Population differences in behaviour are explained by shared within-population trait correlations

Authors


Jonathan N. Pruitt, Department of Ecology and Evolutionary Biology, University of Tennessee, Knoxville TN 37996-1610, USA.
Tel.: 863 289 4078; fax: 865 974 3067; e-mail: jpruitt6@utk.edu

Abstract

Correlations in behavioural traits across time, situation and ecological context (i.e. ‘behavioural syndromes’ or ‘personality’) have been documented for a variety of behaviours, and in diverse taxa. Perhaps the most controversial inference from the behavioural syndromes literature is that correlated behaviour may act as an evolutionary constraint and evolutionary change in one’s behaviour may necessarily involve shifts in others. We test the two predictions of this hypothesis using comparative data from eighteen populations of the socially polymorphic spider, Anelosimus studiosus (Araneae, Theriidae). First, we ask whether geographically distant populations share a common syndrome. Second, we test whether population differences in behaviour are correlated similarly to within-population trait correlations. Our results reveal that populations separated by as much as 36° latitude shared similar syndromes. Furthermore, population differences in behaviour were correlated in the same manner as within-population trait correlations. That is, population divergence tended to be along the same axes as within-population covariance. Together, these results suggest a lack of evolutionary independence in the syndrome’s constituent traits.

Introduction

Behavioural correlations across time, situation and ecological context have received a wealth of attention in recent years (Gosling, 2001; Dall et al., 2004; Sih et al., 2004; Bell, 2007). The phenomenon has been referred to as personality, temperament, behavioural tendency and, most recently, behavioural syndromes (reviewed in Sih et al., 2004). Behavioural correlations across situations and contexts are intriguing to evolutionary biologists because they imply a lack of trait independence, and this lack of independence might generate adaptive trade-offs (Sih et al., 2004; Johnson & Sih, 2005, 2007; Pruitt & Riechert, 2009a,b). However, some theory and data suggest that behavioural syndromes can be readily formed and disintegrated over brief periods of evolutionary time (Bell, 2005; Bell & Sih, 2007; Dingemanse et al., 2007; Wolf et al., 2007, 2008; Herczeg et al., 2009) or even ontogeny (Soldz & Valliant, 1999; Bell & Stamps, 2004), and therefore, have little constraining effect on behavioural evolution. For instance, populations of three-spined sticklebacks differ tremendously in their syndromes (Bell, 2005; Dingemanse et al., 2007), and experimentally imposed selection regimes generated trait correlations in populations typically lacking a behavioural syndrome (Bell & Sih, 2007). Thus, the effect of behavioural syndromes as an influence on evolution, over even moderate time scales, is in question.

If behavioural syndromes are influential for behavioural evolution, we might expect them to have some explanatory power for population or species differences in behaviour. In a hypothetical example, suppose activity level and aggressiveness are positively correlated (e.g. owing to pleiotropy) within populations. If the syndrome is important for trait evolution, we might predict population differences in mean trait values to be explained by their shared syndrome. Thus, if population ‘A’ exhibits a higher activity level than population ‘B’, we would predict, based on the syndrome, that population ‘A’ would necessarily exhibit greater aggressiveness as well. In our study, we use the socially polymorphic spider Anelosimus studiosus (Araneae, Theridiidae) to test this prediction.

