Geographical variation in song frequency and structure: the effects of ...

ANIMAL BEHAVIOUR, 2007, 74, 1573e1583 doi:10.1016/j.anbehav.2007.03.022

Geographical variation in song frequency and structure: the effects of vicariant isolation, habitat type and body size AN NA H. K OETZ *, DA V ID A . WEST COTT † & B RA DLEY C . CONGDON*

*School of Marine and Tropical Biology, James Cook University yCSIRO Sustainable Ecosystems (Received 26 November 2006; initial acceptance 23 February 2007; final acceptance 26 March 2007; published online 27 September 2007; MS. number: 9191)

In this study, we investigated whether historical, refugial isolation may have caused current, large-scale geographical variation in the song frequency and structure in the chowchilla, Orthonyx spaldingii, or whether this variation can be better explained by the influence of isolation-by-distance, vegetation type and/or body size and mass. We recorded songs from 15 locations across the species’ range, covering five historically isolated populations (Pleistocene refugia) and two areas of post-Pleistocene recolonization. We measured six spectrotemporal characteristics of 773 songs and used a multivariate approach to test for differences between refugia. Historically isolated populations could be clearly distinguished by their spectral characteristics, particularly bandwidth and peak frequency. In addition, we found striking song divergence across the Black Mountain Corridor, a known historical climatic barrier. Northern refugia showed significantly narrower bandwidths and higher peak frequencies than southern refugia. Spectral characteristics were not influenced by geographical distance, broad habitat differences or body size. This study shows that spectral song characteristics were at least in part influenced by historical isolation in refugia. Given the known history of population isolation in these refugia, cultural drift is the most likely explanation for the differences in spectral characteristics in chowchilla song. Ó 2007 The Association for the Study of Animal Behaviour. Published by Elsevier Ltd. All rights reserved.

Keywords: bird song; chowchilla; culture; drift; isolation; Orthonyx spaldingii; selection; song variation

Geographical variation in song is a common phenomenon among songbirds, principally because songs are learnt culturally, and inaccurate copying of the tutor song inevitably leads to change across the landscape (Slater 1989). Due to the important functions of song in a bird’s life, geographical variation in song could potentially influence population genetic divergence leading to speciation, by favouring within-dialect mating and natal philopatry and discouraging between-dialect dispersal (Baker & Cunningham 1985; Grant & Grant 1997; Slabbekoorn & Smith 2002a). In songbirds, song is a cultural trait that is nongenetically transmitted from one generation to the next through learning and imitation, and such a cultural change (or cultural evolution) is thought to be a very powerful source of variation (Dawkins 1976; Mundinger 1980; Lynch 1996). Correspondence: A. H. Koetz, School of Marine and Tropical Biology, James Cook University, PO Box 6811, Cairns Qld 4870, Australia (email: [email protected]). D. A. Westcott is at CSIRO Sustainable Ecosystems, PO Box 780 Atherton, Qld 4883, Australia. 0003e 3472/07/$30.00/0

Large-scale processes that are thought to drive song variation are similar to those driving genetic variation: vicariant isolation followed by drift and/or selection of certain song types or song characteristics, and isolation over geographical distances (isolation-by-distance) resulting in the accumulation of small differences due to the reduced probability of distant individuals to interact socially (Cavalli-Sforza 2000). Under the vicariance and drift model, song variation would be the greatest between populations that have been isolated the longest, irrespective of the distance between them. Thus, song is expected to be more similar within than among isolated populations, even when distances within the population are larger than among populations. On the other hand, under a drift only model, song variation would be expected to decrease linearly with distance due to the accumulation of small differences across space (isolation-by-distance). The influence of isolation and subsequent drift on song has been shown in some species that occur both in mainland populations as well as in island populations

1573 Ó 2007 The Association for the Study of Animal Behaviour. Published by Elsevier Ltd. All rights reserved.

1574

ANIMAL BEHAVIOUR, 74, 5

that established through a founder-event. The initial bottleneck effect in small founder populations can lead to a reduced number of song types or song elements due to random sampling effects. Once isolated, subsequent drift may lead to further simplification or modification of the song (Mundinger 1980; Baker & Jenkins 1987; Grant & Grant 1995; Baker 1996; Baker et al. 2001, 2003a, b; Slabbekoorn & Smith 2002a, b). For instance, singing honeyeaters showed a reduced meme pool and lower syllable diversity in an island population compared to the mainland populations, which was attributed to founder effects at the time the island was colonized (Baker et al. 2001). Vicariant isolation followed by drift was also used to explain geographical divergence in the spectral song characteristics of the golden bowerbird, Prionodura newtoniana (Westcott & Kroon 2002). Alternatively, song characteristics may vary geographically due to the acoustic properties of different habitats, or due to different ecological conditions (Slabbekoorn & Smith 2002a, b; Ruegg et al. 2006). Under this selection model, song characteristics would be strongly linked to differences in ecology or habitat, but not with geographical location or distance. Large-scale song variation was clearly linked to habitat differences in the satin bowerbird, Ptilonorhynchus violaceus, as call structure converged in similar habitats but diverged among different habitats independent of geographical location or distance (Nicholls & Goldizen 2006). Similarly, Ruegg et al. (2006) showed that song characteristics were strongly correlated

with ecological (climatic) variables after correcting for genetic distance in the Swainson’s thrush, Catharus ustulatus. Baker (2006) also found a change in the spectral characteristics with changing ambient sound environments between mainland and island populations of both red capped robins, Petroica goodenovii, and western gerygones, Gerygone fusca. Despite the growing number of studies focusing on geographical variation in bird song, the relative importance and strength of the different forces driving song variation have still not been resolved unambiguously. Few studies have systematically recorded song across the entire range of a species to establish the pattern and extent of geographical song variation to test different theories of the origin of song variation. Endemic songbirds of the tropical rainforests in northeastern Australia (the ‘Wet Tropics’) are especially suitable for studying the forces driving song variation due to the known contraction of rainforests into island-like remnants (‘refugia’) during the Pleistocene climatic fluctuations. These climatic fluctuations and their impact on rainforest distribution are particularly well documented for the most recent glacial cycle, which began approximately 135 000 years ago (Kya) and reached its glacial maximum 18 Kya (Webb & Bartlein 1992; Hopkins et al. 1993; Kershaw 1994). At the height of the last glacial maximum, the Wet Tropics rainforests were restricted to a number of small, montane rainforest refugia (Webb & Tracey 1981; Nix & Switzer 1991; Fig. 1a). Furthermore, evidence also

