Diversity and evolution of sound production in the social behavior of Chaetodon butterflyfishes

Many fish produce context-specific sounds during social interactions that are used in recognition of individuals, sex or species, during the defense of food resources or mates, and courtship or spawning (Ladich and Myrberg, 2006; Myrberg and Lugli, 2006). Fish sounds are produced by a diversity of motor behaviors (Ladich and Fine, 2006) but all produce an acoustic field that includes both a localized hydrodynamic flow that may stimulate the lateral line and inner ear at close distances, and a sound pressure wave that may independently stimulate the inner ear at larger distances from the source (Kalmijn, 1988; Coombs and Montgomery, 1999). Studies on the evolution of fish sounds used for intraspecific communication are based largely on meta-analyses of sound production mechanisms compared in broad taxa (e.g. Ladich and Bass, 2003; Ladich and Fine, 2006; Parmentier and Diogo, 2006) and the characteristics of sounds produced by different genera within fish families (Amorim, 2006). A few studies on sympatric congener species show evidence for conservation or divergence in temporal and spectral features of sounds: African cichlids (Amorim et al., 2004,, 2008; Lobel, 1998), sunfishes (Gerald, 1971), gobies (Lugli et al., 1997), mormyrids (Crawford et al., 1997) and marine damselfishes (Kenyon, 1994; Lobel and Mann, 1995; Myrberg and Riggio, 1985; Myrberg et al., 1993; Spanier, 1979). However, the diversity of natural sounds remains to be assessed in relation to social behavior and hearing capabilities in a phylogenetic evolutionary context.

Butterflyfishes of the family Chaetodontidae include approximately 130 species and 10 genera that are abundant members of coral reef fish communities and well known for their diverse color patterns, social behaviors and food habits (Hourigan, 1989). Most are highly social species, which form long-term monogamous pairs or harems that defend food and other resources from competitors, or larger associations for planktivorous feeding or movements across the reef. Phylogenetic studies (Blum, 1988; Smith et al., 2003; Fessler and Westneat, 2007; Bellwood et al., 2010) show a well-defined phylogeny that consists of the bannerfish clade and the Prognathodes–Chaetodon clade (Fig. 1). The genus Chaetodon includes approximately 90 species distributed within four clades with 10 subgenera (Bellwood et al., 2010) all of which have a diagnostic laterophysic connection (LC) – a novel morphological association of the swim bladder, anterior swim bladder horns and the cranial lateral line canal system (Webb, 1998). Several Chaetodon subgenera show distinct differences in the length, width and proximity of the paired anterior swim bladder horns to a medial opening in the lateral line canals in the supracleithral bone of the head and in the morphology of the swim bladder (Webb and Smith, 2000; Webb et al., 2006, 2010). The LC system was proposed to enhance hearing capabilities during social interactions via the transduction of sound pressure waves by the anterior swim bladder horns that may stimulate fluid motion in the lateral line canal and particle motion at the inner ear (Webb, 1998; Webb et al., 2006; Tricas et al., 2006). However, the acoustic behaviors and sound characteristics used in communication by Chaetodon remain unknown for all but one species, and are needed to interpret the putative function and evolution of the LC within the family.

Recent studies show that sound production during social interactions occurs widely among butterflyfish genera (Fig. 1) but details on the mechanics of sound production for Chaetodon species come only from field observations. Acoustic behaviors are known but not quantified for C. multicinctus, C. ornatissimus, C. kleinii (Tricas and Boyle, 2014) and C. ulietensis (Lobel et al., 2010) with kinematic data available for only C. multicinctus (Tricas et al., 2006). Lab and field studies on bannerfishes (all of which lack the LC) reveal two primary acoustic behaviors. The head bob sound is produced by Forcipiger species and is driven by the epaxial muscles (Boyle, 2011; Boyle and Tricas, 2011; Tricas and Boyle, 2014). The swim bladder buckling sound is produced by Hemitaurichthys species, which is driven by contraction of anterior hypaxial muscles (Boyle and Tricas, 2010; Boyle et al., 2013; Tricas and Boyle, 2014). A similar internal sound production mechanism occurs in Heniochus chrysostomus that is associated with contraction of obliquus superioris hypaxial muscles (Parmentier et al., 2011). These studies show that sound production involves distinct acoustic behaviors, includes different muscle systems and likely preceded the evolution of the swim bladder horns and LC character in the Chaetodon genus. However, sound production behaviors and associated sound characteristics must be identified for Chaetodon to estimate their ancestral character states and evolution. In addition, the potential adaptive value of acoustic signal communication in Chaetodon, such as providing honest signals about the size, condition or health of the individual as shown for Forcipiger flavissimus (Boyle and Tricas, 2011) remains unexplored.

