Interaural Timing Cues Do Not Contribute to the Map of Space in the Ferret Superior Colliculus: A Virtual Acoustic Space Study

Transcription

1 J Neurophysiol 95: , First published September 14, 2005; doi: /jn Interaural Timing Cues Do Not Contribute to the Map of Space in the Ferret Superior Colliculus: A Virtual Acoustic Space Study Robert A. A. Campbell, Timothy P. Doubell, Fernando R. Nodal, Jan W. H. Schnupp, and Andrew J. King University Laboratory of Physiology, University of Oxford, Oxford United Kingdom Submitted 16 June 2005; accepted in final form 12 September 2005 Campbell, Robert A. A., Timothy P. Doubell, Fernando R. Nodal, Jan W. H. Schnupp, and Andrew J. King. Interaural timing cues do not contribute to the map of space in the ferret superior colliculus: a virtual acoustic space study. J Neurophysiol 95: , First published September 14, 2005; doi: /jn In this study, we used individualized virtual acoustic space (VAS) stimuli to investigate the representation of auditory space in the superior colliculus (SC) of anesthetized ferrets. The VAS stimuli were generated by convolving broadband noise bursts with each animal s own headrelated transfer function and presented over earphones. Comparison of the amplitude spectra of the free-field and VAS signals and of the spatial receptive fields of neurons recorded in the inferior colliculus with each form of stimulation confirmed that the VAS provided an accurate simulation of sounds presented in the free field. Units recorded in the deeper layers of the SC responded predominantly to virtual sound directions within the contralateral hemifield. In most cases, increasing the sound level resulted in stronger spike discharges and broader spatial receptive fields. However, the preferred sound directions, as defined by the direction of the centroid vector, remained largely unchanged across different levels and, as observed in previous free-field studies, varied topographically in azimuth along the rostrocaudal axis of the SC. We also examined the contribution of interaural time differences (ITDs) to map topography by digitally manipulating the VAS stimuli so that ITDs were held constant while allowing other spatial cues to vary naturally. The response properties of the majority of units, including centroid direction, remained unchanged with fixed ITDs, indicating that sensitivity to this cue is not responsible for tuning to different sound directions. These results are consistent with previous data suggesting that sensitivity to interaural level differences and spectral cues provides the basis for the map of auditory space in the mammalian SC. INTRODUCTION The superior colliculus (SC) is a midbrain nucleus involved in the control of reflexive orienting movements (King 2004). The superficial layers of the SC receive visual inputs, whereas the deeper layers also receive auditory and tactile inputs, which often converge on individual neurons to generate multisensory response properties (Stein et al. 2004). Visual, auditory, and tactile inputs to the SC are arranged to form topographically aligned maps of space, the registration of which tends to be maintained even when the eyes move (Hartline et al. 1995; Jay and Sparks 1984; Peck et al. 1995; Populin et al. 2004). As a consequence, multisensory signals are integrated at the single neuron level to facilitate the control of motor commands that give rise to orienting movements of the eyes, head, and body. Whereas visual and somatosensory maps are formed by topographic projections from the retina and body surface, Address for reprint requests and other correspondence: A. J. King, University Laboratory of Physiology, Parks Road, Oxford OX1 3PT, UK ( andrew.king@physiol.ox.ac.uk). respectively, a topographic representation of auditory space has to be computed using acoustic localization cues generated by the head and outer ears. These cues comprise interaural time and level differences (ITDs and ILDs), together with the direction-dependent spectral filtering of sounds by the head and external ears (King et al. 2001; Wightman and Kistler 1993). Previous studies have shown that mammalian SC neurons are sensitive to a combination of ILDs (Hirsch et al. 1985; Middlebrooks 1987; Middlebrooks and Knudsen 1987; Palmer and King 1985; Wise and Irvine 1983, 1985) and spectral cues (Carlile and King 1994; King et al. 1994; Palmer and King 1985). Sensitivity to ITDs, the dominant cue for auditory localization of low-frequency sounds by humans (Wightman and Kistler 1992), has been demonstrated in the cat SC by closed-field stimulation but only to values outside the physiological range (Hirsch et al. 1985; Yin et al. 1985). The contribution of this binaural cue to the formation of the auditory space map in mammals therefore remains uncertain. The spatial receptive fields (SRFs) of auditory neurons are typically measured by presenting sounds in the free field. More recently, SRFs in the auditory nerve (Poon and Brugge 1993), lateral superior olive (Tollin and Yin 2002a,b), inferior colliculus (IC) (Behrend et al. 2004; Delgutte et al. 1999; Euston and Takahashi 2002; Sterbing et al. 2003), auditory cortex (Brugge et al. 1994, 1996; Mrsic-Flogel et al. 2001, 2005; Nelken et al. 1998; Schnupp et al. 2001), and SC (Sterbing et al. 2002) have been mapped with virtual acoustic space (VAS) stimuli. This approach involves delivering via earphones sounds that are digitally manipulated to simulate the filtering effects of the head and outer ears (King et al. 2001; Wightman and Kistler 1989). Because sounds can be rapidly presented from randomized directions without having to employ large speaker arrays or physically move a speaker to different locations, VAS stimulation enables SRFs to be measured at high spatial resolution. Furthermore, sound localization cues can be independently manipulated in ways that are impossible with free-field stimulation. In this study, we used individualized VAS stimuli to measure the SRFs of auditory units in the ferret midbrain. The stimuli were first validated by comparing both acoustical measurements and neuronal responses obtained with free field and VAS stimulation. We then used VAS stimuli to obtain more detailed information about the auditory spatial response properties of SC neurons than provided by previous free-field studies as well as to explore their dependence on ITDs. The costs of publication of this article were defrayed in part by the payment of page charges. The article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact /06 $8.00 Copyright 2006 The American Physiological Society

