Neural binaural sensitivity at high sound speeds: Single cell responses in cat midbrain to fast-changing interaural time differences of broadband sounds

Abstract

Relative motion between the body and the outside world is a rich source of information. Neural selectivity to motion is well-established in several sensory systems, but is controversial in hearing. This study examines neural sensitivity to changes in the instantaneous interaural time difference of sounds at the two ears. Midbrain neurons track such changes up to extremely high speeds, show only a coarse dependence of firing rate on speed, and lack directional selectivity. These results argue against the presence of selectivity to auditory motion at the level of the midbrain, but reveal an acuity which enables coding of fast-fluctuating binaural cues in realistic sound environments.

1. Introduction

In several sensory systems, relative motion between receptor organs and the outside world is processed by dedicated neural systems, as it yields much information regarding the subject, the external world, and their mutual relationship. While neural selectivity to aspects of motion is well-established in vestibular, visual, somatosensory, and electrosensory systems, it is controversial in hearing (Carlile and Leung, 2016; Kaczmarek, 2005; Lutfi and Wang, 1999; Magezi et al., 2013; McAlpine et al., 2000; Poirier et al., 2017; Smith et al., 2004, 2010). For humans, the dominant source of auditory spatial information is the interaural time difference (ITD) at low frequencies, which they can discriminate at values of tens of microseconds. But humans are strikingly insensitive to dynamic changes in ITD, so-called “binaural sluggishness” (Grantham and Wightman, 1978).

Neural sensitivity to ITDs arises in the brainstem through a form of coincidence detection (Franken et al., 2015). At a further level, in the midbrain's inferior colliculus (IC), neurons show dynamic sensitivity to tones rocking back and forth in ITDs, roughly corresponding to sounds oscillating in azimuthal position (Spitzer and Semple, 1991, 1993). This midbrain sensitivity to changes in ITDs rather than to the absolute or instantaneous ITDs has been interpreted as sensitivity to motion, and is not found at the lower brainstem level (Spitzer and Semple, 1998). Also, directional selectivity—a hallmark of motion processing in the visual modality—has been observed in some IC neurons, to tones with dynamically changing ITDs (Yin and Kuwada, 1983). However, these phenomena were studied with pure tones, which are virtually nonexistent in natural environments, are difficult to localize in space, and evoke only a weak percept of motion.

We recorded the response of IC neurons to broadband noise, on which we imposed linear sweeps of ITD. Informal listening reveals that this stimulus causes a much stronger sense of a moving, punctate source. We found that neurons lack tuning to direction or to specific velocities but track instantaneous ITDs up to unnaturally high speeds. We surmise that the lack of motion selectivity but acuity to very fast auditory stereo is dictated by the physics of sound and lack of spatial mapping at the receptor organ, and that it allows mammals to deal with multiple sound sources and reflections, which are an integral part of natural environments.

2. Methods

A total of 566 datasets at different speeds and 2 directions was collected from 64 single neurons recorded from the central nucleus of the IC in 13 pentobarbital-anesthetized cats. General methods are as described in Joris (2003). Procedures were approved by the University of Wisconsin Animal Care Committee. Sounds were delivered by two dynamic speakers coupled to the ear canals and were calibrated with probe tubes near the eardrum. Stimuli were generated digitally with a 16-bit custom-built digital stimulus system. Continuous, linear changes in ITDs were created by manipulating the system's digital-to-analog conversion (DAC) rate. The magnitude and sign of the difference in the DAC rate for the left and right ear channels determined the speed and direction of simulated motion. By also superimposing an onset-ITD, the range of ITDs over which motion occurred was specified. The stimulus was Gaussian pseudorandom noise (100–4000 Hz), generated with a random number generator. The overall stimulus level was usually 80 dB sound pressure level, which was on average 30 dB above the threshold to static noise. Begin- and end-ITDs were symmetrical around 0 μs and were tailored to bracket the main features of the stationary ITD curve. At speeds >1000 μs/s, larger ITD ranges were often used (up to ±5000 μs) to have a sufficient stimulus duration. Nominal values are stated for speed and end-ITDs: actual speeds differed slightly (<5%) due to hardware limitations. For a given neuron, the ITD-trajectory at different speeds was kept constant (i.e., the same start- and end-ITDs were used at all speeds), and the duration of the stimulus decreased proportional to increasing speed. The number of repetitions of each stimulus condition was increased with speed to keep the total recording time roughly constant. Figures 1(A) and 1(B) illustrate how the responses are converted to functions of instantaneous ITDs. It is well-known that the response of IC neurons to pseudorandom noise contains a temporal structure (e.g., Joris et al., 2006), so that post-stimulus time histograms (PSTHs) show pronounced peaks and pauses which reflect locking of spikes to the fine-structure and envelope of the effective stimulus waveform. This temporal structure hinders examination of the responses to changes in ITDs, particularly at very high speeds [cf. Figs. 1(A), 1(B), and 2(A)]. To mimic free-running dichotic noise, different repetitions used different starting points of a long noise waveform stored in memory, effectively “unfreezing” the pseudorandom noise so that the response peaks and pauses were not lined up across repetitions.

