Evolution of vocal patterns: tuning hindbrain circuits during species divergence

The complex temporal dynamics of motor patterns arise from finely tuned circuits in the central nervous system. Differences in courtship behaviors between populations can contribute to pre-mating isolation and thus to speciation (Hoskin and Higgie, 2010). Within extant species of African clawed frogs (Xenopus and Silurana), males produce a species-specific advertisement call that coordinates courtship (Tobias et al., 2011). Differences in temporal features of advertisement calls can serve as species identifiers (Tobias et al., 2011). Female anurans, including Xenopus laevis, prefer conspecific to heterospecific advertisement calls (e.g. Ryan and Rand, 1993; Gerhardt, 1994; Picker, 1983). Thus, call divergence during speciation could contribute to prezygotic reproductive isolation, reinforcing the genetic divergence of populations and driving speciation. While the ultimate results of divergence in courtship signaling have been extensively examined (e.g. West-Eberhard, 1983; Kirkpatrick and Ryan, 1991; Gerhardt, 1994), specific changes in the neural circuits responsible for the evolution of signaling have only recently been identified (e.g. Chakraborty and Jarvis, 2015; Leininger and Kelley, 2015; Katz, 2016). Neural circuits controlling stereotyped courtship behaviors produced by closely related species can reveal these proximate mechanisms (Katz and Harris-Warrick, 1999).

Each species of Xenopus produces a distinct advertisement call consisting of a train of several sound pulses (Tobias et al., 2011; Leininger and Kelley, 2015). The laevis clade – X. laevis sensu lato – consists of four geographically separated species that diverged ∼8.5 Mya (Furman et al., 2015; Evans et al., 2004; Fig. 1A), each with a distinctive advertisement call (Tobias et al., 2011; e.g. Fig. 1B–D). Because these species diverged relatively recently, the neural circuits producing these temporally distinct patterns are likely to be similar, allowing us to identify mechanisms for tuning specific temporal call features.

Xenopus vocal behavior and its underlying neural circuitry have been studied most extensively in X. laevis (reviewed in Zornik and Kelley, 2017). The male X. laevis advertisement call consists of highly stereotyped, alternating trains of fast and slow rate sound pulses (trills: see Fig. 1B). Compound action potentials (CAPs), produced by groups of hindbrain vocal motor neurons in the laryngeal motor nucleus IX-X (LMN), exit the central nervous system via the fourth rootlet of cranial nerve IX-X (laryngeal nerve) and travel to the vocal organ, the larynx, generating muscle contractions and sounds (Tobias and Kelley, 1987; Yager, 1992). The pattern of CAPs recorded from the laryngeal nerve of calling X. laevis males corresponds to the pattern of sound pulses during in vivo calling (Yamaguchi and Kelley, 2000). Fictive advertisement calling – CAP patterns that parallel in vivo call patterns – can be evoked by application of serotonin to the in vitro male brain (Fig. 2A; Rhodes et al., 2007), providing a powerful preparation for understanding the underlying neural circuits.

Using this preparation, a central pattern generator (CPG) located in the hindbrain was identified as responsible for fictive call patterns in X. laevis (Rhodes et al., 2007). The primary premotor nucleus in the CPG is DTAM (in earlier literature, an abbreviation for the dorsal tegmental area of the medulla, now used as a proper name), a putative homolog of the para-trigeminal respiratory group in lampreys and the parabrachial complex in mammals (see Discussion). DTAM projects to the caudal hindbrain LMN (Wetzel et al., 1985; Zornik and Kelley, 2007), homologous to the nucleus ambiguus of mammals (Albersheim-Carter et al., 2016), which also contributes to CPG function (Fig. 2A). Neurons in DTAM monosynaptically excite LMN motor neurons that produce vocal patterns (Zornik and Kelley, 2008; Yamaguchi et al., 2008; Zornik et al., 2010; Zornik and Yamaguchi, 2012). A local field potential (LFP) wave recorded from DTAM coincides with fast trill and is NMDA receptor dependent (Zornik et al., 2010). Taken together, available evidence suggests that in X. laevis, DTAM neurons drive the LFP wave and control call duration and period (Zornik and Yamaguchi, 2012).