2 AUDITORY RESPONSES OF SUPERIOR COLLICULUS NEURONS 243 METHODS General Seven adult (aged 4 mo) pigmented ferrets (Mustela putorius) with normal hearing (assessed by otoscopic examination and tympanometry) were used in this study. All surgical procedures were approved by local ethical review committee and licensed by the UK Home Office. The animals were anesthetized by an intramuscular injection of alphaxalone/alphadolone acetate (Saffan, 2 mg/kg, Mallinckrodt Veterinary, Uxbridge, UK). During surgery, bupivacaine hydrochloride (Marcain Polyamp, Astrazeneca UK, Luton UK) was applied topically, and supplementary doses of Saffan were given as required via a cannula implanted in the radial vein. Body temperature was monitored by a rectal probe and maintained with a feedback electric blanket at 38 C. The first stage of each experiment involved measuring the animal s head-related transfer function (HRTF) so that individualized VAS stimuli could be constructed, followed by electrophysiological recordings from either the IC or the SC. Preparation for acoustical recordings A damped polythene probe tube ( 30 mm long, 0.86 mm ID, 1.52 mm OD) was passed through each ear canal wall in such a way as to emerge caudally behind the pinna. The tubes were secured internally with a small flange, which abutted against the canal wall, and externally with a rubber O ring that was pressed against the skin. The animal was then placed in a stereotaxic frame and fitted with blunt ear bars, and the skull was exposed. A steel bar (7 mm diam) was attached to the skull with steel screws and dental cement (Simplex Rapid, Austenal Dental, Harrow, UK), so that the head could be supported from behind. At this stage in the SC experiments (n 6), the stereotaxic frame was removed, the incisions in the scalp closed, and the external ears carefully repositioned according to measurements made prior to surgery. Condenser microphones (miniature KE microphone capsules, Sennheiser, High Wycombe, UK) were attached to the probe tubes, and the animal was transferred to an anechoic chamber for free-field acoustic recordings, which were carried out prior to preparation for electrophysiological recording. To allow a more direct comparison between free-field and VAS SRFs, the acoustical measurements for the IC experiment were conducted immediately prior to recording with the craniotomy performed and all recording equipment in place. Acoustical recording A loudspeaker (KefT27, KEF Audio, Maidstone, UK) mounted on a computer-controlled motorized hoop (radius: 65 cm) was used to present broadband signals (512-point Golay codes) (Zhou et al. 1992) from 63 different directions at 16 intervals in azimuth from 160 to 160 and at six vertical angles from 80 to 60 elevation. The sampled positions were arranged so that their diagonal separation was 34. The generation of the Golay codes and the recording of the microphone signals were performed digitally using TDT system 2 A/D and D/A converters (sample rate of 80 khz, Tucker-Davis Technologies, Alachua, FL) and 30-kHz anti-alias filters. The microphone signals were analyzed for each stimulus direction to calculate a spectral transfer function containing both the animal s HRTF and the transfer characteristics of the loudspeaker and probe microphones. The ITDs were extracted from the microphone signals by crosscorrelation of the impulse responses after low-pass filtering (0 4 khz). An in situ calibration to remove the transfer functions of the probe microphones and in-ear headphones used for presenting the VAS stimuli was then carried out. Minimum phase filters were calculated from the equalized amplitude spectra using the Hilbert transform. VAS stimuli consisted of short (100 ms) Gaussian noise burst with 5 ms raised cosine onset and offset ramps, which were convolved with the appropriate minimum phase filters for each direction, and delayed to generate the appropriate ITD. Although frequency-dependent ITDs are excluded by this approach, psychophysical studies in humans have shown that the minimum-phase-plusdelay method adequately approximates the HRTF phase spectrum as long as the low-frequency ITD is appropriate (Kulkarni et al. 1999). The VAS stimuli were not frozen because new Gaussian noise bursts were used for each stimulus presentation. Electrophysiological recording At the conclusion of the acoustical measurements, anesthesia was switched to an intravenous infusion (Perfusor Secura FT infusor, B. Braun Melsungen, Germany) of ketamine/medetomidine (Ketaset, 5 mg kg 1 h 1, Fort Dodge Animal Health, Southampton, UK; Domitor, 10 g kg 1 h 1 ) in Hartmann s solution. A tracheal cannula was implanted, and the animal was ventilated (7025 respirator, Ugo Basile, Milano, Italy) with oxygen-enriched air. Atropine sulfate (0.06 mg kg 1 h 1, Animal Care, York, UK) and dexamethasone (0.5 mg kg 1 h 1, Dexadreson, Intervet UK, Milton Keynes, UK) were administered intramuscularly to reduce mucus secretions in the airways and minimize cerebral edema, respectively. In six ferrets, a craniotomy was performed over the occipitoparietal cortex above the right SC. In the remaining animal, the craniotomy was positioned more caudally, so that recordings could be made from the IC. The dura was removed and the exposed cortex was protected by 2% agar in saline. The agar was supported by a rim of dental acrylic. The left eye was dilated with atropine and fitted with a zero-refractive power contact lens. To eliminate eye movements, pancuronium bromide muscle relaxant was added to the infusate (0.2 mg kg 1 h 1, Pavulon, N.V. Organon, the Netherlands). We monitored the depth of anesthesia and the physiological condition of the animal by continuous measurement of the electrocardio- and electroencephalograph (ECG and EEG; using either custom-built amplifiers or a Datex-Ohmeda anesthesia monitor, Hatfield, UK), end tidal CO 2 (using either a 47210A capnometer, Hewlett Packard GmbH, Boeblingen, Germany or Datex-Ohmeda monitor), and arterial oxyhemoglobin saturation (Datex-Ohmeda monitor). Single-unit activity was recorded extracellularly using a tungsten-inglass electrode ( M ) lowered vertically through the cortex and into the midbrain. The electrode signals were band-pass filtered (500 Hz to 5 khz), amplified ( 15,000 times) and digitized at 25 khz. Electrolytic lesions ( 5 A for 5 s) were made in most electrode penetrations in which acoustically responsive units were isolated, to allow for histological confirmation of recording sites. Visual and auditory stimuli The SC was located by presenting a diffuse flashing light positioned a few centimetres from the contralateral eye. The superficial visual layers of the SC were usually encountered at a depth of 6 mm below the cortical surface. We determined the direction of the maximum visual response of multiunit activity recorded in these layers using an LED mounted on the motorized hoop. The electrode was then advanced into the deeper layers of the SC to search for auditory responses by presenting contralateral broadband noise bursts. These stimuli were delivered via commercial Panasonic earphone drivers (RP-HV297, Bracknell, UK), coupled to an otoscope speculum that was inserted into each ear canal. When an auditory response was identified, the threshold of the unit was determined by presenting unfiltered noise bursts to the contralateral ear (100-ms duration with an inter-stimulus interval of 1,000 ms) at a range of sound levels. Threshold was taken as the lowest sound level to elicit an increase in firing rate that was significantly greater (P 0.05) than the unit s resting level. The spatial receptive fields (SRFs) of most units were measured at two sound levels, one near threshold (typically 5 15 db above unit threshold) and a second at a level well above threshold (typically db above unit threshold). SRFs were

