Sons learn songs from their social fathers in a cooperatively breeding bird

Abstract

Song learning is hypothesized to allow social adaptation to a local song neighbourhood. Maintaining social associations is particularly important in cooperative breeders, yet vocal learning in such species has only been assessed in systems where social association was correlated with relatedness. Thus, benefits of vocal learning as a means of maintaining social associations could not be disentangled from benefits of kin recognition. We assessed genetic and cultural contributions to song in a species where social association was not strongly correlated with kinship: the cooperatively breeding, reproductively promiscuous splendid fairy-wren (Malurus splendens). We found that song characters of socially associated father–son pairs were more strongly correlated (and thus songs were more similar) than songs of father–son pairs with a genetic, but no social, association (i.e. cuckolding fathers). Song transmission was, therefore, vertical and cultural, with minimal signatures of kinship. Additionally, song characters were not correlated with several phenotypic indicators of male quality, supporting the idea that there may be a tradeoff between accurate copying of tutors and quality signalling via maximizing song performance, particularly when social and genetic relationships are decoupled. Our results lend support to the hypothesis that song learning facilitates the maintenance of social associations by permitting unrelated individuals to acquire similar signal phenotypes.

1. Introduction

Signals may convey several levels of information to conspecifics, such as species identity, kinship, motivation or quality [1]. In many animals, signals are innate, but in humans, cetaceans, some bats and some birds, acoustic signals are learned, so there is a reduction in genetic influence over the signal phenotype [2]. Why some animals should learn their vocal signals is a long-standing question with no single answer [3], but several functional hypotheses may explain vocal learning in songbirds, such as signalling of genetic adaptations to local conditions [4], social adaptation to local neighbourhoods [5], more accurate kin recognition [6] and quality signalling based on learning abilities [7]. Although there may be multiple benefits to song learning, these hypotheses have received differing levels of support. For example, the idea that learned songs signal genetic adaptation to local environments can be ruled out in those systems where songs are learned post-dispersal (discussed in Podos & Warren [8]), but post-dispersal learning lends support to the hypothesis that social adaptation to a neighbourhood is a significant benefit of song learning [5,9]. Understanding when and from whom a bird learns, therefore, can inform us about what benefits they may receive by learning at all.

Establishing and maintaining social associations by copying local songs is thought to influence the fitness of many songbirds, for example, by indicating territory tenure or the quality of their learning abilities [5,7,10–12]. Maintaining social associations is particularly consequential in cooperative breeders [13,14], and the few studies that have investigated vocal learning in cooperative breeders have demonstrated cultural transmission of calls from parents to offspring, supporting the role of vocal learning as a kin recognition mechanism [15–17]. In those studies, social association was correlated with relatedness, so benefits of kin recognition could not be disentangled from benefits of being in a cooperative group. Systems in which social associations do not accurately reflect relatedness (e.g. when extra-pair paternity levels are high), however, provide an opportunity to evaluate the relative importance of vocal learning as a mechanism for kin recognition versus a mechanism for non-kin social association. If social group cohesion is more important than kin-selected benefits of sociality, we predict that song learning will facilitate recognition of socially affiliated individuals irrespective of relatedness. Additionally, in cooperative social systems, where individuals are surrounded by non-kin [18,19], it may be disadvantageous for signallers to advertise relatedness, and signals identifying kinship may be unlikely to evolve [20]. Thus, in cooperative systems with high levels of extra-pair paternity, we expect patterns of song transmission that facilitate signalling social rather than genetic relationships to predominate [21], but song transmission in such a system has not previously been studied.

Understanding how song is acquired is not only important for studies of kin recognition, but is also relevant to understanding how song functions as a sexual signal and potential indicator of individual quality. The importance of avian song for mate attraction, mate stimulation and territory defence is well established [22], but the underlying benefits gained by females for having preferences for particular songs remain controversial [23–26]. From the perspective of a female choosing a mate, quality may be based on either direct (resources) or indirect (genetic) benefits [23], and song may signal these benefits differentially depending on how it is acquired. If song has a significant genetic contribution, males that possess superior singing ability (e.g. performance-related acoustic characteristics such as songs with a particularly fast trill) may pass that ability on to their sons irrespective of the son's tutor [27]. Alternatively, aspects of singing ability may be condition dependent [28], and such a signal need not be heritable to be useful to receivers. There may be a tradeoff, however, between demonstrating singing ability (e.g. maximizing trill rate) and accurately imitating a tutor, for example, if a bird's tutor sings a low performance song [29]. Signallers may therefore have to optimize the degree to which they signal quality, kinship and non-kin social association, because some information is facilitated by genetic or condition-related contributions to songs, and other information is facilitated by accurate imitation of unrelated tutors.

