Gap junction mediated feed-forward inhibition ensures ultra precise temporal patterning in vocal behavior

Precise neuronal firing is especially important for behaviors highly dependent on the correct sequencing and timing of muscle activity patterns, such as acoustic signalling. We show that extreme temporal precision in motoneuronal firing within a hindbrain network that directly determines call duration, pulse repetition rate and fundamental frequency in a teleost fish, the Gulf toadfish, depends on gap junction-mediated, feed-forward glycinergic inhibition that generates a period of reduced probability of motoneuron activation. Super-resolution microscopy confirms glycinergic release sites contacting motoneuron somata and dendrites. Synchronous motoneuron activity can also induce action potential firing in pre-motoneurons, a feature that could figure prominently into motor timing. Gap junction-mediated, feed-forward glycinergic inhibition provides a novel means for achieving temporal precision in the millisecond range for rapid modulation of an acoustic signal and perhaps other motor behaviors.


47
Complex behaviors often depend on temporally precise neuronal firing that coordinates 48 network activity at brain levels ranging from cortical microcircuits to hindbrain pattern 49 generators (Kros et al., 2017;Llinás, 2014;Sober et al., 2018). Mechanisms known to 50 increase precision at single cell and network levels include feed-forward inhibition in audition 51 (Grothe, 2003), recurrent inhibitory input in cortex (Kapfer et al., 2007), and neuronal 52 synchrony in cortical and sensory neurons (Tiesinga & Sejnowski, 2001;Uhlhaas et al., 53 2010). Synchronous, concurrent activation of neurons is widely distributed in the brain 54 (Llinás, 2014) and especially important for behaviors requiring both rapid and precise 55 motoneuron activation such as electrogenesis in fishes (Bennett, 1971), and vocalization in 56 fishes (Chagnaud et al., 2012) and tetrapods (Kwong-Brown et al., 2019; Mead et al., 2017). 57 square pulse current injection that enhances synchronized activity in the vocal network. In situ 100 pharmacology shows the AHP is mediated by a glycinergic, feed-forward inhibition 101 dependent on gap junctional coupling; glycinergic release sites in contact with VMN somata 102 and dendrites are confirmed with super-resolution microscopy. Synchronous motoneuron 103 activity can also generate feedback activation of glycinergic VPN neurons. We propose that a 104 gap junction-mediated, feed-forward glycinergic inhibition provides a novel means to enhance 105 temporal precision in the activation of neural networks underlying acoustic signalling, and recorded intra-cranially from ventral occipital nerve roots (Fig. 1b, d). We refer to the highly-113 stereotyped pattern of nerve potentials that occur in the absence of vocal muscle activity as a 114 fictive vocalization (VOC). Each VOC potential directly reflects synchronous VMN activity 115 (Chagnaud et al., 2012). The rate (or frequency) and duration of VOC potentials directly 116 determine the rate and duration of vocal muscle contractions that set, in turn, call pulse 117 repetition rate and duration, respectively ( Fig. 1a) Skoglund, 1961). 120 VOCs occurred spontaneously or were evoked by brief trains of low amplitude, 121 electrical microstimulation in midbrain sites in a region comparable to the periaqueductal grey 122 of birds and mammals (Kittelberger & Bass, 2013). Like other toadfishes, both vocal nerves 123 fired in phase and individual VOC potentials were matched 1:1 with the activity of individual 124 VMN motoneurons (Fig. 1d) (Bass & Baker, 1990;Chagnaud & Bass, 2014). Motoneurons 125 had three to five main dendritic branches and an axon arising from a primary dendrite or soma 126 that lacked axon collaterals and exited the brain ipsilaterally via the vocal tract (VoTr, Fig. 1c) 127 (Bass & Baker, 1990;Chagnaud & Bass, 2014). 128 129

Midbrain-evoked vocal motoneuron physiology 130