In this study, we determine the acoustic behavior of different Chaetodon subgenera and identify basic evolutionary patterns of sound production within the family. We compare sound production in five species with a range of LC morphologies to that of the related forcepsfish, F. flavissimus, which lacks the LC. Herein, we (1) assess the external motor patterns that are associated with sound production; (2) quantify the sound characteristics among species; (3) test the hypothesis of honest acoustic signals related to sex and body size in Chaetodon acoustic communication; and (4) assess the ancestral character states and independent contrasts of sound characteristics within the family based on their published molecular phylogeny. These findings are further interpreted in relation to the auditory sensitivity of these species in a companion paper (Tricas and Boyle, 2015).

A total of 41 butterflyfish were tested for sound production after the introduction of a conspecific individual to the test aquarium. Sound production was recorded for 30 subjects: adult Chaetodon multicinctus (n=10, 56% of tested fish, SL=71.2±14.2 mm, mean±s.d.), Chaetodon auriga (n=5, 100%, SL=111.0±7.0 s.d. mm) including one immature individual, Chaetodon kleinii (n=4, 100%, SL=78.0±4.6 mm), Chaetodon ornatissimus (n=3, 60%, SL=78.7+5.9 s.d. mm), Chaetodon unimaculatus (n=4, 100%, SL=122.0+29.2 mm) and F. flavissimus (n=4, 80%, SL=99.7+12.1 mm). In total, 59% of resident fish and 16% of introduced fish produced sound. The majority (79%) of total sounds were produced by the resident fish as it approached in close proximity to the conspecific usually within 1–2 body lengths. Some body movements associated with sound production were not visible because of the orientation of the fish, the camera angle or blocked view by the body of the second fish.

Two sound types were recorded for F. flavissimus during social interactions with conspecifics. The head bob pulse sound occurred during a very brief dorsal–ventral motion of the head that was sometimes only slightly visible in the video and produced most frequently by the introduced non-resident fish (Table 1). This sound was of short average duration (29 ms) with an average peak frequency of 254 Hz (Table 2, Fig. 2). The head bob behavior sometimes included a slight movement of the jaw that was neither prominent nor frequent in the video. Thus, we assume jaw movement in this species was a passive action associated with the upward motion of the head and not involved in sound production. All resident fish produced the anal fin retraction pulse sound during aggressive interactions with the introduced individual. This sound occurred after extension of the anal fin spine but was coincident with the retraction motion. It was of longer average duration (54 ms) and much lower average peak frequency (27 Hz) than the head bob pulse.

Jaw protrusion was the only sound type recorded for C. kleinii. This acoustic behavior was shown by both resident and introduced fish and most commonly included a clear extension and retraction of the oral jaws and occasionally a brief shake (lateral motion) of the head (Table 1). This sound was of very short average duration (17 ms) and showed an average peak frequency (516 Hz) (Table 2, Fig. 2). We observed no vertical motion of the head, although it may be detected at higher video frame rates.

Two sound types were produced by C. unimaculatus. Both resident and introduced fish produced the head bob–jaw protrusion pulse sound that was a very fast motion and visible in only 1–2 frames of video. This behavior was always associated with a clear extension and retraction movement of the jaws, a frequent vertical head bob motion and an infrequent lateral head shake as for C. kleinii above (Table 1). It appeared to be of longer average duration (52 ms) and peak frequency (1031 Hz) than that for the jaw protrusion sound of C. kleinii, although our sample size was too small for statistical comparisons (Table 2, Fig. 2). The tail slap sound was recorded for one resident C. unimaculatus and involved motion of the whole body directed at the nearby conspecific. This sound waveform was of long duration (424 ms) and very low peak frequency (<1 Hz) as reported for tail slaps by other Chaetodon species described below (Table 2).

Four pulse sounds were produced by C. multicinctus during social interactions with conspecifics. The body pulse sound was produced most frequently by the resident fish in association with lateral motions of the caudal peduncle, lateral body shake, flick of the pelvic fin or rarely without any observed body motion (Table 1). This acoustic behavior was often preceded by erection of the dorsal and anal fins, which did not appear to be associated with sound production. The body sound occurred in trains or single pulses that had a short average duration (25 ms), covered a wide range of peak frequencies (range, <1465 Hz; mean, 137 Hz; median, 231 Hz), and showed a wide average bandwidth of 350 Hz (Table 2, Fig. 2). A single resident fish produced a novel body shake sound with a peak frequency of 29 Hz that involved brief lateral oscillations of the body in close juxtaposition to the introduced fish. The tail slap sound was produced primarily by resident fish, of relative long duration (418 ms), and very low average peak frequency (9 Hz) and bandwidth. Although movement was not quantified, the displacement motion of the tail slap ranged from weak to strong and sometimes displaced the body of the receiver fish by the generated hydrodynamic flow. We also recorded the previously reported ‘click’ sound pulse that occurs after the tail slap (Tricas et al., 2006), but these were relatively infrequent in the lab setting (Tables 1 and 2, Fig. 3). This distinct broadband sound was of very short average duration (16 ms) and much higher peak frequency (2752 Hz) than any other sound recorded.