In this paper, we evaluate genetic and social underpinnings of song variation in splendid fairy-wrens (Malurus splendens), which are cooperatively breeding yet reproductively promiscuous [30], and which therefore permit a natural contrast between social association and genetic relatedness. First, songs may be learned from males who rear the offspring irrespective of genetic relatedness. Second, there may be genetic transmission of song characters, which can be studied by examining young who are not reared by their genetic father. Third, if song transmission is horizontal or oblique, then songs will not be correlated with either the genetic or foster fathers’ songs. Additionally, we relate several metrics of male quality to song variation, including song characters that may indicate quality such as trill rate, song length and song output (reviewed in Gil & Gahr [28]). If song signals genetic quality, there should be significant similarity between songs of genetic fathers and sons and song should correlate with phenotypic indicators of quality. If song signals male condition, then song should correlate with phenotypic quality but may not have a genetic signature. If song signals social association but not quality, we expect strong similarity between socially affiliated individuals but no genetic or phenotypic correlations.

2. Material and methods

(a). Study species and population monitoring

Splendid fairy-wrens are non-migratory Australian passerines that breed in family groups consisting of a socially dominant male and female, often with one to two auxiliary (typically male) helpers [30,31]. Auxiliary males assist with territory defence and provisioning of young and may sire a small proportion of offspring [30]. Approximately 50 per cent of broods contain extra-pair offspring [30]. Male offspring may disperse if breeding opportunities arise in nearby (often neighbouring) territories, or they may inherit their natal territory if the dominant male dies [31,32]. Female offspring, in contrast, tend to disperse in their first year, to a greater distance than males [32]. Both sexes produce a standard display song (type I song, or ‘TI’) during territorial encounters [31,33]. Males, but rarely females, produce an additional display vocalization (type II song, or ‘TII’) in response to the vocalizations of avian predators [33,34]. Males incorporate TII songs into the dawn chorus and append them to TI songs (‘TI + trill’) [33]. Both primary and helper males (but rarely females) participate in the dawn chorus [33].

The work presented here was conducted during the breeding seasons (October–December) from 2005 to 2009 in mallee eucalyptus scrubland habitat at the Brookfield Conservation Park in South Australia. Each season, we monitored from 120 to150 banded splendid fairy-wrens (M. s. melanotus) following published field methodologies [30,32]. For each male, we collected data on: (i) age: either known or minimum age, (ii) social status: breeder or helper, (iii) body condition: residuals of a regression of mass on tarsus length, (iv) cloacal protuberance volume [35], and (v) per cent blue on breast: ranging from 0 per cent in completely brown males to 95 per cent in very blue individuals [36]. We extracted DNA from blood samples using the GentraSystems DNA extraction kit and genotyped all individuals at six microsatellite markers [30]. Sizing reactions were carried out on the Applied Biosystems 3130xl Genetic Analyzer at the University of Chicago Core DNA Sequencing and Genotyping Facility. We manually scored genotypes using Peak Scanner Software v. 1.0 (Applied Biosystems) and assigned parentage using the program Cervus [37] following published protocols [30].

(b). Father–son pair identification and classification

We classified fathers as one of three types: (i) social and genetic father (‘social genetic’, n = 20 father–son pairs); (ii) social but not genetic father (‘foster’, n = 28 father-son pairs); and (iii) genetic but not social father (‘cuckolding’, n = 22 father–son pairs). In total, we had song recordings from 50 sons and 40 fathers; sample sizes differed because fathers appeared multiple times in the dataset if they sired more than one recorded son, and sons appeared multiple times if their foster and cuckolding fathers had both been recorded. Additionally, seven males that we first recorded as sons became fathers in subsequent years, so this dataset comprised a total of 83 individuals. We used a different recording, sampled from the appropriate year, when males were first sons and then fathers. In our analysis, we accounted for non-independence of samples using randomization techniques (described below).

We used father and son recordings that were made in the same year whenever possible (n = 54 father–son pairs), but we included 16 father–son pairs recorded 1 year apart (social genetic, n = 4; foster, n = 2; cuckolding, n = 10). Male songs do not appear to change significantly across years [38], and excluding these 16 pairs did not qualitatively change our results. Sons were 1-year old at the time of recording in all but eight cases, where they were 2-years old. Although we are unsure of the precise timing of the sensitive phase for song learning in fairy-wrens, 1-year old males sang fully crystallized songs (figure 1), implying that the sensitive phase occurs during a male's first year of life. Because sons generally either remained as a helper on the territory where they were raised (n = 25) or adjacent to it (n = 23), most social genetic and foster fathers were the nearest available tutor up until the time of recording. In contrast, most cuckolding males were not readily available tutors, because they were often living in distant or non-neighbouring territories (n = 17).

