Phylogenetic signal, feeding behaviour and brain volume in Neotropical bats

Authors


Abstract

Comparative correlational studies of brain size and ecological traits (e.g. feeding habits and habitat complexity) have increased our knowledge about the selective pressures on brain evolution. Studies conducted in bats as a model system assume that shared evolutionary history has a maximum effect on the traits. However, this effect has not been quantified. In addition, the effect of levels of diet specialization on brain size remains unclear. We examined the role of diet on the evolution of brain size in Mormoopidae and Phyllostomidae using two comparative methods. Body mass explained 89% of the variance in brain volume. The effect of feeding behaviour (either characterized as feeding habits, as levels of specialization on a type of item or as handling behaviour) on brain volume was also significant albeit not consistent after controlling for body mass and the strength of the phylogenetic signal (λ). Although the strength of the phylogenetic signal of brain volume and body mass was high when tested individually, λ values in phylogenetic generalized least squares models were significantly different from 1. This suggests that phylogenetic independent contrasts models are not always the best approach for the study of ecological correlates of brain size in New World bats.

Introduction

Evolutionary studies of brain size have increased our understanding of brain functioning and evolution in relation to behavioural ecology and energetic trade-offs. Over decades, different hypotheses have been proposed to explain the current diversity in brain size. However, identifying the selective pressures that have led to differences in brain size is yet a contentious issue. One of the proposals, the social brain hypothesis, states that sociability (i.e. group size and complexity of social relationships) correlates positively with relative brain size. This hypothesis has found support in different mammalian orders, in addition to primates (Dunbar & Shultz, 2007). Another hypothesis is related to life-history characteristics. Larger brains in mammals correlate with longer gestations, slower maturation and increased lifespans. Evolutionary changes in pre- and post-natal brain growth correlate with the duration and rate of maternal investment, whereas slower life histories in large-brained species appear to be a direct consequence of developmental costs (Barton & Capellini, 2011). A third group of assumptions refer to ecological correlations of brain size. For instance, the environmental change hypothesis predicts that larger brains are correlated with higher survival in novel or changing environments (Sol et al., 2010). Studies conducted in birds and mammals support this hypothesis and suggest that enlarged brains enhance the flexibility behaviour of individuals in potentially adaptive ways (Sol et al., 2005, 2008). Diet and foraging behaviour are also related to brain size. Shifts to more energy-rich or more easily processed diets may be crucial for significant increase in brain size (Aiello & Wheeler, 1995). In small mammals and primates, for example, frugivorous species have larger brain than folivorous species (Clutton-Brock & Harvey, 1980; Harvey et al., 1980). Because fruit is, in general, less predictably and more patchily distributed than leaves, these results suggest that large brains may reflect selection on the cognitive abilities that are involved in exploitation of the environment (Barton, 2006).

Bats have been used in the last decade as a model group in many comparative studies exploring the relationship of brain size with a set of variables (ecological traits, behavioural complexity and energetic variables) that are assumed to act as selective pressures of brain structure and functioning (reviewed in Dechmann & Safi, 2009). Evidence from studies with this group of mammals suggests that the evolution of brain size is a nondirectional process (Safi et al., 2005) that occurs under selective selection, which operates differentially in neurocognitive centres of the brain that are involved in sensory ecology (Safi & Dechmann, 2005). Other surveys have shown that the expensive tissue hypothesis (i.e. brain size is affected by other energetically expensive tissues) does not hold when ecological variables such as habitat complexity are added to the evolutionary models in comparative studies (Dechmann & Safi, 2009).

The recent increase in information on the diet and evolutionary history of bats makes it possible to examine the relationship between feeding habits and brain volume in a phylogenetic framework. One of the earliest studies that examined the relation between foraging strategies and brain size in vertebrates was conducted in 14 families of bats (Eisenberg & Wilson, 1978). The authors found that frugivores of the phylogenetically distant families Pteropodidae and Phyllostomidae had larger brain volumes for a given body weight than aerial insectivores from other families. Nectarivores showed a similar pattern to frugivorous species. These studies confirmed earlier results that were obtained with a reduced group of species (Pirlot & Stephan, 1970). More recently, phylogenetic comparative analyses have shown that phytophagous taxa in general and frugivorous species in particular have larger brains than animalivores (Hutcheon et al., 2002; Jones & MacLarnon, 2004; Ratcliffe, 2009). This pattern has been explained on the basis that food availability and habitat complexity impose greater sensory challenges to phytophagous than to animalivorous bats (Jones & MacLarnon, 2004; Safi & Dechmann, 2005). In fact, the size of brain regions processing auditory stimuli and sensory input related to spatial memory is a function of habitat complexity in echolocating bats (Safi & Dechmann, 2005).