Two sounds were produced infrequently by resident C. ornatissimus, which was the most timid species in the laboratory setting. The body pulse sound occurred primarily during movements of the caudal peduncle and/or extension of the pelvic fins (Table 1). This was similar to the body sound motion of C. multicinctus, but lacked movements of several additional body parts. It had an average duration of 134 ms and a very low average peak frequency of 10 Hz (Table 2, Fig. 2). Resident fish produced both weak and strong tail slap behaviors (as described above) which were of long average duration (366 ms) and low average peak frequency (6 Hz).

Only the head bob–jaw protrusion pulse was produced by C. auriga. This sound was associated primarily with a clear vertical elevation of the head and a strong protrusion of the jaws. The head bob motion was very similar to that observed for F. flavissimus (which did not show active jaw protrusion) and the jaw motion was very similar to that observed in C. kleinii (which did not show evidence of a vertical ‘head bob’ motion). It was produced frequently by the resident fish (Table 1) and occurred in single pulses of relatively long duration (93 ms) and low peak frequency (23 Hz) (Table 2, Fig. 2), and less frequently in a two-pulse train (Table 3).

The acoustic stimuli produced by Chaetodon species span at least four frequency decades (from <1 to >1000 Hz) and show a strong and non-linear inverse relationship between average sound duration and peak frequency (Fig. 4). Extreme low-frequency sounds with an average peak frequency of <10 Hz and long duration (400 ms) were associated with the tail slap behavior in C. ornatissimus, C. multicinctus and C. unimaculatus. The tail slap sound did not differ among C. multicinctus and C. ornatissimus in either median duration (Mann–Whitney U-test, U=0.8, n=7 and 3, P=0.67), average peak frequency (t-test, d.f.=8, t=0.834, P=0.43), average 6 dB median frequency (t-test, d.f.=8, t=0.467, P=0.467) or average 6 dB bandwidth (t-test, d.f.=8, t=0.171, P=0.87). Other pulsed low-frequency sounds that had peak frequencies from ∼10 to 30 Hz and durations near or <100 ms were produced by C. multicinctus, C. ornatissimus, C. auriga and F. flavissimus. Sounds with peak frequencies from 100–1000 Hz were readily produced in social interactions in the lab by C. kleinii, C. unimaculatus, C. multicinctus and F. flavissimus, with pulse durations ≤50 ms. Analysis of variance identified no differences in peak or median frequency, or bandwidth among these sounds. The unique click sound of C. multicinctus showed a much higher average peak frequency of 2732 Hz. These data show that the acoustic fields produced by fish at close distances span a wide frequency range and that longer pulses carry information at lower frequencies.

There were several similarities and differences in the bandwidth of sounds produced among species (Fig. 5). With the exception of C. kleinii, all Chaetodon species produced at least one sound type that had a peak frequency or bandwidth that fell within the <1–30 Hz and the 10–100 Hz ranges. The lowest frequency sounds included the tail slap behaviors produced by C. unimaculatus, C. ornatissimus and C. multicinctus. The 10–100 Hz band included several sound types produced among species, and often extended to higher frequencies. Sounds within the >100 Hz band included different sound types produced by F. flavissimus (head bob), C. kleinii (jaw protrusion), C. unimaculatus (head bob–jaw protrusion) and C. multicinctus (body motion and click), but sounds in this band were not recorded for C. auriga or C. ornatissimus (which is known to produce sounds in this band in the wild, see Discussion). These comparisons show that the acoustic repertoire for some species covers a wide bandwidth through the production of several sound types (as in C. unimaculatus and C. multicinctus) or may have a more limited vocalization bandwidth.

Several pulsed sounds were produced by F. flavissimus and Chaetodon species in trains that included 2–12 discrete pulses separated by <1 s (Fig. 6, Table 3). The most common pulsed trains occurred in the body and tail slap sounds of C. multicinctus. The average rate of pulse train production in Chaetodon ranged from a low of 2 Hz for the tail slap sound of C. ornatissimus to a high average of 6 Hz for the body pulse by C. multicinctus, but both trains were highly variable in pulse number (Table 3). The sequential production of multiple head bobs was shown by both F. flavissimus and C. auriga and the sequential production of multiple tail slap sounds by both C. multicinctus and C. ornatissimus.

The average pressure levels of sounds produced by individual subjects (as measured within a few cm of the source) ranged from a low of 107 dB re. 1 µPa for the anal fin retract pulse of F. flavissimus to a maximum of 134 dB re. 1 µPa for the tail click pulse by C. multicinctus (Table 2). Pressure levels for the tail slap sound were of the strongest magnitude recorded and did not differ among C. ornatissimus (131 dB re. 1 µPa) and C. multicinctus (131 dB re. 1 µPa, t-test, t=−0.13, d.f.=8, P=0.90,). However, within species we confirm that the intensity of the tail slap sound (131 dB re. 1 µPa) was greater than the body motion pulse sound (120 dB re. 1 µPa) produced by individual C. multicinctus subjects (paired t-test, t=−4.15, d.f.=5, P<0.01), which collectively span a broad frequency spectrum (Fig. 7A).

