Functional network properties of the auditory cortex

Abstract

The auditory system transforms auditory stimuli from the external environment into perceptual auditory objects. Recent studies have focused on the contribution of the auditory cortex to this transformation. Other sets of studies have yielded important insights into the contributions of neural activity in the auditory cortex to cognition and decision-making. However, despite this important work, the relationship between auditory-cortex activity and behavior/perception has not been fully elucidated. Two of the more important gaps in our understanding are (1) the specific and differential contributions of different fields of the auditory cortex to auditory perception and behavior and (2) the way networks of auditory neurons impact and facilitate auditory information processing. Here, we focus on recent work from non-human-primate models of hearing and review work related to these gaps and put forth challenges to further our understanding of how single-unit activity and network activity in different cortical fields contribution to behavior and perception.

Introduction

A fundamental goal of the auditory system is to parse an unlabeled mixture of environmental auditory stimuli into coherent perceptual units (Bregman, 1990). The auditory system accomplishes this task, in part, by computations that group together stimuli with similar spectral and temporal features, while simultaneously segregating stimuli with different spectral and temporal features into different groups. Groupings with similar spectral and temporal features lead to the generation of coherent auditory perceptual units, which are also known as auditory objects (Bizley and Cohen, 2013; Griffiths and Warren, 2004).

Our understanding of these stimulus-to-perception computations has been facilitated by elegant anatomical, psychophysical, electrophysiological, and imaging studies in humans and in non-human animal models. However, there are still substantial gaps in our understanding of the relationship between neural activity (in particular, spiking activity) in the auditory cortex, perception, and behavior (Banno et al., 2020; Cohen, 2012). Here, we discuss and review two gaps in our understanding: (1) the specific and differential contributions of different fields of the auditory cortex to auditory perception and behavior and (2) the way network activity impact and facilitate auditory information processing. In particular, we emphasize recent work that has identified underlying computational principles in the cortex of non-human primates and when appropriate we note studies from other animal models of hearing.

In the 1990s, Hackett, Romanski, Kaas, Rauschecker, and others (Hackett et al., 1998, 1999; Kaas and Hackett, 1998, 1999; Romanski and Goldman-Rakic, 2002; Romanski et al., 1999; Romanski et al., 2000) revolutionized our understanding of cortical anatomical connectivity as it relates to auditory processing. First, Hackett and Kaas demonstrated a new organizational principle for the auditory cortex; namely, a “core” that is surrounded by a “belt”, which, in turn, is surrounded by a “parabelt”. From those initial anatomical studies, two cortical processing streams were identified: the so-called “dorsal” and “ventral” pathways. Because these pathways were hypothesized to be analogous to the visual dorsal and ventral pathways (Ungerleider and Mishkin, 1982), it was thought that they contribute to auditory perception and spatial/audiomotor behaviors, respectively.

However, despite the elegance of these studies, we still do not have a coherent theory of how auditory and related cognitive information is processed within and across defined auditory fields. For example, what information is preferentially processed in the primary auditory cortex and how is this processing different from processing that occurs in other core auditory fields? Further, we do not have a good understanding of how information is transformed and represented across different cortical fields (e.g., from the core to the belt and ultimately to the prefrontal cortex). Finally, at a higher hierarchical level of processing, the unique functional role(s) of the auditory dorsal and ventral pathways still remains an active area of research (Rauschecker and Scott, 2009).

