Understanding the neurophysiological basis of auditory abilities for social communication: A perspective on the value of ethological paradigms

Abstract

Acoustic communication between animals requires them to detect, discriminate, and categorize conspecific or heterospecific vocalizations in their natural environment. Laboratory studies of the auditory-processing abilities that facilitate these tasks have typically employed a broad range of acoustic stimuli, ranging from natural sounds like vocalizations to “artificial” sounds like pure tones and noise bursts. However, even when using vocalizations, laboratory studies often test abilities like categorization in relatively artificial contexts. Consequently, it is not clear whether neural and behavioral correlates of these tasks (1) reflect extensive operant training, which drives plastic changes in auditory pathways, or (2) the innate capacity of the animal and its auditory system. Here, we review a number of recent studies, which suggest that adopting more ethological paradigms utilizing natural communication contexts are scientifically important for elucidating how the auditory system normally processes and learns communication sounds. Additionally, since learning the meaning of communication sounds generally involves social interactions that engage neuromodulatory systems differently than laboratory-based conditioning paradigms, we argue that scientists need to pursue more ethological approaches to more fully inform our understanding of how the auditory system is engaged during acoustic communication.

1. Introduction

Acoustic communication between individuals of the same (conspecific) or different (heterospecific) species is an essential ingredient for the reproductive success and survival of many animals, yet our understanding of its underlying neurobiological mechanisms is still limited. In particular, the neural bases for the functional abilities that enable auditory communication are not fully understood. For example, in order to communicate, animals have to detect, discriminate and categorize a vocalization before being able to use it within their ecological and social contexts to guide behavior (Cheney et al., 1980; Fischer et al., 2011; Ghazanfar et al., 2004). What neural-, circuit-, and systems-level properties within the auditory pathway facilitate these abilities?

For the complex task of acoustic communication between individuals, especially in mammals, our knowledge of its neural basis is limited because of the technical and methodological challenges in designing appropriate experiments. Instead, most auditory neurobiology studies have utilized more controlled laboratory paradigms, which investigate “general” auditory processing of sounds, and use behaviors that do not involve social communication. From these studies, scientists often infer how an animal and its brain would function in natural communication contexts. Such “generalist” approaches have certainly yielded important insights into the basic mechanisms and limits of auditory processing but applying that knowledge to elucidate what occurs in the natural communication context remains an open area of investigation.

This leads to the key motivation for our review. We make two overall arguments in favor of a continued expansion of a complementary, neuroethological approach to studying auditory processing with ever-more naturalistic communication paradigms, despite their technical challenges. Indeed, a long tradition of auditory neuroethological research in model organisms like crickets, frogs, bats and songbirds (Covey et al., 1999; Gentner et al., 2003a; Hahnloser et al., 2010; Konishi, 2004; Moss et al., 2011; Neuweiler, 1990; Pollak, 1992; Razak et al., 2010; Suga, 1988; Tsuzuki et al., 1988; Woolley, 2012) has established a conceptual foundation that can be used to approach similar questions in what some have considered more generalist laboratory animal models, like non-human primates and rodents. For those studying these latter species, it may not always seem necessary to employ natural vocal stimuli and behaviors to understand the general auditory-processing abilities of sound detection, discrimination or categorization (Dylla et al., 2013; Heffner et al., 2001; Klink et al., 2004; Ohl et al., 2001; Recanzone et al., 1993; Talwar et al., 2001). Here, we first try to counter this impression by using a few recent examples to illustrate how ethologically motivated approaches using vocalizations (or other “ethologically valid” stimuli with acoustic features derived from natural sounds) can provide insights into neural mechanisms that might otherwise be difficult to uncover.

Building on this, our second main point is that an important new direction in vocalization studies is to develop experimental designs that comprehensively incorporate behavioral contexts into neurophysiological research. This is important because behavioral context can profoundly affect neural processing. Thus, if we are to make progress understanding the neural mechanisms underlying acoustic communication, we must try to study how these mechanisms are engaged in more naturalistic communication behaviors.