The phylogenetic tree map of butterflyfish acoustic behaviors from this and previous studies (Boyle and Tricas, 2010,, 2011; Parmentier et al., 2011) indicates the possibility for both unique and shared sound production traits among the bannerfishes and Chaetodon species examined thus far (Fig. 8). The head bob occurs in Chaetodon clades 2 and 4 (but not clade 3) and is similar to that for F. flavissimus in which the epaxial muscles elevate the cranium. In contrast, the sound production mechanisms in Heniochus and Hemitaurichthys species are driven by mechanisms that involve the hypaxial muscles. Preliminary EMG recordings during production of the similar body pulse sound in C. multicinctus (our unpublished data) tentatively show activity in the ventral anterior epaxial musculature lateral to the anterior swim bladder (note that the bilateral swim bladder horn extensions project dorsal to the horizontal septum). In comparison, the tail slap behavior occurs in 4 of 5 Chaetodon species, but not in the bannerfishes studied so far.

We estimated the ancestral character states for the head bob and tail slap behavior using the maximum-likelihood and parsimony methods (Fig. 9). The anal fin and body motion behaviors, and jaw protrusion in C. kleinii currently appear to be autapomorphies. The jaw protrusion in other species appears to be a by-product of the head bob. Maximum-likelihood analysis shows that the ancestral state for the head bob is undetermined, with equivocal likelihoods at all ancestral nodes, whereas the maximum parsimony method indicates independent evolution of this trait among species (Fig. 9A). The observation of the tail slap behavior only in Chaeotodon showed similar ancestral state patterns for both the likelihood and parsimony methods with high probability at all clade nodes (Fig. 9B). The lack of the observed tail slap behavior in C. kleinii indicates the potential loss of this character. In addition, it is equivocal as to whether the common ancestor of the bannerfish and Chaetodon clades showed tail slap behavior.

Independent contrast analyses of the continuous variable sound characteristics (duration, peak frequency, median frequency, bandwidth and intensity) for the tail slap or most prominent sound type observed for each species show evidence for correlated changes in sound characteristics (Table 4). We found no significant relationship in the slope of the independent contrasts and corrected branch length for any character. Further analysis showed a correlated change between sound duration and sound pressure characteristics (R=0.86, P=0.03) and the expected strong correlation between peak and median frequency (R=0.98, P<0.001) characters. We did not include continuous sound characteristic data available for other butterflyfish species beyond this study because of differences in the quantitative measures that may be recorded in different test aquaria.

Analyses of the acoustic characteristics for male and female C. multicinctus of different sizes show several trends in their body motion and tail slap pulse sounds. The intensity of both sound pulses was correlated with body size (R=0.73, P<0.003 and R=0.71, P<0.001, respectively) (Fig. 7B). Both sexes produced these two sound types in very similar proportions during the test trial periods (body pulses: males, 68.4% vs females, 68.1% of sounds; tail slaps: males, 30.1% vs females, 26.8% of sounds) although there was considerable variation of total number of sounds produced among individual fish (males, 1–34 total sounds per trial; females, 9–20 total sounds per trial). For the body pulse sound, we found no difference between sexes in average pulse duration (Mann–Whitney U-test, U=2.0, n=3 and 5, P=0.14), peak frequency (t-test, t=0.67, d.f.=6, P=0.53), median frequency (t-test, t=1.02, d.f.=6, P=0.35), or 6 dB bandwidth (t-test, t=0.60, d.f.=6, P=0.57), but sample size for tail slaps by females was too small for statistical comparison (Fig. 10).

Butterflyfishes are well known for their diverse feeding ecology and social behaviors (Hourigan, 1989; Roberts and Ormond, 1992) and all species used in this study spend large amounts of time in close social affiliations in the wild. In Hawaii, the forcepsfish F. flavissimus forms social harems (Boyle, 2011); the teardrop butterflyfish C. unimaculatus and threadfin butterflyfish C. auriga form social pairs and larger groups (Reese, 1975; Fricke, 1986; Hourigan, 1989); the multiband C. multicinctus and ornate C. ornatissimus butterflyfish are found in monogamous pairs (Reese, 1975; Tricas, 1989; Kosaki, 1999); and the blacklip butterflyfish C. kleinii forms plankton-feeding associations in the water column or social pairs on the reef (Hourigan, 1989). These close associations among individuals may facilitate the exchange of important acoustic information with mates, other conspecifics and heterospecifics in the noisy reef environment (Tricas and Boyle, 2015).

