Evolutionary Patterns in Sound Production across Fishes

One hundred and forty-one references derived from the published technical and scientific literature (journal articles, books, unpublished dissertations and theses, conference proceedings, and technical reports) were used to build a family-level database on the occurrence of fish sounds. This resulted from an extensive search employing three methods to find documentation of sounds at the family level. First, we reviewed a series of seminal works and comprehensive reviews in the area of fish bioacoustics to populate the initial family-level list (e.g., Dufossé, 1874; Sörensen, 1894–1895; Tavolga, 1968, 1971; Fish and Mowbray, 1970; Fine et al., 1977; Tavolga et al., 1981; Amorim, 2006). Second, we employed a series of standardized search terms in scientific search engines (“All Databases” in Web of Science, Google Scholar) using the search operator: “(fish* AND (sound production OR acoustic OR vocal* OR bioacoustic* OR sonic OR soniferous OR sound*))”. Third, a heuristic approach similar to “snowball sampling” (sensuJohnson, 2014; Doohan et al., 2019) was adopted to track both citing literature and the literature cited of papers. Lastly, in cases where there was no evidence for sound production for a particular family returned through the broader keyword-based search, targeted searches were conducted in species or family guides (e.g., FishBase, FAO Guides) for any information on the occurrence of sound production. Documented evidence for the absence of soniferous behavior based on attempts to record sounds from such species with hydrophones was very limited (Fish and Mowbray, 1970; Kaatz et al., 2010).

The above search resulted in three principal sets of evidence identifying the occurrence of a single character—soniferous behavior (see Table S1 for summary, including sources; see Data Accessibility): 1) quantitative or pictorial documentation of acoustic recordings; 2) morphological characters strongly predictive of sonic ability (Fine and Parmentier, 2015); 3) qualitative descriptions of acoustics or morphological characters indicative of sonic ability.

As stated above, morphological characters strongly predictive of sonic ability were one source of evidence identifying the occurrence of soniferous behavior within a family (Table S1; see Data Accessibility). Although the database was not as large as that available for acoustic recordings, it was still formidable. Given this, we separately mapped the phylogenetic distribution of three major categories of sonic mechanisms based on morphological (structural) characters reported in the literature that are considered to produce soniferous behavior (Table S1; see Data Accessibility). (1) Swim bladder vibration resulting from contractions of striated muscle attached to the swim bladder either directly or indirectly via movement of another skeletal part in close proximity. Examples include contraction of “drumming” muscles directly attached to the walls of the swim bladder of toadfishes and the sonic muscles of catfishes that indirectly induce swim bladder vibration via an elastic spring mechanism (Fine and Parmentier, 2015). (2) Stridulation resulting from movement of skeletal parts against each other. Examples include movements of oral jaw or pharyngeal teeth in anemonefish (Fine and Parmentier, 2015), fin spines in catfish (Fine and Parmentier, 2015), and pectoral girdle parts in sculpin (Barber and Mowbray, 1956; Fine and Parmentier, 2015). Other skeletal elements were also included, such as the snapping of pectoral fin tendons in croaking gouramis (Trichopsis vittata, family Osphronemidae; see Ladich et al., 1992; categorized as “buckling” by Bradbury and Vehrencamp, 2011). (3) Non-swim bladder vibration resulting from vibration of the entire body, large parts of the body, or of muscles not associated with the swim bladder. In some families, multiple sonic mechanisms have been identified (e.g., Kaatz, 2002), and these were included in analyses.

The presence or absence of soniferous behavior was scored as a binary character (Table S1; see Data Accessibility). To be conservative, we coded all families lacking such evidence as 0 (silent). Ancestral states were calculated using stochastic character mapping with the make.simmap function in the phytools package for R (version 0.7-70; Revell, 2012), with 1,000 MCMC generations, and the ARD (“All Rates Different”) model to not assume that rates of behavioral evolution are the same among all clades. Clade values and transition rates were calculated by simulation, and posterior probabilities were mapped using the densityMap function in phytools (Revell, 2012; Figs. 1, 2). Probability values ≥50% were considered to be indicative of soniferous behavior being ancestral for a clade.