(b) Current

(a) 18 000 Years ago

Cooktown BT TU WU

CU

BMC

BMC Cairns AU

LR WHw WHe CR

SR PR Townsville 100 km

Figure 1. Map showing (a) Pleistocene refugia (black shading) as proposed by Webb & Tracey (1981) and northern/southern isolates (grey shading) as proposed by Nix & Switzer (1991) and (b) locations of recording sites (white circles) across the current Wet Tropics rainforests (grey shading) in north-eastern Australia. BT: Big Tablelands; TU: Thornton Uplands; WU: Windsor Uplands; CU: Carbine Uplands; BMC: Black Mountain Corridor; LR: Lamb Range; AU: Atherton Uplands; WH: Walter Hill Ranges (w ¼ west; e ¼ east); CR: Cardwell Range; SR: Seaview Range; PR: Paluma Range; BMC and PR represent areas of postPleistocene recolonization.

KOETZ ET AL.: GEOGRAPHICAL VARIATION IN CHOWCHILLA SONG

suggests the existence of a climatic barrier (the ‘Black Mountain Corridor’, BMC; Fig. 1a) predating the Pleistocene, separating the proposed remnants into northern and southern refugia (Webb & Tracey 1981; Nix & Switzer 1991; Hopkins et al. 1993). This known history of range contraction and the subsequent rapid expansion of the rainforest and its inhabitants provides an ideal system that has been used to test theories of the driving forces of genetic divergence in several species of mammals, birds, reptiles and invertebrates (Joseph et al. 1995; Winter 1997; Schneider et al. 1998; Schneider & Moritz 1999; Schneider et al. 1999; Hugall et al. 2002). The chowchilla, Orthonyx spaldingii, is an endemic songbird that occurs only in the upland rainforests of the Australian Wet Tropics and, thus, would have experienced strong population fluctuation during the Pleistocene. The chowchilla is a ground-dwelling insectivore that lives in small groups of two to eight birds, which cooperatively feed and defend their territory (Jansen 1999). This species is of special interest for a study on the forces driving geographical song variation as, unlike most songbirds, it has a single-song repertoire, both males and females sing the same territorial song, and it shows exceptional geographical variation at larger geographical scales (A. H. Koetz & D. A. Westcott, personal observation) and a mosaic distribution of song dialects at a smaller scale (McGuire 1996; Koetz et al. 2007). Despite the intriguing variation in song in this species, geographical variation across its range has not been quantified. Limited genetic and morphological evidence suggests a distinct split across the BMC in the chowchilla (Joseph et al. 1995; Schodde & Mason 1999), separating the species into two subspecies: O. s. spaldingii south of the BMC, and O. s. melasmenus north of the BMC (Schodde & Mason 1999). This divergence, in addition to the wellestablished history of refugial isolation, provides an ideal natural experimental design, which allows for a unique comparison of the forces driving song variation with those suggested causing genetic and morphological divergence. In this study, we quantify the extent and pattern of geographical variation across the species’ range to test whether historical population isolation contributed to

the current landscape-scale geographical variation in song. Thus, under the vicariance and drift model, we would expect strong differences in the song characteristics between historically isolated refugia and particularly across the older BMC irrespective of the distance between them. On the other hand, if geographical distance, rather than vicariant isolation, contributes to the current song variation, we would expect a significant, positive relationship of song dissimilarity with distance within refugia. If habitat type influences song characteristics, we should find clear song divergence between the different habitat types and convergence of song characteristics in the same habitat types, irrespective of the location or distance. Finally, because some morphological variation across the BMC has been reported in the chowchilla previously (Schodde & Mason 1999), and song frequencies have been shown to be inversely correlated with body size and mass for several bird species (e.g. Ryan & Brenowitz 1985), we also investigated the effect of body size and mass on song frequencies at the population level.

METHODS

Study Sites Recordings of chowchilla song were carried out at 15 locations across the species’ range, covering five Pleistocene refugia as proposed by Webb & Tracy (1981; see Fig. 1a, b): Thornton Uplands (TU), Carbine Uplands (CU), Lamb Range (LR), Walter Hill Ranges (WH) and Cardwell Range (CR). In addition, song was recorded from two areas of postglacial recolonization: the BMC and the Paluma Range (PR). Latitudes and longitudes of all locations are given in Table 1. Songs were recorded between April and November 2004 and August and November 2005. Geographical variation was sampled at different spatial scales that form a hierarchical, or nested, order of scale. At the largest scale, chowchilla populations are divided into northern and southern isolates on either side of the BMC (Fig. 1a). Within these northern and southern isolates,

Table 1. Latitudes and longitudes of recording locations and refugia in the Wet Tropics of northeastern Australia Refugia Thornton Uplands, TU Carbine Uplands, CU Black Mountain Corridor, BMC* Lamb Range, LR

Walter Hill Ranges, WH Cardwell Range, CR Paluma Range, PR*

*Areas of postglacial recolonization.