Anelosimus studiosus is a common tangle-webbed spider with a remarkable range, from mid-latitudes in North America to southern Argentina (Agnarsson et al., 2007; Jones et al., 2007; Riechert & Jones, 2008). This species is unusual among spiders because it exhibits a social behaviour polymorphism; adults exhibit an aggregative grouping (‘social’) or repulsive aggressive (‘asocial’) phenotype (Jones et al., 2007; Riechert & Jones, 2008). Social individuals form multi-female colonies and cooperate in shared web maintenance, prey capture and alloparental care. Asocial individuals establish solitary webs and defend them from intrusion from conspecifics (Furey, 1998). However, it is not uncommon that multi-female webs harbour asocial individuals, which may act as social parasites on colony function (Pruitt et al., 2008; Pruitt & Riechert, 2009c). Northern populations (34–36o latitude) of A. studiosus exhibit a behavioural syndrome, where increased social tendency is phenotypically correlated with a decrease in aggression towards predators, prey and mates, as well as reduced activity level and dispersal tendencies; these correlations add a number of nonintuitive costs and benefits to sociality (Pruitt et al., 2008; Pruitt and Riechert, 2009a–c). Furthermore, even in populations where all females are generally asocial, small differences in social tendency correlate with variation in other behaviour (Pruitt et al., 2008), suggesting that the observed syndromes are not merely artefacts of a ‘two-point regression’ distinguishing discrete phenotypes. Taken together, our previous data suggest that social tendency might be evolutionarily linked with other behaviour in A. studiosus and a shift in social tendency may necessarily involve changes in other behaviour. To test this hypothesis, we expand our syndrome analyses to include eighteen populations over much of A. studiosus’ range. We examine (i) whether geographically disparate populations exhibit similar correlations between social tendency and other behaviour (e.g. activity level, foraging behaviour) and (ii) whether population differences in mean trait values are predicted by within-population syndromes.

Materials and methods

Collection sites and lab maintenance

Individuals were collected by placing a plastic bag over their web and trimming the supporting foliage. A minimum of 48 individuals were sampled from each of 18 populations, spanning half of the species’ range (total = 907). Anelosimus studiosus are forest-edge, disturbance specialists (e.g. recent landslides, roadways), and our population selection was opportunistic. Only mature preparturition females were included in our analysis. Spiders were then transported from their collection sites to nearby air conditioned hostels where the behavioural assays were performed (C° 21.1–24.4). Each individual was housed in a 59 mL cup and fed once ad libitum before the trials were initiated. Behavioural trials were initiated within 24 h of their initial feeding, and 24 h time elapsed between trials. The behavioural trials were completed in the order presented here. Populations come from a variety of source habitats, e.g. high elevation temperate (TN, USA), low elevation temperate (GA, USA), inter-Andean xeric (EC), temperate riparian (TN, USA), sub-tropical riparian (FL, USA), swamp (FL, USA), scrub habitat (FL, USA) (Table 1).

Table 1.   Latitude and longitude of the 18 source populations
PopulationGPS
A36°00′N 84°08′W
B34°17′N 83°49′W
C26°05′N 81°20′W
D30°15′N 84°40′W
E31°59′N 85°8′W
F35°56′N 85°59′W
G32°54′N 83°38′W
H0°5′S 78°26′W
I30°40′N 84°50′W
J28°55′N 81°74′W
K0°18′N 78°10′W
L25°45′N 80°50′W
M35°47′N 78°38′W
N27°57′N 82°27′W
O35°35′N 82°34′W
P25°57′N 81°31′W
Q28°35′N 81°21′W
R31°5′N 91°4′W

Inter-individual distance test

To determine females’ social tendency, two females of unknown social tendency were individually marked with florescent powder and placed in the centre of clear, plastic container (13cm × 13.5cm × 2.5cm). After 24 h of settling time, we measured the distance between them. All females that exhibited an inter-individual distance greater than zero were run through a second confirmatory test with a known highly social female (i.e. one that previously exhibited an inter-individual distance score of zero). This is because more aggressive females, which demand space, may chase away social females. We used the between-individual distance measure from the second, confirmatory trial when calculating the correlations between social phenotype and other behavioural trait scores. By using inter-individual distance, we allow for the possibility of intermediate social phenotypes. Females’ inter-individual distance scores are both repeatable and heritable (Pruitt et al., 2008; Pruitt & Riechert, 2009a,b; J.N. Pruitt, unpublished data).

Exploratory behaviour and boldness

Each female was placed in a new clear plastic container (13.5cm × 13cm × 2.5cm) lined with paper marked with a grid pattern. All females ceased movement in response to our closing the lid forcefully. We observed the subsequent behaviour of the test females over a 5-min period. We recorded (i) time lapsed between the female’s freezing behaviour and the first movement as measured by stopwatch (boldness), (ii) the number of times the female moved in a 5-min period following its first movement (activity) and (iii) the total distance the female traversed in this 5-min interval following its first movement (exploration). Movements were scored as independent if separated by 5 s or more of quiescence.