In this study, we consider sound to be the acoustic field produced by fish motor behaviors that forms both a local hydrodynamic flow field and a propagating sound pressure wave (Kalmijn, 1988). At distances close to the source, the hydrodynamic flow field can stimulate the inner ear of a receiver by whole body accelerations and at very close distances also the mechanosensory lateral line by the pressure gradient across the surface of the skin. In contrast, the sound pressure wave passes through the body but can cause compression of a gas-filled swim bladder (or other structure), which can impart local particle motions upon an adjacent or coupled inner ear. Thus features of the acoustic field can stimulate both the inner ear via hearing and the lateral line via mechanoreception by a variety of mechanisms (Coombs and Montgomery, 1999; Braun and Grande, 2008).

The genus Chaetodon is distinguished by the presence of paired anterior swim bladder horns that form the LC and project towards the inner ear (Webb and Smith, 2000; Webb et al., 2006). The LC is proposed to translate sound pressure fluctuations from the swim bladder horns to fluid motion in the lateral line canal and also secondary particle motions to the ear (Webb, 1998) that potentially enhance the perception of the acoustic field during social interactions (Tricas et al., 2006). The gas-filled swim bladder and swim bladder horns are thought to be relatively insensitive to infrasonic sound pressure and have complex resonant properties at higher frequencies between 100–2000 Hz that can vary considerably among species and with depth (Fine et al., 2009; Lewis and Rogers, 1996; Sand and Hawkins, 1973). Below, we discuss the detection of acoustic fields by the ear and lateral line in the context of butterflyfish social interactions, and the potential contributions to each by the LC.

Butterflyfish sound pulses cover a very broad frequency spectrum from <1 Hz to >1000 Hz that can directly stimulate the inner ear. With the exception of C. kleinii, each species produced at least one sound type that showed a very low peak frequency and bandwidth within the infrasound frequency range of <20 Hz (sensu Sand and Karlsen, 1986). The tail slap behaviors of C. unimaculatus, C. multicinctus and C. ornatissimus showed peak frequencies <1–9 Hz and were commonly produced as part of a visual lateral display behavior by resident fish. The visual features of their bright color patterns enhance the recognition of conspecifics (Boyle and Tricas, 2014). The lateral display is an agonistic visual stimulus commonly used during disputes by territorial Chaetodon species (Sutton, 1985; Yabuta, 2002; Zumpe, 1965) that precedes the tail slap behavior and is delivered at a distance of about 1–2 body lengths in the wild (Tricas et al., 2006). A tail slap at low intensity by C. multicinctus can impose weak hydrodynamic vortex rings to the side of the receiver's body (Hanke et al., 2008), whereas a high-intensity tail slap can displace the body of the receiver fish (Tricas et al., 2006). Low-frequency hydrodynamic motions can stimulate the otolith organs of the inner ear, which are highly sensitive to low-frequency linear accelerations of the body (Karlsen, 1992; Sand and Karlsen, 2000). The relative action of sound pressure on the swim bladder is thought to be negligible at these low frequencies (Sand and Hawkins, 1973). The potential importance of low-frequency sound during social interactions is further indicated by the body shake (C. multicinctus), body (C. ornatissimus), head-jaw (C. auriga) and anal fin (F. flavissimus) motions that are also associated with low-frequency stimuli. The detection of low-frequency hydrodynamic stimuli by the butterflyfish ear needs to be determined, but the sensitivity of the fish ear to acceleration at frequencies <100 Hz is probably equivalent to particle motion sensitivity at higher frequencies (Sand and Karlsen, 1986,, 2000). In addition, the biological relevance of these low-frequency stimuli needs to be tested by manipulation and playback experiments on behaving fish.

Several pulsed sounds in the 137–1031 Hz band were produced by F. flavissimus, C. kleinii, C. unimaculatus and C. multicinctus and add to a growing body of work on their acoustic repertoires. Sound with a peak frequency >100 Hz was not recorded from C. ornatissimus in this study, but does occur during social interaction (346 Hz) in the field (Tricas and Boyle, 2014). Thus, with the exception of C. auriga, all species in this study are now known to produce pulsed sounds within the 100–1000 Hz band. Many sounds share temporal and spectral characteristics with single ‘pop’ and pop train sounds produced by damselfishes (Amorim, 2006; Lobel et al., 2010; Maruska et al., 2007) and other reef species (Lobel et al., 2010), and are all within the hearing range of the inner ear (Ladich and Fay, 2013). Recent manipulation experiments using the auditory evoked potential technique show that Chaetodon has enhanced sound pressure sensitivity in the 100–700 Hz frequency band that is mediated by the swim bladder horns associated with the LC (Tricas and Boyle, 2015). Thus, the particle acceleration component of the acoustic field may directly stimulate the ear, whereas the swim bladder horns add additional sensitivity, likely by the transduction of sound pressure that secondarily stimulates the inner ear as in cichlids (Schulz-Mirbach et al., 2012) and other species that lack an otophysic connection (Ladich and Fay, 2013).