Locations (no. of sites)

Latitude

Longitude

Creb track (1) Mt Lewis 1 (2) Mt Lewis 2 (1) Black Mountain (2) Douglas Track (1) Clohesy Fig (2) Clohesy Rd (1) Lake Morris (2) Davies Ck (2) Kauri Ck (3) Sth Johnstone (2) Misty Mountains (2) Koombooloomba Dam (2) Paluma Dam (2) Paluma town (2)

16 06.05960 S 16 35.16120 S 16 32.24710 S 16 36.66000 S 16 52.98130 S 16 56.24950 S 16 58.69370 S 16 58.62220 S 17 02.33590 S 17 07.69200 S 17 38.44730 S 17 41.17630 S 17 50.38190 S 18 57.60360 S 19 00.49850 S

145 20.20020 E 145 16.13310 E 145 17.14160 E 145 27.37820 E 145 37.97180 E 145 36.94150 E 145 39.00680 E 145 41.75880 E 145 36.73280 E 145 36.32830 E 145 42.80530 E 145 31.30670 E 145 35.69260 E 146 08.87200 E 146 12.65440 E

1575

1576

ANIMAL BEHAVIOUR, 74, 5

Songs were recorded from as close to the groups as possible using a Tascam DA-P1 digital audio tape recorder and a Sennheiser ME-67 directional microphone. Recordings were digitized on a notebook computer using a Sigma Tel Audio 5.10 soundcard at a sampling frequency of 22 050 Hz with 16-bit precision. Recordings from each morning were stored as individual sound files. Songs were analysed using Avisoft SASLab Pro Specht 2005. Spectrograms for each song were produced with a 512-pt fast Fourier transform (frequency resolution: 43 Hz, time resolution: 2.90 ms, flat top window and 87.5% overlap). Each song was given an identifying code. Up to 10 best-quality songs per group (average number of songs per group  SD ¼ 7.89  3.6) were chosen by visually assessing quality on the spectrograms, selecting only those songs that showed low background noise, low distortion of songs and elements, and high purity of the sound trace on the spectrogram. All songs were subjected to a low-pass band filter set at 0.60 kHz to eliminate any low frequency background noise.

most of the suggested Pleistocene refugia (Webb & Tracey 1981) were sampled in this study (Fig. 1b). Within each refuge, we visited one to six different locations, and at each of these locations, we recorded chowchilla vocalizations at one to two sites. At each of these sites, we recorded songs from as many groups of chowchillas as possible during the dawn chorus (2e15 groups per site, average number of groups per site  SD ¼ 6.2  4.1). Thus, we sampled at decreasing scales from the largest scale (northern and southern populations divided by the BMC), to medium scales (refugia), to the smallest scale (locations and sites).

Song Recordings At each location, chowchilla vocalizations were recorded at two sites between 1 and 4 km apart. At each site, songs of as many independent groups of chowchillas as possible were recorded. Chowchillas start singing at dawn and continue for about 30e60 min. During this time, a minimum of 10 songs were recorded for each group of chowchillas before moving on to the next group. This resulted in transects of 500e1000 m at each site. All groups recorded at distances less than 1000 m were assigned to the same site. Each group’s position along the transect was recorded. During the morning chorus, all chowchilla groups in an area will sing at the same time with members of each group taking turns. Chowchillas tend to sing from their roosting trees within their territories before descending to the ground (Jansen 1993), and group members tend to be very close to one another (A. H. Koetz, personal observation). Therefore, songs from single groups could be confidently located and identified, but the recognition of individual birds in each group was not possible. Although the exact number of birds per group was unknown, it was clear that in all cases there were at least two birds in a group, singing identical songs. At some sites, chowchillas sang for too short a time, or were too far away for good-quality recordings during the dawn chorus. In these instances, playback was successfully used to tempt the birds into approaching and singing to record vocalizations of better quality. Chowchillas did not alter their song when responding to playback compared to the song used during dawn chorus (A. H. Koetz, unpublished data).

Song Measurements and Analyses Temporal and spectoral measurements of each song were recorded at the mean spectrum of each whole song, using the automatic parameter measurement tool in Avisoft SASLab Pro (Specht 2005). The parameters measured included song length (s), peak frequency (the frequency at the maximum amplitude, Hz), minimum frequency (Hz), and bandwidth (the difference between maximum and minimum frequency, Hz; Fig. 2). In addition, the number of elements within each song was counted from the spectrograms and used to calculate the element rate (number of elements divided by the song length; Fig. 2). To compare the overall spectral similarity between locations and to reduce the dimensionality of the data, nonmetric multidimensional scaling (NMS) was performed on the six variables described above. NMS is an ordination technique based on ranked distances, and thus is better suited for data that are non-normal or for data of different scales (Clarke 1993; McCune & Mefford 1999). As our data were non-normally distributed and, hence, violated the assumptions of more commonly

kHz

1V

6 5 4 3 2 1 0

a

b c

0.5

1

1.5

d

e 2

2.5

3

3.5

s

Figure 2. Song spectrogram of the song of a chowchilla (LR dialect), showing the song’s waveform (upper figure) and power spectrum (left figure); the frequency parameters measured across the whole song are highlighted: (a) start of song; (b) end of song; (c) peak frequency (frequency with the most energy as shown on the power spectrum, Hz); (d) minimum frequency (Hz); (e) bandwidth (difference of maximum and minimum frequency, Hz). Each black trace on the spectrogram denotes one element, which was counted to determine the number of elements and the element rate of that song.

KOETZ ET AL.: GEOGRAPHICAL VARIATION IN CHOWCHILLA SONG

used multivariate approaches (such as principal components analysis), NMS was deemed more appropriate. A two-dimensional solution yielded the least stress (stress ¼ 0.966) and thus was most suitable when arranging the data points in multivariate space. Correlation coefficients (loadings) were used to determine which two variables described most of the variation between locations. A multi-response permutation procedure (MRPP) was then used to test for multivariate differences among the a priori grouping variables (refuge and location). MRPP is a nonparametric procedure for testing the hypothesis of no difference between two or more groups of entities, and also does not require adherence to assumptions such as normality (Zimmerman et al. 1985; Mielke & Berry 1994; McCune & Mefford 1999). For clarification of the patterns observed, averaged NMS scores (2 SE) for each refuge were plotted separately for both dimensions and compared using one-way ANOVAs and Tukey HSD post hoc tests.