Figure 1.

Spectrograms of a TI (left) and TII (right) songs from two sons (centre), their social fathers (top) and their genetic fathers (bottom).

(c). Song recording and acoustic analysis

We recorded males during dawn chorus song bouts (equipment described in the electronic supplementary material). In addition to dawn song bouts, during 2006–2009, we recorded predator-context type II songs elicited with playbacks of grey butcherbirds (Cracticus torquatus) [33]. For each male, we calculated song output as the first principle component of a principle component analysis using song rate and duty cycle (ratio of the time spent singing to total time sampled) as factors (PC1 explained 82.8% of the variation). We measured both song rate and duty cycle during an uninterrupted sample of the dawn chorus (mean sample duration: 481 s ± 146 s). We also calculated the proportion of TII songs in the chorus sample and the proportion of songs with a trill (TI + trill). For these measurements, we used spectrograms digitized in Raven Pro v. 1.2 [39].

To efficiently and objectively extract all other acoustic characteristics from our recordings, we created a separate audio file for each song and used the software SoundPoints [40] to automatically describe each song (settings described in the electronic supplementary material). SoundPoints transforms acoustic data from spectrograms to a series of time, frequency and amplitude points called landmarks, which can then be used to calculate a variety of acoustic descriptors. The final dataset contained 86 727 notes from 2219 TI and TI + trill songs, and 11 370 notes from 379 TII songs, from which we calculated: song duration, bandwidth (difference between the highest and lowest frequency), peak frequency (mean loudest frequency throughout the vocalization) and note rate (number of notes per second). For TI + trill songs, we also calculated the per cent of song duration that was a trill (‘percent trill’), and the ratio of the trill amplitude to the non-trill amplitude (‘amplitude ratio’). We also used the landmarks from SoundPoints to classify note types based on agglomerative hierarchical clustering and a model selection approach using Akaike information criterion (AIC) values to choose the best number of clusters [40]. For TI and TI + trill songs, we identified a total of 77 note types and for TII songs 15 note types; we used these classifications to calculate the average number of note types in each song. We summarized acoustic characters for each male by calculating means using all songs for the year that male was recorded (mean n = 25.3 ± 13.1 songs per male for TI and TI + trill songs, and 4.2 ± 3.5 songs per male for TII songs). All acoustic characters were repeatable [41] and significantly different among individuals (r > 0.21, F > 3.06 and p < 0.001 for all characters).

(d). Statistical methods

We used partial Mantel tests [42] with 25 000 permutations to assess the relative contributions of social association and genetic relatedness on song similarity. Mantel tests assess correlations based on pair-wise distances, so we expressed song similarity, social association and genetic relatedness as types of distance [43]. We calculated song similarity between males for each song type by z-transforming each variable (subtracting the mean and dividing by the standard deviation), and then calculating the pair-wise Euclidean distances between songs in the resulting unit-less space. We used three different states for social distance (self, 0; social son, 1 and unfamiliar, 2) and for genetic distance (self, 0; genetic son, 1 and unrelated, 2). Differences in ‘self’ songs were comparisons between years. We expected song differences within males to be smaller than differences between males, so we constructed a third distance matrix, with two states (self, 0 and non-self, 1), to use in partial Mantel tests so that the effects being tested by the social and genetic distance matrices were the differences between sons and unfamiliar or unrelated males. To assess which acoustic characters were most strongly correlated between fathers and sons, we compared father and son songs for each acoustic character and song type using simple regressions. For cuckolding fathers, the regression slopes estimate half the genetic heritability of each song character [44].

We determined if variation in song was associated with male age, social status, body condition, cloacal protuberance volume and per cent blue on breast using multiple regressions on each song character. We used each male only once [n = 60 males for TI, 51 for TI + trill and TII (predator context), 40 for TII (dawn context) and 65 for chorus-level acoustic characteristics] and used quality measurements that were made in the same year that the analysed songs were recorded. The variables age, per cent blue on breast and cloacal protuberance volume were significantly positively correlated with one another (r > 0.45, p < 0.001 for all three pair-wise correlations), so represented phenotypic changes in males as they aged. All statistical tests were two-tailed and were conducted in Jmp v. 5 (SAS Institute Inc., Cary, NC, USA) or R [45].