Prey attack sequence

Females were offered a single termite worker. The termite was placed on the female’s web 2.0 cm from the test female, and the capture sequence was recorded. We recorded (i) the time taken to orient to the prey item or the first intention movement exhibited by the female (‘latency of response’), (ii) the time from first response until a female made contact with the prey (‘latency to attack’) and (iii) the number of bites issued before feeding.

Statistical methods

To address the question of whether individual-level trait correlations (the behavioural syndrome) predict population-level trait differences, we first used nonparametric anovas (Kruskal–Wallis test) to test whether populations were significantly different for the individual traits studied. We then evaluated within- and between-population trait correlations to test the consistency of the behavioural syndrome.

Nonparametric Spearman’s correlations were used to assess within- and between-population trait correlations. We used ancova to determine whether the relationship between inter-individual distance (our measurement of ‘sociality’) and other behaviour differed among populations (Main effect: population, Covariate: Inter-individual distance). A significant main effect indicates one or more populations shifted in one trait type without a corresponding shift in the other (i.e. that individuals of the same social tendency are more or less aggressive among populations). A significant interaction term between inter-individual distance and population indicates the slopes of the lines differ among populations (i.e. the magnitude or direction of the relationship). If both individual differences and population differences reflect a single strongly integrated behavioural syndrome, we would expect a strong effect of the covariate and no significant population or interaction terms. Visual inspection of the residuals for our within population analyses indicated right skew and heteroscedasticity for several traits, and thus, our response variables were log-transformed for our ancova analyses.

Because similarities among populations might be caused by shared history and/or geography rather than conserved syndromes, we tested for spatial autocorrelation in between-population analyses using the autoregression model yi = α + βxi + ρΣjyjwij + ε, where yi is a response variable mean for population i, xi is the mean inter-individual distance for population i and wij is a decreasing function of distance between populations i and j (we used 1/distance and 1/distance2, following Lichstein et al., 2002). We used AIC to compare models with and without one or both autoregression terms (Akaike, 1987; Burnham & Anderson, 2004). This model selection tool allowed us to compare models without hierarchical relationships (e.g. the model with only 1/distance vs. only 1/distance2).

Results

Using Kruskal–Wallis test, we detected significant among population differences in all traits studied: Inter-individual distance (inline image = 132.83, P < 0.0001), latency of response to prey (inline image = 38.55, P = 0.002), latency of attack (inline image = 78.96, P < 0.0001), number of bites issued to prey (inline image = 38.79, P = 0.002), boldness (inline image = 31.41, P = 0.018), activity level (inline image = 61.46, P < 0.0001) and exploratory behaviour inline image = 51.37, P < 0.0001).

None of our autoregression models supported spatial autocorrelation as having any predictive value. Model comparisons with AIC never supported including the autoregression terms, autoregression terms never had significant P-values (all were > 0.11), and including autoregression terms did not substantially affect the value or statistical significance of the regression coefficient for inter-individual distance (for AIC scores and P-values see Online Supporting Information). These results indicate that the shared patterns among populations are not confounded by shared geography. For simplicity, we have omitted further consideration of geographical distance in our analyses.

Our within-population syndrome analyses revealed similar syndromes among even geographically isolated populations (summarized in Table 2). For all populations, increased social tendency was positively correlated with a greater latency of response to prey items, longer latency of attack and reduced boldness towards simulated predator cues. Social tendency was correlated with fewer bites during prey capture in 12 of 18 of our populations, reduced activity level in 15 of 18 of our populations and increased exploratory tendencies in 17 of 18 of our populations. Our population-level correlations revealed that populations with higher social tendencies, on average, were slower to respond to and attack prey, issued fewer bites during prey capture, exhibited reduced boldness and lower activity levels (Table 3, Fig. 1). We failed, however, to detect a significant correlation among populations between social tendency and exploratory behaviour (distance moved).

Table 2.   Spearman’s ρ correlation coefficients between inter-individual distance and latency of response to prey (LOR), latency of attack (LOA), number of bites issued to prey items (Bites), latency to return to activity after a simulated predator cue (Boldness), activity level (activity) and exploration in a novel environment (exploration) for eighteen populations of A. studiosus
 A (n = 48)B (n = 49)C (n = 52)D (n = 47)§ E (n = 51)F (n = 51)
  1. *Denotes a significant correlation at α = 0.05.

