Use of bioacoustics in species identification: Piranhas from genus Pygocentrus (Teleostei: Serrasalmidae) as a case study

The genus Pygocentrus contains three valid piranha species (P. cariba, P. nattereri and P. piraya) that are allopatric in tropical and subtropical freshwater environments of South America. This study uses acoustic features to differentiate the three species. Sounds were recorded in P. cariba, two populations of P. nattereri (red- and yellow-bellied) and P. piraya; providing sound description for the first time in P. cariba and P. piraya. Calls of P. cariba were distinct from all the other studied populations. Red- and yellow-bellied P. nattereri calls were different from each other but yellow-bellied P. nattereri calls were similar to those of P. piraya. These observations can be explained by considering that the studied specimens of yellow-bellied P. nattereri have been wrongly identified and are actually a sub-population of P. piraya. Morphological examinations and recent fish field recordings in the Araguari River strongly support our hypothesis. This study shows for the first time that sounds can be used to discover identification errors in the teleost taxa.


Introduction
Sounds for communication purposes are emitted by a high number of different zoological groups. Many species have evolved calls that are assumed to be species-specific [1][2][3] but that can sometimes show intraspecific variability since age, sex, and size can affect different call parameters such as frequency and amplitude [4][5][6][7][8][9][10][11]. The specific features of calls easily explain why sounds were successfully used to differentiate species of orthopterans [12], teleosts [13], birds [14,15], bats [16,17], and cetaceans [18][19][20]. Bioacoustics is also used as part of the diagnosis in bird species [14] where call features complement genetic and morphological data [21][22][23][24][25][26][27] simply because the use of acoustic signals is far easier to distinguish the species in the field. To the best of our knowledge, the consolidation of a fish species diagnosis with acoustic data has only been done in Encheliophis chardewalli (Carapidae) in which the comparison with other sympatric species of the same tribe supports a distinct repertoire [28]. Helvécio Lake (Parque Estadual do Rio Doce; 19˚46 0 29@ S, 42˚35 0 57@ W; Doce River Basin) where 150 specimens (SL: 178 ± 24 mm) were captured with gillnets. Specimens of yellow-bellied P. nattereri were from Araguari River (18˚39 0 37@ S, 48˚26 0 20@ W; Upper Paraná River Basin) where eleven specimens (SL: 328 ± 18 mm) were captured with gillnets and by hookand-line. There was no effect of the fishing method on the sounds recorded. There is no geographical natural connection between red-bellied and yellow-bellied P. nattereri. Pictures of a specimen from each geographical group can be found in S1 Fig

Recordings & sound-analysis
The sounds were recorded in a glass-tank (108 L), with a hydrophone (HTI-96-Min, High Tech Inc., USA; sensitivity: -164.4 dB re 1V μPa -1 ) placed in the middle of the aquarium and  connected to a TASCAM DR5 recorder (TEAC Corporation, USA). The fish were gently hand-held between the thumb and the index of the left hand without any pressure at approximatively five centimetres from the hydrophone in order to homogenize as much as possible the sounds recorded. During recordings in the University of Liège, the water temperature was 26 ± 1˚C and the filters, heater, bubblers and the light of the aquarium were turned off to reduce non-biological noises. For the specimens recorded in Brazil, the water temperature in the aquarium was 26 ± 2˚C. The resonance frequency was 3252 Hz (for details on recordings in tanks, see [50]). In total, 1713 sounds were recorded and analysed. For each vocal specimen, ten sounds were analysed, except for P. cariba and yellow-bellied P. nattereri from the Araguari River for which 26 and 107 sounds have been analysed, respectively. The sounds were digitised on mono-channel at 44.1 kHz with a 16 bit-resolution and then they were sub-sampled at 4000 Hz and a high-pass filter at 50 Hz was applied. The analysis was manually carried out with Avisoft SAS-Lab Pro 5.2 (Avisoft Bioacoustics, Germany).
The temporal features were measured on oscillograms (Fig 2A) while the fundamental frequency (F 0 , in Hz) was obtained from the power-spectra of the sounds (Fig 2B). The temporal features were the sound duration (d, in ms); the number of peaks in a sound (N), and the period between consecutive peaks (p, in ms). The period of the first peaks has been shown to be longer than that of the rest of the sound in some piranha species [10]. In consequence, the same three features were measured in the most energetic zone (abbreviated "ez"), i.e., the continuous area with an intensity of maximum 3 dB less than the maximum intensity peak (as the dB scale is a logarithmic scale, it corresponds to a decrease of a factor 1.4, see Fig 2A) and named d ez , N ez , and p ez .

