NEW ROLES FOR SYNAPTIC INHIBITION IN SOUND LOCALIZATION

Transcription

1 NEW ROLES FOR SYNAPTIC INHIBITION IN SOUND LOCALIZATION Benedikt Grothe The arrival times of a sound at the two ears are only microseconds apart, but both birds and mammals can use these interaural time differences to localize low-frequency sounds. Traditionally, it was thought that the underlying mechanism involved only coincidence detection of excitatory inputs from the two ears. However, recent findings have uncovered profound roles for synaptic inhibition in the processing of interaural time differences. In mammals, exquisitely timed hyperpolarizing inhibition adjusts the temporal sensitivity of coincidence detector neurons to the physiologically relevant range of interaural time differences. Inhibition onto bird coincidence detectors, by contrast, is depolarizing and devoid of temporal information, providing a mechanism for gain control. SENSORY SYSTEMS INTERAURAL TIME DIFFERENCE (ITD). The difference in the arrival time of a sound at the two ears. Depending on the position of the sound source and the individual inter-ear distance, ITDs can be up to about 120 µs in the Mongolian gerbil, and up to about 650 µs in humans. Max-Planck-Institute of Neurobiology, Auditory Processing Group, Am Klopferspitz 18a, D Martinsried, Germany. bgrothe@neuro.mpg.de doi: /nrn1136 Just over 200 million years ago, archosaurs the ancestors of birds evolved tympanic ears (including a tympanic membrane and a middle ear), allowing them to hear and to localize airborne sounds. Some million years later, early mammals independently evolved tympanic ears 1,2 (FIG. 1). The subsequent evolution of sophisticated auditory systems to allow the elaborate analysis of airborne sound occurred completely independently in birds and mammals, although it was driven by the same evolutionary constraints, including the limited number of physical cues that are available to compute the position of a sound source. One of these cues is the tiny differences in the arrival time of a sound at the two ears, known as INTERAURAL TIME DIFFERENCES (ITDs; FIG. 2a). The neural circuits that encode these ITDs in birds and mammals exemplify how the evolution of the vertebrate brain came up with different solutions for a single problem. For both birds and mammals, ITD processing represents the ultimate challenge in temporal processing. No other neural system in vertebrates comes close to the temporal resolution required to encode ITDs, except for the electric system in some fish. Consider that the duration of an action potential is almost two orders of magnitude greater than the minimal ITDs that barn owls or humans can resolve (<10 µs). Particular structural and functional adaptations were necessary to allow the successful processing and encoding of these ITDs. This fact makes the ITD-processing structures in birds and mammals attractive model systems for studying the rules that underlie precise temporal processing in the vertebrate brain in general, and it has stimulated many experimental studies and computational models. The current textbook view of ITD processing resembles closely what Jeffress proposed more than 50 years ago in his elegant model 3. This model was based on three main assumptions (BOX 1): bilateral, TIME-LOCKED or PHASE-LOCKED inputs into the ITD-processing system; COINCIDENCE DETECTION by ITD detector neurons; and an arrangement of delay lines to adjust coincidence detector neurons to different preferred ITDs, creating a topographic representation of AZIMUTHAL SPACE. The bird ITD-encoding system seems to have evolved in a way that closely matches Jeffress s predictions (FIG. 3a,b). Neurons in the nucleus laminaris, the first NATURE REVIEWS NEUROSCIENCE VOLUME 4 JULY

2 Quaternary Gymnophiona Urodela 1.8 Anura Sphenodontidae Testudines Squamata Crocodilia Aves Mammalia Cenozoic Tertiary 65 Ichthyosauria Plesiosauria Pterosauria Dinosauria Cretaceous Mesozoic Jurassic Palaeozoic Triassic 250 Permian 290 Carboniferous 360 Early amphibians 'Stem reptiles' Thecodontia Pelycosauria Amphibians Therapsida Archosaurs Devonian 410 Rhipidistia Reptilians Appearance of tympanic ear Figure 1 The evolution of tympanic ears. The evolution of tympanic ears (tympanum and middle ear), a prerequisite for localizing airborne sound, occurred independently and almost simultaneously in several clades of tetrapods. During the Triassic period, tympanic ears evolved in frogs (Anura), several lines of reptiles, including those leading to archosaurs and later from archosaurs to birds, and in early mammals. So, there is no common ancestor of birds and mammals that had the anatomical substrate for localizing airborne sounds by means of interaural comparison. The neural system for interaural time difference processing evolved independently during parallel evolution in birds and mammals. Number show millions of years ago. Reproduced, with permission, from REF. 105 (1994) Saunders College Publishing. TIME-LOCKED Action potentials of many auditory neurons are locked to specific events of acoustic stimuli, such as onsets, offsets, prominent fluctuations in frequency or amplitude, or even a specific phase-angle of sinusoidal low-frequency sounds. PHASE-LOCKED The most extreme case of timelocking in auditory neurons. Many low-frequency neurons (in mammals up to a few khz, in barn owls up to 8 khz) synchronize their discharge to a specific phase-angle of tones. COINCIDENCE DETECTION The activation of neurons not by single inputs, but only by the simultaneous activity of several inputs. Coincidence detector neurons can be found throughout the nervous system. The most extreme case of coincidence detection is found in the binaural auditory system where the time windows for coincidence detection are in the range of microseconds. AZIMUTHAL SPACE The definition of auditory space independent of the elevation of a sound source. The task of localizing a sound in azimuthal space is often referred to as 'lateralization'. station of binaural processing in the bird auditory brainstem, act as coincidence detectors 4 8 and receive their excitatory inputs through axons of systematically varying lengths and with systematically varying interaural delays 5,9 12. As a consequence, neurons that are tuned to the same characteristic (best) frequency show different best ITDs (the ITD at which a neuron fires maximally), ITD Figure 2 The interaural time difference (ITD). The difference in the time of arrival at the two ears, the ITD, is the main cue for localizing low-frequency sounds. If a sound source is straight ahead, the ITD is zero. If a sound source comes from one side, the sound will reach the ear on that side first, creating an ITD. Depending on the head size (the distance between the ears), ITDs are not greater than 100 ms for the gerbil, but can be up to several hundred microseconds for larger animals. Lowfrequency hearing mammals, such as the Mongolian gerbil, evolved ITD detection to avoid predators such as the eagle owl. Reproduced, with permission, from Nature Neuroscience REF. 106 (2002) Macmillan Magazines