What mechanisms underlie vocal divergence in the laevis clade? While all species in the clade produce fast trills, the duration and period (time from the onset of one call to the onset of the following call) of their calls differ substantially (Tobias et al., 2011). We used the in vitro brain preparation originally developed in X. laevis (Rhodes et al., 2007) to compare motor and premotor activity during fictive calling in X. laevis, X. petersii and X. victorianus. In this study, we tested the hypothesis that species-specific tuning of the DTAM circuitry underlies call divergence in the laevis clade.

All animal care and experimental procedures conformed to guidelines set forth by the National Institutes of Health and were approved by Columbia University's Institutional Animal Care and Use guidelines (protocol no. AC-AAM1051). Sexually mature male X. laevis (mean±s.d. mass: 34.5±6.0 g, n=10), X. petersii (11.9±0.7 g, n=10) and X. victorianus (11.2±1.2 g, n=5) were used in experiments. Animals were raised in the laboratory colony or purchased from Xenopus Express (Brooksville, FL, USA), group housed in polycarbonate tanks, each with 10 l of filtered water changed twice weekly following feeding with frog brittle (Xenopus Express), and kept on a 12 h light:12 h dark schedule at 20°C (X. laevis) or 23°C (X. petersii and X. victorianus).

Male frogs (n=5 each for X. laevis, X. petersii and X. victorianus) were injected via the dorsal lymph sac with 200 μmol l⁻¹ human chorionic gonadotropin 48 and 24 h before calls were recorded. Each frog was placed in a glass aquarium (60×15×30 cm, water depth 25 cm, temperature 20–22°C) in a room illuminated dimly with red light. After 10 min of habituation, a conspecific female was placed in the tank for 30–45 min or overnight during a sound-activated recording (Audacity^®, http://www.audacityteam.org/). Vocalizations were recorded with a hydrophone (CA30 or H2a, Aquarian Audio Products, Shoreline, WA, USA); the signal was digitized (Omega SV, Lexicon Pro), and saved to a personal computer in .wav format (Amadeus Pro, HairerSoft, Kenilworth, UK or Audacity^®).

Male frogs (X. laevis n=12, X. petersii n=14, X. victorianus n=5) were deeply anesthetized by injection of 1.3% tricaine methanesulfonate (MS-222, Sigma; X. laevis: 500–700 μl, X. petersii and X. victorianus: 200 μl) into the dorsal lymph sac and brains were removed from the cranium in ice-cold saline (in mmol l⁻¹: 96 NaCl, 20 NaHCO₃, 2 CaCl₂, 2 KCl, 0.5 MgCl₂, 10 Hepes and 11 glucose, pH 7.8, oxygenated with 99% O₂). Following recovery for 1 h in saline at room temperature, the brains were pinned in a silicone elastomer-lined recording dish (Sylgard, Dow Corning, Midland, MI, USA) and constantly superfused with fresh oxygenated saline at room temperature. CAPs were recorded with an extracellular suction electrode placed on the posterior rootlet of cranial nerve IX-X (laryngeal nerve) that contains the axons of the laryngeal motor neurons (Simpson et al., 1986). To elicit a fictive motor pattern on the laryngeal nerve, serotonin (Sigma) was bath applied to the recording dish to a final concentration of 30 or 60 μmol l⁻¹ (Zornik and Yamaguchi, 2012). The signal from the nerve was amplified ×1000 (Model 1700, A-M Systems, Carlsborg, WA, USA) and band-pass filtered (10 Hz to 5 kHz), digitized at 10 or 20 kHz (Digidata 1400A, Molecular Devices, Sunnyvale, CA, USA), and recorded on a personal computer. After 5 min of application, serotonin was washed out by reinstating saline superfusion. Serotonin was then re-applied every 60 min for the next 4–8 h.

LFPs in DTAM were recorded using a low-impedance carbon fiber extracellular electrode (300–500 kΩ, Kation Scientific, Minneapolis, MN, USA) while simultaneously recording activity in the laryngeal nerve. The signal was amplified ×100, digitized at 10 or 20 kHz and bandpass filtered at 0.1 Hz to 10 kHz. Either the pia covering the cerebellum was removed to allow the DTAM electrode to penetrate or the cerebellum and tectum were transected sagittally at the midline, reflected laterally to expose DTAM, and pinned. The DTAM electrode was placed ∼100 µm posterior to the tectum–cerebellum border, ∼200 µm from the lateral edge of the fourth ventricle and ∼650 µm below the pial surface. Five X. laevis, five X. petersii and two X. victorianus brains were transected in the transverse plane, caudal to the laryngeal nerve, to isolate DTAM from the LMN. Two X. laevis brains and three X. petersii brains were transected in the transverse plane just rostral to DTAM, in addition to the caudal transection, to also remove any descending inputs from the midbrain and forebrain.

