Central circuitry and function of the cochlear efferent systems

The cochlear efferent system comprises multiple populations of brainstem neurons whose axons project to the cochlea, and whose responses to acoustic stimuli lead to regulation of auditory sensitivity. The major groups of efferent neurons are found in the superior olivary complex and are likely activated by neurons of the cochlear nucleus, thus forming a simple reﬂex pathway back to the cochlea. The peripheral actions of only one of these efferent cell types has been well described. Moreover, the efferent neurons are not well understood at the cellular- and circuit-levels. For example, ample demonstration of descending projections to efferent neurons raises the question of whether these additional inputs constitute a mechanism for modulation of relay function or instead play a more prominent role in driving the efferent response. Related to this is the question of synaptic plasticity at these synapses, which has the potential to differentially scale the degree of efferent activation across time, depending on the input pathway. This review will explore central nervous system aspects of the efferent system, the physiological properties of the neurons, their synaptic inputs, their modulation, and the effects of efferent axon collaterals within the brainstem.


Introduction
The cochlear efferent system is composed of several neuronal populations that receive central auditory input and project to the inner ear. One of these populations, the medial olivocochlear (MOC) neurons, located in the ventral nucleus of the trapezoid body (VNTB), are likely to play several roles, including enhancing cochlear sensitivity to salient sounds in noisy environments and dampening the response of the cochlea to loud sounds ( Guinan, 2018 ). A second population, the lateral olivocochlear (LOC) neurons, includes a set of neurons within and surrounding the lateral superior olive (LSO), and have a projection pattern in the cochlea distinct from that of the MOC neurons ( Brown, 1989 ;Guinan et al., 1983 ;Liberman and Brown, 1986 ). More recently, a smaller set of neurons in the ventral nucleus of the lateral lemniscus have also been identified as a bona fide cochlear efferent population and termed the dorsal efferents (DE) ( Suthakar and Ryugo, 2021 ). Unlike the function of the MOC system, the functions of the LOC and DE systems in hearing are unknown.
Much of what we have learned about efferent function comes from studies in which efferent fibers are activated and their effects on otoacoustic emissions assessed. Additional fruitful avenues include in vitro studies demonstrating the effects of the efferent transmitter acetylcholine on hair cell function. Far less is known about the central efferent neurons themselves, that is, their physiology, molecular biology, and neural circuitry. Accordingly, it is not known how efferent neurons are recruited in response to different acoustic stimuli or how this recruitment varies in different behavioral states associated with neuromodulatory pathways. In this review, we will describe the current state of understanding of efferent neurons, their synaptic inputs, and transmitter sensitivity, with a focus on animals post-onset of hearing. A major message in this overview is that the diversity of efferents, and the complexity of the inputs that control their firing, suggests that the functions of the efferent system may be much greater or more refined than previously proposed.

Efferent subtypes and anatomy
Olivocochlear efferents form two distinct groups of neurons that provide divergent innervation of the cochlea, termed LOC and MOC due to their relatively lateral and medial somatic locations within the olivary complex ( Fig. 1 A ) ( Warr and Guinan, 1979 ). Unlike LOC neurons, which have thin, unmyelinated axons, MOC axons are thicker and heavily myelinated ( Guinan et al., 1983 ). Together, these mixed-diameter efferent fibers enter the basal turn of the cochlea alongside afferent spiral ganglion neurons (SGNs). Before the discovery of these two separate efferent systems, olivocochlear fibers were originally categorized as either "crossed" or "uncrossed" with respect to the midline of the brainstem ( Rasmussen, 1953 ). While olivocochlear fibers are now better described as LOC or MOC, the "crossed" and "uncrossed" terminology is still useful for distinguishing fibers within each efferent system, as neurons within each group send projections to either cochlea. In most species, nearly all projections from LOC neurons are to the ipsilateral cochlea, while MOC projections skew more toward the contralateral than ipsilateral cochlea (see Warr, 1992 , Table 7.1). In rare instances ( ∼5%), individual MOC fibers bifurcate and terminate in both cochleae ( Robertson et al., 1987 ).
The somata of LOC neurons reside in and around the lateral superior olive (LSO) ( Fig. 1 A ), and terminate onto type I SGN dendrites directly beneath inner hair cells (IHCs) in the ipsilateral cochlea ( Fig. 1 B ) ( Brown, 1987 ;Robertson, 1985 ;Warr and Guinan, 1979 ;Warr et al., 1997 ). The LOC neurons can be further divided into two subgroups: smaller "intrinsic" neurons, which are restricted to the LSO, and larger "shell" neurons, which are sparse in number and border the margins of the same nucleus ( Vetter and Mugnaini, 1992 ). Shell LOC neurons extend their dendrites into the LSO and the surrounding reticular formation, whereas intrinsic LOC neurons align parallel to isofrequency bands within the LSO. The axons from shell LOC neurons innervate larger portions of the cochlea compared to those of intrinsic LOC neurons, and have been observed to span over a third ( ∼41%) of the cochlear length in rats ( Warr and Boche, 2003 ). The somatic morphology, location, and cochlear innervation patterns of these two subgroups suggest that the action of shell LOC neurons on auditory nerve fibers are less frequency specific than those of intrinsic LOC neurons. A third group of efferent neurons has recently been identified that are morphologically similar to LOC neurons, the so-called DE neurons ( Suthakar and Ryugo, 2021 ). DE neurons reside in the ventral nucleus of the lateral lemniscus (VNLL) and send unmyelinated projections to the ipsilateral cochlea. While they have been reported in adult mice (1 -4 months old), they have yet to be reported in other species. Further studies are needed to elucidate the actions of DE efferents in the auditory system. MOC neurons reside in medial nuclei of the superior olivary complex (SOC), predominantly in the VNTB ( Fig. 1 A ) ( Brown and Levine, 2008 ;Warr, 1975 ). In the adult cochlea, they directly synapse onto outer hair cells (OHCs) ( Fig. 1 B ) ( Warr and Guinan, 1979 ), but initially contact IHCs early in development ( Simmons, 2002 ;Simmons et al., 1996 ). MOC neurons are generally associated with their ability to modulate OHC activity; however, they also directly synapse onto the dendrites of type I SGNs in adult mice and guinea pig ( Fig. 1 B ) ( Hua et al., 2021 ;Robertson and Gummer, 1985 ), suggesting that individual MOC neurons directly modulate both auditory nerve activity and OHC function. When unilaterally stimulated with an electrode near their somata, MOC neurons act most strongly on the contralateral ear ( Gifford and Guinan, 1987 ), reflecting their anatomical preference for contralateral projections. Additionally, the location of MOC neuron somata extend more rostrally than LOC neurons, spanning the entirety of the rostral periolivary region (RPO) ( Vetter and Mugnaini, 1992 ;Warr et al., 2002 ). While the total number of olivocochlear neurons vastly differs between and within different species ( Warr, 1992 ), the estimated ratio of MOC compared to LOC neurons is generally low. For example, out of approximately 1,365 olivocochlear neurons in humans, about 26% are MOC and 74% are LOC ( Arnesen, 1984 ). Similarly, out of approximately 475 olivocochlear neurons in mice, about 35% are MOC and 65% are LOC ( Campbell and Henson, 1988 ).