Statistics
The description of the sounds was carried on the totality of the sounds. For each species, the different features were compared using Wilcoxon signed-rank tests and correlations between features were calculated using Bland-Altman repeated measures correlations.
The comparison between the different species required knowledge of the effect of the fish size because this could affect acoustic features [10]. For this reason, the effect of size was removed in order to conduct suitable comparisons. All the statistics relative to the effect of body size and the species comparison were realised on means per individuals, i.e., the ten sounds were averaged to obtain a single value per specimen. Correlation matrices were calculated using Spearman correlation coefficients (r S ) and their associated P-value matrices corrected by the Holm-Bonferroni method. When a correlation between an acoustic feature and the SL was found, a linear regression was performed. Logarithmic and polynomial regressions were also tested.
Wilcoxon-Mann-Whitney tests (also named Mann-Whitney U tests) and Kruskal-Wallis tests, with the Dunn's test as post-hoc with a Benjamini-Hochberg correction on the P, were used to compare acoustic features between species. The values of the statistics W and Z are presented with P and adjusted P when necessary. To reduce the multivariate aspect of the data, principal component analyses (PCAs) were conducted on the correlation matrices with all acoustic features except those whose R 2 � |0.60|, for which the features divided by the SL were used instead in order to remove the effect of size. In parallel, we carried out the same analyses on the residuals of the linear regressions. As the statistical significances and the biological conclusions with the two methods were identical, we present only the first method in this manuscript. We did not perform phylogenetic corrections because the number of species was small and it would have not been possible to take into account the difference between red-and yellow-belly P. nattereri. The principal components (PCs) are presented with their associated percentage of variance and their cumulative percentage of variance. PCs for each group were compared using Wilcoxon-Mann-Whitney and Kruskal-Wallis tests with the Dunn's test as post-hoc. All the statistics were carried-out using R 3.3.0. (GNU General Public License) and the significance level was α = 0.05.

Authorisations
The experiment was approved by the ethical commission of the University of Liège (case 1532) and the capture of the Brazilian specimens was achieved under the license 10306-1 from the Brazilian Ministry of the Environment. In the Paraopeba River, the fish were captured near the Retiro Baixo Hydropower Dam with an authorisation n˚PT-04/07/2018 issued by Retiro Baixo Energética. In the Parque Estadual do Rio Doce, the fish were sampled under the license 049/2018 from the Instituto Estadual de Florestas of the State of Minas Gerais. In the Araguari River, the fish were captured near the Capim Branco II Hydropower Dam with an authorisation from Consórcio Capim Branco. Six specimens have been euthanasied with an overdose of eugenol (CAS: 97-53-0) and imported to Belgium with the authorisation of the Federal Agency for the Safety of the Food Chain (COBT/IEC/CMY/1546595), the Federal University of Minas Gerais (MTA/TTM n˚02/2019) and the Conselho de Gestão do Patrimônio Genético of the Ministério do Meio Ambiente (n˚R2B8FF7 et n˚A57A7E9).

Description of the sounds
Sounds of P. cariba, two populations of P. nattereri (red-and yellow-bellied) and P. piraya were recorded in this study. Sonic muscles were similar in the three species. All specimens produced sounds except four specimens of P. piraya coming from the Aquário da Bacia do Rio São Francisco of Belo Horizonte, for which body size was between 390 and 430 mm.
Description of Pygocentrus cariba sounds. Pygocentrus cariba SLs were between 87 and 92 mm. Their sounds (n sounds = 26) were low-frequency harmonic drumming sounds and consisted of 16 ± 7 (mean ± standard deviation) continuous peaks with a peak period of 11.5 ± 0.5 ms (Fig 3A). Sounds lasted from 54 to 388 ms (183 ± 74 ms) with a fundamental frequency of 89 ± 6 Hz. The most energetic zone contained between 2 and 18 peaks (7 ± 5 peaks) that corresponded to 15 to 80% (45 ± 16%) of the total number of peaks. The peak period in the most energetic zone was equivalent to the mean peak period of the sound (Wilcoxon signed rank test: V = 208, P = .42). The sound duration was positively correlated with the number of peaks (r = 0.99, P < .0001) but not with the period (r = -0.22, P = .30). In other words, the number of peaks determines the sound duration. N EZ and d were correlated with d EZ (r = 0.94 and 0.78, both P < .0001).
Description of Pygocentrus piraya sounds. Pygocentrus piraya SLs were between 90 and 281 mm. Their sounds (n sounds = 80) were made of 9 ± 3 continuous peaks with a peak period of 15 ± 5 ms (Fig 3B). Sounds lasted from 71 to 266 ms (131 ± 50 ms) with a fundamental frequency of 79 ± 13 Hz. The most energetic zone of P. piraya sounds was much smaller than in P. cariba sounds (34 ± 30 ms vs. 94 ± 63 ms) with 4 ± 2 peaks and covered 38 ± 17% of the total number of peaks. The period in the energetic zone was 14 ± 5 ms. The number of peaks was independent of the duration of the energetic zone (r = 0.16, P = .18) and weakly correlated to the number of peaks in the energetic zone (r = 0.24, P = .04).
Description of Pygocentrus nattereri sounds. The specimens of yellow-bellied P. nattereri were much bigger than the specimens of red-bellied P. nattereri (Wilcoxon-Mann- Whitney: W = 0, P < .0001). Red-bellied P. nattereri sounds (N sounds = 1500) were made of 11 ± 3 continuous peaks with a peak period of 14 ± 2 ms. Sounds lasted from 24 to 417 ms (147 ± 43 ms) with a fundamental frequency of 79 ± 12 Hz. The most energetic zone contained between 1 and 23 peaks (4 ± 2 peaks) that corresponded to 42 ± 16 of the total number of peaks ( Fig 3C). Yellow-bellied P. nattereri sounds (N sounds = 107) were made of 6 ± 3 continuous peaks with a peak period of 31 ± 8 ms. Sounds lasted from 40 to 466 ms (194 ± 90 ms) with a fundamental frequency of 70 ± 9 Hz. The most energetic zone contained between 1 and 7 peaks (2 ± 1 peaks) that corresponded to 36 ± 22% of the total number of peaks (Fig 3D).

