An Overview of Stimulus-Specific Adaptation in the Auditory Thalamus
- First Online:
- Cite this article as:
- Antunes, F.M. & Malmierca, M.S. Brain Topogr (2014) 27: 480. doi:10.1007/s10548-013-0342-6
- 662 Views
In the auditory brain, some populations of neurons exhibit stimulus-specific adaptation (SSA), whereby they adapt to frequently occurring stimuli but retain sensitivity to stimuli that are rare. SA has been observed in auditory structures from the midbrain to the primary auditory cortex (A1) and has been proposed to be a precursor to the generation of deviance detection. SSA is strongly expressed in non-lemniscal regions of the medial geniculate body (MGB), the principal nucleus of the auditory thalamus. In this account we review the state of the art of SSA research in the MGB, highlighting the importance of this auditory centre in detecting sounds that may be relevant for survival.
KeywordsAuditoryThalamusDeviance detectionSSAMMNCorticofugal modulation
The Medial Geniculate Body
The MGB is the main auditory centre of the thalamus and, together with the cortex, establishes the circuitry necessary for the extraction and decoding of afferent acoustic information ascending from the brainstem (de la Mothe et al. 2006; Lee and Winer 2008, 2011). Growing evidence from the auditory, visual, and somatosensory systems suggests that the thalamus actively regulates the flow of sensory information and modulates sensory signals that serve as the inputs to the cortex (Alitto and Usrey 2003; Bartlett and Wang 2007; Cappe et al. 2009a; Han et al. 2012; Sherman 2007; Sherman and Guillery 2002; Winer et al. 2005). This modulation in the MGB is enabled by dynamic interactions between ascending inputs from the inferior colliculus (IC) in the midbrain (Ito and Oliver 2012; Peruzzi et al. 1997; Wenstrup et al. 1994; Winer et al. 1996), projections from the thalamic reticular nucleus (TRN; Kimura et al. 2012; Zhang et al. 2008), and corticofugal projections (Fig. 3) (Llano and Sherman 2008; Ojima 1994; Winer et al. 1999, 2001; Winer and Prieto 2001). MGB neurons receive both excitatory and inhibitory inputs from the IC (Bartlett and Smith 1999; Bartlett et al. 2000; Ito and Oliver 2012; Winer et al. 1996), excitatory inputs from the cortex (Ojima 1994; Sherman and Guillery 1996; Winer et al. 1999, 2001), and inhibitory inputs from the TRN (Kimura et al. 2012; Zhang et al. 2008). Thalamic neurons employ several mechanisms to modulate the synaptic inputs they receive from these multiple sources; for example, they are able to fire in two different response modes, termed the “tonic” and “burst” modes (Hu et al. 1994; Jahnsen and Llinas 1984; Sherman 2001a, b). It has been suggested that the tonic mode is better suited for transferring information to the cortex, while the burst mode acts as a “wake up” call to the cortex (Sherman 2001b).
The MGB has three main subdivisions: the ventral (MGV), dorsal (MGD) and medial (MGM) subdivisions, defined on the basis of cytoarchitectonic and functional differences (Fig. 4a) (Bordi and LeDoux 1994a, b; Clerici and Coleman 1990; Morest 1964; Winer and Morest 1983a, b). The three subdivisions are distinguished by different neuronal types (defined morphologically), internal architecture and input connections, and contain neurons with different response properties to auditory stimuli (Bartlett and Smith 1999, 2002; Bordi and LeDoux 1994a, b; Calford and Aitkin 1983; Hu et al. 1994; Yu et al. 2004). The MGV and the MGD are the two largest parts of the auditory thalamus (Fig. 4a). In rats, almost all neurons in these subdivisions are thalamocortical (Clerici and Coleman 1990; Winer et al. 1999). The MGM, the third subdivision, possesses several unique features; for example, many of its outputs project to non-auditory centres (Fig. 4a). The MGM and the MGD have been implicated in multisensory interactions, learning induced plasticity, and processing of communication signals and emotional content of auditory stimuli (Bordi and LeDoux 1994a, b; Cappe et al. 2009a; Edeline and Weinberger 1991, 1992; Kimura et al. 2007; Komura et al. 2001, 2005; Ledoux et al. 1987). The MGM and the MGD constitute the non-lemniscal part of the MGB, whereas the MGV is the main recipient of lemniscal input and provides precise information about stimulus frequency, intensity and timing to A1 (Fig. 3) (reviewed in De Ribaupierre 1997).
The MGV contains tufted, thalamocortical ‘relay’ neurons that typically receive input from the ipsilateral tonotopically-organized central nucleus of the IC and respond transiently, sensitively, and discretely to pure tone stimulation of the contralateral ear (Winer et al. 2005). MGV neurons project exclusively to A1, primarily to deep layer III and layer IV (Fig. 3) (Smith et al. 2012). MGV tufted neurons possess highly oriented dendritic arbors arranged in parallel with the afferent fibres from the brachium of the IC, resulting in a laminar organization (Bartlett and Smith 1999; Cetas et al. 2003; Winer et al. 1999), and providing the basis for the tonotopic map of the MGV (Winer et al. 1999).