Efferent action in the cochlea is facilitated by a diverse collection of neurotransmitters
Peripheral auditory function is orchestrated by a combination of mechanical processes, afferent output, and centrally mediated efferent input. Sound-induced vibrations are detected and converted into electrical conductances by tonotopically arranged sensory hair cells using a process termed mechanotransduction ( Hudspeth and Jacobs, 1979 ). Both cochlear hair-cell types exhibit specialized responses to depolarizations caused by mechanotransduction. In IHCs, this causes the opening of voltage-dependent calcium channels on their basolateral surface which initiates graded release of the excitatory transmitter glutamate onto type I SGNs ( Koyano and Ohmori, 1996 ;Ottersen et al., 1998 ). Similarly, OHCs release glutamate in response to depolarization, but onto type II SGNs within the outer spiral bundles ( Martinez-Monedero et al., 2016 ;Weisz et al., 2009Weisz et al., , 2021; however, they also respond with voltage-dependent contractions and elongations ( Brownell et al., 1985 ) mediated by a voltage-sensitive membrane protein, prestin Zheng et al., 20 0 0 ). This rapid conversion of membrane potential to movement, known as electromotility, generates the mechanical force upon the basilar membrane responsible for cochlear amplification-thereby boosting the traveling wave at the same frequency being detected ( Dewey et al., 2019( Dewey et al., , 2021. Olivocochlear efferents directly impact cochlear function by exert- JID: HEARES [m5G;16:44 ] ing control over the cochlear amplifier and output of the auditory nerve. MOC efferents regulate OHC electromotility by controlling the membrane potential of the cell and its potassium conductance. Terminals from these efferent fibers are cholinergic and transmit to postsynaptic α9/ α10 nicotinic acetylcholine receptors on the basolateral surface of OHCs that have a high permeability for calcium over sodium ( Katz et al., 20 0 0 ;Weisstaub et al., 2002 ). This calcium influx into OHCs leads to an inhibitory efflux of potassium through small-and large-conductance calcium activated potassium channels (SK2 and BK, respectively) Murrow, 1992a , 1992b ;Housley and Ashmore, 1991 ;Kong et al., 2008 ;Rohmann et al., 2015 ). The activation of these channels is likely enhanced by calcium-induced calcium release from intracellular stores ( Lioudyno, 2004 ;Wersinger and Fuchs, 2011 ). The resulting hyperpolarization diminishes the degree of electromotility in response to mechanotransduction, therefore reducing cochlear amplification ( Mountain, 1980 ;Siegel and Kim, 1982 ).