Electrical midbrain stimulation led to membrane depolarizations in motoneurons that 131 increased in amplitude until a single action potential (AP) was fired (Fig. 2a1,2). This AP 132 coincided with the presence of a single VOC nerve potential. With increasing stimulus 133 strength, additional APs were detected that matched 1:1 with additional VOC potentials (Fig.  134 2a3,4,5) that together mimicked the pulse repetition rate of natural grunts (Fig. 1a) (Elemans et 135 al., 2014;Maruska & Mensinger, 2009;Tavolga, 1958;Winn, 1967). 136 Across repetitions, the amplitude of the second VOC potential always exceeded the 137 first (65.1 ± 15.3 µV vs. 36.6 ± 11.3 µV; n= 25 from five neurons; Mann-Whitney U test 138 p<0.001) (Fig. 2a3,4,5). The reduced amplitude of the first VOC potential reflected a less 139 synchronous and/or a partial activation of the motoneuron population. The first and last VOC 140 potentials generally showed activity distributed over a broader time course ( Fig. 2b; also see 141 inset, amplifying end of vocal nerve record, for example, of "weak" synchrony). The 142 amplitude of the first and last VOC potentials thus directly reflected the extent of synchronous 143 motoneuron activation. 144 Intracellular motoneuron recordings showed broad depolarizations often present 145 during the first and last VMN APs ( Fig. 2a3-5, b). These depolarizations often displayed 146 spikelets, strongly suggestive of asynchronous motoneuron activity ( Fig. 2a4, Fig. 2c). 159 The strong variability in amplitude of motoneuron APs and nerve potentials during a 160 given VOC (e.g., Fig 2a2-5) was further reflected by differences in half-width of the first 161 motoneuronal AP. Half-widths of the first APs riding on the aforementioned broad 162 depolarizations were significantly wider than those of subsequent APs whose amplitude 163 correlated with a higher amplitude in the corresponding VOC potential (first AP half width: 164 1.42 ± 1.07 ms; second: 0.72 ± 0.11 ms; n=40 from eight neurons; Mann-Whitney rank sum 165 test: p<0.001). Larger VOC potential amplitudes thus indicated more extensive synchronous 166 firing across the VMN population and correlated with narrower motoneuron APs. 167

Motoneuron electrical coupling and after-hyperpolarization (AHP) 169
To investigate the origin of different motoneuronal AP and AHP amplitudes observed for 170 antidromic-evoked APs and those during VOC activity, motoneurons were stimulated 171 antidromically at varying amplitudes via the ipsilateral vocal nerve root (ad-ipsi, blue traces; 172 Fig. 3a). At low amplitudes, we detected small depolarizations (Fig. 3a2) whose amplitude 173 gradually increased with stimulation strength, i.e., with increasing recruitment of motoneuron 174 axons and with shapes and peak latencies indicative of electrotonic coupling (3.7 ± 0.6 ms; 175 n=30 from six neurons analyzed). Collision experiments using antidromic activated and 176 intracellular evoked APs (via intracellular current injection) revealed these APs could not be 177 blocked, i.e., they resulted from electrical coupling (Kiehn & Tresch, 2002;Pappas & 178 Bennett, 1966) (Supp. Fig. 1). 179 With increasing amplitude of vocal nerve stimulation, the axon of the respectively 180 recorded motoneuron was eventually recruited and an antidromic AP invaded the recorded 181 motoneuron as shown by the significantly shorter peak latency (2.0 ± 0.3 ms after stimulation; 182 n=6) compared to the previously mentioned subthreshold depolarization (signed rank test: 183 p<0.001) and the faster rise time (Fig. 3a3). These APs showed no AHP, but instead were 184 characterized by a slow decay (double exponential fit; average time constant τ1: 0.85 ± 0.27; 185 time constant τ2: 7.45 ± 3.17; n=21 from five neurons) back to the resting membrane potential 186 (example shown in magenta trace in Fig 3a4). Surprisingly, an AHP started to appear with 187 increasing antidromic stimulation amplitude (Fig. 3a5). The AP and AHP peak amplitudes 188 increased with stimulation strength until each reached a plateau ( Fig. 3a6-8, c, d). An AHP was 189 never observed during intracellular square pulse current injections (Fig. 2d), raising the 190 question on the origin of this AHP. 191 Phase plane plots of the membrane potential during ipsilateral antidromic activation 192 revealed further changes in motoneuronal activity upon increasing stimulation