Geographical and individual variation in echolocation calls of the ...

ethology international journal of behavioural biology

Ethology

Geographical and individual variation in echolocation calls of the intermediate leaf-nosed bat, Hipposideros larvatus Tinglei Jiang*, , Rong Liuà, Walter Metzner§, Yuyan You*, Shi Li–, Sen Liu* & Jiang Feng* * Key Laboratory for Wetland Ecology and Vegetation Restoration of National Environmental Protection, Northeast Normal University, Changchun, China Department of Integrative Biology and Physiology, UCLA, Los Angeles, CA, USA à Center for the Environmental Implications of Nanotechnology and Department of Chemical and Biomolecular Engineering, UCLA, Los Angeles, CA, USA § Department of Integrative Biology and Physiology, UCLA, Los Angeles, CA, USA – Key Laboratory of Vegetation Ecology of Education Ministry, Institute of Grassland Science, Northeast Normal University, Changchun, China

Correspondence Jiang Feng, Key Laboratory for Wetland Ecology and Vegetation Restoration of National Environmental Protection, Northeast Normal University, Changchun 130024, China. E-mail: [email protected]

Received: December 31, 2009 Initial acceptance: February 15, 2010 Final acceptance: March 24, 2010 (G. Beckers) doi: 10.1111/j.1439-0310.2010.01785.x

Abstract The cause and significance of variation in echolocation call frequency within hipposiderid bats is not well understood despite an increasing number of allopatric and sympatric examples being documented. We examined variation patterns in the resting frequency (RF) of echolocation calls emitted by the intermediate leaf-nosed bat, Hipposideros larvatus, on a broad geographical scale. Data mining technology and Kruskal–Wallis test both showed substantial variation with a longitudinal pattern in RF in H. larvatus among colonies, and this variation was associated with geographical distance and not body size. In addition, we found that a high degree of variability between individuals was hidden under the geographical variation. The results support an effect of random cultural drift, and challenge the prey detection hypothesis. Moreover, an acoustic difference among local island colonies may be indicative of a vocal dialect. We found that each colony of H. larvatus seems to maintain a ‘private bandwidth’, which could be used for colony identity and individual communication thus helping individuals and colonies to get a number of fitness benefits.

Introduction Most social behaviours, such as predatory behaviour, individual recognition, mate selection, etc., are dependent on intraspecific communication (Wilczynski & Ryan 1999). Geographical variation in animal vocalisations provides a best opportunity to clarify the myriad factors shaping the evolution and divergence of communication signals (Endler 1983). Moreover, geographical comparisons are important to test a variety of hypotheses about the selective pressure of variation in vocal signals (Carroll & Corneli 1999). Expanding these scenarios, variability of acoustic signals can facilitate population divergence, reproductive isolation and speciation (Slabbekoorn & Smith 2002). Ethology 116 (2010) 691–703 ª 2010 Blackwell Verlag GmbH

Variation in intraspecific communication signals can occur among populations or between individuals within populations. Geographical variation in intraspecific acoustic signals has been observed in a range of animal groups, including invertebrates (Eiriksson 1992), birds (Soha et al. 2004; Wright et al. 2005) and mammals (Mitani et al. 1992). In geographically separated populations, difference in acoustic signals may be a result of genetic differentiation (Catchpole & Slater 1995), learning or cultural drift (Yoshino et al. 2008; Chen et al. 2009) or adaptation to local environmental conditions (Wilczynski & Ryan 1999; Slabbekoorn & Smith 2002; Gillam & McCracken 2007). Within populations, variability in signal structure makes individuals to respond to varying 691

