Gain Control

Available online at www.sciencedirect.com

Gain control mechanisms in the auditory pathway Benjamin Louis Robinson and David McAlpine Belying the apparent ease with which the acoustic world is perceived, the sheer vastness of the range of sounds and sound parameters that must be encoded represents a challenge to traditional models of neural coding in audition. Here, we review recent evidence suggesting that a process of gain control, operating at multiple stages in the auditory pathway, helps maintain coding accuracy to prevailing sound conditions over a wide range of behavioural and sensory contexts. Together, these processes imbue the system with its staggering representational capacity, underpinning everything from the perception of a tiger’s near-silent tread to its triumphant roar, demonstrating once more the principle of efficient coding that underlies sensory processing. Address UCL Ear Institute, 332 Gray’s Inn Road, London WC1X 8EE, UK Corresponding author: McAlpine, David ([email protected])

Current Opinion in Neurobiology 2009, 19:402–407 This review comes from a themed issue on Sensory systems Edited by Leslie Vosshall and Matteo Carandini Available online 6th August 2009 0959-4388/$ – see front matter Published by Elsevier Ltd. DOI 10.1016/j.conb.2009.07.006

Introduction The environment imposes two conflicting demands upon sensory systems. On the one hand, natural signals are often of extremely small magnitude, favouring the evolution of sensory systems with high sensitivity [1]. On the other hand, the range over which these signals varies is anything but small: the difference between a whisper and a roar, for example, is 10 or 12 orders of intensity [2]. Given biological constraints, how do sensory systems cope with the simultaneous requirements of high sensitivity and accuracy over a wide dynamic range? A possible solution to the conflicting demands of sensitivity and accuracy lies in the implementation of gain control, a solution variously instantiated throughout many sensory systems, including the auditory system. Strictly, the term ‘gain’ refers to the slope of an input–output function, and it is well established that a neuron’s response or output gain can be rapidly adjusted in a process referred to as divisive normalisation [3]. However, the evident complexity of any relationship between intracellular, sub-threshold activity and spike generation suggests that, for purely technical reasons, it is not always Current Opinion in Neurobiology 2009, 19:402–407

possible to distinguish changes in gain from other changes that influence neural output functions, especially when relying on extracellularly recorded neural data to make such distinctions [4]. What has become clear, however, is that changes in gain often imply neural adaptations that result in the recoding of sensory information according to current environmental demands. Here, we review recent evidence demonstrating the importance of gain control in the auditory pathway — and the adaptive recoding such gain control implies — from the external surface of the ear to the primary auditory cortex. We propose that the multiple timecourses over which changes in neural gain are implemented, as well as the burgeoning number of candidate effectors of gain change, challenge the traditional means of assessing auditory coding using relatively low-level acoustic cues. What emerges is an exquisite degree of coding control fitted to the stimulus context, even at very early stages of neural processing [5].

Gain and the peripheral auditory system Aside from its protective function, the outer ear performs at least two significant acoustic functions in hearing. Not only does its complex geometry provide for a frequency-dependent increase (or decrease) in sound energy reaching the eardrum, but also this transformation in sound energy (the head-related transfer function or HRTF) affords the means by which sound-sources in the vertical plane are located, including the ability to distinguish sources to the front from those to the back [6]. In a clever demonstration of the auditory brain’s sensitivity to the spatial cues provided in the HRTF, Hofman et al. [7] found that by altering the shape of the outer ear with prosthetic molds they could modulate the extent to which human listeners were able to identify the location of the source of a sound along the vertical dimension. Within a matter of weeks, however, subjects chronically fitted with outer ear prostheses had adapted to their new auditory cues, and were once more able to distinguish sources above the horizon from those below, indicating a significant recoding of the brain’s representation of the surface of the external ear. This, relatively long-term, process of adjustment clearly implicates even the most peripheral components of the sensory end organs in gain control. Interestingly (and unlike the effect of employing prisms to adjust visual spatial cues), upon removal of the prostheses, subjects were found to have retained normal sensitivity to the original acoustic cues generated by their own ears. A more rapid means by which the auditory system demonstrates the influence of gain lies in the process of sensory www.sciencedirect.com