Although single-unit recordings have shed considerable light on the functional properties of the auditory cortex and the representation of stimulus-, task-, and cognitive-related variables in individual auditory fields (Cohen, 2012; Fritz et al., 2013; King et al., 2018; Recanzone, 2018; Recanzone and Sutter, 2008; Shamma et al., 2013; Shamma et al., 2011; Tsunada and Cohen, 2014; Wang and Walker, 2012), the cortex is not just composed of billions of individual neurons, but instead, it is composed of networks of interconnected neurons (Cohen and Kohn, 2011; Kohn et al., 2016; Semedo et al., 2020). These cortical networks operate over different computational scales: network ensembles exist within a single cortical field (e.g., the primary auditory cortex), across cortical fields (e.g., primary and non-primary auditory cortex), and across brain systems (e.g., the dorsal and ventral auditory pathways or the auditory and visual pathways) (Atencio and Schreiner, 2016; Bastos et al., 2012; Fries, 2009, 2015; See et al., 2018; See et al., 2021). These spatial scales include not only feedforward patterns of connectivity but also complex patterns of feedback connectivity (Banno et al., 2020). Indeed, the richness and extent of auditory feedback connectivity distinguishes it from other sensory systems, like the visual system (Brugge, 1992; Felleman and Van Essen, 1991; Kaas and Hackett, 2000). Across these spatial scales, we can also consider different temporal scales of neural activity, which range from single-neuron spiking activity to local-field potentials to neural oscillations. Unfortunately, our understanding of the contribution of network activity to auditory information processing is relatively nascent.

Functional and hierarchical organization of auditory cortex

The auditory cortex is made up of the core, belt, and parabelt (Fig. 1). These cortical areas are based on architectonics and corticocortical connectivity (Hackett, 2011; Rolls et al., 2022; Romanski and Averbeck, 2009). Anatomically, each cortical field is connected most strongly to and reciprocally with its immediate neighbors in all directions (Hackett et al., 1998). For example, the core and belt auditory cortices are highly connected with one another. This is consistent with a general cortical principle that stronger connectivity exists between immediately adjacent neighboring fields (Averbeck and Seo, 2008). On the other hand, the core and parabelt auditory cortices have minimal connectivity.

Figure 1: Anatomical locations and organization of auditory cortical fields.

A schematic side view (anterior is to the left and dorsal is toward the top) of a rhesus brain. The auditory cortex is located on the supratemporal plane and in the lateral sulcus (LS). In the cut-out, which is a top-down view of the temporal lobe, we schematize the relative organization of the core auditory cortex (dark blue), belt auditory cortex (lighter blue), and parabelt auditory cortex (0range). The solid red line indicates the border of the lateral sulcus. Superior temporal sulcus (STS), rostral superior temporal gyrus (STGr), rostral parabelt (RPB), caudal parabelt (CPB), primary auditory cortex (A1), rostral (R), rostral temporal (RT), caudal medial (CM), middle medial (MM), rostral medial (RM), medial rostral temporal (RTM), caudal-lateral (CL), middle lateral (ML), rostral-lateral (RL), lateral rostral temporal (RTL), rostrotemporal polar (RTp), and temporoparietal (TPT).

Beyond cortical-cortical connectivity, these auditory fields are also distinguished by their thalamic inputs. All the fields in the core auditory cortex (A1, R, and RT) receive direct input from the ventral division of the medial geniculate nucleus; A1 receives the strongest projections, whereas RT receives the weakest projection (Jasmin et al., 2019). Because of this connectivity, each is considered to be a “primary” (A1) or “primary-like” (R and RT) auditory field. We do not know fully the computational advantage of multiple such cortical fields (Jasmin et al., 2019), unlike a single primary field that is ubiquitous in the primate and mammalian visual systems. Thalamic input to the belt and parabelt auditory cortices arises mostly from the anterodorsal and posterodorsal divisions of the medial geniculate nucleus (Hackett, 2011). The medial division of the medial geniculate projects widely to all fields in the auditory cortex.

Both the core and belt auditory cortex lie on the dorsal surface of the temporal lobe, which is buried in the lateral sulcus. The belt auditory cortex surrounds the core and has multiple divisions such as the middle-lateral, caudolateral, and rostrolateral belt regions. The parabelt auditory cortex lies lateral to the belt on the superior temporal gyrus. The belt is divided into the caudal, middle, anterior, and rostral auditory fields, as well as medial and lateral auditory fields. The parabelt auditory cortex includes caudal and rostral components. Caudal and rostral auditory fields are also anatomically connected with regions of the prefrontal cortex: the caudal auditory fields project to dorsolateral prefrontal cortex (dorsal area 46) and the rostral fields project to ventral lateral prefrontal cortex (ventral area 46 and area 12) (Romanski et al., 1999).