ARTICLE IN PRESS
In addition to acetylcholine, MOC neurons also release an inhibitory transmitter, γ -aminobutyric acid (GABA) ( Eybalin, 1993 ;Eybalin and Altschuler, 1990 ;Maison et al., 2003aMaison et al., , 2006. In the developing cochlea, GABA release from MOC terminals activates presynaptic GABA B receptors, which downregulates the release of acetylcholine ( Wedemeyer et al., 2013 ). While it is unknown if MOC neurons directly release GABA onto SGNs in mature animals, or co-release GABA with acetylcholine, both SGN subtypes are potential targets ( Hua et al., 2021 ;Robertson and Gummer, 1985 ), and express receptors for both transmitters ( Drescher et al., 1993 ;Maison et al., 2006Maison et al., , 2010Malgrange et al., 1997 ;Shrestha et al., 2018 ). Indeed, it was recently demonstrated that type II SGNs are inhibited by bath application of GABA through GABA A receptors ( Kitcher et al., 2022 ). This suggests that activation of MOC neurons may concomitantly reduce the excitability of type II SGNs, and the extent of electromotility in OHCs, using two different neurotransmitters: GABA and acetylcholine, respectively. LOC neurons likely modulate the postsynaptic activity of IHCs by altering the sensitivity and firing rate of auditory nerve fibers, as they directly synapse onto the dendrites of type I SGNs ( Brown, 1987 ;Warr and Guinan, 1979 ). The majority of LOC neurons are cholinergic ( Maison et al., 2003a ;Safieddine and Eybalin, 1992 ); however, they utilize a number of additional neurotransmitters and peptides: dopamine, calcitonin gene-related peptide (CGRP), GABA, urocortin, and opioid peptides (encephalins and dynorphins) ( Darrow et al., 2006a ;Eybalin, 1993 ;Kaiser et al., 2011 ;Vetter et al., 1991 ;Wu et al., 2020 ). While LOC neuron modulation of the auditory nerve has not been directly demonstrated, it is thought that acetylcholine release from their terminals would have an excitatory effect on type I SGNs ( Felix and Ehrenberger, 1992 ;Ito and Dulon, 2002 ;Maison et al., 2010 ). Knocking out urocortin or its receptor, corticotropin-releasing factor receptor type 1, produces abnormal auditory function and increased susceptibility to noise induced hearing loss in rodents ( Graham and Vetter, 2011 ;Graham et al., 2010 ;Vetter et al., 2002 ). Additionally, sound-evoked activity in the auditory nerve is reduced in mice lacking CGRP alpha ( Maison et al., 2003b ); however, this effect could also be attributed to MOC neurons, which also express CGRP ( Cabanillas and Luebke, 2002 ). Together, this suggests that peptidergic LOC neurons play roles in maintaining cochlear function during normal hearing and in response to environmental stressors, such as acoustic trauma. Even so, there is little agreement on how LOC efferents utilize their many transmitters, the mechanisms by which they modulate auditory nerve activity, or the roles played by LOC neuron subtypes ( Darrow et al., 2007 ;Groff and Liberman, 2003 ;Le Prell et al., 2003 ;Liberman, 1990 ;Maison et al., 2012 ). This is likely due to (1) the difficulty in specifically activating or recording from LOC fibers in vivo , and (2) the recent dis-covery by Wu et al., (2020) that they dynamically regulate transmitter expression in response to sound exposure. In that study, expression of tyrosine hydroxylase, an enzyme required for dopamine synthesis, was upregulated in a sound intensity-dependent manner, building upon previous observations in the cochlea ( Niu and Canlon, 2002 ). Additionally, bath application of dopamine reduces auditory nerve activity through pre-and postsynaptic mechanisms ( Ruel et al., 2001 ;Wu et al., 2020 ). Overall, these results suggest that the relative content and effects of transmitter released by LOC terminals in the cochlea are influenced by an animal's most recent auditory experience, possibly confounding the interpretation of previous studies.

Intrinsic properties of efferent neurons
The strength, timing, and duration of efferent activity in the cochlea depends on the intrinsic membrane properties of olivocochlear neurons as these determine their activity at rest, and how they respond to synaptic input. A handful of studies have reported these properties using whole-cell or intracellular recordings from putative or identified olivocochlear somata ( Fujino et al., 1997 ;Robertson, 1996 ;Romero and Trussell, 2021 ;Sinclair et al., 2017 ;Sterenborg et al., 2010 ;Torres Cadenas et al., 2019Robertson, 1997 , 1998 ). However, few studies recorded from MOC or LOC neurons that were genetically tagged with a fluorophore or retrogradely labeled from the cochlea-currently the most reliable ways of identifying them in vitro . For the purpose of this review, we will focus on results from studies that identified auditory efferents in these ways.
MOC neurons are homogeneous in their spike firing properties and are well suited to encoding the overall power of their inputs. They linearly increase their action potential firing rate in response to somatic current injections of increasing intensity ( Fujino et al., 1997 ;Romero and Trussell, 2021 ), which well-reflects MOC fiber responses to sound in vivo , as determined by axonal single-unit recordings at the level of the auditory nerve ( Brown, 1989 ;Cody and Johnstone, 1982 ;Fex, 1962 ;Liberman, 1988 ;Liberman and Brown, 1986 ;Robertson, 1984 ;Robertson and Gummer, 1985 ). Also reflecting in vivo recordings, MOC neurons are typically silent at rest in brain slices prepared from rodents either pre-or post-onset of hearing ( Fujino et al., 1997 ;Romero and Trussell, 2021 ), but may fire spontaneously at ages closer to hearing onset ( Torres Cadenas et al., 2019 ). While in vitro recordings determined that MOC neurons fire action potentials up to ∼300 Hz in response to somatic current injections, in vivo recordings of MOC fibers rarely report rates above 100 Hz ( Brown, 1989 ;Liberman, 1988 ). This is likely because in vivo recordings of efferent fibers are often captured from animals under deep anesthesia, which produces systemic changes in neural transmission and may alter the excitability of MOC neurons and their inputs. In fact, when compared to anesthetized mice or chinchilla, MOC mediated inhibition of distortion product otoacoustic emissions (DPOAEs) is markedly more potent in awake animals ( Aedo et al., 2015 ;Chambers et al., 2012 ), indicating that their firing rate is also increased.
Generally, LOCs linearly encode somatic current injections, but their response is more variable and depends greatly on membrane potential and stimulus intensity. The only study to publish wholecell recordings from identified LOC neurons, Fujino et al., (1997) , excluded neurons whose resting membrane potential were above action potential threshold, -55 mV. This criterion leaves open the possibility that spontaneously firing neurons were not included, suggesting that at least a subset of LOC neurons may fire at rest. Indeed, recent work by Hong and Trussell (2022) demonstrated that, in acute brain slices from juvenile mice, LOC neurons spontaneously fire in bursts at rest, and this activity is driven by calcium- JID: HEARES [m5G;16:44 ] dependent membrane potential oscillations. It may be that, in vivo , many LOC neurons tonically release transmitter in the cochlea, and that their rate of output is modulated, not driven, by their central inputs. This possibility supports the idea that LOC neurons play a homeostatic role in normal hearing ( Darrow et al., 2006b ), likely maintaining a constant level of cochlear sensitivity that dynamically accommodates for the animal's most immediate, or anticipated needs.