One common characteristic in comparative analyses of brain evolution in bats is that species are treated as members of discrete guilds and dietary categories (Hutcheon et al., 2002; Jones & MacLarnon, 2004; Safi & Dechmann, 2005; Safi et al., 2005; Ratcliffe et al., 2006; Ratcliffe, 2009). Although phytophagous bats apparently have larger brains than animalivorous bats, it is not clear whether brain size increases with increasing proportion of plant material in the diet (Dechmann & Safi, 2009). Recent detailed studies show that the diet of many phytophagous bat species also includes arthropods and meat (Mancina & Herrera, 2010; Rex et al., 2010). Foraging behaviour in these species may be more complex than that in specialist bats, because generalist species perform different feeding strategies and explore greater habitat diversity, including dense habitats. Furthermore, evidence suggests that brain size is not influenced by the nature of the food but by the variation in information associated with diet that has to be processed as part of foraging behaviour (Eisenberg & Wilson, 1978; Harvey & Krebs, 1990; Safi & Dechmann, 2005). If this hypothesis is true, changes in brain size should reflect the effects of levels of diet specialization instead of the effects of diet categories. Levels of diet specialization accommodate the variation in foraging behaviour that is not expressed in diet category. For example, species that feed mainly or strictly on fruit would be classified as frugivorous. This category informs about the main component of the diet of the species, but says little about foraging behaviour. However, when a species is classified as partially specialized for a certain item (e.g. facultative frugivorous), this indicates that the taxon has more diversity of cognitive and ecological demands (it has a wider niche breadth) than a species classified as strictly specialized.

On the other hand, in comparative analyses, it is critical to control for nonindependence among species trait values due to their phylogenetic relatedness (Felsenstein, 1985; Harvey & Pagel, 1991). This nonindependence is characterized by the phylogenetic signal (Blomberg & Garland, 2002) and can be accounted for with different statistical methods (Revell et al., 2008). Pagel (1999) introduced a metric (λ) to estimate how strongly does shared ancestry affect species trait values. Freckleton et al. (2002) showed in a simulated study that this metric correctly predicts the strength of the phylogenetic signal. Phylogenetic generalized least squares models (PGLS) are used to test to what extent the phylogeny predicts variation in species trait values (Pagel, 1997, 1999; Freckleton et al., 2002). The phylogeny is transformed into a variance–covariance matrix of shared evolutionary time between any pair of species, using tree topology and branch lengths. In this matrix, diagonal elements are calculated as the root-to-tips distances and off-diagonal elements as the distance from the root to the most common ancestor of each pair of species (Pagel, 1997, 1999; Freckleton et al., 2002). The λ parameter is quantified by maximum likelihood (ML) in PGLS, and it is used to scale the variance–covariance matrix. In theory, λ varies between 0 and 1. Species can be treated as independent when λ = 0 (i.e. the off-diagonal values are equal to 0 and the phylogeny becomes a ‘star’ tree). When λ = 1, the similarity among species is proportional to their shared evolutionary time (Pagel, 1997, 1999; Freckleton et al., 2002).

The most recent study of brain size as a function of feeding behaviour shows that relative brain volume in species of the family Phyllostomidae is greater in frugivorous species than that in predatory species (Ratcliffe, 2009). Analyses were performed with standardized independent linear contrasts (PIC, i.e. assuming λ = 1) and assuming full independence of species (species-level analysis, i.e. λ = 0). Although those results confirmed previous findings, it should be noted that comparative analyses conducted so far about the evolution of brain size in bats rely on PIC (Hutcheon et al., 2002; Jones & MacLarnon, 2004; Ratcliffe, 2009). However, when λ is intermediate between 0 and 1, the PIC and the species-level analysis are not ideal methods, because they may respectively overestimate and underestimate the effect of the phylogeny. This was tested by Capellini et al. (2010) in a study of the scaling of metabolic rates in mammals. The authors found that the method that is used to quantify phylogenetic signal affects the outcome of the comparative analyses and may lead to select a weak supported model over the model that fits the data best. Conversely, when testing the association between variables and regression analysis, PGLS simultaneously estimates λ, thus accounting for the precise extent the phylogeny predicts variation in species. Unlike species-level comparisons and PIC, PGLS finds the best-fitting model with the appropriate λ value.