Although the architectonic divisions and corresponding connectivity of the auditory cortex are well defined, much less is known about its functional properties and organization (Brewer and Barton, 2016; Escabí and Read, 2003; Gerstein and Kiang, 1964; Hackett, 2011; Linden and Schreiner, 2003; Miller et al., 2002). Historically, single neurons in the auditory cortex have been probed to identify how they represent fundamental features of acoustic stimuli, such as frequency tuning and bandwidth. However, it is becoming increasingly clear that the computations occurring in the auditory cortex are not simply restricted to spectral decompositions of auditory stimuli. Indeed, the frequency tuning of an auditory-cortex site is not hard wired to a specific value but is dependent, to a degree, on the demands of an ongoing behavior task (David et al., 2012; Fritz et al., 2013; Kilgard, 2012; Yin et al., 2014). Moreover, this frequency tuning may not even represent the physical features of a stimulus but may be more closely tied with its perceptual qualities (e.g., its pitch) (Bendor and Wang, 2006; Kikuchi et al., 2019). Further and consistent with a role of the auditory cortex in perception, several studies have demonstrated a role for the core and belt auditory cortices in segregating and organizing the external auditory environment into perceptual units. This notion is supported by studies that have examined the contributions of these cortical fields to auditory streaming and prediction (Christison-Lagay and Cohen, 2018; Kikuchi et al., 2018; Selezneva et al., 2018) as well as those that examine how listeners extract an auditory figure from a noisy background (Christison-Lagay et al., 2017; Schneider et al., 2021; Schneider et al., 2018; Teki et al., 2016).

These representations of frequency, pitch, the auditory scene, etc. are also modulated by cognitive- and task-related variables (Fig. 2). For example, several studies have identified neural signatures of auditory working memory in the early auditory cortex (Bigelow et al., 2014; Huang et al., 2016; Plakke et al., 2013; Poremba et al., 2013), which can also be seen in the prefrontal cortex (Plakke et al., 2015). Other studies have shown that A1 neurons are modulated by visual signals that cue the start of an auditory task, by motor movements associated with the task (i.e., the grasping and release of a touch-sensitive bar), and the delivery of juice rewards (Brosch et al., 2005; Huang and Brosch, 2020; Huang et al., 2019; Knyazeva et al., 2020; Werner-Reiss et al., 2003; Wikman et al., 2019).

Figure 2: Extra-auditory influences on the auditory cortex.

Schematic of the auditory cortex showing the core (central ellipse) and surrounding belt auditory cortex (grey ellipse). Classic work focused primarily on the (1) tonotopic maps in each core field (A1, R, and Rt), which is schematized by changes in color (see color bar) and (2) increased tuning sensitivity for vocalizations, spatial location, etc. between the core and belt, which is schematized by differently colored broad Gaussian “tuning curves” in the core auditory cortex and narrower Gaussians in the belt auditory cortex. More recent studies have highlighted that auditory-cortex responsivity is subject to multiple influences including extrasensory inputs, attention, choice, reward, memory, motor etc. (as schematized by arrows). The challenge is to identify how, when, and why these inputs modulate auditory perception and ongoing behavior.

There is also evidence for hierarchical information processing in the auditory cortex. For example, neurons in rostral fields tend to be more sensitive to stimulus identity (e.g., different monkey vocalizations) than caudal-field neurons (Fukushima et al., 2014; Kikuchi et al., 2018; Rauschecker, 1998; Rauschecker and Tian, 2000; Rauschecker et al., 1995). These auditory-cortex representations of vocalizations are further refined in those regions of the prefrontal/frontal cortex that receive input from the auditory cortex (Diehl et al., 2022; Jovanovic et al., 2022). Computational and imaging studies also suggest hierarchical processing of object (perceptual) information in the human auditory cortex (Kell et al., 2018). Spatial information is also hierarchically organized: single neurons in the caudal belt auditory cortex, including CL and CM, tend to encode more information about the spatial location of an auditory stimulus than those in the core auditory cortex. The spatial tuning of these caudal-belt neurons is sharp enough to support a listener’s spatial acuity (Miller and Recanzone, 2009).

