Abstract and Keywords
The cochlear nucleus contains the circuits through which information about sound is coupled to the brain. In the cochlear nucleus, fibers of the auditory nerve contact neurons that form multiple, parallel representations of the acoustic environment. These circuits vary from simple synapses that preserve the timing of auditory events to complex neuropils that are sensitive to features that identify sounds. This chapter discusses the organization of the cochlear nucleus, covering cell types, synaptic connections, membrane properties and integration of inputs, models of somatic and dendritic properties, and circuit functions.
The cochlear nucleus contains the circuits through which information about sound is coupled to the brain. In the cochlear nucleus, fibers of the auditory nerve contact neurons that form multiple, parallel representations of the acoustic environment. These circuits vary from simple synapses that preserve the timing of auditory events to complex neuropils that are sensitive to features that identify sounds. The parallel pathways each perform a different analysis of the auditory signal. Thus calculations such as the localization of a sound source in space or the identification of a sound are separated in the cochlear nucleus and performed in parallel as signals ascend through the brainstem auditory nuclei.
A tonotopic map of sound is generated in the cochlea, shown schematically in Fig. 4.1. Sound causes a traveling-wave of displacement on the basilar membrane. The location of the largest displacement depends on the frequency of the sound, with low frequency sounds causing displacement at the apex and high frequencies at the base of the cochlea. The resulting map of stimulus frequency as a function of place along the basilar membrane is shown in Fig. 4.1 for the cat cochlea (for descriptions of cochlear anatomy and physiology, see Pickles, 1988; Patuzzi, 1996). Transduction of basilar membrane motion occurs in hair cells (Kros, 1996). Fluctuations of hair-cell membrane potential reflect motion of the basilar membrane. Depolarization in hair cells excites dendrites of auditory-nerve fibers (ANFs), synaptically evoking action potentials whose rate reflects the degree of excitation. Thus any given ANF responds to sound in pro portion to the degree of basilar membrane motion at the point on the membrane innervated by the fiber. Each ANF responds to a limited range of frequencies. The frequency to which a fiber is most sensitive is its best frequency (BF). The best frequencies of ANFs vary systematically, forming a tonotopic map. Perceptually, position along the tonotopic map, or frequency of sound, corresponds to the sense of highness or lowness of the sound, which is like its position along a musical scale (Moore, 1997). In contrast with the visual and somatosensory systems, the tonotopic organization of the auditory system does not reflect location in space. The auditory system’s frequency map is more like the olfactory system’s map of odor molecular structure (Chap. 5).
There are two groups of hair cells and ANFs (Spoendlin, 1973; Kiang et al., 1982; Ryugo, 1992). Inner hair cells and type I ANFs form the major afferent pathway to the cochlear nucleus; each type I fiber innervates one inner hair cell. Type I fibers are (p.126)
NEURONAL ELEMENTS AND THEIR SYNAPTIC CONNECTIONS
The modern definitions of the cell types in the cochlear nucleus were developed by Kirsten Osen on the basis of cytoarchitecture (Osen, 1969) and modified by Merest (p.127) and colleagues (Brawer et al., 1974; Cant and Merest, 1984) on the basis of Golgi material. In some cases the correspondence between cell types was initially unclear so that multiple terms have come to be used to describe the same group of cells. Figure 4.2 shows the distribution of cell types in the cochlear nucleus of the cat in a sagittal view of the nucleus. Incoming ANFs bifurcate in the central region of the ventral cochlear nucleus (VCN) and send an ascending branch (a.b.) rostrally to the anterior division of the VCN (AVCN) and a descending branch (d.b.) caudally to the posterior division of the VCN (PVCN); the descending branch curves back rostrally to innervate the dorsal cochlear nucleus (DCN). The innervation of the cochlear nucleus by ANFs is orderly and reflects the tonotopic organization of the cochlea: low BF fibers bifurcate proximally and innervate the most ventral and lateral portions of the nucleus, whereas high BF fibers innervate more dorsal and medial portions of the nucleus (Osen, 1970a; Bourk e tal., 1981).
The symbols in Fig. 4.2 show the distribution of eight cell types that can be defined on the basis of cytoarchitecture. These are principal cells except for the granule cells and some of the small cells. Note that the principal cells are arranged in such a way that each type receives input from ANFs over the whole tonotopic range. In this sense, each principal cell type carries a separate but complete representation of the sound coming to the ear on that side of the head.
In projecting to different targets in the brain stem, the principal cell types form separate, parallel pathways. Figure 4.3 shows the projection patterns of six of the cell types
Multipolar (M), giant (Gi), and pyramidal (Py) cells project directly to the inferior colliculus (1C; Adams, 1979; Brunso-Bechtold et al., 1981). One population of multi-polar cells, the T-multipolars, project through the trapezoid body (TB) to the 1C, whereas another population, the D-multipolars, is inhibitory and projects to the contralateral cochlear nucleus (Cant and Gaston, 1982; Wenthold, 1987; Schofield and Cant, 1996). The octopus cells (O) project to the superior paraolivary nucleus (Schofield, 1995) and to the ventral nucleus of the lateral lemniscus (VnLL; Adams, 1997). These nuclei are not shown in Fig. 4.3 but are located dorsal to the LSO, between the superior olive and 1C; the nLL receive collateral projections from most of the cells shown. All parallel ascending auditory pathways through the brain stem converge in the 1C. It receives direct projections from the cochlear nucleus and superior olive and also projections from periolivary nuclei (PON) and from the nLL. The axons of most of these projections end in partially segregated bands within the colliculus (Oliver and Huerta, 1992).
Two nonprincipal cell regions of the cochlear nucleus are shown in Fig. 4.2: the granule cell areas (dotted) and the small cell cap (cap). The granule cells resemble and are developmentally related to those in the cerebellum (Mugnaini et al., 1980b; Berrebi et al., 1990; Fimfschilling and Reichardt, 2002); their axons project to the molecular, layer of the DCN (m.l. in Fig. 4.2), where they form parallel fibers. The small cell cap is a collection of small multipolar cells that lies between the principal cell regions of the VCN and the granule cells (Osen, 1969; Cant, 1993). This region receives collaterals of ANFs (Brown and Ledwith, 1990; Liberman, 1991) and collaterals of efferent neurons that project from the superior olive to the cochlea (Benson and Brown, 1990; Benson et al., 1996). These olivocochlear efferent neurons receive input from small-cell-cap axons (Ye et al., 2000) as well as from T-multipolar cells (Thompson and Thompson, 1991; Smith et al., 1993). It has been suggested that the olivocochlear efferent feedback circuit regulates the sensitivity of the cochlea (Guinan, 1996) and of the T-multipolars (Fujino and Oertel, 2001).
Each cochlear nucleus cell type has a unique pattern of response to sound, consistent with the idea that each type is involved in a different aspect of the analysis of the information in the auditory nerve. The diversity of these patterns can be accounted for by three features that vary among the principal cell types: (1) the pattern of the innervation of the cell by ANFs, (2) the electrical properties of the cells that shape synaptic inputs, and (3) the interneuronal circuitry associated with the cell. In the following (p.130) sections, these aspects of the principal cell types are described and related to the cells responses to sound.
CELL TYPES IN THE VENTRAL COCHLEAR NUCLEUS
Figure 4.4 shows examples of four neuron types found in the VCN and one from the DCN. These are taken from studies in which cells were filled with horseradish peroxidase after intracellular recording was performed to correlate their physiological and anatomical properties.
Bushy cells have short (<200 μm), bushy dendritic trees. Their synaptic input is located mainly on the soma, with few synapses on their dendrites (Ostapoff and Merest, 1991). Two subtypes of bushy cells are recognized. As discussed earlier, SBCs and GBGs differ in their locations (see Fig. 4.2) and their projections (see Fig. 4.3). They also differ in the number of converging ANF inputs. Figures 4.4A and 4.4B show ex amples of their cellular morphology; SBCs have one or a few short dendrites that terminate in a dense bush-like structure near the soma (Fig. 4.4A; Brawer et al., 1974; Cant and Merest, 1979a), and GBCs have more ovoid somata and larger, more diffuse dendritic trees (Fig. 4.4B; Tolbert and Morest, 1982a; Smith and Rhode, 1987). Bushy cells are innervated at the cell body by ANFs through large synaptic terminals, described later.