Ltd. covering roughly the entire physiologically relevant range 13 (FIG. 3a,b). Accordingly, there is evidence for a systematic representation of azimuthal space at higher stations in the owl auditory system, the external nucleus of the auditory midbrain and the optic tectum. There, the best ITD of the neurons systematically shifts with the position of the neurons, from rostral to caudal However, our understanding of the bird ITD-encoding system has to be extended by one important factor: as I will discuss later, GABA (γ-aminobutyric acid)-mediated inhibition operates as a differential gain control system to keep the coincidence detector in an appropriate working range and seems to compensate for other binaural influences. The mammalian auditory system has also long been considered as a Jeffress-type system The evidence to support this view included the connection patterns with excitatory inputs from both sides (details provided later), the cyclic nature of the ITD sensitivity when tested with pure tones (BOX 1), and the fact that the best ITD could be predicted on the basis of the time delay of the response to the monaural inputs from both ears 17, Last but not least, results from many human psychophysical studies can be explained by the concept of binaural crosscorrelation, as indicated by the Jeffress model 23,24. However, increasingly conflicting evidence began to raise considerable concerns about the validity of the Jeffress model for the mammalian ITD encoder. A careful anatomical reconstruction of single neurons that project to the mammalian ITD detector, the medial superior olive (MSO), revealed that far more anatomical projection patterns did not fit the concept of delay lines than did (see figure 6 in REF. 25; only 3 out of 16 MSO inputs resemble Jeffress-type delay lines. Note that the authors of REF. 25 did not interpret their data to cast doubt on the Jeffress model). Reconstructions from extracellular 2 JULY 2003 VOLUME 4

3 Box 1 The Jeffress model of interaural time difference (ITD) processing a b Left ear Time-locked response response Stimulus waveform Stimulus waveform Time Delay line Delay line response Ipsilateral leading Contralateral leading ITD (cycles) Stimulus waveform Right ear The current textbook view of ITD processing is based on the seminal model put forward by Jeffress in 1948 (REF. 3).It has three fundamental assumptions. First, the temporal pattern of an acoustic stimulus is preserved in the firing pattern of the excitatory projections to the ITD-encoding structure. For low-frequency stimuli up to a few khz (which we mainly localize by means of ITDs) this is mainly achieved by phase-locking action potentials (red lines in upper panel of a) only occur correlated with a specific phase-angle of the sinusoidal stimulus waveform. Alternatively, action potentials could be time-locked to the stimulus onset, or to prominent changes in amplitude or frequency. Second, the ITD-encoding neurons receive such excitatory, time-locked inputs from both ears, and fire maximally when the action potentials from the two sides arrive simultaneously. So, the ITD-processing neurons act as coincidence detectors that show a sinusoidal ITD sensitivity (lower panel in a) in response to pure tones (maxima are correlated with in-phase, minima with out-of-phase occurrence of the inputs). Depending on the stimulus frequency, and therefore the duration of one stimulus cycle and the correlating discharges, the cyclic ITD functions are narrower (for high frequencies; light green) or wider (for low frequencies; dark green). Third, the excitatory inputs are arranged as delay lines that project to an array of neurons in the ITD detector that all respond to the same stimulus frequency (b). The axonal length defines the conductance time that an action potential needs to travel to a coincidence detector neuron, introducing a specific, fixed delay. Symmetrical axonal inputs from both sides (the neuron in the middle of the vertical array of coincidence detector neurons) would provide coincident inputs only when the stimulus is straight ahead, reaching both ears simultaneously (0 ITD; see function below the neuron). A neuron with a short delay line from the left, but a long delay line from the right ear (top neuron in the array) would respond maximally when an ITD compensates for the mismatch in delays from the two ears. So, it would respond maximally to stimuli in the right hemifield. By contrast, short delays from the right and long delays from the left ear would tune a neuron to respond maximally to stimuli in the left hemifield. A systematic arrangement of these delay lines (as shown) would create a systematic map of best ITDs, and so a topographic map of azimuthal space. tracer injections provided only weak evidence for such delay lines 26. More importantly, despite many attempts, no ITD map has been convincingly shown in the mammalian auditory system, including the cat auditory cortex 27. Most recently, evidence that is in considerable conflict with the Jeffress model arose from studies in the rabbit brainstem 28 and auditory cortex 29, the guinea-pig inferior colliculus (auditory midbrain) 30 and the gerbil MSO 31. For example, many 29 or even most neurons 30,31 showed maximal ITD sensitivity (the peak in the measured ITD functions) outside the physiologically relevant range of ITDs. Previously, maximal ITD sensitivity outside the physiologically relevant range of ITDs has been found in single neurons. As they might have been involved in processing of other spatial temporal cues, such as reverberations 29,32, they were not interpreted as being in conflict with the Jeffress model. It might still be true that the ITD sensitivity of a subset of neurons functions to process the reverberations that occur in all natural habitats. However, a careful analysis of a large population of neurons in the guinea-pig inferior colliculus 30 argues against an agreement of these findings with the Jeffress model. There, best ITDs are not found to distribute across the relevant range of ITDs, as proposed by Jeffress. Instead, there is a systematic arrangement of ITDs that is of a different nature. As in the gerbil MSO 31, all neurons in the guinea-pig inferior colliculus that respond best to the same sound frequency show best ITDs that are narrowly distributed around the same best ITD. Overall, neurons that are tuned to high stimulus frequencies (which consequently have narrow NATURE REVIEWS NEUROSCIENCE VOLUME 4 JULY

