Surgery and animal care
We describe here the surgical procedures used for both behavioral and neurophysiological experiments. The owls used in behavioral experiments were not the same as the physiology owls. All owls were used in other parallel behavioral and neurophysiological studies. All surgical instruments were sterilized. Owls were anesthetized with intramuscular injections of ketamine (10–20 mg/kg/h, Phoenix Pharmaceutical, St. Joeseph, MO, USA) and diazepam (0.25–0.5 mg/kg/h, Steris Laboratories, Phoenix, AZ, USA). A scalp area about 10 mm × 10 mm was incised after subcutaneous injection of the local anesthetic xylocaine (0.2 mL 2% lidocaine HCL; Astrazeneca Pharmaceuticals, Wilmington, DE, USA). The first layer of the skull was removed within this area by using a pair of rongeurs, and a stainless steel post for the behavioral owls, or an inverted T-shaped post for the physiology owls, was fixed to the underlying trabeculae with dental cement (Hygenic, Perm Reline & Repair Resin Type II Class I). The procedure lasted about an hour. After the surgery, owls were encased in a snugly fitting cylinder to prevent struggling and kicking as they recovered from anesthesia. Owls were observed in small cages in a separate recovery room until they came out of the cylinder. When owls recovered well enough to fly, they were returned to their living cages where the owls for neurophysiological experiments were fed two mice a day. The weights of the owls for behavioral experiments had to be maintained at about 90% of their free-feeding values to ensure well-motivated behavioral performance.
The surgery for neurophysiological experiments involved the removal of a scalp area and the underlying skull area about 7 mm × 7 mm in size to expose the brain surface. The surgery was performed under anesthesia as described above. After each experimental session, which lasted several hours, the craniotomy was cleaned with the antibacterial agent: Chlorhexiderm (0.05% chlorhexidine gluconate, DVM Pharmaceuticals, Miami, FL, USA) and resealed with a small plastic sheet and dental cement. The scalp wounds were sutured shut. Sutures and the plastic sheet were removed before each experiment and replaced afterward. The owls for behavioral experiments were not subjected to any additional surgery.
Owls were anesthetized with ketamine throughout the experiment and lidocaine was applied to the edges of the skin incision once per hour. Since ketamine caused a temporary reduction in neuronal discharge, we did not collect data for approximately 15 minutes after each booster injection of ketamine (20 mg/kg/h). Single units were isolated in the optic tectum with parylene-insulated tungsten microelectrodes (250 µm thick shaft, 5 MΩ impedance at 1 kHz), A-M Systems, Carlsborg, WA, USA.
Neural waveforms were amplified and filtered from 1 to 10 kHz (Microamp 200, Beckman Electronics, Fishers, IN, USA) and simultaneously monitored with an oscilloscope (5110, Tektronix, Beaverton, OR, USA), and an audio amplifier (AM8, Grass, West Warwick, RI, USA). A spike discriminator (SD1, Tucker–Davis Technologies, Gainesville, FL, USA) converted neural spikes to computer-compatible TTL pulses, the times of occurrence of which were stored in a computer file. The identity of optic-tectum neurons was also confirmed by their histological locations, some of which were marked with electrolytic lesions and visualized in cresyl-violet-stained sections.
The auditory stimuli were bursts of broadband Gaussian noise, 0.5–12 kHz, 100 ms in duration with left and right channels simultaneously gated with 5-ms rise/decay times. The noise bursts contained different values of ITD and interaural level differences (ILD) to which tectal neurons are tuned. All sound stimuli were digitally synthesized by a Dell Dimension XPS Pro200n computer and delivered by a digital signal processor equipped with a 16-bit, 48-kHz digital-to-analog converter (Tucker–Davis Technologies). ITDs were computed online, whereas ILDs were set by two digital attenuators (PA4, Tucker Davis Technologies) which were controlled by the computer. ILDs were varied in 5-dB steps from -40 to 40 dB. Once the neuron's best ILD was determined, all ITD measurements were made at that ILD. Best ILDs were within the range of ±15 dB. ITDs could be varied in 30-µs steps in either direction, i.e., leading or lagging to each ear.
