Conceived and designed the experiments: AB. Performed the experiments: MS AB. Analyzed the data: AB. Contributed reagents/materials/analysis tools: MS AB. Wrote the paper: TT AB. Other: Supervised the experiments and analysis: TT. Contributed to analysis: TT. Provided significant insights: TT. Helped in editing and amending the manuscript: TT.
The authors have declared that no competing interests exist.
The relationship between neuronal acuity and behavioral performance was assessed in the barn owl (
The ability to discriminate between stimuli is hypothesized to depend on the reliability of the change in activity of individual sensory neurons (e.g.,
An opportunity to further test the hypothesis is found in the auditory system of the barn owl (
To test this hypothesis, we measured the minimum audible angle (MAA), which quantifies the abilities of owls to detect changes of sound source location in azimuth and elevation. The method of estimating the MAA is based on the pupillary dilation response (PDR). In the owl, the pupil dilates upon presentation of a sound and habituates with repetition of the same sound from the same location. The PDR recovers, however, if the sound source's location is perceptibly changed. The magnitude of the recovered PDR is proportional to the angular displacement of the source, making the PDR similar to a psychophysical rating task. Since pupillary dilation is mediated by the same motor circuits regardless of whether the source is displaced vertically or horizontally, differences in behavioral performance should only reflect differences in sensory resolution. Motor performance would be of concern, by contrast, in a gaze-directing task
Previously
Below, we demonstrate that the vertical and horizontal MAAs differ by a ratio of about 2. We then show that azimuthal discrimination of ICx neurons is finer than elevation discrimination by a ratio of about 2, and that comparisons of the spatial resolution abilities of ICx neurons–assessed by incorporating magnitude as well as variance of firing rate changes in azimuth and elevation–yield a similar ratio. Just as studies based on lesions and microstimulation have implicated the ICx and its efferent target, the optic tectum, in acoustically-guided orienting behavior
Spatial discrimination in azimuth and elevation was measured using habituation and recovery of the PDR in 3 owls. The magnitude of pupillary dilation evoked by a noise burst from the habituating location (
(A) A pupillometer, consisting of an IR detector and emitter (marked), is placed close to the cornea of the owl. The detector is placed about 6 mm from the eye, while the emitter is about 20 mm away. The owl is held immobile in a stereotaxic apparatus, allowing us to position the owl, repeatedly, in the same orientation
Symbols and dotted lines represent discrimination values for each subject. By noting the points of intersection of each of the dashed lines with the discrimination functions, we can extract a ratio of elevation to azimuthal discrimination for a given value of
Thus the ratio between discrimination in elevation and in azimuth is about 2.5 (7.5/3), when the respective MAAs are compared. In addition, by interpolating between measured separations (
We recorded from isolated space-specific neurons to determine whether the height/width ratios of neuronal tuning functions can explain the ratios of behavioral MAAs. Our conclusions are based on recordings from 62 neurons whose SRFs were characterized completely in virtual auditory space. Examples of recorded SRFs are shown in
Black represents the spontaneous firing rate. Colors represent the firing rate, increasing from blue through to red. Note that for three of the neurons, the SRF is elongated in elevation. The neuron in (d) is atypical, in that the receptive field is more elongated in azimuth than it is in elevation. Such neurons were always tuned to low elevations.
After charting the entire SRF in 5° increments, we examined the responses in 1° increments along vertical and horizontal transects through the neuron's best area (blue, red lines
(a) The SRF of the cell is shown; lighter shades correspond to higher response rates. The red and blue lines represent the locations used to estimate the azimuthal and elevational response functions, respectively. (b) Response profiles in azimuth (red) and elevation (blue) show that tuning in azimuth is finer than tuning in elevation. (c) Discrimination functions for a 5° separation were computed using data shown in (b), as per Eqn. 1. Response profiles for both azimuth and elevation are shown for reference. Note that maximal discrimination, especially as seen for azimuth, was achieved where rate of change of firing rate was maximal.
A common way to estimate the spatial tuning of a neuron is to measure the width of its SRF at half maximal firing rate
The width at half the maximal firing response for each neuron in azimuth is plotted against its tuning in elevation. Neurons better tuned in elevation also tend to be better tuned in azimuth. The slope of the best fit line (dotted line) is 2.08, confirming that the width of elevation tuning is about twice that in azimuth.