Multipolar cells have multiple, long dendrites that extend away from the soma in several directions (Fig. 4.4C; Brawer et al., 1974; Cant and Morest, 1979a). Two major classes of multipolar cells have been described (Cant, 1981, 1982; Smith and Rhode, 1989; Doucet and Ryugo, 1997; Fujino and Oertel, 2001). The cells of one group, T-multipolars (planar), have a stellate morphology with dendrites aligned with ANFs, suggesting that these cells receive input from a restricted range of best frequencies. Their axons project through the trapezoid body (hence the “T”) to the contralateral 1C (dashed line from cell M in Fig. 4.3B). Cells of the second group, D-multipolars (radiate), have dendritic fields that are not aligned with ANFs. Their axons leave the nucleus through the intermediate acoustic stria and project to the contralateral cochlear nucleus (solid line from cell M in Fig. 4.3B). The axons of both multipolar types have collaterals that terminate locally near the cell body and in the DCN to form intrinsic circuits (Smith and Rhode, 1989; Oertel et al., 1990). T-multipolar cells are excitatory, and D-multipolar cells are inhibitory and glycinergic (Wenthold, 1987; Ferragamo et al., 1998a; Doucet et al., 1999). Small cells in the cap are also sometimes referred to as “multipolar cells,” and multipolar cells are sometimes referred to as “stellate”
Octopus cells (see Fig. 4.4D) occupy a teardrop-shaped area in the posterior and dorsal PVCN (Fig. 4.2) where the ANFs converge to form a tonotopically organized bundle that courses toward the deep layer of the DCN (Osen, 1969; Kane, 1973; Oertel et al., 1990). Octopus cell dendrites are oriented, inspiring their name (the cell illustrated in Fig. 4.4D has a single dendrite extending opposite to the remaining dendrites, which is not typical). The orientation is perpendicular to the ANFs so that they receive input from fibers with a relatively wide range of BFs. The cell bodies encounter fibers with the lowest best frequencies, and the large dendrites extend toward fibers that encode higher frequencies.
CELL TYPES IN THE DORSAL COCHLEAR NUCLEUS
In contrast to the VCN, the DCN is layered and contains a substantial interneuronal neuropil that is important in generating its responses to sound. Figure 4.5 is a drawing of a Golgi preparation showing the neural elements of the DCN (Osen et al., 1990). The first and outermost layer, called the superficial or molecular layer, contains the axons of granule cells (gr), along with other populations of small interneurons. The second layer, called the pyramidal cell layer, is defined by the cell bodies of the pyramidal cells (Py), the most numerous of the DCN’s principal cell types, and cartwheel (Co) and granule cells. The innermost, deep layer is sometimes subdivided into layers (p.132)
The two major afferent systems of the DCN, the ANFs and the parallel fibers (granule cell axons), are othogonal to one another and innervate separate layers. The principal cells integrate inputs from both systems of afferents (Manis, 1989; Zhang and Oertel, 1994; Kanold and Young, 2001). Auditory nerve fibers innervate only the deep layer (layer 3) and do so tonotopically. Fibers with low BFs innervate the most ventral and lateral part of the DCN, whereas those with high BFs innervate the most dorsal and medial part (see Fig. 4.2B). Strips of cells receiving input from fibers with similar BFs form is ofrequency sheets (Osen, 1970a; Wickesberg and Oertel, 1988; Spirou et al., 1993). In the deep layer the dendrites and terminal arbors of many DCN neurons are confined to the is ofrequency sheets, so that the deep layer is organized into a succession of modules, each dealing with a narrow range of frequencies. The drawing in Fig. 4.5 is made in a plane roughly parallel to the path of ANFs through the (p.133) DCN. Parallel fibers innervate only the molecular layer running orthogonally to ANFs, thus crossing the is ofrequency sheets.
Pyramidal (fusiform) neurons (Py) are bipolar (see Figs. 4.4E and 4.5), with a spiny apical dendritic tree in the molecular layer and a smooth basal dendritic tree in the deep layer (Kane, 1974; Rhode et al., 1983b; Smith and Rhode, 1985). The cell bodies of pyramidal cells form a band in the pyramidal cell layer. The basal dendrites have few branches and are flattened in the plane of the isofrequency sheets, where they receive input from the auditory nerve (Blackstad et al., 1984; Zhang and Oertel, 1994). The apical dendrites branch profusely and span the molecular layer. They are densely covered with spines that are contacted by parallel fibers.
Giant cells (Gi) are large multipolar cells located in the deep layers of the DCN. They are the DCN’s second principal cell type (Kane et al., 1981; Zhang and Oertel, 1993b). Giant cells have large, sparsely branching, dendritic trees that cross isofrequency sheets. Most of the dendrites are confined to the deep layers and are smooth, but where the tips of dendrites reach into the molecular layer, they are covered with spines. Giant-cell axons join those of the pyramidal cells to form the DAS and project to the contralateral 1C (see Fig. 4.3; Adams, 1979; Ryugo and Willard, 1985).
Vertical (tuberculoventral) cells (V) are inhibitory interneurons found in the deep layer (Lorente de No, 1981; Zhang and Oertel, 1993c; Rhode, 1999). Their cell bodies and dendrites are intermingled among the basal dendritic trees of pyramidal cells. Their smooth dendritic trees are flattened in the plane of the isofrequency sheet so that when they are examined in tissue cut coronally, they appear to be oriented vertically, perpendicular to the plane of the layers (Osen, 1983). Vertical cell axons ramify parallel to their own isofrequency sheet within the DCN; many, perhaps most, of them also project to the VCN, where axons terminate tonotopically on cells with BFs corresponding to the vertical cells (Wickesberg and Oertel, 1988). Cells that project from the DCN, or tuberculum acousticum, to the VCN are called tuberculoventral cells. Vertical cells stain prominently for glycine (Wenthold et al., 1987; Saint Marie et al., 1991; Wickesberg et al., 1994). They are a prominent source of glycinergic inhibition in both DCN and VCN (Voigt and Young, 1990; Wickesberg and Oertel, 1990).
Cartwheel (Ca) cells are inhibitory interneurons whose numerous cell bodies lie in the pyramidal cell layer of the DCN (Wouterlood and Mugnaini, 1984). Their dendrites span the molecular layer and are densely covered with spines that are contacted by parallel fibers. Although these cells stain for glycine as well as for GABA and GAD (Osen et al., 1990; Kolston et al., 1992), they contact pyramidal, giant, and other cartwheel cells through glycinergic synapses (Golding and Oertel, 1997; Davis and Young, 2000). Cartwheel and cerebellar Purkinje cells share many features. Like Purkinje cells, cartwheel cells fire complex action potentials (Zhang and Oertel, 1993a; Manis et al., 1994). Both contain PEP19 (Berrebi and Mugnaini, 1991), cerebellin (Mugnaini and Morgan, 1987), and GAD (Mugnaini, 1985). Both cartwheel and Purkinje cells have mGluRla located in spines (Petralia et al., 1996; Wright et al., 1996). The dendrites of both contain IP3 receptors (Rodrigo et al., 1993; Ryugo et al., 1995). Genetic mutations affect Purkinje cells and cartwheel cells similarly (Berrebi et al., 1990)
Granule cells are microneurons whose axons, the parallel fibers, provide a major excitatory input to DCN through the molecular layer (Mugnaini et al., 1980a,1980b). They have short dendrites whose claw-like endings receive input from large mossy terminals (p.134) in glomeruli. Granule cells are found in domains around the VCN (see Fig. 4.2) and in the pyramidal cell layer of the DCN (see Fig. 4.5). In the molecular layer, the granule cell axons run perpendicularly to the isofrequency sheets from ventrolateral to dorsomedial. As in the cerebellum, the granule cells in the cochlear nuclei are associated with inhibitory interneurons.
Golgi cells provide both feed forward and feedback inhibition to granule cells (Mugnaini et al., 1980a; Ferragamo et al., 1998b). Golgi axons ramify locally and extensively within the granule cell domains. They contain both GABA and glycine (Mugnaini, 1985; Kolston et al., 1992) but are assumed, by analogy with cerebellar Golgi cells, to be GABAergic.
Stellate (St) cells are confined to the molecular layer. They have smooth dendrites and their axons arborize extensively in the plane of the molecular layer (Wouterlood et al., 1984; Zhang and Oertel, 1993a). Their cell bodies are labeled by GABAergic markers (Osen et al., 1990; Kolston et al., 1992).