The MGD contains two types of neurons: tufted neurons with a similar structure, although not identical, to those of the MGV, and stellate neurons (found only in the MGD) with radially extending dendrites that diverge to form a star-like configuration, (Bartlett and Smith 1999; Clerici and Coleman 1990; Winer et al. 1999). Tufted neurons in the MGD do not display oriented dendritic trees and this subdivision is not tonotopically organized (Clerici et al. 1990). The MGD projects mainly to two discrete non-primary regions of the AC in rat: the dorsal fringe of the caudal primary auditory area, designated as the posterodorsal area, and the ventral margin of the rostral primary auditory area, designated as the ventral area (Fig. 3) (Donishi et al. 2006; Kimura et al. 2007). Both areas relay information from the MGD to higher cortical areas, namely the posterior parietal cortex and the insular cortex.
The MGM extends the full rostrocaudal length of the MGB as a flat lentiform nucleus and is the smallest subdivision of the MGB (Fig 4a) (Winer and Wenstrup 1994). Compared with the MGV and MGD, neurons here have the least uniformity of soma size (Clerici and Coleman 1990; Morest 1964; Winer et al. 1999). MGM neurons often have long, sparsely branching dendrites, encompassing a large area, and their axons can give rise to collaterals that branch locally within the same or nearby nuclei (Smith et al. 2006). The inputs to the MGM are multimodal, arriving not only from the lateral, rostral and dorsal cortices of the IC but also from somatosensory and visual centres, as well as other auditory areas. The neurons may respond to one or more of these multiple modalities (Bordi and LeDoux 1994b; Wepsic 1966) and their responses adapt rapidly to repeated auditory stimuli (Bordi and LeDoux 1994a). Neurons in the MGM provide thalamic input to all auditory cortical areas, terminating in layers I, III, IV and VI (Fig. 3) (Huang and Winer 2000), but also to other subcortical (e.g., LeDoux et al. 1985a) and cortical areas (Campi et al. 2010; Cappe et al. 2009a) participating in multisensory integration (Cappe et al. 2009a, b; Komura et al. 2005). It is well known that MGM is the main direct auditory input to the amygdala (Bordi and LeDoux 1994b; LeDoux et al. 1985a, b). Early reports indicated that the MGM is unique in its ability to show physiological plasticity or long-term potentiation during classical conditioning paradigms that pair somatosensory and auditory stimuli (Gerren and Weinberger 1983; Ryugo and Weinberger 1978). Since then, the MGM has been implicated in the learning-induced expression of the conditioned fear response to auditory stimuli via the amygdala (Antunes and Moita 2010; LeDoux et al. 1984; Weinberger 2011), in the establishment of auditory emotional memories (LeDoux 1995; Ledoux and Muller 1997; LeDoux et al. 1985b), and in the expression of long term plasticity in the AC (Weinberger and Bakin 1998; Weinberger et al. 1995).
The Auditory Cortex and the Corticofugal Pathway
The AC is part of the temporal cortex and represents the site of termination of fibres ascending from the MGB (Fig. 3). The AC shows large variations between species. In the rat, several cortical maps have been proposed (e.g., Doron et al. 2002; Malmierca and Hackett 2010; Polley et al. 2007). On the basis of the patterns of thalamocortical, corticothalamic, and callosal input, as well as physiological response properties, the AC of rat can be divided into a primary auditory cortex (A1) and a surrounding belt region that includes the secondary auditory areas (Doron et al. 2002; Polley et al. 2007). Based on Nissl-stained and Golgi-impregnated material, Games and Winer (1988) analysed the neuronal architecture of A1 in rat. They distinguished six layers, extending over about 1.1–1.2 mm. Layers III and IV are the main recipients of thalamocortical inputs (Huang and Winer 2000). Layers V and VI form part of the projection to the thalamus, subthalamic nuclei and the contralateral cortex (Fig. 3) (Feliciano and Potashner 1995; Games and Winer 1988; Hefti and Smith 2000; Moriizumi and Hattori 1991; Ojima 1994; Saldana et al. 1996; Weedman and Ryugo 1996). Recent studies have challenged the classical view of ‘bottom-up’ hierarchical processing in the auditory system, revealing massive descending pathways that stream from all regions of the AC to subcortical brain structures (Winer et al. 2001; Winer and Lee 2007). These descending projections have been implicated in plasticity (Bajo and King 2012; Bajo et al. 2010), gain control and signal filtering (Liu et al. 2010; Luo et al. 2008; Robinson and McAlpine 2009; Suga and Ma 2003; Sun et al. 2007; Villa et al. 1991) and may contribute to the selective processing of sounds that acquire behavioural significance as a result of learning (Bajo et al. 2010). The corticofugal projections modulate even the lowermost brain centres, such as the cochlear nucleus (Liu et al. 2010; Luo et al. 2008), and presumably, can exert influences even at the level of the auditory nerve and cochlea (Leon et al. 2012; Xiao and Suga 2002), probably implementing a selective processing at the initial levels of sound processing as a way to reduce irrelevant information reaching the brain (Luo et al. 2008).