Rapid lateral motions of a strong tail slap presumably creates a dipole or higher order flow field in the horizontal plane. The steep pressure gradient across the skin of the receiver will quickly dissipate beyond distances of a few body lengths (Coombs and Montgomery, 1999). Weaker Chaetodon tail slaps can create a fluid vortex directed towards the skin of the receiver (Hanke et al., 2008). The motion of these flow fields across the skin can provide somatotopic information about the intensity and frequency components of water velocity by stimulation of superficial neuromasts (Weeg and Bass, 2002) and water acceleration by stimulation of canal neuromasts (Kroese and Schellart, 1992). Furthermore, the acceleration detected by canal neuromasts can provide information about the direction and distance from the source as described for the localization of prey (Coombs et al., 1996). During close social interactions between only two butterflyfish, their locations are evident by visually conspicuous markings in clear coral reef waters. However, in territory border conflicts among pairs (Tricas, 1989; Tricas et al., 2006) discrimination of a sexually monomorphic rival may be facilitated by hydrodynamic information to the lateral line that is integrated with auditory (and visual) input to provide directional information (Braun et al., 2002; Coffin et al., 2014) that can be potentially used to discriminate friend from foe. In addition, the lateral line may be activated by sound pressure stimuli that are transduced by the LC. Thus both a segment of the lateral line and the inner ear may receive coincident sound pressure information for processing, but this hypothesis requires testing. Future experiments such as the pharmacological or physical ablation of neuromasts are needed to demonstrate the relative contribution of the lateral line and LC system to perception of the acoustic field.

Motion analyses in this study show that several motor patterns are shared among species with divergent phylogenetic histories (Fig. 8). Most notable is the head bob motion in F. flavissimus and species in two Chaetodon clades. The head bob motion in F. flavissimus is driven by the action of the epaxial muscles on a ventral linkage between the head and pectoral girdle, which is maintained by simultaneous activity of the adductor mandibulae and sternohyoideus muscles (Boyle and Tricas, 2011). This results in the anterior motion of the pectoral girdle, ribs and rostral swim bladder before the head is released and rotated dorsally. Our study shows that the head bob motion for C. unimaculatus (clade 2) is also rapid and includes a prominent and active protrusion of the jaws that appears independent from the upward motion of the head. In contrast, only the prominent jaw protrusion and not the head bob is seen in C. kleinii (clade 2) and may be an autapomorphy (further studies can be done to test this), whereas jaw protrusion in the other fish appears to be associated with, or a by-product of, the head bob. The apparent diversity in head bob actions among Chaeotodon species leaves the question of the ancestral character state across the phylogeny undetermined (Fig. 9A). The head bob behavior is apparently lacking in C. multicinctus (also clade 2) and C. ornatissimus (clade 3). Preliminary EMG experiments on C. multicinctus demonstrate activity of muscles in the ventral region of anterior epaxial muscles lateral to the anterior swim bladder horns during the production of the body motion sound (our unpublished data), which tentatively appears most similar to that observed for the nearby region of the hypaxial muscles in both Hemitaurichthys (Boyle and Tricas, 2010) and Heniochus species (Parmentier et al., 2011). Further kinematic, anatomical and electromyography analyses are needed to test the hypotheses that the head bob sound is a homologous trait nested deep within the butterflyfish phylogeny, was lost at least once in the bannerfishes and twice in Chaetodon, or has evolved in parallel. Furthermore, it remains to be determined whether these acoustic behavior characters exist for Prognathodes, the sister genus to Chaetodon, the remaining bannerfish genera and their putative outgroups such as the pomacanthid angelfishes and ephippid spadefishes (Fig. 1).

The most distinctive acoustic behavior in Chaetodon was the tail slap, which is not yet reported in the bannerfishes. Reconstruction of ancestral states show this character as prominent at all Chaetodon nodes (Fig. 9B). However, we emphasize that laboratory tests are of limited utility because the tail slap behavior was not observed for C. auriga, but is seen in the field (Tricas and Boyle, personal observations). Similarly, the tail slap behavior was not observed for C. kleinii in the lab but may occur in the wild. Thus, both field and lab studies are needed to better define the evolution of sound production.