Effect of Vegetation Type To test for the effects of different habitat types on song spectrotemporal variables, we averaged the NMS scores for each site and identified each site’s vegetation type following Tracey’s (1982) vegetation classification. Habitat types observed at our study sites included ‘complex mesophyll vine forest’ (Type 1a and b; CMVF), ‘mesophyll vine forest’ (Type 2a; MVF), ‘complex notophyll vine forest’ (Type 6; CNVF), and ‘simple notophyll vine forest’ (Type 8; SNVF). Vegetation types that only occurred at one site (Type 6) were excluded. We then performed a two-way ANOVA on both dimensions separately using site and vegetation type as variables. In addition, to detect any patterns or clustering of vegetation types, site-averaged NMS scores were plotted using vegetation identifiers for each site. Thus, if vegetation type had an effect on song characteristics (NMS scores), we would expect a clustering of sites according to vegetation type.

Effect of Geographical Distance To determine the influence of geographical distance on spectrotemporal differences, pairwise song differences between all the locations were calculated by computing squared Euclidean distances (dissimilarities) from each location’s NMS scores. Pairwise geographical distances (km) between all locations were determined using the Route function in a GPS unit (Garmin etrex Summit Garmin Corp., Olathe, KS, U.S.A.). To test whether spectrotemporal differences are correlated with distance, we used the nonparametric Mantel randomization test, which evaluates the null hypothesis of no relationship between two similarity or dissimilarity matrices (Manly 1997). The Mantel test is an alternative to regressing distance matrices that avoids the problem of partial dependence in the matrices (Manly 1997). The Mantel test statistic r was calculated using a Monte Carlo randomization procedure with 5000 permutations.

Effect of Body Size Peak or dominant frequency is affected by body size, and organisms with larger body sizes generally produce sounds of lower dominant or peak frequencies and narrower bandwidths (Ryan & Brenowitz 1985; Wiley 1991; Badyaev & Leaf 1997; Doutrelant et al. 2001). Chowchillas were captured at 11 locations across their range (N ¼ 2e9 birds per location, mean ¼ 4.6) for a different study on geographical variation in Chowchilla morphology and molecular genetics (Koetz, Westcott & Congdon, personal observation). To detect a possible effect of body size on song frequency, both the mean body mass (g) and mean tarsus length (mm) for each location were compared to the mean peak frequency and mean bandwidth used at that location. Due to the previously suggested divergence in body size across the BMC (Schodde & Mason 1999), only southern locations were used in this analysis (N ¼ 8 locations). If body size had an effect on the song characteristics, we would expect an inverse relationship between the mass/size variables and both song variables. The relationship between body size and mass and peak frequency was tested using a linear regression analysis. All statistical analyses were performed using SPSS (SPSS Inc., Chicago, IL, U.S.A.) and PC-Ord for Windows (McCune & Mefford 1999). Significance levels were set at a < 0.05. All results are given as X  2 SE unless otherwise stated.

RESULTS A total of 773 songs from 93 groups of chowchillas were chosen for analysis (average of eight songs per group and six groups per site). The average duration, peak frequency, minimum frequency and bandwidth  SD are 4.5  1.7 s, 1523.8  448.8 Hz, 809.2  128.2 Hz and 2515  411.8 Hz, respectively (N ¼ 773; Table 2). Songs differed qualitatively in the number and type of elements and broad frequency range (Fig. 3). Chowchilla song was made up of a fixed sequence of pure notes, and songs were characterized by a diversity of different notes. Birds in a locality repeated the same sequence of notes with each song unless the song was cut short due to interference by neighbours or other disturbances. Songs of the southern populations (LR, WH, CR and PR) were characterized by the prevalence of descending notes of broad bandwidths (A-notes), interspersed by shorter notes that may be descending, ascending or without a change in pitch Koetz et al. 2007. The song of most southern populations started with a short descending note followed by a repetition of two to four A-notes (Fig. 3deg). Song of the northern populations (TU and CU) were made up of similar notes but of narrower bandwidths, especially the A-notes. Songs were dominated by short notes with only some A-notes interspersed (Fig. 3a, b). The occurrence of unique notes was common to all populations. Nonmetric multidimensional scaling of the six spectrotemporal song variables showed that variation across Axis 1 (Dimension 1) was mainly influenced by differences in bandwidth, whereas variation across Axis 2 (Dimension 2) was mainly explained by differences in peak frequency

1577

1578

ANIMAL BEHAVIOUR, 74, 5

Table 2. Average spectrotemporal measurements  SD for 773 chowchilla songs recorded in five refugia (TU, CU, LR, WH and CR) and two areas of postglacial recolonization (BMC and PR) in the Wet Tropics, Australia Refugia

Duration (s)

Peak frequency (Hz)

Minimum frequency (Hz)

Bandwidth (Hz)

TU CU BMC* LR WH CR PR*

3.501.27 6.813.32 3.300.84 4.291.43 4.171.13 4.231.20 5.000.99

1496.36401.44 1470.50319.79 1857.96537.40 1600.49439.27 1360.00317.82 1514.14468.29 1407.94485.48

983.18100.94 849.83141.56 803.67102.46 758.66148.44 826.5879.53 820.6177.91 822.9480.54

1535.68110.76 1921.00191.87 2097.76121.30 2671.37276.99 2623.78258.63 2671.11221.54 2734.79199.11

Total

4.471.70

1523.82448.84

809.53128.21

2515.59411.77

*Areas of postglacial recolonization.

in multivariate space, more so across Dimension 1. Northern refugia and the BMC are distinctly divergent from the southern refugia in both dimensions. However, the northern refugia are also quite different from one another across both dimensions, with the BMC falling in between the northern and southern refugia across Dimension 1. The MRPP analysis showed significantly larger variation among refugia than within refugia across multidimensional space, resulting in significant differences between refugia (MRPP test statistic: T ¼ 118.506, N ¼ 773, P < 0.0001). In addition, variation among LR locations

