Cross-correlation in the medial superior olive reexamined. R Batra*, TCT YinÝ. *Department of Anatomy, University of Mississippi Medical Center; İDepartment of Physiology, University of Wisconsin Medical School. The medial superior olive (MSO) is one of the primary sites where a sensitivity to interaural temporal disparities (ITDs) is extracted from the temporal discharge pattern of auditory neurons. Psychophysical modeling assumes that neurons of the MSO cross correlate their inputs to acquire this sensitivity, but tests of this assumption are few.
Here, we reexamine the relationship between the inputs to MSO neurons and their sensitivity to ITDs. We use data from previous studies of extracellular responses of MSO neurons in the cat (Yin & Chan, J. Neurophysiol. 64: 465-488, 1990) and the rabbit (Batra et al., J. Neurophysiol. 78: 1237-1247, 1997). We then model the relationship using an extended version of a model devised by Colburn et al. (Hear. Res. 49: 335-346, 1990).
Cross-correlation of the left and right inputs by an MSO neuron implies a mathematical relationship between the range of ITDs to which it responds and the jitter in the discharge at the left and right inputs. In response to tones of low frequency, auditory neurons synchronize their discharge to one particular phase. The jitter about this phase is described by a synchronization coefficient (SC): the greater the SC, the more tightly synchronized the discharge is to the preferred phase. Similarly, MSO neurons discharge maximally at a preferred interaural phase difference. The degree of preference can also be described by an SC. This interaural SC is a measure of how tightly a neuron is tuned to a particular interaural phase difference. For ideal cross-correlation, which is mathematically similar to convolution, the interaural SC equals the product of the SCs of the left and right monaural inputs.
The SCs of the inputs were estimated in two ways: from the SCs of the MSO neuron to monaural tones and from the SCs to the tones at either ear during a binaural-beat stimulus. A binaural-beat stimulus consists of tones to either ear that differ slightly in frequency, and produce a continuous change in the interaural phase difference. The SCs to the left and right tone during this stimulus, as well as the SC to the beat frequency, were obtained by Fourier analysis of the corresponding frequency components of the response.
The product of the SCs derived from responses to monaural tones overestimated the interaural SC. The SCs derived from responses to binaural-beat stimuli were smaller, and their product more closely matched the interaural SC.
The observation that the SCs from binaural-beat stimuli were better at predicting the interaural SC than SCs from monaural stimuli was puzzling. The cross-correlation hypothesis implies that the product of the SCs of the input fibers from the two sides should predict the interaural SC of the MSO neuron; however, it is unclear whether the SCs of the input fibers are better reflected in the response of the neuron during monaural stimulation or during a binaural-beat stimulus. To investigate this matter further, we modeled the responses of neurons in the MSO to monaural and binaural-beat stimuli.
Modeling of the response and varying the parameters involved indicated that the SCs derived from responses to monaural tones typically matched the SCs of the inputs. This, coupled with the observation that the product of these SCs overestimates the interaural SC, implies that neurons of the MSO do not precisely cross-correlate their inputs, but are more broadly tuned to ITDs than anticipated. The modeling also indicated that the weaker-than-expected interaural SC was a result of the MSO neuron discharging in response to activity at only the left or right input, in addition to discharging when responses were present at both inputs. This effect can also explain why the product of SCs to tones derived from responses to binaural-beat stimuli more closely matches the interaural SC.
Supported by NSF grant IBN 9807872 and NIH grant DC 00116. The original studies in the cat and rabbit were supported by NIH grants DC 02840 to T.C.T. Yin and DC 01366 to S. Kuwada.
The eight mammalian genes Kcna1 through Kcna8 are related to the single Drosophila gene Shaker and code for the voltage-gated potassium (K+) channel subunits Kv1.1 to Kv1.8. Four Kv1 subunits combine to form channels underlying low threshold K+ currents, which in vivo can begin to activate at potentials as negative as -60 mV, i.e. at or near the resting potential. Such Kv1 channels are thought to be important for limiting excitability and reducing temporal summation in auditory neurons that receive and transmit phase-locked information, including principal neurons of the medial nucleus of the trapezoid body, or MNTB, which express both Kv1.1 and Kv1.2 subunits.