In one X. laevis brain (to confirm previous findings: Zornik et al., 2010), three X. petersii brains and two X. victorianus brains, a selective NMDA antagonist, dl-2-amino-5-phosphonopentanoic acid (APV, Sigma), was bath applied to the recording dish at concentrations of 100–500 μmol l⁻¹. Changes in the LFP wave in response to drug application were recorded in DTAM and fictive calling was monitored via the laryngeal nerve suction electrode. After 5 min of APV application, 30 or 60 μmol l⁻¹ serotonin was added to the bath for 10 min. One hour after the reinitiation of saline superfusion, serotonin was applied alone to determine whether the effect of APV was reversible.

Calls were sampled from five males of each species. A single call (Fig. 1B–D) is the smallest vocal unit containing a characteristic and repeating pattern of sound pulses (Tobias et al., 2011). For all species, samples were randomly selected from continuous bouts of calling and lasted a minimum of 10 s and a maximum of 30 s. Samples contained a minimum of 15 calls and 100 pulses for each animal. Pulses were detected and analyzed using custom-written MATLAB scripts (MathWorks, Natick, MA, USA; all MATLAB scripts are available upon request from C.L.B.). Pulse rates ranging from 0 to 100 Hz were assigned to frequency bins (2 Hz width) and then analyzed to determine the threshold rate for fast trill. Rate histograms revealed a minimum at ∼50 Hz in all three species (Fig. 1E) and thus 50 Hz was selected as the threshold for fast trill. In all species, slow trill onset was defined as two or more successive pulses with an instantaneous pulse rate <50 Hz (20 ms) and >10 Hz (100 ms). Fast trills followed slow trills. Fast trills with two or more successive pulses had a pulse rate >50 Hz (20 ms) and <100 Hz (10 ms). Call duration was measured from call onset to call offset. Call period was measured from the onset of a call to the onset of the following call.

Fictive vocal CAP patterns produced following serotonin application were collected and analyzed (X. laevis n=5, X. petersii n=6, X. victorianus n=3) using custom-written MATLAB scripts. For all species, samples were randomly selected from continuous bouts of fictive calling and lasted a minimum of 20 s and a maximum of 30 s. Samples contained a minimum of 15 calls and 100 pulses for each animal. Instantaneous CAP rate was calculated as the reciprocal of the time interval between two successive CAPs. As for in vivo calls, fictive slow trills followed fictive fast trills and typically consisted of a series of two or more pulses with an instantaneous pulse rate <50 Hz and >10 Hz. Fictive fast trills consisted of a series of two or more CAPs >50 Hz and <100 Hz. Fictive call duration was measured from fictive call onset to call offset. Fictive call period was measured from the onset of a fictive call to the onset of the following call.

The DTAM LFP was low-pass filtered to reveal a wave corresponding to fictive fast trill. Because the 5 Hz low-pass filter used for X. laevis does not accurately identify wave onset in species with shorter calls, a 20 Hz filter was used for X. petersii and X. victorianus. Wave onset was defined as the first peak in the second derivative of the low-pass filtered signal prior to the wave peak. Wave offset was defined as the first peak in the second derivative of the low-pass filtered signal following the wave peak. Onset and offset were visually confirmed and manually adjusted if necessary. Call and wave durations were measured from onset to offset time. Call and wave periods were measured from one call onset time to the onset of the following call.

Evoking fictive calling from the isolated brain requires the preservation of laryngeal nerve connections to the LMN during dissection, an intact hindbrain and effective oxygenation to preserve neural function. The success rate for these experiments is variable and can be low. Because the availability of mature male X. victorianus was extremely limited, fictive calling data points are reported individually in cases where n<5.

