Does predation select for or against avian coloniality? A comparative analysis

Authors


  • Present address:É. Danchin, Laboratoire d’Évolution et Diversité Biologique, Université Paul Sabatier, CNRS-UMR 5174, 118 Route de Narbonne, Bât. IV R3, 31000 Toulouse, France

Susana A. M. Varela, Laboratoire d’Évolution et Diversité Biologique, Université Paul Sabatier, CNRS-UMR 5174, 118 Route de Narbonne, Bât. IV R3, 31000 Toulouse, France.
Tel.: 0033561556758; fax: 0033561557327; e-mail: varela@cict.fr

Abstract

Some studies have supported predation as a selective pressure contributing to the evolution of coloniality. However, evidence also exists that colonies attract predators, selecting against colonial breeding. Using comparative analyses, we tested the reduced predation hypothesis that individuals aggregate into colonies for protection, and the opposite hypothesis, that breeding aggregations increase predation risk. We used locational and physical characteristics of nests to estimate levels of species’ vulnerability to predation. We analysed the Ciconiiformes, a large avian order with the highest prevalence of coloniality, using Pagel's general method of comparative analysis for discrete variables. A common requirement of both hypotheses, that there is correlated evolution between coloniality and vulnerability to predation, was fulfilled in our data set of 363 species. The main predictions of the reduced predation hypothesis were not supported, namely that (1) solitary/vulnerable species are more prone to become colonial than solitary/protected species and (2) colonial/protected species are more likely to evolve towards vulnerability than solitary/protected species. In contrast, the main predictions of the increased predation hypothesis were supported, namely that colonial/vulnerable species are more prone (1) to become protected than solitary/vulnerable species and/or (2) to become solitary than colonial/protected species. This suggests that the colonial/vulnerable state is especially exposed to predation as coloniality may often attract predators rather than provide safety.

Introduction

Colonial breeding is a form of social reproduction in which individuals aggregate within densely distributed breeding territories that contain no resources other than nesting sites (Perrins & Birkhead, 1983; Wittenberger & Hunt, 1985; Siegel-Causey & Kharitonov, 1990). It is a widespread phenomenon in animals, occurring frequently in birds in which approximately 13% (Lack, 1968) to 19% (Crook, 1965) of species breed in colonies. Despite decades of study, however, we still lack a consensus on how coloniality has evolved (see reviews in Pulliam & Caraco, 1984; Wittenberger & Hunt, 1985; Siegel-Causey & Kharitonov, 1990; Danchin & Wagner, 1997; Brown & Brown, 2001; Danchin & Giraldeau, 2005).

The oldest explanation of coloniality is that high density breeding reduces nest predation (e.g. Kruuk, 1964; Patterson, 1965; Alexander, 1974; Hoogland & Sherman, 1976; Pulliam & Caraco, 1984; Krebs & Davies, 1993; Wiklund & Andersson, 1994; Brown & Brown, 1996). Reduced nest predation in colonies might result from dilution effects (e.g. Darling, 1938; Hamilton, 1971; Ims, 1990; Murphy & Schauer, 1996), group vigilance (e.g. Brown & Brown, 1987; Møller, 1987; Ferrière et al., 1996; Roberts, 1996) and communal defence (e.g. Smith & Graves, 1978; Gotmark & Andersson, 1984; Elliot, 1985; Robinson, 1985). Evidence of such benefits has been used to hypothesize that colonies have an anti-predator function. Despite such studies, however, support for the reduced predation hypothesis has been contradicted by examples of predation increasing with nesting density (e.g. Tinbergen et al., 1967; Birkhead & Hudson, 1977; Burger, 1984; Bellinato & Bogliani, 1995; Brown & Brown, 1996). Increased nest predation in colonies might result from the fact that colonial nests are more conspicuous to predators than solitary nests because of visual, acoustic and olfactory cues, leading to higher attack rates in colonies (Lack, 1968; Wittenberger & Hunt, 1985; Rodgers, 1987; Brown & Brown, 2001). Hence, rather than providing a defence against predators, colonies may increase the net vulnerability of individual nests (e.g. Hunter & Morris, 1976; Burger, 1977; Rodgers, 1987; Clode, 1993; Birkhead & Nettleship, 1995; Brunton, 1997). Therefore, nest predation may be either reduced or increased by coloniality, leaving this question unresolved.

All of these studies have examined the role of nest predation within species and are valuable for estimating current selective pressures. However, the observation of current predation rates in colonies does not allow us to distinguish between selective pressures operating at the origin of coloniality from mechanisms that developed after colonies formed and that may now participate in maintaining coloniality. Phylogenetic analyses provide a way to examine the effect of selection on the evolution of various traits (Harvey & Pagel, 1991), such as coloniality. Given the limitations of studies of extant systems, phylogenetic comparative analyses are probably the best method for answering the historical questions of how colonial breeding has evolved in relation to species structure, function and behaviour (Brown & Brown, 2001).

To our knowledge, only three studies have applied these methods to analyse coloniality within avian phylogenies. Eberhard (2002) analysed how nest site limitation may constrain coloniality in lineages of cavity-nesting species that do not excavate their own holes, and Dubois et al. (1998) studied the relationship between coloniality and mate fidelity in waterbirds, but neither study examined predation. Rolland et al. (1998) analysed the evolution of coloniality in relation to a number of selective pressures. They were able to analyse the effects of food and habitat on the evolution of coloniality; however, they were unable to draw conclusions about causality between nest predation and coloniality. As Rolland et al. (1998) suggested, this inability was probably due to a lack of power of their test because they quantified nest predation risks by combining two types of data, nest predation rates when available and nest characteristics when rates were not available.