Input from Cochlear Nucleus to MOC neurons
The simplest model of the MOC efferent system is a reflex pathway using sound-driven input to provide inhibitory feedback to the cochlea ( Guinan, 2018 ). Sound delivered to one ear elicits both ipsilateral and contralateral MOC-mediated cochlear effects ( Brown, 1989 ;Liberman and Brown, 1986 ;Robertson, 1984 ). However, the range of central cell types participating in this process is likely to be diverse, thus reflecting multiple levels of control. The basic pathway consists of a three-neuron relay that begins and ends in the organ of Corti ( Fig. 1 A and B ), starting with type I SGNs, the first order neurons that project to the cochlear nucleus (CN). The second order neurons of the CN then project to MOC (and probably LOC) neurons, which in turn project back to the cochlea. The precise identity of the CN neurons mediating this relay has been investigated for nearly 50 years. Mounting evidence suggested that they are ascending projection neurons that originate from the posteroventral cochlear nucleus (PVCN) ( Benson and Brown, 2006 ;De Venecia et al., 2005 ;Robertson and Winter, 1988 ;Thompson and Thompson, 1991 ). Using tract-tracing techniques, Tstellate (also called planar multipolar) neurons have been proposed as the second order neuron of this reflex ( Darrow et al., 2012 ;Warr, 1972 ). Anatomically, T-stellate neurons are well-positioned to transmit acoustic signals to efferent neurons as they receive auditory nerve input, their somata primarily reside in PVCN, and they innervate VNTB ( Oertel et al., 2011 ). In response to tones, T-stellate neurons fire sustained action potentials that increase monotonically with sound intensity ( Rhode and Smith, 1986 ;Smith and Rhode, 1989 ), similar to in vivo responses recorded from MOC fibers ( Brown, 1989 ;Liberman and Brown, 1986 ).
We examined this problem recently using transgenic mice, optogenetics and electrophysiology ( Romero and Trussell, 2021 ). Mouse MOC neurons were identified by driving expression of the fluorophore tdTomato in cholinergic neurons. T-stellate neurons were specifically labeled by combining retrograde transport of creexpressing AAV injected into the IC (a target of T-stellate axons) with injection into ventral cochlear nucleus (VCN) of a second virus that expresses a cre-dependent fluorophore and channelrhodopsin. Fluorescent somata were observed in the VCN, and fibers were observed in the VNTB as well as the dorsal cochlear nucleus (DCN). During voltage clamp recording of MOC neurons, excitation of channelrhodopsin with light elicited excitatory postsynaptic currents (EPSCs). These experiments demonstrated 1) that T-stellate neurons do provide excitatory input to MOC neurons, and 2) that these are the same neurons that also project to the IC.
Beside T-stellate neurons, there may be additional input to MOC neurons from the cochlear nuclei. Bushy cells of the VCN send projections with calyceal-like endings to VNTB and RPO ( Smith et al., 1993 ). Using injected tract-tracers, Ye et al., 20 0 0 , determined that the marginal shell of the anteroventral cochlear nucleus (AVCN) may also project to MOC neurons. As discussed in that paper, the cell types of the marginal shell include small neurons generally known for only local projections within the CN, such as granule cells, as well as larger stellate neurons; moreover, neurons, such as T-stellate, with axon collaterals or dendrites near the injection site may also have been labeled. Hockley et al., (2022) used retrograde dye injection in VNTB and nearby regions in guinea pig to suggest that multipolar cells in the small cell cap may project to the VNTB. By contrast, retrograde transneuronal tract-tracing from MOC neurons using pseudorabies virus abundantly labeled many VCN neurons, but not the marginal shell region ( Brown et al., 2013a ). Additionally, kainic acid lesions of PVCN, but not AVCN, impeded the ipsilateral MOC reflex ( De Venecia et al., 2005 ). Cell specific markers of the different projection neurons of the CN will be required to identify the full spectrum of cell types innervating MOC neurons.
The ipsilateral reflex pathway is often described as "double crossed" with respect to the midline of the brain ( Guinan, 2010( Guinan, , 2018Warr, 1992 ), as (1) the second order interneuron axons cross the midline and synapse onto contralateral MOC neurons, which then (2) cross back to the ipsilateral cochlea (crossed MOC, Fig. 1 A ). MOC neurons that receive crossed interneuron input but do not cross the midline themselves are thought to be involved in the contralateral reflex (uncrossed MOC, Fig. 1 A ). However, the interneuron projections are incompletely described and likely also provide ipsilateral input to MOC neurons ( Thompson and Thompson, 1991 ). Indeed, at least some Tstellate neurons project to ipsilateral VNTB ( Darrow et al., 2012 ;Doucet and Ryugo, 2003 ;Warr, 1995 ) and synapse onto ipsilateral MOC neurons ( Romero and Trussell, 2021 ).
In silent backgrounds, most olivocochlear fibers respond only to ipsilateral or contralateral sound stimuli but not the other ( Brown, 1989 ;Liberman and Brown, 1986 ). However, a great majority of these same unilaterally activated fibers respond more robustly when sound is presented to both ears simultaneously ( Liberman, 1988 ). This suggests that many MOC neurons either receive interneuron input from ipsilateral and contralateral VCN, or that binaural stimuli utilize a pathway which differs from that of the canonical reflex pathway.
Physiological studies have sought to characterize the transmitter and receptors used by CN input to MOC neurons. Tstellate cells are glutamatergic neurons, expressing the vesicular transporters VGluT1 and VGluT2 ( Ito and Oliver, 2010 ). Robertson (1996) recorded from rat VNTB neurons and observed excitatory postsynaptic potentials (EPSPs) triggered by midline stimulation. These were blocked by the α-amino-3-hydroxy-5methyl-4-isoxazolepropionic (AMPA) receptor antagonist CNQX. Romero and Trussell (2021) recorded excitatory postsynaptic currents (EPSCs) in voltage clamped MOC neurons upon optogenetic activation of T-stellate cell axons. These were also blocked by AMPA receptor antagonists. Notably, EPSCs had inwardly rectifying current-voltage relations and were blocked by IEM1925, both properties of AMPA receptors lacking the GluA2 subunit. As GluA2 lacking receptors are calcium permeable, it is suggested that ongoing auditory synaptic transmission may challenge MOC neurons with a significant calcium load that could potentially mediate homeostatic plasticity, normalizing synaptic strength in the face of varying levels of activity.