3. Results

Social distance was strongly and significantly correlated with song distance for most song types (table 1). These relationships remained significant after correcting for possible effects of genetic distance and comparisons of individuals to themselves (table 1). In contrast, genetic distance showed weaker correlations with song distance, which became non-significant when we corrected for the effect of social distance and self (table 1). In summary, social affiliation between fathers and sons was associated with more similar songs than would be expected by chance, but genetic relatedness was not.

Table 1.

Mantel correlations (r) between song distance and uncorrected social distance (social), social distance corrected for correlation with genetic distance (partial social), uncorrected genetic distance (genetic) and corrected genetic distance (partial genetic). Significant p-values are highlighted in bold and indicate that songs between fathers and sons with that association (social or genetic) are more similar than would be expected by chance. Number of individuals contributing to each matrix is given in italics for each song type. TII songs are separated into dawn (dawn) and predator (pred) contexts.

song type		social	partial social	genetic	partial genetic
chorus	r	0.079	0.068	0.040	0.002
(74)	p	<0.001	<0.001	0.046	0.939
TI	r	0.010	0.001	0.019	0.016
(69)	p	0.529	0.953	0.263	0.334
TI+trill	r	0.125	0.104	0.070	0.010
(59)	p	<0.001	<0.001	<0.001	0.611
TII (dawn)	r	0.111	0.095	0.057	−0.003
(48)	p	<0.001	<0.001	0.026	0.910
TII (pred)	r	0.069	0.073	0.010	−0.026
(65)	p	<0.001	<0.001	0.575	0.151

Many song characters were significantly correlated in social genetic and foster father–son pairs in bivariate regressions, but few were correlated in cuckolding father–son pairs (table 2). The mean effect size (Pearson's r ± 95% CI) for all correlations was higher in social genetic (0.318 ± 0.133) and foster (0.393 ± 0.125) father–son pairs than in cuckolding (0.11 ± 0.127) father–son pairs. After sequential Bonferroni correction [46] for 90 comparisons, no cuckolding father–son pairs showed significant correlations, but seven acoustic characters remained significantly correlated in either social genetic or foster father–son pairs (table 2). In summary, many father and son song characteristics were significantly correlated (and thus songs were most similar) for socially affiliated pairs regardless of relatedness (figure 1), but song characteristics were weakly correlated for fathers and sons with a genetic, but no social, association, indicating that songs have a significant cultural contribution.

Table 2.

Simple regression slopes of son on father acoustic characters for each song type and father–son pair type. TI + trill songs are separated into non-trill and trill portions, and TII songs are separated into dawn (dawn) and predator (pred) contexts. Sample sizes for social genetic, foster and cuckolding father–son pairs are given in italics (in that order) for each song type. Bold font indicates a regression significant at p < 0.05, *p < 0.01 and **p < 0.0003 (significant after sequential Bonferroni correction for 90 comparisons).

song type	acoustic character	social genetic	foster	cuckolding
chorus	song output	0.46*	0.21	0.20
(18, 24, 17)	percent TI + trill	0.70**	0.81**	0.25
	percent TII	0.05	0.07	−0.01
TI	bandwidth	0.51	0.33	−0.51
(14, 17, 14)	duration	−0.50	0.13	−0.53
	note rate	1.03*	−0.29	−0.19
	note types	0.45	0.00	−0.66
	peak frequency	−0.06	0.27	−0.33
TI+trill (non-trill)	bandwidth	0.43	0.54	−0.01
(14, 18, 11)	duration	−0.17	0.64*	0.64
	note rate	0.58	0.10	0.33
	note types	0.27	0.70*	0.48
	peak frequency	0.28	0.24	−0.05
	amplitude ratio	1.01*	1.23**	0.79
	per cent trill	−0.19	0.61	0.07
TI+trill (trill)	bandwidth	0.23	−0.02	−0.53
(14, 18, 11)	duration	−0.24	0.34	−0.06
	note rate	0.78**	1.16**	0.81
	note types	−0.05	0.92*	0.12
	peak frequency	0.37	0.73*	0.65
TII (dawn)	bandwidth	−0.04	0.65	0.68
(9, 11, 5)	duration	0.35	0.47	0.17
	note rate	0.41	0.51	−0.08
	note types	0.33	−0.35	0.25
	peak frequency	0.49*	0.49*	0.29
TII (pred)	bandwidth	0.15	0.15	−0.03
(16, 20, 15)	duration	−0.09	−0.10	0.22
	note rate	0.52*	0.42**	0.18
	note types	0.19	0.58*	−0.09
	peak frequency	0.52	0.49**	0.13

We found few relationships between song characters and male quality, even when considering characters such as trill rate, song length and song output. Five partial regressions were significant at p < 0.05, but after sequential Bonferroni correction for 150 comparisons (five metrics of quality for each of the 30 acoustic measurements, listed in table 2), only one relationship remained significant (peak frequency of dawn TII songs and per cent blue, n = 40, F = 19.1, p < 0.0001), and it was heavily driven by one outlying data point. Overall, therefore, the song characters studied here were weakly (if at all) associated with male social status, age or breeding condition.