3 244 R.A.A. CAMPBELL, T. P. DOUBELL, F. R. NODAL, J.W.H. SCHNUPP, AND A. J. KING measured by presenting VAS stimuli in a random order from the same sound directions used for measuring the HRTF. This process was repeated until responses for each virtual stimulus direction had been collected. The mean evoked spike rate at each stimulus position was then used to estimate the unit s SRF (see following text). Two stimulus conditions were used to investigate the contribution of ITDs to the spatial tuning of SC units (Fig. 1). The natural cue condition involved presenting sounds from each position with all sound localization cues co-varying naturally, as they would with a real free-field sound source. In the fixed cue condition, the ITDs were maintained at a fixed value and did not co-vary with the ILDs and spectral cues. This was achieved by delaying the sound reaching the left ear by 200 s, close to the maximum value produced by the separation of the ears, for all 63 virtual sound locations. The sound therefore arrived at the right ear first (ipsilateral to the SC from which the recordings were made) and had an interaural delay corresponding to a source at 90 to the animal s right. This ITD was chosen because it represented a value to which no units in the right SC would be expected to be tuned. Consequently, the ITDs were fixed at a value inappropriate for the normal location of an SRF. For the IC recordings, we searched for units and determined threshold in the same manner as in the SC. Once a unit s threshold was determined, we recorded a VAS SRF then carefully removed the earphone drivers and recorded a free-field SRF by sequential presentation of noise bursts from corresponding stimulus locations. Where possible, we recorded the VAS SRF and then the free-field SRF again FIG. 1. Schematic illustrating the natural and fixed interaural timedifference (ITD) conditions. The contribution of ITDs to the spatial response properties of superior colliculus (SC) units was investigated by presenting virtual acoustic space (VAS) stimuli in which ITDs were fixed at a single value for all sound-source directions. The natural ITD condition replicates free field stimuli: all localization cues co-vary with sound-source direction. Sounds presented at the anterior midline will reach both ears simultaneously and be equally intense in both ears. A sound on either side of the midline will reach the near ear before the far ear and be more intense at the near ear. In the fixed condition, the ITDs had a constant value of 200 s for all virtual soundsource directions: stimuli always arrived at the right (ipsilateral) ear before the left ear, whereas level and spectral cues co-varied naturally with stimulus direction. For clarity, the noise-bursts and axes are not drawn to scale. to control for response stability. IC SRF recordings were made at sound levels of 5 20 db above unit threshold. We ensured that the amplitude of the VAS and free-field stimuli were equivalent at the entrance of the ear canal by calibrating our stimuli using the implanted probe microphones prior to each recording. Owing to the time taken to move the speaker, we sampled only three elevations ( 24, 6, and 12 ), at either 10 or 20 different azimuths (i.e., 30 or 60 sound-source directions). Stimuli were presented times at each location. Analysis of results Stimulus generation and data acquisition were controlled using BrainWare (Tucker-Davis Technologies). This software stored the latency and shapes of all spikes that crossed an arbitrary amplitude threshold determined by the user. These data were saved for off-line analysis. Whenever possible we digitally isolated single units by sorting evoked spikes based on their shape. The response period for each unit was individually determined from the poststimulus time histogram (PSTH). In all cases, firing rates had returned to spontaneous background levels by 400 ms after stimulus onset. Response magnitude was measured relative to the spontaneous activity of the neuron, which was obtained from a second window drawn between 500 and 1,000 ms after stimulus onset. The raw data were exported to Matlab R14 (Mathworks, Natick, MA) with which all further analysis was carried out. SRFs were visualized by producing a smoothed map projection showing the average response rate for each sound direction. Smoothing was done by interpolation of the averaged responses over a uniform grid of 7.5 resolution using biharmonic spline interpolation. To avoid discontinuities due to extrapolation over positions above and behind the animal (along the dateline and at the north pole of our spherical coordinate system), we extended the matrix maps to cover a -200 to 200 azimuth range by copying values across from the opposite edge, i.e., from 160 to 200 and from 160 to 200. This ensured that the algorithm could interpolate smoothly and without discontinuities across the full 180 azimuthal range. For each SRF, we calculated the 50 and 75% response areas (rad 2 ) corresponding to the total angular extent of the regions within which the response exceeded a stated percentage of the unit s maximal response (estimated from the mean response at the 5 most effective virtual stimulus positions). The 50 and 75% response areas provide a measure of the overall responsiveness of a cell across different spatial locations but do not indicate whether the SRF is focused around a single preferred stimulus direction. For example, a multi-peaked SRF could still have a small total 75% area. We therefore derived the centroid (or center of mass) for each SRF. The centroid provides a way of quantifying the preferred sound direction of each unit as well as determining the sharpness of this tuning. The centroid was calculated by modeling the response field as a sphere of unit radius, where the mass density in each direction was given by the observed response strength in the corresponding direction (see Mrsic-Flogel et al for the full derivation of the centroid). The direction of the centroid vector was calculated as a volume integral by approximating the model sphere as a sum of pyramids, the bases of which are at the surface and the apices of which are at the center of the sphere. The direction of the centroid vector summarizes the overall directional preference of the SRF, whereas its length gives an indication of how sharply tuned the SRF is in that direction. The theoretical maximum length of the centroid (for an SRF responding to a single sound direction) is 0.75, the distance of the center of a pyramid of unit height from its apex. The interpolated map, centroid direction vector and visual best direction recorded in the superficial layers of the same electrode penetration were then displayed using a Kavraisky 5 equal-area projection (see Fig. 7).