Fig. 1.

(Color online) (A) and (B) Construction of tuning curves to instantaneous ITD. The histograms show responses of one neuron for one direction of motion and three speeds. (A) Responses plotted as PSTHs on an identical time scale. Each doubling in speed results in a halving of stimulus duration. (B) PSTHs scaled to equal length for stimulus duration (10000, 5000, and 2500 ms). Bottom panel: scaled PSTHs plotted as curves, for various speeds (color code in left column), with the abscissa relabeled to instantaneous interaural time delay (ITD), obtained as (start ITD + post-stimulus-time*speed). (C)–(F) Coding of instantaneous ITD in four midbrain neurons. Responses to two directions of motion (arrows) are shown for a range of speeds between 100 and 1000 μs/s (color bar). Responses are to correlated noise, except in (E) (bottom panel) which is to anticorrelated noise. Start- and end-ITD (up- and downward arrowheads on abscissa) were fixed at values corresponding to a sound source on the far ipsilateral (ITD <0) or contralateral (ITD >0) side (relative to the side of the recording). The dashed line shows the response to static ITDs, changed in steps indicated by the filled circles. The frequency to which the cell is most sensitive is indicated next to the panel label. For reference, ITD curves show a vertical line at 0 ITD and at the ITD generating the largest response to static ITDs.

Fig. 2.

(Color online): Preservation of ITD-sensitivity at extreme speeds. (A) Example for a neuron in which speeds over >3 orders of magnitude were tested. Conventions as in Figs. 1(C)–1(F). (B)–(G) Population data showing preservation of ITD-sensitivity at high speeds and lack of directional tuning. (B) Maximal firing rate tends to increase with increasing speed. (C) Particularly at high speeds (>1000 μs/s), maximal firing rates tended to be higher to moving than to static stimuli. The horizontal line is the line of equality. (D) Modulation depth (max-min/max) of the ITD curve tends to increase with speed. (E) The tuning to moving ITD is well-correlated with that to stationary ITDs at low speeds. Correlation decreases with increasing speed but many neurons retain high correlation up to very high speeds. (F) Delay increases linearly with speed. For reference, the solid line without symbols indicates the slope that would result from a 10 ms delay. (G) Absence of direction selectivity: maximal firing rates were well-matched for the two directions of motion and thus cluster around the diagonal of equality (solid line). The dashed lines indicate the criterion of Yin and Kuwada (1983) for directional selectivity (rate in preferred direction is twice that in the anti-preferred direction). In (B) and (E), lines connect values from the same cell. Data from 61 neurons in 13 animals.