Call variation in Hipposideros larvatus

behavioural and ecological factors, such as change in predation risk (Endler 1987) or local habitat type (Koetz et al. 2007). Bats within the families Rhinolophidae and Hipposideridae emit echolocation pulses dominated by a constant frequency (CF) component (Neuweiler 2000), which in non-flying, i.e. ‘‘resting’’ bats, is matched to their acoustic fovea (Schuller & Pollak 1979). The CF portion emitted at rest is also called the bat’s ‘‘resting frequency’’ (RF). RFs are highly variable between species and also individuals. Numerous studies showed that differences in the RF among populations were associated with geographical barriers in some species; such as in Rhinolophus ferrumequinum (Schnitzler et al. 1976; Taniguchi 1985; Huihua et al. 2003), Hipposideros cervinus (Francis & Habersetzer 1998) and H. ruber (Guille´n et al. 2000). It was suggested that CF differences between the regions of distribution of R. cornutus pumilus (Yoshino et al. 2008) and R. monoceros (Chen et al. 2009) might have developed through random cultural drift. In some species, variation in RF was correlated with body size (Yoshino et al. 2006) or local environmental condition (e.g. humidity; Guille´n et al. 2000). Although biosonar signals are also known to vary among individuals (Neuweiler et al. 1987; Fenton et al. 2004; Hiryu et al. 2006; Gillam & McCracken 2007), little is known about how much of the inter-individual variability may be hidden under geographical variations. Bats alter their sonar signals in the presence of echolocating conspecifics (Obrist 1995; Hiryu et al. 2006; Chiu et al. 2008), and this can influence the behaviour of other bats (Fenton 2003). Moreover, Yovel et al. (2009) suggested that bats can learn the average call characteristics of individuals and use them as a reference for classification of individual identity. Recently, increasing evidence indicates that echolocation has probably evolved from acoustic communication (Ma et al. 2006; Chiu et al. 2008; Jones 2008; Kazial et al. 2008; Yovel et al. 2009). Jones & Barlow (2004) suggested that the major factor promoting acoustic divergence in cryptic pipistrelle species was correlated with facilitating communication with conspecifics, rather than resource partitioning. As a result, the divergence in call frequency is primarily driven by social character displacement, allowing each species to have a ‘private bandwidth’ for species recognition (Russo et al. 2007). However, in geographically separated populations, it is not clear whether each population maintains a ‘private bandwidth’ for identification and communication of members. 692

T. Jiang et al.

Machine learning and Data Mining are becoming increasingly important areas of engineering and computer science, and have been successfully applied to a wide range of problems in science. It can be applied to a variety of situation where data are noisy, incomplete, heterogeneous, or even nonstructural and aims to discover meaningful rules and patterns which might be ignored to the strict statistical analysis. A number of studies showed that data mining can be successfully applied to the problem of automatic species identification (Gaston & O’Neill 2004; Moyo et al. 2006; Mayo & Watson 2007). Moreover, much progress has been made in sound classification technology. For instance, program such as MATLAB and WEKA (Witten & Frank 2005) can be used to automatically classify large datasets in a very short time. Melendez et al. (2006) suggested that classification and regression tree analysis (CART) can accurately classify the communication signals of the little brown bat. CART is a type of decision tree (DT) analysis which splits all of the dependent variables using the optimal predictor variables. In other words, the DT approach is most useful in classification in the data mining techniques (Dunham 2003). There are many advantages of using the DT approach for classification. First of all, DTs are easy to use and efficient. Second, rules can be generated and easy to interpret and understand. Third, DTs scale well for large data sets because the tree size is independent of the size of data. Finally, trees can be constructed even for high dimensional data (Han et al. 2005). Therefore, in this study, we tried to classify the colonies by data mining techniques based upon resting frequency (RF) of Hipposideros larvatus. The intermediate leaf-nosed bat (H. larvatus) is widespread in Asia with a geographical distribution that includes Bangladesh, China, Indonesia, Malaysia, Myanmar and Northeast India (Bates & Harrison 1997). In Mainland China, the species is found in South and Southwest China, including Yunnan, Guizhou, Guangdong, Guangxi, Hainan (Wang 2003; Smith & Xie 2008). Moreover, the bat species, which on average emits RFs 85 kHz in resting frequency (RF) (Fig. 1), used different echolocation call frequency from those in Malaysia and should therefore be considered a distinct subspecies of H. larvatus, that is H. l. poutensis (Thabah et al. 2006). In this study, we report intraspecific patterns of variability in the RF of H. larvatus in Mainland China and attempt to understand the causes and meaning underlying this variation in RF. The results of our analysis led us to answer four primary questions: Ethology 116 (2010) 691–703 ª 2010 Blackwell Verlag GmbH


T. Jiang et al.

(a)

(b)

Fig. 1: Spectrogram (a) and power spectrum (b) of typical echolocation calls of Hipposideros larvatus. The power spectrum was constructed based upon the second call on the spectrogram. The measurement point (indicated by arrow) of the resting frequency (RF) is shown in the power spectrum. The RF value of each call was evaluated according to the frequency containing the maximum energy.