The massive corticofugal projections received by the MGB represent one of its most remarkable features (reviewed in Ojima and Rouiller 2011). In rat, these projections outnumber the ascending projections by a factor of 10, and are exclusively excitatory (Kimura et al. 2007; Ojima and Rouiller 2011; Winer 2006; Winer et al. 2001; Winer and Lee 2007). All three MGB subdivisions receive corticofugal projections, many of which terminate on MGB principal neurons (Fig. 3). GABAergic projection neurons in the TRN also receive cortical input and in turn project to the MGB. Lemniscal AC areas target lemniscal MGB divisions preferentially, whereas non-lemniscal areas project largely to non-lemniscal thalamic nuclei, although there are corticothalamocortical projections that allow interactions between the non-lemniscal and the lemniscal regions (e.g., the projections from A1 to the MGD; Fig. 3) (Bajo et al. 1995; Llano and Sherman 2008; Winer et al. 1999; Winer et al. 2001). Poly-modal MGB regions, such as the MGM, receive input from all AC areas and also from non-auditory cortex (Winer et al. 2001, 2005). The corticothalamic projection originates mainly in layer VI pyramidal neurons, whose terminals are small and modulatory in nature (Fig. 3). A numerically smaller corticothalamic projection arises from layer V neurons and produces large terminals that may influence the receptive field properties of the post-synaptic neuron (Fig. 3) (Bajo et al. 1993; Llano and Sherman 2008; Ojima 1994; Rouiller and Welker 2000).
It is well known that the corticofugal pathway strongly modulates the responses of MGB neurons (Figs. 6, 8). Previous studies using cooling techniques (Orman and Humphrey 1981; Palmer et al. 2007; Ryugo and Weinberger 1976; Villa et al. 1991, 1999) and electrical stimulation (He 1997, 2003a, b; He et al. 2002; Watanabe et al. 1966) have demonstrated that the AC can modulate the MGB either by facilitation or by suppression (Figs. 6, 8). Similar corticofugal modulation occurs in other sensory modalities, as demonstrated in the visual (Rushmore et al. 2005; Sillito et al. 1994) and somatosensory systems (Ghosh et al. 1994). Hence, the cortical ‘‘feedback’ to the thalamus has been suggested as a gain-control mechanism modulating the transmission of sensory information en route to higher levels (He 2003a; Suga and Ma 2003; Villa et al. 1991).
Deviance Detection in the Auditory System
Deviance detection in the human brain has been linked to the evoked potential known as the mismatch negativity potential (MMN) (Näätänen et al. 1978), a response recorded by electroencephalogram. The MMN is elicited by sounds violating some feature in the regularity of a stimulus sequence, e.g., in frequency (Sams et al. 1985), intensity (Näätänen et al. 1987), duration (Jacobsen and Schröger 2003;Paavilainen et al. 1993), more complex forms of irregularities (Carral et al. 2005a; van Zuijen et al. 2006), and discrimination of complex stimuli (Jacobsen et al. 2004; Näätänen et al. 1993) such as musical stimuli (Virtala et al. 2011). MMN is traditionally measured in an oddball design, where rare sounds (deviants) are randomly embedded within sequences of common sounds (standards), and is defined as the difference between the event-related potentials evoked by deviants versus standards (Näätänen et al. 1978). The MMN occurs in humans even in the absence of attention or behavioural task (Escera et al. 1998; Näätänen et al. 2010), e.g., in comatose patients (Fischer et al. 2000; Kane et al. 1996), sleeping newborns (Carral et al. 2005b; Sambeth et al. 2009), and fetuses (Huotilainen et al. 2005).
The neuronal mechanisms underlying MMN are not well understood, and different hypotheses are under discussion in the literature (for reviews see Näätänen et al. 2005; May and Tiitinen 2010). The prevailing theoretical frameworks rely on a “sensory memory” explanation for the MMN, in which the MMN does not just reflect adaptation, or refractoriness, of the neuronal population to the repetition of the standard stimulus but reflects true deviance detection (Jacobsen and Schröger 2003; for reviews see Grimm and Escera 2012; Näätänen et al. 2005). According to this view, the enhanced responses to the deviants are interpreted as an indication that the brain effectively stores a ‘memory trace’ of the standard stimuli, to which the incoming stimuli are compared, leading to stronger responses elicited by the mismatched deviants.
However, it remains unresolved whether the MMN indexes true deviance detection or is due to adaptation or refractoriness of the neuronal population coding for the standard stimulus. Some authors argue that the “sensory memory” interpretation of MMN leaves many fundamental questions unanswered (e.g., what is the physiological manifestation of the sensory memory trace underlying MMN? How does the comparison process work?). An alternative explanation for the MMN is the “neural adaptation” model (May and Tiitinen 2001; for a review see May and Tiitinen 2010). According to this model, the attenuation of the responses to the standard is due to adaptation or refractoriness of the neuronal population coding for the standard, and the MMN is, in essence, generated by “fresh-afferent” activity of cortical neurons that are under non-uniform levels of adaptation (May and Tiitinen 2010). The controversy surrounding this issue goes beyond the scope of the present review. From now on, when we mention deviance detection in this review, we are referring to studies that are based on the “sensory memory” view of the MMN.