Effect of the fish size
For all species, all the frequency features and the peak periods were correlated to the SL (all P < .001). The p and p ez were highly positively correlated to the SL (r S = 0.64 & 0.73) while the fundamental frequency was highly negatively correlated to the SL (r S = -0.62). When p and p ez were predicted, it was found that the SL was a significant predictor (both P < .001, both R 2 = 0.68). To remove the effect of the size, the features divided by the SL (written "X SL -1 " for the feature "X") were used instead of the features themselves. In fish, some sound features, such as pulse duration or fundamental frequency, are known to be affected by their body size [8,9,[51][52][53][54]. We combined our results with data also recorded in glass tanks from [55] and [10] to study the frequency as a function of size with 182 different specimens of red-bellied P. nattereri ranging from 44 to 270 mm (Fig 4). Besides linear regression, other models such as logarithmic models (Table 2) could explain the relation between the size and the fundamental frequency, with larger specimens having a smaller fundamental frequency.  (Table 3 and Fig 5). The comparison between the studied specimens of red-bellied and yellow-bellied P. nattereri showed that their sounds differ on all the features (Wilcoxon-Mann-Whitney: W = 401, 1328, 1246, 1486, 1534, 0, 6; P = .004, < .001, .005, < .001, < .001, < .001 & < .001; Fig 6) except d ez (Wilcoxon-Mann-Whitney: W = 962, P = .36). Similar results were obtained after dividing by the SL. No difference was found between the acoustic features of yellow-bellied P. nattereri and P. piraya.

Discussion
This study described for the first time sounds of hand-held P. cariba and P. piraya. These results are important for future passive acoustics monitoring (PAM) of those species. However, caution need to be taken when compared fish sounds recorded in tanks (see [50] for details). Another important consideration is the fish size effect. Non-linear regression models has also been suggested in Balistidae [8,56] but with fewer specimens. In red-bellied P. nattereri, old adult specimens would be mute [55] because sonic muscles have been invaded by fat cells and there are several modifications in the ultrastructure of the myofibrils and