(1) does body size constraint the evolution of geographical variation of RF; (2) are variations in RF associated with geographical distances and longitudinal gradient, therefore indicating cultural drift as a primary determinant of RF variation; (3) is there a high degree of variability between individuals existing under the geographical variation and (4) does each colony have a ‘‘private bandwidth’’ for colony identity and individual communication. Materials and Methods Field Sampling

Our preliminary survey detected only a small acoustic variation between H. larvatus from single caves. We therefore assumed a priori that all individuals from each cave belong to a single colony. Surveys were carried out from April 2008 to September 2008 at nine caves, covering most of the species range in Mainland China (Fig. 2). A total of 100 adult bats was collected and forearm length measured (FAL) to within 0.01 mm using digital calipers. A bat was considered to be adult when the phalangeal epiphyses were fused with the diaphyses as assessed by trans-illuminating the wing with a bright flash light (Racey 1974). Reproductive status of female bats was evaluated according to Racey (1974, 1982). In the study, pregnant and lactating females were excluded from analyses. Body mass is not a reliable indicator of age in bats, mostly because it varies with food supply and can be seasonably different, especially in hibernating species. For each cave, we also determined its latitude, longitude and elevation using GPS (eTrex Vista, Garmin International Inc., Olathe, KS, USA). Ethology 116 (2010) 691–703 ª 2010 Blackwell Verlag GmbH

Echolocation Call Recording and Analysis

Bat echolocation calls were recorded with a Real-time ultrasonic detector (UltraSoundGate 116; Avisoft Bioacoustics, Berlin, Germany) that was connected to a laptop computer positioned approx. 30 cm in front of the hand-held bats. The sampling frequency was 441 kHz. Bats were hand-held to avoid Doppler effects, which obviously would have blurred studying individual CF variability: only under these recording conditions are RFs individually characteristic and stable and the CF portion of the echolocation pulses emitted by the motionless bat matches the bat’s ‘acoustic fovea’ (Schuller & Pollak 1979). Conversely, free-flying bats emit calls with variational frequency compensated for flight-induced Doppler-effects, so CFs recorded from flying bats do not match the individual’s resting frequency (Schnitzler 1970). We used the software package Avisoft SasLab Pro (length of FFT: 1024 points with 94% overlap, spectral resolution: 195 Hz; Avisoft Bioacoustics) to determine RF from at least 100 high-quality calls (signal-to-noise ratio >10:1) from call sequences of each individual based upon the following criteria: First, initial calls within a call series were not considered for analysis, because such calls may show transient, lower frequency values before reaching the final RF level (Siemers et al. 2005). Second, when calls were emitted in groups (doublets, triplets, etc.), only the second call per group was chosen (Russo et al. 2001). Statistical Analysis

We performed Kruskal–Wallis and Dunn’s multiplecomparison tests (Zar 1999) to determine acoustic 693


T. Jiang et al.

Fig. 2: Study area with the caves of sampled localities. In the map, scale of 400 km is for mainland China; scale of 80 km is only for Hainan province. HNK, Hainan Kuang cave; HNX, Hainan Xianan cave; HNS, Hainan Shilun cave; HNR, Hainan Xianren cave; YND, Yunnan Dahei cave; GDW, Guangdong Wuxing cave; YNF, Yunnan Fangkong cave; GXP, Guangxi Pozhi cave; JXL, Jiangxi Luohan cave.

variation among colonies and between individuals in each colony. In this analysis, data from females and males were initially tested separately, and then combined. We conducted a Mann–Whitney U-test to evaluate gender differences in RF in each cave. Variation in the RF related to the sexual and geographical structure of the colonies was studied with a linear model that included SEX and CAVE as fixed effects. Since changes in RF may be associated with variation in FAL, we used FAL as reference for assessing the importance of variation in RF. Therefore, an ANCOVA model was used to determine the effects of FAL to changes in RF. In this model, the means of RF and FAL per CAVE (Table 1) were

Table 1: Resting frequency (RF) in echolocation call of adult female (F) and male (M) Hipposideros larvatus from nine caves in China Geographical position