In making these two arguments, an important question that arises is whether the neural substrates for vocalization processing differ from those used to process other types of sounds. For example, vocalizations may be more intrinsically arousing than other sounds and, thus, increase engagement of limbic areas relative to those other sounds (Ehret, 2005). Do these differences arise from evolutionarily tuned innate mechanisms present at birth or because of extensive experience during development and/or adulthood for ethologically important communication sounds? In other words, are vocalizations processed just like any another complex sound, or are they “special” in engaging unique (or, at least, different) processing mechanisms? Whereas recent findings begin to speak to this question, it is likely irresolvable in many species. Indeed, the ideal experiment would require the prohibitively burdensome task of providing an animal from birth as much experience hearing and responding to a category of non-vocal (e.g. synthetic) sounds as a “natural” category of species-specific vocalizations, and then performing neurophysiological studies to uncover differences.

Instead, we advocate that hearing scientists embrace the basic principle from neuroethology (Ewert et al., 1983): neural activity needs to be considered in the full context of the elicited behaviors. Following this principle, we believe that the more tractable answer to the aforementioned problem is to identify how (1) vocalizations and (2) synthetic stimuli that become behaviorally salient through operant conditioning differentially modulate neural transformations along the auditory pathway from an acoustic input to behavioral output. In fact, several studies have already suggested that the neural representations of non-vocalizations (Bieszczad et al., 2010b; David et al., 2012; Polley et al., 2006) and vocalizations (Gentner et al., 2003b) depend on the details of a trained behavioral task (e.g. contingencies, rewards, strategy). It is precisely because of the importance of such details in the manifestation of auditory processing that we argue that to truly understand how such processing proceeds in the context of real acoustic communication, we must move towards experimental designs that may capture more of this natural behavioral context (DiMattina et al., 2006; Fortune et al., 2011).

To proceed, we first adopt a framework for testing how the neural representation of an acoustic-communication signal is linked in the brain to behavior that could also be used for other sounds that are not used in vocal communication. We make no attempt to ascribe any of the subsequent processing functions to specific brain regions. Instead, we discuss the computational and processing steps that we hypothesize must take place for an animal to use a communication signal to guide behavior (Griffiths et al., 2004).

Consider an animal hearing a species-specific vocalization that signals both the presence of food and contains information about the identity of the vocalizer. The first stage of auditory processing is to transduce the vocalization’s acoustic energy into a neural signal reflecting the vocalization’s spectral and temporal properties. Complex sounds such as vocalizations can be thought of as specific combinations of individual features, like a call’s pitch or temporal envelope. The neural representations of such acoustic features are then bound together through sequential and simultaneous grouping principles to form an “auditory object”, which is the fundamental perceptual unit in audition (Griffiths et al., 2004; Shamma et al., 2011; Winkler et al., 2009). This representation must then be interpreted in a framework that converts this information into a behavioral judgment (i.e., an auditory decision). Throughout this process, the representation of the vocalization is presumably compared with memory stores to categorize this call and recognize its “meaning”. Once this information has been referenced, it has to be interpreted within the context of an animal’s current behavioral goals. Ultimately then, the bottom-up representation of the acoustic signal must be interpreted in the top-down framework of the animal’s behavioral and social experiences.

Importantly, these hypothesized steps could relate to any auditory stimulus that has behavioral relevance to the animal (i.e., those that inform an animal’s current or future behavior). As outlined above, studying these steps for acoustic communication not only provides a concrete, ethologically relevant context for elucidating functional auditory processing abilities but also adds value to our ability to clarify neural mechanisms. To make this point explicit, we use the remainder of this review to discuss some of these general auditory steps (acoustic-feature encoding and categorization) in more detail and illustrate how ethological approaches have contributed to revealing their underlying mechanisms.

2. Value of ethological approaches to studying acoustic-feature encoding

In early stages of processing, the auditory system encodes the spectrotemporal features of a stimulus. These features include tonal components, noise components, amplitude modulations, and frequency modulations (Attias et al., 1997; Kanwal et al., 1994; Liu et al., 2003; Morton, 1977). These features, which are universally present in many species’ vocalizations, are classically thought to correlate with a sender’s internal motivation level arising from specific hostile or friendly contexts (August et al., 1987; Morton, 1977). For example, low-frequency noisy utterances in many species transmit information about hostility. In contrast, high-frequency tonal calls transmit information about fear. Importantly, this convergence in acoustic structure arises irrespective of whether or not non-human animals intentionally transmit motivational or other types of information (Cheney et al., 1992).