PLOS ONE
Use of bioacoustics in species identification: Piranhas as a case study sarcoplasmic reticulum [55]. In the present study, the four biggest specimens of P. piraya did not produce any sounds, supporting the hypothesis that older piranha specimens lose their sound-producing ability. Deep analysis of sonic muscles from these specimens did not take place because these fish were from a public aquarium and were returned after this experiment. The challenge of this study was to discriminate three piranha species using their sound features. The easiest to discriminate was P. cariba which differs from all the others in d ez , N, and N ez . The greater differences in acoustic features between P. cariba and congeners could be related to their higher genetic distance [57,58]. Pygocentrus cariba and the potential ancestor of both the other sister species (P. nattereri and P. piraya) would have separated around 8 Ma, whereas Pygocentrus nattereri separated from P. piraya only more recently [48,57], at 2.63 ± 0.2 Ma [57][58][59]. Interestingly, the studied specimens of yellow-bellied P. nattereri and P. piraya cannot be distinguished using acoustic features, whereas the studied specimens of yellow-bellied and red-bellied P. nattereri can. This observation does not correspond to the theory that sounds are species-specific [60][61][62][63]. This paradox could be solved with two complementary hypotheses: yellow-bellied and red-bellied P. nattereri are two different species and yellow-bellied P. nattereri have been wrongly identified, being in fact P. piraya.
The current yellow-bellied P. nattereri was first described as Serrasalmo ternetzi by Steindachner in 1908 [64] before being considered the holotype of the Gastropristis genus by Eigenmann 1915 [65,66]. The main distinctive feature was the morphology of the anal fin. Unfortunately, the holotype was mutilated and is lost to science [42]. This yellow-bellied population of piranha was then described as Serrasalmus ternetzi [67] based on three specimens that differ from P. nattereri at the level of the anterior profile, the length of the dorsal-fin base and the distance of dorsal fin from adipose fin [42,67]. Subsequent studies on a higher number of individuals confirmed the status of S. ternetzi. The diagnosis used the yellow coloration (on about 100 specimens), the convexity of the head and other meristic data in thirteen specimens [42,68]. However, using the body shape and head convexity of 60 specimens, Fink (1993) considered S. ternetzi as a junior synonym of P. nattereri [42]. More recently, a study observed differences in meristic data from north and south populations of P. nattereri: yellow-bellied P. nattereri possessed a deeper body relative to body length than red-bellied P. nattereri [43]. They considered this species as a nonlinear cline [43] and like in our study, their specimens of yellow-bellied P. nattereri were again bigger than those of red-bellied P. nattereri. Molecular studies show that specimens from different geographical areas (the red-bellied from Amazonas and yellow-bellied from Paraná-Paraguay) have exclusive haplotypes, supporting that a split happened at around 1.8 Ma [48, 57,69]. Morphological, meristic, colour pattern, and molecular data all support deep differences between yellow-bellied and red-bellied P. nattereri. On the studied specimens, acoustic features clearly reinforce this assertion.
According to Fink (1993), "Pygocentrus piraya is diagnosed by the presence of adipose fin rays and is restricted to the Rio São Francisco of Brazil" whereas "P. nattereri is undiagnosed and highly variable in pigmentation" [42]. Fink (1993) reported that P. nattereri can be distinguished from P. cariba by the lack of a black humeral spot and from P. piraya by a lack of adipose fin rays [42]. We did not observe black humeral spots but we observed adipose fin rays in the so-called yellow-bellied P. nattereri (S1 Fig) sampled in the Araguari River (Paraná River basin). However, Pygocentrus is not native to this location [70,71] and was recorded for the first time in 2009 (S2 Fig; Godinho, personal data) [72][73][74] and identified as yellow-bellied P. nattereri by Langeani & Lacerda-Rêgo (2014) [71]. According to the characters they use in their study, it is however not possible to precisely identify the P. nattereri species but only the Pygocentrus genus. According to our results, having a population of yellow-bellied P. nattereri in Araguari River is deeply questionable and could result from misidentification. Based on historical morphological description and on the main argument that they both have the same acoustic features, the so-called yellow-bellied P. nattereri of the Araguari River should be P. piraya. Yearly records show a progressive migration from East to West (S2 Fig), suggesting fish were introduced to the east part of Araguari River [49] (S2 Fig; Godinho, personal data). This greatly supports our sonic-based assertion: P. piraya are now found in the Paraná River basin. Our study supports that acoustic features can be important to reveal misidentified fish species. Up to now, this tool has been only used in other zoological groups. The distinction between species appears to be possible because acoustic signalling may be more flexible than morphology [17]. In birds, sounds were used for the distinction of three species of Phylloscopidae 30 years before their official description as different species [75]. Songs were also used to clarify systematic relationships in Passeriformes [76] and to contribute to the description of new species in Rhinocryptidae [77] and Tyrannidae [78]. Insects (Orthoptera, Hemiptera, and Neuroptera) [79][80][81] and anurans (Leptodactylidae, Ceratobatrachidae, Ranidae) [82][83][84][85][86][87][88] also possess many cryptic species that can be mainly differentiated acoustically. This role of bioacoustics in taxonomy supports the application of PAM in conservation programmes. However, more studies on frequency discrimination by fish are still necessary to know if acoustic differences are distinguishable by the fish.

Conclusions
Calls of P. cariba (Orinoco River basin) were distinct from all the other, red-and yellow-bellied P. nattereri (Parque Estadual do Rio Doce and Araguari River respectively) calls were different, and yellow-bellied P. nattereri calls were similar to those of P. piraya (São Francisco River basin). As the yellow-bellied P. nattereri found in the Araguari River basin possess adipose fin rays and the same acoustic features as Pygocentrus piraya, we have to consider that the population from Araguari River is most probably Pygocentrus piraya. This means that the geographical area of Pygocentrus piraya is larger than expected; and more likely, P. piraya has been introduced to the Paraná River basin. The use of acoustic features was made possible due to the removal of the effect of the fish size.