Recent studies demonstrated that the MMN, which usually peaks at around 150–200 ms post-stimulus, is not the earliest electrophysiological manifestation of deviance detection in the auditory brain (Althen et al. 2011; Grimm et al. 2011; Leung et al. 2013; Recasens et al. 2012; Slabu et al. 2010, 2012; Sonnadara et al. 2006). Auditory deviance detection can occur as early as 20 ms after the onset of change, suggesting that early change detection processes occur prior to MMN generation and include the participation of subcortical regions (Althen et al. 2011; Grimm et al. 2011, 2012; Leung et al. 2013; Slabu et al. 2010). This supports the idea of a hierarchically organized system serving auditory deviance detection at different levels of complexity (reviewed in Grimm and Escera 2012).
MMN-like activity has been demonstrated to occur in several species other than human (e.g., monkeys: Fishman and Steinschneider 2012; Javitt et al. 1992, 1994; cats: Csepe et al. 1987a, b; Pincze et al. 2001; rabbits: Ruusuvirta et al. 2010; guinea pigs: Kraus et al. 1994a, b; rats: Ahmed et al. 2011; Astikainen et al. 2006, 2011; Nakamura et al. 2011; Roger et al. 2009; Ruusuvirta et al. 2013; mice: Umbricht et al. 2005). Although most of these studies were dedicated to the AC (e.g., Csepe et al. 1987a, b; Fishman and Steinschneider 2012; Javitt et al. 1994; Pincze et al. 2001; Umbricht et al. 2005), some of them studied MMN-like activity in the MGB (Kraus et al. 1994a, b) and the hippocampus (Ruusuvirta et al. 2010; 2013). In the MGB studies, the non-lemniscal but not the lemniscal MGB represented acoustic change and contributed to MMN-like activity in guinea pig (Kraus et al. 1994a, b). However, whether the responses recorded in animal models are homologous to the MMN recorded from the scalp in humans remains unclear. Fishman and Steinschneider (2012) investigated this issue in the AC of the awake monkey and concluded that homologues of MMN, insofar as it is conceived as a distinct response component reflecting deviance detection, are not present in A1 when examined using a frequency oddball paradigm. The authors note that these results do not preclude, however, the existence of genuine sensory memory processes in the brain, as thought to be reflected by the MMN in humans (according to the “sensory memory” model).
Stimulus Specific Adaptation
SSA is rapid and highly sensitive to stimulus statistics: it depends on the contrast between the two stimuli, on the relative probabilities of the standard and deviant stimuli, and on the presentation rate (Fig. 5) (Antunes et al. 2010a, b; Lumani and Zhang 2010; Malmierca et al. 2009; Reches and Gutfreund 2008; Ulanovsky et al. 2003; Zhao et al. 2011). SSA is usually quantified using the indices developed by Ulanovsky et al. (2003): (1) the frequency-specific SSA index (SI), SI(fi) (i = 1 or 2), calculated as SI(fi) = [d(fi) − s(fi)]/[d(fi) + s(fi)], where d(fi) and s(fi) are responses (in spike counts/stimulus) to frequency fi when it is deviant or standard, respectively; (2) the amount of SSA for both frequencies (common SSA index, CSI), calculated as CSI = [d(f1) + d(f2) − s(f1) − s(f2)]/[d(f1) + d(f2) + s(f1) + s(f2)]. These indices reflect the extent to which the response to a tone, when standard, was smaller than the response to the same tone, when deviant. The indices range between −1 and +1, being positive if the response to a tone, when deviant, was greater than the response to the same tone, when standard (Figs. 2, 5, 6). Since these indices are quantified as a ratio, a small absolute difference in response to a standard and a deviant tone might lead to a large ratio if the absolute size of the responses is small. Conversely, when the absolute size of the responses is large, a small difference between them yields a small ratio. This is a fact that we should bear in mind when interpreting the data. However, we believe that the output of the CSI index does not reflect a measurement artefact but has a biological meaning by measuring the contrast between the standard and the deviant stimuli: the contrast is sharpened when the absolute size of the responses is low, enhancing SSA.