Statistical analyses were carried out using MATLAB and R (R Core Team, Vienna, Austria). Because the data were not normally distributed, but the coefficient of variance was roughly constant across groups, generalized linear models (GLMs) with gamma distributions were used to compare in vivo and in vitro call characteristics across species (fast trill rate, slow trill rate, call duration and call period). Linear regression models were used to compare in vivo fast trill duration and period with in vitro fast trill duration and period, respectively. To compare the relationship between the onset and offset of the DTAM LFP with fast trill, the median for each species was used. A GLM was used to compare animal means for DTAM LFP wave duration and period in intact brains and for transected brains between X. laevis and X. petersii.

We recorded advertisement calls from five adult male X. laevis, X. petersii and X. victorianus. All three species produce a distinctive male advertisement call consisting of repeated trains of sound pulses or trills (Fig. 1B–D, Table 1). Xenopus laevis calls alternate between a slow trill (∼30 Hz) of ∼25 sound pulses and a fast trill (∼60 Hz) of ∼12 pulses (Fig. 1B). Similarly, X. petersii calls (Fig. 1C) consist of a slow trill with ∼3 pulses followed by a fast trill made up of 2 pulses (Fig. 1Ci, percentage of fast trills that are doublets: 35.2±26.2%, mean±s.d.) or a single pulse (Fig. 1Cii, identified as following a slow trill). The advertisement call of X. victorianus, however, consists of only fast trills made up of 3 or 2 pulses (Fig. 1Di,ii); slow trills are absent. Fast trills thus occur in all three species.

We used GLMs to compare vocal characteristics of advertisement calls across all three species (Fig. 2, Table 1). Slow trill pulse rates (Fig. 2A, Table 1) did not differ significantly between X. laevis and X. petersii (P>0.05; X. victorianus does not have a slow trill). Fast trill pulse rates did not differ significantly between X. laevis and X. petersii or between X. petersii and X. victorianus (P>0.05 for each comparison; Fig. 2B, Table 1). Fast trill rate was significantly faster in X. victorianus compared with X. laevis (P<0.01).

Call duration (P<0.0001 for all comparisons; Fig. 2C, Table 1) and period (P<0.0001 for all comparisons; Fig. 2D, Table 1) differed significantly between all three species, with X. laevis having the longest call duration and period and X. petersii having the shortest call duration and period.

When the X. laevis brain was removed and placed in an oxygenated saline solution (Fig. 3A), bath application of serotonin resulted in species-specific activity patterns recorded from the laryngeal nerve (Fig. 3B–D). In male X. laevis, the temporal features of these in vitro CAPs parallel the pattern of sound pulses in the in vivo advertisement call (compare Fig. 1B with Fig. 3B) and have thus been termed fictive advertisement calling (Rhodes et al., 2007). If fundamental neural circuit properties for calling are conserved across the X. laevis clade, we hypothesized that serotonin should also elicit fictive advertisement calling in X. petersii and X. victorianus.

We found that bath application of serotonin to isolated brains also elicited patterned laryngeal nerve CAPs in X. petersii (n=6) and X. victorianus (n=3, due to limited availability of adult males of this species; Fig. 3C,D). As in X. laevis (n=5; Fig. 3B), these CAP patterns closely resembled the patterns of each species' advertisement call (compare Fig. 1 with Fig. 3). CAPs were produced at fast (∼60 Hz; Fig. 3, yellow) and slow (∼30 Hz: Fig. 3, blue) rates in X. laevis and X. petersii; X. victorianus brains produced only fast CAPs. Four of the six X. petersii brains produced slow trill CAPs and all produced fast trill CAPs. We conclude that fictive calling elicited by serotonin is a fundamental neural circuit property conserved across these three species.

We used GLMs to compare characteristics of in vitro laryngeal nerve activity across all three species (Fig. 4, Table 1). Slow trill CAP rates (Fig. 4A, Table 1) did not differ significantly between X. laevis and X. petersii (P>0.05; X. victorianus does not have a slow trill). Fast trill CAP rates did not differ significantly between X. laevis and X. petersii or between X. petersii and X. victorianus (P>0.05 for each comparison; Fig. 4B, Table 1). Although fairly similar, fast trill rate was somewhat faster in X. victorianus than in X. laevis (P<0.01).

Fictive call duration (P<0.0001 for all comparisons; Fig. 4C, Table 1) and period (P<0.0001 for all comparisons; Fig. 4D, Table 1) differed significantly between all three species, with X. laevis having the longest fictive call duration and period and X. petersii having the shortest fictive call duration and period.