4 a c Birds Mammals Excitation Inhibition b Right ear leading Left ear leading ITD (µs) d Right ear leading Left ear leading ITD (µs) Figure 3 Different strategies for encoding interaural time differences (ITDs). a The bird Nature Reviews Neuroscience ITD-encoding structure, the nucleus laminaris (shown is the arrangement as found in the chick), operates in a way that is similar to the model suggested by Jeffress in 1948 (REF. 3). Arrays of coincidence detector neurons (coloured circles) on both sides of the brainstem receive excitatory inputs (red lines) from the two ears (only inputs to the left nucleus laminaris are shown). Depending on the axonal length (delay line) of the two inputs, each coincidence detector neuron responds maximally to a particular ITD, which compensates for differences in the internal neural delays. A systematic arrangement of the delay lines from the contralateral ear creates a map of horizontal auditory space. b The maxima of the ITD functions of different neurons in one nucleus laminaris distribute across the physiologically relevant range (shaded area). The width of ITD functions depends on the stimulus frequency and, for lower frequencies, might be broader than shown. c The mammalian ITD encoder is the medial superior olive (MSO; coloured circles), as described for guinea-pigs and gerbils. MSO neurons that are tuned to the same frequency preferentially respond to the same ITD. The ITD functions are adjusted to bring the maximal slope close to zero. The opposite MSO is adjusted like a mirror image. Therefore, the relative activity of the entire population of MSO cells, rather than the distribution within an MSO, represents the horizontal position of a sound in space. Stimuli in the left hemifield mainly activate the right MSO, whereas stimuli in the right hemifield mainly activate the left MSO. Stimuli straight ahead activate both MSOs. The ratio of blue and yellow representing the auditory space in the figure indicates the relative activation of the left and right MSO. The ITD tuning is achieved by a complex interaction of binaural excitatory and inhibitory inputs. d ITD functions of MSO neurons are adjusted so that the maximal slope preferentially occurs close to zero ITD. For a given frequency, the ITD tuning of the population of MSO neurons in the left MSO mirrors that of the corresponding population of cells in the right MSO. ITD functions) (BOX 1) prefer short ITDs. Neurons with lower best frequencies prefer longer ITDs. Across the population of recorded cells, the mean best ITD corresponds to about cycles of a sine wave at each neuron s best frequency. This arrangement means that the steepest slope of ITD functions of most neurons occurs close to zero ITD and slightly shifted into the contralateral hemifield. So, there are considerable problems with the idea of an azimuthal space map in mammals that is based on best ITDs as proposed by Jeffress (FIG. 3c,d; compare REFS 27,33). Not only is the representation of ITDs in mammals inconsistent with the old textbook view, but there is also a conspicuous binaural inhibition innervating MSO neurons. Recent studies revealed a surprising role of this inhibition in ITD processing that was not anticipated by any ITD model 28,31. Glycine inhibition in the mammalian MSO The major ITD-encoding structure in the mammalian auditory system, the MSO, receives direct excitatory inputs from the ventral cochlear nuclei (VCN) on both sides of the brainstem (FIG. 4a). The source of this bilateral is the spherical bushy cells (SBCs), which faithfully time-lock their discharges to the temporal pattern of sounds for example, they provide phaselocked inputs in response to pure tones up to a few khz 34,35. The SBCs project to the bipolar MSO cells, with ipsilateral inputs making synapses on the lateral MSO dendrites and contralateral inputs making contact onto the medial dendrites Such an arrangement is thought to improve binaural coincidence detection 39. MSO cells show high-fidelity ITD sensitivity in all low-frequency hearing mammals tested so far 17,20,22,28. Although the first indirect evidence for inhibition acting on low-frequency MSO cells goes back to the 1960s 20, undisputable anatomical evidence was not provided before the early 1990s 40 44, and was shortly followed by direct physiological and pharmacological confirmation in vitro The inhibition acting on MSO cells is predominantly, if not exclusively, mediated by glycine and is conveyed through two pathways. The dominant pathway is from the MEDIAL NUCLEUS OF THE TRAPEZOID BODY (MNTB; compare REF. 48; FIG. 4a). MNTB cells are driven by sound at the contralateral ear through a fast and effective pathway: they receive projections from globular bushy cells (GBCs) in the VCN, which follow the temporal structure of sounds with exceptional precision (with high fidelity phase-locking to pure tones 49 ). The axon diameters of these projections are the largest in the auditory brainstem 50,51 and contact MNTB neurons through the largest, fastest and temporally most secure synapses in the mammalian brain the calyces of Held 52. As a result, MNTB cells also reliably phase-lock to frequencies beyond 1 khz 44,53 (BOX 2). Therefore, they convert temporally precise into temporally precise inhibition. Until recently, the MNTB represented an enigma, largely because the only target of MNTB cells that has been intensively studied is the lateral superior olive (LSO), a nucleus juxtaposed to the MSO. The LSO is the initial site of interaural intensity difference processing, which is used to localize high-frequency sounds. The underlying mechanism of this intensity difference processing was considered to be pure subtraction of contralateral, glycine-mediated inhibition, provided by the MNTB, from ipsilaterally driven 54,55. With few exceptions 56,57, interaural intensity processing was thought to require moderately fast but not outstandingly precise inhibition, and the existence of the calyx of Held remained functionally unexplained. More recent studies in the LSO show that timing does matter in adjusting the binaural properties of these cells 58,59 and there is indirect evidence that its kinetics are extraordinarily fast 28,60,61. But even more important is the finding that MNTB neurons also project into two other areas, the ventral nucleus of the lateral lemniscus 62,63 and the MSO (see above), both of which are mainly, if not exclusively, involved in temporal processing 53,64,65. 4 JULY 2003 VOLUME 4

5 Box 2 Different firing patterns of inhibitory inputs in mammals and birds MEDIAL NUCLEUS OF THE TRAPEZOID BODY (MNTB). Its neurons receive their inputs through the largest and temporally most secure synapse, the calyx of Held. MNTB neurons contain the highest concentration of the inhibitory transmitter glycine in the mammalian brain and project to several brainstem structures, among them the medial superior olive. No structural or functional analogue of MNTB is known in birds. DEPOLARIZING INHIBITION Inhibition is thought to function by hyperpolarization of the membrane potential of the target cell owing to the opening of Cl channels. However, in some neurons, release of inhibitory transmitters can cause depolarization, which in turn activates other channels that prevent the cell from reaching spike threshold. Whether the opening of Cl channels hyperpolarizes or depolarizes a cell depends on its Cl reversal potential. The inhibitory inputs to the bird and mammalian interaural time difference (ITD)-encoding structures have fundamentally different temporal properties. The glycinergic neurons in the medial nucleus of the trapezoid body (MNTB) provide the main, hyperpolarizing inhibition to the mammalian ITD