Principles of auditory information-processing derived from neuroethology

For capture of prey (flying insects) and orientation, the mustached bat (Pteronotus parnellii) emits orientation sounds (biosonar pulses), each of which consists of a long constant-frequency (CF) component followed by a short frequency-modulated (FM) component. Since each orientation sound contains four harmonics (H₂), there are eight components that can be defined (CF_2-4, FM_1-4). In the emitted sound, the second harmonic (H₂) is always predominant and the frequency of CF₂ is about 61kHz (Fig. 1A). The frequency of the CF component is different among subspecies and to some extent among individuals of the same subspecies. It also differs between males and females. For FM₂, the frequency sweeps down from 61 kHz to about 49 kHz. H₃ is 6-12 dB weaker than H₂, and H₁ and H₄ are 18–36 and 12–24 dB weaker than H₂, respectively.

Echoes eliciting behavioral responses in the mustached bat always overlap temporarily with the emitted sound, e.g. Fig. 1A. As a result, biosonar information must be extracted from a complex sound potentially containing up to 16 components. The CF component is an ideal signal for target detection and the measurement of target velocity (relative movements and wing beats), because the reflected sound energy is highly concentrated at a particular frequency. The mustached bat uses the CF component for this purpose and performs an interesting behavior called Doppler-shift compensation (Schnitzler, 1970). The short FM component, however, is suited for ranging, localizing and characterizing a target because of the distribution of its energy over many different frequencies. Different parameters of echoes received by the bat carry different types of information about a target (Fig. ID).

The eight components (CF_1-4 and FM_1-4) of the orientation sound of the mustached bat all differ from each other in frequency, and are analyzed in parallel at different regions of the basilar membrane (Fig. 2, bottom). The signals are then coded and sent into the brain by peripheral neurons. In the brain, the signals are sent up to the auditory cortex through many auditory nuclei where signal-processing takes place. For simplicity, we may consider that there are eight channels for the processing of these signal elements: CF₁ channel, CF₂ channel, and so on. The CF₂ channel is very large relative to any other channel and is associated with an extraordinarily sharply tuned local resonator in the cochlea for fine frequency analysis (Fig. 2).

In the CF₁, CF₂ and CF₃ channels (Fig. 2), frequency-selectivity is increased, and amplitude-selectivity is added by inhibition to some neurons in the cochlear nucleus and also to many neurons at higher levels. In a certain region of the medial geniculate body, part of the CF_X channel and part of the CF₂ or CF₃ channel are integrated, so that neurons in this region respond poorly to the CF_1; CF₂ or CF₃ tone when delivered alone, but respond strongly when the CF₁ tone is delivered together with the CF₂ or CF₃ tone. A deviation of the CF₂ or CF₃ frequency from the exact harmonic relationship with the CF₃ frequency, i.e. an amount of Doppler shift, is a critical parameter for their excitation (Suga, 1984). These CF/CF combination-sensitive neurons project to the CF/CF area of the auditory cortex. In the CF/CF area, two types of CF/CF neurons, CF₁/CF₂ and CF₁/CF₃, are clustered separately and form frequency-vs-frequency coordinates within each cluster for the representation of Doppler shifts, i.e. velocity information (Fig. 3). CF/CF neurons show sharp ‘level-tolerant’ frequency-tuning curves and are remarkably specialized to respond to a particular frequency relationship between the two CF tones (Suga & Tsuzuki, 1985). The signal-processing in the CF channels is thus ‘parallel-hierarchical’.