4 AUDITORY RESPONSES OF SUPERIOR COLLICULUS NEURONS 245 To quantify the effects of changing sound level or fixing ITD, we compared SRF features such as response area and centroid parameters (length and direction) between the two conditions, e.g., fixed and natural ITDs. Paired t-tests were used to look for a systematic effect of the stimulus property in question over the whole population of units. We also used a Monte Carlo test to analyze the results on a unit by unit basis to see if the responses of a subset of cells were influenced by sound level or ITD condition. This was done by pooling all repetitions from both conditions at each sound-source direction. Pairs of simulated SRFs were generated by random resampling of the pooled responses (with replacement) for each sound-source direction until two SRFs, based on the same number of stimulus repetitions as those used for collecting the data, were obtained. The centroid statistics were then calculated and the difference in simulated values was determined. This process was repeated 10,000 times, to estimate the distribution of differences in centroid statistics that are to be expected by chance. If an observed difference fell into the highest or lowest 2.5% of this distribution, then we considered this to be significant at the 5% level. Units with low spike rates or unreliable responses often have centroids with unreliable directions. For example, a burst of spikes at a single stimulus presentation could spuriously displace the centroid direction toward that sound-source location. To eliminate units with misleading centroids we used a Monte Carlo approach to resample (with replacement) each SRF 10,000 times and derive the centroid statistics for each. SRFs were excluded if the SE of the simulated centroid directions exceeded 3. RESULTS Validation of virtual acoustic space stimuli To show that our VAS stimuli accurately replicate the animal s own HRTF, we have to demonstrate that the HRTFfiltered noise bursts faithfully recreate the amplitude spectra of the corresponding free-field stimuli. Furthermore, we must show that comparable neuronal spatial tuning can be recorded with VAS and free-field stimulation. The implanted probe microphones made it possible to record our VAS stimuli at the same location within the ear canal where the HRTF was initially measured. Figure 2 shows the amplitude spectra measured from each ear for VAS and freefield stimuli from three different sound-source directions. The spectra are almost identical in each case. Figure 3 shows the differences between the VAS and free-field amplitude spectra FIG. 2. Comparisons of free-field (black) and VAS (red) amplitude spectra from 3 different sound-source directions as indicated at the top of each column. FIG. 3. Differences between the amplitude spectra of free-field and VAS stimuli measured in the left and right ears for 30 sound-source directions. These directions correspond to the 10 different azimuths and 3 elevations shown in Fig. 4. Azimuth is plotted along the ordinate with the data from the 3 elevations sampled at each azimuth presented in adjacent bins. Most directions exhibited a constant amplitude difference across frequency. However, for each ear, the higher frequency components of sounds presented on the contralateral side were greatly attenuated by the head causing the recorded amplitude spectrum to approach the noise floor of our microphones and amplifier. over 30 sampled sound directions (same directions at which the neural responses are shown in Fig. 4). At most locations, these differences were negligible ( 1 db in 80% of cases; 3 dbin 90%) and constant across frequency, indicating that the VAS stimuli faithfully replicate the spectral cues arising from real free-field stimuli. For some directions, however, the amplitude spectra exhibited larger differences at higher frequencies. This is caused by the poor signal-to-noise ratio at higher frequencies for stimuli presented on the side contralateral to the microphone. Because the SC contains topographically aligned maps of visual and auditory space (e.g., King and Hutchings 1987), it is possible to make inferences about the fidelity of our VAS stimuli from a comparison of the virtual SRF centroid directions with the best direction of the overlying visual responses (see following text). In addition, we performed a direct physiological validation by comparing VAS SRFs with those obtained using free-field stimulation. This was done using a population of 11 units recorded from the central nucleus of the IC in one animal. We chose the IC rather than the SC because the more robust responses obtained in this nucleus were suited for making quantitative comparisons of data recorded over the longer periods of time required to map the SRFs using both forms of stimulation. Figures 4 and 5 show examples of three of these units where SRFs were recorded at corresponding sound levels using both free-field and VAS stimuli. The responses of the unit shown in Fig. 4, where two sets of VAS and free-field recordings were conducted around an hour apart from the same unit, clearly illustrate the stability of the SRFs. Indeed, the coefficients derived from cross-correlation of the spike rates in the response window for each mode of stimulation were almost identical for the two sets of recordings. The SRFs for the two units shown in Fig. 5 were quite different from one another, with the response illustrated in Fig. 5B being more sharply tuned than that in Fig. 5A. Once again, however, the responses obtained using free-field and VAS stimuli were extremely similar, with correlation coefficients close to one in each case.