Unipolar brush cells and chestnut cells are excitatory interneurons, slightly larger than granule cells, that reside in the two most superficial layers (Floris et al., 1994; Weedman et al., 1996). Not only do unipolar brush cells receive input through large, mossy terminals, but also their axons terminate in mossy endings, presumably supplying input to granule cells. The unipolar brush cells have been shown in the cerebellum to convert a single action potential from mossy fiber inputs into a prolonged train of action potentials (Rossi et al., 1995). Chestnut cells have a single stubby dendrite that receives multiple inputs (Weedman et al., 1996)
ANFs and mossy terminals to granule cells (and their parallel fiber axons) comprise the two major systems of extrinsic excitatory inputs to the cochlear nucleus. ANFs bring acoustic information from the cochlea. Through mossy terminals on granule cells, perhaps through unipolar brush cells, parallel fibers bring multimodal input from widespread regions of the brain, including somatosensory (Itoh et al., 1987), vestibular (Burian and Gstoettner, 1988), auditory (Brown et al., 1988a; Caicedo and Herbert, 1993; Feliciano et al., 1995; Weedman and Ryugo, 1996), and pontine motor systems (Ohlrogge et al., 2001).
In addition neurons receive excitatory and inhibitory input from local arborizations of cochlear nuclear neurons and excitatory and inhibitory efferent inputs from other brain stem nuclei (Cant, 1992). Through local collateral terminals, pyramidal and T-multipolar cells provide glutamatergic excitation (Smith and Rhode, 1985, 1989; Ferragamo et al., 1998a). D-multipolar, vertical, and cartwheel cells are inhibitory and glycinergic (Wickesberg and Oertel, 1990; Golding and Oertel, 1997; Ferragamo et al., 1998a), and stellate and Golgi cells are probably inhibitory and GABAergic (Osen et al., 1990; Kolston et al., 1992). Collaterals of olivocochlear neurons in the ventral nucleus of the trapezoid body provide cholinergic excitation to some neurons (Brown et al., 1988b; Horvath et al., 2000; Fujino and Oertel, 2001; Mulders et al., 2002). Many of the efferent inputs from the periolivary regions are inhibitory (Potashner et al., 1993; Ostapoff et al., 1997).
ANFs make synapses on all of the cell types in the cochlear nucleus, except the cells of the molecular layer of the DCN and those in the granule cell regions. Their terminals range in size from small boutons to large endbulbs (Rouiller et al., 1986). Figure 4.6 shows a reconstruction of an ANF that illustrates the variety of synaptic contacts (p.135)
SYNAPSES IN VENTRAL COCHLEAR NUCLEUS
Neurons in the ventral cochlear nucleus characteristically respond to sound by firing well-timed action potentials. Both the characteristics of the synaptic current and the
In the AVCN individual ANFs wrap the cell bodies of bushy cells with calyceal endings (eb in Fig. 4.6) or with clusters of boutons and fmger like endings (meb in Fig. 4.6). The SBCs in the AVCN receive input through endbulbs of Held from about three ANFs per cell (see Fig. 4.7B; Brawer and Morest, 1975; Ryugo and Fekete, 1982; Ryugo and Sento, 1991). The GBCs are contacted by smaller modified endbulbs from between 4 and 40 fibers (Tolbert and Merest, 1982b; Smith and Rhode, 1987; Ostapoff and Morest, 1991; Nicol and Walmsley, 2002), mainly on the soma and proximal dendrites (Fig. 4.7C). Endbulbs (EB) contain 100 or more synaptic release sites (Fig. 4.7A, arrowheads; Lenn and Reese, 1966; Cant and Morest, 1979b; Nicol and Walmsley, 2002). Both SBCs and GBCs also receive terminals other than those from ANFs; in GBCs, where this question has been examined directly (Ostapoff and Morest, 1991), ANF endings predominate on the soma and initial segment of the axon (27/52 somatic endings per cell), whereas inhibitory-type endings predominate on the dendrites (27/38 dendritic endings per cell). Small numbers of excitatory-type endings that are not from ANFs are also seen scattered everywhere on the cell (≈4 per cell).
Multipolar cells receive bouton endings from collaterals of ANFs (be in Fig. 4.6). In cats differences in the somatic innervation of T- and D-multipolar cells define a clear distinction between two types of multipolar cells (Cant, 1981; Smith and Rhode, 1989). The cell bodies of D-multipolars are densely covered with terminals, about half of which arise from ANFs (Fig. 4.7F). The cell bodies of T-multipolars, in contrast, receive few terminals on the soma and only one-tenth of those arise from ANFs (Fig. 4.7E). In multipolar cells many ANFs terminate on proximal dendrites. The remaining synapses are a mixture of the inhibitory types. The difference in somatic innervation is less clear in rats (Alibardi, 1998).
The number of terminals made by ANFs on multipolar cells has not been counted anatomically, but the electrophysiological evidence indicates that only about five ANFs contact a T-multipolar cell in mice (Ferragamo et al., 1998a). At left in Fig. 4.8A are shown the EPSPs evoked in a T-multipolar cell by electrical stimulation of the auditory nerve. As the stimulus strength increased in small increments, EPSPs increased in size by jumps until action potentials were produced. The scatter plots at the right show that the EPSP amplitudes evoked by a range of stimulus strengths clustered into about four groups, for subthreshold potentials (filled symbols in the bottom plot). When the slope of the rise of the EPSP was considered, two additional groups could be recognized in suprathreshold responses (open circles in the top plot). The five jumps in the rate of rise reflect the recruitment of five ANF inputs as the stimulus strength was increased.
Octopus cells are contacted by short collaterals of large numbers of ANFs through uniformly small terminal boutons; the innervation of the soma is dense (see Fig. 4.7D), but terminals are also found on the dendrites (Kane, 1973). Octopus cell dendrites extend across the auditory nerve array, receiving inputs from roughly one-third of the (p.138)
Most evidence suggests that the auditory-nerve neurotransmitter is glutamate: glutamate levels and levels of its precursor, glutamine, are high in the terminals of ANFs (Hackney et al., 1996), its release is Ca2+ dependent (Wenthold, 1979), its extrinsic application mimics the action of the neurotransmitter (Raman et al., 1994; Gardner et al., 2001), and the receptors clearly belong to the family of glutamate receptors (GluRs; Petralia et al., 1996; Wang et al., 1998).
The postsynaptic GluRs at ANF synapses in the VCN are mainly of the AMPA sub type. In young animals NMDA receptors contribute significantly to responses evoked by stimulation of the auditory nerve, but their contribution decreases with development and becomes insignificant in mature animals (Wickesberg and Oertel, 1989; Bellingham et al., 1998). Auditory AMPA receptors have rapid kinetics and high unitary conductances. It was first shown in targets of avian ANFs that the GluRs of the AMPA subtype have exceptionally rapid kinetics (Raman et al., 1994). The finding has been confirmed in mammals. Spontaneous release of neurotransmitter activates miniature synaptic currents whose time constants of decay are shorter than 1 msec (Gardner et al., 1999).
(p.140) GluRs are tetramers formed from any combination of the GluR family of subunits. In addition, alternative splicing gives these subunits “flip” or “flop” configurations. In VCN, most receptors associated with ANF terminals contain GluR3 and GluR4 sub-units in the flop configuration (Petralia et al., 2000; Schmid et al., 2001). The presence of flop suggests that the gating kinetics of these receptors are fast (Mosbacher et al., 1994) and the absence of GluR2 should give them high single-channel conductances and make them permeable to Ca2+ (Swanson et al., 1997). Figure 4.8C shows currents evoked by rapid application of glutamate recorded from three typical VCN cells (Gardner et al., 2001). The kinetics are fast for activation, deactivation, and desensitization; in fact, they match the most rapid that have been measured anywhere. Conductances of single GluRs on VCN cells average 28 pS. On average 40–50 receptors are activated in a miniature synaptic event.
The large conductance and rapid kinetics of receptors in the VCN can be understood in terms of the properties of the postsynaptic cells. Bushy and octopus cells have low input resistances that give them short time constants and allow them to encode brief events with temporal precision but also require large synaptic currents to bring them to threshold (Oertel, 1983; Manis and Marx, 1991; Golding et al., 1995). In the avian homolog of SBCs, activation of an action potential in one ANF opens ≈10,000 AMPA receptors and produces a conductance increase of over 200 nS through the release of ≈100 quanta (Zhang and Trussell, 1994). In rodent GBCs, each ANF activates an average conductance between 34 and 45 nS, evoking a synaptic current of over 10 nA (Isaacson and Walmsley, 1995; Bellingham et al., 1998).