Next, we mapped the presence of soniferous behavior onto a recent phylogeny of the Actinopterygii that included species from 470 families (Rabosky et al., 2018; Fig. 1, Table S1; see Data Accessibility). All species included in the phylogeny (Rabosky et al., 2018) were assigned to their currently recognized families using Eschmeyer's Catalog of Fishes (Fricke et al., 2020). The Rabosky et al. (2018) tree was then pruned to include only one member of each family based on genus, using a list of valid genera in each family from the Eschmeyer Catalog of Fishes (Fricke et al., 2020). A rapidly expanding body of work has demonstrated that in actinopterygian families where sound production has been investigated, most species examined to date are soniferous, including toadfishes (Batrachoididae; e.g., Mosharo and Lobel, 2012), drums (Sciaenidae; e.g., Ramcharitar et al., 2006), damselfishes (Pomacentridae; e.g., Parmentier et al., 2016), butterflyfishes (Chaetodontidae; e.g., Tricas and Boyle, 2015), sturgeons (Acipenseridae; e.g., Johnston and Phillips, 2003), gobies (Gobiidae; e.g., Zeyl et al., 2016), cichlids (Cichlidae; e.g., Lobel et al., 2021), squirrelfishes (Holocentridae; e.g., Parmentier et al., 2011), and cods (Gadidae; e.g., Hawkins and Picciulin, 2019). Based on the existing and abundant evidence for extensive soniferous behavior within families, we decided that it was reasonable to assume that this behavior is conserved at the family level.

We recognize that our use of presence-only categorical character data, rather than continuous data, greatly limits the different types of phylogenetic comparative methods suitable to calculate and evaluate our data. We also recognize that the sampling of soniferous fishes represents a possible sampling bias (e.g., pelagic, deep-water, or rare species are undersampled) and that our results may be sensitive to uncertainties in character data that can influence outcomes using comparative methods in phylogenetic analysis (Paterno et al., 2018). Two principal sources of error may influence our interpretations: 1) soniferous behavior is conserved at the family level (Type I error, or false positive, where we incorrectly infer that soniferous behavior is conserved at the family level), and 2) absences of data represent true absences of soniferous behavior and not just the result of sampling bias in previous studies (Type II error, or false negative, where we incorrectly infer that soniferous behavior is absent at the family level). Missing or uncertain character-state data is recognized to create uncertainty in phylogenetic inference and interpreting evolutionary patterns and processes through ancestral-state reconstruction (e.g., Maddison, 1993; Kearney, 2002). There have been several approaches to dealing with character state uncertainty, ranging from conservatively coding uncertain data as “absent” (e.g., Baliga and Law, 2016; Chen and Wiens, 2020) to modeling the probability of uncertainty or misclassification (Paterno et al., 2018). To account for possible uncertainties about our inference that sound production is conserved at the family level, we performed a sensitivity analysis (sensuGrant and Kluge, 2003) to evaluate the robustness of our stochastic character mapping results to changes in the underlying data. Following an approach similar to Odom et al. (2014), we conducted an iterative set of simulations where we artificially varied the proportion of families having or lacking evidence of soniferous behavior. We chose the range of 0–12% for sensitivity analysis for ancestral-state reconstruction, because clade values tended to converge on 50% probability for all clades at or beyond 12% uncertainty, especially for the false absence tests. We randomly assigned 2, 4, 6, 8, 10, and 12% of those missing families as having (false absence, or Type II error) or not having (false presence, or Type I error) evidence for sound production, and used stochastic character mapping (as above) to simulate character evolution with these missing data with 1,000 iterations for each permutation. With new ancestor reconstructions based on those randomizations, we then qualified our ancestral-state probability for key clades (Table 1).

Evidence of soniferous behavior (direct or indirect) was identified in 175 of the 470 families represented in the phylogenetic hypothesis presented by Rabosky et al. (2018; Fig. 1, Table S1; see Data Accessibility) based on our three sources of evidence (Fig. 2, also see Materials and Methods): 52 families were supported by acoustic recordings or analysis, 26 families by inference based on morphological characters well known to be associated with sound production, 66 families by both acoustic and morphological evidence, and 31 families by qualitative descriptions indicative of soniferous behavior.

Stochastic character mapping simulates the distribution of a character along branches of a phylogeny (Bollback, 2006; Revell, 2012), and summaries of many simulations (n = 1,000 in this study) are used to compute probabilities of a character being ancestral for clades (also see Materials and Methods). Figure 2 reconstructs ancestral states of soniferous behavior across actinopterygian phylogeny showing probabilities ranging from 0% to 100% that the family is characterized by sound production; Table 1 presents a summary of probability values at key nodes.