5 246 R.A.A. CAMPBELL, T. P. DOUBELL, F. R. NODAL, J.W.H. SCHNUPP, AND A. J. KING FIG. 4. Raster plots comparing free-field (blue) and VAS (red) responses from an inferior colliculus unit (unit 3 in Fig. 6) measured at a range of sound-source directions. The recordings shown are the 1st (A) and last (B) pairs from this unit and were made about an hour apart. The similarity between plots A and B illustrates recording stability throughout this period. Each plot shows the 1st 300 ms after stimulus onset. Space was sampled in a triangular array, so stimuli at 6 elevation were presented at azimuth positions 18 to the left of (anti-clockwise from) the labeled positions on the lowest row ( 12 elevation). Each sound direction was sampled 10 times in each condition. The correlation coefficients, r, between the VAS and free-field recordings are 0.82 for A and 0.91 for B. Recordings were conducted at 15 db above unit threshold. The SRFs measured from all 11 IC units with both VAS and free-field stimulation are illustrated in Fig. 6 by plotting the response as a function of sound azimuth at a single elevation. In some cases (units 4, 6, 7, 9, and 10), SRFs were obtained in VAS and in the free-field only once, whereas in the others (units 1, 2, 3, 5, 8, and 11), at least one of these measurements was repeated to help control for changes in the response properties over time. For each unit, the VAS SRF was recorded first followed by a free-field recording. Further recordings on the same unit were conducted when possible. As in Keller et al. (1998), we compared the SRFs by calculating correlation coefficients between the free-field and VAS data. The r values in the top right of each panel indicate the correlation coefficients of VAS and free-field recordings (based on all the sound directions sampled, including the other elevations) conducted at adjacent points in time. For example, unit 8 with two VAS and two free-field recordings has three r values listed; from top to bottom: VAS1/free-field1, VAS2/free-field1, and VAS2/ free-field2. Although all responses were predominantly contralateral, the widths and locations of these azimuth response profiles differed across units but were similar within individual units. For example, unit 3 responded reproducibly and fairly selectively to both real and virtual sound directions in front of the animal, whereas unit 6 was more broadly tuned and responded best to stimuli located well into the contralateral hemifield. To assess the significance of the measured r values, we conducted a resampling test to estimate the likelihood of the observed value occurring. Correlation coefficients were calculated between the observed free-field data and 10,000 simulated VAS SRFs created by re-sampling the evoked spike counts from all VAS recordings. Cases where 5% of the simulated r values were greater than the observed value are indicated by * next to the correlation coefficients given in Fig. 6. The likelihood of

6 AUDITORY RESPONSES OF SUPERIOR COLLICULUS NEURONS 247 FIG. 5. Raster plots comparing free-field (blue) and VAS (red) responses from the inferior colliculus measured at a range of sound-source directions. A: unit 9 (r 0.96; recorded at 15 db above threshold). B: 1st pair of recordings from unit 11 (r 0.96; recorded at 10 db above threshold; see Fig. 6). Each plot shows the 1st 300 ms after stimulus onset. these r values occurring was considered to be no greater than chance. Units 3 and 11 showed excellent correspondence between VAS and free-field and were stable on re-recording. The r values of these recordings ranged from 0.81 to Those units in which the SRF was determined only once with each form of stimulation exhibited no evidence of response drift and had high r values in the range of those found in units 3 and 11. The response of unit 8 was less stable between the two pairs of recordings, responding more strongly during the second VAS and free-field recordings. Despite this, the correlation between the VAS and free-field responses within each recording pair was high (r 0.85). Only unit 5 showed systematic differences between VAS and free-field across recordings. Of the four r values (ranging from 0.76 to 0.79) for this unit, two were not significantly greater than chance, confirming that, in this instance, the correlation between the free-field and VAS recordings was poor. The range of r values for all 11 IC units was 0.63 to 0.96, with a mean of These values are comparable to those reported by Keller et al. (1998). Auditory receptive fields of SC neurons measured with normal localization cues A total of 48 acoustically responsive units were isolated from the SC of six ferrets. Three of these were excluded from further analysis according to the criteria described in METHODS. A total of 88 recordings were made, 79 of which passed our exclusion criteria. The SRFs of most units comprised a single peak, usually in the contralateral hemifield (Fig. 7). There was considerable variation in the size of the SRFs and in the magnitude of the maximum response obtained at corresponding sound levels relative to unit threshold. In keeping with previous free-field measurements of spatial tuning in the ferret SC (King and Hutchings 1987), the narrower axis of the 75% area typically covered 30. Effect of altering sound level Of the 45 remaining SC units, we recorded 13 at one sound level only and 32 at two sound levels. The difference in sound level over all units was 18 7 (SD) db. For two neurons, we

7 248 R.A.A. CAMPBELL, T. P. DOUBELL, F. R. NODAL, J.W.H. SCHNUPP, AND A. J. KING FIG. 6. Comparison of VAS and free-field responses from all 11 IC units. For clarity, each line shows an azimuth sweep from only 1 of the 3 recorded elevations ( 6 ). Multiple recordings of the same condition are shown using different colors and symbols (e.g., the 2nd VAS recording is shown as a red line with open circles). VAS and free-field recordings were obtained in an interleaved fashion to help control for changes in the response over time. Numbers inthe top right of each subplot indicate the r values for each pair of recordings. The r values were calculated using the mean evoked spike counts from the full set of sound-source directions (as shown in Figs. 4 and 5). Cases in which 5% of simulated VAS SRFs had r values greater than the observed value are indicated (*). See main text for further details. re-recorded one of the sound levels to test for unit stability. Our level analyses are therefore conducted on 34 pairs of SRFs. Increasing sound level resulted in a significant (paired t-test, t 3.271, df 33, P 0.003) increase in the maximum response of the units (Fig. 8A) as well as a significant (paired t-test, t 3.591, df 33, P 0.002) increase in the size of the 75% area (Fig. 8B). Increasing the sound level also led to a small decrease in centroid vector length, although this decrease was not significant over the population as a whole (paired t-test, t 1.138, df 33, P 0.263; Fig. 8C). A shorter centroid vector would have indicated a less sharply tuned response. Although there was no systematic effect of sound level on the centroid length over the population, Monte-Carlo analysis of individual units showed that 17/34 SRF pairs (16/32 units) had significantly different centroid lengths between the two conditions (P 0.05). Of these 16 units, 12 showed a significant decrease in centroid length as the sound level was increased. Changing sound level did not result in systematic shifts in centroid direction (see Fig. 9) for either azimuth (paired t-test, t 1.779, df 33, P 0.085) or elevation (paired t-test, t 1.138, df 33, P 0.263) for the population as a whole. Indeed, the mean shift was only 6 in azimuth and 3 in elevation. However, like centroid length, a subset of units exhibited a small (Fig. 9C) but significant shift in centroid direction with increasing sound level (13/32). Topography and alignment of visual and auditory maps Previous free-field studies of the ferret SC (e.g., King and Hutchings 