitat types (Table 1). In general, marine populations had lower ratio of responders/nonresponders than pond populations.
The aim of our experiment was to provide measures of complex behaviour in adult fish unbiased by former experience with predation or aggression from populations that differed markedly in predation pressure. Among the 12 possible behavioural correlations within populations (four populations × three behaviours) we found only one showing a significant trend. Further, we could not detect statistical differences among the correlation coefficients. Hence, the ‘constraint’ hypothesis of behavioural syndromes was rejected. On the other hand, comparing behavioural types between the different habitats, we found that isolated small ponds were characterized by fish with a higher drive to feed, more aggression and boldness than fish from marine habitats. The habitat dependent occurrence of the same phenotype irrespective of population origin implies natural selection as the causal agent (e.g. Clarke, 1975; Endler, 1986).
The ‘constraint’ hypothesis evoked to explain the existence of behavioural syndromes predicts strong genetic correlations between different behaviours. This hypothesis can be rejected based on the lack of behavioural correlations. The ‘adaptive’ hypothesis predicts that correlations between different behaviours might be present in environments where the particular combinations of behaviours are selected for but can be reversed or uncoupled in other environments. The only study testing the predictions of these hypotheses explicitly by statistically comparing the behavioural correlations between high and low predation risk three-spined stickleback populations (six population replicates per habitat type) supported the ‘adaptive’ hypothesis (Dingemanse et al., 2007). We found minimal evidence for behavioural syndromes. This result clearly contradicts the ‘constraint’ hypothesis, while the lack of syndromes did not allow us to test the ‘adaptive’ hypothesis. Our result also contradicts earlier studies reporting numerous behavioural correlations (e.g. Bell & Stamps, 2004; Bell, 2005; Dingemanse et al., 2007). Two possible reasons for the lack of behavioural syndromes are conceivable. Firstly, previous experience with predators or conspecifics is needed for the ontogenetic emergence of behavioural correlations. Bell & Sih (2007) has shown that exposure to predation indeed resulted in the emergence of previously lacking behavioural syndromes, a phenomenon partly caused by plasticity. It is also easy to imagine that dominant individuals within a social group might develop to aggressive, bold and active, whereas subordinates peaceful, shy and inactive. Secondly, the lack of behavioural syndromes in our markedly different populations might simply mean that natural selection did not favour correlated behaviours in any of them. Further experimental studies separating the genetic and environmental contribution to behavioural syndromes following Dingemanse et al. (2008), considering both group living and predation are needed.
Geographic variation in behaviour is common (e.g. Foster & Endler, 1999), and there are several studies showing covariation between predation pressure and certain behaviours among populations (e.g. Magurran & Seghers, 1991, 1994; Brown et al., 2005). Predation related behavioural traits have also been shown to be heritable (e.g. Seghers, 1974; Giles, 1984; Magurran, 1990; Brown et al., 2007). However, it also known that experience has a large (in some cases population specific) effect on the actually expressed behaviours (e.g. Dill, 1974; Magurran, 1990). Our results based on completely conspecific- and predator-naïve P. pungitius demonstrate that the studied populations differed in their average behavioural type. We found that fish from the different habitat types differed consistently in a measure of complex behaviour: sticklebacks from ponds without predation were more aggressive, bolder and quicker to feed than their conspecifics from marine habitats. Further, in a somewhat coarse scale, we found that many more pond fish responded to our treatments (i.e. did actually feed, exhibit aggression, or leave the refuge during the observation period) as compared with marine fish. Even though only two replicate populations per habitat type were tested, the large geographical (> 500 km) and genetic (Shikano, Herczeg & Merilä, unpublished work) distance between replicates made them truly independent. It is noteworthy that marine vs. pond habitats differ in several other aspects than predation, for instance in salinity, habitat complexity, interspecific competition, etc., so we can only suggest predation as the most likely factor behind the observed patterns. However, the considerably smaller antipredator defence apparatus (pelvic spines) we observed both in the wild-caught and common garden pond fish when compared with marine fish suggests that the difference in predatory pressure is an important driver of evolutionary divergence between the habitat types. At any rate, the fact that the differences were driven by habitat type and not population origin suggests that the pattern is a result of natural selection (e.g. Clarke, 1975; Endler, 1986; Schluter & Nagel, 1995).
Interestingly, the different behaviours were independent from each other within the populations (= lack of behavioural syndromes), but shifted together in response to presumed differences in the selection pressures across populations. Hence, we suggest that in our case drive to feed, aggression and boldness evolved together, but on an independent genetic basis as a response to the changes in predation pressure. The population differences in complex behaviour might be related to the different life-history strategies of P. pungitius populations living in different habitats. Results from our related studies have revealed that pond fish have evolved to grow faster and longer than lake or marine fish, sometimes reaching giant sizes (Herczeg, Gonda & Merilä, unpublished work), and that group living has considerable developmental costs for pond fish, but not for marine fish (Herczeg et al., 2008; Gonda, Herczeg & Merilä, unpublished work). Hence, the life-history evolution of these markedly different and isolated P. pungitius populations warrant further investigations.
There are at least two different ways how animals can change their behaviour in response to an increase (or reduction) in predation risk. One would intuitively predict that as the mortality risk imposed by predation increases, prey become shyer and less active, as found by e.g. Bell (2005) and Brydges et al. (2008). Interestingly, Brown et al. (2005, 2007) reported the opposite: Brachyraphis episcopii individuals showed higher activity and were bolder under heavy than low predation risk. The interpretation was that fish in the high predation risk populations have to be bolder to carry on with the necessary activities (Brown et al., 2005, 2007). According to our results, P. pungitius behave in the intuitively expected way: increased predation risk makes them risk-averse, and hence, they become shyer and less aggressive, decreasing their feeding activity.
In summary, we found negligible evidence of behavioural syndromes (population level correlations between different behaviours) in first laboratory generation and predator- and conspecific-naïve adult P. pungitius from four geographically and genetically isolated populations differing markedly in predation regime. However, the behavioural types (an individual-based estimate of complex behaviour) differed between habitat types, but not between populations within habitat types. Fish from low predation risk populations were quicker feeders, and also more aggressive and bolder than their conspecifics from high predation environments. We suggest that (i) further studies are needed to separate the environmental and genetic components of behavioural syndromes and (ii) that independent evolution of different behavioural traits as a response to different levels of predation can result in population level differences in complex behaviours.
Victor Berger, Göran Englund, Tuomas Leinonen, Daniel Lussetti and Pirkko Siikamäki helped us in organizing and executing the field sampling. Special thanks to the Oulanka Research Station and the White Sea Biological Station helping us with the fieldwork. We are highly indebted to Niels Dingemanse for his constructive comments and John Loehr for correcting the English. The authors received financial support from the Academy of Finland and Centre for International Mobility (CIMO). The experiments were conducted under the license of the Helsinki University Animal Experimentation Committee.