(Table 3). Higher Dimension 1 values correspond to wider bandwidths, whereas higher Dimension 2 values are associated with lower peak frequencies (Table 3). When plotting average NMS scores for each refuge in multidimensional space, there is a strong separation between northern and southern refugia, with song from the northern refugia (including the BMC) having narrower bandwidths and higher peak frequencies than the southern refugia (Figs 4 and 5). Southern refugia are characterized by wide bandwidths and lower peak frequencies (Figs 4 and 5) and cluster together across both dimensions kHz 6 5 4 3 2 1 0 6 5 4 3 1 2 0 6 5 4 3 2 1 0

kHz 6 5 4 3 2 1 0 kHz 6 5 4 3 2 1 0 kHz 6 5 4 3 2 1 0 kHz 6 5 4 3 2 1 0

(a) CU

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

5.5

6

6.5

7

7.5

s

(b) TU

0.5

1

1.5

2

2.5

3

3.5

4

s

(c) BMC

BMC 0.5

1

1.5

2

2.5

3

3.5

0.5

1

1.5

2

2.5

3

3.5

0.5

1

1.5

2

2.5

3

3.5

0.5

1

0.5

1

s

(d) LR

s

(e) WH

4

4.5

5

5.5 s

(f) CR

1.5

2

2.5

3

3.5

4

4.5

5

s

(g) PR

1.5

2

2.5

3

3.5

4

4.5

5

5.5

6

6.5

7 s

Figure 3. Example spectrograms of different chowchilla dialects from five different refugia and two areas of recolonization (a) CU; (b) TU; (c) BMC; (d) LR; (e) WH; (f) CR; (g) PR within the Wet Tropics of Australia. Position of the Black Mountain Corridor (BMC) is indicated by a dashed line. For abbreviations, see text and Table 1.

KOETZ ET AL.: GEOGRAPHICAL VARIATION IN CHOWCHILLA SONG

Table 3. Correlation matrix showing r, r2 and t for dimensions 1 and 2 of the NMS analysis using six spectrotemporal characteristics (duration, peak frequency, minimum frequency, bandwidth, notes/s and number of notes) of chowchilla song 1 Dimension Duration Peak frequency Min frequency Bandwidth Notes/s No. of notes

2

r

r2

t

r

r2

t

0.103 0.422 0.457 0.891 0.039 0.113

0.011 0.178 0.209 0.793 0.002 0.013

0 0.274 0.287 0.676 0.05 0.005

0.086 0.918 0.368 0.426 0.129 0.04

0.007 0.842 0.135 0.182 0.017 0.002

0.074 0.72 0.275 0.36 0.094 0.045

Bold numbers indicate those characteristics that explain most of the variation in multidimensional space.

Two-way ANOVAs showed that for both Dimension 1 (bandwidth) and Dimension 2 (peak frequency), geographical location had a significant effect on NMS scores (ANOVA: Dimension 1: F18,663 ¼ 65.711, P < 0.001; Dimension 2: F18,663 ¼ 15.372, P < 0.001) but vegetation did not (Dimension 1: F2,663 ¼ 0.555, P ¼ 0.574; Dimension 2: F2,663 ¼ 0.273, P < 0.761). When plotting average NMS

0.4 0.2

(a)

0 Axis 1 – (bandwidth)

was significantly greater than within locations, indicating that LR locations are also significantly different in multidimensional space (T ¼ 9.622, N ¼ 284, P < 0.0001). Separate one-way ANOVAs confirmed that refugia were significantly different for both dimension scores (ANOVA: Dimension 1: F6,773 ¼ 147.551, P < 0.001; Dimension 2: F6,773 ¼ 30.241, P < 0.001; Fig. 5a, b). Moving from north to south, bandwidth increased (Fig. 5a), and peak frequency decreased (Fig. 5b). For Dimension 1 (bandwidth), southern refugia (LR, WH, CR and PR) were not significantly different from one another with the exception of LR and WH but were all significantly different from the northern refugia (TU and CU) and the BMC (Fig. 5a). Similarly for Dimension 2 (peak frequency), the northern refugia and the BMC were again significantly different from the southern refugia, although the pattern was less clear. Three of the southern refugia (WH, CR and PR) were not significantly different from one another. However, LR was intermediate between the northern and these three southern refugia (Fig. 5b).

–0.2 –0.4

TU

CU BMC

LR

North of the BMC

WH

CR

PR

South of the BMC

–0.6 –0.8 –1 –1.2

0.6

–1.4 –1.6 N=

0.4 PR

0.2

CU

–0.6

TU BMC

–0.8 –1 –2

–1.5

–1

–0.5

0

0.5

49

284

CU

BMC

LR

111

99

126

WH

CR

PR

0.2

–0.2 –0.4

60

(b)

LR

0

44

0.4

CR

Axis 2 – (peak frequency)

Axis 2 (peak frequency)

WH

1

Axis 1 (bandwidth) Figure 4. Ordination plot of the two significant NMS dimensions (Dimension 1: bandwidth; Dimension 2: peak frequency) using six spectrotemporal song variables of the chowchilla song; songs from each refuge are represented by their centroid  2 SE. For abbreviations, see text and Table 1.

0 TU –0.2

North of the BMC

South of the BMC

–0.4 –0.6 –0.8 –1 –1.2 N=

44

60

49

284 Refuge

111

99

126

Figure 5. Mean NMS scores (2 SE) of Dimension 1 (a) and Dimension 2 (b) for each refuge; refuges sorted from north to south (left to right. For abbreviations, see text and Table 1.