Synapses in many parts of the brain show short-term and long-term changes in strength, including facilitation, depression, long-term potentiation (LTP), and long-term depression (LTD). Synaptic transmission by mature ANFs is remarkable in showing little plasticity, a feature that is useful for transmitting ongoing acoustic information faithfully and with minimal distortion by preceding sounds. In responding to sounds, mammalian ANFs fire up to 300 action potentials/sec in vivo (Sachs and Abbas, 1974). In vitro studies show that synaptic depression is prominent in very young animals and decreases as animals mature (Wu and Oertel, 1987; Brenowitz and Trussell, 2001). In relatively mature animals synaptic depression is detectable only at the highest natural firing rates (Wu and Oertel, 1987; Golding et al., 1995; Oertel, 1997). The depression seen in these records is likely to be greater than under physiological conditions, because these in vitro recordings were made at lower than physiological temperatures and synaptic depression decreases as temperature increases and because other mechanisms further decrease depression (Brenowitz et al., 1998).
SYNAPSES IN DORSAL COCHLEAR NUCLEUS
Pyramidal cells, the principal cells of the DCN, integrate input from two different circuits, one in the molecular layer and the other in the deep layer, through separate apical and basal dendrites. In the molecular layer the parallel fibers excite pyramidal cells through spines on apical dendrites (Smith and Rhode, 1985). The shafts of apical den drites and the cell bodies of pyramidal cells are densely packed with inhibitory-type terminals (Juiz et al., 1996). Many of these inhibitory terminals arise from glycinergic cartwheel cells. Cartwheel cells are themselves excited by parallel fibers on dendritic spines in the molecular layer and are also interconnected among themselves (Wouterlood (p.141) and Mugnaini, 1984; Golding and Oertel, 1997). In the deep layer, terminals from ANFs end on the mostly smooth basal dendrites, where they account for ≈10% of the terminals on the proximal dendrites and ≈38% on distal dendrites; the remaining terminals on the basal dendrites are inhibitory type. Indirectly ANFs drive glycinergic in hibition on basal dendrites through vertical cells (Voigt and Young, 1990; Zhang and Oertel, 1993c, 1994).
The AMPA receptors that mediate excitation to DCN neurons differ in the various neurons and, in the case of pyramidal cells, differ in the apical and basal dendrites. GluR2 and GluR3 subunits are found in both pyramidal and cartwheel cells (Petralia et al., 2000). The basal dendrites of pyramidal cells also contain GluR4 subunits but the apical dendrites do not (Rubio and Wenthold, 1997). Cartwheel cell dendrites do not contain GluR4 subunits but do contain GluRl subunits (Petralia et al., 1996). The dendrites receiving terminals from ANFs, both vertical cells and basal pyramidal den drites, show faster receptor currents than those receiving terminals from granule-cell axons, the cartwheel and apical pyramidal dendrites (Fig. 4.8C; Gardner et al., 2001). This difference may reflect the presence of GluR4 subunits, as in the VCN. In the case of vertical cells, the receptor kinetics are almost as fast as in VCN neurons, but the underlying receptor subunits have not been studied. A similar pattern is seen in miniature synaptic currents (Gardner et al., 1999).
There are dramatic differences in the way excitatory, glutamatergic synapses in the deep layer and those in the molecular layer of the DCN are modulated by activity (Fujino and Oertel, 2002). Synapses between parallel fibers and their pyramidal and cartwheel cell targets in the molecular layer show LTP and LTD. Examples are shown in Fig. 4.9. When parallel fibers are induced to fire rapidly (100 Hz) and their synaptic inputs are associated with a depolarization of the target neurons, synaptic currents increase in amplitude (Fig. 4.9A, B); when parallel fibers are induced to fire slowly (0.1 Hz) and their activity is paired with postsynaptic depolarization, synaptic currents decrease in amplitude (Fig. 4.9C, D). The strength of synapses between parallel fibers and their targets is thus continually modulated upward and downward by synaptic traffic. Those between ANFs or T-multipolar neurons and dendrites in the deep layer do not show LTP or LTD (Fig. 4.9E, F).
Plasticity in the molecular layer of the DCN has much in common with long-term plasticity in the hippocampus and cerebellum (see Chaps. 7 and 11). Synapses on spines, but not those on the shafts of dendrites, show LTP and LTD, supporting the idea the spines serve as biochemical compartments. Furthermore, long-term changes in synaptic strength are governed by changes in the intracellular Ca2+ concentration, because plasticity is not observed when recordings are made with patch-pipettes that contain high concentrations of Ca2+ buffers (9 mM EGTA). Both LTP and LTD require a rise of intracellular Ca2+. There are several potential sources of Ca2+ in fusiform and cartwheel cells. Some Ca2+ enters neurons from the extracellular space through the NMDA subtype of GluRs (Manis and Molitor, 1996). Blocking these receptors with DL-2-amino-5-phosphonovaleric acid (APV) reduces LTP in pyramidal and cartwheel cells, and it consistently blocks LTD in cartwheel cells (Fujino and Oertel, 2003). Antagonists of metabotropic GluRs also reduce LTP and LTD in pyramidal and cartwheel cells. Voltage-sensitive Ca2+ channels are prominent in cartwheel (Zhang and Oertel, 1993a; Manis et al., 1994; Golding and Oertel, 1997) and pyramidal (Hirsch and (p.142)
Granule cells receive inputs at specialized glomerular synapses in which a large synaptic terminal (mossy fiber) is partially surrounded by claw-like granule cell dendrites, which also receive inhibitory synapses from other cells, probably including Golgi cells (Mugnaini et al., 1980a).
(p.143) As was discussed for the VCN, neurons in the DCN are influenced by inhibition both from intrinsic inhibitory neurons and from descending inputs from periolivary regions. Stellate neurons in the molecular layer of the DCN label for GABA, cartwheel and Golgi cells double-label for GABA and glycine, and vertical and D-multipoIar cells label for glycine (Osen et al., 1990; Saint-Marie et al., 1991; Kolston et al., 1992; Doucet et al., 1999). One question raised by these findings is whether neurons that are labeled for both GABA and glycine release both neurotransmitters. Cartwheel cells have been shown to produce purely glycinergic synaptic potentials in their targets (Golding and Oertel, 1996, 1997). Whether this results from only glycine being released or from only glycine receptors being present postsynaptically is not known. Both GABA and glycine receptors that are typical of such receptors elsewhere in the brain are present in DCN (Wu and Oertel, 1986; Harty and Manis, 1996). In vivo antagonists of both glycine and GABA affect responses to sound, although the effects of glycine antagonists are usually stronger (Caspary et al., 1987; Evans and Zhao, 1993; Caspary et al., 1994; Davis and Young, 2000). A second question concerns why there should be two types of inhibition. Glycinergic, strychnine-sensitive IPSPs are prominent in circuits that can be activated in slices; GABAAergic inhibition is subtle (Ferragamo et al., 1998a; Lim et al., 2000). It is possible that GABA mediates mainly presynaptic inhibition through GABAB receptors and that, when GABA and glycine are released together, the glycine serves as the signaling molecule to the postsynaptic cell and the GABA serves to regulate the synapse presynaptically (Lim et al., 2000).
NUMBERS OF CELL TYPES AND CONVERGENCE
In cat, there are ≈3000 IHCs and ≈50,000 type I ANFs (Ryugo, 1992). The total population of cells in the cochlear nucleus is similar in number to the ANFs. Most numerous are the bushy cells, with 36,600 SBCs and 6300 GBCs (Osen, 1970b). These numbers are consistent with the innervation ratios described above for SBCs (1–3 ANFs per cell) and GBCs (≈35 ANF-type terminals per cell), given that each ANF contacts, on average, 1 SBC soma and 3–6 GBC somas (Liberman, 1991). There are ≈9400 multipolar cells (Osen, 1970b), most of which are T-multipolars; the ratio of T- to D-multipolars is ≈15:1 (Doucet and Ryugo, 1997). Anatomical convergence ratios onto multipolar cells have not been estimated, but it was argued earlier based on physiological data to be ≈5 ANFs/T-multipolar (Fig. 4.8A). There are ≈1500 octopus cells (Osen, 1970b), and each ANF contacts ≈2 somas in the octopus cell area (Liberman, 1993), yielding a convergence ratio of ≈67 ANFs/octopus cell, similar to the minimal value estimated earlier for mouse. In DCN, there are ≈4400 pyramidal cells and ≈1300 giant cells (Osen, 1970b; Adams, 1976). The most numerous DCN cell types are the granule cells (60,000) and cartwheel cells (18,000), with fewer stellates (4000) and Golgi cells (2300) (Wouterlood and Mugnaini, 1984). The degree of convergence onto and among DCN cell types cannot be estimated from existing data.