Inhibitory inputs to MOC neurons
It is likely that MOC neurons receive synaptic inhibition from neurons within the SOC. In rat, Robertson (1996) JID: HEARES [m5G;16:44 ] from the MNTB, as IPSCs were evoked by glutamate uncaging over this nucleus, thereby locally activating somata and dendrites but not axons. As the MNTB is thought to be exclusively glycinergic ( Fischer et al., 2019 ), the remaining GABAergic component must have another source. Interestingly, Albrecht et al., (2014) revealed a population of glycinergic neurons within the mouse VNTB whose axonal projections make synaptic contact on ipsilateral MNTB principal neurons. They suggested that these had a mixed transmitter phenotype in young mice, graduating to being predominantly glycinergic with age. Although this study did not show single-axon projection patterns, it is conceivable that these neurons could have collaterals that provide further input to MOC neurons, and perhaps account for some GABAergic components observed by Torres Cadenas et al., (2019) . Indeed, Kulesza and Berrebi (20 0 0) note that, in rat SOC, labeling for glutamic acid decarboxylase (GAD) is most dense in the VNTB. However, other olivary GABAergic sources could be the SPN ( Kulesza and Berrebi, 20 0 0 ) or neurons distributed ventral to the MNTB and LSO ( Adams and Mugnaini, 1990 ). Elucidating the sources of inhibition to MOC neurons and their pattern of activation in vivo will be a key aspect of efferent research.

Auditory cortex
Tracer experiments have provided strong evidence that the SOC receives abundant descending input from auditory cortex, and one of the most prominent targets of these fibers is the VNTB ( Coomes and Schofield, 2004 ;Feliciano et al., 1995 ), although cortical fibers are found in most divisions of the SOC. The presence of these fibers in the VNTB suggests that some or most may terminate on MOC neurons. Multiple auditory cortical regions participate in this projection, and a variety of cortical axon terminal types were found in the SOC ( Coomes and Schofield, 2004 ), suggesting heterogeneity in these descending signals. Doubletracer experiments by  showed that these descending inputs do indeed terminate on MOC neurons. Horvath et al., (2003) and Brown et al., (2013) used pseudorabies virus to specifically infect olivocochlear neurons, thereby transneuronally labeling presynaptic partners, and observed obvious labeling of layer 5 pyramidal neurons of auditory cortex. Finally, doubletracer injections showed that, while cortical neurons descend both to the SOC and the CN, these projections arise from different neuronal populations ( Doucet et al., 2002 ). Functional studies are needed to determine the actions of this well established and specific projection from auditory cortex to cochlear efferent neurons.
In an EM study, Benson and Brown (2006) reported three types of endings on MOC neurons: terminals with large round vesicles, with smaller size round vesicles, and those with irregular, pleiomorphic vesicles. They speculated that the terminals with smaller vesicles might originate from the second order neurons in the CN. Assuming the pleiomorphic vesicles are inhibitory, the population of terminals with larger vesicles might represent descending inputs. Suthakar and Ryugo (2017) reported that inputs from IC featured large round vesicles but did not quantify them. Still, these differences are in line with the distinct presynaptic physiology of CN-vs IC-sourced inputs, as described below. On the postsynaptic side, however, EPSPs mediated by IC activation utilize the same type of transmitter receptor as CN inputs, i.e., GluA2-lacking, Ca 2 + -permeable AMPA receptors ( Romero and Trussell, 2021 ). Finally, careful tracer studies indicate that the descending input from IC is tonotopically organized in the VNTB/RPO, with a high-to-low frequency map extending medial to lateral ( Caicedo and Herbert, 1993 ;Malmierca et al., 1996 ;Suthakar and Ryugo, 2017 ). Thus, there are tonotopic projections from CN and IC onto MOCs, but it remains unclear whether the narrowness of tuning of these projections is similar.
A striking difference between input to MOC neurons coming from the CN or IC is in their short-term plasticity ( Romero and Trussell, 2021 ). We found that CN inputs, when activated at 20-50 Hz, exhibited a synaptic depression whose recovery required several seconds. By contrast, IC inputs exhibited a buildup of response with the same frequency of stimulation, also lasting several seconds. Using conductance clamp, we showed that, in fact, the facilitating IC input was more potent in driving MOC activity than the CN input, suggesting that top-down control of efferents may be a more dominant function than previously appreciated.