1987; King et al. 1998) have shown that the best azimuths of visual units recorded in the superficial layers are arranged to form a map that corresponds closely in its spatial extent, magnification, and orientation with the auditory representation in the deeper layers of the nucleus. We have explored this relationship further for two reasons. First confirming that auditory virtual SRFs are aligned with the visual responses would provide further evidence as to the fidelity of our VAS stimuli. Second, by using the visual map as a template against which to compare the auditory SRFs, we can examine the contribution of ITDs to the topographic organization of the auditory representation. In this study, we focused mainly on how the auditory SRFs vary with the location of the units along the rostrocaudal axis of the SC. This is the axis along which stimulus azimuth is represented. For each acoustically responsive unit, we compared the azimuth of the SRF centroid vector with the visual best azimuth recorded in the same electrode penetration (Fig. 10). The centroid directions of auditory units recorded in the rostral third of the SC spanned the anterior quadrant with the most rostral units being tuned to ipsilateral sound directions. This same region of the SC also represents frontal visual space. As the location of the recording electrode was moved toward

8 AUDITORY RESPONSES OF SUPERIOR COLLICULUS NEURONS 249 naturally were compared with those obtained with a fixed ITD of 200 s for all sound directions (see METHODS). This time delay would correspond to a sound originating from 90 to the right (ipsilateral to the recorded SC). Unless the virtual soundsource originated from this direction, the fixed ITD cues conflicted with the other localization cues. Because most of the units were recorded at more than one sound level, we obtained FIG. 7. A E: spatial receptive fields of 5 auditory units recorded from the right SC. Color scale indicates the mean evoked spike rate per stimulus presentation. Response windows were drawn individually for each unit. Maximum response is indicated by the red region. The solid black contour line encloses the area within which the spike rate is 75% of the unit s maximum response. The black cross shows the direction of the centroid vector, which indicates the unit s preferred sound direction (see METHODS). The white circle represents the visual best position of multiunit activity recorded in the superficial layers of the electrode penetration from which the auditory unit was recorded. Stimulus levels above unit threshold: A, 30 db; B, 25 db; C, 0 db; D, 7.5 db; E, 25 db. the caudal end of the SC, the auditory centroid vectors and visual best azimuths shifted systematically toward more posterior regions of the contralateral hemifield. Because we did not undertake any free-field SC recordings in the present study, we have included in Fig. 10 data from a previous study (King et al. 1998) in which the preferred sound directions were derived from auditory spatial response profiles mapped at a single elevation (gray crosses in this figure). These measures of spatial selectivity are not equivalent, as the centroid vector is based on the relative response of the unit throughout the SRF and can therefore be located away from the sound location evoking the strongest response (see Fig. 7A). Nevertheless, the great majority of auditory centroid vectors fell within or very close to the distribution of best azimuths from the earlier free-field study. Effect of manipulating ITD cues The virtual SRFs of 32 of 45 SC units were recorded using two randomly interleaved ITD conditions. The purpose of this was to investigate the possible contribution of ITDs to the generation of the map of auditory space in the SC. The SRFs recorded when all sound localization cues were allowed to vary FIG. 8. Effects of changing sound level on the SC responses. Circled data points are units which showed a significant change in the measured parameter according to our Monte-Carlo simulation. A: maximum response (mean evoked spike rate per stimulus presentation) of each recorded unit at high and low sound levels. Suprathreshold recordings were performed at an average sound intensity of 34 db above threshold across all units. Near-threshold recordings were performed at an average intensity of 17 db above threshold. B: 75% response area (rad 2 ) of each unit at high and low sound levels. C: effect of sound level on the length of the centroid vector. Higher sound levels led to a significant increase in area and maximum response (P 0.001) but did not affect centroid length over the population as a whole (P 0.245).

9 250 R.A.A. CAMPBELL, T. P. DOUBELL, F. R. NODAL, J.W.H. SCHNUPP, AND A. J. KING both suprathreshold (Fig. 11A) and near-threshold (B) sound levels. We also looked for an effect of ITD on centroid direction, centroid length, 50% SRF area and maximum response strength (Fig. 12). Of these parameters, only the length of the centroid vector (Fig. 12B) showed a significant overall change, decreasing in value in the fixed ITD condition (paired t-test, t 3.583, df 53, P 0.001). To examine whether fixing the ITD altered the response properties of individual units, we performed a Monte Carlo simulation where spike counts from the two ITD conditions were pooled and used to generate simulated SRF pairs from which we calculated differences in various descriptive statistics. This analysis showed that manipulating the ITD resulted in significant changes (P 0.05) in 17/32 units for at least one of the four measured parameters. A smaller subset of units showed a significant effect for at least one measure of spatial tuning (Monte Carlo analysis, indicated by circled data points in Fig. 12). For example, 7/32 units displayed a significant change in SRF area with 2/32 units showing a change in centroid direction when the ITD was fixed. Examples of individual SRFs recorded in the two ITD conditions are shown in Fig. 13. These illustrate cases where holding the ITD constant had either no effect (Fig. 13, A and B) or induced a FIG. 9. Effect of sound level on centroid direction of SC units. A: comparison of the centroid azimuth direction for each unit at high and low sound levels. B: comparison of the centroid elevation direction for each unit at high and low sound levels. C: difference in the centroid at high and low sound intensities. Gray square, the mean shift in centroid direction. As in Fig. 8, circled points are judged to have changed significantly across sound levels, according to our Monte Carlo analysis. There was no systematic shift in centroid direction over the population as a whole. a total of 54 recordings in the fixed ITD condition and 79 in which ITDs were allowed to vary naturally. We compared the topographic alignment of the visual and auditory maps using SRFs generated with natural cues and with fixed ITDs (Fig. 11). Despite the reduction of spatial information available in the fixed ITD condition, the auditory centroid vectors continued to co-vary with the visual best positions at FIG. 10. Comparison of auditory best azimuth (determined by the centroid direction vector) of deep SC units with the visual best azimuth measured in the superficial layers of the same electrode penetration. Near- and suprathreshold recordings from the same neuron are distinguished using blue and red points. Also shown are free-field data from King et al. (1998) in which the best azimuth was defined as the loudspeaker location producing the maximum response within a spatial response profile obtained at a single elevation. The free-field data exclude broadly tuned or multi-peaked units. Units from the present study were only excluded if a permutation test judged the centroid direction to be too variable (see METHODS). Units where the 50% area covered more than half of space have been labeled broad and surrounded with a square. Units with four or more isolated 75% regions are termed multipeaked and surrounded with a green circle.