  2. **Denotes a significant correlation at α = 0.008 (significant after Bonferroni correction).

  3. §Denotes behaviourally monomorphic populations.

LOR−0.42**−0.64**−0.38*−0.36*−0.33*−0.44**
LOA−0.49**−0.80**−0.37*−0.32*−0.31*−0.41**
Bite number0.37*0.64**0.34*0.07−0.100.26
Boldness−0.66**−0.37*−0.47**−0.68**−0.44**−0.57*
Activity0.250.33*0.29*0.41**0.190.18
Exploration0.47**0.31*0.26*0.51**0.33*0.40**
 G (n = 50)§ H (n = 51)I (n = 50)J (n = 50)K (n = 50)L (n = 51)
LOR−0.46**−0.44**−0.55**−0.53**−0.61**−0.37**
LOA−0.44**−0.47**−0.54**−0.42**−0.47**−0.34*
Bite number0.37*0.42**0.43**0.220.47*0.45**
Boldness−0.38*−0.62 **−0.35 *−0.47**−0.58**−0.48**
Activity0.43**0.44**0.38 *0.33 *0.43**0.34 *
Exploration0.46**0.41**0.35 *0.56**0.49**0.41**
 M (n = 57)N (n = 52)§O (n = 53)§ P (n = 51)Q (n = 49)R (n = 50)
LOR−0.45**−0.31 *−0.35 *−0.41**−0.71**−0.51**
LOA−0.41**−0.34 *−0.32 *−0.35 *−0.41**−0.53**
Bite number0.32 *0.200.210.45**0.32 *0.14
Boldness−0.57**−0.28*−0.29 *−0.28 *−0.56**−0.66**
Activity0.35*0.47**0.60**0.50**0.47**0.78**
Exploration0.40**0.44**0.200.51**0.44**0.58**
Table 3.   Spearman’s ρ correlation coefficients testing for among population correlations of mean trait values between inter-individual distance and latency of response to prey (LOR), latency of attack (LOA), number of bites issued to prey items (Bites), latency to return to activity after a simulated predator cue (Boldness), activity level (activity) and exploration in a novel environment (exploration)
TraitSpearman’s ρ
  1. *Denotes a significant correlation at α = 0.05.

  2. **Denotes a significant correlation at α = 0.01.

LOR−0.86**
LOA−0.67**
Bites0.51*
Boldness−0.64**
Activity0.67**
Exploration0.33
Figure 1.

 Scatter plots between the mean inter-individual distance measure of each population and other mean trait values. Each point represents a population. Lines represent standard errors of the population means.

Inter-individual distance (our measure of social tendency) was a significant covariate for all of the ancova (Table 4, Fig. 2). Neither the main effect of population nor the interaction term (population x inter-individual distance) was significant for latency of response to prey, the number of bites issued during prey capture, boldness towards simulated predator cues or exploratory behaviour. A significant main effect of population was detected for latency of attack (F1,17 = 2.24, P = 0.004), but the interaction term was nonsignificant (F1,17 = 1.28, P = 0.202). A significant main effect of population was also detected for activity level (F1,17 = 2.09, P = 0.007), and for this trait, the interaction term was also significant (F1,17 = 3.79, P < 0.0001).

Table 4. ancova for each trait in the syndrome
TraitTermDFβF RatioP-value
Latency of responsePopulation17 1.180.277
Inter-individual Distance1−0.051214.91< 0.0001
Population* Inter-Individual Distance17 0.980.476
Latency of attackPopulation17 2.240.004
Inter-individual Distance1−0.041134.8< 0.0001
Population* Inter-Individual Distance17 1.280.202
Bite numberPopulation17 1.510.091
Inter-individual Distance10.01367.02< 0.0001
Population* Inter-Individual Distance17 1.380.144
BoldnessPopulation17 1.280.204
Inter-individual Distance1−0.059178.46< 0.0001
Population* Inter-Individual Distance17 0.870.602
Activity levelPopulation17 2.090.007
Inter-individual Distance10.03162.42< 0.0001
Population* Inter-Individual Distance17 3.79< 0.0001
ExplorationPopulation17 1.470.113
Inter-individual Distance10.03418.63< 0.0001
Population* Inter-Individual Distance17 1.560.074
Figure 2.