Recordings in brainstem slices from mice lacking Kcna1 showed that their MNTB neurons were strongly hyperexcitable and had reduced amplitudes of low threshold K+ currents (Brew et al, 2003). This was not unexpected given that Kv1.1 was thought to be one of the major subunits contributing to these K+ currents. Heterozygous mice with only a single copy of the Kcna1 gene had MNTB neurons with normal excitability. From the high sequence similarity between Kcna1 and Kcna2, and the similar K+ currents they produce when expressed in oocytes, we expected that Kcna2-null MNTB neurons would also exhibit hyperexcitability and reduced K+ currents. However, we found that Kcna2-null MNTB neurons were actually hypoexcitable, as were the Kcna2-heterozygous MNTB neurons (Brew et al, 2000 and 2001). I will present simulations (using the software NEURON) of these neurons' excitability, showing how very subtle alterations in the voltage-dependence of activation of Kv1 channels can strongly influence excitability and synaptic responses. I will also show the effects of varying the relative proportions of different types of K+ currents, and relate this to the K+ current differences that have been noted between rat and mouse auditory neurons as well as across tonotopic gradients. These simulations support the idea that precise differential regulation of Kcna genes may be used to fine-tune neuronal excitability for different tasks.
Experimental evidence indicates that synaptic modification depends upon the timing relationship between the presynaptic inputs and the output spikes that they generate. Results are presented for models of spike timing-dependent plasticity (STDP) with additive potentiation and multiplicative depression. Previous studies of this form of STDP found that spike-timing correlations among a subset of the synaptic inputs increased the corresponding synaptic weights while leaving the weights of the uncorrelated inputs unchanged . Competition between synapses is introduced by activity-dependent scaling of the synaptic weights , which occurs on a much slower time scale than STDP. However, it was also found that higher-rate synaptic inputs with no spike-timing correlations are not selectively potentiated , a result that is at odds with the classical studies of long-term potentiation and depression. Our study re-examines these results when the effect of two biologically important refinements of the time dependence of synaptic plasticity are included: (i) time asymmetry of the STDP time window, in which the time constant associated with synaptic depression is larger than that associated with potentiation , and (ii) the limited time-extent of the interaction between input excitatory postsynaptic potentials (EPSPs) and output action potentials (APs), whose importance was highlighted in recent experimental studies . Four classes of models are identified according to the time-extent over which EPSP-AP interactions can occur.
The results indicate that when the STDP time constants for potentiation and depression are equal there is no selective potentiation of synapses that have higher rates of input (and no spike-timing correlations). Selective potentiation of synapses with higher input rates is only possible for models in which: (i) the STDP time constant for depression is larger than for potentiation, and (ii) the time-extent of the EPSP-AP interactions is "input restricted" (i.e., restricted to the interspike interval of the inputs).
The analysis is based upon the Fokker-Planck approach and is carried out using a conductance-based leaky integrate-and-fire neuronal model. Averaging over the EPSP-AP interaction time distributions for many independent small amplitude weight increments provides a unified account of spike-timing [1,3] and rate-based models  of synaptic plasticity. This approach enables not only the average of the distribution of excitatory conductances to be calculated but also the standard deviation of the distribution.
Many aspects of auditory perception utilize fine timing information provided by the phase-locking of auditory nerve fibers to acoustic stimuli. The population-wide distribution of interspike intervals at the level of the auditory nerve forms an autocorrelation-like temporal representation of the stimulus whose properties explain many diverse aspects of pitch perception (e.g. missing fundamentals, level-invariance, pitch equivalence, octave similarity, pitch shifts of inharmonic complex tones). Population-interval distributions (PIDs) also provide robust representations for those aspects of timbre that are related to stationary power spectra (e.g. formant structure and vowel quality). PID-based models of masking and harmonic resolvability that are based on the competition of interval patterns integrate this information across cochlear territories in a manner that reflects cochlear excitation patterns.
A central question for auditory neurophysiology concerns the means by which the central auditory system might make use of such timing information. The bandpass modulation tunings that have been observed in the auditory pathway fail to account for the level-invariant nature of pitch and pitch shifts of inharmonic tones. Thus far, except for a few possible exceptions, central pitch detectors that could account for pitch equivalences between pure tones and harmonic complexes have not been found. It is therefore difficult to envision how processing schemes for periodicity analysis based on time-rate coding transformations in the ascending auditory pathway could work in practice (we look forward to an open and wide-ranging discussion of these issues).