3. Results

Responses from four representative neurons to ITD sweeps are shown in Fig. 1. The dashed line and filled circles [Figs. 1(C)–1(F)] show the response to stationary ITDs, varied in discrete steps: here the speed is zero and the same response curve is shown in the upper and lower panels for each cell. Clearly, the neurons are sensitive to ITD. Two directions of motion were presented (arrows, top and bottom panel), at speeds between 100 and 1000 μs/s (black to bright shading). The most striking feature of the responses is the preservation of ITD-sensitivity at all speeds. For the neurons in Figs. 1(C) and 1(F), which are of the “peak-type” and “trough-type,” respectively (reflecting a coincidence and “anti-coincidence” process), the response is virtually independent of speed and direction. Motion affects the response of the neuron in Fig. 1(D) by a shift of the central peak along the ITD axis, and by an increase in response rate with speed in Figs. 1(D) and 1(E). However, the overall response pattern is highly correlated between the two directions, as well as to the stationary response. In none of these neurons was there a strong asymmetry in the firing rate between the two directions of motion, which is a traditional criterion for directional selectivity. The neuron in Fig. 1(E) (bottom) was also tested with anticorrelated noise. The static ITD curve to such noise shows an out-of-phase pattern to that of correlated noise. Again, the response to motion was similar to the response to static ITDs except for a scale factor. The neural data thus show that auditory motion can cause a change in gain of the response, but the most striking finding is the preservation of sensitivity to instantaneous ITDs even at high speeds (1000 μs/s). This seems at odds with behavioral binaural sluggishness.

In a natural environment, the fastest auditory motion is caused by head rotation. Note that movements of the external ear in animals with mobile pinnae are not relevant because they have little effect on the size of low-frequency ITDs (Roth et al., 1980); moreover, they would tend to counteract the effect of head motion through the vestibulo-auricular reflex (Tollin et al., 2009). Head velocity during natural behavior is mostly below 100°/s for cats, and is lower for humans (Einhäuser et al., 2009). Acoustic measurements show that ITD increases monotonically with azimuth and is largest for positions close to the interaural axis, where it reaches values of ∼400 μs in a cat (Roth et al., 1980) and ∼700 μs in a human (Blauert, 1983). The largest change in ITD occurs for sounds crossing the subject's mid-sagittal plane, and is ∼10 μs/°. Thus, the highest speed used in Fig. 1, 1000 μs/s, likely covers the majority of speeds of the ITD cue that cats—and humans, who have slower head dynamics—would commonly experience. However, recent measurements of human head movement during active motion find peak rotations of several hundred °/s (Carriot et al., 2014). Large head velocities, though only rarely exceeding 500°/s, have also been measured in cats trained to make head saccades to auditory targets (Tollin et al., 2009). A head rotation of 500°/s would generate a maximum ITD change of approximately 5000 μs/s. We therefore further explored the upper speed limit in neural coding of instantaneous ITD.

Examples of responses of a neuron to higher speeds are illustrated in Fig. 2(A). The ITD-sensitivity obtained to static noise and to noise moving at 500 and even at 8000 μs/s, far outside the natural range of head rotations, is virtually identical. An upper speed limit in neural coding of instantaneous ITD was not found. Even at the highest speed tested (128 000 μs/s, corresponding to a linear speed “through the head” of ∼20 m/s or ∼75 km/h, about 100 times faster than the fastest changes in ITDs that a cat naturally experiences), ITD-tuning is preserved. Nevertheless, some aspects of the response are not invariant with speed. Besides an increase in instantaneous firing rate with speed, the responses in Fig. 2(A) also show an increasing onset response and horizontal shifts in the positioning of the ITD-curve. However, such changes are as expected from known properties of firing rate adaptation and neural latency.

These changes were quantified with several measures and are summarized in Figs. 2(B)–2(G). Because the same start- and end-ITDs were used at all speeds, the duration of the stimulus halved with every doubling in speed. Responses of IC neurons tend to adapt over the stimulus duration (Ingham and McAlpine, 2004), so a shortening of stimulus duration is expected to cause a rate increase. Across all datasets, the response rate indeed tended to increase with speed (Spearmans ρ = 0.16, p < 0.001) and to be higher than to stationary ITDs [Figs. 1(B) and 1(C)] in most neurons. The presence of motion did not lead to a degradation of ITD-sensitivity in the sense of blurring or “smearing” of peaks and troughs of the ITD curves, which would be a straightforward correlate of binaural sluggishness. Rather, responses were well-modulated by ITDs even at the highest speeds tested [Fig. 2(D)]. In fact, rather than blurring, ITD-tuning became more pronounced with motion: the modulation depth of the ITD-curves increased with increasing speed. Also, the dynamic ITD cue tended to bring out detailed but consistent features in the tuning to ITD. For example, in Fig. 1(E), the location of peaks and troughs was consistent for different speeds, but these features were less marked for the responses to static ITDs. In Fig. 1(F), detailed features (e.g., the narrow peak above 0 ITD) are consistently present to the moving sound but less prominent in the static response.