GLM analyses comparing fictive calling in the in vitro preparation with in vivo calling found no significant effects for fast rate, slow rate, call duration or call period. We conclude that fictive calling elicited by serotonin in vitro corresponds to in vivo calling in all three species.

We recorded DTAM activity during in vitro fictive calling in X. laevis (n=5), X. petersii (n=6) and X. victorianus (n=2). A LFP wave in DTAM (Fig. 5A) was associated with fictive fast trill (Fig. 5B) in all three species. This wave began before fictive fast trill (median±s.d. difference: X. laevis: −160.0±115.8 ms, X. petersii: −59.2±13.6 ms, X. victorianus: −47.6±10.0 ms) and terminated after fictive fast trill (median±s.d. difference: X. laevis: 106.7±73.4 ms, X. petersii: 49.4±8.3 ms, X. victorianus: 28.3±12.2 ms) (Fig. 5C).

Because in vivo and in vitro call durations and periods were significantly longer in X. laevis than in X. petersii and X. victorianus (above, Fig. 4D,E), and the DTAM LFP coincides closely with fast trill in each species, we examined the relationship between the duration and period of the LFP wave and the duration and period of fictive fast trill using linear regression in pooled individuals of all three species (X. laevis n=5, X. petersii n=6, X. victorianus n=2). Wave duration was linearly (Fig. 5D) and strongly (R²=0.947, slope=0.6618, P<0.0001) related to fictive fast trill duration. Wave periods in all three species were also linearly (Fig. 5E) and strongly (R²=0.991, slope=1.0028, P<0.0001) related to periods between successive fast trills. We conclude that the temporal relationship of the DTAM LFP to fictive fast call duration and period is conserved across the three species.

In X. laevis, fictive fast trills and the DTAM LFP wave are both NMDA dependent (Zornik et al., 2010). We hypothesized that if similar mechanisms underlie calling in X. petersii and X. victorianus, as is suggested by the similarities in the underlying DTAM activity, calling would be NMDA dependent in all species examined here. We thus applied APV (a competitive NMDA receptor antagonist) to fictively singing brains in X. laevis (n=1, to confirm previous findings; see Zornik et al., 2010), X. petersii (n=3) and X. victorianus (n=2) to determine whether NMDA receptor dependence is conserved. One hour prior to APV application, all preparations produced fictive calling accompanied by a LFP wave in DTAM (Fig. 6A) in response to serotonin application. Pretreatment for 15 min with APV blocked both fictive calling and the LFP wave in response to serotonin (Fig. 6B). An hour after APV washout, serotonin-induced fictive calling and the DTAM wave returned (Fig. 6C). The NMDA dependence of the DTAM wave is thus conserved within the laevis clade.

Next, we set out to determine whether DTAM alone is responsible for generating the LFP wave, rather than other components of the vocal circuit, such as interneurons located in the anterior portion of the LMN (red portion of LMN, Fig. 2A). These neurons innervate DTAM (Zornik and Kelley, 2007), raising the possibility that the LMN controls the timing of the LFP wave. To address this question, we transected the in vitro brain between DTAM and the LMN, just caudal to the laryngeal nerve. After transection, serotonin application produced a DTAM LFP wave that closely resembled the DTAM LFP wave produced by intact brains in X. laevis (n=5), X. petersii (n=5) and X. victorianus (n=2) (Fig. 7A,B). We used GLMs to examine wave duration and period across species and found that both were significantly longer in X. laevis than in X. petersii and X. victorianus (Fig. 7C,D, P<0.0001 for all comparisons). Xenopus petersii wave period (P<0.0001) but not duration (P>0.05) was significantly longer than that of X. victorianus. Transections did not significantly alter wave duration in X. laevis, X. petersii or X. victorianus (P>0.05, means±s.d.: X. laevis: 444.2±29.6 ms intact, 465.3±28.9 ms transected; X. petersii: 108.7±7.3 ms intact, 118.0±2.3 ms transected; X. victorianus individual means: 119.8 and 105.4 ms intact, 113.2 and 121.1 ms transected; Fig. 7C). Wave periods in all species were also unaffected by transection (P>0.05, means±s.d.: X. laevis: 1115.6±188.8 ms intact, 1016.2±71.4 ms transected; X. petersii: 695.7±157.5 ms intact, 733.9±194.4 ms transected; X. victorianus individual values: 372.9 and 396.6 ms intact, 442.5 and 463.3 ms transected; Fig. 7D). We conclude that the duration and the period of the DTAM LFP wave do not reflect inputs from the LMN in these three species.