 The within population regression lines between inter-individual distance measure and other behavioural traits. The Y-axes have been log-transformed. Letters correspond with outlier populations.

Discussion

While behavioural syndromes have been documented for a variety of behavioural traits in a variety of taxa (Keller & Ross, 1993, 1999; Ross & Keller, 1998; Gosling, 2001; Dingemanse et al., 2004; Sih et al., 2004; Johnson & Sih, 2005, 2007; Pruitt & Husak, in press), evidence for their shaping behavioural evolution is lacking. Our data comparing the behavioural syndromes of 18 populations revealed that populations separated by as much as 36o latitude shared common elements, both in sign and in magnitude (Table 2; Fig. 2), of their syndrome. Furthermore, our population-level comparisons of mean trait values revealed significant correlations in five of the six components of the syndrome, with the among population correlations being in the same direction as the within-population correlations. Thus, the A. studiosus syndrome, relative to other study systems, is remarkably stable and holds explanatory power for population-level variation in behaviour. These data are consistent with the hypothesis that trait correlations have played a role in shaping population-level trait divergence in this system (adaptive or otherwise) and might act as constraints on evolution. Here and throughout our discussion, we use the term ‘constraint’ to mean a quantitative resistance to change in certain directions, not an absolute limit. Given enough time, a large population size, a significant mutation rate and strong selection, evolution can certainly uncouple correlated traits; however, we emphasize that even modest constraints can influence the direction of evolution and sometimes ‘trap’ a population on a local, but globally suboptimal, fitness peak on an adaptive landscape (Wright, 1932; Schluter, 1996).

Comparative data sets on behavioural syndromes are rare, and the existing data have revealed remarkable inter-population variation in syndromes (Snyder, 1988; Snyder & Dingle, 1989; Dingle, 1994; Bell, 2005; Bell & Sih, 2007; Dingemanse et al., 2007; Herczeg et al., 2009). These population comparisons have been interpreted as evidence that syndromes are readily generated and dissolved over evolutionary time and their importance in shaping evolutionary trends in behaviour is trivial. Our system exhibits the opposite trend (i.e. remarkable consistency in syndromes across populations), even though our data come from a large number of source populations and habitat types. All populations of A. studiosus exhibited correlations between social tendency and latency of response towards prey, latency of attack, and boldness towards predators; the majority of the populations also exhibited significant correlations between social tendency and the number of bites issued to prey items, activity level, and exploratory tendency. We are, however, reluctant to suggest that the syndrome is ‘rigid’ against selection, because the observed behavioural correlations might represent a complex, adaptive plastic response to environment in all the populations sampled in our study. However, this interpretation is at odds with the inference, from other studies, that even minute changes in selective regime can alter behavioural syndromes (Bell & Sih, 2007; Dingemanse et al., 2007; Wolf et al., 2007, 2008). For instance, Dingemanse et al. (2007) sampled 12 populations of the three-spined stickleback (Gasterosteus aculeatus) and found the presence of behavioural syndromes varied predictably with the presence of predators: populations without piscivorous predators lacked a syndrome and populations with piscivorous predators expressed behavioural syndromes. Similarly, Bell & Sih (2007) found that one could experimentally generate behavioural syndromes by adding predators to populations of three-spined stickleback which normally lack syndromes. Although we sampled opportunistically, our source habitats doubtlessly vary in their susceptibility to predators, prey availability, frequency of disturbance, risk associated with dispersal and any number of abiotic factors (e.g. precipitation, temperature, disturbance type). Why selection would favour similar trait correlations among functionally dissimilar behavioural traits, and in such diverse habitat types is unknown, and therefore, the potential of the syndrome as an adaptive constraint cannot be completely disregarded.