Current rates of nest predation during reproduction are the result of the interactions among three different factors: the prevalence of predators, the adaptations evolved by animals to escape predation, and the species’ intrinsic risk of nest predation. The prevalence of predators is likely to vary both in space and time and no documentation exists of predator abundance in the distant past. It is thus impossible to evaluate directly the impact of variation in predator density over evolutionary time. The evolutionary arms race between predators and prey also makes it difficult to infer the past from the current predation rates. Furthermore, coloniality itself may be a predator avoidance strategy. Thus the use of current predation rates to quantify the impact of predation on evolutionary processes could generate spurious correlations with coloniality, which might have limited the explanatory capacity of Rolland et al.’s (1998) study. One way to escape these limitations is to use the variables that determine intrinsic predation risks such as nest characteristics. Such variables include the vulnerability of nest contents, and the nest's placement among the different layers of the habitat.

Several authors have shown that the locational and physical characteristics of nests can reflect vulnerability to predators (Lack, 1968; Ricklefs, 1969; Collias & Collias, 1984; Rodgers, 1987; Martin, 1988a,b, 1993a,b, 1995; Martin & Li, 1992; Owens & Bennett, 1995; Martin & Clobert, 1996; Poiani & Pagel, 1997). In particular, Martin (1995) has shown that traits associated with reproductive success, such as annual fecundity, number of broods, clutch size and adult survival vary predictably among nest sites and nest types. Furthermore, Owens & Bennet (1995) suggest that the diversification in nest types among ancient avian lineages has led to differential nestling mortality over evolutionary time and subsequently, to an explosive radiation in life history strategies among species.

Such studies demonstrate that analysing nest characteristics as a proxy of predation risk is better than using predation rates, because predation rates are likely to vary more rapidly than nest characteristics, so that nest characteristics are probably the best predictor of mean predation pressures over evolutionary time. Moreover, unlike predation rates, nest characteristics are available for the vast majority of bird species. Many species can thus be classified as being vulnerable to predation or not. Changes in nest characteristics can then be traced over phylogenies in parallel with changes in coloniality.

Here we use a comparative approach (Pagel, 1994) to test the two opposite hypotheses that predation selects for or against coloniality. We used a new coding of nest vulnerability to test predictions on a large portion of the avian phylogeny, the order Ciconiiformes, which constitutes a polyphyletic group showing the highest incidence of coloniality in birds (Siegel-Causey & Kharitonov, 1990).

Implications and predictions

We first tested the common requirement of both the reduced and the increased predation hypotheses that there is a significant correlation between coloniality and nest exposure to predation (Table 1).

Table 1.   Predictions of the reduced (RPH) and the increased (IPH) predation hypotheses. The requirement of correlated evolution between coloniality and nest vulnerability to predation is common to both hypotheses. Predictions 1a and 1b of contingent evolution from the RPH, state that species aggregate for the benefit of reduced predation. Predictions 2a and 2b of contingent evolution from the IPH, state that colonial breeding at vulnerable sites increases nest predation risk. See also Fig. 1a.
 Hypotheses tested
Requirement of both hypotheses
 Correlated evolutionLR ∼ X2 with 4 d.f.Coloniality is not independent from nest vulnerability to predation
Predictions
  Contingent evolution
   Reduced predation hypothesis1a (q13 > q24)Solitary/vulnerable spp. are more prone to become colonial than solitary/protected spp.
 1b (q43 > q21)Colonial/protected spp. are more prone to become vulnerable than solitary/protected spp.
   Increased predation hypothesis2a (q31 > q42)Colonial/vulnerable spp. are more prone to become solitary than colonial/protected spp.
 2b (q34 > q12)Colonial/vulnerable spp. are more prone to become protected than solitary/vulnerable spp.

The reduced predation hypothesis is that coloniality reduces the vulnerability of individual nests because of group defence, implying that solitary/vulnerable species are the most exposed to predation. This predicts two contingent evolutions across species over evolutionary time (Table 1 and Fig. 1a). Prediction 1a: Solitary/vulnerable species are more prone to become colonial than solitary/protected species. Prediction 1b: Colonial/protected species are more likely to evolve to a vulnerable state than solitary/protected species.

Figure 1.

 Flow diagrams illustrating correlated evolution between coloniality and species nest vulnerability to predation. (a) The predictions of the reduced (RPH) and the increased (IPH) predation hypotheses, and (b)–(d) show the best alternative models presenting the transition rates involved in the evolution of coloniality in relation to (b) nest position, (c) nest accessibility and (d) nest type (for results on nest approachability see Appendix 1 online). The subscripts in the qij’s refer to the beginning and ending character-states: state 1 for solitary/vulnerable species, state 2 for solitary/protected species, state 3 for colonial/vulnerable species and state 4 for colonial/protected species. In (a), colours refer to the different predictions. Green arrows are for prediction 1a of the RPH (q13 > q24), blue for prediction 1b of the RPH (q43 > q21), red for prediction 2a of the IPH (q31 > q42), and yellow for prediction 2b of the IPH (q34 > q12). The thickness of the arrows indicates which transition is expected to be the most common (see formalization in the middle of the flow diagrams, and Table 1). In (b)–(d), colours refer to predictions of (a), and arrows’ thickness indicates results (see formalization in the middle of the flow diagrams). Black arrows indicate the transitions for which contingency was rejected.

The increased predation hypothesis assumes that coloniality increases the intrinsic vulnerability of individual nests because of increased nest conspicuousness in colonies, implying that colonial/vulnerable species are the most exposed to predation. This also predicts two contingent evolutions across species over evolutionary time (Table 1 and Fig. 1a). Prediction 2a: Colonial/vulnerable species are more prone to become solitary than colonial/protected species. Prediction 2b: Colonial/vulnerable species are more likely to evolve to protected nests than solitary/vulnerable species.