1579

ANIMAL BEHAVIOUR, 74, 5

0.8

6 PRpt LRcf PRpd

0.4

5

LRdc

LRkc2 CRkd

Song dissimilarity

WHe

Axis 2 (peak frequency)

1580

LRkc1

0 CUml

LRlm

LRcr

–0.4

LRkc3 TUcreb2

BMC1 LRdt

–0.8

4

3

2

TUcreb1

1 BMC2

–1.2

0 –1.6 –2

–1.5

–1

–0.5

0

0.5

1

Axis 1 (bandwidth) Figure 6. Ordination plot of the two significant NMS dimensions (Dimension 1: bandwidth; Dimension 2: peak frequency) of the chowchilla song; songs from each location are represented by their centroid  2 SE and identified by their respective vegetation type following Tracey’s (1982) vegetation classification (Type 1a and b: open circle; Type 2a: filled circle; Type 8: grey circle).

scores for each site/vegetation type in multidimensional space, there was also no clear clustering of sites according to vegetation type (Fig. 6). Variation within vegetation types was as great as among vegetation types (Fig. 6). Song dissimilarity was significantly correlated with geographical distance for between-refugia comparisons (Mantel test: r ¼ 0.393, P ¼ 0.011), but not for within-refugia (LR) comparisons (r ¼ 0.213, P ¼ 0.264), indicating that song dissimilarity was influenced by isolation in refugia but not by geographical distance within refugia. There was no correlation between body mass or size and peak frequency or bandwidth (linear regression: mass versus peak frequency: r2 ¼ 0.001, F ¼ 0.001, P ¼ 0.977; mass versus bandwidth: r2 ¼ 0.015, F ¼ 0.0891, P ¼ 0.776; tarsus versus peak frequency: r2 ¼ 0.272, F ¼ 2.241, P ¼ 0.185; tarsus versus bandwidth: r2 ¼ 0.007, F ¼ 0.046, P ¼ 0.837; Fig. 7).

DISCUSSION This study clearly shows that chowchilla song varies greatly across larger scales, and that refugial isolation indeed influenced spectral song characteristics in the chowchilla. Visible differences in chowchilla song on spectrograms as well as easily audible differences include changes in the type, order and proportion of elements in the song, and also changes in the frequency range of the songs (Figs 3e5). Refugia differed significantly in song frequency characteristics, with a broadening of the bandwidth, as well as a decrease in peak frequency when moving from north to south within the species’ range (Fig. 5). Southern refugia cluster together closely in multivariate (song) space

50

150 200 250 100 Geographical distance (km)

300

350

Figure 7. Pairwise comparison of song dissimilarity (squared Euclidean distances) versus geographical distance (km) for between/ within-refuge comparisons (filled circles: within refuge; open circles: between refuge; line of best fit for between-refuge comparison: Mantel r ¼ 0.393, P ¼ 0.011).

(Figs 4 and 5), indicating that, at this large scale, songs within the refugia are very similar, particularly in bandwidth. Spectral characteristics of the northern refugia are distinctly set apart from the southern refugia, although the northern refugia are also significantly different from one another (Figs 4 and 5). Song from the BMC is more similar to the northern than the southern refugia, which may be explained by the fact that the recording sites were at the northern end of the BMC and could have been influenced by the northern refugia through recolonization or simply proximity. These results support the view that vicariant isolation can lead to a change in song characteristics. Isolation-bydistance does not seem to have an effect on the spectral characteristics as shown by our results. In addition, our results indicate that frequency characteristics in the chowchilla may be influenced by processes similar to those driving genetic divergence. The distinct split in song characteristics across the BMC is indicative of the long isolation between the northern and southern populations. This divergence in song characteristics across the BMC mirrors the genetic and morphological divergence suggested previously in the chowchilla (Joseph et al. 1995; Schodde & Mason 1999). These findings raise the question about which forces drove the change in song frequency after the populations were isolated. Spectral characteristics often change in different habitats, or due to the effect of body size or mass, none of which seemed to have influenced song frequencies of the chowchilla (Figs 6 and 8). Previous studies on the influence of habitat differences on song were most often conducted across vastly different habitats (e.g. dense rainforest versus more open ecotone forest; Slabbekoorn & Smith 2002b). In this study, all recording locations were within very similar rainforest habitats, and although these were of different types as classified

KOETZ ET AL.: GEOGRAPHICAL VARIATION IN CHOWCHILLA SONG

Song frequency (Hz)

3500 3000 2500 2000 1500 1000 43

44

45

46

47

48

Tarsus length (mm) Figure 8. Linear regression of chowchilla tarsus length (mm) versus peak frequency (filled square) and bandwidth (open square).

by Tracey (1982), their classification is probably too broad a scale to draw definite conclusions. Nevertheless, we found no effect of broad habitat type at the large scale of our study. At this larger scale, songs sung from within the same habitat type differed vastly between refugia (e.g. TUcreb & LRdc; Fig. 6); yet in some instances songs sung from different habitats within the same refuge were almost identical in their spectrotemporal characteristics (e.g. LRcf & LRkc; Fig. 6). Although we did not test for and thus cannot discount the possible effect of local-scale habitat differences on song characteristics, Kroon & Westcott (2006) did not find any evidence for local song dialects of the golden bowerbird, P. newtoniana, to be better adapted to the local acoustic environment. The golden bowerbird shares its range and upland rainforest habitat with the chowchilla, and both show song characteristics superiorly adapted for long-range communication in dense forests (Richards & Wiley 1980; Endler 1992; Ryan & Kime 2002). Therefore, although we cannot dismiss the effects of small-scale habitat structure on local song variation, they seem unlikely to affect the large-scale song divergence found in this study. The second alternative is that song characteristics changed due to morphological differences north and south of the BMC. Southern populations have larger body sizes than the northern populations (A. H. Koetz, unpublished data), and larger body size and body mass have been shown to inversely influence dominant or peak frequencies and bandwidth (Ryan & Brenowitz 1985; Wiley 1991; Badyaev & Leaf 1997; Doutrelant et al. 2001). However, within the southern refugia neither the peak frequency nor the bandwidth was correlated with body size or body mass in the chowchilla, and thus the influence of body size on spectral characteristics is unlikely. A different hypothesis is that of cultural drift in isolated populations, resulting in a random change of song characteristics over time. Such cultural drift can lead to the random fixation of a few song types or note types in the isolated population (Lynch & Baker 1994; Grant & Grant 1996). Cultural drift seems the most likely explanation for the differences in song spectral characteristics between refugia, given the known history of population