Auditory neurophysiologists have generally approached the neural encoding of such acoustic features in one of several different complementary ways. The most traditional approach is simply to define a discrete feature class, like a noise band or a tone frequency, and then acquire neural spike-count responses to systematically parameterized versions of these features (e.g., changes in frequency) to produce a tuning curve. Alternatively, without first presuming the classes of relevant features, one can use synthetic stimuli of varying mathematically parameterized complexity – like ripples (Calhoun et al., 1998; Depireux et al., 2001; Escabi et al., 2003; Kowalski et al., 1996a; Kowalski et al., 1996b) or random chords (deCharms et al., 1998; Linden et al., 2003) – to probe neural responses in a more unbiased fashion and produce spectrotemporal receptive fields (STRFs) (Ahrens et al., 2008; Calabrese et al., 2011; Holmstrom et al., 2007).

These approaches have been broadly used to test auditory sensitivity of neurons from the auditory periphery to the prefrontal cortex (Abeles et al., 1972; Cohen et al., 2007; Escabi et al., 2002; Linden et al., 2003; Miller et al., 2002; Phillips et al., 1981; Sachs et al., 1968; Temchin et al., 2005). However, in practice, they have two limitations that can impede further progress in elucidating the encoding of acoustic features. First, passively stimulating animals with just tone bursts or even band-limited noise often does not evoke strong neural responses in higher cortical areas (Rauschecker et al., 1995; Stiebler et al, 1997; Romanski and Goldman-Rakic, 2002;), making the calculation of tuning curves and STRFs difficult (Averbeck et al., 2006; Cohen et al., 2007). Interestingly though, when tone bursts are presented in the context of a behavioral task, they can actually elicit strong responses, even in the prefrontal cortex (J Tsunada and YE Cohen, unpublished observation). This emphasizes the importance of the behavioral relevance of sounds in order to evoke responses.

In this sense, ethologically valid sounds may have an advantage for probing neural encoding in an animal’s auditory system, since this system has presumably evolved to more efficiently transmit information about the natural variability of acoustic features in sounds that are intrinsically relevant (Cohen et al., 2007; Margoliash, 1983; Margoliash, 1986; Ohlemiller et al., 1996; Rauschecker et al., 1995; Scheich et al., 1979; Singh et al., 2003; Woolley et al., 2005). For example, Rieke and colleagues (1995) found that the spiking activity of bullfrog auditory-nerve fibers encodes (in the information-theoretic sense) natural amplitude spectra (i.e., spectra that mimic a bullfrog call) better than spectrally matched broadband Gaussian (random) noise. Likewise, in the cat inferior colliculus, ethologically valid amplitude-modulation rate (Attias et al., 1998) and amplitude-contrast distributions (Escabi et al., 2003) also yield better information transmission than non-natural distributions. In the auditory cortex, the natural rhythm of vocalizations generally evokes stronger responsiveness than temporally compressed or expanded versions of calls (Carruthers et al., 2013; Chandrasekaran et al., 2009; Wang et al., 1995). A bias for the ethologically valid range of temporal modulation may also explain why it is easier to drive certain forms of auditory cortical plasticity for sound rhythms that are typically found in natural vocalizations (Chandrasekaran et al., 2009; Kim et al., 2009; Liu et al., 2006). These data reinforce the idea that the auditory system is adapted to the statistical structure of acoustic features that are present in natural stimuli, including vocalizations.

A second limitation of the standard approaches to characterizing auditory neurons stems from the fact that auditory neurons exhibit substantial nonlinearities in the time and frequency domains, thereby diminishing the predictive ability of the linear STRF or tuning curve to explain auditory encoding (Bar-Yosef et al., 2002; Carruthers et al., 2013; Cohen et al., 2007; Escabi et al., 2002; Machens et al., 2004; Theunissen et al., 2000). Whereas findings have identified methods to describe the multiple acoustic dimensions that drive neural activity and their associated nonlinearities (Atencio et al., 2008), these methods are often too computationally intensive to be widely used.