SSA differs from other forms of adaptation, such as ‘fatigue’ (e.g., Carandini 2000; Sanchez-Vives et al. 2000a, b) or ‘exhaustion’ (Avissar et al. 2007) that rely on intrinsic neuronal mechanisms. The responses of neurons showing SSA are scaled according to the probability of a stimulus in the past, in a way that cannot be determined by intrinsic constraints or any mechanisms that affect the responses to all inputs to the same extent. Although the neuronal intrinsic properties and/or synaptic mechanisms that lead to SSA are unknown, most probably SSA depends on mechanisms operating at the inputs of the neuron rather than on intrinsic activity-dependent mechanisms operating at the output of the neuron (Duque et al. 2012; Reches et al. 2010; Ulanovsky et al. 2004; Zhao et al. 2011). A recent study demonstrated that SSA is not constant within the neuronal receptive field of IC neurons, but is biased toward the high-frequency regions (Duque et al. 2012). This supports the hypothesis that SSA, at least in the IC, is not created by intrinsic membrane properties of the neurons, and instead may be related to segregation of excitatory and/or inhibitory inputs (Duque et al. 2012). One possible model to explain SSA is that the two stimuli (standard vs. deviant) activate different paths to the recorded neuron and that adaptation mechanisms act at the level of these separated activation paths, as originally suggested by Eytan et al. (2003). In view of this, synaptic depression could be a potential mechanism underlying SSA (Chung et al. 2002; Rothman et al. 2009). Synaptic depression is input-specific, causing the responses of a neuron to depend on the previous history of afferent firing, thus enhancing its sensitivity to non-repeated stimuli (Abbott et al. 1997; Rothman et al. 2009). Synaptic depression scales the sensitivity of the neuron to all of its driving inputs (Abbott et al. 1997; Rothman et al. 2009), can explain a variety of time scales of adaptation (Varela et al. 1997), and is compatible with the participation of additional mechanisms, either linked to neuronal intrinsic properties (Abolafia et al. 2011) or to modulatory effects caused by neurotransmitters (Duque et al. 2013; Perez-Gonzalez et al. 2012) and/or neuromodulators. Indeed, recent findings suggest that GABAA mediated inhibition modulates SSA, in both the IC (Perez-Gonzalez et al. 2012) and the MGB (Duque et al. 2013). Their results suggest that GABA acts as a gain control system maintaining the responses of the neurons within a range such as to maximize the deviant to standard ratio, therefore enhancing SSA. This is accomplished by rendering the excitation due to the standard stimulus subthreshold or near threshold thus sharpening the contrast between standard and deviant stimuli, as in the “iceberg effect” (Isaacson and Scanziani 2011; Rose and Blakemore 1974). Altogether, these mechanisms can hypothetically transform SSA at each level of the auditory pathway.
Another possibility is that SSA goes beyond simple adaptation and reflects deviance detection. In this case, a neuron exhibiting SSA would be sensitive to the deviant but also to the regularity of the standard. Such a neuron would create a predictive model of the future, based on the regularities from the past, detecting deviant features in the environment (Winkler et al. 2009). SSA would therefore function as a kind of deviance detection mechanism, adjusting the strength of the response of a neuron based on the information that it receives (Wark et al. 2007). Indeed, recent findings demonstrated that responses evoked by a deviant sound are larger than would be predicted from refractoriness alone in A1, suggesting that SSA in A1 is related to the presence of true deviance detection (Taaseh et al. 2011). On the other hand, Fishman and Steinschneider (2012) suggest that deviance detection likely resides in cortical areas outside A1. This assumption is based on the fact that a control condition where deviants were interspersed among many tones of variable frequency (‘many standards’ control) replicated the larger responses to deviants observed under the oddball condition, in A1 of the macaque. However, as acknowledge by the authors in their discussion, it should be noted that the probability of each tone in the ‘many standards’ control condition (~5 %) was about half the probability of the deviants in the oddball condition (10 %). Therefore, while the control condition prevented the modelling of background regularities, it did not control for adaptation appropriately. The responses to deviants in the control condition may have been less adapted than in the oddball condition, and this fact can be a flaw in the interpretation of the data. The authors believe, however, that the discrepancies between the two conditions are inconsequential with respect to distinguishing SSA from genuine deviance detection. They suggest that despite these differences, the fact that the results in the oddball condition can be mimicked using rare tones that are not perceptually distinguishable as deviants indicates that deviant responses in the oddball condition do not uniquely reflect genuine deviance detection (Fishman and Steinschneider 2012). Although this question needs to be investigated in future studies, it is possible that SSA occurs prior to deviance detection. In A1, SSA appears to occur on several time scales concurrently, from hundreds of milliseconds to tens of seconds (Ulanovsky et al. 2004), paralleling the behavior of large neuronal populations recorded in human event-related potentials (Costa-Faidella et al. 2011). SSA occurs at much shorter latencies (Anderson et al. 2009; Antunes et al. 2010b; Malmierca et al. 2009; Ulanovsky et al. 2003; von der Behrens et al. 2009) than the MMN component reported in human studies (Nelken and Ulanovsky 2007). This suggests that SSA occurs prior to MMN generation, at multiple levels in the auditory pathway, similar to what occurs with deviance detection in humans (Althen et al. 2011; Grimm et al. 2011, 2012; Leung et al. 2013; Recasens et al. 2012; Slabu et al. 2010, 2012; Sonnadara et al. 2006).