Materials and methods

Quantifying nest vulnerability to predation and coloniality

To measure the intrinsic predation risk of species we followed Martin (1988b, 1993b, 1995) by quantifying the physical properties of the immediate environment surrounding the nest (Table 2). Although Martin mainly examined North American passerines, similar conclusions were drawn for palearctic birds and colonial species in particular (Lack, 1968; Rodgers, 1987). In forests, nest predation is on average higher in shrub nests, lower in ground nests and intermediate in sub-canopy/canopy nests. In shrub/grassland habitats, nest predation is on average higher in ground nests than in above ground nests. Among shrub nests, nest predation does not vary significantly between forest, shrub/grassland and marsh habitats (Martin, 1988b, 1993b, 1995). We also addressed other habitats, such as deserts, savannahs, mountains, seacoasts, islands, tundra, steppe, several other types of wetlands other than marshes and human settlements. Because nest predation is generally higher in nests with scarce foliage concealment and low foliage density around the nest (Martin, 1988b), we assumed that ground nests in the open (any type of uncovered terrain placed in any type of habitat) are at least as vulnerable as ground nests in shrub/grassland habitats. In vertical habitats, such as mountains, seacoasts and human settlements, nests placed on cliffs, steep slopes, ravines, canyons and buildings are known to be relatively safe (Lack, 1968). The same is assumed for nests placed very high on trees, on pylon platforms and other similar natural or artificial structures (Lack, 1968; T. E. Martin personal communication). Nests that are situated on islands, or the Antarctic mainland, as well as nests that are surrounded by water or placed over water, generally have low predation (Lack, 1968; Rodgers, 1987). Finally, closed nests (domed, hole and cavity nests) are known to suffer much less predation than any type of open nests (Martin & Li, 1992; Martin & Clobert, 1996). We thus built a nine state ordered variable that we dichotomized, for methodological reasons, in four ways according to where we placed the limit between vulnerable and protected nests (Table 2).

Table 2.   Distribution of the species in our data set relative to coloniality and nest vulnerability to predation. The last four columns provide the four thresholds that we used to define the limit between vulnerable and protected species. Cases 2–5 and case 9 are based on Martin (1988b, 1993b, 1995); Martin & Li (1992) and Martin & Clobert (1996). Cases 6–8 are based mostly on Lack (1968). To classify case 1 we extrapolated from case 2. Protection against predators increases from case 1 to 9, with ground nesting species in the open being the most vulnerable situation and closed nests being the most protected situation.
Nest placement descriptionNest vulnerability to predation
Nest positionNest ApproachabilityNest AccessibilityNest type
1. Ground nests in the open (any type of uncovered terrain placed in any type of habitat)Exposed position (92 spp.)Easy to approach (163 spp.)Accessible nests (185 spp.)Open nests (288 spp.)
2. Ground nests in semi-open habitats (heathland, grassland, steppe, savannah, tundra, etc.)    
3. Shrub nests (< 2 m off the ground)    
4. Subcanopy and canopy nests (> 2 m off the ground)Less exposed position (271 spp.)   
5. Ground nests in forest (and other concealed) habitats    
6. Nests placed on tops of trees and features, on cliffs, steep slopes, buildings and other similar structures Difficult to approach (200 spp.)  
7. Nests surrounded by water or placed over water  Inaccessible nests (178 spp.) 
8. Nests placed anywhere on isolated sea islands    
9. Domed and cavity nests (on burrows, hollows, holes, crevices and caves) placed anywhere   Closed nests (75 spp.)

Nest position:  Estimates predation risk at the scale of the vegetation that immediately surrounds the nest and its position within the different layers of the habitat. Ground nests in open and semi-open habitats, as well as shrub nests in any kind of habitat (coded 0), are assumed to be more exposed, and hence more vulnerable to predation, than nests placed elsewhere (coded 1) (Martin, 1988b, 1993b, 1995). Nests were classified as shrub nests if they were < 2 m above the ground (Martin, 1993b).

Nest approachability: Estimates the degree of nest approachability by predators within the different layers of the habitat. Ground, shrub and subcanopy/canopy nests in any kind of habitat (coded 0) are assumed to be easier to approach and hence more vulnerable to predation, than nests placed elsewhere (coded 1) (Martin, 1988b, 1993b, 1995).

Nest accessibility:  Estimates the degree of nest accessibility to predators when nests are surrounded by water or placed over water (nests placed anywhere on islets in rivers, lakes and lagoons, or on oceanic islands and the Antarctic mainland), as well as when nests are closed (coded 1). These are assumed to be less accessible than nests placed elsewhere (coded 0) (Lack, 1968; Rodgers, 1987).

Nest type:  Estimates the vulnerability of the nest contents once the predator has approached within detection distance. Any type of closed nests (coded 1) is assumed to be much less vulnerable to predators than any type of open-nests in any type of habitat (coded 0) (Martin & Li, 1992; Martin & Clobert, 1996).

Each of these variables estimates nest vulnerability to predation at different scales, with nest vulnerability increasing in the following order (Table 2): closed nests < nests that are protected by water < nests that are hard to approach from the ground < nests in concealed habitats < nests in semi-open and open habitats. As nest approachability and nest accessibility yielded similar results, we chose to show the results only concerning nest position, nest accessibility and nest type. For further details about approachability, see Appendix 1 on the online edition of J. Evol. Biol.

To code coloniality, we applied Perrins & Birkhead's (1983) definition of coloniality, which is that species are colonial when their breeding territories are aggregated and contain no resources other than nesting sites. Species were assigned 0 when solitary and 1 when colonial. Species that breed in small and/or loose colonies were coded as colonial breeders.

Data were collected from the literature (Palmer, 1962; Cramp & Simmons, 1977–1983; Perrins & Birkhead, 1983; Cramp, 1985–1992; del Hoyo et al., 1992–1997; Cramp & Perrins, 1993–1994), and can be obtained from the online edition of J. Evol. Biol. (Appendix 2).

