Climatic patterns predict the elaboration of song displays in songbirds

Summary

Climatic variability and unpredictability [1] affect the distribution and abundance of resources and the timing and duration of breeding opportunities. In vertebrates, climatic variability selects for enhanced cognition when organisms compensate for environmental changes through learning and innovation [2–5]. This hypothesis is supported by larger brain sizes [6], higher foraging innovation rates [7–9], higher reproductive flexibility [10–12], and higher sociality [13] in species living in more variable climates. Male songbirds sing to attract females and repel rivals [14]. Given the reliance of these displays on learning and innovation, we hypothesized that they could also be affected by climatic patterns. Here we show that in the mockingbird family (Aves: Mimidae), species subject to more variable and unpredictable climates have more elaborate song displays. We discuss two potential mechanisms for this result, both of which acknowledge that the complexity of song displays is largely driven by sexual selection [15, 16]. First, stronger selection in more variable and unpredictable climates could lead to the elaboration of signals of quality [14, 17–20]. Alternatively, selection for enhanced learning and innovation in more variable and unpredictable climates might lead to the evolution of signals of intelligence in the context of mate attraction [14, 21–23].

Results and Discussion

Mockingbirds and their close relatives are found in climatically diverse habitats throughout the New World. This group exhibits exceptional interspecies variation in singing behavior (Figure 1), spanning species with simple to hyperdiverse repertoires and zero to extensive heterospecific mimicry (see the Supplemental Data available online). We used this wide range of variation in habitat and singing behavior to explore the potential effects of climatic variability on the complexity of song displays.

Figure 1.

Examples of variation in singing behavior in the Mimidae. Spectrograms of acoustic specimens from adult breeding males singing alone during the breeding season. Note the repetition of standard sounds (i.e. ‘syllable types’) in all species and the grouping of syllables into temporally discrete clusters (i.e. ‘songs’).

Climate data were obtained from the Global Historical Climatology Network (http://www.ncdc.noaa.gov/oa/climate/ghcn-monthly/index.php). Variability was measured as the mean coefficient of variation (CV) and the mean range of variation of monthly averages in precipitation and temperature throughout each species' breeding range (see Supplemental Data). These variables reflect the diversity of physiological challenges and the extent of variation in resources to which species are exposed [24]. We also measured the predictability of climatic variables across years via Colwell's predictability index [1]. This index reflects the variation in onset and extent of weather patterns across years (high interyear variation equals low predictability). In our data set, range of variation in temperature was strongly correlated with range of variation in precipitation (Pearson's r = −0.659, degrees of freedom [df] = 27, p < 0.001), predictability of precipitation (Pearson's r = −0.755, df = 27, p < 0.001), and CV of temperature (Pearson's r = 0.770, df = 27, p < 0.001). To avoid multicollinearity, we combined these four climatic variables through factor analysis into a single composite factor. This factor, termed “climatic composite,” explains 91% of the variance in temperature range (factor loading = 0.953), 69% of the variance in CV of temperature (factor loading = 0.827), 78% of the variance in predictability of precipitation (factor loading = −0.609), and 80% of the variance in range of precipitation (factor loading = −0.636). Higher climatic composite scores imply more limiting climatic conditions (i.e., larger temperature ranges, more variable temperatures, and less predictable precipitation). The negative loading of range of precipitation is also an indication of less limiting conditions because in our database, environments with larger ranges of variation in precipitation were also the wettest and presumably less limiting habitats (Pearson's correlation between mean annual precipitation and range of variation in precipitation, r = 0.836, df = 27, p < 0.001). To summarize, the independent variables considered in the models presented below were CV of precipitation, predictability of temperature, climatic composite, and two additional variables of known importance for the evolution of bird song, namely habitat (open versus forested) [25] and migratory behavior (none, facultative migrant, or obligate migrant) [26, 27]. Information on habitat and migration was obtained from [28].

We measured the complexity of song displays via ten parameters that describe the presentation style and vocal diversity of each species (see Supplemental Data). Among others, these measurements include the number of syllable types present in three minutes (i.e., the duration of our acoustic specimens; see Experimental Procedures), the average number of songs per minute, the duration and complexity of songs, the tendency to copy heterospecific sounds, and the mean similarity between renditions of any given syllable type (also known as syllable type consistency [19]). To account for correlated behaviors, these ten parameters were reduced to three composite factors (Table 1) via Bayesian factor analysis [29]. Factor 1 was termed “mimicry syndrome” because it reflects behaviors related to the tendency to copy heterospecific sounds. Tandem repetition and syllable switching rate are important contributors to this factor because heterospecific mimics among the Mimidae tend to repeat each syllable type multiple times before introducing a new one [30]. A higher tendency to copy heterospecific songs corresponds to higher factor 1 scores. Factor 2 was termed “short-term diversity” because it reflects song output and syllable diversity within our three-minute samples. Species that sing more and present a greater number of syllable types per minute have higher factor 2 scores. Factor 3 was termed “song complexity” because it describes the duration and composition of songs. Species with longer songs and a higher number of syllable types per song have higher factor 3 scores.