amplitudes 193 ( Fig. 3a, black). The gradual appearance of the AHP, together with the absence of the AHP 194 upon initial AP firing (Fig. 3a3,4), suggested that a further recruitment of motoneurons via the 195 antidromic stimulation underlies AHP generation ( Fig. 3A5-8 (Eccles et al., 1954;Renshaw, 1941), could be 203 ruled out as the origin of the AHP given the lack of motoneuron axon collaterals (Fig. 1c) 204 (Chagnaud & Bass, 2014) and the short onset of the AHP. To exclude that a motoneuronal 205 axon collateral could arise at the periphery and enter via one of the nearby dorsal roots, the 206 dorsal roots were bilaterally cut in two experiments. There was no difference in AHP 207 amplitude (% of baseline) between cut and uncut recordings during VOCs (before cut: 17.04 208 ± 9.21, n= 150 from nine neurons; after: 15.19 ± 6.07, n=161 from nine neurons; Mann 209 Whitney U-test: p=0.23; Supp. Contralateral antidromic stimulation also revealed electrotonic potentials whose 214 amplitude depended on stimulation strength (ad-contra, red traces; Fig. 3a). Electrotonically 215 mediated potentials eventually reached threshold and evoked an AP. The peak latency of 216 these APs was significantly longer (3.71 ± 0.36 ms; n=39 for six neurons) than ones elicited 217 ipsilaterally ( Fig. 3b; Signed rank test: p=0.001), while the peak latency of the subthreshold 218 depolarization did not differ, consistent with their common origin from electrotonic coupling 219 (see above). As tract tracing and intracellular neuron fills showed that motoneurons only 220 innervate the ipsilateral muscle (Chagnaud & Bass, 2014), electrical coupling alone is thus 221 able to drive AP firing, independent of whether motoneurons belong to the ipsilateral or 222 contralateral VMN population. 223 As with the ipsilateral antidromic activation, an AHP component could clearly be 224 distinguished in the contralateral antidromic stimulation experiments ( Fig. 3a4-8). This AHP 225 occurred independent of AP firing, emphasizing the independence of the two events within a 226 given motoneuron. 227 Consistent with our findings in midshipman (Chagnaud et al., 2012), the ability to 228 initiate an AP via electrotonic coupling was in strong contrast to our intracellular current 229 injections that failed to initiate an AP in most cases, even at high current intensities (> 5 nA). 230 In cases where intracellular current injection elicited an AP, motoneurons showed rapid 231 adaptation of AP firing, likely due to weak somatic repolarization ability (see Fig. 2d). Gulf 232 toadfish are, however, able to contract their vocal muscles for several hundred milliseconds 233 (Fig. 1a). How can the muscle do this if motoneurons cannot fire for extended time periods 234 due to the rapid AP adaptation seen during square pulse current injections? To test whether 235 the AHP was required to de-inactivate motoneurons, we stimulated the motoneurons in which 236 current injection led to AP firing with pulse trains of different frequencies. In contrast to long 237 (> 50 ms) duration pulses (Fig. 2c), motoneurons showed no signs of AP adaptation to pulse 238 trains with brief (< 5 ms) pulses, indicating the necessity of membrane repolarization for 239 sustained motoneuron firing (Supp. Fig. 3). Stimulation was reliable into the behaviorally 240 relevant physiological range (the pulse repetition rate/ fundamental frequency of toadfish 241 vocalizations) with train frequencies tested up to 110 Hz. The weak repolarization capability 242 and low excitability of the motoneurons thus provide the means to prevent sustained AP 243 firing, which would decrease the extent of firing synchrony and precision across the VMN 244 population. 245 246

Network activity induces AHP 247