Instead, ethological approaches that explicitly test for nonlinear sensitivities to naturally arising combinations of ethologically valid acoustic features (Arnold et al., 2006; Balaban, 1988; Holy et al., 2005) have proven to be informative. Neurons in multiple brain regions have been found to be more responsive to combinations of acoustic features than to the individual contributions of each feature (Fuzessery et al., 1983; Kanwal et al., 2007; Leppelsack, 1978; Margoliash, 1983; Margoliash et al., 1992; Portfors, 2004; Portfors et al., 1999; Rauschecker et al., 2000; Rauschecker et al., 1995; Suga, 1978). For example, in the bat auditory cortex, neural activity in response to a sequence of stimuli that simulate an echolocation pulse and its reflection exceeds the sum of the responses to the individual stimuli when the sequence is presented within an ethologically appropriate temporal window (Suga et al., 1979). Similarly, neurons in the auditory cortex have been reported to be more responsive to a species-specific vocalization when it is presented in the normal forward direction than when it is presented backwards (Rauschecker et al., 1995; Wang et al., 1995). Both of these examples highlight the fact that the auditory system not only cares about the spectral content of acoustic features but also their non-linear temporal organization. More importantly, they demonstrate a neural sensitivity to the natural complexity of behaviorally relevant sounds, which motivates studying acoustic feature encoding with ethologically valid stimuli.

3. Value of ethological approaches to studying acoustic categorization

The neural representation of acoustic features enables functional auditory decisions to be made to guide behavior, including the detection, discrimination and categorization of sounds. Here, we discuss categorization in more detail and again argue that ethological approaches may offer advantages for revealing underlying neural mechanisms.

The need to categorize sounds arises because discriminating acoustic variability may not always be necessary or desirable for a required behavioral judgment. In the context of communication, this variability may appear due to intrinsic differences in the vocalizer, such as male vs. female, large vs. small or young vs. old (Liu et al., 2003; Peterson et al., 1952; Riede et al., 1999); the vocalizer’s emotional state, such as stressed vs. relaxed (Bachorowski et al., 1995; Streeter et al., 1983); or simply environmental conditions like room reverberation (Houtgast et al., 1980). Depending on the behavioral situation, variability could be important for guiding differential responses or should be ignored to drive a specific stereotyped response triggered by the sound category.

Progress in understanding the neural basis of acoustic categorization has typically used sounds that are not ethologically relevant for the species being studied. In particular, speech phonemes, which humans perceive categorically (Harnad, 1987), have been used in many auditory neurophysiological studies conducted in animals from rodents to non-human primates (Engineer et al., 2008; Lee et al., 2009; Mesgarani et al., 2008; Russ et al., 2007; Russ et al., 2008; Steinschneider et al., 2003; Tsunada et al., 2011; Tsunada et al., 2012). The assumption has been that basic auditory mechanisms shared across species contribute to this ability (Kuhl, 1981). For example, once trained to discriminate speech sounds, monkeys can readily categorize their morphs (Lee et al., 2009; Russ et al., 2007; Tsunada et al., 2011; Tsunada et al., 2012). These categories are represented in the ventral prefrontal cortex. Importantly, neural activity in the prefrontal cortex during categorization also predicts behavioral choice (Russ et al, 2008). Earlier in the stimulus-behavior transformation, these categories are differentially represented between interneurons and pyramidal neurons in the belt region of the auditory cortex (Tsunada et al., 2012), but this belt activity is not modulated by the monkey’s categorical choices. (Tsunada et al., 2011). Finally, a study in mongolian gerbils even suggests that primary auditory cortex can exhibit categorical neural activity after animals learn to categorize frequency sweeps (Ohl et al., 2001).

Whereas these studies provide some important insights into neural mechanisms for acoustic categorization, they are limited by the fact that the sounds do not have intrinsic meaning for the animals. To further advance our understanding of brain mechanisms for categorization in the context of acoustic communication, the communication stimuli used and the behaviors studied should be species-specific. This helps ensure that the animals have been exposed to the natural acoustic variability that triggers the natural, stereotyped response (although the range of exposure of animals raised in a laboratory-based vivarium may be more limited than their wild counterparts). In contrast, most of the paradigms described above using human speech sounds in animals require explicit training (often with considerable time and effort) in a new behavior so that experimenters can infer perception. Such paradigms generally involve instrumental conditioning rather than social interactions for learning the meaning of a sound category. This could alter the nature of the learned categorical representations, as other studies have shown that the way in which stimuli are learned affects how they are encoded (Bieszczad et al., 2010a; David et al., 2012; Gentner et al., 2003b; Polley et al., 2006). Finally, the range of stimuli is more limited than would naturally be encountered if those stimuli were part of a natural communication repertoire, potentially affecting the way in which categorization is achieved.

