Glycinergic axonal inhibition subserves acute spatial sensitivity to sudden increases in sound intensity

Locomotion generates adventitious sounds which enable detection and localization of predators and prey. Such sounds contain brisk changes or transients in amplitude. We investigated the hypothesis that ill-understood temporal specializations in binaural circuits subserve lateralization of such sound transients, based on different time of arrival at the ears (interaural time differences, ITDs). We find that Lateral Superior Olive (LSO) neurons show exquisite ITD-sensitivity, reflecting extreme precision and reliability of excitatory and inhibitory postsynaptic potentials, in contrast to Medial Superior Olive neurons, traditionally viewed as the ultimate ITD-detectors. In vivo, inhibition blocks LSO excitation over an extremely short window, which, in vitro, required synaptically-evoked inhibition. Light and electron microscopy revealed inhibitory synapses on the axon initial segment as the structural basis of this observation. These results reveal a neural vetoing mechanism with extreme temporal and spatial precision and establish the LSO as the primary nucleus for binaural processing of sound transients.


30
A key component of the neuron doctrine is the unidirectional propagation of action potentials, 31 formulated as the "law of dynamic polarization" by Cajal and van Gehuchten (Berlucchi, 1999;32 Shepherd, 1991). As the site where action potentials are typically initiated, the axon initial 33 segment (AIS) has a pivotal role in this process (Bender and Trussell, 2012; Kole and Brette,34 2018; Leterrier, 2018) and is a bottleneck where inhibition can have an "outsized" effect on a 35 neuron's output, as proposed for chandelier and basket cells (Blot and Barbour, 2014;Nathanson 36 et al., 2019). Disruption of such synapses is associated with severe brain disorders (Wang et al.,37 2016), but their exact functional role in the normal brain is speculative because physiological 38 studies of these synapses have been limited to in vitro recordings. Even the basic physiological 39

Sharp ITD-sensitivity to clicks in LSO but not MSO 87
We obtained in vivo whole-cell recordings while presenting clicks at different ITDs in 19 LSO 88 neurons and 11 MSO neurons. Responses to tones for these cells have been reported before 89 . We were surprised to find sharp sensitivity to ITDs of clicks in LSO 90 but not MSO neurons. Figure 1A shows sensitivity to ITDs of transients in a principal LSO neuron 91 excitation is unopposed and reliably triggers a single spike. Likewise, at large positive ITDs, the 106 leading contralateral (inhibitory ear) click is not able to suppress spiking to the lagging ipsilateral 107 click, even for lags between IPSP and EPSP as small as 0.25 ms. Figures 1C and 1D show data 108 for a non-principal LSO neuron (IID function in Figure 1-figure supplement 1B): here the 109 intracellular traces are more complex than a stimulus-like stacking of PSPs, but nevertheless 110 tuning to ITDs is present, be it with shallower slopes (-1.0 and 2.0 spikes per click/ms), wider 111 trough (halfwidth 1350 s), and higher variability, yielding an ITD-SNR of 0.63. Figures 1E and  112 1F show data for an MSO neuron. As expected, the main feature in the response is an excitatory 113 peak near 0 ms. Even though this is one of the steepest-sloped ITD-functions of our MSO sample 114 (2.1 spikes per click/ms for slope at ITD < 0ms), the ITD-tuning lacks the acuity observed in 115 principal LSO cells, with an ITD-SNR of only 0.38. The intracellular data ( Figure 1F) reveal that, 116 surprisingly, spiking is not restricted to ITDs where the two events coincide, but also occurs at 117 other ITDs, where the click at either ear can elicit a suprathreshold response.  classical role of MSO neurons as "ITD detectors", LSO neurons show superior ITD-tuning for  136 transient sounds. We also tested LSO neurons with broadband noise, which has a flat amplitude 137 spectrum like clicks but with a random phase spectrum. ITD-sensitivity to noise was generally 138 weak ( Figure 1-figure supplement 3A), which was also the case for responses to dynamic 139 interaural phase differences in modulated or unmodulated pure tones (   Legend in K applies also to J and L. 183 The online version of this article includes the following figure supplements for    pairs. 238