As just discussed, and as alluded to above, there is both anatomical organization and physiological response properties in support of a “what versus where” distinction in dorsal and ventral auditory-cortex circuits. We caution, however, that, with the exception of a single study in cats (Lomber and Malhotra, 2008), there is not any direct causal evidence, to our knowledge, favoring a dissociation between object- (“what”) and spatial- (“where”) related processing in the ventral and dorsal pathways, respectively, in primate models of hearing. Indeed, in non-human primates, neural sensitivity to spatial and non-spatial processing is seen in both pathways, with almost equal degrees of tuning sensitivity (Cohen et al., 2004; Gifford III and Cohen, 2005). Similarly, other studies have also found that neurons in the dorsal pathway are modulated by both non-spatial and spatial auditory information (Belin and Zatorre, 2000; Cusack, 2005; Engel et al., 2009; Lewis et al., 2005; Pizzamiglio et al., 2005; Rauschecker, 2011; Recanzone, 2008; Walker et al., 2011; Warren et al., 2005; Zatorre et al., 2002). Although it is not a primate study, it is worth noting that a recent ferret study also failed to identify a clean disassociation between auditory object and spatial behavior in these two pathways (Town et al., 2022).

We would like to offer that this parcellation of “what” and “where (perceptual and non-perceptual) information into parallel pathways may be simplified and does not account for the complexities of audition. Spatial information, like non-spatial information (see above), is critical to the parsing of the auditory scene into distinct perceptual objects: two stimuli that are far apart are more likely to be heard as two distinct auditory objects (sounds) versus two stimuli that are close together. This suggests that spatial information in the dorsal pathway could be used for parsing the auditory scene into perceptual objects (Cusack, 2005). Could the dorsal pathway have a privileged role in those situations in which “where” information is needed to parse the auditory scene? A non-exclusive alternative possibility is that, in these situations, there may be enhanced functional connectivity between the dorsal and ventral pathways.

In addition to processing different types of auditory information, the auditory cortex also has pervasive connections with other sensory cortices, which may contribute, in part, to multisensory behavior and perception (Fig. 2) (Caruso et al., 2021; Ghazanfar and Schroeder, 2006; Khandhadia et al., 2021; Raposo et al., 2012; Schmehl and Groh, 2021). For example, the belt auditory cortex (mainly the caudal fields) has considerable connections with the secondary visual cortex (Falchier et al., 2009) and the secondary somatosensory cortex (Cappe and Barone, 2005; Smiley et al., 2007). In contrast, the core auditory cortex receives sparse input from these extrasensory fields (Cappe and Barone, 2005; Falchier et al., 2009). Consistent with these patterns of connectivity, belt neurons have more robust multisensory responses than core neurons (Bizley and Dai, 2020; Bizley et al., 2006; Ghazanfar et al., 2005; Morrill and Hasenstaub, 2018). However, in general, the differential contribution of the core and belt auditory cortex to multisensory processing is still not fully understood (Ghazanfar et al., 2005; Lehmann et al., 2006; Merrikhi et al., 2023).

These extrasensory signals may also serve to facilitate and enhance auditory processing. For example, when somatosensory and auditory stimuli are presented simultaneously, somatosensory signals arrive at A1 faster than the auditory responses. Because these somatosensory signals reset ongoing neural oscillations in the auditory cortex, when the auditory signals reach A1, they are coupled to an optimal oscillation phase. This coupling, in turn, facilitates auditory processing (Lakatos et al., 2007). This facilitation is largest when the auditory response is the weakest. Similar coupling and enhanced processing have also been found using audiovisual stimuli (Kayser et al., 2008; Mégevand et al., 2020).