isolation of these refugia that also resulted in genetic divergence in many Wet Tropics endemics (Joseph et al. 1995; Schneider et al. 1998; Schneider & Moritz 1999). Our results also show that song is more divergent between northern refugia than between southern refugia. TU is distinctly different from CU, especially in Dimension 1. In contrast, none of the southern refugia are clearly as divergent as the northern refugia. This could be due to possibly longer isolation between northern refugia, or it may suggest higher levels of gene flow between the southern refugia. Comparatively higher levels of gene flow south of the BMC are possible, given that most of the southern refugia were recolonized from the LR (Schneider et al. 1998). However, molecular genetic analyses are needed to confirm this hypothesis. Using whole-song spectral characteristics may not give the necessary resolution to clearly determine the processes driving divergence within refugia and between and within locations. The level of resolution of whole-song data is similar to that of previous genetic data, and further study is needed to resolve the forces driving song, genetic and morphological divergence in the chowchilla in more detail. In conclusion, this study clearly shows that spectral song characteristics were at least in part influenced by historical isolation in refugia, as historically isolated populations can be clearly distinguished by their spectral characteristics. Given the known history of population isolation in these refugia, cultural drift is the most likely explanation for the differences in spectral characteristics in chowchilla song. Rainforests of the Wet Tropics are currently largely connected, and finding such clear divergence in chowchilla song despite the current population connectivity raises many interesting questions about the maintenance and consequences of such variation. For instance, do genetic and morphological patterns of divergence match the pattern of song divergence, and are these also mainly influenced by drift? Do large-scale patterns of song element sharing and song syntax reveal similar evolutionary processes as found in this study? The similarity and order of song elements may give a greater resolution of geographical patterns than the spectral characteristics. Determining the effect of small-scale differences in the habitat structure on spectral song characteristics would clarify whether habitat structure is important in shaping song divergence in this species. Finally, an intriguing question is whether chowchillas discriminate between different song variants, and hence whether such large-scale song divergence, as found in this study, may act as a barrier to gene flow.

Acknowledgments We thank the volunteers who helped record and observe chowchillas in the field as well as two anonymous referees for their constructive comments. This research was funded by a James Cook University Graduate Research Scheme, Rainforest CRC Research Support Scheme, Stuart Leslie Bird Research Award and a Birds Queensland grant. Research ethics approval was granted on 20 January 2004 (James Cook University, A870_04).

1581

1582

ANIMAL BEHAVIOUR, 74, 5

References Badyaev, A. V. & Leaf, E. S. 1997. Habitat associations of song characteristics in Phylloscopus and Hippolais warblers. Auk, 114, 40e46. Baker, M. C. 1996. Depauperate meme pool of vocal signals in an island population of singing honeyeaters. Animal Behaviour, 51, 853e858. Baker, M. C. 2006. Differentiation of mating vocalizations in birds: acoustic features in mainland and island populations and evidence of habitat-dependent selection on songs. Ethology, 112, 757e771. Baker, M. C. & Cunningham, M. A. 1985. The biology of bird-song dialects. Behavioral and Brain Sciences, 8, 85e133. Baker, M. C. & Jenkins, P. F. 1987. Founder effects and cultural evolution of songs in an isolated population of chaffinches, Fringilla coelebs, in the Chatham Islands. Animal Behaviour, 35, 1793e1803. Baker, M. C., Baker, E. M. & Baker, M. S. A. 2001. Island and island-like effects on vocal repertoire of singing honeyeaters. Animal Behaviour, 62, 767e774. Baker, M. C., Baker, E. M. & Baker, M. S. A. 2003a. Songs of the red-capped robin, Petroica goodenovii: comparison of acoustic features in island and mainland populations. Emu, 103, 329e335. Baker, M. C., Baker, M. S. A. & Baker, E. M. 2003b. Rapid evolution of a novel song and an increase in repertoire size in an island population of an Australian songbird. Ibis, 145, 465e471. Cavalli-Sforza, L. L. 2000. Genes, Peoples and Languages. New York: North Point Press. Clarke, K. R. 1993. Non-parametric multivariate analyses of changes in community structure. Australian Journal of Ecology, 18, 117e143. Dawkins, R. 1976. The Selfish Gene. Oxford: Oxford University Press. Doutrelant, C., Lemaitre, O. & Lambrechts, M. M. 2001. Song variation in blue tit Parus caeruleus populations from Corsica and mainland southern France. Ardea, 89, 375e385. Endler, J. A. 1992. Signals, signal conditions, and the direction of evolution. American Naturalist, 139, S125eS153. Grant, P. R. & Grant, B. R. 1995. The founding of a new population of Darwin’s finches. Evolution, 49, 229e240. Grant, B. R. & Grant, P. R. 1996. Cultural inheritance of song and its role in the evolution of Darwin’s finches. Evolution, 50, 2471e2487. Grant, P. R. & Grant, B. R. 1997. Mating patterns of Darwin’s Finch hybrids determined by song and morphology. Biological Journal of the Linnean Society, 60, 317e343. Hopkins, M. S., Ash, J., Graham, A. W., Head, J. & Hewett, R. K. 1993. Charcoal evidence of the spatial extent of the Eucalyptus woodland expansions and rainforest contractions in North Queensland during the late Pleistocene. Journal of Biogeography, 20, 357e372. Hugall, A., Moritz, C., Moussalli, A. & Stanisic, J. 2002. Reconciling paleodistribution models and comparative phylogeography in the Wet Tropics rainforest land snail Gnarosophia bellendenkerensis (Brazier 1875). Proceedings of the National Academy of Sciences, U.S.A., 99, 6112e6117. Jansen, A. 1993. The ecology and social behaviour of chowchillas, Orthonyx spaldingii. Ph.D. thesis, James Cook University of North Queensland. Jansen, A. 1999. Home ranges and group-territoriality in chowchillas Orthonyx spaldingii. Emu, 99, 280e290. Joseph, L., Moritz, C. & Hugall, A. 1995. Molecular support for vicariance as a source of diversity in rainforest. Proceedings of the Royal Society of London, Series B, 260, 177e182. Kershaw, A. P. 1994. Pleistocene vegetation of the humid tropics of northeastern Queensland, Australia. Palaeogeography, Palaeoclimatology, Palaeoecology, 109, 399e412.