Stimulus Specific Adaptation in the Medial Geniculate Body
The pioneering study of SSA found very low levels of SSA in the MGB of the cat (presumably the MGV) for the same parameter range that elicited strong SSA in A1 (Ulanovsky et al. 2003). Low levels of SSA in the MGV were later confirmed in the mouse (Anderson et al. 2009), rat (Antunes et al. 2010a), and Mongolian gerbil (Bäuerle et al. 2011). Yu and co-workers (Yu et al. 2009) studied SSA in the rat MGB and TRN, a subdivision of the thalamus that lies outside the MGB. They demonstrated strong SSA in the TRN and weaker SSA in MGB. Anderson et al. (2009) studied SSA in the different subdivisions of the MGB in mice, and showed substantially weaker levels of adaptation than those found in AC (Ulanovsky et al. 2003; von der Behrens et al. 2009) or IC (Malmierca et al. 2009) neurons. This study showed some SSA in the MGM and in the lemniscal MGV but not in the non-lemniscal MGD subdivision. However, SSA is expressed strongly in the MGB of the rat, with MGM and MGD neurons exhibiting the strongest SSA (Figs. 2, 4, 5, 6, 7) (Antunes and Malmierca 2011; Antunes et al. 2010b). These findings link thalamic SSA to the non-lemniscal pathway (Antunes and Malmierca 2011; Antunes et al. 2010b), as is also the case in the midbrain (Duque et al. 2012; Lumani and Zhang 2010; Malmierca et al. 2009; Patel et al. 2012; Perez-Gonzalez et al. 2005; Reches and Gutfreund 2008; Zhao et al. 2011), and in the zebra finch auditory forebrain (Beckers and Gahr 2012). This is in line with earlier studies that showed stronger adaptation in non-primary auditory cortices (Schreiner and Cynader 1984) and in association cortices (Irvine and Huebner 1979) than in A1. It is also in accordance with early reports that studied MMN-like activity in MGB of the guinea pig and demonstrated that the non-lemniscal MGB represents acoustic change detection to deviant tones (Kraus et al. 1994b) and discriminates speech contrasts (Kraus et al. 1994a). Although a link between MMN-like activity in animals and SSA remains elusive, it is worth mentioning that the non-lemniscal but not the lemniscal MGB responded more strongly to a deviant sound than to a standard in the evoked potential activity (Kraus et al. 1994b), paralleling the behavior of SSA neurons in the MGB and IC. The reasons for the lack of SSA in the MGD of mice (Anderson et al. 2009) are unclear, but may include species differences, methodological differences (e.g., anesthetic used, definition of subdivisions), and/or sampling constraints (MGD recordings in the mouse study were based on a small sample (n = 13) and included both multi-unit and single unit recordings).
As for the other levels of the auditory pathway (Lumani and Zhang 2010; Malmierca et al. 2009; Reches and Gutfreund 2008; Taaseh et al. 2011; Ulanovsky et al. 2003; Zhao et al. 2011), SSA in the MGB varies with the stimulation parameters, including the frequency difference between the two stimuli, their relative probabilities, and the repetition rate (Fig. 5) (Anderson et al. 2009; Antunes et al. 2010a, b). In general, SSA in the rat MGB (Antunes et al. 2010b) is evoked using stimulation parameters similar to those used to evoke SSA in cat AC (Ulanovsky et al. 2003) and rat IC (Ayala and Malmierca 2013; Malmierca et al. 2009; Zhao et al. 2011). The stimulation parameters that evoke the strongest SSA in MGB neurons are frequency ratios of 0.141 to 0.526 octaves, with an interstimulus interval (ISI) of 250 to 500 ms (onset to onset, for a stimulus duration of 75 ms; Fig. 5) (Antunes et al. 2010b). Under these conditions, a high percentage of MGM and MGD neurons show strong SSA (Fig. 5). Moreover, the rat MGB exhibits strong SSA even under extreme testing conditions, e.g., a frequency interval of 0.14 octaves combined with an ISI of 2,000 ms (Fig. 5) (Antunes et al. 2010b). An interesting finding is that SSA in some non-lemniscal MGB neurons can be observed for frequency intervals as small as 0.057 octaves (Fig. 5) (Antunes et al. 2010b). This suggests that some neurons in the MGB can “discriminate” between two very close frequencies, both of which lie well within their frequency response areas, with a substantially smaller difference between them than the typical peripheral tuning width (Antunes et al. 2010b). This indicates that MGB neurons show hyperacuity (Condon and Weinberger 1991; Moore 1993) in the frequency dimension, as previously demonstrated to occur in cat AC (Ulanovsky et al. 2003) and rat IC (Malmierca et al. 2009) neurons. It should be mentioned that the observation that a neuron can segregate two spectrally very close frequencies as reflected in its output could simply be the result of the activation of different inputs to the neuron that are narrowly tuned to different frequencies and which adapt largely independently. This possibility is related to the mechanisms underlying SSA, and needs to be studied further.