10 AUDITORY RESPONSES OF SUPERIOR COLLICULUS NEURONS 251 of SC neurons than has previously been carried out with free-field stimulation. Moreover, VAS stimuli can be manipulated to assess the contribution of different acoustic cues to the SRFs. We have shown here by holding ITDs constant while leaving ILDs and spectral cues to vary naturally that sensitivity to timing differences between the ears is not a major determinant of the map of auditory space in the SC of the ferret. Fidelity of the virtual acoustic space stimuli FIG. 11. Effect of fixing the ITD at 200 s on the visual-auditory correlation in the SC. A: data obtained at suprathreshold sound levels. B: data obtained at near-threshold sound levels. Circled data points are those for which the centroid direction vector changed significantly as a function of ITD condition according to a Monte Carlo analysis. Validation of the VAS stimuli was achieved by comparing the amplitude spectra of the free-field and VAS signals measured with probe microphones implanted into the ear canals and also by a comparison of SRFs of the same IC units recorded with each form of sound presentation. Although the central nucleus of the IC does not contain a map of auditory space, the SRFs found there are sufficiently discrete and inhomogeneous (see also Behrend et al. 2004; Sterbing et al. 2003) to justify using the stronger responses recorded in this nucleus to compare the spatial tuning obtained with free-field and VAS stimulation. We did find a discrepancy between some of the very highfrequency components of the amplitude spectra of the free-field and VAS stimuli for sounds presented on the contralateral side, which arose because these signals were attenuated by the head to the noise level of our recording system. However, the bandwidth of these differences is almost certainly too narrow to be preserved after the sound has been filtered by the cochlea. Overall, the correlation between both acoustical and physiological measures was very high and comparable to that reported in other studies (Behrend et al. 2004; Keller et al. 1998; significant change (Fig. 13, C F) in one of the spatial tuning measures. These results suggest that, at least in some cases, sensitivity to ITDs might contribute to the formation of the SRFs of SC neurons. The limited effect of fixing the ITD on centroid direction indicates, however, that the contribution of this cue to the map of sound azimuth must be minor. This is further highlighted in Fig. 14, in which we combined all the SRFs recorded within four different visual azimuth ranges, i.e., from different regions of the SC. In both the natural (Fig. 14A) and fixed ITD (Fig. 14B) conditions, this resulted in a singlepeaked SRF at each location, which varied in azimuth from the rostral to the caudal end of the nucleus. Moreover, despite the variation in centroid vector direction for individual units, particularly in rostral SC (see Fig. 10), these average SRF plots revealed a much tighter correlation with the pooled visual best azimuth. DISCUSSION In this study, we used individualized VAS stimuli, derived from acoustical measurements of each animal s own HRTF, to provide a more detailed characterization of the auditory SRFs FIG. 12. Effect of fixing ITDs on the centroid azimuth direction (A) and length (B) and the SRF response area (C) and maximum response (D). Circled data points exhibited a significant difference between the 2 ITD conditions according to a Monte-Carlo simulation.