In the FM_1; FM₂, FM₃ and FM₄ channels (Fig. 2), frequency-selectivity is increased and amplitude-selectivity is added to some neurons by inhibition. Interestingly, FM-selectivity is also added to some neurons by disinhibition, so that these ‘FM-specialized’ neurons respond to FM sounds, but not to CF tones or noise bursts. In a certain region of the medial geniculate body, part of the FMi channel and part of the FM₂, FM₃ or FM4 channels are integrated, so that neurons in this region respond poorly to these FM sounds when delivered alone, but respond strongly to the FM₁ sound combined with the FM₂, FM₃ or FM4 sound. The delay of the FM₂, FM₃ or FM₄ sound from the FM₁ sound, i.e. echo delay, is the critical parameter for their facilitative responses. These FM-FM combination-sensitive neurons act as ‘delay-dependent multipliers’ for processing target range information (Suga, 1989). The delay lines utilized by these neurons are created by neurons responding to the FM₁ sound in the inferior colliculus and also the medial geniculate body. The FM-FM neurons in the medial geniculate body project to the FM-FM area of the auditory cortex. In the FM-FM area, three types of FM-FM neurons, FM₁-FM₂, FM₁-FM₃ and FM₁-FM₄, are clustered separately and form an echo-delay axis in each cluster for the representation of target range information (Fig. 3; Suga, 1984). Therefore, the signal-processing in the FM channels is also parallel-hierarchical.

As described above, part of one channel is integrated with part of the other channel in the medial geniculate body. The remaining parts of these channels project to the auditory cortex, which is not described above. For instance, part of the CF₂ channel projects to the DSCF (Doppler-shifted CF processing) area of the auditory cortex which has frequency-vs-amplitude coordinates to represent target velocity information and subtended target angle information. The DSCF area overrepresents frequencies between the CF₂ resting frequency (about 61 kHz) of the bat’s own sound and 1·0 kHz above it. The DSCF area has two subdivisions which predominantly contain either I-E or E-E binaural neurons (Figs 2, 3). Fig. 2 is only to illustrate the parallel-hierarchical processing of biosonar information which has thus far been explored.

Almost all frequencies found in the biosonar signals are represented not only in the areas which appear to be important for echolocation, but also in the other areas which do not appear to be important for echolocation. These other areas are probably important for processing communication sounds. Except for the CF₂ channel, which is specialized for processing biosonar information, from the periphery to the auditory cortex, a clear separation of biosonar signal processing from non-biosonar signal processing first appears in the medial geniculate body.

The auditory cortex of the mustached bat shows multiple cochleotopic (tono-topic) representation, which is directly related to representation of different types of biosonar information. Fig. 3 shows several functional areas that have been explored electrophysiologically. In these areas, certain response properties of single neurons arranged orthogonally to the cortical surface are identical. (For example, each column in the CF/CF area is characterized by a particular combination of two frequencies.) In this sense, there is a columnar organization. Along the cortical surface, however, the response properties vary systematically and form axes for representation of particular types of biosonar information, as shown in Fig. 3. Among the various functional areas, the CF/CF, FM-FM, DF, VF and VA areas consist of combination-sensitive neurons, so that these areas are particularly interesting in terms of neural mechanisms for processing complex sounds. [For further information of the auditory cortex of the mustached bat, see Suga (1984, 1988).]

The FM-FM area, representing target ranges of up to 310cm, projects to the DF and VA areas of the cerebrum as well as other regions of the brain (Fig. 3). The DF area consists of three clusters of FM-FM neurons. In each cluster, target ranges of up to 140 cm are systematically represented. The DF area projects to the VF area, as well as other areas in the brain to which the FM-FM area does not project. The VF area also consists of three clusters of FM-FM neurons and represents target ranges of up to 80 cm. We do not yet know the functional significance of these multiple-range (delay) axes. One may hypothesize that these three different areas are related to echolocation behavior at different distances to targets. The H₁-H₂ area, part of the VA area, contains combination-sensitive neurons that are different from FM-FM and CF/CF neurons. They show facilitative responses to the CF₂ and/or FM₂ of an echo when these are combined with the CF₁ and/or FM₁ of the biosonar pulse.