N

CF

Caves Longitude ⁄ latitude ⁄ F M F HNK HNX HNS HNR YND GDW YNF GXP JXL

109.45 109.43 109.48 110.21 101.24 111.94 103.77 107.82 114.09

18.58 18.59 18.63 19.94 21.97 22.43 22.69 22.86 25.46

5 4 4 8 4 3 6 9 9

5 9 3 5 3 3 7 5 8

87.29 86.19 89.38 88.76 91.78 87.03 89.07 87.68 85.59

M

0.95 0.94 2.42 0.95 0.69 1.50 0.82 2.08 2.04

86.62 87.09 88.55 88.31 90.01 83.58 88.70 87.84 84.53

DIF

1.31 1.92 1.89 2.25 0.53 0.35 1.07 0.97 2.28

Values are given as x SD. The results of the Mann–Whitney U-test for sexual difference are shown as DIF. Significant levels are: +, p < 0.001 and –, p > 0.05.

694

+ + + + + + + ) +

the dependent variable and covariate, respectively, and CAVE and SEX were the main fixed effects. To assess if differences in RF were related to geographical distance, we first calculated a dissimilarity matrix of acoustic distances using RF differences in kHz between caves. We then calculated a geographical distance matrix from the latitude and longitude of each cave, and compared the acoustic and geographical distance matrices using a non-parametric Mantel test of matrix association (Mantel 1967; Schneider et al. 2000) in Arlequin (Schneider et al. 2000). The Mantel test was performed with 1000 permutations. To test if geographical location is associated with RF, we tested for associations between observed RF and longitude and latitude of each cave. The linear regressions of the means of RF against each geographical variable were conducted. To asses patterns of call variability among-colonies, between-individuals and within-individuals, coefficients of variation (CV ¼ SD 100=x) were calculated for each level in different geographical scales. In this study, the entire sample area is separated into Mainland and Hainan Island by the Qiongzhou Strait. Moreover, the geographical distance between Hainan Island colonies (HNK, HNX, HNS) ranged only from 2.39 to 6.90 km. Therefore, we divided the sampled area into four different categories. (1) All caves: all caves within the entire study area; (2) Mainland caves: all caves on the mainland excluding the caves on Hainan island; (3) All island caves: excluding all caves on the mainland and (4) Local island caves: including HNK, HNX and HNS. Within-individual Ethology 116 (2010) 691–703 ª 2010 Blackwell Verlag GmbH


T. Jiang et al.

CV (CVw) were estimated for each individual from the means of RFs emitted by this individual. Between-individual CV (CVb) were calculated from the grand mean and standard deviation of RFs based upon average values of individual bats. Among-cave CV (CVa) are estimated based on the grand mean and SD of RFs over all caves. CVa was initially calculated for all colonies combined. In addition, we calculated the ratio of CVb to CVw (CVb ⁄ CVw) as a measure of relative inter-individual variability. If the CVb ⁄ CVw ratio is >1.0 for a given property, then this property is relatively more variable between individuals and may function as individual recognition cue (Robisson et al. 1993; Jouventin et al. 1999;). Data Mining

A classification analysis is performed on the data to predict the cave of each item based on its attributes such as frequency or frequency with gender. The CART algorithm, as implemented in WEKA, was used to confirm the accuracy of best features and boundaries included in the automatic classification algorithm (Witten & Frank 2005). In the modelling process, the data set is partitioned into 10 mutually exclusive subsets. Of the 10 subsets, a single subset is retained as the validation data for testing the model, and the remaining nine subsets are used as training data to develop the classification model. The cross-validation process is then repeated 10 times (the folds), with each of the 10 subsets used exactly once as the validation data. The 10 results from the folds then can be averaged to produce a single estimation. Results From the nine sampled caves, we analysed 17074 calls from 100 individuals with an average of 170 calls per bat. For the entire sample, RF ranged from 85.59 (SD = 2.04) to 91.78 kHz (SD = 0.69) in females and from 84.53 (SD = 2.28) to 90.01 kHz (SD = 1.78) in males (Table 1). The RF values of the bats differed significantly among caves for both genders (Kruskal–Wallis test: H8 = 5703.47, p < 0.001, in females; H8 = 5162.09, p < 0.001, in males; H8 = 1152.46, p < 0.001, for all samples; Table 2 and Fig. 3). Moreover, Dunn’s multiple-comparison test revealed significant geographical variation between colonies by pairwise comparisons except few cases (Colony YNF vs. Colony HNR, Colony HNX vs. Colony JXL, in females; Colony YNF vs. Colony HNS, Ethology 116 (2010) 691–703 ª 2010 Blackwell Verlag GmbH