Representation of variables related to decision making in the auditory cortex

Neural correlates of decision-making have been found in early auditory fields (Fig. 2). For example, in a spatial-delayed match-to-sample task, there was a strong representation of several decision and cognitive variables in A1 (Napoli et al., 2021). In that study, neurons were recorded in A1 and dorsolateral prefrontal cortex (dlPFC; caudal, dorsal area 46 and adjacent area 8) while monkeys carried out the spatial-delayed match-to-sample task. Cue-specific activity was found in A1, which preceded similar cue-related activity in the dlPFC. There was also robust delay- period activity in both A1 and dlPFC. Interestingly, decision-related activity was also found in A1 and dlPFC but earlier in A1 than in dlPFC. The A1 decision-related activity also shifted with reaction times. This choice-related activity was earlier when reaction times were shorter but later when reaction times were longer. Furthermore, the choice-related activity on error trials represented the actual choice of the animal in A1 but not dlPFC. Thus, this study found substantial choice-related activity in A1, which appeared to precede similar signals in dlPFC. The finding of choice-related activity in A1 is consistent with other studies that identified choice-related neural modulation in A1 (Bizley et al., 2013; Ceballo et al., 2019; Christison-Lagay and Cohen, 2018; Kilian-Hütten et al., 2011; Mohn et al., 2021; Niwa et al., 2012, 2013).

These results may seem surprising, given findings that early visual fields have little activity related to decision making (Freedman and Assad, 2016; Jasper et al., 2019; Krishna et al., 2021). Indeed, the finding of choice-related activity in the early auditory fields is not universal. For example, in an auditory flutter discrimination task, Lemus et al. did not find choice-related activity in the auditory cortex (Lemus et al., 2009a). Similarly, Tsunada et al. did not identify choice-related signals in the middle-lateral belt of the auditory cortex during a categorization task nor did they identify choice activity during a frequency-discrimination task (Tsunada et al., 2011; Tsunada et al., 2016). Using a combination of modeling, electrophysiological recordings, and computational modeling, Tsunada et al. argued that activity in the anterolateral belt causally contributes to the current auditory decision (Tsunada et al., 2016). Lemus et al. and Tsunada et al. did, however, identify choice-related activity in regions downstream from the auditory cortex (Lemus et al., 2009b; Tsunada et al., 2019). Interestingly, because the PFC choice activity that was identified by Tsunada et al. related to the next trial, it most likely reflected an evaluative process of the previous trial and/or the biasing of subsequent trials.

How can we reconcile these different sets of findings? We do not have a definitive answer, but there are several non-exclusive possibilities. One possibility may relate to the different perceptual and cognitive demands of the different auditory tasks/stimuli and how each task/stimulus engages different auditory fields. Another possibility relates to the analysis of the choice activity itself, including clearly differentiating between choice and stimulus signals and differentiating between causal feedforward signals that relate to the ongoing decision versus attention-related feedback signals (Nienborg et al., 2012; Nienborg and Cumming, 2009; Tsunada et al., 2016). Finally, neural selection bias (e.g., only analyzing neurons with high stimulus sensitivity or other response properties) may limit how well we can extrapolate from the properties of a specific recorded population to a general statement on the contribution of an auditory field to decision-making.

Population codes matter in the AC

The combined activity of auditory-cortex neurons –that is, its network or ensemble-level properties (Eggermont, 2007)– is a better predictor of behavior than single-unit activity alone (Bathellier et al., 2012; Christison-Lagay et al., 2017; Engineer et al., 2008; Ince et al., 2013; Miller and Recanzone, 2009; Pachitariu et al., 2015), consistent with models of population coding (Averbeck and Romanski, 2006; Bartolo and Averbeck, 2020). But how do populations of neurons in the auditory cortex encode stimuli? One popular approach is to test the noise correlations between pairs of auditory-cortex neurons (Cohen and Kohn, 2011; Downer et al., 2021; Gourévitch and Eggermont, 2010; See et al., 2018); noise correlations reflect the functional connectivity of the underlying network (Aertsen et al., 1989; Cohen and Kohn, 2011; Greschner et al., 2011). More specifically in the auditory cortex, neurons with similar frequency-tuning profiles tend to be more strongly correlated than those with dissimilar frequency tuning and have more synchronous activity (Fukushima et al., 2012). Because networks of correlated neurons are stable over time and stimulus conditions, it is possible that such networks reflect fundamental units of information processing in the auditory cortex (See et al., 2018). Indeed, neurons with similar frequency-tuning profiles tend to be more strongly correlated than those with dissimilar frequency tuning and have more synchronous activity (Fukushima et al., 2012).