Here, we examined the performance of two comparative methods, PGLS and PIC, in the study of the role of diet on the evolution of brain size. We used diet categories, levels of diet specialization and handling behaviour as ecological factors. We performed the analyses in two families of Neotropical bats. We included handling behaviour because motor skills to manipulate food are evolutionarily associated with the size of different structures of the brain (Reader & Laland, 2002; Barton, 2004), and this variable has not been analysed before in bats. First, we tested whether there is a phylogenetic signal in brain volume and in its association with body mass and the ecological traits. Then, we tested which model, PGLS, PIC and models that did not account for phylogeny (i.e. species-level analysis), fit the data best. Simultaneously, we elucidated whether changes in brain size are a function of diet per se (i.e. diet categories) or are related to feeding behaviour (i.e. levels of diet specialization and handling behaviour). If the latter is true, we should find brain volume is larger in species with intermediate levels of specialization (i.e. omnivores). In addition, we should find that species that manipulate their food before swallowing (i.e. frugivores and carnivores) have larger brains than species that do not manipulate or barely manipulate their food (i.e. sanguivores, nectarivores and insectivores).

Materials and methods

Molecular phylogeny and divergence time estimation

We estimated time-calibrated phylogenies of seven species of mormoopids and 54 species of phyllostomids by analysing two nuclear protein coding regions (exon 28 of the von Willebrand factor gene and recombination activating gene 2) and five mitochondrial genes (cytochrome b, cytochrome c oxidase subunit 1, 16S ribosomal RNA, 12S ribosomal RNA and tRNA valine). Sequences of Noctilio leporinus (Noctilionidae) were used to root the trees. Sequences were downloaded from GenBank (Table S1 in the Supporting Information), and individual loci were aligned in Muscle 3.6 (Edgar, 2004). The Hasegawa–Kishino–Yano (HKY) model with allowance for gamma distribution of rate variation and for some proportion of invariant sites (HKY + Γ + I) best fitted the alignment of cytochrome c. All other loci were fitted by the general time reversible model of substitution (GTR) + Γ + I, as determined in jModelTest 0.1.1 (Posada, 2008).

We obtained time-calibrated phylogenies in Bayesian evolutionary analysis by sampling trees (BEAST) 1.6.1 (Drummond & Rambaut, 2007) on concatenated sequences (6557 bases) previously partitioned by locus and by codon position (1 + 2, 3) in the case of the nuclear regions. We used the Yule process as the tree prior and the uncorrelated lognormal relaxed clock to account for lineage-specific rate heterogeneity (Drummond et al., 2006). We used the HKY + Γ model unlinked for all the loci as a trade-off between computational power and model complexity. We applied two calibration constraints: the lower limit of the Whitneyan stage (30.8 Ma) and the lower limit of the Laventan stage (11.8 Ma) as minimum age constraints of a lognormal distribution (Morgan & Czaplewski, 2002; Czaplewski et al., 2003). Maximum age constraints were set to the upper limit of divergence between Mormoopidae and Phyllostomidae (40.99 Ma) and to the limit of divergence between Lophostoma and Mimon (17.1 Ma) (Teeling et al., 2005; Hoffmann et al., 2008). We conducted four independent analyses with 60 million steps, with sampling every 6000 generations. Convergence of the chains to the stationary distribution was confirmed in Tracer 1.5 (http://tree.bio.ed.ac.uk/software/tracer/). The effective sample size of the parameters varied between 942.1 and 9413.1. We combined the last 2500 trees from each independent analysis and built a maximum clade credibility tree with the program TreeAnnotator 1.6.1 (http://beast.bio.ed.ac.uk/TreeAnnotator).