Although soniferous behavior occurs in the three extant clades of non-teleostean actinopterygians (Polypteriformes, Acipenseriformes, Holostei; Fig. 2), the reconstruction reveals that it is unlikely to be ancestral with probability values of only 32.3% for Actinopterygii and 16.6% for Teleostei, which comprises >99.8% of actinopterygian species (Nelson et al., 2016). Otocephala, a species-rich subclade of actinopterygians exhibiting morphological adaptations to enhance hearing (Braun and Grande, 2008), and Ostariophysi, a large subgroup of Otocephala, have even lower probabilities of 9.8% and 9.1%, respectively, that soniferous behavior is ancestral. Euteleostei, a second large subclade of Teleostei that includes two-thirds of living fish species, also shows a low probability of 10.7% that soniferous behavior is ancestral.

We find much stronger support for soniferous behavior as a character at the base of several key clades within teleosts. Osteoglossiformes, an early-diverging clade of teleosts, contains four soniferous families and has a 54.3% probability that soniferous behavior is ancestral. Siluroidei, a subclade of catfishes, and Curimatoidea, a subclade of the Characiformes, have probabilities of 93.9% and 63.6%, respectively, that soniferous behavior is ancestral (Figs. 2, 3A). Acanthomorpha, which includes 85% of actinopterygian species in marine habitats (Wainwright and Longo, 2017), has a low probability of 24.8%, but the subclades Percomorphaceae and Eupercaria (e.g., “surgeonfishes,” “drums,” “grunts,” scorpaenoids) have probabilities of 75.3% and 78.1%, respectively, that soniferous behavior is ancestral (Fig. 3B). An even higher probability of 97.8% supports soniferous behavior as ancestral for a crown group within Eupercaria (Fig. 3B)—Hexagrammidae (greenlings) + Cottoidei (e.g., sculpins) + Zoarcoidei (e.g., wolffishes).

In aggregate, our results indicate that soniferous behavior has a high probability, i.e., more than 50%, of being ancestral for approximately 33 independent clades across Actinopterygii identified in our analyses (Fig. S1; see Data Accessibility). We interpret this as evidence of widespread, independent evolution of sound production in actinopterygian fishes.

Morphological information on three broad categories of sonic mechanisms (see Materials and Methods for more clarification) was available for 89 actinopterygian families. As illustrated in Figure 4 and listed in Table S1 (see Data Accessibility), families with more than one mechanism have two filled circles in different colors (one color/mechanism; see legend and Table S1; see Data Accessibility). A single mechanism is reported for 67 families; swim bladder vibration (SBV) is the most common (n = 41 families), followed by stridulation (STR, n = 19 families) and non-swim bladder vibration (non-SBV, n = 7 families). The following families with a non-SBV mechanism include identified species within the family that have lost their swim bladder: Hexagrammidae (n = 4, Shinohara, 1994), Aploactinidae (n = 4, Matsubara, 1943; Imamura, 2004), and Synanceiidae (n = 4, Matsubara, 1943; Imamura, 2004).

Swim bladder vibration occurs with a second mechanism in an additional 22 families (Fig. 4). Both SBV and STR occur within 18 of these families: Tetraodontidae, Balistidae, Tetrarogidae, and 15 families in the Siluroidei (see Fig. 4 for family names); individual species exhibiting both mechanisms are found in Balistidae and the 15 siluroid families. Both SBV and non-SBV are reported for four families: Aploactinidae, Synanceiidae, Hexagrammidae, Rhamphocottidae.

Sensitivity analysis incorporating uncertainty in the number of soniferous families did not change the outcomes of our principal findings (Fig. 5). Clades in which there is a high probability that sound production is ancestral, namely Osteoglossiformes, Curimatoidea, Siluroidei, Percomorphaceae, Eupercaria, and Hexagrammidae + Cottoidei + Zoarcoidei, still had a >50% probability of sound production as an ancestral character even when up to 12% of soniferous families are removed (Table 1, Fig. 5). Assignment of ancestral-state probability for key clades did not change significantly, even with up to 12% of Type I or Type II error, which is less likely than perhaps 1% or 5% of families with missing data that may ultimately be found to be acoustic.