These noise correlations are specifically relevant in population coding because they affect the amount of information that can be encoded by the population, which, in turn, impacts behavioral performance (Abbott and Dayan, 1999; Averbeck et al., 2006; Downer et al., 2021; Panzeri et al., 2022). Despite the existence of large noise correlations in the auditory cortex (Rothschild et al., 2010), which are often associated with information limiting neural encoding (Bartolo and Averbeck, 2020; Kafashan et al., 2021; Panzeri et al., 2022), certain groups of neurons exhibiting strong noise correlations were also shown to improve auditory neural encoding. For instance, neural ensembles constructed from neurons that have coincident firing have higher information capacity than individual neurons and random groupings of neurons (See et al., 2018) These findings can possibly be reconciled by considering the scale of the neuronal populations studied. Information limiting correlations in large neuronal population (>100 neurons) tend to be detrimental to neural encoding (Abbott and Dayan, 1999; Bartolo and Averbeck, 2020; Kafashan et al., 2021; Zohary et al., 1994), whereas the information limiting effects are minimal in smaller populations (Averbeck et al., 2006; Averbeck and Lee, 2006). However, effects in populations do not tend to reverse, such that positive effects in small populations become negative effects in large populations. Additionally, it is always possible to find some pairs that show positive effects of noise correlations, but at the population level the effects of noise correlations have only been found to be negative.

Despite the ubiquitous existence of noise correlations in the auditory cortex (Downer et al., 2015; Rothschild et al., 2010; Winkowski and Kanold, 2013), the role and the effect of noise correlations on auditory encoding are still unclear. For example, motivated by findings in the monkey visual cortex showing that attention seems to facilitate behavioral performance by decreasing noise correlations (Cohen and Maunsell, 2009), several studies have investigated the relationship between task engagement and auditory-cortex noise correlations (Downer et al., 2015, 2017a; Issa and Wang, 2013). In one study, Downer et al. measured the noise correlations between neurons in the core auditory cortex while monkeys listened passively to amplitude-modulated tone bursts or selectively attended to these tone bursts (Downer et al., 2017a). This study found that the noise correlations were lower during active listening relative to the passive-listening condition. Although these findings point to a straightforward inverse relationship between attention and auditory-cortex noise correlations, the story is more complicated: in the belt, task engagement increased the noise correlations (Downer et al., 2017a). The relationship between engagement and noise correlations is further nuanced by the finding that changes in neural-correlation structure depend not only on the neural sensitivity to the attended stimulus feature but also on the properties of non-attended stimulus features (Downer et al., 2017b). To further complicate this issue, noise correlations in the auditory cortex increase with age (Shilling-Scrivo et al., 2021; Shilling-Scrivo et al., 2022) and after parturition (Rothschild et al., 2013), at least in rodent species. Thus, the relationship between noise correlations, behavior, and neural representations in the auditory cortex is complex and requires further investigation.

Information in the form of “temporal regularities” are particularly important in audition as a means to parse the auditory scene (Bregman, 1990; Darwin, 1997). Stimuli with similar temporal regularities tend to group together and are heard as a single auditory object. For example, a series of inharmonic tone bursts with simultaneous (synchronous) temporal onsets are heard as one sound that is distinct (i.e., perceptually segregated) from other tone bursts in the environment (Christison-Lagay and Cohen, 2014; Elhilali et al., 2009; Krishnan et al., 2014; Lu et al., 2017; O’Sullivan et al., 2015; Shamma et al., 2013; Teki et al., 2016; Teki et al., 2013; Teki et al., 2011; Thakur et al., 2015).

This temporal information may, in part, be encoded by neural correlations over various lengths of time windows (Panzeri et al., 2022). Longer time windows could filter out faster neural fluctuations that are uncorrelated across neurons, while simultaneously emphasizing the slower, correlated variability (Averbeck and Lee, 2003). In sensory areas, the relevant time windows vary greatly: from less than a millisecond to several hundreds of seconds (Panzeri et al., 2010). This variability in encoding-window length may be a mechanism by which neural ensembles can encode different information at different timescales (Norman-Haignere et al., 2022; Runyan et al., 2017). In particular, longer time windows may facilitate the processing of natural (ethological) auditory stimuli, which are characterized by low rates of temporal modulation(Cohen et al., 2007; DiTullio et al., 2022; Singh and Theunissen, 2003).