Although adaptations to fruit, nectar and pollen consumption evolved in Phyllostomidae, we also included the sister family Mormoopidae in the analyses because the feeding habits of mormoopids represent the ancestral state of phyllostomids (Rojas et al., 2011).

Brain volume, body mass and ecological traits

Cranial volumes were determined with the method of Eisenberg & Wilson (1978) from adult individuals of both sexes (= 315; Table S2), deposited in the Mammalian Collection of the Institute of Ecology and Systematics in Cuba (IES) and in the National Mammalian Collection of the Institute of Biology, National Autonomous University of Mexico (IBUNAM). Data of body mass (in grams) were obtained from the tags of the specimens. Each skull was weighted to the nearest 0.01 g. Small pellets were introduced through the foramen magnum, and the skull was reweighted. This procedure was repeated four times, and we calculated an average pellet weight for each specimen. This mean value was divided by an empirically derived constant (6.5) and was converted to volume (in cubic centimetres). The brain nearly fills the cranial cavity, and the specific gravity of the brain nearly equals the specific gravity of water. Therefore, cranial volume in cubic centimetres should approximate the brain weight in grams (Eisenberg & Wilson, 1978). To test the accuracy of this method, we compared our data with the values of brain volume, as published by Baron et al. (1996) (Table S2). The two set of values have a strong positive and significant correlation (= 0.9895, < 0.001, = 28 species).

To classify the 61 species according to the ecological traits, we built a database of feeding habits (Table S3). Classification of species followed the procedure explained in Rojas et al. (2011, 2012). We used four levels of specialization (absent, complementary, predominant or strict) for animalivory, nectarivory, frugivory and phytophagy (i.e. nectarivory + frugivory). We also coded species in a broad sense of feeding specialization. Species that feed predominantly or strictly on a food type were classified accordingly as animalivorous, phytophagous or omnivorous. Then, animalivorous and phytophagous taxa were classified as specialists (they feed on a single food type), whereas omnivores were classified as generalists. Finally, carnivorous species and predominantly and strictly frugivorous species were classified as ‘food handlers’. The other species were classified as ‘non-handlers’.

Statistical analyses

Analyses were performed in BayesTraits (Pagel et al., 2004) on the maximum clade credibility tree. Continuous variables were log-transformed, and ecological traits were used as factors. We used an analysis of covariance (ancova) within PGLS framework to assess variation in slope and intercept across the levels of diet specialization for each feeding type, diet specialization (in the broader sense) and levels of handling behaviour (or food manipulation), with body mass as predictor variable to account for allometric effects. First, we assessed the strength of the phylogenetic signal for cranial volume alone. Then, we performed the analyses while simultaneously estimating λ. In PGLS, model parameters and λ are found by ML, and λ is incorporated in the error term of the model to account for the species’ shared evolutionary history (Pagel, 1997, 1999; Freckleton et al., 2002). We compared models including the ecological traits to the allometric model to test whether additional independent variables increased the fit of the model to the data. For this, we used a likelihood ratio test: LR = −2 × [Lh (better-fitting model) – Lh (worse-fitting model)]. In this case, the best-fitting model has the highest log-likelihood score (Lh). The significance of LR was tested with a χ2 distribution with degrees of freedom corresponding to the difference in number of parameters between the two competing nested models (Pagel, 1999; Freckleton et al., 2002; Quinn & Keough, 2002).

We repeated the analyses with λ forced to be equal 1 (i.e. models that produce similar results to those obtained with phylogenetically independent contrasts, hereafter PIC models) and with λ forced to be equal 0 (species-level analysis). To identify the best-fitting model in the estimation of the phylogenetic signal for brain volume and in the phylogenetically controlled ancova, we compared PGLS models, PIC models and species-level models with likelihood ratio tests. The model was tested for significance with 1 degree of freedom (Pagel, 1997, 1999; Freckleton et al., 2002). In the ancova, the ecological traits were quantified with dummy variables (Quinn & Keough, 2002). As reference level, we choose ‘predominant’ for both specialization for phytophagy and specialization for frugivory and ‘nonnectarivorous’ for specialization for nectarivory. The choice of the reference level does not affect the results (Quinn & Keough, 2002), that is, selecting any level would lead to the same conclusions. Dummy variables quantify differences in intercepts. The interaction terms between dummy variables and body mass indicate differences in slopes relative to the reference level. The significance of intercepts and slopes can be tested with t statistics, using information on intercept and slope values and associated standard errors. In all analyses, we used 0.05 as significance level.