DTAM also receives input from the central amygdala located in the ventral forebrain (Brahic and Kelley, 2003; Hall et al., 2013). Because previous work has also shown that removal of the midbrain and forebrain does not disrupt fictive calling in X. laevis (Yu and Yamaguchi, 2010), we predicted that species-specific LFP wave patterns are generated autonomously within the hindbrain compartment that includes DTAM. To test this prediction, we combined a transection that eliminated LMN inputs with an additional transection just anterior to the cerebellum that removed inputs from the midbrain and forebrain in two X. laevis and three X. petersii brains. We found that following both transections, the LFP waves persisted in response to serotonin application (wave duration: X. laevis: 395.5, 334.3 ms, X. petersii: 141.5, 166.5, 130.4 ms; wave period: X. laevis: 1212.7, 1092.6 ms, X. petersii: 724.2, 1024.0, 662.1 ms; compare with values above for intact and transected animals). We incorporated these values into our GLMs for wave duration and period and found no significant effect of this anterior transection (P>0.05 for each GLM). These more limited observations suggest that DTAM LFP wave parameters also do not reflect inputs from midbrain or forebrain in the X. laevis clade, reinforcing the idea that intrinsic features of the DTAM LFP govern species differences in vocal patterns.

Three relatively recently (8.5Mya) diverged species within the laevis clade – X. laevis, X. petersii and X. victorianus (Furman et al., 2015) – produce male advertisement calls with distinct temporal properties (Tobias et al., 2011). All of these species produce advertisement calls that include fast trills, but call duration and period are significantly different across all three species, with X. laevis having the longest call duration and period and X. victorianus having the shortest. We show here that key features of the neural circuitry responsible for vocal patterning are conserved across these species. Species-specific fictive advertisement calls (in vitro nerve activity patterns that parallel in vivo sound pulse patterns) can be elicited by applying serotonin to the isolated brain in each species. However, as is the case for actual advertisement calls recorded in vivo, fictive call duration and period recorded in vitro vary significantly across species.

Next, we examined premotor activity that might be responsible for species differences in call duration and period. In each species, fictive fast trills coincide with a LFP wave recorded from the rostral hindbrain nucleus DTAM. Wave duration and period are tightly correlated with fictive fast trill duration and period across individuals of each species. Furthermore, the LFP wave is NMDA receptor dependent in all species. Collectively, these findings support conserved involvement of NMDA-dependent DTAM activity in regulating the duration of fast trill and call periods across all three species.

Differences in DTAM LFP wave duration and period across species might be due to autonomous properties of the nucleus or instead might reflect differences in connectivity with caudal or rostral components of vocal neural circuitry. In support of the former hypothesis, we found that in all species the DTAM LFP wave persists after isolation from caudal LMN inputs. Furthermore, in X. laevis and X. petersii, the wave persists when rostral inputs from the forebrain and midbrain are also removed (X. victorianus was not tested). These observations suggest that species-typical characteristics of the DTAM LFP wave are intrinsic to the hindbrain compartment that includes DTAM, rather than reflecting activity elsewhere in the vocal circuit, and support an autonomous role for DTAM in control of call duration and period. Differences across species in DTAM activity could be the result of intrinsic cellular properties of DTAM neurons, network differences in synaptic connectivity within the nucleus or adjacent regions of the anterior hindbrain, or a combination of these factors.

DTAM was initially identified as a target for androgens (Kelley et al., 1975; Kelley, 1980), essential hormones that support advertisement calling (Wetzel and Kelley, 1983), and as a source of input to vocal motor neurons (Wetzel et al., 1985; Zornik and Kelley 2007, 2008). More recent studies identified a population of vocal premotor neurons in X. laevis DTAM (fast trill neurons, FTNs) with phase-locked spikes preceding each fast trill CAP (Zornik and Yamaguchi, 2012). FTNs exhibit a long-lasting depolarization (LLD) during spiking that coincides with fictive fast trill and with the DTAM LFP wave. The DTAM LFP wave thus likely reflects the activity of a population of FTNs that depolarize synchronously during fast trill. Our data suggest that a homologous population of FTNs in X. petersii and X. victorianus also generate a DTAM LFP wave by producing synchronous LLDs. However, because the waves in X. petersii and X. victorianus are shorter than those in X. laevis, we predict that FTN LLDs will terminate more rapidly in X. petersii and X. victorianus, leading to a brief fictive fast trill and thus a shorter call.