As a consequence, we have explored alternative, time-domain processing strategies for utilizing stimulus-driven fine timing information. In the spirit of Licklider's duplex model, they are meant as heuristics that illustrate functional principles rather than descriptions of the input-output behavior of particular neuronal populations. Feedforward neural timing nets are coincidence arrays that function as asynchronous temporal pattern sieves to pass interval patterns that are shared across inputs. In effect they multiply the autocorrelations of their input spike trains to extract pitch irrespective of timbre (and vice versa). Recurrent timing nets are coincidence arrays with delay loops that build up and separate auditory objects with different fundamentals (e.g. different voices, instruments) on the basis of temporal pattern invariances. These recurrent nets suggest alternative correlation-based strategies for scene analysis that are based on temporal pattern coherence rather than on feature detection and binding.
Ascending information from brainstem structures converges in the central nucleus of the inferior colliculus (ICC) and is rerouted to the primary auditory cortex (AI) via the auditory thalamus (MGBv). This hierarchical organization leads to a significant reduction in temporal modulation preferences from the ICC to AI. Neurons in the ICC, MGBv and AI also exhibit significant selectivity to spectral, intensity, and aural dimensions of complex stimuli although the interactions for simultaneously processing these stimulus dimensions are currently not well understood. A spectro-temporal Gabor analysis technique is presented that we are using to study spectro-temporal preferences and the transformations in these neuronal stations. We use this technique to quantify the interrelationships, trade-offs, and differences between temporal, spectral and aural aspects of processing. Comparisons of auditory spectro-temporal receptive fields (STRFs) in the ICC, MGBv, and AI shows a relative conservation of spectral modulation preferences but a large shift in the preferred temporal modulations. Temporal modulation transfer functions are significantly overlapped in the thalamus and AI, although they extend to higher modulation rates in the ICC. A novel trade-off in spectral and temporal modulation preferences is seen only at the level of the ICC. Aural selectivities are explored via comparisons of left versus right ear STRF. We are finding that the binaural structure of the auditory STRF is adept to a spectro-temporal binaural disparity analysis, analogous to the spatio-temporal binocular disparity analysis proposed by Anzai et al. (1997) for processing motion and depth in the primary visual cortex. Such a joint analysis of spectro-temporal and binaural information may enable auditory neurons to simultaneously and independently encode head related spatial cues and contextual information found in complex environmental stimuli.
Ian Forsythe, Matt Barker, Brian Billups, Paul Dodson, Bruce Graham & Adrian Wong. Department of Cell Physiology & Pharmacology, University of Leicester, P.O Box 138, Leicester LE1 9HN. UK.
Information is encoded as trains of action potentials that in the binaural auditory pathway are relayed and integrated to perform specific computations associated with sound source localisation. The temporal fidelity of this information is a crucial factor. The calyx of Held synapse with its postsynaptic target, the medial nucleus of the trapezoid body (MNTB) is considered a 'simple' relay synapse in which an excitatory input is converted into an inhibitory projection to the contralateral auditory brainstem. The efficacy of transmission at this synapse is dependent on multiple presynaptic and postsynaptic factors that provide insight into auditory processing and the more general limitations of information transmission at central synapses. Since most physiologically relevant information is transmitted as action potential trains we have examined some of the factors that influence frequency-dependent changes in synaptic efficacy.
The first question concerns the role of presynaptic potassium conductances in action potential firing. It is clearly established that fast spiking neurones express Kv3 potassium conductances that aid AP repolarisation, yet nodes of Ranvier express little or no functional Kv channels, although Kv1 are present in juxtaparanodal regions. Intriguingly, immunohistochemistry clearly shows that Kv1 (and Kv3) are highly expressed at many synaptic terminals. So what do they do? Using subunit-specific toxins we show that homomeric Kv1.2 channels are located in the last 20 um of the axon. They take no part in AP repolarization, but serve to reduce axonal hyperexcitability during the depolarising after-potential (DAP) that accompanies all APs in myelinated axons. Thus presynaptic Kv1 channels crucially maintain the AP pattern of the presynaptic train by blocking aberrant APs generated by the passive spread of capacitive current in myelinated axons.
