Evolution of vocal performance and song complexity in island birds

26 Oceanic islands share distinctive characteristics thought to underlie a set of parallel 27 evolutionary trends across islands and taxonomic groups – including life history traits, 28 morphology and visual signals. To which extent acoustic signals also change in parallel on 29 islands is less clear. Some important processes associated with insularity, such as founder 30 effects and reduced sexual selection, could lead to a decrease in vocal performance and 31 song complexity on islands. In a field-based study, we recorded 11 insular species and their 32 closest mainland relatives. Out of the 11 species pairs, 6 live in the tropics (São Tomé / 33 Mount Cameroon), and 5 in the temperate region (Madeira / southern France). For each 34 species we measured two proxies of vocal performance (song duration and syllable rate) 35 and one proxy of song complexity (syllable diversity). This study did not recover a clear 36 relationship between the island environment and song traits. If as expected, syllable rate 37 was lower in island species than in their mainland counterparts, the two other proxies 38 showed no clear island-mainland pattern of divergence. Several factors may explain the 39 absence of reduction for song duration and syllable diversity. Among those, relaxation of 40 interspecific competition on islands may have led to an increase in syllable diversity, or 41 correlations between song variables may have constrained song evolution. More studies on 42 island species are needed to draw a better picture of divergence patterns and go beyond 43 the confounding ecological factors that could explain peculiar song characteristics in 44 islands. 45


Introduction
Clegg & Owens 2002; but see Meiri et al. 2008) could lead to peculiar song features in 126 island populations (Potvin, 2013). Thus, we also controlled for differences in body mass 127 between insular and continental species in our statistical analyses. The song of some of 128 these focal species is already known as differing in terms of frequency ranges (Robert et al. 129 2021). Yet, insular characteristics of the others features of their songs have not been 130 studied. We measured vocal performance (song duration and syllable delivery rate) and 131 song complexity (syllable diversity), and predicted lower values on islands compared to the 132 mainland, especially for vocal performance.

Study site and species pairs 136
Our study was based on the comparison of vocal performance and song complexity proxies 137 of insular and continental related species pairs. We focused on two pairs of insular and 138 continental acoustic communities. The tropical pair comprised São Tomé Island, hereafter 139 referred to as 'tropical island', and Mount Cameroon, hereafter referred to as 'tropical 140 mainland'. The temperate pair comprised Madeira Island, hereafter referred to as 141 'temperate island', and southern France, hereafter referred to as 'temperate mainland'. São 142 Tomé and Madeira are located 250 km and 657 km away from the mainland, respectively. 143 We worked on comparable mature or primary forests (mature laurel and evergreen oak 144 forests in the temperate zone, and primary rainforests in the tropics) with similar 145 propagation properties (Robert et al. 2019). For each community (tropical island, tropical mainland, temperate island and temperate mainland), we worked on three sites (separated On the islands we recorded the common vocalising species. We then recorded their 149 closest vocalising relative on the mainland (see Appendix 2 in Supporting Information for an  When only ranges were reported or when masses were reported separately for males and 191 females, we assumed the midpoint median was the mean of the two values. Body mass did 192 not differ significantly between island and mainland species (pairwise comparisons using 193 Wilcoxon rank-sum test, P > 0.35). song as a measure of syllable diversity. We divided the number of syllables per song by the 219 song duration to obtain syllable rate (Appendix 4 in Supporting Information). Duration and 220 syllable diversity were significantly and positively correlated (r=0.45, P<0.001, Appendix 5 syllable diversity (r = 0.13, P <0.001) and duration (r = -0.058, P <0.05). 223 To assess whether we had recorded enough individuals to characterise species song 224 characteristics in an accurate way, we computed accumulation curves of the means of song 225 duration and maximum sampled syllable rate (Appendix 6 in Supporting Information). This 226 was not done for syllable diversity, as this would require labelling all syllables from all songs 227 to see whether adding a new song would add to an existing accumulated syllable diversity, 228 which was technically unfeasible with our means. 229 230