Phylogeny and selected taxa

We chose the Ciconiiformes (Fig. 2) because it is the portion of the avian phylogeny with the highest prevalence of coloniality (Siegel-Causey & Kharitonov, 1990), where 23 of the 30 families contain colonial species. This order includes all the seabirds, in which nearly all species are colonial (Lack, 1968; Wittenberger & Hunt, 1985; Siegel-Causey & Kharitonov, 1990), but also a large number of solitary species and nonmarine colonial species, as well as a large diversity in nest types. Such variability across species provides maximum statistical power for comparative analyses. Although there is an important correlation between marine life and colonial breeding, it was shown that the appearance of coloniality preceded the transition to the marine habitat (Rolland et al., 1998), suggesting that despite the high percentage of aquatic species in the Ciconiiformes, we can use this taxon to study the evolution of coloniality independently of habitat type.

Figure 2.

 Phylogenetic tree of the 30 Ciconiiformes families involved in this study. The first column includes the family names, followed by species common names and species number within each family group. The last column lists the number of species included in this study, specifying how many are colonial.

We used two different phylogenies as a way to examine the robustness of the results. Both are composite phylogenies of the Ciconiiformes. We used Sibley & Ahlquist's (1990) phylogeny as the trunk of our composite trees which is the only one available at the scale of the entire Ciconiiformes’ order. We used it to branch families. For the genus and species branches, we used more recent and complementary phylogenies, with most of the transitions of our composite-trees coming from these more recent data. For all analyses, we used the composite trees resolved to the species level. The first composite phylogeny involved eleven sub-phylogenies based on nuclear and mitochondrial DNA hybridization (Moum et al., 1994; Nunn et al., 1996; Bretagnolle et al., 1998; Nunn & Stanley, 1998; Slikas, 1998; Griffiths, 1999; Crochet et al., 2000; Whittingham et al., 2000; Wink & Sauer-Gürth, 2000). The second composite phylogeny involved 10 different sub-phylogenies based on either molecular or morphological information (Friesen & Anderson, 1997; Slikas, 1998; Kennedy et al., 2000; Kennedy & Page, 2002; Giannini & Bertelli, 2004; Kennedy & Spencer, 2004; Thomas et al., 2004; Wink & Sauer-Gürth, 2004; Lerner & Mindell, 2005). The second phylogeny incorporates the most recent data available in the literature, namely the shorebird (Thomas et al., 2004) and the seabird (Kennedy & Page, 2002) supertrees. Despite several substantial differences between the two phylogenies, results were very similar using either one. We thus present results obtained from the most recent phylogeny. That phylogeny comprised 363 species of all 30 Ciconiiformes families (Fig. 2), representing 35% of all species belonging to this order. It can be obtained from the online edition of J. Evol. Biol. (Appendix 3) in a Mesquite-version 1.05 format file (Maddison & Maddison, 2004).

As in other comparative studies (Garland et al., 1993; Dubois et al., 1998; Rolland et al., 1998; Cézilly et al., 2000) we did not account for branch lengths because our tree was built from different phylogenies that were derived from different methods. The assumption of equal branch lengths has been demonstrated to be conservative (Garland et al., 1993; Pagel, 1994). Furthermore, several authors have specifically tested the effect of this assumption and found that it does not affect the results qualitatively (e.g. Poiani & Pagel, 1997; Møller et al., 1998; Nunn, 1999; Poulin, 1999).

Translating predictions in terms of comparative methods

To analyse predictions of both the reduced and the increased predation hypotheses, we used the general method of comparative analysis for discrete variables (Pagel, 1994). This method analyses evolutionary changes between two binary characters along branches of a phylogeny using a continuous-time Markov model. An advantage of this approach is that uncertainty in the ancestral state reconstructions is automatically taken into account in all likelihood calculations. The method estimates the probabilities of any kind of change in any branch of the tree, which can be used to chart the most likely course of evolution from the ancestral state to the derived state (the flow diagram of the evolutionary changes) (Pagel, 1994) (Fig. 1). Transition rate parameters (qij) in the flow diagram denote the rate of change from state i to state j. The subscripts refer to the beginning and end character states for each particular transition, where 1 = 0,0 (i.e. the solitary/vulnerable state); 2 = 0,1 (solitary/protected); 3 = 1,0 (colonial/vulnerable); 4 = 1,1 (colonial/protected).

Predictions of both hypotheses can thus be formulated as transition parameters (qij) that should take different transition rates. Testing Prediction 1a thus boils down to the testing of q13 being higher than q24. Testing prediction 1b leads to q43 > q21. Testing Prediction 2a leads to q31 > q42. And testing Prediction 2b leads to q34 > q12 (Table 1 and Fig. 1a).

Applying Pagel's comparative method

We used the likelihood ratio (LR) test statistic (Pagel's, 1994 omnibus test) to analyse correlated evolution between nest vulnerability and coloniality. This test compares the likelihood of a model of independent evolution (with four parameters, LI4) to the likelihood of a model of dependent evolution (with eight parameters, LD8), and selects the one that better fits the data. If the additional parameters in the model of dependent evolution are significant, the LR statistic will be asymptotically distributed as a chi-squared variate with four degrees of freedom (Pagel, 1994, 1997). If the model of correlated evolution with eight parameters was significant (compared to the null hypothesis of independent evolution), we determined whether it incorporated the predictions of the reduced predation (q13 > q24 or q43 > q21) and the increased predation (q31 > q42 or q34 > q12) hypotheses (Table 1 and Fig. 1a). To accomplish this, we tested different restricted models of contingent evolution with less than eight parameters (Pagel, 1994), in which the specific transitions were restricted by imposing the following equalities: q13 = q24; q43 = q21; q31 = q42; q34 = q12. We built these models with different numbers of restrictions, starting from models with only one restriction to models incorporating all restrictions (Table 3).