The presence of the AHP only at high antidromic stimulation amplitudes, i.e., high levels of 248 motoneuron recruitment, strongly suggested a network-dependent activation of the AHP. To 249 test this hypothesis, we ipsilaterally evoked an AP antidromically in a VMN motoneuron (ad-250 ipsi) at low threshold stimulation (i.e., without an AHP), followed by stimulation of the 251 contralateral nerve (ad-contra), which resulted in a small electrotonic depolarization (Fig. 4a,  252 b; also see supplemental movie 1). Subsequently, the delay of this second stimulation was 253 reduced up to the time point of the first ipsilateral nerve stimulation (Fig. 4b). Once close to 254 the antidromic AP, an AHP started to appear that increased in amplitude the closer the 255 contralateral evoked potential came to the ipsilateral evoked antidromic AP ( Fig. 4b; see heat 256 map in 4a). These experiments suggested that an increase in overall depolarization in the 257 vocal network is needed to generate the AHP. To test this hypothesis, we took advantage of 258 the prominent, wide depolarization that often appeared in motoneurons at the end of a VOC 259 (see Figs. 1d; 2a). Similar to the above, we moved an ipsi-or contralateral, antidromically 260 evoked depolarizing potential into this depolarization occurring at the end of a VOC (ad-ipsi 261 and ad-contra, Fig. 4c and d, respectively). With decreasing lag between the antidromically 262 evoked potential and the depolarization during the VOC, an AHP started to appear in the 263 contralateral evoked potential that increased in amplitude (Fig. 4c, d). This again showed the 264 necessity of a network-wide depolarization in order to elicit the AHP. 265 266

Necessity of gap junctional coupling for AHP activation 286
Having observed that the AHP is highly dependent on activation of the VMN population and 287 not on single neuron AP firing (Figs. 3-5a,b), we next blocked gap junctional coupling to 288 determine if the AHP originated from a network activation. A combined superfusion of the 289 exposed vocal hindbrain region coupled with pressure injection of carbenoxolone (CBX, a 290 gap junction blocker) directly into VMN severely impared the vocal network's ability to 291 generate synchronized motor discharges as evidenced by barely detectable VOC activity ( Fig.  292 5c, red trace). However, low amplitude VOC-related activity could still be detected in 293 intracellular recordings from motoneurons upon midbrain stimulation, showing that the vocal 294 network could still be activated (Fig. 5c). Upon antidromic stimulation (Fig. 5d), APs had a 295 similar amplitude as during control conditions, showing that the loss of vocal-related activity 296 during midbrain activation was not due to CBX impairment of motoneuron action potential-297 generating capacity, but to decreased electrotonic input to the motoneurons (% change from 298 baseline: control: -31.96 ± 9.31; CBX: -33.67 ± 9.68; n=54 from five neurons; Mann Whitney 299 U test: p=0.079). No AHP could be elicited in antidromic activated VMN neurons. This 300 showed that gap junctional coupling is indeed required to activate the AHP. 301 302

Dependence of AHP activation on inhibitory input to VMN 303
Having identified that gap junctional coupling was essential for the AHP, we next tested 304 whether inhibitory input could provide a source of this AHP. 305 Bicuculline injections into VMN. Injections of bicuculline, a potent GABA(A) receptor 306 antagonist, into VMN led to an increased rise time of the onset depolarization and in AP 307 amplitude during VOCs (Fig. 6 a, The amplitude of the antidromically evoked motoneuron AP increased (baseline % 317 change; control: -11.37 ± 7.08, n=74 from 5 neurons; bicuculline: -29.97 ± 12.61, n= 98 from 318 seven neurons; Mann Whitney U test: p<0.001), but no significant difference was found in the 319 antidromically evoked AHP (baseline % change; control: 12.50 ± 6.63, n=74 from 5 neurons; 320 bicuculline: 13.97 ± 10.13, n= 98 from seven neurons; Mann Whitney U test: p=0.306; Fig. 6a  control: 19.31 ± 7.14, n=62 from five neurons; strychnine: -1.15 ±4.65, n=48 from five 332 neurons; Mann Whitney U test: p<0.001; Fig. 6d, e). Similarly, and in contrast to the 333 bicuculline experiments, the AHP during antidromic stimulation was completely abolished 334 (baseline % change; control: 11.14 ± 5.89, =62 from five neurons; strychnine: 0.34 ± 0.79, 335 n=48 from five neurons; Mann Whitney U test: p<0.001; Fig. 6d inserts, e) and accompanied 336 by a longer depolarization of the antidromic AP (Fig. 6d inserts). These results suggested that 337 glycinergic neurons, activated via gap junctional coupling, were responsible for generating the 338 AHP during antidromic stimulation. 