detector, the medial superior olive (MSO). MNTB cells receive their input (calyciferous axon in a) from globular bushy cells through the largest and temporally most precise synapse in the mammalian nervous system, the calyx of Held 102 (a). For each MNTB cell, there is only one of these giant synapses covering a large portion of the postsynaptic membrane. The synaptic transmission in the MNTB ( µs synaptic delay) is twice as fast as that normally found in mammalian synapses 103 and the transmission has an extraordinarily low jitter 104. Accordingly, MNTB cells in vivo show high temporal fidelity 44,53. This high temporal fidelity is even visible in intracellular recordings in vitro from brain slices from 14-day-old animals, Therefore, the reason for the highly specialized inhibition might be not interaural intensity difference processing, but rather processing of temporal cues that use temporally accurate inhibition 48. The MNTB projection is not the only glycinemediated input to MSO cells. A second source of glycinergic inhibition onto MSO neurons, which is driven by the ipsilateral ear, is the lateral nucleus of the trapezoid body (LNTB) 43,45,66. LNTB neurons also receive inputs from GBCs, which contact their targets through large synapses, which resemble the so-called end bulbs of Held 67. Hence, the anatomy of this pathway is also consistent with temporally precise inhibition. a MNTB principal cell b 100 Hz 600 Hz 800 Hz Conventional synaptic bouton Calyx of Held Calyciferous axon 20 mv 30 mv 2.5 ms immediately after hearing onset (b). Stimulation of afferent fibres using trains of stimuli elicit action potentials that exactly follow the pattern of activation even at high rates (shown for 100, 600 and 800 Hz; reproduced, with permission, from REF. 103 (2000) Society for Neuroscience). So, temporal summation is avoided in this pathway. The pattern of activation in MNTB is then relayed as inhibition to the targets of the MNTB, including the MSO 46. The inhibition to the avian ITD detector, the nucleus laminaris, is fundamentally different. The GABA (γ-aminobutyric acid) neurons in the superior olivary nucleus (SON) in birds, which provide DEPOLARIZING INHIBITION (see text) to nucleus laminaris neurons, combine excitatory inputs over a long time window. Panel c shows how repetitive sub-threshold stimulation of the fibres that provide excitatory input to the SON in a chick brain slice preparation causes a pronounced summation of postsynaptic potentials, which can be seen by the progressive increase in the membrane potential. The stimulus pattern is visible as small stimulation artefacts in the traces. Even rates of only 25 Hz (top trace) cause summation that results in an action potential after the third stimulus (reproduced, with permission, from REF. 89 (1999) Society for Neuroscience). At higher repetition rates the duration until the first action potential is elicited is significantly shortened. These results indicate that SON neurons cannot reliably follow temporal input patterns. The effect of several inputs onto a nucleus laminaris neuron (or other SON targets) will therefore be tonic, and the overall release of GABA will depend on the sum of input activity, which is by and large related to the sound intensity at the ipsilateral ear. 10 ms c 25 Hz 50 Hz 100 Hz 200 Hz 50 ms 20 mv But what is the role of glycinergic inhibition in ITD processing, and does its timing really matter? Recordings from the rabbit superior olive provided indirect evidence that at least a subset of ITD-sensitive MSO neurons receives inhibitory inputs that are driven by the contralateral ear 28. However, there is no evidence for a differential distribution of inhibitory inputs in the MSO, so all MSO neurons should receive these inputs. But extracellular recordings alone would not reveal what exactly the function of these inhibitory inputs might be. Studies including pharmacological approaches using the Mongolian gerbil, which has been shown to use ITDs for sound localization 68 and which NATURE REVIEWS NEUROSCIENCE VOLUME 4 JULY

6 a Mammalian ITD-detection system To midbrain MSO Phase locked inhibition LNTB Phase locked inhibition SBC GBC MNTB GBC SBC VCN Left cochlea Right cochlea b Avian ITD-detection system To midbrain Postsynaptic depolarization and shunting Presynaptic decorrelation; postsynaptic depolarization and shunting NL NM NA SON Depolarization SON NM Left inner ear Right inner ear Figure 4 The interaural time difference (ITD)-encoding systems in mammals and birds. a The mammalian ITD encoder is the medial superior olive (MSO), an auditory brainstem structure (inset). MSO neurons are bipolar and their somata are arranged in one parasagittal plane. Excitatory projections originate from the left and right cochlea. Auditory nerve-fibre discharges time-lock to the stimulus onset or phase-lock to ongoing pure tones up to about 2 khz. They excite spherical and globular bushy cells (SBC and GBC, respectively) in the ventral cochlear nuclei (VCN). SBCs and GBCs also provide phase-locked to their targets: SBCs project to the MSO on both sides, GBC cells to the contralateral medial nucleus of the trapezoid body (MNTB) and the ipsilateral lateral nucleus of the trapezoid body (LNTB). The inputs to MNTB and LNTB are specialized to preserve temporal information: the MNTB is driven by a gigantic synapse, the calyx of Held. MNTB and LNTB neurons provide phase-locked, glycinergic, hyperpolarizing inhibition to the ipsilateral MSO. MSO neurons extract the ITD from a comparison of their four inputs and project to higher auditory centres. The well-timed inhibition is crucial for adjusting the ITD sensitivity of MSO cells to the physiological relevant range of ITDs. b The avian counterpart of the MSO is the nucleus laminaris (NL, here shown for the chick). NL neurons are also bipolar and their somata are arranged in one horizontal plane. They receive phase-locked from bushy cells in the nucleus magnocellularis (NM), the equivalent of the VCN in mammals. NL neurons also receive GABA (γ-aminobutyric acid)-mediated inhibition from the superior olivary nucleus (SON). SON neurons receive inputs from the NM, NL and nucleus angularis (NA) and provide inhibitory feedback to all of their input sources and to the contralateral SON. The inhibition by SON is decorrelated, thereby eliminating phase-locking, and acts through depolarization and shunting of its target cells. This inhibition provides a differential gain control for the ITD detector mechanism. 6 JULY 2003 VOLUME 4