A few neurons in MGB show SSA levels as high as those reported in the TRN (Yu et al. 2009), even with ISIs twice as long. Some neurons in the MGB (Antunes et al. 2010b) exhibit higher values of SSA than A1 neurons (Ulanovsky et al. 2003), for the largest ISI (=2,000 ms), but only outside the MGV. The MGV receives input from the central nucleus of the IC and is the main source of ascending input to A1 (He 2003a; Lee and Winer 2008). In this context, it is worth mentioning that SSA found in the MGV (Antunes et al. 2010b) and the central nucleus of the IC (Malmierca et al. 2009) is relatively large only for short ISIs (125–250 ms, onset-to-onset, 75 ms stimulus duration). As previously mentioned, the MGB and IC data tightly link SSA in subcortical regions to the non-lemniscal pathway (Figs. 4, 5) (Antunes and Malmierca 2011; Antunes et al. 2010b; Duque et al. 2012; Lumani and Zhang 2010; Malmierca et al. 2009; Perez-Gonzalez et al. 2005). In the lemniscal pathway, SSA levels reported in A1 of the cat (Ulanovsky et al. 2003) and rat (Taaseh et al. 2011; Yaron et al. 2012) are far in excess of the values found in the MGV of the rat (Antunes and Malmierca 2011; Antunes et al. 2010b; Taaseh et al. 2011; Yaron et al. 2012), guinea pig (Bäuerle et al. 2011), and mouse (Anderson et al. 2009). In all these studies, the SSA values found in the MGV are similar and very close to zero. In this context, A1 emerges as the first lemniscal station in which SSA is widespread and strong.
Is SSA in the MGB Inherited via Corticofugal Projections?
To examine whether the AC is necessary for the expression of SSA in MGB neurons, Antunes and Malmierca (2011) studied SSA throughout the MGB of the anesthetized rat, before, during and after reversibly deactivating the AC by cooling (using the techniques described by Carrasco and Lomber 2009; Coomber et al. 2011; Lomber et al. 1999). During deactivation of AC, the responses of MGB neurons were significantly modified but the levels of SSA and the dynamics over time were mostly unaffected (Figs. 6, 7, 8) (Antunes and Malmierca 2011). These findings demonstrate that SSA in the MGB is not inherited from the AC, but instead may be inherited from lower levels such as the IC and/or be generated de novo in the MGB. Similar findings were obtained by Anderson and Malmierca (Anderson and Malmierca 2013) for the IC. It may well be that SSA generated at lower levels is transmitted in a bottom-up manner and can be modulated intrinsically at each level of the auditory pathway (Antunes and Malmierca 2011). Such bottom-up transmission of SSA would enable each level of the auditory system to shape the responses of the previous level in order to eliminate as much statistical redundancy as possible (Schwartz and Simoncelli 2001). SSA in A1 may therefore express the combined result of the rather weak SSA found in MGV augmented by intracortical mechanisms (Szymanski et al. 2009) and possibly by the weaker (but still present) non-lemniscal input to A1, either directly from the MGM (Kimura et al. 2003) or indirectly through feedback connections from higher auditory areas.
The insensitivity of SSA in the MGB to AC deactivation is all the more remarkable given the significant alteration of many other properties of MGB neurons during AC deactivation, such as their frequency response areas, spontaneous activity, discharge rate and/or latencies (Figs. 6, 8) (Antunes and Malmierca 2011). These findings confirm previous studies of corticofugal projections that utilized cooling techniques (Nakamoto et al. 2010; Ryugo and Weinberger 1976; Villa et al. 1991; 1999) or electrical stimulation (He 2003a; Ojima and Rouiller 2011), and corroborate a strong corticofugal modulation in the MGB (Bajo et al. 2010; Nakamoto et al. 2010; Villa et al. 1991; 1999; Yu et al. 2009). We conclude that the corticofugal modulation does not account significantly for the SSA exhibited in MGB neurons, but rather, modulates the discharge rate of these neurons affecting the responses to both the standard and the deviant stimulus similarly, suggesting a gain control mechanism (Figs. 6, 7, 8) (Antunes and Malmierca 2011). As a result, the degree of SSA quantified as a ratio of driven rates, is largely unaffected by cooling (Fig. 7). This conclusion is in line with previous studies that suggest the corticofugal system participates in a gain control process that leads to improved coding of salient stimuli (Robinson and McAlpine 2009), and possibly underlies auditory attention (He 2003a) and learning-induced plasticity (Bajo et al. 2010). Thus, the results are consistent with a role of the corticofugal pathway in scaling the sensitivity of the MGB neurons to its driving inputs by controlling their gain (Antunes and Malmierca 2011).
An interesting finding is that although SSA in the MGB is only weakly affected by cortical deactivation, there is a significant relationship between SSA and changes elicited by cooling (Fig. 8d, e). The corticofugal modulation of the discharge rate of MGB neurons varies significantly with the SSA level that they exhibit, such that the facilitation exerted by the AC on MGB neurons is reduced as SSA increases (Fig. 8d, e). This relation is not dependent on the anatomical subdivision to which the MGB neurons belong but rather on the level of SSA that they exhibit (Antunes and Malmierca 2011). The general view of cortico-thalamic interactions is one of very large variability with many effects. Therefore, the fact that the amount of cortical modulation is correlated to the level of SSA constitutes strong evidence for a rule that relates cortical modulation to neuronal properties: low SSA is related to strong cortical gain (Fig. 8d, e) (Antunes and Malmierca 2011). These findings are striking and should be addressed in future studies, including using different ways of quantifying SSA in order to avoid any inherent limitations of the indices used here.