Each AP triggers calcium influx (through P-type channels), exocytosis of glutamate and activates postsynaptic AMPA receptor-mediated EPSCs. During high frequency trains (>10Hz) EPSC magnitude rapidly declines due to presynaptic vesicle depletion and postsynaptic AMPA receptor desensitisation. Distinction between these two mechanisms of short-term depression is a major (and ubiquitous) physiological problem. We have developed a new method to minimise postsynaptic desensitisation based on the use of low affinity competitive antagonists such as kynurenate and -D-glutamaylglycine (avoiding use of cyclothiazide which has many non-specific actions). A simple model of transmission at the calyx confirms the mechanism involves 'diversion' of AMPA receptor kinetics away from desensitisation. This method shows that desensitisation makes little contribution to short-term depression at frequencies below 10Hz, but makes increasing contributions during higher frequency trains. We conclude that desensitisation contributes to short-term depression at synapses both before and after hearing onset and suggest that estimates of the readily releasable pool of synaptic vesicles have been underestimated by around 40%.
Our data suggest that there is nothing 'simple' about transmission at a relay synapse: multiple pre- and postsynaptic adaptations contribute to maintain and modulate the efficacy of transmission at the calyx of Held.
For a longtime, ITD processing in birds and mammals has been thought to function as suggested by L. Jeffress' seminal model1. This model incorporates excitatory projections from both ears that faithfully time-lock to the temporal structure of sounds and convert onto binaural coincidence detector neurons. The latter fire maximally, when the two inputs arrive exactly simultaneously. Additionally, the model assumes that via a systematic arrangement of the length of the input fibers (delay-lines), different conductance delays can be achieved that tune different coincidence detector neurons to different favorable ITDs. Such a system could then create a map of best ITDs, hence of azimuthal space, by means of the distribution of peak firing rates.
Recent results from the rabbit2, gerbil3 and guinea pig4 auditory systems, however, revealed more or less unexpected features. First, best interaural time differences (ITDs) in the gerbil3 and the guinea pig4 strongly correlate with the best/characteristic frequencies (BF) in a way, that adjusts the maximal slopes rather than the peaks of ITD functions to the physiologically relevant range. Second, indirect2 and direct3 evidence from MSO recordings suggest a pronounced influence of inhibitory projections on the ITD tuning of single cells.
Our recent recordings of ITD sensitivity in the dorsal nucleus of the lateral lemniscus (DNLL) and its ontogenetic development in the gerbil confirm the relationship of BF and best ITD. Moreover, our new results indicate that initially, shortly after hearing onset, the excitatory inputs create a sensitivity that has its maximum around zero ITD. Hence, juvenile ITD functions are similar to adult ITD functions during blockade of inhibition in the MSO3. Moreover, the adjustment of the ITD sensitivity and the development of the glycinergic inputs5 both depend on early auditory experience and can both be inhibited by rearing animals in omnidirectional white noise.
An open question is whether the principles underlying detection of ITDs and their neuronal representation found in the small gerbil MSO are a feature of all low frequency hearing mammals, or whether they are different in different phylogenetic groups of mammals. The fact that the distribution of glycinergic inputs is similar in ITD using animals like cats6, chinchillas7 and gerbils5, but different in non-ITD users like bats, short tailed opossums or rats5, argues for identical functions. Similar arguments hold for the way ITDs are neurally represented. The fact that there is coherence between the neural representation in guinea pig IC4 and the Mongolian gerbil MSO3/DNLL present study), even though these two species are not closely related8 and evolved low frequency hearing independently9 indicates that our findings, again, are of general relevance for ITD using mammals. Recent results from the cat IC10 strongly support this notion.
Taken together, there is strong evidence that mammals evolved one particular mechanism of encoding and one way of representing ITDs - and these are significantly different form those suggested by Jeffress.
We are exploring the representation of sound-source location by neurons in the auditory cortex. We employ an "ideal observer" approach, in which we test the accuracy with which one could localize a sound source given only the information that is available in the firing pattern of one auditory cortical neuron or a small ensemble of neurons. We use artificial neural networks to classify neural responses. Using this approach, we have evaluated the stimulus-related information that is transmitted by neurons in various auditory cortical fields and by specific features of spike pattern. Results show that the temporal firing patterns of single neurons can vary systematically with sound-source location throughout as much as 360° of space. That is, single neurons are panoramic and, by implication, any particular source location is represented by widely distributed populations of neurons. The timing of spikes appears to transmit as much or more information than does the mean spike rate. Sounds that elicit localization illusions in human listeners elicit rather parallel location-specific responses patterns from cortical neurons. In comparisons of various auditory cortical areas in the cat, we have found qualitatively similar location-coding properties in fields A1, A2, and AES. In contrast, recent results suggest some specialization for location coding in the posterior auditory field (PAF). Compared to neurons in other cortical fields, PAF neurons show a greater diversity of spatial sensitivity, sharper location selectivity, and markedly greater modulation of spike timing by variation in sound-source location.