7 has an MSO with all of the typical features of the mammalian ITD-processing system 22,69, revealed an unexpected role of inhibition. In vitro recordings in a gerbil brainstem preparation had already shown that the timing of the inhibition in the MSO is important for setting a time window for the coincidence detection of the excitatory inputs from both sides, and indicated a fast kinetic of the inhibitory postsynaptic potentials 46.The effect of the inhibition, however, remained unclear because the in vitro approach did not allow the relative timing of the four inputs to be assessed. Our recent in vivo recordings from single MSO neurons with and without pharmacological blockade of glycine-mediated transmission revealed a surprising function of the inhibition to the MSO 31. Under control conditions, the ITD sensitivity of most neurons in the gerbil MSO was tuned as described earlier (FIG. 5a), with the maximal slope close to zero ITD and the best ITDs corresponding to an interaural delay of 0.12 cycles of the neurons best frequencies. The application of strychnine an antagonist of glycinergic inhibition caused an increase in MSO firing rates. Such an increase in firing was expected, independent of any specific timing of the inhibition. However, this increase was large at some ITDs but small or absent at other ITDs. As a consequence, the ITD functions shifted. The nature of the shift was always the same: it moved the maximum of the ITD function close to zero ITD. As a consequence, the maximal slopes moved out of the physiologically relevant range of ITDs (FIG. 5a). From these results, two conclusions can be drawn. First, the excitatory inputs are generally adjusted so that the overall conductance delay from both sides is more or less identical (how this is achieved during ontogeny is unknown). As a consequence, the MSO coincidence detection based on causes ITD functions to peak around zero ITD. Second, the glycinergic inhibition adjusts the slope of the ITD function to the physiologically relevant range, which gives a maximum amount of information because small changes in ITD cause maximal changes in discharge rates 70. But how can this be explained in respect to the inhibitory inputs? A tonic, not phase-locked inhibition could account for such a shift only if the inhibition itself were already ITD sensitive. But the glycinergic inputs themselves are monaural, so they are insensitive to ITDs. The only plausible explanation, therefore, is that the relative timing of one or both of the inhibitory inputs is the defining factor. FIGURE 5b shows how this timing could have the effect observed in vivo. In this scenario, contralaterally driven inhibition precedes the from the same side, thereby delaying the net excitatory postsynaptic potential. This would cause a lowering and a shift in the maximum of the ITD function to more positive ITDs (contralateral stimuli leading). Given the fast pathway from GBCs through the calyx synapse to the MNTB, which then contacts only the MSO somata, such a temporal lead of inhibition might be possible. In the bat MSO, the contralaterally driven glycinergic inhibition has been shown to arrive simultaneously or even ahead of the contralaterally driven in a significant subset of neurons 64. In addition, ipsilaterally driven inhibition lagging the ipsilaterally driven would shorten the net excitatory potential that arises from ipsilateral stimulation, thereby adding to the effect of preceding contralateral inhibition. Preliminary data confirm that contralateral inhibition precedes contralateral and that ipsilateral inhibition is lagging ipsilateral 71. An earlier binaural model by Batra and colleagues 28, and our own binaural model 31, which was based on a modified Hodgkin Huxley model 72,73, incorporated phase-locked inhibition from the MNTB. Such an arrangement could simulate the shift of the ITD function as indicated earlier 31. FIGURE 5c shows the results of one of these simulations. The crux of the models so far is that one has to assume unusually fast conductance times for the inhibitory input, with a time constant, τ decay, in the range of only a few hundred microseconds 31. Estimations of in vitro recordings in rats indicate relatively fast conductance times, around 2 ms, for the glycinergic inhibition 74. This is slower than required for the proposed ITD-encoding mechanism, but there are two reasons to question whether this conductance reflects that in adult gerbils. First, the measurements were performed in brain slices from juvenile animals around the time of hearing onset, several days before the inhibitory system in the superior olive matures 69,75. Second, they were performed in rats, but it is unclear whether rats use ITDs. The only published recordings from rat MSO cells indicate that their temporal resolution is in the range of a few milliseconds, by far inferior to that in animals that are known to use ITDs 76 and more like that found in other non-itd users, such as bats 53. In accordance with this finding, the rat (and bat) MSO differs structurally from that in ITD-using mammals (such as gerbils or cats). The gerbil or cat MSO shows several anatomical specializations that probably represent a structural correlate of, if not the basis for, the functional adaptation to process temporal information in the microsecond range 48,69. The most obvious structural specialization is the strict alignment of the bipolar cell bodies in one sagittal plane, bringing the dendrites into spatial register 77. This might enable the axons to grow into the MSO in a uniform manner, to allow the right timing (which does not necessarily mean that they represent delay lines as proposed by Jeffress). By contrast, in non-itd-using animals, MSO cells are not aligned in this way and the dendrites are therefore not in register 48. The second structural specialization concerns the inhibitory inputs. In ITD-using mammals the glycine inputs are confined to the somata of MSO cells 69,78 80, whereas they are evenly distributed across somata and dendrites in non-itd-using animals, including rats 69. This confinement of inhibitory inputs is established only several days after hearing onset and is activity- and experience-dependent 69 (FIG. 6). This development significantly departs from what is known from the development of MNTB inputs to the LSO 81,82. Functionally, the refinement of the inhibition on MSO neurons should increase the temporal acuity of the NATURE REVIEWS NEUROSCIENCE VOLUME 4 JULY

8 b a Spikes s 1 Contralateral inputs Ipsilateral inputs EPSP 21% 83% Strychnine (without inhibition) Control (with inhibition) Ipsi leading Contra leading ITD (µs) PSP IPSP PSP IPSP EPSP c Spikes s 1 Spikes s G I,max (ns) ITD (ms) F signal (Hz) ITD (ms) Figure 5 The role of inhibition in the medial superior olive (MSO). a The effect of blocking inhibition at a single MSO neuron. Under control conditions, the MSO neuron s interaural time difference (ITD) sensitivity is maximal when the lead of the contralateral stimulus corresponds to about 0.12 cycles of the neuron s preferred stimulus frequency. In the case of the gerbil (shown here), the best ITD is outside the physiologically relevant range of ITDs (blue shading). However, the maximal slope is inside the physiological range. In the presence of strychnine, which blocks glycinergic inhibition, the MSO cell fires at a higher discharge rate at some ITDs and with an equal rate at others. The ITD function is shifted to the left. Maximal discharges occur around zero ITD and the steep slope is shifted outside the physiologically relevant range of ITDs. b The most likely explanation for the effect of the glycinergic