Modulatory inputs
The SOC receives neuromodulatory input from many brainstem nuclei ( Beebe et al., 2021 ;Thompson and Hurley, 2004 ), some of which directly contact MOC and LOC neurons, and these inputs presumably regulate efferent action in response to changes in global brain state. Serotonergic and noradrenergic varicosities have been detected in close apposition to both types of olivocochlear neurons using tract-tracing and histochemical techniques ( Mulders and Robertson,20 0 0b , 20 05 ; Thompson and Thompson, 1995 ;Woods and Azeredo, 1999 ), and likely originate from raphe nuclei and locus coeruleus, respectively ( Brown et al., 2013a ;Horvath et al., 2003 ;Mulders and Robertson, 20 01 , 20 05 ). There is also evidence for peptidergic input, as terminals containing substance P, urocortin 3, cholecystokinin and encephalin appear to be in close apposition to MOC and/or LOC somata and dendrites Mulders and Robertson, 20 0 0b ). The source of these peptidergic inputs is unknown, but substance P input may come from the dorsal raphe or IC ( Ljungdahl et al., 1978 ;Wynne et al., 1995 ).
Physiological evidence suggests that auditory efferents are engaged by cognitive tasks, such as selective attention to auditory or visual stimuli ( Aedo et al., 2015 ;Giraud et al., 1997 ;Oatman, 1971Oatman, , 1976Smith et al., 2012 ;Terreros et al., 2016 ;Walsh et al., 2015 ). Noradrenergic inputs from the locus coeruleus possibly mediate these actions as this region is implicated in arousal and attention ( Aston-Jones and Bloom, 1981 ;Aston-Jones et al., 1999 ). Supporting this idea, Robertson (1997 , 1998 ) found that bath application of noradrenaline and substance P onto voltage-clamped MOC neurons generally elicits an excitatory response, suggesting these types of input sensitize or initiate efferent signaling. It is unknown when or how serotonin mediated input effects efferent function; however, serotonergic receptors are present on MOC neurons ( Ohata et al., 2021 ), and their activation increases MOC excitability ( Suthakar and Weisz, 2022 ). As modulatory agonists seem to generally trigger an excitatory response, it may be that efferent sensitivity is indeed heavily dependent on an animal's global brain state. Moreover, concurrent activation of multiple modulatory systems may be required to achieve maximal MOC mediated inhibi- JID: HEARES [m5G;16:44 ] tion of the cochlear amplifier, allowing MOC fibers to fire at much higher spiking rates than previously reported (see section 3.3 ).

Excitatory and inhibitory inputs to LOC neurons
The sources of input to LOC neurons is less well understood than for the MOC system. Sterenborg et al., (2010) explored the types of fast synaptic inputs to LOC in a mouse brain slice preparation. Although they did not identify LOC neurons by acetylcholine expression or cochlear targeting of axons, they were able to distinguish presumptive LOC neurons from presumptive LSO principal cells by their differences in size and intrinsic properties. Electrical stimulation of fibers in the MNTB elicited GABA/glycinergic IPSCs while stimulation of fibers lateral to the LSO evoked glutamatergic EPSCs. The picture here is that, like MOC neurons, the LOC system is excited by the CN and inhibited through the MNTB. Recent optogenetic experiments confirm this hypothesis in part, showing that T-stellate cells provide excitatory input to cholinergic neurons within the LSO ( Hong and Trussell, 2021 ). This source of input is consistent with a tracer study showing T-stellate neurons projecting to LSO, in addition to bushy cells ( Doucet and Ryugo, 2003 ).
Results from an elegant tracing study from Gómez-Álvarez and Saldaña (2016) are consistent with these views. Dye injections in rat LSO labeled, as expected, cell bodies of spherical bushy cells of CN and principal cells of MNTB. However prominent labeling was also observed in T-stellate cells, suggesting they also project to LSO, and thus could be an excitatory source to LOC neurons. Intriguingly, these injections also labeled a subset of small neurons in VNTB, distinct from MOC cells, opening the door to an additional source of excitation or inhibition in LOC neurons.
The descending control of LOC neurons has received little attention. Among the projections from cortical pyramidal cells, the input to LSO is sparse, although consistent ( Coomes and Schofield, 2004 ;Feliciano et al., 1995 ). In contrast to the dense input from IC to the VNTB, anatomical studies have not reported substantial IC input to the LSO ( Caicedo and Herbert, 1993 ;Faye-Lund, 1986 ;Suthakar and Ryugo, 2017 ;Thompson and Thompson, 1993 ;Vetter et al., 1993 ). However, Groff and Liberman (2003) observed that stimulation of IC could sometimes exert long lasting efferent effects that were attributed to the LOC system.
Regarding the physiology of postsynaptic responses in LOC neurons, LOC IPSCs described by Sterenborg et al., (2010) are of similar duration to those reported for MOC neurons by Torres Cadenas et al., (2019) , with decay constants of about 10 ms, while their EPSCs decay much more slowly (LOC: 4-5 ms; MOC: < 1 ms). The differences in excitation suggest different integrative capacities of the two efferent cell types. Regarding receptors mediating these currents, Sterenborg et al., (2010) reported that glycine receptors, not GABA receptors, mediated inhibition in LOC neurons, in contrast to the results from MOC neurons ( Torres Cadenas et al., 2019 ). These authors also reported that EPSCs were mediated by AMPA receptors, and that these showed inward rectification characteristic of Ca 2 + -permeable AMPA receptors. More recently, Hong and Trussell (2021) , recording from cholinergic neurons in LSO (presumably LOC neurons), found that EPSCs did not show inward rectification, and moreover, that with repetitive stimulation, the synaptic current was mediated by both AMPA receptors and kainate receptors. The presence of kainate receptors confirms an earlier study ( Vitten et al., 2004 ) that demonstrated kainate receptor-evoked currents in most neurons of the LSO. Caicedo and Eybalin (1999) explored AMPA receptor subunit distribution in LSO, observing expression of all 4 subunit genes, but noting a preponderance of GluA1 in "small fusiform cells", which might correspond to LOC neurons. Future studies will need to distinguish receptor expression with the source of excitatory and inhibitory input.