With increasing speed, responses also showed an apparent shift along the ITD axis and changes in response onset, particularly at very high speeds [Fig. 2(A)]. The shift along the ITD axis is the result of the presence of a neural latency combined with the temporal zoom-in that is inherent in plotting the data on an ITD-abscissa [see Sec. 2 and Figs. 1(A) and 1(B)]. We assessed this with a cross-correlation analysis between the static ITD-curve and the responses to moving ITDs. The height of the cross-correlation function was near 1 at low speeds and decreased with increasing speed [Fig. 2(E)]. This is as expected because with an increase of speed and a shortening of stimulus duration, the transient response dominates and reduces the correlation [Fig. 2(A)]. However, correlation remained near maximal for many neurons even at speeds as high as 10 000 μs/s. The delay in the peak of the cross-correlation function increased with increasing speed, with a slope as expected for the neural latency of the response [Fig. 3(F)] (Joris et al., 2006). Finally, there was no directional selectivity in the firing rate. Across the population and over the range of speeds tested, firing rates were similar for the two directions of motion [Figs. 1(C), 1(D), and 1(F)] and were near the diagonal of equality [Fig. 2(G)].

Fig. 3.

(Color online) Lack of effects of motion on coincidence detectors in ITD-circuit. Motion of the stimulus causes differential compression and expansion of the waveforms to the two ears (A), and concomitant changes in the temporal distribution of evoked spike trains (B). Binaural neurons track the pattern of coincidences, and therefore the instantaneous ITD, at any speed (C). (A) Stimulus waveforms for the two ears are depicted at low (left column) and high (right column) speed. They start and end at the same virtual position (start ITD and end ITD), which, for clarity, is greatly exaggerated relative to the duration of the stimulus (the proportions shown for the “low speed” roughly equal to those at the highest speeds used in this study). (B) Schematic responses. At a monaural level (e.g., the left and right auditory nerve), spikes are temporally locked to the same features of the acoustic waveform in the two ears. Motion causes a compression of the pattern at the ear toward which the sound is moving, relative to that at the other ear. (C) All instantaneous ITDs occurring at low speed also occur at high speed. The number of spikes coding the temporal pattern of the stimulus is sparser at high speed, but there is nothing faster in the operation of binaural coincidence detectors to a high-speed stimulus compared to a low-speed stimulus. The three circles represent three coincidence detectors that receive the same monaural spike trains but with different delays (schematically indicated by the different length of the “axons” converging on them) so that they respond to different ITDs [blue traces in (C)]. Note that these responses are the same at high and low speeds.

In summary, instantaneous ITD is exquisitely coded at the level of the IC despite the integrative properties of this nucleus reported earlier (Spitzer and Semple, 1991, 1993). Dynamic changes in ITD do not lead to blurred tuning to this cue. To the contrary, motion enhances coding of instantaneous ITD, even at unnaturally high speeds.

4. Discussion

In spatially mapped senses, changes in the spatial relationship between receptor organ and the outside world result in motion of the stimulus pattern striking the receptor surface. An object moving relative to the eye causes retinal motion. Likewise, a touching object which changes position relative to the body causes motion on the skin. In both of these major sensory systems, motion detection is an important and well-characterized neural property at early levels of neural processing. But a moving sound causes no “cochlear motion” because there is no direct spatial coding at this sensory organ. In humans, the dominant cue for spatial hearing involves the comparison of instantaneous pressure fluctuations at the two ears in the form of ITDs. We simulated simple relative motion by monotonic changes in ITD and find that neurons at the level of the midbrain encode instantaneous ITD up to very high speeds. This is not to say that responses are invariant with motion, but there is no indication of integrative properties toward motion detection. Direction tuning, which is generally seen as a hallmark of motion-sensitivity, is absent. These findings contrast with earlier work that implies such integrative properties (Spitzer and Semple, 1991, 1993, 1998; Yin and Kuwada, 1983; Zuk and Delgutte, 2017).