Supported by NIH grants RO1 DC00420, PO1 DC00078, and P30 DC05188.
I will describe here the representations of some simple and complex sounds in the inferior colliculus, auditory thalamus and primary auditory cortex. I will show that whereas sound representation in the inferior colliculus is reasonably well described by filtering operations in frequency and time, in auditory cortex the representation of weak and rare acoustic components is non-linearly enhanced. Some of the processes responsible for this differential enhancement operate already in the auditory thalamus, whereas others, mostly those with longer time constants and high resolution in terms of physical variables, first appear in auditory cortex. I will argue that these are correlates of auditory scene segregation, both simultaneous and concurrent, in auditory cortex.
Human auditory cortex is typically studied in a non-invasive manner by using hemodynamic (PET, fMRI) or electromagnetic (EEG, MEG) techniques. The disadvantage of these techniques is that their granularity is very coarse relative to the questions one can investigate with animal preparations. The advantage is that one can perform psychophysical studies concurrently with noninvasive recording, thereby permitting rather constrained interpretations of the relation between neuronal activity and behavior.
The research program we are pursuing aims ultimately towards an understanding of the cortical basis of speech perception. A major subgoal in defining reasonable biological models must be to bridge the gap between the data we are able to obtain from 'awake behaving humans' and the data deriving from animal work. In two areas of inquiry, (i) functional anatomy and (ii) the relevance of timing information, some progress has been made. From (i) the perspective of functional anatomy, two major insights are that there exist (unsurprisingly) multiple areas that appear to be organized along parallel, hierarchically organized pathways and, moreover, that there is a much richer contribution of the non-dominant hemisphere to the analysis of speech signals. With regard to (ii) the relevance of timing information (both in the signal and the neuronal response), it is increasingly clear that human auditory cortical responses reflect extraordinary sensitivity to temporal structure in complex signals and, moreover, may use temporal mechanisms to represent the information.
I will discuss experiments based on the various noninvasive recording techniques that, cumulatively, highlight these features of the functional anatomy and physiology. I argue that specific temporal integration constants form the basis for complex sound processing. The model, asymmetric sampling in time (AST), derived from psychophysical and imaging data, suggests that at the cortical level there are two privileged integration windows (~20-50ms and ~150-250ms) and that these two integration times are differentially weighted in left versus right non-primary auditory cortices. Experiments using both speech (audio-visual syllables; filtered speech) and non-speech stimuli (click-trains; concatenated noise-bands) are presented in support of this hypothesis.
Supported by NIH DC 05660 to DP.
Distinct biophysical properties including multiple voltage-dependent membrane conductances and well-timed transient inhibition contribute to the temporally precise processing characteristics of auditory neurons. We investigate the underlying mechanisms of coincidence detection through in vitro experiments (gerbil MSO) using dynamic clamp stimuli and with computational models of the Hodgkin-Huxley type. We focus particularly on what makes these neurons fire, i.e. on how they integrate subthreshold signals in the presence of a noisy synaptic (excitatory and inhibitory) background, as is typical in vivo. Consistent with previous reports, the partial blockade of low threshold potassium currents (IKLT) reduced coincidence detection (as well as reduced phase-locking and signal-to-noise ratio). We used analysis by spike triggered reverse correlation for injected current Irevcor to evaluate and interpret our results. Blockade of IKLT slowed the rise of Irevcor, indicating a less precise time window for integration. Presumably the faster rise, in control, is required to reach threshold before IKLT is activated. Also, spike generation was associated with a preceding (by a few msec) hyperpolarization ("dip") in Irevcor, suggesting a drop in excitatory current or increase in inhibitory current to promote spiking. Multiple factors pointed towards the involvement of a second, novel mechanism. Even in the presence of an IKLT antagonist, the dip in Irevcor persisted; cells did not convert to tonic mode, but remained phasic; rebound action potentials were produced after termination of a hyperpolarizing stimulus with 30% larger amplitudes as compared to spikes evoked by depolarization. We suggest that the sodium current (INa) is substantially inactivated at rest and describe some manipulations of INa in experiments and in computations to further support this suggestion. Our computer model, including conductances for spike generation and for IKLT, shows decreased coincidence detection when IKLT is reduced or when INa is increased (compensating for substantial inactivation at rest). We hypothesize that favored (on average) temporal combinations of synaptic inputs transiently reduce the inactivation of INa and deactivate some of IKLT to create the brief temporal window for coincidence detection of small signals in noise.