Efferent collaterals in the cochlear and vestibular nuclei
It has long been thought that collaterals of cochlear efferents release acetylcholine in the brain ( Comis and Guth, 1974 ), and thus a fascinating aspect of the efferent system is the potential impact of central efferent collaterals on auditory processing. These secondary branches of efferent neurons are interesting because they may potentially refine the actions of auditory processing at a level that is specific for the different parallel processing streams set up in the brainstem. For example, it may be that excitatory actions of MOC transmission in CN could play a role in the improved perception of signal in noise that is a hallmark of MOC function ( Benson and Brown, 1990 ). Below we review the anatomical and physiological properties of these branches.

Axon pathways
The collaterals of MOC neurons in brainstem have been extensively studied. The initial target of these axons is the vestibular nucleus ( Fig. 1 A ). Baashar et al., (2015) , Brown et al., (1988) , and  , described thick branches of MOC axons entering the internal vestibular nucleus of cat and rodent, where they terminate onto dendrites with multiple endings per axon. They suggested that this input might modify vestibular activation by intense sound, potentially enhancing muscle reflexes as a feedback mechanism.
MOC fibers branch further as they proceed laterally, and these branches enter the CN through several routes, a strial, subpeduncular and ventral approach ( Osen et al., 1984 ), with the major branch entering just medial to the VCN. The primary projections of MOC axons are in granule cell domains and regions of CN near those domains. Auditory granule cell domains of the CN are distributed across up to 7 subregions and are found broadly across mammalian species ( Mugnaini et al., 1980 ). Auditory granule cell somata essentially surround the VCN medially, dorsally and laterally, extend into the DCN, and are found in the peduncular zone above DCN. The primary course of MOC axons is within a medial granule domain and into the granule cell lamina dividing VCN and DCN, although some fibers are also seen in the main body of VCN and DCN, including the region of VCN known as the small cell cap, adjacent to the granule cell lamina ( Baashar et al., 2015 ;Brown and Benson, 1992 ;Brown et al., 1988 ;Shore et al., 1991 ).
What cell types are innervated by these collaterals? Brown and colleagues indicated that collaterals primarily target thick dendrites of multipolar neurons extending into the lamina/medial regions, including the small-cell cap ( Benson and Brown, 1990 ;Brown, 1993 ). Comparatively few terminals were on 'small cells' or granule cells. Baashar et al., (2019) also concluded that multipolar, possibly D-and T-stellate cells, receive MOC input. Using an analysis of vesicle diameters in EM preparation,  showed that the majority of these terminals have small round vesicles. Remarkably, the terminals with the large vesicles typical of afferent fibers did not appear on these dendrites, suggesting either that some neurons only receive MOC and not afferent input, or that these inputs are distributed on different dendrites of the same cells. Presumptive granule cell dendrites ('varicose dendrites') hosted of mix of terminals with either small-round or pleomorphic vesicles, the latter presumably being inhibitory. These results suggest that despite the presence of collaterals within or near the granule cell lamina, larger multipolar VCN cells are primary targets of MOC collaterals. Benson and Brown (2004) used electron microscopy to characterize the terminations of type II auditory nerve fibers in mouse, concentrating on the region of VCN just ventral to the granule cell lamina. This thin region is composed of small and large cells and both received type II input. Additionally, it was suggested based on terminal profiles, that olivocochlear efferent terminals converged JID: HEARES [m5G;16:44 ] onto these same cells. Ending of type I auditory nerve fibers were only found on the larger dendrites, presumably of the large multipolar cells. They suggested that signaling from the type II fibers report OHC function while the efferent contacts present a corollary of information sent back to the inner ear. Gómez-Nieto et al., (2008) have described a population of cholinergic neurons in the VNTB that project to, and synapse on, cochlear root neurons, a cell population embedded in the auditory nerve just outside the CN. These projection neurons seem not to be MOC neurons, and their function appears to be in mediation of the acoustic startle reflex rather than cochlear tuning.

ARTICLE IN PRESS
The thin axons of LOC intrinsic and shell neurons make few collaterals before exiting the brain. Experiments with dye tracing showed only a small fraction of cases in mouse where collaterals were observed within or near LSO, and in the vestibular nucleus ( Brown, 1993 ). Horváth et al., (20 0 0) suggested that rat LOC neurons make no collaterals in the CN, but shell neurons do. By contrast, a different study in gerbils, using 3 H-aspartate, argued for both LOC and MOC projections to the CN ( Ryan et al., 1990 ). In balance, it seems that LOC collaterals probably do not play a major role in modulating brainstem level auditory processing compared to the prominent projections from MOC axons.