The strategy of using species-specific vocalizations has led to important progress understanding the neural basis of human speech categorization (Chang et al., 2010), but so far only at the level of local field potentials. A clearer picture at the single neuron level will, practically speaking, require animal models. Thus, developing ethological paradigms using species-specific vocalization categories that elicit stereotyped responses is essential. Some intriguing approaches are antiphonal calling in non-human primates (Miller et al., 2009), spontaneous categorization of food vocalizations in non-human primates (Gifford et al., 2005), and dueting in songbirds (Fortune et al., 2011), wherein animals respond to hearing conspecific vocalizations by vocalizing themselves.

Another paradigm being actively pursued is the maternal response to infant vocalizations in rodents, particularly mice (Cohen et al., 2011; Fichtel et al., 1999; Liu et al., 2006; Liu et al., 2003). Mouse pups emit ultrasonic calls when isolated from their nest; these calls elicit a search and retrieval behavior in mothers (Ehret, 2005). This behavior likely involves some experience-dependent acquisition, since virgin females begin preferring calls after helping to raise pups (Ehret et al., 1987; Lin et al., 2013). In terms of acoustic categorization, the pup calls form a natural acoustic category in the frequency and duration domains that can be discriminated from other mouse ultrasonic vocalizations (Liu et al., 2003), and mouse mothers actually perceive ultrasonic sound frequency, bandwidth and duration categorically (Ehret, 1992; Ehret et al., 1981).

The behavioral findings have motivated the use of the maternal model to explore how the ultrasonic pup call category is represented in the auditory cortex (Liu et al., 2007; Liu et al., 2006). This has helped uncover novel mechanisms, like inhibitory plasticity (Galindo-Leon et al., 2009; Lin et al., 2013; Vogels et al., 2013) and multimodal olfactory-auditory integration (Cohen et al., 2011). In the latter result, pup-naïve virgin females exhibited only minimal multimodal integration in contrast to mothers, thereby emphasizing the importance of using behaviorally relevant, species-specific communication sound categories to reveal novel neural mechanisms.

4. Do vocalizations engage “special” processing or plasticity mechanisms?

Given our argument for the use of ethological paradigms to study communication, it may seem that we advocate the position that the auditory system has a preferential bias towards a specific class of acoustic stimuli, namely vocalizations. In fact, though, whether vocalizations have some sort of processing or plasticity “privilege” above and beyond other natural stimuli is still an active debate, and we cannot make any definitive conclusions. Nevertheless, there is some evidence in favor of preferential processing. For example, behavioral studies in non-human primates point to preferential processing of vocalizations over other natural sounds: a monkey’s performance during a working-memory task is better when he is required to remember a species-specific vocalization than when he is required to remember other sounds, like human voices and music (Ng et al., 2009).

Perhaps, the best evidence for vocalization specialization comes from work showing that in the primate, regions in the anterior portion of the temporal lobe respond selectively not only to vocalizations but also to the identity of the vocalizer (Perrodin et al., 2011; Petkov et al., 2008). Consistent with these findings, human patients with damage in the temporal and parietal cortices have deficits in voice recognition and discrimination (i.e., phonagnosia) (Van Lancker et al., 1982; Van Lancker et al., 1988). However, this kind of selectivity lies at the apex of a series of computational steps that have yet to be fully identified, including the learning of perceptual categorization and computations that create neural invariance (tolerance) to “identity-preserving” (e.g., changes in loudness, location etc.) changes in a sound. It may be that those same steps are employed for general auditory processing of sound categories that are not vocalizations, albeit potentially in other brain areas. In the case of either vocal or non-vocal stimuli though, to better understand the nature of category specialization, it will be important to differentiate between neural variability that is elicited by different vocalizers or sources producing the same sound (i.e., inter-source variability) from variability that is elicited by the same source producing different sounds (intra-source variability).