Another line of evidence that indicates behavioural syndromes have the potential to influence evolution in our system is the finding that population-level differences in trait values are correlated in the same direction as within population correlations. Shifts in mean inter-individual distance are correlated with shifts in aggressiveness towards predators and prey, as well as activity level (Table 3). These correlations imply a lack of trait independence for population-level evolution (i.e. a shift in one trait value results in a correlated shift in others). This hypothesis is further supported by recent neurochemical work on A. studiosus which linked two neurocirculatory hormones (octopamine and serotonin) with many of the behaviours in this study (Jones TC unpublished data). Our interdependence interpretation is somewhat weakened, however, by our finding that population differences in latency of attack and activity level are not fully explained by differences in social tendency (Fig. 2; Table 4). This result is driven by three populations ‘N’, ‘O’ and ‘P’. These populations span the North American range of A. studiosus and occupy divergent habitats (Florida Everglades, Temperate Riparian Habitat, Urban Ecosystem). Expression of both traits can be heavily influenced by local weather patterns (e.g. humidity, ambient temperature, rainfall) as well as individuals’ proximity to moulting. Moulting is a particularly likely explanation for population differences because A. studiosus populations commonly moult synchronously, and thus, population differences in behaviour might be confounded with differences in the timing of measurement with respect to moult cycles. Whether the observed trends in behaviour are the influence of genes or environment is unresolved, because subjects of our study were not raised in a common garden (see Herczeg et al., 2009 for discussion). However, whether by genetic influences or environmental, the traits behave nonindependently (i.e. the environmental cues influencing social tendency appear to jointly influencing other types of behaviour), and thus, the syndrome has the potential to generate adaptive trade-offs.

The weight of the evidence presented here supports significant concordance of within- and between-population trait correlations. Although these correlations need not imply absolute constraints on behavioural evolution, it appears that populations are most likely to diverge along multivariate ‘lines of least resistance’ (Schluter, 1996) generated by interdependence of traits constituting the behavioural syndromes. This interdependence is noteworthy because it could influence how social behaviour evolves in A. studiosus, and perhaps for spiders in general. For instance, if selection were to favour increased social tendencies in A. studiosus for some populations, selection must be ‘relaxed’ enough in associated traits to permit that change. Thus, sociality might not readily evolve in populations experiencing selection for aggressiveness towards prey and/or predators. The relationship between social tendencies and prey availability has been investigated in related species (A. eximius, A. guacamayos), and as expected, sociality is more common and pronounced in environments with greater prey densities, and particularly, larger prey sizes (Avilés et al., 2007; Guevara & Avilés, 2007; Powers & Avilés, 2007; Purcell & Avilés, 2007; Yip et al., 2008). One interpretation of these data is that sociality is only permitted to evolve in environments where selection for aggressiveness towards prey is reduced, such as the lowland tropics where prey abundance is high. As it happens, ≈98% of social spider species are restricted to the tropics.

Conclusions

Arguably, the most debated development from the behavioural syndromes literature is whether syndromes can influence behavioural evolution over long-time scales, such as the divergence of populations. Previous studies have suggested that behavioural syndromes have little explanatory power for population-level character divergence. In contrast, we document similar syndromes in a large number of geographically disparate populations. Moreover, population differences in behaviour are correlated in the same direction as the within-population syndromes. This finding is consistent with the hypothesis that correlated behavioural traits might be ‘dragged along’ by selection. While correlative, to the best of our knowledge, this represents the strongest evidence for behavioural syndromes potentially influencing large-scale evolutionary patterns of behaviour.

Acknowledgments

We thank the National Science Foundation Animal Behaviour Program (Grant Number: 0235311) and the University of Tennessee Ecology and Evolutionary Biology Department (summer research grant issued to JNP) for their economic support. We acknowledge the Ministerio del Ambiente of Ecuador for issuing our permits (Permit numbers: 013-2008-IC-FAU-DRFP/MA) and the Escuela de Ciencias Biológicas, Pontificia Universidad Católica del Ecuador, for their sponsorship of our research in Ecuador. We are also indebted to Kyle Demes, Alison Bell, the editor, and one anonymous reviewer for their thoughtful comments on a previous version of this manuscript. We thank Thomas Jones for sharing his unpublished data.

Ancillary