This process of encoding different information on different time scales is sometimes referred to as temporal multiplexing (Panzeri et al., 2010). It has been observed in multiple sensory areas, including the auditory system in which neurons in the core and belt auditory cortex encode formants earlier than pitch (Caruso et al., 2018; Panzeri et al., 2010; Walker et al., 2011). Interestingly, the neurons that multiplex information at different time scales are not limited to the cortex: fluctuations in feature encoding are coordinated across neurons in the inferior colliculus. This multiplexing of information at different time scales may be a means by which the brain efficiently codes information.

Beyond the correlation structure that can be gleaned from the spiking activity of groups of neurons, there is also evidence for population coding at larger spatiotemporal scales. This coding may be particularly relevant to multisensory processing. For example, the simultaneous presentation of auditory and non-auditory stimuli increases oscillatory power at different hierarchical levels of the auditory cortex (Karthik et al., 2021; Keil et al., 2013; Schroeder et al., 2008) (Kuroki et al., 2018; Maier et al., 2008; Wang et al., 2019). This cross-areal phase/power synchronization is correlated with improvements in multisensory behavior and appears to play a causal role in multisensory behavior (Hipp et al., 2011; Mercier et al., 2015). Indeed, experimentally induced increases in gamma-band synchronization slows multisensory response times (Misselhorn et al., 2019).

Conclusion and open questions

As discussed throughout, although substantial progress has been made, several open and fundamental questions remain regarding the relationship between neural activity, perception, and decision-making in the auditory cortex. For example, although the contribution of corticofugal activity to audition has received recent attention (Blackwell et al., 2020; Clayton et al., 2020; King et al., 2018; Williamson and Polley, 2019; Yin et al., 2020), we still know relatively little about how feedback activity in the auditory cortex supports hearing. Relatedly, we need to differentiate whether early choice-related activity reflects causal contributions to the current decision or whether it reflects bias- or evaluative-related activity that occurs after the outcome of a decision (Fig. 3). Similarly, what is the relationship between a stimulus, a task, and the contribution of a particular auditory field to decision-making and behavior? Do simple choices engage earlier auditory fields, whereas more complex choices engage later auditory fields? The contribution (if any) of the dorsal auditory pathway to perception need further study. Finally, in addition to the study of pairwise correlations, further work is needed to identify how larger networks of neurons contribute to perception (Francis et al., 2018) and whether these larger networks can best be described by lower-dimensional subspaces (Fitzgerald et al., 2013; Ganguli et al., 2008; Gao et al., 2017; Ni et al., 2018).

Figure 3: Schematic of how and when neural activity can be modulated by behavioral choice.

In each schematic, neural activity is plotted (arbitrary units [a.u.]) relative to the onset of a target stimulus. The first vertical dotted line indicates target onset, whereas the second vertical line indicates target offset. The red and black activity traces in each panel represent possible outcomes for a listener’s different choices (e.g., choosing “high frequency” versus choosing “low frequency”, respectively) in response to the same target stimulus. A: Choice does not modulate activity (black and red traces overlap). B: A particular choice (red curve [e.g., high-frequency choice]) modulates neural activity prior to stimulus onset and throughout target presentation, which is suggestive of some form of expectation or attention. C: Different choices (red and black curves) modulate neural activity only during target-stimulus presentation. D: A particular choice (red curve) modulates neural activity after stimulus onset and could reflect motor processing and/or post-hoc processing of the previous trial and evaluation of the subsequent trial.

Highlights.

1. We review the current understanding of the functional organization of auditory cortex.

2. While much work has been done in auditory cortex, basic question remain unanswered, including why are there multiple primary auditory fields? What is their differential contribution to auditory perception?

3. How do neural ensembles within and across auditory areas contribute to auditory perception?