Effective inhibition of LSO neurons is limited to a short initial part of the IPSP 239
Prior to our recordings, published LSO in vivo intracellular recordings were limited to a few traces 240 (Finlayson and Caspary, 1989). To gain insight into the sharp ITD-tuning in LSO and its lack in 241 MSO, we compared intracellular synaptic responses to monaural and binaural clicks from these 242 neurons (Figures 2 and 3). As illustrated for 2 LSO neurons (Figures 2A and 2B), they receive a 243 well-timed EPSP in response to monaural ipsilateral clicks, which reliably trigger spikes, and a suggesting that consistent, precise response timing is already present at the presynaptic level. 256 Strikingly, the IPSP duration extends to almost 5 ms in principal cells (Figure 2A), close to an 257 order of magnitude larger than the halfwidth of the tuning function to ITDs ( Figure 1A and 2D). 258 This is consistent with in vitro data (Sanes, 1990;Wu and Kelly, 1992), where the effective window 259 of inhibition was also reported to be much smaller than the IPSP duration. The availability of the 260 monaural responses allows us to examine this window. Figure 2E shows comparisons of binaural 261 and monaural click responses, for a principal LSO neuron (ITD-function in Figure 2D). In each 262 panel, the intracellular response is shown at one ITD (black traces), with the monaurally-recorded 263 responses to ipsi-(blue) and contralateral clicks (green) superimposed, incorporating the stimulus 264 ITD. At large negative delays ( Figures 2E1 and 2E2), the leading EPSP reliably triggers spiking, 265 unhindered by the ensuing IPSP. More surprisingly, when the IPSP leads and significantly 266 overlaps with the EPSP (Figure 2E6), it also fails to inhibit spiking. Only when the early steep 267 slope of the IPSP coincides with the early steep slope of the EPSP are spikes completely blocked 268 ( Figure 2E4). Comparison of binaural responses for a fuller range of ITDs with the monaural IPSP, 269 is shown in Figure 2F. Responses from large negative to large positive ITDs reveal the 270 exceedingly narrow range of ITDs over which spikes are suppressed, near the onset of the IPSP. 271 monaural spike rates to clicks were substantial and could even equal spike rates to binaural clicks 305 (here delivered at ITD = 0 ms, generating a spike rate >90% of the peak of the click ITD function). 306 We calculated the summation ratio (Goldberg and Brown, 1969), i.e. the ratio of the spike rate to 307 binaural stimulation to the sum of monaural responses, where values > 1 indicate facilitation, as 308 expected for a coincidence detector. MSO summation ratios in response to clicks ( Figure 3F, 309 magenta) were all < 1.3 and had a median of 0.97, indicating that the binaural response rates are 310 similar to the sum of monaural response rates. In contrast, MSO summation ratios to tones were 311 substantially higher than those to clicks ( Figure   In vitro recordings reveal powerful inhibition for synaptically-evoked but not for simulated 362

IPSPs 363
To better understand the narrow, sub-millisecond ( Figures 1A, 1K   plotted against the ITD-SNR (as in Figure 1L). Data from principal LSO cells recorded in vivo are 407 shown for comparison (13 data sets from 7 cells). N.D.: not defined (slope was not defined when 408 20% of maximal spike rate was not reached after smoothing (see Materials and Methods)). 409 The online version of this article includes the following figure supplement for Figure 4:

LSO, but not MSO, neurons have inhibitory innervation of the axon initial segment 436
The connectivity of LSO neurons has been extensively studied (Cant, 1991 That the actual inhibitory synaptic input is more powerful than somatic current injection suggests 444 a specialization distal from the soma, possibly the AIS. To visualize the spatial pattern of 445 glycinergic terminals on LSO neurons, we immunostained the LSO for gephyrin and ankyrin G, 446 markers for postsynaptic glycine receptors and the scaffolding of the AIS/nodes of Ranvier, 447 respectively ( Figure 5). We additionally stained for DAPI or for synaptophysin-1 (SYN1) to 448 visualize somata or synaptic boutons, respectively. We analyzed samples from the mid-frequency 449 region of the LSO (Figures 5A and 5B), where electrophysiological recordings were typically 450 made, selecting neurons with a complete, relatively planar AIS that could be unambiguously 451 connected to an axon hillock and soma ( Figures 5C-H). A high density of gephyrin-positive puncta 452 covered the soma and proximal dendrites. In several cases, gephyrin-positive puncta could also 453 be seen extending onto the axon hillock ( Figure 5, filled yellow arrowheads), and/or along the AIS 454 itself ( Figure 5, open yellow arrowheads). Clear overlap of gephyrin-positive puncta with a putative 455 synaptic terminal labeled by synaptophysin-1 on an interrupted AIS was sometimes seen (Figure  456 5F, open white arrowhead). 457 To obtain conclusive proof of innervation of the AIS, we performed electron microscopy (EM) on 458 3 principal LSO neurons labeled with biocytin. Figure 6A shows a camera lucida drawing of a 459 principal LSO neuron, with indication of parts of the axon that were examined with EM. A section 460 at a distance of several tens of µm from the soma, shows the myelinated axon ( Figure 6B). A 461 section through the AIS shows indeed 3 synaptic profiles ( Figure 6C, enlarged in Figure 6D1-3). 462 The same was true for the two additional principal LSO neurons (Figure 6-figure supplement 1A). shown in the electron micrographs with the same numbers. 1 illustrates the initial segment (IS) 499 close to the cell body and three synapses colored for clarity, 2 the myelinated axon approximately 500 70 µm from the cell body and 3 two pieces of the dendrite (d) at the locations pointed to by the 501 arrows as well as the axon (a) initial segment seen enlarged in 1. Scale bar in 1 applies to 1 and 502 2. In addition to this cell and the one shown in Figure 6, we observed similar synaptic terminals in 503 a third LSO principal cell (data not shown). (B) Absence of innervation of the initial segment of an 504 MSO principal cell. Top middle: camera lucida drawing of an MSO principal cell that was 505 intracellularly recorded from and labeled in vivo, and its location in a coronal section of the MSO. 506 Numbers and arrows point to locations on the axon shown in the electron micrographs with the 507 same numbers. 1 illustrates the initial segment (asterisk) close to the cell body, 2 the transition 508 from initial segment to myelinated axon approximately 20 µm from the cell body and 3 the 509 myelinated axon approximately 50 µm from the cell body. A second MSO principal cell that was 510 processed for EM also did not show innervation of the axon initial segment (data not shown). 511 512