Table 1.

Factor loadings and proportion of unaccounted variance (ψ) in a Bayesian Factor Analysis of mockingbird song parameters. Primary contributors to each factor are presented in boldface.

Variable a, d	Factor 1 b	Factor 2	Factor 3	ψ c
	‘Mimicry Syndrome’	‘Short-term Diversity’	‘Song Complexity’
Heterospecific copying	0.911	−0.193	0.448	N/A
(Syllable Switching Rate)ˆ2	−0.845	0.273	−0.112	0.024
log(Tandem Repetition)	0.836	−0.283	0.099	0.040
log(Number of syllable types)	0.216	0.813	0.115	0.033
log(Syllable Versatility)	−0.117	0.680	−0.266	0.437
Song Rate	−0.007	0.602	−0.297	0.549
(Consistency)ˆ2	0.345	−0.487	0.182	0.704
log(Song duration)	0.245	−0.133	0.739	0.310
log(Syllable types per song)	−0.255	0.214	0.739	0.034
log(Syllable duration)	−0.330	−0.024	−0.619	0.499

^aSpecies averages and definitions are presented in the Supplementary Data. Variables were transformed as indicated (following [44]) and standardized prior to the analysis

^bThe loadings were inverted in Factor 1 to facilitate the interpretation of scores

^cError (i.e. unexplained) variance. MCMCpack does not estimate this parameter for the ordinal variable.

^dThe negative item difficulty parameter (λ₁) for heterospecific copying was 0.0103, whereas its Γ (i.e. the cutpoint used to convert latent continuous variables into this ordinal variable) was 1.033.

Given that climatic variability and unpredictability may select for enhanced learning and innovation [2–5], our a priori hypothesis was that species living in more variable and unpredictable environments should be able to learn or invent more syllable types and should be able to copy syllables that are harder to produce or that are heard less often. In the context of our measurements, this predicts greater syllable diversity, more complex songs, and a higher tendency to copy heterospecific sounds. The latter prediction is based on the mechanical difficulties of imitating heterospecific sounds [31] and the rarity with which some of the mimicked sounds are typically heard [32].

We used a phylogenetically informed approach to explore the relationship between singing behavior, climatic variability, and climatic unpredictability (see Experimental Procedures and Supplemental Data). Our phylogenetic hypothesis is based on a species-level molecular phylogeny generated from both nuclear and mitochondrial DNA sequences (see Supplemental Data). In the analysis of each song parameter, we used a multimodel inference approach [33] (for statistical hypothesis testing, see Supplemental Data). Two of the advantages of multimodel inference over traditional model selection methods are the ability to incorporate model uncertainty into the process of formal inference [34] and the ability to estimate unconditional parameters from a set of candidate models with different evolutionary assumptions. The evolutionary assumptions considered are star phylogeny (i.e., no phylogenetic signal), Brownian evolution, and the Ornstein-Uhlenbeck model of evolution (for descriptions of these models, see [35]).

Our results show that climatic variables are important predictors of the elaboration of song displays in the Mimidae (Table 2). The predictive value of our models was high for mimicry syndrome (R² = 0.531), intermediate for short-term diversity (R² = 0.191), and moderate for song complexity (R² = 0.102). Models that account for phylogenetic correlation were better supported for mimicry syndrome and short-term diversity, suggesting a moderate to strong effect of phylogenetic history in these behaviors (see Tables S5 and S6). Models that assumed a star phylogeny were generally better supported in the case of song complexity (see Table S7).

Table 2.

Multimodel inference on the effects of climate, migration, and habitat on male singing behavior in the Mimidae. We present the relative variable importance ^a and, in parentheses, the model-averaged coefficients, ${\stackrel{\sim }{\overline{\beta }}}_j$ , for each independent variable. The proportion of variability in the data that is accounted for by each model (R²) is presented at the bottom. Interpretation of the coefficients in this table is facilitated by inspection of bivariate correlations in Table 3.

Independent variable	Song parameter
	Mimicry Syndrome	Short-term Diversity	Song Complexity
Migration b	0.981 (−0.088, 0.649)	0.746 (−0.256, −0.514)	0.074 (0.006, 0.014)
Habitat c	0.178 (0.013)	0.214 (−0.020)	0.423 (0.129)
CV in precipitation	0.189 (−0.019)	0.230 (−0.032)	0.399 (0.164)
Predictability of temperature	0.416 (−1.217)	0.207 (0.144)	0.334 (0.959)
Climatic Composite d	0.229 (−0.020)	0.240 (0.020)	0.403 (0.054)
R2	0.531	0.191	0.102

^aRelative importance is computed as the sum of the Akaike weights of the set of models in which the variable appears. This parameter reflects how important is the variable in predicting the song parameter (0=no predictive value, 1=high predictive value).