Gain control mechanisms in the auditory pathway Robinson and McAlpine 403

transduction that occurs at the level of the cochlea of the inner ear. Transmission of sound energy via the bones of the middle ear (themselves subject to a form of input gain control through the action of the middle-ear reflex) to the oval window of the cochlea generates a travelling wave along the coiled basilar membrane. Current investigations focus on understanding cellular mechanisms underlying particular properties of basilar membrane responsiveness, including the appearance of high gain at low intensities, reduced, or compressive, gain responses at mid-intensities to high-intensities, and the contribution of these gain mechanisms to the basilar membrane’s sharply tuned frequency response — none of which is present in the passive properties of the membrane (see [8]). These changes in gain, important in themselves, accrue further interest in that they are thought to underlie certain psychophysical phenomena (see below). So what is it that provides the cochlea with its boost in sensitivity? The likely candidates, the outer hair cells (OHCs), outnumber the primary auditory transducers, the inner hair cells (IHCs), by three to one, and appear to increase cochlear gain by rapidly changing somatic length, thereby altering the micromechanical properties of the basilar membrane, rather like springs of adjustable tension attached to the edges of a trampoline. In this way, the travelling wave undergoes an overall amplification, a compression at high levels and a sharpening at the particular region of the basilar membrane representing the frequency of the incoming sound. Recent research has focused on the precise mechanism by which OHCs feed energy back into the basilar membrane [9]. The details of this mechanism are beyond the scope of the present discussion and have been recently reviewed [10,11].

Gain control in the auditory midbrain With its active elements intact, the dynamic range of the cochlear input–output function matches well with psychophysical data indicating that listeners are able to distinguish changes in sound level in the order of 1 decibel (dB) or so over the full 120 dB range of human hearing. This includes the existence of a ‘mid-level’ hump in discrimination performance — a reduction in sensitivity over the range of sound levels where cochlear compression is greatest [12]. In this context, therefore, measurements of single ANFs, as well as in the midbrain and cortex, have tended to reveal a somewhat surprising picture. Typically, increases in neural spike-counts are reported to extend over only the first 20–30 dB beyond threshold before they reach rate saturation. Given that the vast majority of ANFs show thresholds close to 0 dB sound pressure level (SPL) — elevated threshold generally indicates some form of pathology — the possibility that ANFs represent sound level in the form of a rate code was disputed until only very recently. The ‘dynamicrange problem’, a term coined to express this apparent mismatch between psychophysical and neural performance, suggests an essentially static auditory system, one in www.sciencedirect.com

which response functions at the earliest level of neural coding in the auditory pathway are fixed in their representation of the acoustic environment, suited to detecting and discriminating low-intensity sounds, but incapable of distinguishing sounds at moderate or high levels. Indeed, only recently was it suggested that sound level could not be realised through a spike-rate code alone, and that other forms of information (e.g. temporal) must play a role in encoding sound level at the earliest neural stage of processing. More recent studies, however, suggest that the dynamic-range problem is simply the outcome of employing a form of acoustic stimulation in physiological studies that does not take full account of psychophysical or natural listening conditions. Motivated by studies of contrast gain control in the visual system demonstrating the means by which neurons with relatively small neural dynamic ranges can encode changes in contrast extending over many orders of magnitude, Dean et al. [13] assessed responses of neurons in the auditory midbrain nucleus of the inferior colliculus (IC) to a novel stimulus configuration. Hypothesising that adaptation might result in alterations to coding of sound intensity at the single-neuron level, these authors recorded responses of individual IC neurons to continuous wide-band noise whose level was adjusted every 50 ms to a value drawn from a statistical distribution, defined by a high-probability region (HPR) from which the majority (80%) of sound levels were drawn. When the HPR was shifted to a new value, neurons’ spike-rateversus-level functions also shifted to accommodate the new stimulus distribution, response functions continuing to shift ‘rightwards’ as the HPR was shifted to higher and higher levels. Moreover, as a population, neurons adapted their response functions so as to maximise the accuracy with which sound levels just above the background mean were encoded. Neurons were found to adjust their coding to new sound-level regimes within a few hundreds of milliseconds [14]. In this case, changes in gain are manifest as a precise and rapid reconfiguration of the relationship between the input sound level and the neural output — an adaptive recoding — to maximise the precision with which the sound environment is represented by the auditory system. With recent conference reports now indicating adaptive recoding to be a feature of the responses of primary nerve fibres, the notion of a dynamic-range problem in hearing has been much diminished. What was previously considered a coding problem is now recognised as having constituted a limitation in our understanding of how the brain represents information — a limitation based on previously standard laboratory technique that dictated that sounds be presented in pseudo-random order, with long intervals between successive presentations in order to lessen the presumed malign influence of any adaptive mechanisms on the proposed neural representation of intensity. Released from this constraint, the auditory brain’s ability Current Opinion in Neurobiology 2009, 19:402–407