In addition to the previous analyses, we contrasted the results of Ratcliffe (2009) with those obtained with PGLS models. The author used PIC models as implemented in CAIC 2.6.9 (Purvis and Rambaut, 1995). We used the same data set (see Table 1 in Ratcliffe, 2009) and tested both PIC and PGLS models in BayesTraits. Then, we identified the best-fitting model as described above. For the phylogeny, we accommodated the supertree of mammals published by Bininda-Emonds et al. (2007, 2008) to the subset of species of the data set.

Table 1. Phylogenetic signal and model fit test of ecological correlates of brain volume in New World bats (= 61 species). Models in which the phylogenetic signal (λ) was estimated with maximum likelihood (PGLS or phylogenetic generalized least squares) were compared with a likelihood ratio test (LR) to models in which λ was forced to be 1 (PIC or phylogenetically independent contrasts) or to be 0 (species-level analysis, that is, ordinary least squares). Likelihood values (Lh) are provided for PGLS (λ = maximum likelihood (ML)), PIC (λ = 1) and species-level (λ = 0) models
 λLh λ = MLLh λ = 1PGLS vs. PICLh λ = 0PGLS vs. species level
LR P LR P
Traits
Brain volume0.9914.7514.7501−2.8035.10< 0.001
Body mass0.97−1.57−1.840.540.462−12.0620.98< 0.001
Models
Allometric model0.7377.1171.6410.94< 0.00168.7716.68< 0.001
Feeding specialization0.4781.4274.8513.14< 0.00180.511.820.177
Specialization for phytophagy
Model 10.6985.9182.676.480.01184.313.200.074
SSJ model0.6985.7282.466.520.01184.153.140.076
Model 20.8383.6981.374.640.03178.1911.00< 0.001
Specialization for nectarivory0.7079.0675.217.700.00674.329.480.002
Specialization for frugivory
Full model0.5684.0980.028.140.00483.600.980.322
SSJ model0.5881.9476.1411.6< 0.00181.201.480.224
Food manipulation0.7479.6574.2410.820.00174.0211.26< 0.001

Results

Effects of the ecological traits

We obtained a well-resolved and supported molecular time-calibrated phylogeny comprising main lineages of Mormoopidae and Phyllostomidae, with the exception of the phyllostomid subfamilies Lonchophyllinae and Rhinophyllinae (Fig. S1, Data S1). Tree topology and divergence times were consistent with previous studies (Rojas et al., 2011).

Body mass explained 89% of the variance in brain volume (t58 = 20.02, < 0.001, Lh = 77.11). When we added the factor feeding specialization (i.e. specialist vs. generalist mode) to the allometric model, the new model improved significantly the fit to the data and increased the explained variance by 3% (model summary: Lh = 81.42, λ = 0.47, r2 = 0.92; model fit: LR = 8.62, = 0.013). However, we found no consistent effect of feeding specialization in increasing brain volume across the range of body mass (Fig. 1a; Table S4).

Figure 1.

Brain volume on body mass with fit lines by type of ecological trait. (a) Feeding specialization. (b) Specialization for phytophagy. (c) Specialization for nectarivory. (d) Specialization for frugivory. (e) Food manipulation. (f) Feeding habits according to Ratcliffe (2009).

Specialization for phytophagy also improved the fit to the data and increased in 3% the variance explained by the allometric model. Slopes and intercepts differed significantly, indicating that the effect of specialization for phytophagy on brain volume is not consistent after controlling for body mass (Table S5, Fig. S2a). Based on these results, we merged strictly and predominantly phytophagous in one level and complementary and nonphytophagous in another level that was used as reference, and then, we compared slopes and intercepts (Table S5, SSJ model). This new analysis confirmed our previous results (Fig. 1b; model summary: Lh = 83.69, λ = 0.83, r2 = 0.91; model fit: LR = 13.16, = 0.001).