inhibition is well timed, phase-locked inhibition from both sides. Without inhibition, the excitatory postsynaptic potentials (EPSPs) from the contra- and ipsilateral sides would reach the MSO without a significant interaural delay and would coincide at zero ITD. However, the contralaterally driven inhibition precedes the contralaterally driven, resulting in a delayed net postsynaptic potential (PSP) in response to contralateral stimulation. An ipsilateral inhibitory postsynaptic potential (IPSP) lagging the ipsilateral EPSP will result in a shortened net PSP in response to ipsilateral stimulation. Both effects act together and require positive ITDs (ipsilateral stimuli delayed) to bring the PSPs from both sides into register. The resulting ITD function will be shifted to positive ITDs. The contralateral inhibition is the more pronounced input, and so is probably the more essential part in the scenario. c Simulation of ITD encoding in the MSO with contralateral inhibition (for simplicity the ipsilateral inhibitory input was omitted) with a modified Hodgkin Huxley model (see text). Upper panel: ITD functions with various inhibitory synaptic strength (maximum synaptic conductance, G I,max ) show the influence of the contralateral inhibition on best ITD and overall spike rate (signal frequency 1,000 Hz). Right panel: ITD functions of a simulated 500-Hz neuron to ITDs of tones of various frequencies. The central peak of the ITD functions is largely independent of the stimulus frequency F signal, a feature observed in many MSO cells (G I,max 10 ns). Parts a and c modified, with permission, from Nature REF. 31 (2002) Macmillan Magazines Ltd. inhibitory currents because the problem of temporal summation due to the cable properties of large dendrites should be eliminated. The glycinergic inhibitory inputs to the MSO are structurally and functionally specialized to provide highly accurate timing and this timing seems to be crucial for shaping ITD functions in the mammalian ITD detector, the MSO. Direct evidence for this role of inhibition in the MSO comes from only one species, the gerbil, and is supported by some indirect evidence from rabbits. However, the fact that the anatomical arrangements of the inhibitory inputs are identical in other mammals, such as cats, argues for a similar role in all mammals that use ITDs to localize low-frequency sounds. However, this point has to be proven by further studies. Moreover, the question of whether mammals with significantly larger inter-ear distances represent ITDs in the same way as do guinea-pigs and gerbils remains subject to future experiments. GABA inhibition in the avian nucleus laminaris The bird ITD-encoding structure is the nucleus laminaris. Like MSO cells, nucleus laminaris neurons are bipolar, except for neurons tuned to relatively high frequencies in the barn owl nucleus laminaris 83.In chickens and alligators (also archosaurs), the principal dendrites arise from the dorsal and ventral poles of the cell bodies, which form a sheet of neurons arranged in a horizontal plane 9,11,83 (FIG. 4b). This pattern presumably represents the primary, plesiomorphic status of the nucleus laminaris 83. Bushy cells in the nucleus magnocellularis provide bilateral with well preserved timing, equivalent to the excitatory VCN input to the MSO in mammals. The way in which axons from contralateral bushy cells project to the chick nucleus laminaris, with each axon branching to send collaterals to nucleus laminaris neurons throughout the entire frequency contour (not shown in FIG. 4b), resembles the delay lines proposed by Jeffress 3. In fact, the contralateral delay lines in the chick create a measurable difference in the arrival time of action potentials in the mediolateral extent 5. But, at least in the chick, incoming axons from the ipsilateral nucleus magnocellularis branch in a pattern that causes the overall axonal length to each nucleus laminaris neuron to be roughly equal As proposed by Jeffress, nucleus laminaris neurons fire maximally when their binaural excitatory inputs arrive simultaneously, acting as coincidence detectors 4 8. The is mediated by glutamate The delay-line arrangement of the excitatory inputs results in a topographic map of ITDs in the nucleus laminaris 5. The barn owl nucleus laminaris represents a derived version that is based on the principles found in the chick. However, instead of a monolayer containing a single space map, the barn owl nucleus laminaris is hypertrophied, resulting in a polylayer with multiple representations of azimuthal space. Additionally, not only are the contralateral excitatory inputs arranged as delay lines, as in the chick, but also the ipsilateral inputs 83. In the barn owl, the azimuthal space map created in the nucleus laminaris is preserved at higher stations of the auditory system JULY 2003 VOLUME 4

9 a Juvenile CN b Adult CN c MSO MSO MNTB MNTB Figure 6 Inhibitory inputs to medial superior olive (MSO) cells become refined to cell somata. a In juvenile gerbils around hearing onset (postnatal day 12), excitatory and inhibitory inputs are distributed on dendrites and somata (no excitatory inputs to the somata are shown). b In adult gerbils, glycinergic inhibition is restricted to the cell somata and is absent on dendrites. CN, cochlear nucleus; MNTB, medial nucleus of the trapezoid body. c In adult gerbils, staining for gephyrin (green), a molecule that anchors the glycine receptors in the postsynaptic membrane, outlines MSO cell somata but is absent in the dendritic area (right relative to the area of cell bodies). d In adult gerbils, staining for glycine receptors (yellow) is restricted to the cell somata. The dendrites (blue, stained for MAP2) are devoid of glycine receptors. Parts c and d reproduced, with permission, from Nature Neuroscience REF. 69 (2002) Macmillan Magazines Ltd. d Recent studies in chickens and barn owls also show that nucleus laminaris neurons receive additional inhibitory inputs. However, the function and mechanism of action of the inhibition are fundamentally different from the inhibition that acts on the MSO in mammals. The first difference is that this inhibition is mediated by GABA rather than glycine It originates in the superior olivary nucleus (SON). It is unclear whether this nucleus is homologous with any nucleus in the mammalian superior olivary complex, which contains the MSO, LSO and MNTB. Apparently, there is no functional equivalent in the mammalian auditory brainstem. The SON receives inputs from the ipsilateral nucleus magnocellularis, the nucleus laminaris and the nucleus angularis and projects back to all three of them (FIG. 4b) using GABA. The second fundamental difference between mammalian and avian inhibition is that the SON provides feedback inhibition, whereas MNTB and LNTB provide feedforward inhibition. In addition to the ipsilateral feedback inhibition, there is a weaker projection from the SON to its contralateral counterpart The third difference is that, in contrast to MNTB and LNTB, which are specialized to preserve timing, the GABA-mediated inhibition that originates