Table 3.   Akaike Information Criterion (AIC) values of the various models built to test predictions of contingent evolution between coloniality and nest vulnerability to predation. Species intrinsic predation risk was quantified at four different scales: nest position, nest approachability, nest accessibility and nest type (see also Table 2). The contingency models’ column shows first the restricted models that were tested for three of the four codings of nest vulnerability to predation (for the results on nest approachability see Appendix 1 online), and secondly, the alternative restricted models that were built by incorporating the restrictions from the best restricted models (in bold), as well as by setting the transition parameters with similar estimates to be equal to each other (for a better understanding of the contingency-test models, see also Table 1 and Fig. 1a). The AIC statistic (Akaike, 1973, 1974) was used to select between models. The LDn columns are for the likelihoods of the restricted models, and the AIC columns are the correspondent AIC statistic for each model. Bold AIC values indicate the models that describe the data best at every stage in the model selection phase. AIC values with asterisks indicate the best alternative models that were used to construct the flow diagrams in Fig. 1b–e.
Contingency-test modelsNest vulnerability to predation
PositionAccessibilityType
LDnAICLDnAICLDnAIC
Restricted modelsq13 = q24D7215.65230240.08254193.79208
q43 = q21D7216.65231241.88256196.45210
q31 = q42D7218.97233243.09257195.03209
q34 = q12D7216.95231244.89259193.89208
q13 = q24; q43 = q21D6216.91229242.27254197.13209
q13 = q24; q31 = q42D6219.30231243.17255195.07207
q13 = q24; q34 = q12D6217.38229244.89257193.97206
q43 = q21; q31 = q42D6219.65232248.92261201.61214
q43 = q21; q34 = q12D6217.12229246.88259196.85209
q31 = q42; q34 = q12D6223.31235249.25261195.33207
q13 = q24; q43 = q21; q31 = q42D5219.97230249.08259201.64212
q13 = q24; q43 = q21; q34 = q12D5217.40227246.94257197.45207
q13 = q24; q31 = q42; q34 = q12D5223.40233249.48259195.37205
q43 = q21; q31 = q42; q34 = q12D5223.68234257.42267202.00212
q13 = q24; q43 = q21; q31 = q42; q34 = q12D4223.77232258.96267202.21210
Alternative restricted models
 Nest positionq21 = q42 = q24 = q43 = q13; q12 = q34; q31D3218.866225*    
 Nest accessibilityq24 = q42 = q12 = q43 = q13; q31; q21; q34D4  241.40249*  
q24 = q42 = q12 = q43 = q13 = q21; q31; q34D3  245.18251  
q24 = q12 = q43 = q13; q42 = q21 = q31; q34D3  250.04256  
 Nest typeq42 = q12 = q34 = q24 = q43 = q13 = q31; q21D2    198.96203
q42 = q12 = q34; q24 = q43 = q13 = q31; q21D3    194.30200*

To select the restricted model of contingent evolution that best fitted the data we used the Akaike Information Criterion Statistic (AIC, Akaike, 1973, 1974; de Leeuw, 1992; Burnham & Anderson, 1998) (Table 3). AIC is an efficient and very general method for model selection that accounts for the principle of parsimony. It is defined as twice the maximized log-likelihood for a given model, plus twice the number of parameters in that model. The lower the AIC of a given model the better it fits the data, and the larger the difference between two AIC values [ΔAICi = AICi– min(AICi)] the more significantly different they are. Models with ΔAIC < 2 cannot be considered as fitting the data differently and should all be accepted. When ΔAIC ≥ 2, the higher AIC model has less support and can be ignored (Burnham & Anderson, 1998). As the AIC selection method does not estimate P-values, when possible, we give the P-values provided by the LR statistics selection method (see Appendix 1 online).

We then built what we call alternative restricted models that incorporated further restrictions, by setting parameters with similar transition rates to be equal to each other (Table 3). Finally, we used the AICs of all the restricted and alternative model(s) to select those that best fitted the data (Table 3 in bold with asterisk). Our aim was to characterize for each variable (nest position, approachability, accessibility and type), the most probable relationship of contingent evolution between coloniality and nest vulnerability (Fig. 1). The LR statistics for each of the restricted models is given as

One problem with maximum-likelihood methods for discrete characters is that the results may be biased by the relative frequency of traits across the phylogeny (Nosil & Mooers, 2005). Hence, the frequency rather than the distribution of trait values over the tree may explain results. To rule out this possibility, we followed Kolm et al. (2006)’s randomization method. We randomized the distribution of coloniality and nest vulnerability to predation across our phylogenetic tree 50 times independently while leaving the tree unchanged. We used these 50 randomized distributions of trait values to redo the analyses. We did this for each one of the variables in the study (nest position, approachability, accessibility and type), making a total of 200 tests. Our aim was to investigate how often the random species values alone would yield the same result as those obtained with the species’ actual trait values. As we found no significant correlation between coloniality and nest vulnerability to predation in any of the 50 randomized datasets, we conclude that our results are not due to the frequency distribution of trait values in our data set, and are thus biologically sound (for further details on this analysis see Appendix 1 online).

Finally, for root reconstruction of ancestral states we used Pagel's (1994, 1999a,b) maximum-likelihood reconstruction method. It determines the ancestral state of a binary character at the root of the tree that maximizes the probability of arriving at the observed states in the terminal taxa, given the independent model of evolution.