In X. laevis, intracellular recordings in synaptically isolated FTNs reveal rhythmic oscillations in response to NMDA that are similar in duration and period to LLDs recorded during fictive calling (Zornik and Yamaguchi, 2012). Intrinsic cellular properties of FTNs could thus allow them to act as network pacemakers and contribute to fast trill duration and period. For instance, X. petersii and X. victorianus FTN properties could differ from X. laevis FTN properties, allowing them to produce shorter oscillations.

Outward currents could contribute to call period and duration by terminating depolarization in FTNs. Candidates include: calcium-dependent potassium currents (El Manira et al., 1994; Cazalets et al., 1999; Del Negro et al., 1999; Tahvildari et al., 2008), the Na⁺/K⁺-ATPase hyperpolarizing current (pre-Botzinger complex; Rubin et al., 2009), A-type transient K⁺ currents (lamprey locomotion: Hess and El Manira, 2001) and K⁺ leak currents (respiration: Koizumi et al., 2016).

Species differences in ion channels (Harris-Warrick, 2010; Del Negro et al., 2010) expressed by FTNs could serve as control elements in species divergence of rhythmic oscillations that produce differing LLD durations in FTNs, and thus different LFP wave durations. For example, if outward current channels are more highly expressed or are activated more rapidly in X. petersii and X. victorianus than in X. laevis, the duration of the LLD and LFP should decrease because the neurons would be able to repolarize more rapidly. Beyond species differences in channels, differences in intracellular or extracellular regulation of Ca²⁺ levels by intracellular mechanisms or astrocytes (e.g. Morquette et al., 2015) could affect circuit properties.

Whether the intrinsic features of FTNs in X. petersii and X. victorianus can account entirely for differences in call duration and period remains to be determined. Features of network connectivity and synaptic strength might also play a role in species differences. In the stomatogastric nervous system, for example, altering synaptic strength in a reciprocally inhibitory two-cell circuit leads to large differences in burst duration and period (Sharp et al., 1996). In nematodes, differences in connectivity of homologous neurons result in distinct behaviors: Caenorhabditis elegans feed on bacteria using pharyngeal pumping while a related species, Pristionchus pacificus, uses its jaws to ingest prey. Homologous neurons produce these distinct behaviors but differ extensively in their connectivity (Bumbarger et al., 2013).

In many vocal vertebrates, respiration powers sound production, providing a potential link between hindbrain neural circuits for breathing and calling (e.g. Martin and Gans, 1972; reviewed in Leininger and Kelley, 2015). In terrestrial frogs such as Rana, vocalization is powered by expiration and requires activity in DTAM (aka the pretrigeminal nucleus; Schmidt, 1992, 1993). In Xenopus, aquatic anurans derived from terrestrial ancestors (Cannatella and de Sa, 1993; Irisarri et al., 2011), vocalization has been decoupled from respiration. Male X. laevis call while submerged, without respiration; glottal motor neurons that gate air movements from the lungs to the buccal cavity are inhibited during vocalization (Zornik and Kelley, 2008). A possible evolutionary scenario for the role of DTAM in vocal patterning is the repurposing of a neural circuit element, which functions during expiration in other vertebrates, to drive vocal patterning in the absence of actual breathing.

In the lamprey, a basal vertebrate, expiration is driven by the rhythm-generating paratrigeminal respiratory group (pTRG) (reviewed in Bongianni et al., 2014). The pTRG is intrinsically rhythmically active and glutamatergic, and projects to vagal motor neurons (Cinelli et al., 2013), features shared with the Xenopus DTAM vocal circuit (reviewed in Zornik and Kelley, 2017). Both the pTRG and DTAM are located in the rostral hindbrain compartment derived from embryonic rhombomere 1 (r1; Murakami et al., 2004; Morona and González, 2009). The several shared features described above support the homology of DTAM to the pTRG.