(p.145) The vertical cells are inhibitory interneurons that project to their isofrequency sheets in both DCN and VCN (Wickesberg and Oertel, 1988; Zhang and Oertel, 1993c; Rhode, 1999); they inhibit all of the principal cells in the cochlear nucleus, except the octopus cells (Voigt and Young, 1990; Wickesberg and Oertel, 1990, 1991). Vertical cells are narrowly tuned, and they respond most strongly to tones at a frequency near their BF (Spirou et al., 1999). Consistent with the tonotopic projection of vertical cell axons, cells in both DCN and VCN have a glycinergic inhibitory input whose BF is very close to the cell's own BF (Wickesberg and Oertel, 1988; Caspary et al., 1994; Davis and Young, 2000).
Both T- and D-multipolar cells in the VCN serve the role of interneurons through their axon collaterals. These terminate locally within VCN as well as projecting to the deep DCN (Smith and Rhode, 1989; Oertel et al., 1990). T-Multipolars are excitatory and D-multipolars are inhibitory. T-multipolars are the likely source of polysynaptic EPSPs in vertical and principal cells in DCN and in T-multipolars in VCN (Ferragamo et al., 1998a). D-multipolars seem likely to be the source of IPSPs in the same cells. The properties of D-multipolar cell responses to sound are complementary to those of vertical cells, in that D-multipolars are broadly tuned and respond best to stimuli like noise but only weakly to tones (Palmer et al., 1996; Paolini and Clark, 1999). In this sense, the vertical D-multipolar cells provide complementary inhibitory inputs to cells throughout the cochlear nucleus. The role of their interaction in the DCN is discussed later. Notice that these two inhibitory circuits are mutually inhibitory (Fig. 4.10).
Pyramidal cells in cats (but not in mice) and octopus cells give axon collaterals that terminate locally (shown as recurving arrows in Fig. 4.10); both are excitatory (Smith and Rhode, 1985; Golding et al., 1995). The pyramidal cell axons project to other pyramidal cells, and the octopus cell axons terminate in the octopus cell area and in granule cell areas.
Granule-cell axons (p.f. in Fig. 4.10) terminate on spines of the apical dendrites of pyramidal cells, on spines of cartwheel cells, and on the stellate cells of the superficial DCN. The axons of cartwheel cells end in the pyramidal and deep layers of the DCN, where they produce IPSPs in pyramidal and giant cells (Zhang and Oertel, 1993b, 1994; Golding and Oertel, 1997). In addition, cartwheel cells interact with one another in a network (Wouterlood and Mugnaini, 1984; Berrebi and Mugnaini, 1991; Golding and Oertel, 1996). The reversal potential of the cartwheel-to-cartwheel cell synapses lies a few millivolts above the resting potential and the threshold for firing so that its effect is depolarizing for the cell at rest but hyperpolarizing when the cell has been depolarized by other inputs. Thus, the effects of the cartwheel cell network are likely to be context dependent—excitatory when the cells are at rest but stabilizing when the cells are excited. The cartwheel cells are the most numerous neuron in the DCN, except for the granule cells; by virtue of their numbers, their network of interconnections, and their profuse terminal distribution, cartwheel cells represent a substantial computational resource for the DCN.
Not shown in Fig. 4.10 are some circuits in the granule cell regions, including the unipolar brush and chestnut cells and some of the connections of the Golgi cells. The latter form inhibitory feed forward and feedback connections with granule cells, as in the cerebellum (Mugnaini et al., 1980a) and probably also inhibit T-multipolar cells in VCN (Ferragamo et al., 1998b).
(p.146) MEMBRANE PROPERTIES AND INTEGRATION OF INPUTS
Previous sections have described the morphology, connections, and synaptic physiology of the circuits of the cochlear nucleus. In the remainder of this chapter, the properties of these circuits are related to the cells' responses to sound. Figure 4.11 shows that the electrophysiological properties of neurons influence their responses to sound and thus determines the computational role of cells. Responses to short bursts of sound recorded in vivo using extracellular electrodes are shown in the left column; responses in vitro to intracellular injections of hyperpolarizing and depolarizing currents are shown in the right column. The sounds in this case were tones at the BF of the neuron at a loudness significantly above threshold, where the neurons’ responses are stable and assume their typical form (Pfeiffer, 1966; Bourk, 1976). For comparison, the responses of ANFs to this stimulus resemble those shown in Fig. 4.11C1: following stimulus onset there is a rapid increase in discharge rate to a maximum, followed by a slower decline to a fairly steady discharge, the so-called primary like response. An important aspect of auditory-nerve responses that is not shown is their irregularity, meaning a lack of repetitive firing behavior; the spikes of an ANF occur randomly in time, so the intervals between spikes are highly variable and the pattern of action potentials is different for each stimulus repetition.
CHOPPER RESPONSES FROM T-MULTIPOLAR CELLS
Figures 4.11A and 4.11B show responses characteristic of T-multipolar cells (Rhode et al., 1983a; Wu and Oertel, 1984; Smith and Rhode, 1989; Feng et al., 1994). T-multipolar cells receive irregular, phasic input from ANFs, yet respond to tones by firing tonically at regular intervals, a pattern called chopping. The contrast between the regularity in firing of T-multipolar cells and the irregularity of their ANF inputs indicates that the temporal firing pattern is not imposed by the inputs but arises from the intrinsic properties of the cells themselves. Depolarization with steady current pulses causes T-multipolar cells to fire regularly with the same sort of chopping pattern (Fig. 4.1 IB; Oertel et al., 1988; Manis and Marx, 1991). The reproducibility of firing gives PST histograms of responses to sound a series of characteristic modes that is independent of the fine structure of the sound (Fig. 4.11A). The intervals between nodes are of equal duration and correspond to the intervals between spikes, with one spike per node. The peaks are large at the onset of the response because the latency to the first spike is quite reproducible in chopper neurons; the peaks fade away over the first 20 msec of the response as small variations in interspike interval accumulate and spike times in successive stimulus repetitions diverge (Young et al., 1988). As is shown later, the regular firing is expected from properties of neuronal spike initiation.
PRIMARYLIKE RESPONSES FROM BUSHY CELLS
Figure 4.11C shows typical responses to tones of bushy cells (Rhode et al., 1983a; Rouiller and Ryugo, 1984; Wu and Oertel, 1984; Smith and Rhode, 1987). Large EPSPs from between one and three endbulbs (Fig. 4.7A,B) cause SBCs to fire whenever the ANFs do (except when the cell is refractory). This one-spike-in, one-spike-out mode of processing means that the responses to sound of SBCs resemble those of ANFs, so they are called primary like (Fig. 4.11C1). Evidence that primary like responses reflect (p.147)
GBCs give a similar response, called primary like-with-notch, or pri-N (Fig. 4.11C2). Pri-N responses differ from primary like (and ANF) responses in their behavior at the be ginning of the response, where a precisely timed peak followed by a brief notch is ob served. The convergence of multiple large inputs from ≈35 ANFs (Ostapoff and Merest, 1991) makes the timing of firing and the encoding of the fine structure of auditory stimuli precise (Joris et al., 1994a, 1994b). For example, consider the large peak at the onset of the stimulus in Fig. 4.11C2; this peak contains a spike in every repetition of the stimulus. If the GBC fires in response to the first input spike it receives from any one of its inputs, then the more inputs that converge on the cell, the less time it will take for the first input spike to occur and the sharper will be the onset peak in the PSTH (Rothman et al., 1993). The small notch following the peak results from refractoriness.