4. Discussion

Males learned songs from their social fathers, with no evidence for genetic transmission of acoustic characters. As a consequence, songs indicated genetic kinship poorly, but were an excellent indicator of social association, which in fairy-wrens may be the affiliation with the greatest consequence for reproductive fitness [18,19]. Singing a song that indicates natal origin on a territory may provide numerous benefits to helpers; it may reduce aggression from dominant individuals, facilitate inheriting the territory if the dominant male dies, make cooperative territory defence more effective and facilitate access to extra-pair females [5,12,18,19]. These processes may all operate simultaneously, and they do not require genetic kinship to benefit both the helper and dominant male if unrelated helpers reciprocate tolerance with nestling provisioning and territory defence [47]. Additionally, even if dominant males could experience a net benefit by ejecting unrelated helper males, we might still not expect kin recognition signals to evolve if they are a disadvantage to helpers that are surrounded by non-kin [20]. In summary, vertical cultural transmission of songs in this system may facilitate the maintenance of social associations between fathers and sons that are not necessarily genetically related.

The lack of genetic signatures in songs was consequential not only to kinship signalling, but also to quality signalling. Aspects of song structure such as trill rate, song frequency and song output, which are often thought to indicate underlying aspects of male quality [28], were neither heritable nor condition dependent. Instead, these acoustic traits were learned from a male's social father. Young, socially subordinate males (putatively ‘poor’-quality individuals) sang so similarly to their social fathers (putatively ‘high’-quality individuals) that song was uninformative with respect to male age, breeding condition or social status. This supports the idea that there may be tradeoffs between accurate tutor imitation and quality signalling through song performance [29], and in fairy-wrens, this tradeoff tends towards accurate imitation, rather than maximizing potential aspects of performance.

The decoupling of genetic and vocal transmission in this system means that within one generation, songs can become disassociated from a genetic lineage and may therefore be uninformative with respect to any aspect of heritable male quality. Consider a scenario in which a high-quality male gains extra-pair copulations, perhaps because of high-quality songs; his extra-pair sons would inherit his high-quality genetic traits, but would learn the ‘inferior’ songs of their cuckolded social father. Females may still have preferences for some acoustic characteristics, for example, if songs with those characteristics transmit well through the environment or are attractive for other reasons, but our results suggest that associations between song and male genetic quality likely do not drive female preferences in this system. Complementing this are preliminary analyses that suggest many song characteristics do not predict male mating success in this system [38], although thoroughly assessing female mating preferences in relation to male phenotypic traits is an important direction of future research. Additionally, investigating song transmission in females, as well as other social influences on male song development (such as tutoring by older siblings when present, or by neighbours when social fathers are unavailable), may reveal additional complexities of song transmission in this system.

We have shown that in a cooperative breeder with high levels of extra-pair paternity, many acoustic characters have low genetic transmission and are passed from social fathers to sons irrespective of genetic relatedness. Song variation in this system may therefore convey reliable information about social affiliation and individual identity, but does not appear to be an easily interpretable proxy of individual quality. Previous work has demonstrated the importance of social association for avian vocal transmission in systems where individuals learn from unrelated neighbours or kin [3,16,17], and our study illustrates that living with unrelated members of the same social group is another context in which social learning may facilitate group cohesion, because it allows unrelated individuals to acquire similar signal phenotypes. Additionally, much work has illustrated the value of song as a signal of individual quality (reviewed in Gil & Gahr [28]), but signalling quality using elaborate or high performance songs may not be universal [23,24,48], and our results suggest that when social associations between unrelated individuals are of great consequence, birds may compromise such measures of song ‘attractiveness’ for copying accuracy. More generally, signal information content and signal ontogeny have traditionally been considered separate fields of research, but when both lines of inquiry are undertaken simultaneously, the results may be complementary. Studies investigating song variation in relation to male quality or mating success may therefore benefit by taking the effects of cultural transmission on song into consideration when interpreting results.