404 Sensory systems

to extract information about the sound environment effectively and efficiently, became evident [15]. Recent psychophysical evidence supports these findings [12]: Pienkowski and Hagerman measured intensity discrimination in listening conditions selected so as to prevent central mechanisms of adaptation from having any useful effect on the representation of sound level. They did so by keeping sounds very brief (4 ms), and by employing sequences of sounds widely separated in level. The data from this study are striking: without central adaptive mechanisms, intensity discrimination worsens significantly at mid-range levels. The authors invoke cochlear mechanisms to explain this so-called ‘mid-level hump’, suggesting that the mid-range compressive non-linearity in the cochlear response, described above, is a limiting factor in sound-level discriminability across this range. Central mechanisms, and then operating only for stimuli of sufficient duration, are required to overcome this decrement in performance for mid-level sounds. This illustrates a remarkable point: it appears that peripheral gain changes are necessary to allow the auditory system to embrace as wide a range of levels as possible, but this compression leads to a potential loss of accuracy; however, subsequent, central changes in gain — such as those seen by Dean and colleagues — may largely compensate for these necessary peripheral losses.

Cortical gain We began by proposing gain as a possible solution to the problems inherent in sound perception, namely the need to encode small changes accurately over a large range of sound intensities, including the very quiet. As suggested above, the cochlea and sub-cortical auditory brain structures appear to address these problems to a great degree, and it might therefore be reasonable to ask, in terms of gain control, what role is left for the cortex if its inputs are already well-matched to the statistics of incoming sound stimuli? It should come as no surprise that cortical neurons are in fact subject to significant changes in both the gain and the form of their receptive-field properties, and that these changes are initiated by a remarkable range of sensory and non-sensory factors that, taken together, provoke a profound reconsideration of the notions of compression and dynamic range described for lower brain centres. Before discussing purely cortical forms of gain control, it is important to note that some forms of gain control are similar in sub-cortical and cortical brain centres. For example, Phillips and Hall [16] report that rate-versuslevel functions for pure tones shift to higher levels when presented against background noise. Rees and Palmer [17] observe similar shifts sub-cortically, in the IC. These shifts may aid the neural population in responding to important sounds in noisy environments, such as picking out a voice at a cocktail party. However, the possible coding implications have not been explicitly examined on Current Opinion in Neurobiology 2009, 19:402–407

a population level, rendering the functional implications of these studies less clear than those of Dean et al. [13]. The range of shifts may share certain mechanisms; however, these changes in tone-evoked rate-versus-level functions are thought to be due in part to cochlear two-tone suppression — a very rapid mechanism that is unlikely to contribute to the data of Dean et al., who employed wide-band noise as a stimulus [13]. Further, adaptation of the type seen by Dean et al. operates over a time-scale of hundreds of milliseconds, well beyond the temporal reach of two-tone suppression [14]. This brief comparison illustrates the idea that gain control is useful as a phenomenological description, but that mechanisms and functions are likely to be manifold. Given this complexity, it is important to try and establish clear links between mechanistically similar forms of gain control at different levels of the brain. Indeed, adaptation resulting from the wide-band stimulus used by Dean et al. in the IC has been examined explicitly in the cortex by Watkins and Barbour [18], who reported an additional feature of neural adaptation to sound-level statistics in cortical neurons to that reported for IC neurons. A unique feature of intensity coding in the auditory system is the appearance of non-monotonic rate-versus-intensity functions. Absent at the level of the auditory nerve, the proportion of non-monotonic functions increases at subsequent stations in the ascending pathway, such that by the level of primary auditory cortex, fully half of neurons exhibit some form of tuning for sound intensity. Cortical neurons were found to respond in a manner that preserved coding accuracy for quiet sounds, even in the context of a stimulus comprising mainly high sound levels. Essentially, high sound levels lead to few if any spikes being evoked in these neurons, rendering them in a relatively un-adapted state. Low sound levels evoke neural activity that, whilst maladapted to the high levels in the underlying statistical distribution, is well placed to encode the appearance of quiet sounds within the distribution. The cortex therefore prevents adaptive recoding occurring in some neurons in loud environments, to preserve responsiveness to the sudden appearance of quiet sounds. In addition to such stimulus-driven changes in gain, cortical neurons appear sensitive to non-sensory effectors of gain change in their representation of complex sounds. One of these effectors appears to be behaviour itself. In a powerful demonstration of the effects of task-related receptive-field changes, Fritz et al. [19] measured spectro-temporal receptive fields (STRFs; time-dependent and frequency-dependent receptive fields) of primary cortical neurons in ferrets trained to perform an auditory detection task. Neural STRFs were derived before, during and following training periods in which animals were required to detect the substitution of a complex broadband sound with a pure tone. Remarkably, individual STRFs www.sciencedirect.com