Central efferent actions
Mulders and colleagues have explored the effects of MOC axon stimulation on the excitability of VCN neurons. In experiments using either extracellular or intracellular recordings in which cell types were defined by their characteristic peristimulus time histograms, stimulation of MOC axons resulted in either inhibition or excitation of neurons. Excitation or inhibition were defined either as shifts left or right in input-output curves for auditory stimuli ( Mulders et al.,20 02 ,20 08 ), or directly observed orthodromic EP-SPs/spikes or IPSPs ( Mulders et al.,20 03 ,20 07 ,20 09 ). This mixture of response profiles might be expected if MOC synapses were excitatory and triggered local inhibition by interneurons within the VCN. A caveat is that the fiber stimulation used in these studies may also have activated descending cholinergic inputs from midbrain ( Mellott et al, 2011 ). A consistent theme is the presence of excitation in 'onset-chopper' neurons, neurons that are generally identified as the glycinergic D-stellate cell, a broadly tuned inhibitory neurons that projects to DCN and contralateral CN ( Brown et al., 2013b ;Oertel and Wu, 1989 ).

Chemosensitivity
Since MOC neurons are cholinergic, the presence of cholinergic receptors and responses in VCN provide another clue to central targets and functions of efferent axons. Oertel and colleagues have compared responses of different stellate cells in mouse VCN to cholinergic agonists ( Fujino and Oertel, 2001 ;Oertel et al., 2011 ; see also Bal et al., 2010 ). Depolarizing effects of nicotinic agonists were observed with T-stellate cells but not D-stellate cells ( Fujino and Oertel, 2001 ), an observation confirmed by us ( Ngodup et al., 2020 ). This contrasts with the recordings from Mulders and colleagues described above which highlight D-stellate cells as responding to an excitatory efferent transmitter. It is possible that the in vivo response to MOC stimulation represented disynaptic input, with excitation coming via T-stellate cells. On the other hand, the bath applications used by Oertel and colleagues might have resulted in desensitization of some subtypes of neuronal nicotinic receptor, which is avoided during rapid exocytosis of transmitter. It is also possible that in some cells, activation of cholinergic receptors might arise from spillover from inputs to neighboring neurons, and thus not require a direct synaptic contact.
Although identified MOC inputs to bushy cells have not been described morphologically, Goyer et al., (2016) has examined cholinergic effects on VCN bushy cells, observing a mixture of actions attributable to a mixture of nicotinic and muscarinic receptor subtypes. In brain slices, they observed tonic activation of muscarinic receptors, suggesting a cholinergic 'tone'. In vivo , application of high concentrations of acetylcholine from pipettes enhanced bushy cell spike rates.
In the DCN, a variety of effects of cholinergic agonists on cartwheel and fusiform cells have been described (reviewed by Trussell, 2019 ), but it is not clear if these effects are associated with MOC inputs, since the DCN is not a prominent target of MOC axons. Nevertheless, as the sole target of the granule cell system is the DCN ( Mugnaini et al., 1980 ), it is possible that MOC activation indirectly modulates DCN processing. In this regard, it is intriguing that K őszeghy et al., (2012) showed in rat DCN granule cells that the frequency of spontaneous Ca 2 + transients, which presumably reflect spontaneous spike activity, is enhanced by a muscarinic M3 receptor agonist. Both muscarinic M1 and M3 receptors could be resolved in CN immunohistochemically. Related is the observation of inhibitory effects of muscarine but not nicotine on Golgi cells which populate granule cell domains of CN ( Irie et al., 2006 ). Intriguingly, these authors described an atropine sensitive, slow inhibitory synaptic response upon electrical stimulation of slices, suggesting that Golgi cells are targets of cholinergic synapses.

Convergence with other cholinergic systems
In general, cholinergic fibers from other brain regions may also provide input to CN, complicating the interpretation of studies in which cholinergic agonists are applied. Schofield and colleagues have used tract tracing methods to document the various sources of cholinergic input to CN ( Mellott et al., 2011 ). In terms of numbers of projecting neurons, 74% of input came from the MOC system, with the remainder largely from the pedunculopontine tegmental nucleus and laterodorsal tegmental nucleus. In general, labeling for presynaptic and postsynaptic markers of cholinergic transmission reveals evidence for cholinergic fibers and receptors throughout the CN ( Gillet et al., 2018 ;Yao and Godfrey, 1999 ), including fibers terminating on or near bushy cells. As the MOC projections in the studies cited above appear to be more restricted in their scope, it is likely that this widespread labeling is reflective of non-auditory cholinergic sources.

Summary and speculations
While a tremendous amount of work has examined the functions of the medial efferent neurons in modifying cochlear function, there remain major gaps in our understanding of the efferent system. The function(s) of the lateral efferent system is one obvious gap, made all the more challenging by the difficulty in selectively stimulating their fibers and in the diversity of potential transmitter actions they unleash in the cochlea. Our understanding of the central circuitry of the efferent system suggests an essential area for future work is in how higher brain regions interact with the 'cochlear reflex' to tune efferent activity under different conditions. The action of efferent collaterals in the CN is also a mystery at this time -could they be sensitizing a subset of CN neurons in parallel with a selective dampening of cochlear activity, or does it serve as a feedback control for the efferents themselves? Future studies armed with genetic markers of individual cell types for stimulation, recording and anatomical study, may throw needed light on these issues.