The results presented here for actinopterygians, in combination with recent ones for tetrapods (Chen and Wiens, 2020), broadly indicate strong selection to exploit sound production as a behavioral character across vertebrate evolution. Our findings are significant for several reasons. First, we show that sound production is likely as ancient in Actinopterygii as it is in Tetrapoda (∼100–200 Ma, Chen and Wiens, 2020), given its presence in Acipenseridae, a family originating circa 155 Ma (Shen et al., 2020), and in Osteoglossiformes, a single order of five families (Nelson et al., 2016) with origins circa 100 Ma (see Lavoué et al., 2012). Polypteriformes are also an ancient lineage (Giles et al., 2017) that exhibit soniferous behavior, but extant species of Polypteridae share a more recent origin, 15–25 Ma (Near et al., 2014; Hughes et al., 2018). Second, evidence for (or strongly suggestive of) soniferous behavior is present in a surprising number of families (n = 175) that contain nearly 85% of the estimated 34,000 valid extant species of actinopterygians spread across the tree (Fig. 1; Table S1; see Data Accessibility). Last, the results suggest that actinopterygians independently evolved soniferous ability approximately 33 times (Fig. S1; see Data Accessibility), an estimate that may change as soniferous behavior is either found or not in more species.

One of the largest limitations with our analysis was our ability to confidently infer the absence of sound production within a family, since the absence of evidence is not evidence of absence. There have been relatively few published studies that specifically report on lack of soniferous behavior despite recording attempts or morphological investigation (e.g., Fish and Mowbray, 1970; Kaatz et al., 2010; Hawkins and Picciulin, 2019). For example, over the development of the field of fish bioacoustics, there have been examples of certain taxa being categorized as silent (Sebastes in Fish and Mowbray, 1970) that later recording attempts demonstrated to be soniferous (Širović and Demer, 2009); consequently, if sounds are not recorded during an initial set of observations, it does not always represent definitive proof that the focal species is incapable of soniferous behavior. While phylogenetic comparative methods offer extensive power to quantitatively analyze evolutionary patterns, these approaches are limited for categorical data compared to continuous data. Further, our data should be considered a “presence-only” dataset rather than presence–absence. There is a precedent within the paleontological literature for inferring trait evolution using phylogenetic comparative methods that use either presence-only datasets or datasets with unclear patterns of trait absence (e.g., Finarelli and Flynn, 2006; Benton, 2015; Hunt and Slater, 2016), particularly where appropriate trait data may be limited, such as behavior (Hsieh and Plotnick, 2020), physiology (e.g., Legendre et al., 2016), or development (Organ et al., 2015; Jablonski, 2020). Our method for the ability to infer the identity of non-soniferous families offers an opportunity to further explore these families for the potential of soniferous behavior. We envision that many additional fish species will be shown to produce sounds in the coming years, further supporting our demonstration of the widespread nature of soniferous behavior across fishes.

Our survey of sonic mechanisms based on morphological characters would lead one to conclude that swim bladder vibration (SBV) is the ancestral character state for soniferous actinopterygians, occurring in 63 of the 89 families included in our survey; SBV occurs with a second mechanism for the same or different species within a family in about one-third of these families (Fig. 4, Table S1; see Data Accessibility; see Results for definitions). Tetrapods also exhibit examples where the most prevalent sonic organ—larynx or syrinx in non-avian and avian taxa, respectively—occurs together with a second mechanism of sound production (see Bass and Chagnaud, 2012).

If sound production in two sister taxa is achieved through two different mechanisms defined by different morphological characters, this could be an indication that sound production has independently evolved in these two sister taxa because the mechanism is different. While the evidence remains somewhat limited, we find no such support for this possibility in our survey. As such, the mapping of morphological mechanisms presented here does not change our estimate regarding the approximate number of times (33) sound production has independently evolved among actinopterygians. We recognize, however, that it remains important for establishing ancestral character states to look at close relatives of some of the soniferous clades identified here to see if they exhibit similar morphologies for sound production, especially where one is coded as silent and the other as soniferous, e.g., among the Otocephala and Eupercaria (Fig. 3A, B). We further recognize that evidence of one sonic morphology in a single family or species does not necessarily mean that it is the only one because any one study may have only focused on identifying the presence of that character.