In mammals, however, rhythmic respiratory activity (specifically inspiration) is driven by more posterior hindbrain nuclei that include the parafacial respiratory group (pFRG) and the pre-Botzinger complex (preBotC; Thoby-Brisson et al., 2009). In the mouse, the pFRG forms in r4 and the preBotC in r7 (Tomás-Roca et al., 2016). Thus, although the lamprey pTRG has been considered a homolog of the preBotC (Cinelli et al., 2013), neither its location in r1 nor its prominent role in expiration (as opposed to inspiration, which is passive in lampreys; Bongianni et al., 2014) supports this assignment.

Is there a mammalian homolog of pTRG/DTAM that functions in expiration? The strongest candidate (Wetzel et al., 1985) is the Kolliker–Fuse nucleus of the parabrachial complex (PBC), which includes rhythmically active neurons linked to both inspiration and expiration (Dick et al., 1994). The PBC has been proposed to drive breathing patterns (Forster et al., 2014). Like neurons in DTAM and pTRG, neurons in the PBC originate in r1 (Tomás-Roca et al., 2016), are glutamatergic (Yokota et al., 2007), are rhythmic (Dick et al., 1994; Forster et al., 2014), and project to vagal and laryngeal motor neurons (nucleus ambiguus: Browaldh et al., 2015; the LMN). Thus, DTAM, pTRG and PBC are candidate vertebrate homologs. Neurons in the PBC are also active during vocalization in mammals (Farley et al., 1992; Jürgens, 2002), suggesting a highly conserved role for this hindbrain complex in both respiration and vocalization.

As we show here for the laevis clade, control of duration and period of the DTAM LFP wave could provide a conserved tuning mechanism across the Xenopus phylogeny. For example, within a different clade, the ability of male X. boumbaensis to produce a single sound pulse driven by a short train of CAPs could reflect shortening of the DTAM LFP duration to produce the very short CAP train observed in vitro (Leininger and Kelley, 2013). The major anuran suborders include the Archaeobatrachia (e.g. Xenopus) and the Neobatrachia (e.g. Rana) (Igawa et al., 2008; Roelants and Bossuyt, 2005). A DTAM LFP has also been observed during fictive calling in Rana pipiens (Schmidt, 1992), suggesting an ancient (Permian/Tertiary) role for the DTAM in vocal patterning within the Anura.

Small changes in neural circuits can cause behavior patterns to diverge so that even highly genetically similar species exhibit distinct patterns (Katz and Harris-Warrick, 1999). Here, we identified the circuit differences underlying distinct call patterns in closely related species of Xenopus. Reinforcement of differences between male courtship calls has been proposed as a driver for speciation in recently diverged frog lineages (Hoskin et al., 2005). Species differences in courtship patterns enable individuals to preferentially mate with members of their own lineage and avoid the reproductive costs of hybridization (Hoskin and Higgie, 2010). Behavioral studies have shown that advertisement calls in frogs serve as a strong premating isolation mechanism; in two-choice phonotaxis experiments, females strongly prefer a conspecific advertisement call over that of another species (Gerhardt and Doherty, 1988).

Related species can exhibit behavioral and neural preferences for different call features. For instance, female Hyla chrysoscelis use pulse rate to recognize mates while Hyla versicolor females use pulse duration. These species preferences are reflected in differences in selectivity of interval-counting neurons and long-interval selective neurons in the inferior colliculus (Schul and Bush, 2002; Rose et al., 2015; Hanson et al., 2016). The X. laevis inferior colliculus is also populated by auditory neurons that respond preferentially to sound pulse rate (Elliott et al., 2011) and auditory sensitivity to temporal features of sound pulses could differ in X. petersii and X. victorianus.

Thus, tuning of the duration and period of premotor activity in the rhythmically active premotor hindbrain nucleus, DTAM, is a likely source of the divergence of vocal patterns in the closely related species, X. laevis, X. petersii and X. victorianus. This mechanism is also a candidate for the control of vocal patterns in other Xenopus species as well as more distantly related Anuran species. Changes in DTAM activity of related populations resulting in vocalization differences might have served as pre-zygotic barriers that prevented gene flow and thus facilitated speciation.

We thank Sarah Woolley and Dustin Rubenstein for helpful input on the project, Ben Evans for clarification of phylogenetic terminology, the National Xenopus Resource Center at the Marine Biological Laboratory for hosting scientific consultations, Irene Ballagh for the brain schematic diagram in Fig. 3A, Alexandro Ramirez for input on data analyses and Albyn Jones for statistical consultation.