11 252 R.A.A. CAMPBELL, T. P. DOUBELL, F. R. NODAL, J.W.H. SCHNUPP, AND A. J. KING FIG. 13. A F: effect of fixing ITDs on the SRFs of 6 different units. Left: responses in the natural cue condition. Right: responses of the same units when the ITDs were fixed to 200 s. A and B: units with no significant change in response properties on fixing ITDs. C and D: units exhibiting a significant decrease and increase, respectively, in centroid length. E and F: units showing significant decreases and increases, respectively, in their 75% areas. In all cases, P Sterbing et al. 2003). We can therefore conclude that our VAS stimuli reliably simulated the real sound sources. A similar conclusion was reached by comparing judgements made by human listeners of the apparent locations of free-field and virtual sound sources that had been synthesized in the same way as in the present study (Wightman and Kistler 1989). Representation of space in the SC The fidelity of the VAS stimuli was further confirmed by the similarity between the virtual SRFs measured here and the auditory spatial tuning reported in previous free-field studies of the SC. With the exception of a few units in the ferret SC where the SRFs were mapped in more detail (King and Hutchings 1987), most free-field experiments have measured spatial response profiles along a single azimuth or elevation or attempted to estimate the location of the borders of the SRF. Nevertheless, we found that the properties of the VAS SRFs closely resemble those reported in free-field studies of the ferret (King and Hutchings 1987) and other species (King and Palmer 1983; Middlebrooks and Knudsen 1984). Increasing the sound level resulted in a systematic increase in maximum firing rate and in the response area of the units, confirming results found in previous free-field studies (King and Hutchings 1987; King and Palmer 1983; Middlebrooks and Knudsen 1984). Around 40% of units showed a small but significant change in the direction of the centroid vector, which resulted from a non-uniform increase in SRF area with sound level. There was, however, no systematic shift in preferred sound direction across the population of neurons (Fig. 9) and the distribution of centroid directions with recording site within the SC was very similar at near-threshold and suprathreshold sound levels (Fig. 10). A similar result was noted with free-field stimulation (Carlile and King 1994; King et al. 1994; Middlebrooks and Knudsen 1984). We did observe more scatter in the auditory map than reported in the free-field studies. In particular, the range of centroid direction vectors of units recorded in rostral SC was larger than that of the best azimuths, as defined by the peak of the spatial response profile, in our earlier free-field studies of the ferret SC (e.g., King and Hutchings 1987; King et al. 1998). Our acoustical measurements and recordings in the IC suggest

12 AUDITORY RESPONSES OF SUPERIOR COLLICULUS NEURONS 253 that this apparent mismatch is unlikely to arise because the VAS stimulation did not adequately replicate the free-field sound source. On the other hand, Behrend et al. (2004) showed that the presence of recording equipment around the animal s head can alter virtual SRFs. Such equipment was obviously in place in the free-field experiments but not when the VAS stimuli were generated for the SC experiments in the present study. Acoustic distortions produced by the recording equipment could therefore contribute to the differences between the azimuth maps observed with VAS and free-field stimulation. Another possibility is that these differences might reflect our use of the centroid as a measure of spatial selectivity, which is based on the whole receptive field rather than the region of maximum response only. This latter possibility is likely, given that, in contrast to the free-field studies, we did not exclude units with very broad or multi-peaked SRFs. Although the distribution of centroid direction vectors for individual units was quite scattered, the average SRFs obtained by pooling data within discrete regions of the SC revealed a clear topographic shift in spatial selectivity with recording site. The broad spatial tuning of the auditory units is consistent with the coarsely tuned movement fields of deep SC neurons (McIlwain 1991; Lee et al. 1988; Sparks et al. 1976). Thus the location of the stimulus and the vector of the orienting movements elicited by it both appear to be specified by the spatial distribution of activity across a population of SC neurons. Estimates of neuronal discrimination values from the responses of space-mapped neurons in the external nucleus of the IC in the barn owl also suggest that the ability of the animal to detect a change in soundsource direction is based on a shift in the population response of these neurons (Bala et al. 2003). FIG. 14. Representation of sound azimuth in the SC based on population responses. Plots show normalized auditory SRF data which has been pooled over a range of visual azimuths ( 90 and behind, 80 to 45, 45 to 24, and anterior of 24 ). A: data obtained with normal VAS stimuli. The region of maximum response of the pooled data follows the visual best position (the mean of which is shown by the circle). B: this alignment is not altered by fixing the ITDs to 200 s. Role of ITDs in location selectivity of SC neurons Cue-trading experiments have shown that low-frequency ITDs are the primary cue for localization by humans in the horizontal plane (Wightman and Kistler 1992). Tuning to ITDs is also known to underlie the representation of sound azimuth in the optic tectum of the barn owl (Olsen et al. 1989) as well as the behavioral responses of this species (Poganiatz et al. 2001). In contrast, previous recording studies suggest that the map of space in the mammalian SC is based mainly on ILDs and spectral cues. This is supported by the broad, multi-peaked frequency tuning of SC neurons, which is dominated by high frequencies (Carlile and Pettigrew 1987; Hirsch et al. 1985; King and Carlile 1994; King and Palmer 1983; Middlebrooks 1987; Wise and Irvine 1983) and by the presence of a topographic variation in ILD sensitivity along the rostrocaudal axis of the nucleus (Hirsch et al. 1985; Wise and Irvine 1985). Moreover, the changes in spatial tuning observed following occlusion (Middlebrooks 1987; Palmer and King 1985) or passively moving one ear (Middlebrooks and Knudsen 1987), or after removal of the external ear structures (Carlile and King 1994; Schnupp et al. 1998), are consistent with a combination of ILDs and spectral cues being the primary determinants of the spatial selectivity of mammalian SC neurons. Sensitivity to ITDs has been demonstrated in cat SC neurons (Hirsch et al. 1985), but because response latencies decrease with increasing sound level, this could provide a mechanism underlying the processing of ILDs rather than a basis for the representation of auditory space (Yin et al. 1985). A unique advantage of VAS stimulation is the capacity to manipulate independently the acoustic cues available and therefore to assess their contribution to auditory localization (King et al. 2001). This has been done in human psychophysical (Martin et al. 2004; Wightman and Kistler 1992, 1997) and neurophysiological studies (Delgutte et al. 1999; Nelken et al. 1998; Tollin and Yin 2002a,b). By holding ITDs at a constant value while allowing ILDs and spectral cues to vary naturally, we found that the properties of the SRFs of a minority of SC neurons did change significantly, perhaps reflecting the broad ITD sensitivity demonstrated by Hirsch et al. (1985) in the cat. However, there was no overall shift in the centroid direction vectors, which continued to co-vary with the visual receptive fields mapped at the same SC locations. Although behavioral measurements have shown that ferrets can certainly localize sounds using ITDs (A. Schulz, J.W.H. Schnupp, and A. J. King, unpublished observations), our results indicate that these binaural cues make little contribution to the map of auditory space in the SC and are therefore presumably processed by other pathways within the midbrain. ACKNOWLEDGMENTS We are grateful to J. Bithell and R. Ripley for valuable ideas for the data analysis. GRANTS This work was supported by the Wellcome Trust, through a 4-year studentship to R.A.A. Campbell and a Senior Research Fellowship to A. J. King, and by Biotechnology and Biological Sciences Research Council Grant 43/S19595 to J.W.H. Schnupp.