All analyses were performed with ‘Discrete’ (version 4.0, Pagel, 2000) the details of which can be found in numerous publications (Pagel, 1994, 1997, 1999a,b). Note that during the analyses no parameter restrictions were made other than those stated here. Similarly, we did not make assumptions about the root values of characters. To edit trees into the Discrete format file, to estimate the number of changes (i.e. mutations) along the phylogeny, and to generate the randomized character matrices, we used Mesquite (version 1.05) (Maddison & Maddison, 2004).

Results

Among the 363 species we examined of the 30 Ciconiiformes families, 190 were solitary and 173 were colonial (Fig. 2). Reconstruction of ancestral states for coloniality gave solitary breeding as the most likely ancestral state, with a probability of 99.6% (LR = 5.40, d.f. = 1, P < 0.025). The reconstruction of ancestral states of nest type could not be determined with certainty although the trend suggests that nests were probably open (89.3%; LR = 2.12, d.f. = 1, P < 0.15), exposed (74.5%; LR = 1.07, d.f. = 1, P < 0.30), easy to approach (83%; LR = 1.61, d.f. = 1, P < 0.30) and accessible (68.1%; LR = 0.76, d.f. = 1, P < 0.50). Overall, there were many evolutionary transitions along the tree (a minimum of 23 evolutions and eight reversals for coloniality, 11 evolutions and 14 reversals for nest position, 19 evolutions and 30 reversals for nest approachability, 19 evolutions and 18 reversals for nest accessibility, and 15 evolutions and four reversals for nest type), a situation favourable for the statistical testing of our predictions.

Coloniality and nest position

We found a significant correlation between coloniality and nest position (LI4 = −223.77, LD8 = −215.00, LR = 10.13, d.f. = 4, P < 0.05), fulfilling a requirement of both hypotheses that the evolution of coloniality is not independent from nest position.

Among the possible restricted models of contingent evolution, one significantly differed from all the others (ΔAIC ≥ 2). It involved the following restrictions: q13 = q24, q43 = q21 and q34 = q12 (in bold in Table 3). This model was then used to construct an alternative model, incorporating the same restrictions and by setting parameters with similar transition rates to be equal to each other. We thus obtained an alternative restricted model that explained the data significantly better than all the other models (ΔAIC ≥ 2), where q21 = q42 = q24 = q43 = q13 and q12 = q34 (Table 3 in bold with asterisk). Therefore, the flow diagram of the evolutionary changes resulting from this selected model characterizes the most probable relationship of contingent evolution between coloniality and nest position (Table 3; Fig. 1b).

Thus, among the four predictions, only Prediction 2a (q31 > q42), that colonial/exposed species are more likely to become solitary/exposed than colonial/unexposed species, was accepted. Conversely, Predictions 1a (q13 > q24), 1b (q43 > q21) and 2b (q34 > q12) were all rejected (Table 1 and Fig. 1a and b). Therefore, when using nest position, the reduced predation hypothesis was not supported. On the contrary, the colonial/vulnerable state appeared especially exposed to predation, as expected under the increased predation hypothesis, with species diverging from that state by reversing towards solitary breeding.

Coloniality and nest accessibility

We found a significant correlation between coloniality and nest accessibility (LI4 = −258.96, LD8 = −239.87, LR = 20.19, d.f. = 4, P < 0.001), which fulfils the requirement of both hypotheses that the evolution of coloniality is not independent from nest accessibility.

Three of the restricted models fitted the data equally well (ΔAIC ≤ 1) (Table 3 in bold). All three models were then used to construct alternative ones by setting parameters with similar transition rates to be equal to each other. The AIC of one of these alternative models was significantly lower than the two others (ΔAIC ≥ 2) and incorporated the parameters q24 = q42 = q12 = q43 = q13 (Table 3 in bold with asterisk; Fig. 1c).

Both predictions of the increased predation hypothesis were accepted: Predication 2a (q31 > q42) that colonial/accessible species are more prone to become solitary than colonial/inaccessible species, and Prediction 2b (q34 > q12) that colonial/accessible species are more prone to become inaccessible than solitary/accessible species. In contrast, both predictions of the reduced predation hypothesis were rejected: Prediction 1a (q13 > q24) and Prediction 1b (q43 > q21) (Table 1, Fig. 1a and c).

In summary, the analyses of nest accessibility led to the rejection of the reduced predation hypothesis and provided support of the increased predation hypothesis. Thus the colonial/vulnerable state appears especially exposed to predation, with species diverging from it either by evolving to the colonial/inaccessible state, or by reversing to the solitary/accessible state. Moreover, the best alternative model incorporated q43 < q21, which contradicts Prediction 1b (q43 > q31), providing additional evidence that colonial/accessible species are more exposed to predators than solitary/accessible species.

Coloniality and nest type

We found a significant correlation between coloniality and nest type (LI4 = −202.21, LD8 = −193.66, LR = 10.71, d.f. = 4, P < 0.03), again fulfilling the requirement of both hypotheses that the evolution of coloniality is not independent from nest type.

Two of the restricted models fitted the data equally well (ΔAIC = 1) (Table 3 in bold). Both models were then used to construct alternative ones by setting parameters with similar transition rates to be equal to each other. This led to the selection of a single alternative restricted model where q42 = q12 = q34 and q24 = q43 = q13 = q31 (Table 3 in bold with asterisk) (ΔAIC ≥ 3). Therefore, the flow diagram of the evolutionary changes between coloniality and nest type incorporated q13 = q24; q43 < q21; q31 > q42; q34 = q12 (Fig. 1d).

Only Prediction 2a (q31 > q42), that colonial/open-nesting species are more likely to become solitary than colonial/closed-nesting species, was accepted. Predictions 1a (q13 > q24), 1b (q43 > q21) and 2b (q34 > q12) were rejected (Table 1, Fig. 1a and d).