Table 2: Comparisons of cave samples of Hipposideros larvatus in resting frequency (RF) on Mainland China (HNK–JXL; see Fig. 2 for geographical locations) by the Kruskal–Wallis test (KWT) and Dunn’s multiple-comparison test KWT Samples Female + Male + Both +

YNDa YNFb HNRb GXPc HNXd JXLd HNKe HNSf GDWg YNDa YNFb HNSb GXPc HNXd HNKe HNRf GDWg JXLg YNDa YNFb GXPc HNXd HNKd HNSe HNRf GDWg JXLh

Significance level for KWT is +, p < 0.001. Samples sharing same superscript letters showed no significant differences in Dunn’s multiple-comparison test (p < 0.05). See Table 1 for sample sizes and Fig. 1 for abbreviations of characters.

Colony GDW vs. Colony JXL, in males; Colony HNX vs. Colony HNK, for all samples; Table 2). Variation within caves in RF was generally smaller than among caves (Fig. 3). For each colony, the mean value of RF was consistently higher in females than in males except HNX, and this difference was statistically significant for each cave except GXP (Table 1). Moreover, the RF values of the bats differed significantly between individuals in each colony except between female individuals in colony GDW (Kruskal–Wallis test: H2 = 2.34, p = 0.31). In addition, there was marked overlap in RF between males and females in each colony except GDW (Fig. 3). Whereas CAVES had a significant effect on RF (F8 = 10.639, p < 0.01), but SEX and FAL and their interaction term did not (Table 3). There was a sexual dimorphism in RF with females using slight higher pitched calls in all colonies except HNX where the dimorphism was reversed (Table 1). Resting frequency variations were significantly and positively associated with geographical distances between caves (Mantel test: r = 0.51, p = 0.02; Fig. 4). Linear regressions of RF against longitude and latitude showed that only longitude and RF had a significant negative correlation (r2 = 0.72, p < 0.01; Fig. 3). Resting frequency varied less within-individual than between-individuals and among-colonies in all four categories. In ‘All caves’ and ‘Mainland caves’, variation in RF among-caves was much higher than the variability between-individuals, but an opposite pattern was found in ‘Local island caves’ and ‘All island caves’ (Fig. 5). Moreover, the ratio of CVb ⁄ CVw in all four categories was distinctly >1 (Fig. 5). In addition, RFs showed highest CVb and CVa and lowest CVw for ‘Mainland caves’ compared with any other category. Moreover, CVb ⁄ CVw ratio also indicated the same pattern in ‘Mainland caves’ (Fig. 5). 695


T. Jiang et al.

Fig. 3: Within-individual, within-cave and geographical variation in resting frequency of Hipposideros larvatus. Each point represents a single individual, with error bars showing standard deviations. Solid and hollow points represent female and male individuals, respectively. Individuals are then grouped into caves, as indicated on the vertical axis, which are arranged in order around the longitudinal gradient. The value of longitude of each cave was showed. In addition, comparisons of individual samples of Hipposideros larvatus by the Kruskal–Wallis test (KWT) were showed behind the value of longitude in order around females, males and both sexes. Significance level for KWT is +, p < 0.001 and –, p > 0.05. It is apparent that there are individual differences in resting frequency in each cave, and there is gradual variation around the longitudinal gradient.

The 10-fold cross-validation showed that the average correct classification rate of predicting the cave of each item is 64.43% with frequency attributes. Introducing gender attributes in this model yielded a higher classification accuracy of 74.31%. 696

As shown in Fig. 6, there was some overlap in the classification of caves. In all four inland caves, some data point of RF from each cave was misclassified to the almost all other of caves, especially HNR. However, the classification of data points of RF Ethology 116 (2010) 691–703 ª 2010 Blackwell Verlag GmbH


T. Jiang et al.

Table 3: Effect of SEX, CAVES and FAL on the resting frequency (RF) of Hipposideros larvatus in China Effect

df

MS

F

p

SEX CAVES FAL FAL * SEX ERROR

1 8 1 1 6

1.633 7.310 0.265 1.543 0.687

2.377 10.639 0.386 2.246

0.174