in the chick CN CN SON eliminates timing information. In vitro recordings (BOX 2) show that these cells have high input resistances and combine single excitatory postsynaptic potentials into a long-lasting postsynaptic excitatory potential, resulting in a random pattern of action potentials 89. In addition, synaptic release of GABA by SON neurons creates slow postsynaptic potentials in their target cells 89,94,96. One of the reasons for this might be an accumulation of Ca 2 in the presynaptic terminal, causing a decorrelation (loss of phase-locking) of synaptic vesicle release 96. Whatever the mechanisms are, the inhibitory system in the bird ITD-detecting system is characterized by decorrelation, resulting in a tonic effect that is related to the overall sound intensity but not to the temporal structure of the sound. The fourth and perhaps the most interesting difference between the inhibition in the MSO and in the nucleus laminaris is its direct postsynaptic effects. Both transmitters glycine and GABA act through ligandgated Cl channels. In adult-like MSO cells, the opening of these channels causes an influx of Cl, resulting in a classical inhibitory hyperpolarization 45, which moves the potential away from spike threshold and increases overall conductance. As shown in chick brain slices, the postsynaptic response to GABA release from SON terminals in the bird ITD system is mediated mainly by GABA A receptors, which also open Cl channels. Because of a more positive reversal potential for Cl, however, GABA depolarizes nucleus magnocellularis and nucleus laminaris neurons 87 89,94,97. These depolarizing inhibitory postsynaptic potentials (IPSPs) are even more effective in preventing the cell from firing action potentials than hyperpolarizing IPSPs would be. The reason for the inhibitory effect is probably twofold. First, the depolarization activates a low-threshold K conductance, which prevents the membrane potential from reaching spike threshold. Second, the Cl efflux, and possibly at high input rates also the resulting K conductance, have a shunting effect that reduces the cells input resistance 94,97. During activation of this feedback inhibition, higher input currents are required to activate the coincidence detectors. This effect increases the precision of phaselocking in the nucleus laminaris 88 and nucleus magnocellularis 94. Such a feedback circuit also seems to be well suited as a gain control that prevents coincidence detectors from firing as a result of increased monaural coincidence at higher sound amplitudes, which would cause higher firing rates of the excitatory nucleus laminaris inputs 98. Indeed, the ability of barn owl nucleus laminaris neurons to encode ITDs is resistant to changes in absolute sound level 99. Moreover, the inhibitory effect of one SON onto its contralateral counterpart should decrease the GABA-mediated inhibition onto the nucleus magnocellularis and nucleus laminaris at the side at which the sound amplitude is lower. Thereby, this system could also compensate for weak interaural intensity differences (compare REF. 94). In vivo recordings in the barn owl indicate that ITD sensitivity is relatively tolerant to interaural intensity differences, and this tolerance seems to be implemented NATURE REVIEWS NEUROSCIENCE VOLUME 4 JULY

10 at the level of the nucleus laminaris 100. An open question is whether GABA B receptorsmight also contribute to such gain controlin birds 101. Conclusions In contrast to the traditional textbook view, which assumes that the ITD-processing system is alike in birds and mammals, and functions according to the Jeffress model, recent evidence strongly indicates that birds and mammals have their own, profoundly different solutions for the same problem of how to localize sounds by using ITDs. These differences concern the mode of neural representation of ITDs as well as the role of inhibition in extracting ITDs. It will be important to explain why the systems evolved in such different ways, and to what extent common precursor systems have been crucial for this development. A system such as the ITD-processing one which performs such an extreme task and, therefore, has only a few degrees of freedom might help us to understand why systems evolve in one way and not in another, simply because the questions we can ask are unusually specific. There will also be the challenge to explain, at a cellular level, how the incredible specificity of the role of inhibition in temporal processing can be achieved and how it is adjusted during ontogeny. In particular, the role of auditory experience in the development of glycinergic inhibition in the MSO begs the question of how inhibitory synapses can be selected on the basis of how they influence temporal resolution in a sub-millisecond range. To solve these problems will be an important challenge in the coming years of research in sensory systems. 1. Clack, J. A. The evolution of tetrapod ears and the fossil record. Brain Behav. Evol. 50, (1997). An excellent review of our current knowledge of the independent evolution of the tetrapod middle ears. It revises the old but still often stated view that the mammalian three-ossicle middle ear derived from a reptilian-like one-ossicle middle ear. 2. Clack, J. A. Patterns and processes in the early evolution of the tetrapod ear. J. Neurobiol. 53, (2002). 3. Jeffress, L. A. A place theory of sound localization. J. Comp. Physiol. Psychol. 41, (1948). One of the most elegant and influential models in computational neuroscience. It provided the conceptual framework for several anatomical, physiological, psychophysical and theoretical studies related to sound localization in mammals and birds. 4. Carr, C. E. & Konishi, M. A circuit for detection of interaural time differences in the brain stem of the barn owl. J. Neurosci. 10, (1990). 5. Overholt, E., Rubel, E. W. & Hyson, R. L. A circuit for coding interaural time differences in the chick brainstem. J. Neurosci. 12, (1992). 6. Joseph, A. W. & Hyson, R. L. Coincidence detection by binaural neurons in the chick brain stem. J. Neurophysiol. 69, (1993). 7. Reyes, A. D., Rubel, E. W. & Spain, W. J. In vitro analysis of optimal stimuli for phase-locking and time-delayed modulation of firing in avian nucleus laminaris neurons. J. Neurosci. 16, (1996). 8. Kuba, H., Koyano, K. & Ohmori, H. Development of membrane conductance improves coincidence detection in the nucleus laminaris of the chicken. J. Physiol. (Lond.) 540, (2002). 9. Parks, T. N. & Rubel, E. W. Organization and development of brain stem auditory nuclei of the chicken: organization of projections from n. magnocellularis to n. laminaris. J. Comp. Neurol. 164, (1975). 10. Rubel, E. W. & Parks, T. N. Organization and development of brain stem auditory nuclei of the chicken: tonotopic organization of n. magnocellularis and n. laminaris. J. Comp. Neurol. 164, (1975). 11. Young, E. D. & Rubel, E. W. Frequency-specific projections of individual neurons in chick brainstem auditory nuclei. J. Neurosci. 3, (1983). 