Statistical analyses 231
All three dependent variables (song duration, syllable rate and syllable diversity) were log-232 transformed to ensure their normality (based on graphical assessment of residuals). We 233 used linear mixed-effects models (LMM) using the R package nlme (Bates & Pinheiro 2011) 234 to explore patterns of variation. For each of the three response variables, we ran a model 235 testing insularity (island/mainland), latitude (tropical/temperate), the interaction between 236 latitude and insularity, body mass, and altitude as fixed effects. We included bird mass 237 (from HBW, Billerman et al. 2019) as a proxy for syrinx mass, long recognized as 238 influencing the timing of sound production (e.g., Brackenbury 1978; Gaunt 1987). Syrinx 239 mass primarily determines the minimum frequency that a bird can achieve (Ryan & 240 Brenowitz 1985; Potvin 2013), but it may also affect other variables such as syllable rate as 241 a result of trade-offs between rate and frequency bandwidth (Podos 1996(Podos , 1997. We 242 included altitude because of its negative correlation with temperature and species number, 243 which has been found to influence song properties such as frequency (Morton 1975; 244 Kirschel et al. 2009). Finally, we included "species" nested in "pair" nested in "family" as a 245 random factor to account for the phylogenetic structure of the data (Griffith 2000; the analyses by sample size, i.e., by the number of recordings per individual. We selected 248 the best model by minimising the Akaike Information Criterion (AIC: Akaike 1987), and 249 considered that models differing by less than 2 AIC units were equivalent. 250 We were also interested in assessing which particular species could be driving 251 potential song differences between islands and mainland. For this, we compared song 252 duration, syllable rate, and song complexity for each of the 11 island-mainland species 253 pairs, by conducting t-tests. All statistical analyses were conducted in R v.3.05.1 (R 254

Development Core Team 2017). 255
In order to investigate how phylogenetic relationships among species could influence 256 our results, we also analysed the same variables (song duration, syllable rate and syllable 257 diversity) using a Bayesian phylogenetic mixed model (BPMM) approach (Appendix 7 in 258 Supporting Information). 259 260

Results 261
All three song variables were affected by the interaction between latitude and insularity in 262 different ways (Table 1). 263 Syllable rate -As predicted, syllable rate was lower on islands (Table 1; Figure 2). 264 Syllable rate was not affected by altitude but it was significantly influenced by body mass, 265 with lighter species singing faster (Table 1). When we looked at each species 266 independently, four tropical pairs (cuckoo, oriole, speirops, and sunbird) out of six showed 267 lower syllable rates on the island, thereby driving the insularity x latitude effect (the 268 paradise-flycatcher and the prinia showing higher syllable rate on the island). Three 269 temperate pairs (robin, firecrest, and chaffinch) out of five showed lower syllable rate on the 270 island (the blackcap showing higher syllable rate and the blackbird showing no change, Figure 1). Accumulation curves of the mean of maximum sampled syllable rate (Appendix 6 in Supporting Information) showed that a plateau was reached for 9 out 11 pairs, except for 273 the paradise-flycatcher and the blackcap, for which increasing the sampling effort would 274 have increased their already higher syllable rate on the island, which would have been 275 conservative for our results. 276 Song duration -Contrary to our expectation, song duration was longer in the 277 tropical island than in the tropical mainland and it did not differ between the temperate 278 island and the mainland (latitude by insularity interaction Table 1; Figure 2). It was not 279 affected by altitude but it was significantly influenced by body mass, heavier species 280 producing longer songs (Table 1). When looking at each species independently, we found 281 that four tropical pairs (cuckoo, oriole, paradise-flycatcher, and sunbird) out of six showed 282 longer songs on the island, thereby driving the island x latitude effect (the speirops showed 283 a significantly lower song duration, and the prinia did not show any significant variation). 284 The absence of difference in song duration between island and mainland in the temperate 285 area is due to the fact that only two pairs (firecrest, chaffinch) out of five showed longer 286 songs on the island (the blackbird and the blackcap showed shorter songs and the robin did 287 not show any change, Figure 2). Accumulation curves of mean of song duration maxima 288 (Appendix 6 in Supporting Information) showed that all species were sufficiently sampled, 289 except the insular speirops and blackbird. Since both species presented shorter songs on 290 islands, increasing sampling effort could have shown no variation between island and 291 mainland, or could have reinforced the insularity effect. 292 Syllable diversity -As expected, syllable diversity was lower on the temperate 293 island than on the temperate mainland but, in contrast, it was higher on the tropical island 294 than on the tropical mainland. Syllable diversity was not affected by body mass but it was 295 significantly negatively correlated to altitude (Table 1). The general increase we found in the 296 out of six (the speirops showed lower values and the cuckoo and oriole did not show any 298 variation). The lower syllable diversity found in the temperate island was driven by three 299 species pairs (blackbird, blackcap, and chaffinch) out of five (the firecrest showed a 300 significant increase and the Robin did not change, Figure 3). were equivalent to the LMM ones in terms of fixed-effect structure of the final models. In 304 particular, BPMM found a positive effect of insularity on syllable rate which was lower on 305 islands. As in LMM, syllable rate was significantly affected by mass, lighter species 306 producing faster songs. Contrasting with LMM, altitude had a significant and positive 307 (although very weak) effect on syllable rate. Song duration was not affected by altitude but 308 it was significantly influenced by body mass, lighter species producing longer songs. 309 Results of BPMM indicated a significant effect of interaction between "insularity" and 310 "latitude" on song duration: song duration was higher in the tropical island than in the 311 tropical mainland but it did not differ between the temperate island and mainland. Overall, 312 effects of the interaction "latitude x insularity" were weaker in the BPMMs than in LMMs. 313 This indicates that phylogenetic relationships explain a part of the significant link between 314 our three variables and the interaction between "latitude x insularity". The temperate pairs 315 represent more recent colonisation and divergence events than the tropical pairs (except for 316 the cuckoo pair) which probably enhanced the effect of latitude. As in LMM, BPMM 317 indicated that syllable diversity changed with latitude and insularity (significant latitude by 318 insularity interaction). It was lower on the temperate island than the temperate mainland but 319 it was higher on the tropical island than the tropical mainland. As in LMM, syllable diversity 320 to altitude. 322