Although the close agreement between the height/width ratios of behavioral acuity and neuronal acuity suggests a relationship between neuronal and behavioral sensitivity, it could be argued that such agreement is coincidental, since tuning curve half-width is an arbitrary measure, and its value depends on the chosen response rate criterion. Furthermore, analysis based on half-widths does not allow a direct comparison with the psychometric function. We therefore applied signal detection theory, which considers not only the average response rates of the neurons, but also their variance
A change of stimulus position can occur anywhere in space. A given neuron's contribution to the detection of that change across any arbitrary region can be estimated by computing the average of standard separations (
It is also clear in
In
(a) Individual discrimination values plotted here are from unit 719HL. Note that at lower spatial separations, azimuthal discrimination exceeds elevation, but maximal discrimination values are higher in elevation, albeit they occur at much larger separations. (b) Discrimination values plotted against angular separation. Elevation data are slightly offset to the right for clarity. Lines represent the mean discrimination computed from all neuronal pairs, such as those represented by symbols in (a). Note that azimuthal discrimination is roughly double that for elevation. (c) Comparison between behavioral data from individual birds (black) and neuronal discrimination. Data for azimuthal neuronal discrimination for separations <5° are drawn from a previous study
If we average the individual neurons' average
To mechanistically link the neuronal responses to our behavioral paradigm, we had earlier developed a habituation-based model that replicated the observed behavioral and neuronal discrimination performances in azimuth
In this model, neurons of the space map, which do not habituate, were assumed to project topographically to a layer of habituating neurons, the summed activity of which was assumed to control the state of dilation of the pupil (
The illustration shows responses of SSNs and habituating neurons to sound source displacement along the azimuth. Space map neurons (colored circles) are represented by a series of tuning functions (a) centered over each cell (b). The space map neurons, which do not habituate, are assumed to project topographically to a layer of habituating neurons (c). The summed output of the habituating layer is assumed to be proportional to the pupillary response (not shown). The sound is initially presented from location
When the sampled neuronal responses are used as inputs, the model yields discrimination results that closely approximate behavior (
The slope of the azimuthal discrimination function is about twice that of the elevation function in behavior (solid lines), as well as in the output of the habituation model (dashed lines). Note here that behavioral discrimination, shown by solid lines, is closely matched by the output of this computational model.
We demonstrated above that in the barn owl, spatial discrimination in azimuth exceeds that in elevation by a factor of about two. Because the PDR circuitry–which controls behavioral output–is the same whether the discrimination is made along the vertical or horizontal axes, this difference in acuity is likely to reflect the reliability of only the sensory apparatus and not a combination of the reliability in the motor and sensory segments. Indeed, we found that the RFs of neurons in the auditory space map are about twice as tall as they are wide. This ratio of neuronal acuity in elevation and azimuth was also observed with discrimination metrics derived from signal detection theory, and analysis with a computational habituation model (
Discrimination | Azimuth(°) | Elevation(°) | Ratio (El./Az.) |
Behavioral discrimination ( | 3.5 | 8.4 | 2.4 |
Behavioral discrimination ( | 6.0 | 12.0 | 2.0 |
RF tuning widths | 20.0 | 41.0 | 2.1 |
Habituation model ( | 5.7 | 12.4 | 2.4 |
Behavioral discrimination was assessed at a resolution of 1.5°, and neuronal discrimination was assessed at a resolution of 5°. Neuronal ratios values shown below were obtained by noting the angular separations which corresponded to arbitrarily selected discrimination values.
Our measures of auditory spatial discrimination revealed a clear difference between performance in azimuth and in elevation. Spatial hearing in the barn owl has previously been analyzed by evaluating the owl's ability to orient toward the source. Konishi
Thus, our study of spatial
That SSNs in the barn owl midbrain are better tuned in azimuth than in elevation was first reported by Knudsen and Konishi
The shape of spatial RFs is related to the distribution of the binaural cues across space and the sharpness and variance of neuronal tuning to the binaural cues. In the barn owl, interaural differences in timing and level (ITD and ILD), which are subserved by anatomically parallel and physiologically independent pathways, serve as cues for localization
One possible reason for the vertical elongation might be that the cross-correlation-like process, with which ITD is thought to be computed
The discrepancy in the accuracy with which owls could aim their heads at an auditory target and the half-height width of their spatial RFs had led to assertions that owls gathered information from numerous coarsely-tuned neurons to achieve the high behavioral accuracy
The apparent dichotomy between small-headed mammals, which are thought to use a slope code, and owls, which are argued to use a peak-based, “local code” (i.e., a space map) for sound localization, was recently postulated by
Our analysis postulates that the ability of the owl to discriminate changes in spatial location is dependent on the variance that has accumulated in the sensory pathways up to the ICx. However, we do not have an independent estimate of the contributions of motor error on the behavioral response. A recent analysis of tracking of visual targets in primates
Spatial discrimination behavior was measured using a discrimination assay based on the habituation of the acoustically evoked pupillary dilation response, details of which are available elsewhere
Stimuli consisted of reproducible noise bursts with flat spectra (within 1 dB) between 3 kHz and 11 kHz, presented from speakers arrayed as shown in
Sound sources were aligned as closely as possible such that a source placed at 0° azimuth and 0° elevation would cast an image onto the retina at the visual fovea, or
Neuronal acuity was assessed in three barn owls under nitrous oxide anesthesia. The RF of isolated space-specific neurons was determined by presenting sounds from the frontal hemisphere, using a virtual auditory space paradigm based on individualized HRTFs
To compute the standard separation for neural responses,
The habituation model attempts to reproduce the essential features of our behavioral paradigm, the habituation and recovery of the PDR
The model also incorporates the trial-to-trial variance in firing of the space-specific neurons which diminishes the ability of HL neurons to respond to changes in firing rate induced by a test sound. This variance was included by scaling the output of each ICx neuron by its standard deviation. Thus, space-specific neurons with small variance in their response will contribute more to the HL neurons' response, while those with large variances in their response will have a proportionately reduced contribution to the output of the HL neurons. This property of our hypothetical HL neurons is reminiscent of neurons in the fly's visual system whose firing rates scale with the variance in prior stimulus inputs
Each of the two components, mean and variance, is included in the computation as follows. First, the output of each ICx neuron, R, is normalized by the variance of response to the habituating stimulus (VHS; Eqn. 2):
The output of each HL neuron is summed and averaged
We thank Dr. Kip Keller for technical assistance and Dr. Jagdeep Kaur-Bala for critically reading the manuscript.