12. Young, E. D. & Rubel, E. W. Embryogenesis of arborization pattern and topography of individual axons in N. laminaris of the chicken brain stem. J. Comp. Neurol. 245, (1986). 13. Wagner, H., Mazer, J. A. & von Campenhausen, M. Response properties of neurons in the core of the central nucleus of the inferior colliculus of the barn owl. Eur. J. Neurosci. 15, (2002). 14. Knudsen, E. I. & Konishi, M. Center-surround organization of auditory receptive fields in the owl. Science 202, (1978). 15. Brainard, M. S. & Knudsen, E. I. Experience-dependent plasticity in the inferior colliculus: a site for visual calibration of the neural representation of auditory space in the barn owl. J. Neurosci. 13, (1993). 16. Gold, J. I. & Knudsen, E. I. A site of auditory experiencedependent plasticity in the neural representation of auditory space in the barn owl s inferior colliculus. J. Neurosci. 20, (2000). 17. Yin, T. C. & Chan, J. C. Interaural time sensitivity in medial superior olive of cat. J. Neurophysiol. 64, (1990). A careful description of a significant number of MSO cells in vivo. It provided the strongest physiological evidence so far in favour of a Jeffress-type ITDdetection mechanism in the MSO. 18. Joris, P. X., Smith, P. H. & Yin, T. C. Coincidence detection in the auditory system: 50 years after Jeffress. Neuron 21, (1998). 19. Moushegian, G., Rupert, A. L. & Gidda, J. S. Functional characteristics of superior olivary neurons to binaural stimuli. J. Neurophysiol. 38, (1975). 20. Goldberg, J. M. & Brown, P. B. Response of binaural neurons of dog superior olivary complex to dichotic tonal stimuli: some physiological mechanisms of sound localization. J. Neurophysiol. 32, (1969). 21. Crow, G., Rupert, A. L. & Moushegian, G. Phase locking in monaural and binaural medullary neurons: implications for binaural phenomena. J. Acoust. Soc. Am. 64, (1978). 22. Spitzer, M. W. & Semple, M. N. Neurons sensitive to interaural phase disparity in gerbil superior olive: diverse monaural and temporal response properties. J. Neurophysiol. 73, (1995). 23. Stern, R. M. & Trahiotis, C. in Hearing (ed. Moore, B. C. J.) (Academic, New York, 1995). 24. Palmer, A. R. & Shackleton, T. M. The physiological basis for the binaural masking level difference. Acta Acust. Unit. Acust. 88, (2002). 25. Smith, P. H., Joris, P. X. & Yin, T. C. Projections of physiologically characterized spherical bushy cell axons from the cochlear nucleus of the cat: evidence for delay lines to the medial superior olive. J. Comp. Neurol. 331, (1993). A thorough anatomical study addressing the question of the existence of delay-lines in mammals (although B.G. does not agree with their conclusion). 26. Beckius, G. E., Batra, R. & Oliver, D. L. Axons from anteroventral cochlear nucleus that terminate in medial superior olive of cat: observations related to delay lines. J. Neurosci. 19, (1999). 27. Middlebrooks, J. C., Xu, L., Furukawa, S. & Macpherson, E. A. Cortical neurons that localize sounds. Neuroscientist 8, (2002). 28. Batra, R., Kuwada, S. & Fitzpatrick, D. C. Sensitivity to interaural temporal disparities of low- and high-frequency neurons in the superior olivary complex. I. Heterogeneity of responses. J. Neurophysiol. 78, (1997). A detailed description of different types of ITD sensitivity in the superior olivary complex. This paper is particularly important because it comes from the first (and so far only) recordings from SONs in awake animals. Additionally, it provides the first indirect in vivo evidence for inhibitory effects on ITD functions in a mammal that uses ITDs to localize low-frequency sounds. 29. Fitzpatrick, D. C., Kuwada, S. & Batra, R. Neural sensitivity to interaural time differences: beyond the Jeffress model. J. Neurosci. 20, (2000). 30. McAlpine, D., Jiang, D. & Palmer, A. R. A neural code for low-frequency sound localization in mammals. Nature Neurosci. 4, (2001). A new way of analysing data from ITD-sensitive neurons in the guinea-pig auditory midbrain provided the basis for a new concept concerning the representation of ITDs in the mammalian auditory system. 31. Brand, A., Behrend, O., Marquardt, T., McAlpine, D. & Grothe, B. Precise inhibition is essential for microsecond interaural time difference coding. Nature 417, (2002). The first direct evidence for the involvement of glycinergic inhibition in ITD processing in the MSO in vivo. Pharmacological blockade of glycinergic inhibition caused a significant shift in the ITD sensitivity of single MSO cells. 32. Grothe, B. & Neuweiler, G. The function of the medial superior olive in small mammals: temporal receptive fields in auditory analysis. J. Comp. Physiol. A 186, (2000). 33. McAlpine, D. & Grothe, B. Sound localisation and delay lines do mammals fit the model? Trends Neurosci. (in the press). 34. Warr, W. B. Fiber degeneration following lesions in the anterior ventral cochlear nucleus of the cat. Exp. Neurol. 14, (1966). 35. Osen, K. K. Cytoarchitecture of the cochlear nuclei in the cat. J. Comp. Neurol. 136, (1969). 36. Stotler, W. A. An experimental study of the cells and connections of the superior olivary complex of the cat. J. Comp. Neurol. 98, (1953). 37. Lindsey, B. G. Fine structure and distribution of axon terminals from the cochlear nucleus on neurons in the medial superior olivary nucleus of the cat. J. Comp. Neurol. 160, (1975). 38. Kitzes, L. M., Kageyama G. H., Semple, M. N. & Kil, J. Development of ectopic projections from the ventral cochlear nucleus to the superior olivary complex induced by neonatal ablation of the contralateral cochlea. J. Comp. Neurol. 353, (1995). 39. Agmon-Snir, H., Carr, C. E. & Rinzel, J. The role of dendrites in auditory coincidence detection. Nature 393, (1998). 40. Cant, N. B. in Neurobiology of Hearing: The Central Auditory System (eds Altschuler, R. A., Bobbin, R. P., Clopton, B. M. & Hoffman, D. W.) (Raven, New York, 1991). 41. Covey, E., Vater, M. & Casseday, J. H. Binaural properties of single units in the superior olivary complex of the mustached bat. J. Neurophysiol. 66, (1991). 42. Kuwabara, N. & Zook, J. M. Projections to the medial superior olive from the medial and lateral nuclei of the trapezoid body in rodents and bats. J. Comp. Neurol. 324, (1992). 43. Cant, N. B. & Hyson, R. L. Projections from the lateral nucleus of the trapezoid body to the medial superior olivary nucleus in the gerbil. Hear. Res. 58, (1992). 44. Smith, P. H., Joris, P. X. & Yin, T. C. Anatomy and physiology of principal cells of the medial nucleus of the trapezoid body (MNTB) of the cat. J. Neurophysiol. 79, (1998). 45. Grothe, B. & Sanes, D. H. Bilateral inhibition by glycinergic afferents in the medial superior olive. J. Neurophysiol. 69, (1993). 46. Grothe, B. & Sanes, D. H. Synaptic inhibition influences the temporal coding properties of medial superior olivary neurons: an in vitro study. J. Neurosci. 14, (1994). The first in vitro evidence for well-timed glycinergic inhibition that could influence the ITD tuning of MSO cells. 47. Smith, P. H. Structural and functional differences distinguish principal from nonprincipal cells in the guinea pig MSO slice. J. Neurophysiol. 73, (1995). 10 JULY 2003 VOLUME 4