It was recently demonstrated that GABAA mediated inhibition plays a role in shaping SSA in the MGB, by sharpening the contrast between the standard and deviant stimuli, thus enhancing SSA (Duque et al. 2013; Perez-Gonzalez et al. 2012). Some high SSA neurons from the non-lemniscal MGB receive suppressive influence from the corticofugal pathway (Antunes and Malmierca 2011), and inhibition is a possible mechanism underlying such suppression. A major source of inhibition to the MGB is via the AC-TRN GABAergic input to the MGB (He 2003b; Yu et al. 2009). However, inhibition driven by this corticofugal pathway does not drive or modulate SSA in the MGB, since SSA levels were preserved during cortical deactivation (Antunes and Malmierca 2011). Thus, the recently demonstrated inhibitory effect that plays a role in shaping SSA in the MGB (Duque et al. 2013), must involve other sources of GABAergic input to the MGB, such as those ascending from the IC (Peruzzi et al. 1997), and the MGB–TRN–MGB connections themselves (Yu et al. 2009).
Final Remarks and Future Research Directions
The presence of strong SSA in the non-lemniscal auditory thalamus suggests that SSA is important for the type of processing performed there (Figs. 4, 5) (Antunes et al. 2010b). For example, the very strong SSA found in the MGM is consistent with its role as a major auditory input to the fear circuit in the amygdala (Antunes et al. 2010b; Bordi and LeDoux 1994b; LeDoux et al. 1984; Weinberger 2011). The MGM-to-amygdala pathway participates actively in the mechanisms of fear learning, and has been suggested to be important for suppressing fear of neutral or safe auditory stimuli (Antunes and Moita 2010), paralleling the role of SSA in suppressing responses to repetitive sounds presumed to be irrelevant and safe. Moreover, the AC and the corticofugal pathway do not drive or transform SSA in the MGB but rather modulate the responses of MGB neurons, probably by acting as a gate or gain control mechanism (Figs. 6, 7, 8) (He 2003a; Robinson and McAlpine 2009; Villa et al. 1991). The amount of gain exerted by the AC in the MGB varies from neuron to neuron and depends on the sensitivity of these neurons to rare sounds in the environment (Fig. 8b, c). Some other hypothetical functional roles for SSA include optimization of neural coding (Dean et al. 2008; Schwartz and Simoncelli 2001; Wark et al. 2007), maximization of information transmission (Brenner et al. 2000; Fairhall et al. 2001), enhancements in the discriminability of incoming stimuli (Malone et al. 2002; Muller et al. 1999), and economization of spikes (deCharms and Merzenich 1996). These are functional roles that have been attributed to adaptation in general, and especially to those forms of adaptation that, like SSA, depend on the history of stimulation.
Beyond pure speculation, present research on SSA supports the hypothesis that SSA, at least in its simplest form (i.e., to frequency deviants), is transmitted in a bottom-up manner through the auditory pathway, acting as a gating mechanism involved in confining the auditory stream to behaviourally relevant stimuli. One is tempted to speculate that SSA will have a different shape at the different levels of the auditory pathway, being enhanced for some characteristics along the pathway, and/or participating in more complex tasks at higher levels. SSA has usually been studied using simple acoustic features and paradigms, such as pure tones embedded in random oddball sequences, stimuli that are far removed from the complexity of natural sounds and the real acoustic environment. Maybe for this reason, the expression of SSA appears to be similar at the different levels of the auditory pathway. What are the limitations and abilities of each level of the auditory pathway? Is SSA a process necessary for deviance detection? If so, where are the true neuronal deviance detectors where SSA information from other levels converges? To further understand the mechanisms and functional roles of SSA, future research calls for the use of more complex experimental conditions such as those reported by Yaron et al. (2012) who studied the sensitivity of neuronal responses to statistical regularities in the rat AC. They contrasted the neuronal responses to oddball sequences presented either randomly or periodically, and demonstrated that neurons in the AC are sensitive to the detailed structure of sound sequences. Yet another study examined sensitivity to pattern violation in neurons of the IC (Aguillon et al. 2013). Their preliminary results indicate that some IC neurons show a differential response to expected versus unexpected tones, regardless of their probabilities. These data suggest that adaptation in the IC goes beyond repetition suppression (as measured by the simplest oddball paradigm) and may also encode expectation suppression. Future studies should aim to clarify whether the putative sensitivity of neurons to violations in complex patterns indeed reflects deviance detection or can instead be explained by response interactions due to more basic physiological mechanisms (e.g., forward suppression).
We thank Dr. Nell Cant for her constructive comments and corrections on a previous version. Financial support was provided by the Spanish MEC (BFU2009-07286) and EU (EUI2009-04083, in the framework of the ERA-NET NEURON Network of European Funding for Neuroscience Research) to M.S.M.