Patterns are emerging to indicate that acoustic signals used during social interactions are adaptive in butterflyfish communication. Our preliminary independent contrast analyses show evidence for correlated changes in sound duration and sound pressure characteristics for social signals. Honest signals provide accurate information about the condition of the signaler (Fitch and Hauser, 2002) and are often linked to physical attributes, such as body size or mass. In this study, we have shown that the intensity of both the tail slap and body motion sounds produced by C. multicinctus increases with body size and that both sound types are shared among sexes. Feeding territory size in wild C. multicinctus pairs is highly correlated with body size (Tricas, 1989). Defense of feeding territories towards unfamiliar non-mate adults or juveniles is known for many Chaetodon species (Hourigan, 1989; Roberts and Ormond, 1992; Tricas, 1989; Tricas et al., 2006) and we expect that sound production may be used as an indicator of condition or resource holding potential by many Chaetodon species as shown for F. flavissimus (Boyle and Tricas, 2011) and other fishes (Amorim et al., 2010; Ladich, 1998). The strong visual and acoustic displays during territorial contests may provide useful information about the relative size of an opponent, reduce the likelihood of escalated combat and lower the risk of injury. The acoustic stimuli provided during close social interactions in Chaetodon may provide valuable information that reinforces other visual and chemical signals or cues related to mate or non-mate recognition (Reese, 1975; Boyle and Tricas, 2014; Hanke et al., 2008) in their noisy coral reef environment (Tricas and Boyle, 2015).

Lab experiments were performed to characterize sound production in five Hawaii butterflyfish species that are members of three of the four Chaetodon clades based on their molecular phylogeny (Fessler and Westneat, 2007; Bellwood et al., 2010) and previously placed in four subgenera each of which has different LC variants (Smith et al., 2003; Webb et al., 2006). Species in Chaetodon clade 1 (Chaetodon hoeferi, Chaetodon robustus and Chaetodon marleyi) were not tested. Three test species from clade 2 are the blacklip butterflyfish C. kleinii and teardrop butterflyfish C. unimaculatus [subgenus Lepidochaetodon; indirect LC with mucoid connective tissue, long horns (LC variant Ind1)] and the multiband butterflyfish, C. multicinctus [subgenus Exornator; indirect LC without mucoid connective tissue, long horns (LC variant Ind2)]. The ornate butterflyfish C. ornatissimus is a member of clade 3 [subgenus Citharoedus; indirect LC without mucoid connective tissue, short stubby horns (LC variant Ind3)]. The threadfin butterflyfish C. auriga is a member of Chaetodon clade 4 [subgenus Rabdophorous; direct LC with mucoid connective tissue, long horns (LC variant Dir1)]. We also tested for the acoustic behavior in the forcepsfish F. flavissimus, which is a member of the bannerfish group that possesses no swim bladder horns or LC. Details of sound production by F. flavissimus and Forcipiger longirostris were reported in a previous study (Boyle and Tricas, 2011).

Experimental fish were obtained from commercial collectors, transported to the lab and held for 1–3 days in holding tanks or aquaria (80–125 liter) equipped with fresh flow-through sea water from Kaneohe Bay (annual range, 25–28°C). Fish were supplied with colonies of live corals or sessile invertebrates for food. Adult fish were used in most trials (see Results, Table 4). Most subjects were not killed to determine sex, with the exception of C. multicinctus, which was tested for the effects of body size and sex on the acoustic features of different sounds. Following these experiments, individuals were measured for standard length and sexed by either gonad catheterization or after euthanization. All experimental procedures were approved by the University of Hawaii Institutional Animal Care and Use Committee.

Acoustic behavior experiments were conducted in a flow-through sea water 125 liter aquarium with interior dimensions of 77.5×44.5×35.5 cm, L×W×H (water depth). Water temperature across the duration of the study ranged from 25 to 28°C. A resident test fish was acclimated for 3–24 h in the aquarium that contained at least one live coral head. Sound production was evoked by the introduction of a second conspecific stimulus fish enclosed in a 20×20×20 cm wire frame covered with fine 2.5 cm mesh monofilament netting that allowed visual contact between fish and sound transparency. Video of body movements during sound production events was recorded on a Sony video camera (TRV-950) at 30 f s⁻¹ through a glass window on the tank side. Sounds were recorded with Brüel and Kjaer 8103 (−211.8 dB re. 1 V μPa⁻¹; Naerum, Denmark) and 8104 (−205.4 dB re. 1 V μPa⁻¹) hydrophones connected to a Nexus 2692 conditioning amplifier at 60 dB gain (Naerum, Denmark) and positioned approximately 3 cm from the face of the cage with the stimulus fish. The 0.1–20 kHz frequency response of the hydrophones includes part of the infrasound spectrum (<1–20 Hz). Hydrophone signals were digitized at 48,000 samples s⁻¹ on one audio track of the camera for reference on the video (16 bit) and on one channel of a Sony DAT recorder (PCMM-1, 16 bit, 20 Hz–22 kHz playback response) for waveform analysis. Strong frequency components <20 Hz occurred in several sound waveforms, thus are likely underrepresented in our subsequent analyses because of the low-frequency filter of the DAT recorder. Resident fish would most commonly approach the exposed side of the cage to produce sound, thus intensities for resident and stimulus fish were recorded at approximately 1–10 cm of the source. Water flow was stopped during acoustic recordings to minimize background noise. Trials were typically of 15 min duration but sometimes were extended up to 30 min to obtain representative sounds. Two trials were run for each resident fish although up to eight trials were required to evoke sound production from some individuals.