339 Following strychnine injections, spontaneous VOCs appeared with similar decreases 340 in motoneuron AHP amplitude not seen for spontaneous VOCs under control conditions 341 without prior strychnine injections (Fig. 6f). Since VMN motoneurons lack axon collaterals 342 ( Fig. 1c), the results imply that gap junction coupling between motor and glycinergic VPN 343 neurons is sufficient to drive AP firing of glycinergic neurons. 344 Typically, VOCs show sharp peaks and stable though sometimes variable intervals 345 between successive potentials, similar to the sound pulses within natural grunts (e.g., 346 (Maruska & Mensinger, 2009;McIver et al., 2014). This was also the case for spontaneous 347 VOCs following GABA injections (Fig. 6c) blocking inhibition was essential for the AHP, we next tested whether the effect of inhibition 361 was detectable in individual neurons. At sub-threshold midbrain stimulation levels to induce a 362 VOC compound potential, tonic membrane hyperpolarizations (on average, -2.69 ± 0.73 mV 363 below resting membrane potential; n=44 from five neurons) could be detected (Fig. 6a, gray  364 trace, subthreshold, blue trace suprathreshold), indicating the activation of inhibitory inputs. 365 To test whether these hyperpolarizations were artefacts originating from electrical stimulation, 366 we retracted the electrode from the intracellular space (Fig. 6g, orange trace). This abolished 367 the hyperpolarizing components. Due to the high synchrony of motoneuron APs during vocal 368 activity, field potentials could be detected in the nerve even with our high resistance 369 electrodes (Fig. 6g, black arrow). Changing the chloride reversal potential in the respective 370 neurons by intracellular chloride injections via 3M KCl-filled electrodes revealed a prominent 371 inhibitory input to motoneurons during VOCs as the membrane potential showed significant 372 changes in the degree of repolarization compared to baseline levels (Fig. 6h, blue arrow). The 373 AHP during VOCs was heavily reduced (Fig. 6h, blue trace; % change from baseline; 374 baseline: 21.2 ± 8.7; chloride injected: 2.6 ± 11.9; n=51 from six neurons; paired t-test: 375 p<0.001), thus indicating an inhibitory contribution to vocal behavior. Antidromic stimulation 376 still showed the AHP, however at a reduced amplitude (Fig. 6i;  coupling, as well as APs at higher stimulation amplitudes (Fig. 7b). Thus, antidromic 397 activation in motoneurons could elicit AP firing in VPN premotoneurons. The latency of these 398 APs (5.05 ± 0.48 ms, n=144 from 6 neurons) roughly coincided with the depolarization 399 following antidromically elicited APs in motoneurons (Fig. 3a, magenta arrow). 400 To test if a network component might also be important to activate VPN neurons, we 401 stimulated the ipsilateral and contralateral vocal nerve (at varying latencies) to antidromically 402 generate subthreshold membrane depolarizations in VPN neurons. A decrease in latency 403 between the ipsilateral and contralateral stimulation pulses eventually led to AP firing in VPN 404 neurons, thus showing that AP firing in VPN neurons also depends on gap junctional 405 activation of the vocal network (Fig. 7c). VPN activation likely led to motoneuron 406 depolarization (Fig. 3, magenta arrows) (Capranica, 1992), the vocal and auditory systems "co-evolved and we should expect them to 540 share the same underlying code for signal generation and recognition". 541 From an environmental perspective, transmission distance is limited by the frequency 542 content of calls in shallow water habitats like those where toadfishes defend nests and engage 543 in acoustic courtship (Bass & Clark, 2003;Fine & Lenhardt, 1983;Gerald, 1971). The 544 principal limitation to call PRR and fundamental frequency in species like toadfishes that 545 generate calls with frequency content mainly ≤500 Hz is the contraction rate of muscles that, 546 in this case, drive swim bladder vibration. To overcome such limitations and enhance 547 transmission distance, toadfishes adopted superfast muscles (Elemans et