Therefore, the use of nest type leads to the rejection of both predictions of the reduced predation hypothesis, and provides evidence that the colonial/vulnerable state is especially exposed to predation. Colonial/vulnerable species diverged from that state usually by reversing towards solitary breeding. Moreover, the best alternative model incorporated q43 < q21, which contradicts Prediction 1b (q43 > q31), suggesting that the colonial/vulnerable state is more exposed to nest predators than the solitary/vulnerable state.

Discussion

We used a comparative approach to test the two opposite hypotheses that predation pressure selects for or against avian coloniality. We analysed species from all 30 families of the Ciconiiformes, the order with the highest prevalence of colonial breeding, with 77% of families exhibiting coloniality. In our data set of 363 species, approximately half (48%) breed colonially. The mapping of coloniality and nest vulnerability to predation on our phylogenies revealed numerous transitions which have provided substantial statistical power.

Phylogenetic assumption and methods

We used the criteria identified in various bird taxa for categorizing the vulnerability of nest types (Lack, 1968; Rodgers, 1987; Martin, 1988b, 1993b, 1995; Martin & Li, 1992). Some categories overlapped in nest vulnerability according, for example, to whether ground nests in unconcealed habitats were open or closed, or whether open nests were surrounded by water or not. To reduce the possible effects of such overlaps, we divided vulnerability vs. protection at four different cut-off points (Table 2). These analyses are not independent but were performed to evaluate whether or not the location of cut-off points between vulnerability and protection influence the results. The four different codings of predation risk yielded qualitatively similar results (Fig. 1b–d and Fig. e in Appendix S1), suggesting that our results are robust.

To test the robustness of our results to variation in tree structure we used two different Ciconiiformes composite-phylogenies. Both used the Sibley & Ahlquist (1990) phylogeny as the families’-trunk, because it is the unique phylogeny available for the Ciconiiformes’ families. This phylogeny was once controversial, mainly because it is based on incomplete information (Houde, 1987; Lanyon, 1992; Sheldon & Gill, 1996), but it is now considered to be the best available in birds for broad based comparative analyses and is widely used (Harshman, 1994; Mooers & Cotgreave, 1994; Cézilly et al., 1998; Rolland et al., 1998; Dubois et al., 1998; Kennedy & Page, 2002; Thomas et al., 2004). As our two composite phylogenies produced very similar results, we report the results from the analyses of the most recent one. In those analyses, only 7% of the 363 species were in the Sibley & Ahlquist (1990) phylogeny. The other 93% of the species were branched according to the recent phylogenies.

Evidence that predation does not select for avian coloniality

The reduced predation hypothesis assumes that coloniality reduces the vulnerability of individual nests, implying that solitary/vulnerable species are most exposed to predation. This hypothesis predicts that (a) solitary/vulnerable species are more prone to become colonial than solitary/protected species and (b) colonial/protected species are more likely to evolve to a vulnerable state than solitary/protected species. The increased predation hypothesis assumes that coloniality increases the vulnerability of individual nests, implying that colonial/vulnerable species are most exposed to predation. This predicts that (a) colonial/vulnerable species are more prone to become solitary than colonial/protected species and (b) colonial/vulnerable species are more likely to evolve to protected nests than solitary/vulnerable species.

Our results suggest that across evolutionary time, coloniality has not generally provided safety from predators in the Ciconiiformes. We found that both predictions of the increased predation hypothesis were supported while both predictions of the reduced predation hypothesis were rejected. This implies that the least protected state is colonial/vulnerable rather than the solitary/vulnerable state. This was the case when using two different phylogenies and when coding nest vulnerability in four ways.

Our results thus suggest that coloniality per se does not protect the immobile eggs and chicks. Furthermore, and contrary to prediction 1a of the reduced predation hypothesis, the evolution toward coloniality is equally likely in vulnerable as in protected solitary species (Fig. 1b–d, prediction 1a), suggesting that mechanisms other than protection against predators were involved in the evolution of colonial breeding. Moreover, our findings (Fig. 1c, prediction 2b) support Lack's (1968) claim that to compensate for the increased susceptibility to predation while still benefiting from coloniality, colonies are generally set in relatively protected sites. Indeed, colonial/accessible species tended to adopt inaccessible nesting sites, thus decreasing their vulnerability without loosing coloniality. This is also in agreement with Wittenberger & Hunt's (1985) argument that at higher population densities concealment is less effective and, hence, the use of inaccessible nest sites might offer an efficient defence against predation. Finally, according to Rodgers (1987), safety does not result from coloniality but from the intrinsic properties of the breeding site. In drought years, colonies over water become accessible to predators and predation increases drastically leading to the abandonment of colonial breeding. Our results show that wherever we place the dichotomy between vulnerability and protection, the colonial/vulnerable state was more prone to becoming solitary than the colonial/protected state (Fig. 1b–d, prediction 2a), suggesting that becoming solitary without changing breeding sites tends to lower predation pressures.

Thus, our study provides evidence against predation as a pressure favouring colonial breeding in Ciconiiformes. Nevertheless, given that nest characteristics are only a proxy of predation, it is still possible that a third, unknown, correlated variable was responsible for the observed effects. However, this is unlikely because many studies have shown that nest characteristics correctly reflect nest vulnerability to predators in various ecological contexts (references in the Introduction). Another question is whether these findings are applicable to other avian groups. It may be useful to perform similar analyses on other orders, depending on the availability of a sufficient number of transitions. If similar results are produced with different avian taxa it would suggest that predation may be a general force selecting against coloniality. In contrast, if analyses of other taxa yield contrasting results it would imply that the relationship between predation and coloniality is more complex and subject to factors such as phylogenetic constraints and ecological or morphological differences among avian groups and predators.