This prompts one to ask whether there might be a progression along the ascending auditory system in the weighting of different classes of acoustic communication stimuli, so that neurons in certain auditory regions begin showing greater (or perhaps exclusive) selectivity for the natural acoustic structure of vocalizations over that of other natural valid stimuli. Indeed, Woolley et al (2005) (Woolley et al., 2005) found that the spectrotemporal tuning of neurons in the avian analogues of the auditory thalamus and cortex deemphasizes the spectrotemporal modulations that are common to most natural stimulus classes and, instead, accentuates those unique modulations that allow a listener to discriminate between different stimulus classes. Importantly, though, at this level, there is no exclusive responsiveness to the spectrotemporal features of vocalizations: this exclusivity emerges in the sensorimotor-integration areas of the songbird’s song-production system.

If vocalizations are indeed privileged, it is still an open question whether this specialization is genetically “hard wired” or arises from learning or the animal’s “overexposure” to these behaviorally salient stimuli (Egnor et al., 2004); most likely, it is a combination of both. Recent research has begun to address such questions through the use of neurophysiology in scenarios where vocalizations acquire salience through natural social interactions. For instance, songbird research has begun exploring the neural changes in the auditory forebrain of females as they learn the “meaning” of different male songs. Woolley and Doupe (Woolley et al., 2008) found that female zebra finches prefer their mate’s song more than an unfamiliar conspecific’s song. Moreover, the expression of the immediate early gene zenk in the caudomedial nidopallium (NCM), which is an analog of the belt auditory cortex, is higher when female finches listen to their mate’s song than when they listen to an unfamiliar conspecific’s song. The songbird NCM also happens to be a site where estrogen, a so-called social neurochemical (Choleris et al., 2003) that is upregulated during mating season, can modulate neural activity and plasticity in response to songs (Maney et al., 2006; Remage-Healey et al., 2012; Tremere et al., 2011; Tremere et al., 2009; Tremere et al., 2012). This further emphasizes the importance of exploiting the natural ethological context to study auditory processing, since the involvement of these hormonal neurochemicals could well be different in processing sounds that are not behaviorally salient (Maney et al., 2006).

The importance of hormones in plasticity and the processing of communication signals is also beginning to be implicated in the maternal communication system between mouse pups and their mothers (Ehret et al., 2009; Lin et al., 2013; Miranda et al., 2009). Overall, research in this area is now turning to the roles of social experience with pups and maternal hormonal modulation in mediating neural coding changes, directions that may one day inform strategies to ameliorate deficits in real-life communication learning.

5. Conclusions

If a primary goal of hearing scientists is to elucidate the neural mechanisms for many of the auditory abilities that contribute to natural acoustic communication, then we advocate that research efforts should more fully embrace ethological paradigms that involve species-specific communication sounds and actual communication behavior. While recent ethologically motivated studies outlined here have made great strides on the stimulus side, much more still needs to be done to realize the natural behavioral side. We, therefore, end by briefly considering advances still needed that could help natural ethological paradigms for communication overcome the practical difficulties of its laboratory study. For example, the development of newer wireless neural recording techniques (Roy et al., 2012; Szuts et al., 2011) in non-human primates would allow freer and more natural movement during neural recordings. Second, even though behavior may become less restricted, trial-by-trial behavioral variability would likely persist, motivating the need to be able to derive neural results from individual behavioral trials. Improving the ability to record a greater number of neurons and to apply statistical and computational methods for single-trial analyses would therefore be important (Kass et al., 2005; Yu et al., 2009). The discussion above suggests that adopting natural communication stimuli and natural behavioral paradigms for auditory research would be desirable for studying how the auditory system normally functions and learns to guide behavior. Indeed, the neural basis for action selection is poorly understood for auditory-evoked behaviors in general, and robust natural tasks would, in principle, be greatly advantageous.

Highlights.

1. Processing communication sounds depends on general auditory abilities

2. Advantages to ethological paradigms to study general auditory abilities

3. Neuroethology has elucidated higher-order processing and nonlinear sensitivity

4. Neuroethology can help elucidate categorization of naturally variable calls

5. To test vocalization specialization, must study call-to-behavior transformations

Abbreviations

A1

primary auditory cortex

NCM

caudomedial nidopallium

STRF

spectrotemporal receptive field