LSO neurons show graded latency-intensity changes which disambiguate spatial tuning 513
It has been hypothesized that temporal specializations in the LSO-circuit evolved to generate 514 feature: indeed for at least a sizable fraction of neurons, it is the "right" slope that is closest to 0 525 ms ( Figure 7B, function 1). For cases where the ITD-function is centered near 0 ms ( Figure 7B, 526 function 2, example in Figure 7L (cyan)), there is an additional issue of ambiguity: a rise in spike 527 rate could signal both a leftward or rightward change in horizontal position of the sound source. 528 A similar problem occurs at the population level if some neurons have the "left" slope near 0 and 529 others the "right" slope. However, a natural and elegant solution to these issues is directly 530 embedded in the properties of the LSO circuit. 531 Figures 7D and 7E show how PSPs change with sound intensity for a principal and a non-principal 532 cell. In the principal neuron, the changes in both EPSP and IPSP are extremely reproducible and 533 finely graded in amplitude and latency with increasing SPL, also for individual trials (Figure 7-534 figure supplement 1A). In the non-principal neuron, the changes are complex, with multiple events 535 following each click and a leading IPSP at high intensities. The latency changes are sizeable 536 compared to the relevant ITD range for the animal: they show a steady decrease which is 537 approximately linear over the 30-dB range tested, with a slope amounting to ~10-20 µs/dB 538 ( Figures 7F and 7G). 539 In real-world environments, IIDs and ITDs co-occur and are correlated (Gaik, 1993).  binaural change here is in the relative timing of these fixed PSPs. This is different for changes in 546 IID only (ITD fixed at 0 ms), for which pairs of PSPs are shown in Figure 7H (IIDs of -20, 0, and 547 +20 dB). As expected, the changes in level affect the amplitude of the PSPs, but they also have 548 a large, clear effect on latency: the latency differences between onset of EPSP and IPSP are 549 actually larger than the ITDs (±0.3 ms) imposed in Figure 7I. This results in a nonlinear interaction 550 when both cues are combined, causing a marked functional change in the tuning function ( Figure  551 7J). For the cue combination favoring the ipsilateral ear (both cues < 0; Figure 7J, left panel), the 552 large and early EPSP is not effectively opposed by the small and later arriving IPSP: this results 553 in a higher probability of spiking than for ITD or IID alone. For the combination favoring the 554 contralateral ear (both cues > 0) ( Figure 7J, right panel), a large and leading IPSP opposes a late 555 and small EPSP: this results in a lower probability of spiking than for ITD alone. The effect of cue 556 combination is therefore to remove the "right" slope of ITD-tuning, and to generate a steep "left" 557 slope closer to 0 ITD ( Figure 7C). 558 This is illustrated ( Figure 7K In summary, striking specializations at 3 levels combine to make LSO principal cells spatially 570 tuned to transient sounds. Exquisite timing in afferent inputs supplies these neurons with 571 temporally punctate events; the intrinsic properties of the neurons enable these events to interact 572 at a sub-millisecond timescale; and the opposite sign and strategic location of the inputs enable 573 input from one ear to veto the input from the other ear. The net result is sharp tuning to sound 574 transients, which moreover is coherent with IID-tuning to sustained sounds in non-principal cells.