^bReference category is “obligate migrant”. The two coefficients presented in parentheses refer to non-migratory and facultative migrant respectively.

^cReference category is ‘forest’.

^dHigher scores imply wider temperature ranges, more variable temperature, less predictable precipitation and a narrower range of precipitation (see text for details)

To facilitate the interpretation of the coefficients in Table 2, we present the bivariate correlation coefficients of association between individual song variables and climatic variables in Table 3. Species living in less predictable climates or occupying habitats with more variable temperatures are more likely to engage in heterospecific mimicry and to modify repertoire presentation styles accordingly (i.e., to repeat each syllable type several times before introducing a new one [30]). Also interesting are the strong effect of migration on mimicry and the higher mimicry syndrome scores of facultative migrants as compared to nonmigrants and obligate migrants (compare β_j values in Table 2). These results suggest a potential link between vocal learning and other cognitive processes because facultative migrants show higher rates of innovation in foraging [8].

Table 3.

Pearson's product-moment correlation coefficients as measures of association between singing behaviors and climatic variables ^a. These data are presented as a posteriori tests to facilitate the interpretation of the coefficients in Table 2. Significant and marginally significant correlations are presented in boldface ^b.

	CVp	Pt	‘Climatic Composite’
			log(Ranget)	log(Rangep)	Pp	CVt
‘Mimicry syndrome’ c
Tandem repetition	−0.006	−0.367*	0.608**	−0.338	−0.422**	0.466**
Switching Rate	0.045	0.263	−0.456**	0.211	0.423**	−0.337*
‘Short-term diversity’ c
Types per Sample	−0.249	0.024	0.141	−0.137	−0.180	0.434**
Syllable Versatility	−0.285	0.146	−0.275	0.11	0.103	−0.002
Song Rate	−0.170	−0.169	0.697**	−0.53**	−0.575**	0.626***
Syllable Consistency	0.404**	−0.337*	0.059	−0.116	0.027	−0.197
‘Song complexity’ d
Song Duration	0.248	0.067	0.188	−0.124	−0.11	0.172
Syllable Duration	−0.364*	−0.496**	−0.235	0.05	0.253	−0.157
Syllable Types per Song	−0.013	0.077	0.285	−0.276	−0.247	0.421**

^aCV = coefficient of variation; P = Colwell's predictability index; Range = range of variation; Subscript p = precipitation; Subscript t = temperature

^b* p<0.08, ** p<0.05, *** p<0.001

^cCorrelations based on phylogenetically independent contrasts [44] given strong evidence for a phylogenetic signal (df = 26)

^dCorrelations based on raw measurements given low evidence of a phylogenetic signal (df = 27)

Short-term diversity also increases with more variable and unpredictable climatic conditions. Variability in precipitation and unpredictability in temperature are associated with higher syllable consistency (note the negative loading of syllable consistency in short-term diversity in Table 1). Higher unpredictability in precipitation and more variable temperatures are likewise associated with higher song rates and greater short-term diversity. There is, however, a negative correlation between song rate and range of variation in precipitation; this somewhat counterintuitive result reflects the wetter and presumably less limiting nature of environments with higher ranges of precipitation in our database (see above).

In terms of song complexity, more variable climates are associated with songs with a higher number of syllable types. The correlation between climate and syllable duration is difficult to interpret in this context given the lack of an a priori hypothesis for the expected direction of change in this song parameter.

To summarize, we have shown here that there is a strong general trend in the Mimidae toward more elaborate song displays in more variable and unpredictable breeding environments. This result is largely driven by an improvement in song-learning ability in the form of more consistent singing, more diverse displays, and a greater ability to copy heterospecific sounds. There are at least two potential mechanisms that could explain this correlation. First, environmental variability and unpredictability can create limiting conditions that might increase the intensity of competition for mates [36] or other resources [37], leading to an elaboration of signals of general male quality. Bird songs contain information about physiological condition, immunocompetence, developmental stress, and territory quality [14, 17–20] and may consequently undergo stronger positive interor intrasexual selection in variable and unpredictable climates [17]. This hypothesis would be supported by larger sexual differences in noncognitive/non-songrelated traits in more variable and unpredictable climates. However, we found that a simple index of size dimorphism (see Supplemental Data) was not correlated with temperature (Pearson's correlation of phylogenetically independent contrasts with df = 25: CV, p = 0.536; range of variation, p = 0.342; predictability, p = 0.714) or precipitation variables (CV, p = 0.084; range of variation, p = 0.306; predictability, p = 0.769). Furthermore, even though plumage reflectance data are not available for this group, all mimids are monomorphic for plumage pattern and color to the human eye, and most species are achromatic gray or rufous brown (see [28]).