Gain control mechanisms in the auditory pathway Robinson and McAlpine 405

were mutable during the behavioural testing, particularly for frequencies at or close to that of the test tone. In general, Fritz et al. found facilitation in STRF frequencies close to the target frequency, changes which were swift in onset and, in some cases, lasted for many hours. Whilst the power normalisation procedure employed in this study precludes any clear differentiation between changes in STRF gain and changes in receptive-field shape, the results do suggest highly localised increases in gain close to the tone frequency, and decreases away from it. The link between behaviour and STRF change was further demonstrated by the observation that the magnitude of changes was correlated with task performance — better performance equating with larger changes in the gain and shape of receptive fields close to the frequency of the target tone. Further studies by the same laboratory have extended these findings to reveal a host of factors that evoke plasticity in

cortical receptive fields, including the type of task undertaken (detection versus discrimination [20]), task difficulty (high versus low signal-to-noise [21]) and stimulus complexity (single versus complex tones [22]). Overall, a picture is emerging indicating that auditory cortical neurons can rapidly adjust their gain and reconfigure their receptive-field properties in response to the moment-tomoment demands of the sensory–motor milieu. Research in species other than the ferret, and from other laboratories, supports this notion, demonstrating a range of effects from pure stimulus-driven gain changes [23–25] to alterations in A1 receptive fields induced by self-stimulation of cortical pleasure centres [26]. To summarise, it appears that gain changes in the auditory periphery apparently linked to the most basic of stimulus properties become, by the level of the cortex, part of a much larger neural project of dynamicrange compression where the effector of the shift in

Figure 1

Gain in action. Mechanisms of auditory gain control operate at multiple levels of the auditory system when an animal is alerted by a novel sound. Clockwise from bottom right: (A) The anatomy of the outer ear boosts and modifies spectral cues, including those for source location [7]; (B) membrane responses of cochlear outer hair cells generate longitudinal cellular forces that amplify vibrations of the basilar membrane [10]; (C) rapid adaptation of responses of midbrain neurons to match of neural output to statistics of the incoming sound [14]; (D) Cortical adaptation in response to loud sounds preserves sensitivity to quiet sounds — such as the approach of a second tiger [18]; (E) cortical facilitation of STRFs to match incoming sound frequency/level [20]; (F) medial olivo-cochlear efferents alter rate-versus-sound-level functions to increase output dynamic range for the representation of discrete signals (tiger) within background (environmental) noise [29]. www.sciencedirect.com

Current Opinion in Neurobiology 2009, 19:402–407

406 Sensory systems

response range is not only the underlying statistical distribution of the acoustic cues but, potentially, any environmental cue rendered contextually important by the conjunction of instantaneous demand, prior experience, and the receptive-field properties of a given neuron (see Figure 1).