627
Our data lead to a new view of brainstem binaural processing, departing strongly from the 628 previously accepted roles of the MSO as a timing comparator and the LSO as an intensity 629 comparator. We find that both excel as timing comparators, be it for different types of sounds, 630 complementary in frequency range and temporal characteristics. Our data show that principal 631 LSO cells are significantly more temporally specialized than was previously appreciated, towards 632 one specific, highly ecologically relevant form of ITD-sensitivity which has received little attention: Traditionally, the LSO is viewed as the brainstem nucleus underlying behavioral sensitivity to IIDs. 640 A long-standing problem with this depiction is that it lacks a rationale for the extreme features of 641 the LSO-circuit, which hinder, rather than help IID sensitivity and which suggest a key role for 642 timing. These features include large axosomatic synapses such as the calyx of Held, differential 643 axon diameters on ipsi-and contralateral side, and fast membrane properties of monaural inputs 644 (Joris and Trussell, 2018). Despite these features, ITD-sensitivity of LSO neurons is weak (Caird 645 and Klinke, 1983;Joris, 1996;Joris and Yin, 1995;Tollin andYin, 2005, 2002), except to sound 646 transients as documented in vivo for a limited number of neurons (Caird and Klinke, 1983;Irvine 647 et al., 2001;Joris and Yin, 1995;Park et al., 1996) and in vitro with bilateral electrical shocks 648 (Sanes, 1990;Wu and Kelly, 1992). It was recently argued that spatial sensitivity to high- that have been undersampled in extracellular studies , the data underscore 655 that temporal aspects of binaural sensitivity are an essential feature of this nucleus. 656 LSO neurons show acute tuning to ITDs of transient stimuli to an extent that surpasses that of 657 neurons in the MSO, which is classically regarded as the nexus of ITD-sensitivity (Figures 1-3). 658 Intracellular traces to monaural stimulation reveal the presence of extraordinarily well-timed 659 excitation and inhibition in LSO neurons (Figure 2). We discovered a "prepotential" (Figure 2A, 660 In response to binaural stimulation, inhibition is remarkable in its depth, temporal acuity, reliability, 663 and limited duration of its effect. Effective interaction between EPSP and IPSP occurs over a time 664 window which is only a small fraction of the latter's duration, and generates steeply-sloped and 665 narrow ITD-tuning (Figures 1 and 2). Application of inhibition in vitro by somatic conductance 666 clamp (Figure 4) or current injection (Figure 4-figure supplement 1), was ineffective to completely 667 suppress spiking, as opposed to synaptically driven inhibition. This suggested that at least some 668 synaptically-evoked inhibition acts electrotonically closer to the spike initiation region in the axon. 669 Indeed, morphological examination at the light ( Figure 5) and EM ( Figure 6) level revealed 670 glycinergic terminals at the AIS of LSO but not MSO neurons (Figure 6-figure supplement 1B). 671 It has often been proposed that IIDs are translated to ITDs through a peripheral latency 672 mechanism (the "latency hypothesis" (Jeffress, 1948)). Response latency generally decreases 673 with sound level, so an acoustic IID would generate a neural ITD pointing to the same side. Human 674 psychophysical studies do not support a simple IID-to-ITD conversion for low-frequency, ongoing 675 sounds (Domnitz and Colburn, 1977). Indeed, for such sounds, IIDs are small (Maki and 676 Furukawa, 2005), and the relationship between intensity and latency is complex (Michelet et al., 677 2010). However, EPSPs and IPSPs show large and systematic latency changes in response to 678 transient sounds (Figures 7D-7G). Physiological evidence for an interaction between IID and ITD 679 has been observed for transient responses in a variety of species and anatomical structures 680 (Irvine et al., 2001;Joris and Yin, 1995;Park et al., 1996;Pollak, 1988;Yin et al., 1985), but in 681 these extracellular recordings the underlying cellular mechanisms could not be assessed. Our  and identical for monaural stimulation (of either ear) and binaural stimulation, so that little 705 response increment is gained with binaural stimulation. In contrast, in LSO neurons the response 706 sign is opposite for the two ears and maximal binaural interaction is obtained when ITD causes 707 an alternation in sign so that the response is fully modulated between the excitation to monaural 708 ipsilateral stimulation and inhibition to contralateral stimulation. In this subtractive mechanism, 709 strong monaural responses (of opposite sign) yield maximal binaural interaction. 710 In conclusion, the LSO pathway is not a simple IID pathway but consists of at least two 711 The methods for in vivo and in vitro patch clamp recording and electron microscopy have been 737 previously described (Franken et al., , 2016 and are briefly summarized here. 738 739