Sound stimuli were delivered through an earphone assembly consisting of a Knowles (Itaska, IL) ED-1914 receiver as a sound source and a Knowles BF-1743 damped coupling assembly for smoothing the frequency response of the receiver. These components were encased in an aluminum cylinder 7 mm in diameter and 8.1 mm in length that fit into the external ear canal. In addition, a Knowles 1939 microphone for monitoring sound pressure levels in the ear canal was included in the earphone cylinder. The cylinder was inserted into the ear canal and the gaps between the cylinder and the canal were filled with a silicon compound (Earmold & Research Laboratories, Whichita, KS, USA). The microphone was initially calibrated against a 12-mm Brüel & Kjær (Norcross, GA, USA) microphone with a probe tube whose tip was placed close to the center of the eardrum. This procedure allowed translation of the voltage output of the Knowles microphone into dB SPL. A standard value of 20 dB above a neuron's threshold was chosen as the stimulus sound level in all cases.
Behavioral methods for owls
We used three tame adult owls for behavioral experiments. The head-turning response did not need any training, but it had to be reinforced by food for repeated trials. We trained the owls to feed from an apparatus which dispensed a small amount of mouse meat at a time, allowing 20–30 trials in an hour. Test sessions seldom continued more than 2 hours. We initially used a small freefield speaker ("hoop speaker") mounted on a semicircular track to encourage the owls to localize it at various azimuthal angles. The owls had to initially orient to another source ("zero speaker") placed straight ahead and wait for the signal from the hoop speaker.
When the owls became consistent in localizing both speakers, they were trained with earphones. We used earphone assemblies similar to the ones used for neurophysiological experiments. A metal bar was attached to the head post and held the left and right earphones in place. All sound stimuli were digitally synthesized in an IBM-compatible PC using Matlab and presented, after appropriate lowpass filtering, through 16-bit D/A converters at a rate of 40 kHz (Sound Blaster Live, -120 dB noise floor; Milpitas, CA). The sound pressure level for all stimuli used in the behavioral experiments was 20 dB above the owl's threshold, which is about 0 dB SPL between 3 and 8 kHz at the eardrum (Dyson et al. 1998
). Interaural level differences (ILDs) were always kept zero so that the owls turned their heads only in the horizontal plane in response to an ITD (Moiseff 1989
). The onset and offset of the sounds were not interaurally delayed and, therefore, the detection of ITD was based on the ongoing part of the waveforms. Sound stimuli were delivered by an earphone assembly consisting of Knowles components as described earlier.
The experiments consisted of two parts. The rationale for the first part was to determine the largest ITD that could be detected by owls. For this part, the acoustic stimulus was a single burst of broadband Gaussian noise, 100 ms in duration with a 10-ms rise/decay time. The value of ITD was varied on each presentation. The owls performed in complete darkness in an IAC anechoic chamber (5 × 3 × 3 m) and were monitored with an infrared video camera. On each trial, a broadband click was presented from a loudspeaker directly in front of the owl. The purpose of this click was to have the owl fixate at 0°. After the owl had fixated at 0°, the target stimulus was presented through headphones. For the target sound, the computer would randomly select one of 14 ITDs. The ITD had an equal a priori probability of leading to the left or the right ear. The 14 values were 150, 200, 300, 400, 500, 750, and 1000 µs leading to the left or the right ear. The owl's task was to orient to the side (hemifield) corresponding to the sign of the ITD, i.e., if the stimulus had an ITD leading to the left ear, the owl's task was to orient its head to the left and pause for a minimum of 0.5 s. If the head orientation was to the correct direction, the owl was immediately rewarded with a small amount of mouse meat.
To prevent experimenter bias, the computer did not display the value or the sign of the ITD until after the owl had responded and the trial had been scored. The experimenter observed the owl on a monitor outside the anechoic chamber and scored a left or a right head turn by watching the monitor. Once the experimenter recorded the head-orienting response, the computer displayed the sign and magnitude of ITD. If the direction of head-turning was consistent with the sign of ITD, the experimenter would dispense the food reward. Only the direction of the response (left vs. right) and not the absolute angular magnitude of the head orientation was used in analysis. When the owls were not motivated to localize the sound, response latencies were long and head-turning responses were slow. This occurred only on a small proportion of trials near the end of an experimental run. Since these responses were always correlated with decreased accuracy, all trials in which time to head fixation exceeded 1 s were excluded from further analysis.