The presence and absence of soniferous behavior among actinopterygians likely includes secondary loss (Miles and Fuxjager, 2019). Within speciose clades where soniferous behavior has a high probability of being ancestral, non-soniferous clades may have secondarily lost this character. For example, the clade Hexagrammidae + Cottoidei + Zoarcoidei have 97.8% probability that sound production is ancestral. Fish and Mowbray (1970) comment on the absence of sound production and swim bladders in Zoarcidae (their Zoarchidae). If further research provides conclusive evidence for absence, then our tree (Fig. 3B) likely indicates secondary loss. As an additional example, sound production is likely ancestral for Gadiformes (Hawkins and Picciulin, 2019), which use specialized swim bladder muscles to make sounds. Yet several species of gadiforms have reduced swim bladder muscles and do not produce sounds, likely representing secondary loss of this behavior (Hawkins and Picciulin, 2019). Other places to investigate potential loss of soniferous capacity are between sister groups where one is coded as silent and the other as soniferous (e.g., Fig. 3).

A particularly fascinating case of secondary loss concerns mochokid catfishes in the genus Synodontis; some species are only soniferous and other more recently diverged species are both soniferous and weakly electric or only weakly electric (Boyle et al., 2014). Weakly electric species of Synodontis have reduced sonic muscles, but share features of myogenic electric organs (Kéver et al., 2020, 2021). Neuroanatomical studies show that the general pattern of central nervous system organization of populations of neurons is highly conserved among all species of Mochokidae so far investigated (Kéver et al., 2020, 2021). How the electrophysiological properties (physiological characters of Wiley, 1981) of these neuronal populations vary depending on their role in sound production or electric organ discharges has only begun to be explored (see Kéver et al., 2020).

Our results indicate that within Actinopterygii, soniferous behavior occurs across the most speciose clades and is estimated to have evolved independently about 33 times. This is a quickly expanding field (Lobel et al., 2010; Lindseth and Lobel, 2018), and there are surely more soniferous fish species and families to be recorded. In a broader sense, together with the recent report showing evolutionary patterns of sound production in tetrapods (Chen and Wiens, 2020), our findings highlight the important role that this behavior has played in the history of vertebrates.

Supplemental material, including Table S1, Figure S1, and PDF versions of all figures, is available at https://www.ichthyologyandherpetology.org/i2020172. Unless an alternative copyright or statement noting that a figure is reprinted from a previous source is noted in a figure caption, the published images and illustrations in this article are licensed by the American Society of Ichthyologists and Herpetologists for use if the use includes a citation to the original source (American Society of Ichthyologists and Herpetologists, the DOI of the Ichthyology & Herpetology article, and any individual image credits listed in the figure caption) in accordance with the Creative Commons Attribution CC BY License.

Thanks to K. Bemis, T. Grande, H. W. Greene, L. Hughes, G. Ortí, L. Page, R. Grosberg, E. Schuppe, M. Wilson, and K. R. Zamudio for discussion and helpful comments on the manuscript. We thank Jon D. Fong for providing a list of valid genera for each family from the Eschmeyer Catalog of Fishes. Funding provided by National Science Foundation awards OCE-1736936 (A.N.R.), DBI-1523836 (S.C.F.), and IOS-1656664 (A.H.B.); the U.S. Bureau of Ocean Energy Management M21AC00003 (A.N.R.); the Tontogany Creek Fund (W.E.B.); and Cornell Lab of Ornithology (A.J.M.). While we have made every effort to include all known soniferous fish families in our analysis, we apologize if we have overlooked taxa not represented here. A previous version of this manuscript was posted on a preprint server (Rice, A. N., S. C. Farina, A. J. Makowski, I. M. Kaatz, P. S. Lobel, W. E. Bemis, and A. H. Bass. September 2020. Evolution and ecology in widespread acoustic signaling behavior across fishes. bioRxiv DOI:10.1101/2020.09.14.296335). A.H.B., W.E.B., A.N.R. (listed alphabetically) conceived the study. All authors aggregated data. A.H.B., W.E.B., S.C.F., A.N.R. (listed alphabetically) analyzed the data. S.C.F. and A.N.R. conducted statistical analyses. A.N.R. wrote initial draft; A.H.B., W.E.B., S.C.F., A.N.R. (listed alphabetically) revised. All authors approved the final version of the manuscript.