The electrical characteristics of bushy cells are shown in Fig. 4.1 ID (Wu and Oertel, 1984; Manis and Marx, 1991). When depolarized, these cells fire one to three spikes at stimulus onset and then their membrane potential settles to a steady, slightly depolarized value (filled square). This behavior is produced by a low-threshold potassium conductance that is in part activated at rest and is strongly activated by depolarization, producing a membrane rectification (Manis and Marx, 1991; Reyes et al., 1994). The rectification (low input resistance in the physiological voltage range) has two important consequences: (1) large synaptic currents, provided by the calyceal endings of ANFs, are required to overcome the input resistance when the cell is depolarized; and (2) the membrane time constant of the cell is fast (2–4 msec at rest). The fast time constant can be seen from the rapid initial drop of the response to hyperpolarizing current and from the fall in voltage at the end of the depolarizing current pulse in Fig. 4.1 ID (compare with Fig. 4.1 IB). The short time constant of bushy cells blocks temporal integration of synaptic inputs (Oertel, 1983; Smith and Rhode, 1987; Paolini et al., 1997). Figure 4.12 shows postsynaptic potentials produced in a multipolar cell
ONSET RESPONSES FROM OCTOPUS CELLS
Octopus cells respond to the synchronous firing of groups of ANFs that occurs at stimulus transients, such as the sudden increase in loudness at the onset of a syllable. The tones that are usually used to characterize neurons activate relatively few ANFs and bring octopus cells to threshold only at the beginning of the tone, which has led to the term onset units (Fig. 4.13A; Godfrey et al., 1975a; Rhode et al., 1983a). When presented with a train of broadband transients, such as a train of clicks, octopus cells fire to each click up to very high rates (>500 spikes/sec). Octopus cells also respond vigorously to low frequency tones, tones that evoke synchronous firing of ANFs to each cycle of the tone (Rhode and Smith, 1986). This behavior reflects the unusual biophysical properties of octopus cells that allow EPSPs to be brief and the responses well timed.
Octopus cells are like bushy cells in that the resting input resistance of octopus cells is low, ≈6 MΩ (Bal and Oertel, 2000). As was shown for bushy cells in Fig. 4.12, the low input resistance prevents temporal summation of nonsynchronous inputs. Instead of receiving large inputs from few fibers, however, octopus cells receive small inputs from many ANFs. Thus simultaneous firing of many inputs is necessary to drive an octopus cell to fire.
Octopus cells show a strong membrane rectification, like bushy cells (Fig. 4.13B). In the case of octopus cells, the low input resistance results from the activation of two opposing voltage-sensitive conductances: a hyperpolarization-activated, mixed cation conductance (gh) that pulls the membrane potential toward −40 mV and a depolarization-activated K+ conductance (gKL) that pulls the membrane toward −80 mV (Golding et al., 1995; Bal and Oertel, 2000, 2001). Both are large conductances. The maximal gh is 150 nS; the maximal gkl is over 500 nS. The voltage sensitivity of these conductances overlaps at rest so that both conductances are partially activated, each with a resting current over 1 nA. The voltage sensitivity of these conductances is high near the resting potential, so that each conductance is activated steeply by small changes in the membrane potential. These voltage-sensitive conductances shape EPSPs and affect the firing of octopus cells in three important ways. (1) The low input resistance of octopus cells makes EPSPs small (<1 mV) and brief (≈ 1 msec), thus making the cell sensitive only to synchronous inputs. (2) When octopus cells are depolarized synaptically, gKL is activated rapidly and cuts short EPSPs, so that the timing of the peaks of synaptic depolarizations is nearly invariant over a large amplitude range. (3) gKL causes octopus cells to be sensitive to the rate at which they are depolarized (Ferragamo and Oertel, 2002). Slow depolarizations activate gKL and prevent the generation of action potentials. This (p.150)
D-multipolar neurons also respond to the onset of sounds but have not been characterized as well as octopus cells. Like octopus cells, D-multipolars receive innervation by many ANFs on their somata and on dendrites that spread across the tonotopic axis (Smith and Rhode, 1989; Oertel et al., 1990; Doucet and Ryugo, 1997), and both are broadly tuned (Jiang et al., 1996). D-multipolars also respond with a precisely timed onset spike to tones (as in Fig. 4.13A) but differ from octopus cells in that they give some steady discharge after the onset spike (Smith and Rhode, 1989; Rhode and Green-berg, 1994a; Paolini and Clark, 1999). Unlike octopus cells, the firing pattern in response to tones is shaped by inhibition along with the intrinsic electrical properties (Paolini and Clark, 1999).
PYRAMIDAL AND CARTWHEEL CELLS
Pyramidal neurons in the DCN show pauser and buildup responses to sound (Fig. 4.14A; Godfrey et al., 1975b; Rhode et al., 1983b). The examples shown in Fig. 4.14A are typical of anesthetized animals, where the inhibitory circuits of the DCN are weakened. The response shows a poorly timed, long latency onset spike followed by a prominent pause (Fig. 4.14A1) or a slow buildup in response with a long latency (Fig. 4.14A2). The membrane properties of pyramidal cells include the typical complement of potassium, calcium, and sodium channels (Hirsch and Oertel, 1988; Manis, 1990; Kanold and Manis, 1999; Molitor and Manis, 1999). The pauser and buildup characteristics seem to derive from a transient potassium conductance. Figure 4.1 IB shows intracellular responses to depolarizing currents in cells that were held hyperpolarized between depolarizing current pulses; the dotted lines in the current waveforms show the holding current necessary to place the cell at its resting potential. Depolarization following hyperpolarization produced a pauser response (Fig. 4.14B1) or a buildup response (Fig. 4.14B2), depending on the strength of the depolarizing current. If the cell was not hyperpolarized between depolarizations, the pause did not occur and the cell gave a simple tonic response, similar to Fig. 4.1 IB. The pause is produced by activation of a transient potassium conductance that is normally inactivated at rest. If the cell is hyperpolarized, inactivation is removed, so that the transient conductance can be activated by a subsequent depolarization (Manis, 1990; Kanold and Manis, 1999, 2001). In vivo a hyperpolarization that is sufficient to remove the inactivation is observed as an aftereffect of a strong response to an acoustic stimulus (Rhode et al., 1983b) or from inhibitory synaptic inputs. Properties of the transient potassium current are reviewed in Chap. 2.
Figure 4.14C shows responses to sound of cartwheel cells. Cartwheel cells are the only cells in the cochlear nucleus with complex action potentials, which reflect a combined calcium and sodium spike (Fig. 4.14D; Zhang and Oertel, 1993a; Manis et al., 1994; Golding and Oertel, 1997). Many cartwheel cells respond weakly to sound (Fig. 4.14C1), whereas others give robust responses more like those of other cochlear nucleus cells (Fig. 4.14C2; Parham and Kim, 1995; Davis and Young, 1997). No particular (p.152)
(p.153) MODELS OF SOMATIC AND DENDRITIC PROPERTIES
As data on the membrane properties of cochlear nucleus neurons have accumulated, it has become possible to use models to explore the different behaviors described in the previous section. These models are based on the gating of ion channels. Such models make explicit the qualitative hypotheses discussed earlier and provide rigorous quantitative tests of them. For example, the slope sensitivity of octopus cells can be accounted for by gating of the low-threshold potassium conductance (Cai et al., 2000), and the pausing behavior of DCN pyramidal neurons can be accounted for by transient potassium channels (Kim et al., 1994; Hewitt and Meddis, 1995; Kanold and Manis, 2001). In the following, two early models of T-multipolar and bushy cells are described to show how patterns of activation can be affected by the channels present in a membrane.
The transformation in firing pattern that is observed between ANFs and chopper neurons is primarily a change from the irregular discharge of ANFs to regular firing in choppers. Computational models show that the transformation is a property of the action potential generation mechanism itself (Arle and Kim, 1991; Banks and Sachs, 1991; Hewitt and Meddis, 1993).
Figure 4.15A shows the structure of a simple neuronal model consisting of a soma and an axon containing voltage-gated sodium and potassium channels, with characteristics like those in the squid axon, connected to a dendritic tree model (Banks and Sachs, 1991). Auditory-nerve inputs are applied to the model through the excitatory conductances g E. The model ANF spike trains are irregular and accurately duplicate the statistical features of real auditory-nerve spike trains. Figure 4.15B shows the excitatory synaptic conductance of the model (below) and the resulting spike train (above). In this model, there is little correspondence between the time of arrival of auditory-nerve spikes, as judged by the EPSPs, and the postsynaptic spikes. There is, of course, an EPSP preceding each output spike, but the basic pattern of the output is determined by the tendency of the neuron to fire regularly, and not by the time of occurrence of input spikes. Figure 4.15C is a PSTH of the model's output, showing the chopper pattern. Note the correspondence of spike times in Fig. 4.15B and peaks in the PSTH in Fig. 4.15C. Regular firing, or chopping, is thus a property of the voltage-sensitive conductances summarized in the Hodgkin-Huxley membrane model. Real multipolar cells have additional conductances that modulate the cells' behavior, but the basic result shown here does not change (Rothman, 1999).