Auditory information is sent not only to the association cortex from the auditory cortex, but also to the motor system. Both the FM-FM and CF/CF areas project to the pontine motor nuclei which, in turn, project to the cerebellum. In the cerebellar vermis, there are tiny clusters of FM-FM and CF/CF neurons. Biosonar information is also sent to the vocal system. Some neurons in the periaqueductal gray and midbrain reticular formation, for instance, become active prior to vocalization and respond to acoustic stimuli delivered from a loudspeaker.

The projections of the CF/CF area thus far studied do not overlap with those of the FM-FM area. All the data indicate that complex acoustic signals are processed in a parallel-hierarchical way in the ascending auditory system and beyond the auditory cortex.

A comparison of the data obtained from the mustached bat with those obtained from other species illustrates the specialization of the bat’s auditory system for echolocation and also general neural mechanisms that are probably shared by many different species. These mechanisms are listed below. The data indicating the existence of each mechanism were obtained mainly from the animals listed in parentheses. The data obtained from the owl are related to sound localization, and those obtained from the other species are related to sound reception, frequency analysis and/or processing of complex sounds important to a species.

• (1) The peripheral auditory system has evolved not only for the reception of biologically important sounds, but also for frequency analysis of these sounds that fulfil species-specific requirements. The sharpness of a frequency-tuning curve, sensitivity and/or population can be higher for peripheral neurons tuned to frequencies of sounds that are most important to the species [bats (Suga, 1984; Neuweiler et al. 1980), mice (Brown, 1973a,b) and frogs (Narins & Capranica, 1976)]. A population is larger for neurons with sharper frequency tuning [bats (Suga, 1984; Neuweiler et al. 1980)].

• (2) The frequency tuning of some central neurons is sharpened by lateral inhibition, which eliminates the ‘skirt’ of a frequency-tuning curve [bats (Suga, 1973, 1988, 1989), cats (Katsuki et al. 1958; Evans & Nelson, 1973; Young & Brownell, 1976), mice (Ehret & Moffat, 1985) and frogs (Fuzessery & Feng, 1982)]. The more important the frequency analysis of particular components of sounds, the more pronounced is the neural sharpening for neurons tuned to these components [bats (Suga & Tsuzuki, 1985)].

• (3) The frequency tuning of some other central neurons is broadened by ‘excitatory’ convergence. Broadly tuned neurons are clustered separately from sharply tuned neurons in different portions of the auditory system [bats (Suga, 1973) and cats (Aitkin, 1973; Aitkin et al. 1975; Schreiner & Cynader, 1984)].

• (4) A phase-locked or stimulus-locked response is commonly strong and observed up to 3 kHz at the periphery, but it is weak and rarely observed to stimuli higher than 0·3 kHz in the auditory cortex. The population of ‘phase-locking’ neurons is smaller and the degree of phase-locking is progressively lower at higher levels of the auditory system [cats (de Ribaupierre et al. 1972; Rouiller et al. 1979)]. Thus, a temporal code at the periphery can be changed into a place code at higher levels of the auditory system [bats (Suga, 1984, 1989), owls (Konishi et al. 1988) and frogs (Rose & Capranica, 1984)].

• (5) The cochlea, or part of it, projects in parallel to different subdivisions of a nucleus or nuclear complex at each level of the ascending auditory system [bats (Suga, 1984, 1988), cats (Woolsey, 1961), monkeys (Merzenich & Brugge, 1973), owls (Konishi et al. 1988) and frogs (Hall & Feng, 1987)]. These multiple cochleotopic or tonotopic representations result from the divergence of axons. This divergence is usually associated with a convergence of axons for sorting out different types of auditory information. This combined divergence-convergence occurs repeatedly in the central auditory system and is the anatomical basis of parallel-hierarchical processing of information for both acoustic pattern recognition and sound localization. By this divergence-convergence, neural filters are created which are tuned to various information-bearing parameters (IBPs) other than frequency [bats (Suga, 1984,1988), owls (Konishi et al. 1988) and frogs (Hall & Feng, 1987)]. These IBP-tuned neurons (hereafter, IBP neurons or filters) act as cross-correlators which correlate incoming signals with their filter properties, i.e. with neurally stored information. It should be noted that ‘biologically important’ complex sounds are processed by combination-sensitive neurons, i.e. IBP filters tuned to different combinations of signal elements [bats (Suga, 1984, 1988), song birds (Margoliash, 1983) and frogs (Mudry et al. 1977; Fuzessery & Feng, 1983)].