Static vs. dynamic aggregations

Our findings also raise the issue of how predation pressure might affect the formation of other kinds of avian aggregations, such as feeding flocks and communal roosts, as well as aggregations of other animal taxa. Many animals travel in groups such as fish shoals, bird flocks, insect swarms and ungulate herds. It is intuitive that predation selects for such aggregations because of frequent observations of individuals immediately bunching into much tighter groups in reaction to a predator and separating soon after the predator desists. Such cases are well documented (e.g. Seghers, 1974; Kenward, 1978; Duncan & Vigne, 1979; Foster & Treherne, 1981; Couzin et al., 2005).

In contrast, once reproduction has commenced, breeding colonies become static in that breeders cannot immediately further aggregate their immobile nests, offspring or eggs in response to a predator. The fixed nature of colonies may be an important limitation to predator avoidance that dynamic groups do not suffer. This is consistent with studies that have suggested that predation may contribute to the breaking up of dense colonies (e.g. Hunter & Morris, 1976; Burger, 1977; Rodgers, 1987; Birkhead & Nettleship, 1995; Brunton, 1997; reviewed in Brown & Brown, 2001). Thus, we propose considering static and dynamic aggregations separately when examining selective pressures that might favour clustering.

Two approaches to study coloniality

Two general approaches have been used to study coloniality. The first assumes that individuals benefit by joining a group and thus colonies are viewed as serving a function of reduced predation and/or enhanced food finding (e.g. Crook, 1965; Lack, 1968; Ward & Zahavi, 1973; Alexander, 1974; Pulliam & Caraco, 1984; Wittenberger & Hunt, 1985; Siegel-Causey & Kharitonov, 1990; Clode, 1993; Krebs & Davies, 1993; Brown & Brown, 1996, 2001). The second assumes that colonies are byproducts of numerous individuals choosing resources such as breeding habitat and mates in the same location (Danchin & Wagner, 1997; Brown et al., 2000; Wagner et al., 2000; Kosciuch & Langerhans, 2004; Danchin & Giraldeau, 2005; Mikami, 2006). This byproduct approach views benefits of high density breeding as maintaining rather than producing colonies. Our results do not rule out that colonies have provided a measure of safety, but rather that the benefits of group defence have not outweighed the cost of increased conspicuousness to predators.

Predation is one of two selective pressures of the functional approach that have been stressed as a possible explanation of coloniality. The other well studied pressure is produced by the unpredictability of the location of food exploited by most colonial species. The information centre hypothesis (Ward & Zahavi, 1973) proposes that in an unpredictable environment, individuals may benefit from having multiple neighbours which they can follow to food sources. One argument of this hypothesis rests on the observation that most marine birds, which feed on unpredictably located prey, are colonial. This association was interpreted to suggest that solitary marine species had become colonial in order to exploit the unpredictable marine environment by means of information gathering. However, Rolland et al. (1998), using similar phylogenetic analyses on the Ciconiiformes order (as well as other groups), found that current colonial-marine species were already colonial and nonterritorial feeders before exploiting the marine environment. This result contradicts any influence played by the marine environment in the evolution of coloniality in seabirds. However, it does not exclude the possibility that food finding played a role in the evolution of coloniality that took place before the passage to the marine environment, or that the marine environment, and hence food finding, may play a role on the maintenance of coloniality in seabirds. Nevertheless, Rolland et al.’s (1998) study weakens the main arguments for a role of food finding in the evolution of coloniality in general. Rolland et al.’s (1998) study combined with this study provides arguments against the functional approach.

Conclusions

Our results contradict the prediction that reduced predation has favoured coloniality in the Ciconiiformes. We found instead the opposite, that coloniality appears to increase predation risk. Yet colonies exist, which create an enigma. A possible solution may be that when colonies form, the increased conspicuousness of nests may attract predators, leading to colony break-up, as has been observed (see Introduction for references). The repetition of colony formation, breakup and reformation over the short term may ultimately select for solitary breeding across evolutionary time. This may help explain our finding of a large number of reversals from the colonial/exposed state to the solitary/exposed state. We propose that this observable short-term cycle may comprise a mechanism that creates instability at the species level over evolutionary time. In this sense, nest predation played a major role in the evolution and maintenance of coloniality as it was probably responsible for the breaking up of colonies or for the adoption of protected common nesting sites.

Combined with the results of Rolland et al. (1998) on food-finding, our study does not support the assumption that colonies have evolved for their function of reduced predation or enhanced food finding. The alternative approach views aggregation as the by-product of individual decision-making processes such as breeding habitat selection and mate choice (Danchin & Wagner, 1997; Wagner et al., 2000; Danchin & Giraldeau, 2005). Therefore, communal defence and feeding tactics that are frequently assumed to be the primary cause of coloniality, may turn out to be adaptations to group living. Once colonies have formed as byproducts of multiple individuals seeking to benefit from settling in good breeding habitat with good mates, it is possible that benefits of high density breeding (as well as costs) subsequently accrue, possibly contributing to the maintenance (or the disruption) of coloniality over evolutionary time. Thus, a key difference between the functional and byproduct approaches is whether benefits accrue before or after colony formation. Whereas the functional approach implicitly assumes that colonies are byproducts of individuals pursuing benefits such as reduced predation and enhanced food finding, the alternative approach assumes that such benefits are byproducts of coloniality, that is, secondary adaptations to high density breeding. Further phylogenetic analyses incorporating both approaches may help resolve the question of coloniality.

Acknowledgments

We thank J. Clobert, M.D. Pagel and T.E. Martin for help at various stages of this work. S.A.M. Varela has been supported by a PhD grant from the Portuguese Foundation for Science and Technology (FCT) and from the Social European Funding (SEF), as well as the scientific support of the French National Center of Scientific Research (CNRS). E. Danchin and R. H. Wagner have received a CNRS PICS no. 2410 grant. We also thank three anonymous referees for their constructive suggestions on the previous versions of the paper.

Ancillary