Figure 4.15D shows a membrane model for a bushy cell (Rothman et al., 1993). Because the synaptic input to bushy cells is on the soma, the model consists of only one somatic compartment, containing the same conductances as in the multipolar cell model plus a low-threshold voltage-dependent potassium conductance, g M. This conductance is activated at and just above the resting potential and gives the model electrophysiological characteristics similar to those of real bushy cells (see Fig. 4.11D). As the number and strength of independent auditory-nerve inputs are varied, the model accurately duplicates bushy cell PSTHs, producing primary like responses (Fig. 4.11C1) if supplied (p.154)
The model in Fig. 4.15D accounts for the properties of SBCs and for some properties of GBCs. However, there are problems in trying to account for all the properties of GBCs. One issue is that GBCs receive a large number of ANF inputs, up to 50 (Liberman, 1991; Ostapoff and Merest, 1991). Not only do these inputs vary in strength, but their firing may be strongly correlated because they presumably receive input from the same or from neighboring hair cells. Most of these inputs probably come from ANFs with significant spontaneous discharge. Because the inputs have to be suprathreshold to produce an irregular output, the result of applying a large number of spontaneously active inputs to the bushy cell model is to give the model substantial spontaneous activity. However, pri-N neurons frequently have low or no spontaneous activity (Blackburn and Sachs, 1989; Spirou et al., 1990). In addition, there are problems accounting for all aspects of the phase-locking behavior of pri-N neurons (phase locking is explained later in Fig. 4.16). The details of this issue are beyond the scope of this chapter and are discussed elsewhere (Joris et al., 1994b; Rothman and Young, 1996). Inhibitory inputs to GBCs, which are not present in the model, could account for some of the differences.
We have seen that each neuron type of the cochlear nucleus has specific structural and functional properties that, together with its distinctive synaptic connections, enable it to respond in a specific way to auditory stimuli. We are now in a position to assess how the multiple features of the auditory stimulus are encoded by the parallel processing lines of the cochlear nucleus. It is the simultaneous extraction of these multiple features that permits the auditory system to be able to localize sound sources, whether the source is prey or predator or a person to whom one speaks, to interpret the meaning of sounds and, in humans, to understand language and appreciate song and instrumental music.
PHASE-LOCKING IN BUSHY CELLS AND SOUND LOCALIZATION IN THE HORIZONTAL PLANE
The computation of the location of a sound source in the horizontal plane begins in the superior olivary nuclei (see Fig. 4.3B), where neurons compare the time of arrival (MSO; Goldberg and Brown, 1969) and the relative loudness (LSO; Boudreau and Tsuchitani, 1970) of the stimuli in the two ears. Such comparisons are useful because a sound originating on, say, the left will reach the left ear before it reaches the right ear and will be louder in the left ear (Yin, 2002). The inputs to the MSO and LSO are from the bushy cells of the VCN. In the following paragraphs, the anatomical and membrane (p.156)
Interaural time differences are best encoded by phase locking to low frequency sounds because interaural time is encoded with every cycle of a sound. The means by which temporal information about the stimulus is encoded is shown in Figure 4.16A. This figure shows the spike trains of three neurons responding to a low-frequency tone; the responses are phase locked to the stimulus, in that spikes occur near a particular preferred portion of the stimulus waveform. Figure 4.16B shows that phase locking occurs in cat ANFs (line) for frequencies up to ≈5 kHz (Johnson, 1980). The shaded region shows that the phase-locking ability of primary like and pri-N neurons (bushy cells) is similar to that of ANFs (Blackburn and Sachs, 1989; Joris et al., 1994a). Bushy cells actually display enhanced phase locking at low frequencies, below 1 kHz, where they may be entrained precisely to the stimulus waveform (Joris et al., 1994a,b). By contrast, the phase locking of chopper neurons (T-multipolars) is much weaker, essentially disappearing by 2 kHz.
The differences between primary like and chopper neurons derive from their membrane properties. Because of membrane capacitance, the postsynaptic processing of all neurons tends to be low pass; i.e., fast fluctuations in the synaptic inputs are filtered out. In T-multipolars, where the inputs are on the dendritic tree, there is an additiona component of low-pass filtering due to the dendrites (White et al., 1994). As a result, frequencies in the input above a few hundred Hertz are severely attenuated. By contrast, in bushy cells postsynaptic filtering is minimized by placing the synapses on the soma, by making the postsynaptic currents large so as to quickly charge the membrane capacitance, and by minimizing temporal integration of inputs, as described in Fig. 4.12. The tendency of bushy cells to follow the temporal patterns of their inputs, as (p.157) opposed to T-multipolar cells is also apparent in the model results in Figs. 4.15B and 4.15E.
A particularly strong cue for sound localization is the interaural delay in the wave form of the stimulus. In fact, perceptual experiments show that the strongest cue for localization of sound in azimuth is the interaural delay at frequencies below 1 kHz (Wightman and Kistler, 1992). If the waveform of a stimulus like the tone in Fig. 4.16A is delayed in one ear, the ANF spikes that are phase locked to the tone will be delayed by the same amount. In other words, an interaural time delay produces a change in the relative phase-locking point in the two ears. Thus MSO neurons can compare the time of arrival of the stimulus waveforms in the two ears by comparing the time of arrival of phase-locked spike trains from the two cochlear nuclei. MSO neurons accomplish this comparison by functioning as coincidence detectors (Goldberg and Brown, 1969), meaning that they respond when they receive coincident spikes from their bushy cell inputs on the two sides. Coincidence detection is possible in MSO cells only because they share the short membrane time constant and the low threshold potassium channel described earlier for bushy cells (Smith, 1995). Inhibitory inputs from the MnTB and LnTB serve to sharpen the coincidence detection (Brand et al., 2002). Thus the membrane specializations of both bushy and MSO cells allow precision in the timing of firing that is necessary for binaural comparison of interaural time difference.
STIMULUS SPECTRUM REPRESENTATION IN CHOPPER NEURONS
In addition to localizing a sound, it is important to identify it and to extract its meaning. The auditory system identifies sounds based on their frequency content and their temporal fluctuations. An illustration of the importance of frequency content, or spectrum, is provided by the vowels of human speech. Figure 4.17A shows the frequency content of the vowel EH, as in “met.”. There are prominent peaks of energy at 512, 1792, and 2432 Hz (arrows); these peaks are called formants and correspond to the resonant frequencies of the vocal tract. Each vowel is characterized by a different combination of formant frequencies (Peterson and Barney, 1952), and our perceptual recognition of different speech sounds is closely tied to the frequencies of their first three formants (Remez et al., 1981).
Because of the importance of the frequency content of sounds for their perception, it is natural to consider the neural representation of sounds in terms of a plot of neural response versus BF (Pfeiffer and Kim, 1975; Sachs and Young, 1979). That is, we consider the representation of a sound in terms of the tonotopic map established in the cochlea (see Fig. 4.1), where each ANF represents the energy in the stimulus at frequencies near its BF. Figures 4.17B–E compare the tonotopic representation of the frequency spectrum of the EH for two subpopulations of ANFs and for chopper neurons (Blackburn and Sachs, 1990); the chopper neurons appear to derive a stable representation of the stimulus spectrum at all sound levels from the more variable ANF representation.
The plots in Figs 4.17B–E show discharge rate versus BF; these plots were constructed by recording the responses of several hundred neurons to the vowel, plotting each neuron's discharge rate at an abscissa position equal to its BF, and then averaging the data using a moving window filter (Blackburn and Sachs, 1990). The lines show the average values. Response profiles are shown for three populations of neurons: (p.158)
Low spontaneous rate fibers, by contrast, have higher thresholds and wider dynamic ranges. At the lowest level (25 dB; Fig. 4.17B) there is no response from the low spontaneous rate fibers, because the stimulus is below threshold. As stimulus level increases, a good rate representation is provided (dotted line), which is maintained to the highest level shown; there are clearly defined peaks of response near the first and the second/third formant peaks (arrows) with a minimum of response in between.