A second possibility is that the relationship between climate and song is driven by variation in the strength of female preferences for signals that provide specific information about a male's ability to learn and innovate (see [14, 21–23]). Cognitive skill could be an important mate choice criterion in omnivorous species such as the ones considered here because good short-term learning and decision-making abilities may be critical for foraging (and thus provisioning rates), acquisition of all-purpose territories, selection of nest sites, and timing of breeding. This scenario assumes that certain aspects of song learning can be generalized to other types of learning (e.g., that there are general mechanisms for storing memories in different contexts). The possibility of such general mechanisms is suggested by the positive correlation between various measures of cognition in both primates and birds (see [38–40]), by the greater learning proficiency in non-singing tasks in male birds with more complex song types [41], and by the larger brains in birds that mimic versus those that do not [23]. In the Mimidae, however, residual brain size (see Supplemental Data) is not correlated with mimicry syndrome (Pearson's correlation: df = 20, p = 0.924), short-term diversity (df = 20, p = 0.605), or song complexity (df = 20, p = 0.243), and this is also likely to be true in other songbirds (see [42]). Nevertheless, brain size may not be a sufficiently sensitive indicator of the cognitive skills that are relevant to the Mimidae, and thus a lack of correlation between song elaboration and brain size is not conclusive evidence against the hypothesis of selection on indicators of intelligence.

In conclusion, sexual selection theory posits that females will select mates on the basis of signals or cues linked to male traits that provide females with direct or indirect fitness benefits. Although our results add weight to the idea that some aspects of learned vocal displays in songbirds can provide information about general cognitive skill, further resolution of the mechanisms behind the correlations reported in this study will require experimental tests of cognitive skill in different mimids, other indices of sexual dimorphism in noncognitive traits (e.g., differences in plumage reflectance), and field work that identifies the actual fitness benefits of choosing more intelligent mates.

Experimental Procedures

We present a brief description of our methods below. For detailed protocols and additional information on data collection and analysis, please refer to the Supplemental Data.

Song Characters

From a pool of 1738 recordings of mimids obtained from sound archives and private collections, we selected and analyzed 98 acoustic specimens (n = 29 species, 3.4 ± 1.6 specimens per species) of comparable duration, sound quality, and social context (see Supplemental Data for details). On each specimen, we measured nine general song parameters describing vocal diversity and presentation style (see Supplemental Data). The analyses reported above are based on species means for each parameter. To account for correlated behaviors, we summarized the nine mean parameters for each species along with the tendency to copy heterospecific songs (an ordinal variable with levels: 0 = none, 0.5 = some, 1 = high) into three composite factors via Bayesian factor analysis [29]. This technique allows a mix of ordinal and continuous responses and was implemented with MCMCpack in R (http://mcmcpack.wustl.edu).

Comparative Analysis

We used a phylogenetically informed approach when testing for the effects of climatic variables on singing behavior. The phylogenetic hypothesis used for these analyses was reconstructed from a DNA sequence character matrix [43] and is presented in the Supplemental Data. To account for the potential effects of phylogeny on singing behavior, we used multimodel inference [33] considering three types of models: nonphylogenetic models (i.e., ordinary least squares), phylogenetic generalized least squares (PGLS) regression under the assumption of Brownian evolution, and PGLS regression under the Ornstein-Uhlenbeck evolutionary process (all models implemented with the MATLAB program Regressionv2.m [35]). A brief description of the methods for the analysis of each composite song factor follows (for further details, see [33]). First, we evaluated models with all possible combinations of migratory behavior, habitat, CV in precipitation, predictability of temperature, and climatic composite (see Supplemental Data). We ranked these models based on the second-order Akaike information criteria (AICc) and computed their corresponding relative weights of evidence, or Akaike weights (w_i). Relative variable importance was determined as the sum of the Akaike weights across all reasonably supported models (i.e., a difference of less than 10 in AICc with respect to the best model [33]) in which a variable occurred (see Supplemental Data). Model-averaged coefficients were computed as Akaike-weighted averages over all reasonably supported models (i.e., β_j in [33]), assuming β_j ≡ 0 in models in which variable j did not occur. Similarly, we computed the coefficient of determination in each case as the Akaike-weighted average of R² in all reasonably supported models.

Supplementary Material

Supp Data. Supplemental Data.

Supplemental Data include Supplemental Experimental Procedures, seven tables, and one figure and can be found with this article online at http://www.cell.com/current-biology/supplemental/S0960-9822(09)01060-4.