Koetz, A. H., Westcott, D. A. & Congdon, B. C. 2007. Spatial pattern of song element sharing and its implications for song learning in the chowchilla, Orthonyx spaldingii. Animal Behaviour, 74, 1019e1028. Kroon, F. J. & Westcott, D. A. 2006. Song variation and habitat structure in the Golden Bowerbird. Emu, 106, 263e272. Lynch, A. 1996. The population memetics of birdsong. In: Ecology and Evolution of Acoustic Communication in Birds (Ed. by D. E. Kroodsma & E. H. Miller), pp. 181e197. London: Cornell University Press. Lynch, A. & Baker, A. J. 1994. A population memetics approach to cultural-evolution in chaffinch song: differentiation among populations. Evolution, 48, 351e359. McCune, B. & Mefford, M. J. 1999. PC-ORD: Multivariate Analysis of Ecological Data. Gleneden Beach, Oregon: MjM Software Design. McGuire, M. 1996. Dialects of the chowchilla Orthonyx spaldingii in upland rainforest of north-eastern Australia. Emu, 96, 174e180. Manly, B. J. 1997. Randomization, Bootstrap and Monte Carlo Methods in Biology. London: Chapman & Hall. Mielke, P. W. & Berry, K. J. 1994. Permutation tests for common locations among samples with unequal variances. Journal of Educational and Behavioral Statistics, 19, 217e236. Mundinger, P. C. 1980. Animal cultures and a general theory of cultural evolution. Ethology and Sociobiology, 1, 183e223. Nicholls, J. A. & Goldizen, A. W. 2006. Habitat type and density influence vocal signal design in satin bowerbirds. Journal of Animal Ecology, 75, 549e558. Nix, H. A. & Switzer, M. A. 1991. Rainforest Animals: Atlas of Vertebrates Endemic to Australia’s Wet Tropics. Canberra: Australian National Parks and Wildlife Service. Richards, D. G. & Wiley, R. H. 1980. Reverberations and amplitude fluctuations in the propagation of sound in a forest: implications for animal communication. American Naturalist, 115, 381e399. Ruegg, K., Slabbekoorn, H., Clegg, S. & Smith, T. B. 2006. Divergence in mating signals correlates with ecological variation in the migratory songbird, Swainson’s thrush (Catharus ustulatus). Molecular Ecology, 15, 3147e3156. Ryan, M. J. & Brenowitz, E. A. 1985. The role of body size, phylogeny, and ambient noise in the evolution of bird song. American Naturalist, 126, 87e100. Ryan, M. J. & Kime, N. M. 2002. Selection on long-distance acoustic signals. In: Acoustic Communication (Ed. by A. M. Simmons, A. N. Popper & R. R. Fay), pp. 225e274. New York: Springer-Verlag. Schneider, C. & Moritz, C. 1999. Rainforest refugia and evolution in Australia’s Wet Tropics. Proceedings of the Royal Society of London, Series B, 266, 191e196. Schneider, C. J., Cunningham, M. A. & Moritz, C. 1998. Comparative phylogeography and the history of endemic vertebrates in the Wet Tropics rainforests of Australia. Molecular Ecology, 7, 487e498. Schneider, C. J., Smith, T. B., Larison, B. & Moritz, C. 1999. A test of alternative models of diversification in tropical rainforests: ecological gradients vs. rainforest refugia. Proceedings of the National Academy of Sciences, U.S.A., 96, 13869e13873. Schodde, R. & Mason, I. J. 1999. The Directory of Australian Birds: Passerines. Collingwood: CSIRO Publishing. Slabbekoorn, H. & Smith, T. B. 2002a. Bird song, ecology and speciation. Philosophical Transactions of the Royal Society of London, Series B, 357, 493e503. Slabbekoorn, H. & Smith, T. B. 2002b. Habitat-dependent song divergence in the little greenbul: an analysis of environmental selection pressures on acoustic signals. Evolution, 56, 1849e1858.

KOETZ ET AL.: GEOGRAPHICAL VARIATION IN CHOWCHILLA SONG

Slater, P. J. B. 1989. Bird song learning: causes and consequences. Ethology, Ecology and Evolution, 1, 19e46. Specht, R. 2005. Avisoft SASLab Pro: Sound Analysis and Synthesis Laboratory. Version 4.38. http://www.avisoft.com. Tracey, J. G. 1982. The Vegetation of the Humid Tropical Region of North Queensland. Melbourne: CSIRO. Webb, T. & Bartlein, P. J. 1992. Global changes during the last 3 million years: climatic controls and biotic responses. Annual Review of Ecology and Systematics, 23, 141e173. Webb, L. J. & Tracey, J. G. 1981. Australian rainforests: patterns and change. In: Ecological Biogeography of Australia (Ed. by A. Keast), pp. 605e694. The Hague: W. Junk.

Westcott, D. A. & Kroon, F. J. 2002. Geographic song variation and its consequences in the golden bowerbird. Condor, 104, 750e760. Wiley, R. H. 1991. Associations of song properties with habitats for territorial oscine birds of eastern North America. American Naturalist, 138, 973e993. Winter, J. W. 1997. Responses of non-volant mammals to Late Quaternary climatic changes in the Wet Tropics region of northeastern Australia. Wildlife Research, 24, 493e511. Zimmerman, G. M., Goetz, H. & Mielke, P. W. 1985. Use of an improved statistical-method for group comparisons to study effects of prairie fire. Ecology, 66, 606e611.

1583