Discussion 324
A clear relationship between the island environment and song did not emerge from this 325 study. The only supported pattern was a decrease in syllable rate both in the tropical and 326 temperate islands. However, on islands, song duration was not lower (it even increased on 327 the tropical island), and although syllable diversity was lower in the temperate island, this 328 pattern did not hold in the tropics. 329

Syllable rate in island species 330
The most consistent result across latitudes was the lower syllable rates shown by insular 331 species in comparison to their mainland counterparts. As mentioned in the introduction, 332 syllable rate is probably the song feature for which sexual selection towards exaggeration 333 Evolutionary trade-offs could also occur between song features and concur to 362 explain the documented change. In particular, Podos (1996Podos ( , 1997) identified a negative link 363 between frequency bandwidth and trill rate in Emberizidae. In the two species that sing trills 364 in our data set (chaffinch and sunbird), syllable rate may be equivalent to trill rate. 365 Aforementioned trade-off is, however, unlikely to explain the observed decrease, as the 366 frequency range of the chaffinch song decreases, rather than increases, on the island, 367

Song duration in islands species 381
Contrary to our expectation, relatively to the mainland, songs were longer on the tropical 382 island but were similar in length on the temperate one. As explained above, trade-offs or 383 correlations between acoustic traits may counterbalance the predicted effect of insularity on 384 song and explain why we did not find the expected results. Trade-offs between frequency 385 ranges and trill rate (Podos 1997) and between amplitude (loudness) and syllable diversity 386 past studies in the temperate region suggested that song duration is an adequate proxy for 408 song complexity due to its role in mate choice and male-male competition. This role was 409 demonstrated for most of the temperate species we studied here, but it remains to be 410 tested in tropical species. 411 Finally, contrasted results in vocal performance proxies could also be explained by 412 cultural drift during colonization that is susceptible to draw unpredictable island-mainland 413

Song complexity on islands 416
We found that island syllable diversity was, as expected, lower in the temperate area but, 417 against our expectations, it was higher in the tropical area. Apart from the structural 418 correlation with song duration we discussed above, this pattern could be due to changes in Character release is expected to be more pronounced at low latitudes where there are 437 greater differences in species numbers between mainland and islands (Robert et al. 2021); 438 this could explain why we found a higher syllable diversity on the tropical island compared 439 to the tropical mainland. 440 Secondly, two distinct consequences of founder events also predict reduced song 441 complexity in oscines, for which song is a learned behaviour (in our study, 10/11 species 442 pairs are oscines). Reduced complexity could be the consequence of (i) an interruption of Overall, this study confirms some of the predictions on the direction of song evolution on 485 islands but, more importantly, highlights how this process may be affected by several 486 mechanisms that can prevent convergent evolution of song traits on islands. To better 487 understand patterns of changes on islands for song performance and complexity, a larger 488 comparative study is needed. Field data are long to collect but are crucial to control both for 489 latitude and for time since island colonisation. Our study supported previous suggestions 490 that some aspects of the 'island effect' may be stronger on temperate regions. As with 491 genetic variation, studies on the evolution of song complexity on islands will need to 492 distinguish between short-term responses to founder events and long-term adaptations to 493 the island environment. In this respect, studies should sample species pairs across 494 latitudes and spanning a wide time range, from early colonisers to well-differentiated of vocal production in order to better understand trade-offs or structural correlations that 497 could involve proxies of song performance or song complexity. Overall, our study provides 498 additional elements to the debate regarding the predictability of evolutionary trends on 499 oceanic islands and the possibility of convergent evolution of acoustic signals within the 500 'insularity syndrome'.