Adding specialization for nectarivory to the allometric model did not improve the fit to the data (model summary: Lh = 79.06, λ = 0.70, r2 = 0.90; model fit: LR = 3.90, = 0.419). Slopes and intercepts were not significantly different across the levels of this ecological trait (Fig. 1c; Table S6).

Specialization for frugivory improved the fit to the data and increased in 3% the variance explained by the allometric model (Table S7, Fig. S2b). Slopes and intercepts differed significantly between nonfrugivorous species and the other taxa with any level of specialization for frugivory (model summary: Lh = 81.94, λ = 0.58, r2 = 0.92; model fit: LR = 9.66, = 0.008). The effect of this ecological trait in increasing brain volume was not consistent after accounting for body mass (Fig. 1d; Table S7, SSJ model).

Including food manipulation (i.e. handlers vs. nonhandlers) in the model led to similar results to including specialization for nectarivory. The full model did not improve the fit to the data (model summary: Lh = 79.65, λ = 0.56, r2 = 0.92; model fit: LR = 13.96, = 0.030), and slopes and intercepts were not significantly different across the levels of the trait (Fig. 1e; Table S8).

Phylogenetic signal

Brain volume and body mass both exhibited each a strong phylogenetic signal, significantly different from 0 but not from 1 (Table 1). In the allometric model, the phylogenetic signal of brain volume relative to body mass was lower (0.73) than those estimated independently for each trait. It was also significantly different from 0 and 1, suggesting that the ML estimation of λ and the other parameters in a PGLS framework is an appropriate approach. PGLS models provided a better fit to the data than PIC models. PGLS models including specialization for phytophagy (only the model in which strictly and predominantly phytophagous were tested against complementary and nonphytophagous), specialization for nectarivory and food manipulation also provided a better fit than that achieved with species-level analyses (Table 1).

Feeding habits and brain volume in phyllostomid bats

The allometric model applied to the data set of Ratcliffe (2009) explained 93% of the variance in brain volume (t50 = 26.84, < 0.001, = 53 species, Lh = 88). Frugivores, nectarivores and omnivores had significantly larger brains than animalivores after allometric effects were accounted for in the PGLS model, forcing λ to be 1 (Table 2; r2 = 0.94; model summary: Lh = 90.47, λ = 1, r2 = 0.94; model fit: LR = 4.93, = 0.026). This was consistent with the results of Ratcliffe (2009). The additional parameters of the full model did not increase the fit further (Table S9, full PIC model).

Table 2. Brain volume and feeding habits in phyllostomid bats. The simplest statistically justifiable phylogenetically independent contrasts (PIC) model (PIC-SSJ) has one slope and two intercepts (one for animalivorous species and one for frugivorous, nectarivorous and omnivorous species together). The simplest statistically justifiable phylogenetic generalized least squares (PGLS) model (PGLS-SSJ) has two slopes (one for nectarivorous species and one for frugivorous, omnivorous and animalivorous species together) and three intercepts (one for nectarivorous species, one for animalivorous species and one for frugivorous and omnivorous species together)
Predictors (PIC-SSJ model) t 50 P Predictors (PGLS-SSJ model) t 47 P
Intercept44.56< 0.001Intercept61.22< 0.001
Body mass27.62< 0.001Body mass30.86< 0.001
Animalivory−2.220.030Animalivory−3.160.003
Nectarivory2.140.037
Nectarivory × body mass−2.260.028

When we estimated λ in the PGLS framework, the full model (Table S9, full PGLS model) did not improve the fit of the data with regard to the simplest statistically justifiable model (SSJ). The SSJ model showed that frugivores and omnivores had larger brains than animalivores after controlling for body mass. In addition, nectarivores had a significantly higher intercept and a lower slope relative to the other feeding habits (Table 2, Fig. 1f; model summary: Lh = 92.51, λ = 0, r2 =0.96; model fit: LR = 9.02, = 0.029).