Gain control and descending systems Recent anatomical and physiological studies have challenged the view of ‘bottom-up’ hierarchical processing in the auditory system, revealing massive reciprocal, descending pathways that stream from all regions of the auditory cortex to sub-cortical brain structures, by direct or indirect pathways [27]. Whilst the function of this corticofugal system remains to be determined, one of its effects may be gain control. Perhaps the best-characterised descending pathway is the medio-olivocochlear bundle, or MOC pathway, the efferent input to the OHCs. It has long been appreciated that MOC stimulation can alter the gain of rate-level functions in the auditory nerve. Recent studies have extended these findings, showing complex and variable effects of MOC stimulation on rate-level functions in both the ventral cochlear nucleus [28] and the IC [29]. In one of these studies [29], the authors show a significant effect of MOC activation in increasing the slope of rate-level functions of tones played in the presence of background noise. In addition, in some neurons, MOC activation resulted in a decrease in tone thresholds. These changes in threshold and gain render this subset of IC neurons more able to detect tones in noise and more able to discriminate between tones of different frequencies. Clearly, however, any cortical effects on OHCs are likely to be indirect: investigations by Suga et al. in the bat brain reveal, amongst many other effects, direct alterations in gain in the IC as a result of cortical stimulation (reviewed in [30]). Interestingly, whilst contemporary research has demurred from the difficult question of what these efferent-induced changes might mean for behaviour, half a century ago this question was actively under scrutiny. Perhaps indeed it is time to look again at this approach, driven by the work of Raul Hernandez-Peon, who showed reduced or abolished evoked potentials in dorsal cochlear nucleus when cats failed to attend to tones — a gain control phenomenon the authors ascribed to the descending pathways [31]. One fascinating question concerns the extent to which complex changes in cortical gain, of the form described above, impacts on sensory processing in lower brain centres — it remains to be tested whether responses of neurons in the IC, for example, are influenced by behaviour.

Conclusion The function of gain control in the auditory system appears to be one of dynamic-range compression, which Current Opinion in Neurobiology 2009, 19:402–407

results in improved coding of salient stimuli. Factors guiding this compression — both stimulus-specific factors such as the statistical distribution of sound levels, and behavioural factors such as the difficulty of an auditory task — are increasingly tractable to investigation. Whilst it appears that these latter factors are the preserve of ‘higher’ cortical function, the descending auditory pathway may distribute gain control to more peripheral processing sites. It will be fascinating to see how far the notion of dynamic, context-dependent stimulus encoding reaches as we discover precisely what constitutes context, so driving changes in gain, for different parts of the auditory brain.

References 1.

Singh NC, Theunissen FE: Modulation spectra of natural sounds and ethological theories of auditory processing. J Acoust Soc Am 2003, 114:3394-3411.

2.

Baccus SA: From a whisper to a roar: adaptation to the mean and variance of naturalistic sounds. Neuron 2006, 51:682-684.

3.

Heuer HW, Britten KH: Contrast dependence of response normalization in area MT of the rhesus macaque. J Neurophysiol 2002, 88:3398-3408.

4.

Pena JL, Konishi M: From postsynaptic potentials to spikes in the genesis of auditory spatial receptive fields. J Neurosci 2002, 22:5652-5658.

5.

Pressnitzer D, Sayles M, Micheyl C, Winter IM: Perceptual organization of sound begins in the auditory periphery. Curr Biol 2008, 18:1124-1128. This study demonstrates that the presumed ‘high-level’ process of auditory streaming, whereby patterns of tones evolving over time gradually are heard as two independent sound streams can be accounted for in the responses of neurons in the cochlear nucleus. Neural correlates of streaming were previously assessed only at the level of the cortex, as it was assumed that the perceptual attributes of streaming required cortical contributions. 6.

Campbell RA, King AJ, Nodal FR, Schnupp JW, Carlile S, Doubell TP: Virtual adult ears reveal the roles of acoustical factors and experience in auditory space map development. J Neurosci 2008, 28:11557-11570.

7.

Hofman PM, Van Riswick JG, Van Opstal AJ: Relearning sound localization with new ears. Nat Neurosci 1998, 1:417-421.

8.

Cooper N: Compression in the peripheral auditory system. Compression: from Cochlear to Cochlear Implants. New York: Springer; 2004.

9.

Dallos P, Wu X, Cheatham MA, Gao J, Zheng J, Anderson CT, Jia S, Wang X, Cheng WH, Sengupta S et al.: Prestin-based outer hair cell motility is necessary for mammalian cochlear amplification. Neuron 2008, 58:333-339.

10. Ashmore J: Cochlear outer hair cell motility. Physiol Rev 2008, 88:173-210. 11. Dallos P: Cochlear amplification, outer hair cells and prestin. Curr Opin Neurobiol 2008, 18:370-376. 12. Pienkowski M, Hagerman B: Auditory intensity discrimination as a function of level-rove and tone duration in normal-hearing and impaired subjects: the ‘‘mid-level hump’’ revisited. Hear Res 2009, 253:107-115. 13. Dean I, Harper NS, McAlpine D: Neural population coding of sound level adapts to stimulus statistics. Nat Neurosci 2005, 8:1684-1689. This study describes a new means of assessing neural sensitivity to sound level, suggesting a solution to the ‘dynamic-range problem’ in hearing. By presenting distributions of sound levels with different ‘highprobability regions’ these authors demonstrate adaptive coding, with rate-versus-sound-level functions of midbrain neurons shifting to accommodate the current distribution. As a population, neurons are maximally www.sciencedirect.com

Gain control mechanisms in the auditory pathway Robinson and McAlpine 407

informative, as judged from Fisher Information, for sound levels just above the mean level of the distribution.