The rationale for the second part of this study, as elaborated fully in "Psychophysical Predictions" subsection of the Discussion section, was to determine if the perceived position of noise bands may be predicted from a model of binaural interaction with short delay lines. Such a model predicts localization reversals at specific regions of the frequency spectrum, and therefore, we employed narrow bands of noise for this study. All conditions were the same as the first part of this study except for the following changes: The stimulus was a narrow band of noise with a large ITD. The ITD was either 400 or -400 µs, randomly selected on each trial. A calibration sound was used to determine the ITD that produced a 0° head-pointing response. This calibration ITD was +50 µs for one owl and was used to offset a small initial bias. For the other owl, the calibration ITD was 0 µs. The stimulus ITDs were -400 and 400 µs relative to calibration ITD. The bandwidth of the noise was also selected randomly on each trial from a set of seven predetermined values: 2–2.5 kHz, 2.2–2.5 kHz, 2.3–2.5 kHz, 2.5 kHz tone, 2.5–2.7 kHz, 2.5–2.8 kHz, and 2.5–3 kHz. Thus, one edge of the noise band was always at 2.5 kHz and the other edge was either above or below the first edge by 200, 300, or 500 Hz. The owl was trained to point its head to the perceived position of the sound. Head orientation was measured by a magnetic tracker (Polhemus 3Space Isotrack, Colchester, VT, USA) mounted on the owl's head. This system measured orientation to within 0.5°.
Behavioral methods for human subjects
Three normal-hearing individuals served as subjects. They were seated in a dimly lit double-walled steel chamber (IAC with interior dimensions of 1.8 × 1.9 × 2 m) and listened to the stimuli through headphones (Sony Model MDR-V1). Each run consisted of a presentation of 150 trials. Each trial began with the presentation of a 300-ms target stimulus (a Gaussian noise band) followed by a 300-ms acoustic "pointer" (white noise burst). The task of the subject was to adjust the ITD of the pointer such that its perceived location matched that of the test stimulus. This pointing paradigm is a standard method of estimating the perceived position of complex sounds by human subjects (Stern et al. 1988
; Domnitz and Colburn 1977
The design of this experiment was similar to the second part of the behavioral experiments for owls. The test stimulus was assigned one of two ITDs with an equal prior probability on each trial (-4000 or 4000 µs). The frequency band of the test stimulus was randomly selected on each trial from a set of 6 conditions: 200–250 Hz, 225–250 Hz, 240–250 Hz, 250–260 Hz, 250–275 Hz, and 250–300 Hz. Thus, one edge of the band was always at 250 Hz and the bandwidths were 10, 25, and 50 Hz. On each trial, one of two ITDs was paired with one of 6 noise bands. The noise bands were generated by summation of components in the frequency domain. Their amplitudes were selected from a Rayleigh distribution and their phases from a uniform distribution (0, 2π), thus producing a Gaussian band of noise (Rice 1954
). The level of the stimulus was 65 dB SPL measured with a 6-cc coupler and sound level meter (Brüel & Kjær Model 2260, B&K 0.5-in. microphone model 4189). Sounds were presented through D/A converters (Sound Blaster Live, Milpitas, CA, USA) at a rate of 40 kHz.
The initial ITD of the pointer was randomly selected from a range of -600–600 µs. The pointer ITD could not exceed these limits. If the perceived position of the pointer did not match that of the target stimulus, the subject pressed either a left or right key to "move" the pointer closer to the target. A large step size (100 µs) was available to the subjects for rapid pointer adjustment, and a small step size (25 µs) was available for fine tuning. The subject could listen to the target–pointer pair and adjust the pointer ITD as many times as s/he required until satisfied that the perceived positions of the target and pointer were the same. At this time, the subject pressed a separate key to terminate the trial, record the final ITD of the pointer, and proceed to the next trial. Approximately 10 pointing estimates were obtained for each combination of ITD and bandwidth. No feedback was provided during the run.