Surgery for in vivo electrophysiology 740
The animals were anesthetized by an intraperitoneal injection of a mixture of ketamine (80-120 741 mg/kg) and xylazine (8-10 mg/kg) in 0.9% NaCl. Anesthesia was maintained by additional 742 intramuscular injections of a mixture of ketamine (30-60 mg/kg) and diazepam (0.8-1.5 mg/kg) in 743 water, guided by the toe pinch reflex. Body temperature was kept at 37°C using a homeothermic 744 blanket (Harvard Apparatus, Holliston, MA, USA) and a heating lamp. The ventrolateral brainstem 745 responses were obtained to binaural clicks for which IID was varied. Positive ITDs and positive 796 IIDs refer to stimuli for which respectively the contralateral stimulus leads the ipsilateral stimulus 797 or the contralateral stimulus is more intense than the ipsilateral stimulus. measuring slope steepness ( Figure 1J) and halfwidth ( Figure 1K). 805 To quantify the modulation of spike rate as a function of ITD ( Figure 1L) we used the ITD-SNR 806 metric which has been described by Hancock  where ̅ stands for the mean spike count across trials to a stimulus with sound level . 828 is then calculated as the median ratio across sound levels. Because LSO neurons are excited by 829 ipsilateral sounds but inhibited by contralateral sounds, a strong binaural effect means that the 830 response to binaural stimuli is a lot smaller than the sum of monaural ipsilateral and monaural 831 contralateral responses, and this will result in large values of . If instead binaural stimulus 832 presentation results in the same average spike count as the sum of monaural ipsilateral and 833 monaural contralateral stimulation, will be equal to 1. Since MSO neurons are excited by monaural ipsilateral as well as monaural contralateral sounds, 841 a strong binaural effect will result in ̅ , being much larger than the sum of ̅ , and 842 ̅ , . Inverting the ratio in the definition of compared to thus means that both 843 metrics are >> 1 when there is a significant binaural advantage compared to monaural stimulation. NaH2PO4, 1.5mM MgSO4, 1.5mM CaCl2, 5mM N-Acetyl-L-Cystine, 5mM Sodium ascorbate, 856 3mM sodium pyruvate, and 2mM Thiourea (pH adjusted to 7.45 with NaOH, final osmolarity: 310 857 mOsm). Following 30-45 minutes of recovery, slices were maintained at room temperature for 858 >30 minutes before recording. 859 Whole-cell current-clamp recordings were made using Dagan BVC-700A amplifiers. Voltage data 860 was filtered at 5 kHz, digitized at 100 kHz, and stored on computer using in-house custom 861 software written in IGOR-Pro (Wavemetrics). Recording electrodes were pulled from borosilicate 862 glass (1.5mm OD; 4-8MΩ) and filled with intracellular solution containing 115 mM K-gluconate, 863 4.42 mM KCl, 0.5 mM EGTA, 10 mM HEPES, 10 mM, Na2Phosphocreatine, 4 mM MgATP, and 864 0.3 mM NaGTP, osmolality adjusted to 300 mOsm/L with sucrose, pH adjusted to 7.30 with KOH. 865 All recordings were carried out at 35°C with oxygenated ACSF perfused at a rate of ~2-4 mL/min, 866 and bridge balance and capacitance compensation were monitored throughout. All membrane 867 potentials shown are corrected for a 10 mV liquid junction potential. The peak conductance of IPSGs was adjusted so that an individual event elicited a 5-10 mV 875 hyperpolarization from the resting potential. Synaptic stimuli were evoked through glass pipettes 876 (50-100 µM dia.) via a constant current stimulator (Digitimer DS3), and presented with random 877 temporal offset intervals Small current steps were interleaved to monitor input resistance. 878 Synaptic stimulation was ipsilateral to the LSO for excitatory input stimulation, or near the center 879 of the MNTB for inhibitory stimulation. Excitatory and inhibitory responses were isolated through 880 the inclusion of 1 µM strychnine or 10 µM NBQX to the bath, respectively. Stimulation intensity 881 was also adjusted so that action potential probability at optimal synaptic timing was close to, but 882 less than 100%, to avoid saturation. Tissue on slides were rehydrated in 0.1M PBS for 5-10 min. Sections were blocked and 894 permeabilized with PBTGS (10% Goat Serum, 0.3% Triton in 0.1M PBS) for 1.5 hours in a 895 humidity chamber at room temperature on a slow-moving shaker. The tissue was then incubated 896