The remarkable tolerance for speed is explained by the physics of the stimulus and the process of coincidence detection that enables the comparison of temporal spike patterns. Motion causes temporal compression of the waveform in one ear relative to the other [Fig. 3(A)]. At all speeds explored here, the stimulus is a plain broadband noise at either ear, which sets up temporal spike patterns in auditory nerve fibers. These patterns are dictated by the stimulus waveform and cochlear filtering (Joris, 2003). Because they are locked to the stimulus waveform, they are compressed as well [Fig. 3(B)]. The binaural neuron tracks coincidence in the spike patterns, and therefore the instantaneous ITD, at any speed [Fig. 3(C)]. Motion causes a redistribution of spikes, but there is nothing “faster” in the requirements for coincidence detection at high speeds than at low speeds. A crude analogy is the task of detecting whether two travelers, from two separate queues lined up at an airport gate, pass the gate simultaneously. If simultaneity can be detected for a single occurrence of a pair, it can be detected for streams of such pairs, no matter whether these streams are fast, slow, correlated, uncorrelated, etc.

Ultimately, responses to motion are limited by the finite bandwidth of cochlear filters. Temporal compression of the waveform induces a frequency (Doppler) shift: if that shift exceeds the bandwidth of cochlear bandpass filters, the stimuli between the two ears become uncorrelated. However, for the bandwidths of relevance here, this would require speeds amounting to a significant fraction of the speed of sound c. Measurements of binaural bandwidth (Mc Laughlin et al., 2007) yield Q-factors (center frequency/bandwidth) between 1 and 5. The speed at which the Doppler shift would exceed these bandwidths is c/Q, i.e., about 1/5 of the speed of sound (∼250 km/h) for the neurons with the sharpest frequency tuning, which vastly exceeds the range of speeds that is ecologically relevant.

The absence of traditional correlates of neural motion analysis (directional selectivity, speed tuning) in our data contrasts with previous reports of such correlates in the IC (Kuwada et al., 1979; Spitzer and Semple, 1991, 1993, 1998; Yin and Kuwada, 1983). A major difference with these studies is that the present data were collected in response to noise rather than tones. Yin and Kuwada (1983) observed directional selectivity in a small fraction of neurons (14%) in responses to tonal binaural beats in anesthetized cat. Little or no directional preference was seen to such stimuli at the midbrain and cortical level in awake rabbit (Fitzpatrick et al., 2009). The “preprocessing” to motion described in the IC by Spitzer and Semple (1991, 1993, 1998) seems to be tied to their specific stimulus paradigm, in which ITD-ramps in two directions were juxtaposed (Ingham et al., 2001; McAlpine et al., 2000).

Several recent studies of the processing of dynamic binaural cues used broadband rather than tonal stimuli. Zuk and Delgutte (2017) found more limited coding of a dynamic binaural ITD than of amplitude modulations and suggest this as a neural correlate of behavioral binaural sluggishness. On the other hand, fast processing of changes in binaural cues has been observed physiologically and even psychophysically (Joris et al., 2006; Siveke et al., 2008). In line with these latter data, the present results suggest that at the midbrain level, a premium is placed on the coding of instantaneous values of binaural cues, with little temporal integration.

We surmise that the sensitivity to very fast changes in ITDs, demonstrated here, is needed for adequate interpretation of the spatially complex sound patterns found in natural environments. In realistic environments, even a single sound source bombards the binaural system with a complex pattern of ITDs, due to reflections from the ground plane, walls, other objects, body, etc., and this complexity obviously increases when adding multiple sources. The detection of brief “glimpses” of sound during which binaural cues are dominated by one source, is likely an important source in the process of sound segregation (Dietz et al., 2011; Faller and Merimaa, 2004; Nelson and Takahashi, 2010; Schimmel et al., 2008; Yost and Brown, 2013). The processing of short glimpses benefits from ITD-processing that can track very fast changes, as found in this study. Binaural sluggishness is perhaps tied to the readout or binding of such glimpses, which requires integration across time and neurons.