The solid lines in Fig. 4.17 show responses of chopper neurons to the same stimulus. Note that the choppers maintain a representation that is at least as good as that of the better ANF group; there is a clear peak at BFs equal to the formants at all sound levels (Blackburn and Sachs, 1990; May et al., 1998). This behavior could be explained at high levels if T-multipolar cells receive inputs predominantly from low spontaneous rate ANFs. However choppers also respond at low sound levels (Fig. 4.17B) and there fore must receive inputs from high spontaneous rate fibers (Bourk, 1976; Sachs et al., 1993). This raises the question of how choppers avoid saturation of their discharge rates by their high spontaneous rate inputs. It is likely that inhibitory inputs play a role (Blackburn and Sachs, 1992; Caspary et al., 1994; Rhode and Greenberg, 1994b; Palmer et al., 1996). Known sources of inhibition on T-multipolars include D-multipolar and vertical cells in the cochlear nucleus (see Fig. 4.10; Wickesberg and Oertel, 1990; Ferragamo et al., 1998a) as well as neurons in the superior olive (Ostapoff et al., 1997). These inhibitory inputs have diverse responses to sound. Particularly interesting with respect to responses to speech are the D-multipolars, which respond strongly to broad band stimuli like speech (Winter and Palmer, 1995). From Fig. 4.10, it is clear that the D- and T-multipolars form a network in which T-multipolars receive both excitatory and inhibitory recurrent inputs, as well as ANF inputs. Such networks can perform several types of computations, including a winner-take-all computation in which the strongest input dominates, reducing other inputs to zero (Shamma, 1998; Wilson, 1999). In the case of the T-multipolars, the response peak among low spontaneous rate ANFs with BFs near a formant peak could suppress the saturated inputs from high spontaneous rate units, giving the robust chopper representation. Another possibility is that T-multipolars arrange the low and high spontaneous rate inputs on their dendritic trees in such a way that they can switch from one input to the other (Lai et al., 1994). This sort of switching is based on theoretical calculations showing that inhibitory inputs can cancel excitatory inputs located more distally on a dendritic tree (Koch et al., 1983). Thus high spontaneous rate inputs located distally on the tree could be cancelled by inhibitory inputs (from D-multipolars) at high levels to allow the neuron to respond to low spontaneous rate inputs located proximally. The dynamic range of T-multipolars might also be extended by cholinergic modulation. Cholinergic inputs from olivo-cochlear efferent neurons reduce the sensitivity of the cochlea in the presence of loud sounds, reducing saturation (Guinan, 1996). They also excite T-multipolar cells but do not affect D-multipolar cells (Fujino and Oertel, 2001). The cholinergic efferents would thus be expected differentially to boost excitation of those chopper neurons that en code the formants, boosting the encoding of spectral peaks more than other excitation, and to increase the balance of excitation over inhibition.
(p.160) In addition to the frequency information discussed in Fig. 4.17, natural auditory stimuli also contain information in their temporal structure. An example of temporal structure is the fluctuation in sound amplitude corresponding to the sequence of syllables in an utterance. In experiments in which the information encoded in the frequency content of the sound (like Fig. 4.17A) is removed, leaving only the temporal structure, listeners can make many basic speech discriminations (Van Tasell et al., 1987). Indeed the cochlear implant, an auditory prosthesis for the deaf in which the ANFs are directly stimulated electrically, conveys much of its information via the temporal structure of the stimulation (Shannon et al., 1995). Cochlear nucleus neurons are sensitive to temporal fluctuations like those in speech and generally sharpen the representation of temporal information, i.e. give enhanced responses, compared with ANFs, to increases or decreases in stimulus amplitude (Wang and Sachs, 1994; Delgutte et al., 1998; Frisina, 2001). Onset neurons show the largest enhancement, followed by choppers and then primary like neurons.
The representation of temporal and spectral information encoded in onset and chopper neurons allows the identity of sounds to be determined—one speech sound versus another, for example. This information complements the information about sound source location provided by the bushy cell-superior olive pathway. Thus we begin to see how aspects of the acoustic environment are separated out at the brainstem level and selectively processed and represented. It is important to point out, however, that the separation is not complete. Information about the stimulus frequency spectrum is also encoded in the bushy cell pathway, and information about sound localization is encoded in the chopper pathway.
FEATURE DETECTION IN DORSAL COCHLEAR NUCLEUS
As discussed earlier, principal cells of the DCN integrate two systems of inputs (see Fig. 4.10): ANFs and related inhibitory inputs and parallel fibers and their associated inhibitory circuits. The former carry mainly auditory information, but the latter carry a mixture of auditory and nonauditory information. In contrast to the VCN, where effects of inhibition are relatively weak, DCN neurons in unanesthetized animals receive strong inhibition from both sets of inputs. In the DCN, spectrally complex sounds evoke a summation of excitation and inhibition that enables neurons to detect spectral features, which are often the information-bearing elements of sounds (Nelken and Young, 1996; Parsons et al., 2001).
Figure 4.18B shows a tone response map typical of what are almost certainly pyramidal cells (so-called type IV units) in unanesthestized animals (Evans and Nelson, 1973; Young and Brownell, 1976). The map shows discharge rate as a function of frequency and sound level. Two inhibitory areas (gray) are consistently observed in such maps'one located at and below BF (≈13–20 kHz in this case) and a second above BF (>22 kHz here; Spirou and Young, 1991). Excitatory areas are seen at low sound levels at BF (≈18 kHz) and usually, but not always, between the two inhibitory areas (≈21 kHz here) and at low frequencies (<10 kHz). In gerbil DCN, response maps are similar but inhibition is weaker (Davis et al., 1996b; Davis and Voigt, 1997).
Figure 4.18D schematically shows how the excitatory and inhibitory areas in type IV response maps could arise from what is known about the connections and the responses to sounds of interneurons (Spirou and Young, 1991; Davis and Young, 2000). (p.161)
The vertical and D-multipolar cells provide two complementary sources of inhibition to DCN principal cells. Vertical cells are inhibited by D-multipolar neurons (Fig. 4.10); because the latter respond strongly to noise but not to tones (Winter and Palmer, 1995; Palmer et al., 1996), vertical cells have the opposite characteristic, responding to tones but not to noise (Spirou et al., 1999). This circuitry makes the pyramidal cell's responses complex and nonlinear. For example, the response map shown in Fig. 4.18B predicts responses to tones and other narrowband stimuli but fails to predict responses to broadband stimuli like noise (Spirou and Young, 1991; Yu and Young, 2000). For broadband stimuli, the type II inhibitor is itself inhibited, by the D-multipolar, so the large central inhibitory area disappears.
Natural stimuli are mixtures of narrowband and broadband features. Pyramidal cells are inhibited (from their spontaneous rate, averaging ≈40/sec) by either narrowband peaks in the stimulus’ frequency content (as at a formant peak in speech) or notches in the frequency content of a broadband noise (Nelken and Young, 1994). The latter is a stimulus feature produced by the external ear that is used by humans and cats to localize sound sources (Musicant et al., 1990; Middle brooks, 1992; Huang and May, 1996). Cats appear to process the narrowband notch cue in the DCN, because lesions there specifically degrade performance in vertical sound localization, which depends on the notch cue (May, 2000). Thus, DCN principal cells seem to signal “interesting”features in the stimulus spectrum by being inhibited where such features lie near BF.
The second set of inputs to DCN principal cells, from granule cells and their associated inhibitory interneurons, conveys multimodal sensory information. Auditory responses are weak in cartwheel cells (Parham and Kim, 1995; Davis and Young, 1997), which probably reflects a generally weak acoustic response in the granule cell system. However granule cells strongly excite and, through cartwheel cells, inhibit pyramidal cells when the superficial DCN circuitry is activated from the somatosensory spinal nuclei (Davis et al., 1996a; Kanold and Young, 2001). The somatosensory input comes predominantly from the muscles connected to the pinna in cat. This fact raises a number (p.163) of interesting possibilities because of the importance of pinna movements for hearing (or analogously head movements in animals that do not have mobile pinnae). One possibility is that the DCN is involved in coordinating motor and sensory information in sound localization (May, 2000; Young and Davis, 2001).
The similarity of the granule cell circuitry in DCN to that in the cerebellum (Mugnaini and Morgan, 1987) and similar structures in electric fish (Montgomery et al., 1995; Bell, 2002) suggests that the DCN might be performing a role similar to one kind of cerebellar learning (Medina et al., 2000). A wide range of information about movements of the pinna, turning of the head, and movement of the body, which can make noise or cause changes in the environmental sounds reaching the ear, is represented in the parallel fiber array. Synaptic plasticity in the apical dendrites of the cartwheel and pyramidal cells (see Fig. 4.9) could then be used to learn associations between sound and the information present on the parallel fibers. In electric fish, a system like this is used to subtract off self-generated electric fields (Bell et al., 1997). In the auditory system, such information could be important in interpreting self-generated acoustic changes and the DCN could be used to discover when such changes are expected. (p.164)