Joint work with G Svirskis, R Dodla, V Kotak, D Sanes in the Center for Neural Science, NYU.
The processing of auditory stimuli changes significantly during the course of normal maturation or following partial hearing loss. We are interested in the contribution of inhibitory synaptic transmission to the generation of auditory coding properties, particularly the possibility that inhibitory synaptic strength is regulated by spontaneous or environmentally-driven activity. We have explored how patterns of synaptic transmission can alter the strength of an inhibitory projection from the medial nucleus of the trapezoid body (MNTB) to the lateral superior olive (LSO). During postnatal development, individual MNTB arbors become restricted along the LSO frequency axis. These arbors remain in an expanded state when MNTB neurons are functionally denervated, suggesting the involvement of an activity-dependent mechanism. Complementary observations have been made in two other brain stem auditory nuclei, the MSO and the SPN. In each instance, inhibitory synapse refinement is thought to underlay the maturation of a specific auditory coding property. To determine whether there is a period of inhibitory synaptic plasticity during development, whole-cell recordings were obtained from developing LSO neurons in a brain slice preparation. Recordings from P7-19 LSO neurons show that low frequency stimulation of the MNTB leads to a ~50% decline in evoked inhibitory synaptic currents. This form of activity-dependent depression is age-dependent, suggesting that it could support the developmental rearrangement of inhibitory MNTB terminals as they compete with neighboring excitatory and/or inhibitory inputs. Recently, we have examined the cellular mechanism of inhibitory synapse plasticity. One surprising result is that MNTB neurons, which are glycinergic in adult animals, also release GABA during development. In fact, GABA signaling is necessary for activity-dependent inhibitory synaptic depression, and this depression is mediated by GABAB receptor activation on LSO neurons. These results emphasize the dynamic nature of inhibitory synaptic gain, and provide specific cellular mechanisms to account for such properties.
A variety of mechanisms determine whether a neuron is best suited for extracting information about either the intensity or the synchrony of its inputs. Central neurons have been classically described as "integrate-and-fire" or "temporal integrator" (TI) neurons to emphasize that the firing frequency of a typical neuron is proportional to the steady-state rate of synaptic inputs. Contrary to this, a minority of neurons, particularly those found in the auditory system, are understood to function as "coincidence detectors" (CD) in that they do not respond so much to the frequency of input synaptic events as to the clustering of synaptic inputs within narrow time windows. Recent experimental and theoretical work has called into question these distinctions by pointing out that under normal operating conditions, the output of most central neurons does not, in fact, behave like pure TI or CD neurons but as a blend. To examine this conjecture I will discuss results from in vitro recordings and modeling studies on the ability of the two neuronal types to modulate their firing rate in response to systematic variation of input synchrony over a wide range of input intensity. I will show specific examples of how the input-output relation of the two neuron types are modified by factors like dynamic changes in postsynaptic membrane properties, variation of input timing, synaptic inhibition and short term plasticity.
Synthesis and use of neural images of auditory space in the barn owl. Takahashi TT, Bala ADS, Euston DR, Keller CH, Spezio ML, Spitzer MW. Institute of Neuroscience, University of Oregon. Eugene, Oregon USA 97405.
Barn owls localize sounds based primarily on the interaural differences in the sound-level and time of arrival of sounds. Neurons in its auditory space map, found in the external nucleus of the owls inferior colliculus (ICx), have discrete spatial receptive fields (RFs) based on the computation of interaural time and level differences (ITD, ILD). Using virtual auditory space (VAS) techniques, we recently measured the RF of the neurons were the space-map neurons selective for one or the other cue. The RF that a cell would have were it selective for only the ILD typically consists of a horizontal zone of high activity, often flanked above and below by regions where spontaneous activity is inhibited. The ITD component of a cell on the other hand is a vertical strip of activity flanked by inhibitory regions. The normal RF of the cell likes at the intersection of the ITD and ILD-based RFs.