Discussion

In our analyses, we did not find a consistent effect of feeding specialization, specialization for phytophagy and particularly for frugivory and food manipulation on brain volume across the range of body mass of New World bats. We were not able to detect whether brain size increases with increasing proportion of plant material in the diet of New World bats. Analyses based on phylogenetically independent contrasts have suggested that fruit-eating bats have larger brain than non-fruit-eating bats (Jones & MacLarnon, 2004; Ratcliffe, 2009). However, we found that PGLS models did not support these results. We also found no evidence to support the hypothesis of sensory adaptation, that is, brain size is influenced by the information that has to be processed as part of foraging behaviour. We did not detect a pattern in which generalist species have larger brains than specialist after accounting for body mass (see Fig. 1a). Our results did not support the hypothesis of handling behaviour either (see Fig. 1e).

The lack of a consistent pattern (i.e. slopes significantly different) suggests that in small species, adaptations to frugivory (and, in general, to phytophagy) are associated with larger brains, whereas the opposite occurs in bigger species (see Fig. 1). How can such pattern arise? In comparative studies, equivocal results may be due to low sample size, that is, phylogenies with less than 30 taxa (Freckleton et al., 2002). However, our phylogeny and the tree in Ratcliffe (2009) both have more than 50 taxa. There might be another explanation. An increase in the size of visual and olfactory structures in the brain of bats characterizes the evolutionary transition between nonfrugivorous and frugivorous species (Barton et al., 1995). For instance, in Neotropical fruit-eating bats, olfaction plays a key role in locating fruits (Rieger & Jakob, 1988; Thies et al., 1998; Korine & Kalko, 2005). In addition, experimental studies indicate that vision is also important in echolocating phytophagous bats (Chase, 1981, 1983; Heffner et al., 2007). Furthermore, spatial location of food sources and integration of different sensorial channels, spatiotemporal predictability (Fleming & Muchhala, 2008) of floral resources and fruits, habitat complexity (Safi & Dechmann, 2005) and morphological and functional adaptations to flight (Iriarte-Diaz et al., 2012) are putative modellers of brain size. It seems that patterns of brain evolution in bats could be better understood by examining (1) the variation in brain structures involved in sensory ecology and (2) the contribution of such variation to total brain mass. This could provide a more precise evaluation of the factors that influence evolution of brain size, as suggested by the mosaic evolution of brain structure in mammals (Barton & Harvey, 2000).

The strength of the phylogenetic signal in the models we assessed was significantly different from 1. Furthermore, PGLS models provided a better fit than PIC models in all cases (except for brain volume and body mass) and in some cases than species-level models (see Table 1). In addition to this, the re-examination of the analysis of Ratcliffe (2009) supports the suggestion that the relation between feeding habits and brain size should be analysed on the basis of the rigorous estimation of the effect of shared evolutionary history (Freckleton et al., 2002).

In conclusion, feeding behaviour (either characterized as feeding habits, as levels of specialization on a type of item or as handling behaviour) has a variable effect on brain volume after accounting for body mass and controlling for the strength of the phylogenetic signal. This result was consistent for feeding habits for the phylogeny of Phyllostomidae and Mormoopidae that we obtained in this study and the phylogeny of Phyllostomidae that was extracted from the supertree of mammals of Bininda-Emonds et al. (2007, 2008). Although the strength of the phylogenetic signal of brain volume and body mass is high when tested individually, λ values in PGLS models were significantly different from 1 in the phylogeny of Phyllostomidae and Mormoopidae, suggesting that PIC models might not be the best approach for the study of ecological correlates of brain size in New World bats.

Acknowledgments

This work is dedicated to the memory of Elisabeth Kalko. Our gratitude goes to the curators of the collection of IBUNAM and to Silvia Santiago for her assistance with the measurement of specimens. We thank Isabella Capellini for advice and help with the analysis in BayesTraits. We thank Jacek Radwan, Michel Laurin and an anonymous reviewer for their helpful comments on the manuscript. The work of DR was financed through a PhD scholarship from the Xunta de Galicia. The work of CAM was partially financed by Rufford Small Grants. This research was partially supported by AECID (A/023710/09), the Spanish DGICYT (CGL2009-10466), FEDER funds from the European Union, CYTED (409AC0369) and the Xunta de Galicia (INCITE09-3103009PR) to LN.

Authorship

DR and CAM conceived the ideas; DR, CAM and JJFM collected the data; DR, CAM and LN analysed the data; all authors contributed to the writing of the manuscript.

Ancillary