23. Blake DT, Merzenich MM: Changes of AI receptive fields with sound density. J Neurophysiol 2002, 88:3409-3420.

14. Dean I, Robinson BL, Harper NS, McAlpine D: Rapid neural adaptation to sound level statistics. J Neurosci 2008, 28:6430-6438. 15. Kvale MN, Schreiner CE: Short-term adaptation of auditory receptive fields to dynamic stimuli. J Neurophysiol 2004, 91:604-612.

24. Ulanovsky N, Las L, Nelken I: Processing of low-probability sounds by cortical neurons. Nat Neurosci 2003, 6:391-398. This study demonstrates that cortical neurons are sensitive to context in which relatively simple sounds are heard, evoking stimulus-specific adaptation (SSA). Such sensitivity of cortical neurons could underlie the appearance of the mismatch negativity (MMN) response recorded from human subjects.

16. Phillips DP, Hall SE: Spike-rate intensity functions of cat cortical neurons studied with combined tone-noise stimuli. J Acoust Soc Am 1986, 80:117-187.

25. Ulanovsky N, Las L, Farkas D, Nelken I: Multiple time scales of adaptation in auditory cortex neurons. J Neurosci 2004, 24:10440-10453.

17. Rees A, Palmer AR: Rate-intensity functions and their modification by broadband noise for neurons in the guinea pig inferior colliculus. J Acoust Soc Am 1988, 83:1488-1498. 18. Watkins PV, Barbour DL: Specialized neuronal adaptation for preserving input sensitivity. Nat Neurosci 2008, 11:1259-1261. Employing a similar stimulus paradigm to that of Ref. [13] above, this study showed that the high proportion of non-monotonic rate-versus-sound-level functions observed in the auditory cortex had the function of preserving sensitivity to quiet sounds, even in the context of high sound levels. 19. Fritz J, Shamma S, Elhilali M, Klein D: Rapid task-related plasticity of spectrotemporal receptive fields in primary auditory cortex. Nat Neurosci 2003, 6:1216-1223. 20. Fritz J, Elhilali M, Shamma S: Active listening: task-dependent plasticity of spectrotemporal receptive fields in primary auditory cortex. Hear Res 2005, 206:159-176.

26. Hui GK, Wong KL, Chavez CM, Leon MI, Robin KM, Weinberger NM: Conditioned tone control of brain reward behavior produces highly specific representational gain in the primary auditory cortex. Neurobiol Learn Mem 2009, 92:27-34. 27. Winer JA, Lee CC: The distributed auditory cortex. Hear Res 2007, 229:3-13. This article highlights the extensive projections of the auditory cortex, including to lower brain centres where, presumably, it can exert influences to the level of the sensory hair cells in the cochlea. 28. Mulders WH, Seluakumaran K, Robertson D: Effects of centrifugal pathways on responses of cochlear nucleus neurons to signals in noise. Eur J Neurosci 2008, 27:702-714. 29. Seluakumaran K, Mulders WH, Robertson D: Unmasking effects of olivocochlear efferent activation on responses of inferior colliculus neurons. Hear Res 2008, 243:35-46.

21. Atiani S, Elhilali M, David SV, Fritz JB, Shamma SA: Task difficulty and performance induce diverse adaptive patterns in gain and shape of primary auditory cortical receptive fields. Neuron 2009, 61:467-480.

30. Suga N, Gao E, Zhang Y, Ma X, Olsen JF: The corticofugal system for hearing: recent progress. Proc Natl Acad Sci U S A 2000, 97:11807-11814.

22. Fritz JB, Elhilali M, Shamma SA: Differential dynamic plasticity of A1 receptive fields during multiple spectral tasks. J Neurosci 2005, 25:7623-7635.

31. Hernandez-Peon R, Scherrer H, Jouvet M: Modification of electric activity in cochlear nucleus during attention in unanesthetized cats. Science 1956, 123:331-332.

www.sciencedirect.com

Current Opinion in Neurobiology 2009, 19:402–407