How does the activity of space-map neurons relate to spatial acuity measured behaviorally? We have recently measured the minimal audible angle (MAA) in the barn owl using a novel technique that exploits the habituation and recovery of the pupilary dilation response (PDR). The owl's pupil dilates when a sound is presented in a quiet environment. Repeated presentation of the sound habituates this reflex. If some parameter, such as the sound source's position is altered, the PDR recovers, indicating that the change was detected. This procedure is carried out in head-fixed, untrained birds. Measures of behavioral performance obtained with the PDR are similar to those from trained birds in operant tasks. For example, the detection thresholds for a noise burst measured with the PDR and with a head-saccade task are similar, as are frequency-discrimination thresholds. This study revealed that the MAA was 3E in azimuth and 7.5E in elevation.
Application of signal detection theory to the azimuthal component of the neurons' spatial tuning curve revealed that the changes of firing at the slope of a tuning curve predict a finer MAA than that measured behaviorally. We therefore tested the hypothesis that the behavioral MAA reflects the performance of the population of space map neurons. In the PDR task, the PDR was habituated using repeated presentations of sound from a single location, and then probed for recovery by presentation of sounds from different locations during test trials. During a habituating trial, the noise burst from location x is expected to produce a focus of neural activity at the space map site that represents the location x. When a test stimulus is presented from location x+dx, the focus of activity would move, causing the population activity at the space-map's representation of x to change. We suggest that it is this change in activity of the neurons representing location x, x+dx that serves as the cue for discrimination. In other words, the owl responds to a test stimulus when it is able to reliably detect a change in evoked activity in its space map at the site corresponding to location x. To test this hypothesis, we estimated the activity of a neuronal population from single unit recordings taken from 86 space map neurons. Each neuron's azimuthal tuning curve was taken to represent a cross section through the focus of activity on the space map and the change in the level of activity for a shift of source loci from x and x+dx was simulated. By using tuning curves from single units assessed at high resolution (1E) in VAS, we are able to incorporate the trial to trial variance in firing as well as the diversity of response characteristics (firing rate, tonic/phasic, RF breadth) found in the space map. This model predicted the performance of the bird in the MAA task quite accurately, suggesting that the bird relies on the change in the neural image to discriminate habituating and test trials at x and x+dx.
The auditory responses of neurons in the song system, and in particular HVC, have been well characterized. The neurons in the song system exhibit similar response properties and overall are highly selective to the sound of the bird's own song (BOS). In contrast, the ensemble responses properties of auditory neurons in the auditory forebrain and midbrain are more heterogenous and less selective. I will first contrast the selectivity of neurons in the song system and the auditory forebrain for behaviorally relevant sounds. I will then review the basic properties of auditory neurons in the midbrain (Mld), primary auditory forebrain (field L) and secondary auditory forebrain (cHV) from work done in our laboratory and in other labs. I will then describe how we are estimating the Spectral Temporal Receptive Fields (STRF) of auditory neurons at these different stages of processing. By obtaining STRF from an ensemble of sounds that have flat modulation spectrum, we are able to test whether the ensemble of neurons are tuned to the particular spectral and temporal modulations found in song. Our data show that both in the midbrain and in the forebrain the ensemble response properties of neurons are indeed tuned to modulations found in natural sounds. I will also describe how the encoding of complex sounds changes from Mld to field L to cHV and what are the potential theoretical advantages to these different coding regimes and transformations.
A prominent characteristic of cortical responses in the awake condition is the abundance of persistent (sustained) discharges that often span the entire stimulus duration and beyond. In general, neurons are more likely to fire in a sustained manner when stimulated by a nearly optimal stimulus, whereas they tend to display only phasic (onset) responses when stimulated by non-optimal stimuli. Neurons often show greater selectivity to particular stimulus parameters in their sustained discharges than in their onset discharges. Under certain stimulus conditions (such as at high repetition rates), subpopulations of cortical neurons exhibit persistent firings that do not bear stimulus-related temporal structures. Such sustained discharges are likely the results of temporal-to-rate transformations and therefore represent processed stimulus information. Because sustained discharges have longer latencies than onset discharges, they are more likely to reflect properties resulting from or enhanced by cortico-cortical processing both within and across cortical areas. The increase in sustained activities suggests an increased excitability of cortical neurons in the awake condition. At the same time, auditory cortical neurons also exhibit stronger context-dependent inhibition in the awake condition. This inhibition appears to limit the range of stimuli to which a neuron may respond and contributes to a greater degree of non-monotonicity with regard to sound level. The emergence of stronger excitation and inhibition may underlie the increased stimulus selectivity in the auditory cortex in the awake condition.