Digital video and sound recordings were transferred to a PC and stored as uncompressed AVI and WAV files, respectively. Sound files were filtered (60 Hz notch) in Cool Edit Pro 2.1 software with a 100 dB attenuation and ‘super narrow’ setting. Most sound files were low-pass filtered at 2 kHz and downsampled at 4 kHz with Cool Edit Pro 2.1 software set at the ‘high-quality’ setting. The resulting bandwidth (0–2 kHz) was below the minimum resonance frequency of 2871 Hz estimated for the 35.5 cm water depth in the aquarium (Akamatsu et al., 2002), thus sounds recordings of most sound types were not significantly biased in this experimental setup. However, we did record high-frequency clicks from several species as reported by Tricas et al. (2006), which had peak frequencies from 2 to 3 kHz. Thus, we analyzed these sounds both at full sampling frequency and after low-pass filtering at 3 kHz.

Sounds were produced as single pulse events or in trains. We measured the duration (ms) of all pulsed sounds. Sequential sound events separated by pulse intervals <1 s were considered to be part of a pulse train. For pulse trains, we measured the total time of the train, pulse period (time between successive pulse onsets) and interpulse interval (time between pulse events). We counted the number of pulses in the train to calculate the average pulse rate of each sound train (pulses s⁻¹ or Hz). Spectral features of individual pulse events were analyzed using custom MatLab (v7.0) scripts. Power spectra of 3 kHz sampled files were generated from 4096 point fast-Fourier transforms of Hanning windowed data. From these power spectra, we determined the peak frequency, median 6 dB frequency (the median frequency bin of all frequencies ≤6 dB from peak), the minimum 6 dB frequency (the lowest frequency bin ≤6 dB from peak), the maximum 6 dB frequency (the highest frequency bin ≤6 dB from peak) and the 6 dB bandwidth proportion (the percent of frequency bins ≤6 dB from the frequency peak bin).

Body movements associated with sound production were qualitatively described and categorized from frame-by-frame analysis of the video recordings with QuickTime 7. The body part (e.g. jaw, head, body, caudal region; dorsal, anal, pelvic or caudal fin), direction (dorsal, ventral, lateral) and form (e.g. extension, retraction, protraction, shake, slap) of motion by the body part(s) were determined. Subsequent review of the video data sometimes resulted in assignment of more descriptive action patterns such as the ‘head bob’ to indicate motion in the dorsal–ventral plane and ‘head shake’ as oscillation of the head in the horizontal plane. In this analysis, each movement pattern we report to be associated with the acoustic behavior always overlapped movements associated with production of a sound, but was not necessarily coincident with the sound event. Our normal speed video recordings were of relatively low temporal resolution (30 frames s⁻¹=33 ms) and we may not have reliably captured more rapid motions. In addition, we recognize that many of these movements do not define causal mechanisms, rather we assume they are associated with the internal sound production mechanism.

Measures of the acoustic characteristics of individual sounds were averaged for each individual and then averaged among individuals to estimate the population statistic for each species. For comparisons among species, data were tested for the assumptions of normality and homogeneity of variance and, when violated and could not be transformed, non-parametric tests were used. Sound characteristics of low-frequency sound types were compared with a series of one-way ANOVA tests and Student–Newman–Keuls post hoc tests or Kruskal–Wallis tests with Dunn's post hoc tests. t-tests or non-parametric Mann–Whitney U-tests were used to compare characteristics of sounds shared between two species. Characteristics of two sound types produced by individuals of a species were tested for differences with the paired t-test.

We used Mesquite software (v3.02; Maddison and Maddison, 2015) to explore the diversity and evolution of the acoustic behavior characters that are associated with sound types and characteristics across the recent molecular phylogeny of Bellwood et al. (2010). We reconstructed the ancestral character states of the head bob and tail slap behaviors using the Mk1 model that reports the proportional likelihood of the ancestral characters across the phylogeny, and also a parsimony reconstruction (unordered, one step). We used the PDAP:PDTREE module (v1.16; Midford et al., 2005) to conduct a preliminary analysis on the continuous characters of sounds for our six species (e.g. duration, mean frequency, bandwidth, intensity) to get insight on possible correlated sound characteristics across our test phylogeny. We recognize that this is a very small sample size of species (∼5%) within the family and that these analyses will undoubtedly change as more taxa are added. Nonetheless, we view these analysis as providing insight for the development of testable hypotheses in future research on the evolution of sound production.

We thank Jon Dale, Emily Donham, Adam Dewan, Karen Maruska, Luke Herman for their assistance with the lab experiments and analyses; Jaqueline Webb for comments on an early version of this manuscript; Whitlow Au and John Allen for many discussions on butterflyfish sounds, bioacoustics and fish hearing. Finally, we thank Aaron Rice and an anonymous reviewer for their comments on our manuscript, and Bob Thomson and Julio Rivera (Department of Biology) for advice on our phylogenetic analysis.