• (6) IBP filters can be sharpened by lateral inhibition [bats (Suga, 1988, 1989; Suga & Tsuzuki, 1985), owls (Konishi et al. 1988) and frogs (Rose & Capranica, 1984)].

• (7) Different types of IBP filters are clustered separately at particular locations of the central auditory system. In other words, the system contains functional subdivisions or areas specialized for processing particular types of auditory information important to a species [bats (Suga, 1984, 1988), owls (Konishi et al. 1988) and frogs (Hall & Feng, 1987)].

• (8) In each subdivision or area, IBP filters are systematically arranged so that they form an axis or axes representing the IBP or IBPs [bats (Suga, 1984,1988) and owls (Konishi et al. 1988)]. If small differences in IBP values are not biologically important, the IBP axis may not be formed within the subdivision [frogs (Hall & Feng, 1987)]. It should be noted that, with the exception of frequency, there is no peripheral anatomical basis for IBP axes: they are created centrally from neural interactions. That is, they are computational axes or maps.

• (9) The axis and/or population of neurons representing an IBP is apportioned according to the species-specific importance of the IBP [bats (Suga, 1984, 1988) and owls (Konishi et al. 1988)].

• (10) The bandwidth of IBP filters is not so narrow as to express a particular value of an IBP by the excitation of only a few neurons located at a single location along the IBP axis. Even after the sharpening of the tuning of IBP filters by lateral inhibition, it is expressed by a spatiotemporal pattern of excitation of many neurons distributed along the IBP axis [bats (Suga, 1984, 1988) and owls (Konishi et al. 1988)].

• (11) The functional organization of the auditory system can be different among different species, reflecting differences in species-specific auditory behavior and/or the properties of the acoustic signals used by them. The organization can also be different among individuals or between sexes within the same species when the properties of their biologically important acoustic signals are different among conspecifics or sexes [bats (Suga et al. 1987) and frogs (Narins & Capranica, 1976)].

• (12) The auditory cortex consists of specialized areas excited only by biologically important sounds and an unspecialized area (primary auditory cortex) excited by less important and unfamiliar sounds as well as by the biologically important sounds [bats (Suga, 1984)]. The primary auditory cortex is tonotopically organized and contains neurons somewhat similar to peripheral neurons, probably for maintaining ‘raw data’.

• (13) For protection of information-processing during and immediately after vocalization, vocal self-stimulation is reduced not only by the middle ear muscles, but also by inhibition occurring in the central auditory system [bats (Suga & Shimozawa, 1974), monkeys (Müller-Preuss, 1980) and song birds (McCasland & Konishi, 1981)].

• (14) Cortical representation of certain types of auditory information by combination-sensitive neurons is protected from masking by their unique response properties [bats (Suga, 1984)].

The work on the auditory system of the mustached bat has been supported by a research grant from the US Public Health Service, RO1-NS17333 Javits Neuro-science Investigation Award.

This article contains the materials in my two previous articles: Parallel-hierarchical processing of biosonar information in the mustached bat (In “Animal Sonar”, ed. P. E. Nachtigall & P. W. B. Moore, Plenum Press, 1988,149-159) and What does single-unit analysis in the auditory cortex tell us about auditory information processing in the auditory system? (In “Neurobiology of the Neocor-tex”, ed. P. Rakic & W. Singer, John Wiley and Sons, 1988, 331-349).